CN118171744B - Prediction method and device for spatiotemporal distribution, electronic device and storage medium - Google Patents

Abstract

According to the space-time distribution prediction method and device, the electronic equipment and the storage medium, original data are filtered to obtain target data, the target data are constructed to obtain a user stay state vector sequence, and the target data are converted to obtain a user data vector; inputting the user stay state vector sequence and the user data vector into a prediction model to carry out first calculation to obtain a first hidden state, and carrying out second calculation on the user stay state vector sequence and the user data vector by the prediction model to obtain a second hidden state; and obtaining a third hidden state according to the first hidden state and the second hidden state, and converting the third hidden state to obtain a distribution prediction result. Compared with the related art, the method and the device have the advantages that the vectors are input into the prediction model in the positive sequence and the negative sequence to conduct forward prediction and reverse prediction, and the two prediction results are converted into the final distribution prediction result, so that the accuracy of the prediction result of the user distribution is improved.

Description

Prediction method and device for spatiotemporal distribution, electronic device and storage medium

Technical Field

The present disclosure relates to the field of data processing technology, and in particular to a prediction method and device for spatiotemporal distribution, an electronic device, and a storage medium.

Background Art

With the development of science and technology, the rise of emerging technologies such as cloud computing and big data has brought new vitality to all walks of life and derived various new business needs. Cloud computers, as the final form of computer virtualization, are the product of cloud computing. They can project cloud computers onto any device through high-speed network transmission, such as mobile phones, tablets, set-top boxes, etc. Due to their convenience and availability, they have quickly occupied the market of business office and public entertainment. Therefore, it is of great theoretical and practical significance to predict the spatiotemporal distribution of cloud computer users based on the field of deep learning, so as to accurately deliver cloud computer products to the target population and promote usage scenarios.

However, the existing methods for predicting the spatiotemporal distribution of cloud computer users require high data collection accuracy and pay more attention to the spatial location information of mobile users, resulting in low accuracy in the prediction results of the final user distribution.

Summary of the invention

The present disclosure provides a method, device, electronic device and storage medium for predicting spatiotemporal distribution. The main purpose is to solve the problem that the existing method for predicting the spatiotemporal distribution of cloud computer users has high requirements for data collection accuracy and pays more attention to the spatial location information of mobile users, resulting in low accuracy of the prediction results of the final user distribution.

According to a first aspect of the present disclosure, a method for predicting spatiotemporal distribution is provided, comprising:

Filtering the collected original user data to obtain target user data, performing data construction processing on the target user data to obtain a user stay state vector sequence corresponding to the target user data, and performing data conversion processing on the target user data to obtain a user data vector corresponding to the target user data;

Inputting the user stay state vector sequence and the user data vector into a pre-trained prediction model for a first inference process to obtain first hidden state data corresponding to the target user data, and performing a second inference process on the user stay state vector sequence and the user data vector based on the pre-trained prediction model to obtain second hidden state data corresponding to the target user data;

The third hidden state data corresponding to the target user data is obtained according to the first hidden state data and the second hidden state data, and data conversion processing is performed on the third hidden state data to obtain a corresponding distribution prediction result.

Optionally, performing data construction processing on the target user data to obtain a user stay state vector sequence corresponding to the target user data includes:

According to the spatial data information and the temporal data information carried in the target user data, the target user data is converted into data to obtain a user stay state vector;

The user stay state vectors are arranged according to the time data information to obtain the user stay state vector sequence.

Optionally, performing data conversion processing on the target user data to obtain a user data vector corresponding to the target user data includes:

Performing data conversion processing on the geographic data in the target user data to obtain a geographic data vector;

Performing data conversion processing on other data in the target user data by using a preset data conversion algorithm to obtain other data vectors, wherein the other data is all data in the target user data except the geographic data;

The geographic data vector and the other data vectors are combined to obtain the user data vector.

Optionally, the step of inputting the user stay state vector sequence and the user data vector into a pre-trained prediction model for first inference processing to obtain first hidden state data corresponding to the target user data includes:

Activate the user stay state vector sequence and the user data vector through a first preset activation function to obtain first activation input data, where the first activation input data includes the time data information, and the time data information at least includes time period information for generating the user data;

After data calculation processing is performed according to the first activation input data, the first historical hidden state data, and the first weight coefficient and the bias value, activation is performed through the first preset activation function to obtain first forget gate data, wherein the first historical hidden state data is hidden state data calculated according to user data generated in an adjacent time period before the time period information, and the first weight coefficient and the bias value are used to control the data forgetting degree of the first activation input data;

After data calculation processing is performed according to the first activation input data, the first historical hidden state data, and the second weight coefficient and the bias value, activation is performed through the first preset activation function to obtain first input gate data, and the second weight coefficient and the bias value are used to control the new data acceptance degree of the first activation input data;

After data calculation processing is performed according to the first activation input data, the first historical hidden state data, and the third weight coefficient and the bias value, activation is performed through a second preset activation function to obtain first candidate memory data, and the third weight coefficient and the bias value are used to perform data adjustment on the first activation input data;

Performing data calculation processing according to the first forget gate data, the first input gate data, the first candidate memory data, and the first historical update memory data to obtain first update memory data, where the first historical update memory data is update memory data calculated according to user data generated in an adjacent time period before the time period information;

After data calculation processing is performed according to the first activation input data, the first historical hidden state data, and a fourth weight coefficient and a bias value, activation is performed through the first preset activation function to obtain first output gate data, and the fourth weight coefficient and the bias value are used to control the degree of output state generation of the first activation input data;

Data calculation and processing are performed according to the first updated memory data and the first output gate data to obtain the first hidden state data.

Optionally, performing a second inference process on the user stay state vector sequence and the user data vector based on the pre-trained prediction model to obtain second hidden state data corresponding to the target user data includes:

Activate the user stay state vector sequence and the user data vector by a third preset activation function to obtain second activation input data, where the second activation input data includes the time data information, and the time data information includes at least time period information for generating the user data;

After data calculation processing is performed according to the second activation input data, the second historical hidden state data, and the fifth weight coefficient and the bias value, activation is performed through the third preset activation function to obtain second forget gate data, wherein the second historical hidden state data is hidden state data calculated according to user data generated in an adjacent time period after the time period information, and the fifth weight coefficient and the bias value are used to control the data forgetting degree of the second activation input data;

After data calculation processing is performed according to the second activation input data, the second historical hidden state data, and the sixth weight coefficient and the bias value, activation is performed through the third preset activation function to obtain second input gate data, and the sixth weight coefficient and the bias value are used to control the new data acceptance degree of the second activation input data;

After data calculation processing is performed according to the second activation input data, the second historical hidden state data, and the seventh weight coefficient and the bias value, activation is performed through a fourth preset activation function to obtain second candidate memory data, and the seventh weight coefficient and the bias value are used to perform data adjustment on the second activation input data;

Performing data calculation processing according to the second forget gate data, the second input gate data, the second candidate memory data, and the second historical update memory data to obtain second update memory data, where the second historical update memory data is update memory data calculated according to user data generated in an adjacent time period after the time period information;

After data calculation processing is performed according to the second activation input data, the second historical hidden state data, and the eighth weight coefficient and the bias value, activation is performed through the third preset activation function to obtain second output gate data, and the eighth weight coefficient and the bias value are used to control the degree of output state generation of the second activation input data;

Data calculation processing is performed according to the second updated memory data and the second output gate data to obtain the second hidden state data.

Optionally, performing data conversion processing on the third hidden state data to obtain a corresponding distribution prediction result includes:

Inputting the third hidden state data into a preset fully connected structure for full connection processing to obtain fully connected data;

The fully connected data is subjected to data activation processing by using a fifth preset activation function to obtain probability distribution data, and the probability distribution data is used as the distribution prediction result.

According to a second aspect of the present disclosure, there is provided a prediction device for spatiotemporal distribution, comprising:

A filtering unit, used to filter the collected original user data to obtain target user data;

a processing unit, configured to perform data construction processing on the target user data to obtain a user stay state vector sequence corresponding to the target user data, and to perform data conversion processing on the target user data to obtain a user data vector corresponding to the target user data;

an input unit, configured to input the user stay state vector sequence and the user data vector into a pre-trained prediction model for a first inference process to obtain first hidden state data corresponding to the target user data, and to perform a second inference process on the user stay state vector sequence and the user data vector based on the pre-trained prediction model to obtain second hidden state data corresponding to the target user data;

A conversion unit is used to obtain third hidden state data corresponding to the target user data according to the first hidden state data and the second hidden state data, and perform data conversion processing on the third hidden state data to obtain a corresponding distribution prediction result.

Optionally, the processing unit includes:

A conversion module, configured to perform data conversion on the target user data according to the spatial data information and the temporal data information carried in the target user data, so as to obtain a user stay state vector;

The arrangement module is used to arrange the user stay state vector according to the time data information to obtain the user stay state vector sequence.

Optionally, the processing unit further includes:

A conversion module, used for performing data conversion processing on the geographic data in the target user data to obtain a geographic data vector;

The conversion module is further used to perform data conversion processing on other data in the target user data by using a preset data conversion algorithm to obtain other data vectors, wherein the other data is all data in the target user data except the geographic data;

The merging module is used to merge the geographic data vector and the other data vectors to obtain the user data vector.

Optionally, the input unit includes:

A first activation module, configured to activate the user stay state vector sequence and the user data vector by using a first preset activation function to obtain first activation input data, wherein the first activation input data includes the time data information, and the time data information includes at least time period information for generating the user data;

A first calculation module, configured to perform data calculation processing according to the first activation input data, the first historical hidden state data, and the first weight coefficient and the bias value, and then activate the data through the first preset activation function to obtain first forget gate data, wherein the first historical hidden state data is hidden state data calculated according to user data generated in an adjacent time period before the time period information, and the first weight coefficient and the bias value are used to control the data forgetting degree of the first activation input data;

The first calculation module is further used to, after performing data calculation processing according to the first activation input data, the first historical hidden state data, and the second weight coefficient and the bias value, activate through the first preset activation function to obtain the first input gate data, and the second weight coefficient and the bias value are used to control the new data acceptance degree of the first activation input data;

The first calculation module is further used to, after performing data calculation processing according to the first activation input data, the first historical hidden state data, and the third weight coefficient and the bias value, activate through a second preset activation function to obtain first candidate memory data, and the third weight coefficient and the bias value are used to perform data adjustment on the first activation input data;

The first calculation module is further used to perform data calculation processing according to the first forget gate data, the first input gate data, the first candidate memory data and the first historical update memory data to obtain first update memory data, where the first historical update memory data is update memory data calculated according to user data generated in an adjacent time period before the time period information;

The first calculation module is further used to, after performing data calculation processing according to the first activation input data, the first historical hidden state data, and a fourth weight coefficient and a bias value, activate through the first preset activation function to obtain first output gate data, and the fourth weight coefficient and the bias value are used to control the output state generation degree of the first activation input data;

The first calculation module is further used to perform data calculation processing according to the first updated memory data and the first output gate data to obtain the first hidden state data.

Optionally, the input unit further includes:

A second activation module is used to activate the user stay state vector sequence and the user data vector through a third preset activation function to obtain second activation input data, where the second activation input data includes the time data information, and the time data information at least includes time period information for generating the user data;

a second calculation module, configured to perform data calculation processing according to the second activation input data, the second historical hidden state data, and the fifth weight coefficient and the bias value, and then activate the data through the third preset activation function to obtain second forget gate data, wherein the second historical hidden state data is hidden state data calculated according to user data generated in an adjacent time period after the time period information, and the fifth weight coefficient and the bias value are used to control the data forgetting degree of the second activation input data;

The second calculation module is further used to, after performing data calculation processing according to the second activation input data, the second historical hidden state data, and the sixth weight coefficient and the bias value, activate through the third preset activation function to obtain second input gate data, and the sixth weight coefficient and the bias value are used to control the new data acceptance degree of the second activation input data;

The second calculation module is further used to, after performing data calculation processing according to the second activation input data, the second historical hidden state data, and the seventh weight coefficient and the bias value, activate through a fourth preset activation function to obtain second candidate memory data, and the seventh weight coefficient and the bias value are used to perform data adjustment on the second activation input data;

The second calculation module is further used to perform data calculation processing according to the second forget gate data, the second input gate data, the second candidate memory data and the second historical update memory data to obtain second update memory data, where the second historical update memory data is update memory data calculated according to user data generated in an adjacent time period after the time period information;

The second calculation module is further used to, after performing data calculation processing according to the second activation input data, the second historical hidden state data, and the eighth weight coefficient and the bias value, activate through the third preset activation function to obtain second output gate data, and the eighth weight coefficient and the bias value are used to control the output state generation degree of the second activation input data;

The second calculation module is further used to perform data calculation processing according to the second updated memory data and the second output gate data to obtain the second hidden state data.

Optionally, the conversion unit comprises:

A processing module, used for inputting the third hidden state data into a preset fully connected structure for performing fully connected processing to obtain fully connected data;

An activation module is used to perform data activation processing on the fully connected data through a fifth preset activation function to obtain probability distribution data, and use the probability distribution data as the distribution prediction result.

According to a third aspect of the present disclosure, there is provided an electronic device, including:

at least one processor; and

a memory communicatively connected to the at least one processor; wherein,

The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect.

According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute the method described in the first aspect.

According to a fifth aspect of the present disclosure, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the method as described in the first aspect above.

The prediction method and device for spatiotemporal distribution, electronic device and storage medium provided by the present disclosure filter the collected original user data to obtain target user data, perform data construction processing on the target user data to obtain a user stay state vector sequence corresponding to the target user data, and perform data conversion processing on the target user data to obtain a user data vector corresponding to the target user data; input the user stay state vector sequence and the user data vector into a pre-trained prediction model for a first inference processing to obtain first hidden state data corresponding to the target user data, and perform a second inference processing on the user stay state vector sequence and the user data vector based on the pre-trained prediction model to obtain second hidden state data corresponding to the target user data; obtain third hidden state data corresponding to the target user data based on the first hidden state data and the second hidden state data, and perform data conversion processing on the third hidden state data to obtain a corresponding distribution prediction result. Compared with the related art, the embodiment of the present disclosure converts user data into vectors, inputs the converted vectors into the prediction model in forward and reverse order for forward and reverse calculation processing, and converts the two calculation results into final distribution prediction results. It does not require high-precision user data and pays attention to the user's spatial location information and time feature information at the same time, thereby improving the accuracy of the prediction results of user distribution.

It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

FIG1 is a schematic flow chart of a method for predicting spatiotemporal distribution provided by an embodiment of the present disclosure;

FIG2 is a schematic diagram showing the principle of a method for predicting spatiotemporal distribution provided by an embodiment of the present disclosure;

FIG3 is a schematic diagram of a process for constructing a user stay state vector sequence provided by an embodiment of the present disclosure;

FIG4 is a schematic diagram of a conversion process of a user data vector provided by an embodiment of the present disclosure;

FIG5 is a schematic diagram of a flow chart of a first inference process provided by an embodiment of the present disclosure;

FIG6 is a schematic diagram of a flow chart of a second inference process provided by an embodiment of the present disclosure;

FIG7 is a schematic diagram of the structure of a prediction device for spatiotemporal distribution provided by an embodiment of the present disclosure;

FIG8 is a schematic diagram of the structure of another device for predicting spatiotemporal distribution provided by an embodiment of the present disclosure;

FIG. 9 is a schematic block diagram of an electronic device provided by an embodiment of the present disclosure.

DETAILED DESCRIPTION

The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

The following describes the spatiotemporal distribution prediction method and device, electronic device, and storage medium according to embodiments of the present disclosure with reference to the accompanying drawings.

FIG1 is a schematic flow chart of a method for predicting spatiotemporal distribution provided in an embodiment of the present disclosure.

As shown in Figure 1, the method includes the following steps:

Step 101, filtering the collected original user data to obtain target user data, performing data construction processing on the target user data to obtain a user stay state vector sequence corresponding to the target user data, and performing data conversion processing on the target user data to obtain a user data vector corresponding to the target user data.

In the disclosed embodiment, the original user data is the cloud computer user data collected from the database, including but not limited to: the total number of cloud computer users, cloud computer user spatial information, time information, semantic information, user unique ID, geographic data, etc. The way to filter the original user data includes but is not limited to: filters and threshold determination, etc. In addition, it can also be filtered by a filter with customized filter conditions. The main purpose of filtering the original user data is to remove invalid data such as noise data, redundant data, and duplicate data. Regarding threshold determination, for example, the threshold condition set is the time the user uses the cloud computer, and the threshold is 30 minutes. When filtering the original user data, the user data of the time the cloud computer is used for less than 30 minutes will be filtered out.

When constructing the user stay state vector sequence of the target user data, the spatial dimension will be mapped according to the various data contained in the target user data, and the time attribute will be attached to obtain the user stay state vector sequence containing the user's time and space characteristics; when converting the target user data into a user data vector, each data in the original user data is converted separately to obtain the vector data corresponding to each data, and then the user data vector is jointly constructed according to the vector data corresponding to each data.

Step 102: Input the user stay state vector sequence and the user data vector into a pre-trained prediction model for a first inference processing to obtain first hidden state data corresponding to the target user data; and perform a second inference processing on the user stay state vector sequence and the user data vector based on the pre-trained prediction model to obtain second hidden state data corresponding to the target user data.

In the disclosed embodiment, the structure of the pre-trained prediction model includes but is not limited to: Deep Recurrent Neural Network (DRNN), Bi-Long Short Term Memory (Bi-LSTM), etc. When the user stay state vector sequence and the user data vector are inferred by the prediction model, the inference is performed in the Bi-LSTM structure of the prediction model, and it is necessary to perform forward long short-term memory network neuron inference and reverse long short-term memory network neuron inference on the user stay state vector sequence and the user data vector.

Among them, when performing forward long short-term memory network neuron extrapolation, the user stay state vector sequence and the user data vector are input into the Bi-LSTM structure in a forward order input manner, and when performing reverse long short-term memory network neuron extrapolation, the user stay state vector sequence and the user data vector are input into the Bi-LSTM structure in a reverse order input manner. The forward order input is to input the user stay state vector sequence and the user data vector according to the order of time, and the reverse order input is to input the user stay state vector sequence and the user data vector according to the order opposite to the time. For example, the user stay state vector sequence and the user data vector are converted from the target user data between 8:00-12:00. The forward order input is to input the user stay state vector sequence and the user data vector in the order of 8:00-12:00, and the reverse order input is to input the user stay state vector sequence and the user data vector in the order of 12:00-8:00. At the same time, there is no clear limitation on the order of performing forward long short-term memory network neuron extrapolation and reverse long short-term memory network neuron extrapolation. For example, forward long short-term memory network neuron extrapolation can be performed first, and then reverse long short-term memory network neuron extrapolation can be performed; reverse long short-term memory network neuron extrapolation can be performed first, and then forward long short-term memory network neuron extrapolation can be performed; forward long short-term memory network neuron extrapolation and reverse long short-term memory network neuron extrapolation can also be performed simultaneously. Specifically, the embodiments of the present disclosure are not limited.

Furthermore, in order to facilitate the understanding of the embodiments of the present disclosure, in the subsequent embodiments, the first inference processing represents the forward long short-term memory network neuron inference and the second inference processing represents the reverse long short-term memory network neuron inference.

Step 103: Obtain third hidden state data corresponding to the target user data according to the first hidden state data and the second hidden state data, and perform data conversion processing on the third hidden state data to obtain a corresponding distribution prediction result.

In the embodiment of the present disclosure, the third hidden state data may be expressed in a form including but not limited to a vector form, a matrix form, etc. The third hidden state data is obtained by merging the first hidden state data and the second hidden state data, that is, the third hidden state data is obtained by connecting the hidden state of the forward LSTM and the hidden state of the reverse LSTM to obtain the hidden state of the bidirectional LSTM. The third hidden state data may be obtained by formula (1):

Formula (1)

in, is the third hidden state data, is the first hidden state data, is the second hidden state data.

Since the third hidden state data is represented in the form of a vector, a matrix, etc., the third hidden data needs to be converted before the distribution prediction result can be obtained.

Specifically, in order to facilitate understanding of the implementation process of the embodiment of the present disclosure, a principle schematic diagram of a spatiotemporal distribution prediction method is provided, as shown in Figure 2, wherein the data layer is an operation process of filtering the original user data, and the user stay state vector sequence and the user data vector must be converted by the conversion layer before they can be obtained. The model training layer is an operation process of the prediction model processing the user stay state vector sequence and the user data vector, and the result layer is the obtained distribution prediction result.

The present disclosure provides a method for predicting spatiotemporal distribution, which filters the collected original user data to obtain target user data, performs data construction processing on the target user data to obtain a user stay state vector sequence corresponding to the target user data, and performs data conversion processing on the target user data to obtain a user data vector corresponding to the target user data; inputs the user stay state vector sequence and the user data vector into a pre-trained prediction model for a first inference processing to obtain first hidden state data corresponding to the target user data, and performs a second inference processing on the user stay state vector sequence and the user data vector based on the pre-trained prediction model to obtain second hidden state data corresponding to the target user data; obtains third hidden state data corresponding to the target user data based on the first hidden state data and the second hidden state data, and performs data conversion processing on the third hidden state data to obtain a corresponding distribution prediction result. Compared with the related art, the embodiment of the present disclosure converts user data into vectors, inputs the converted vectors into the prediction model in forward and reverse order for forward and reverse calculation processing, and converts the two calculation results into final distribution prediction results. It does not require high-precision user data and pays attention to the user's spatial location information and time feature information at the same time, thereby improving the accuracy of the prediction results of user distribution.

In one possible implementation of the embodiment of the present disclosure, as a refinement of the above step 101, regarding the construction of the user stay state vector sequence, the embodiment of the present disclosure provides a schematic diagram of the construction process of the user stay state vector sequence, as shown in FIG3, including:

Step 301: Perform data conversion on the target user data according to the spatial data information and time data information carried in the target user data to obtain a user stay state vector.

In the embodiment of the present disclosure, it is necessary to map the target user data to the spatial dimension according to the spatial data information carried in the target user data, and at the same time add a time attribute to the mapped target user data according to the time data information, thereby converting it into a single trajectory point of the user and expressing it as a vector to obtain the user's stay state vector.

Among them, the spatial data information includes but is not limited to: the location of the user when logging into the cloud computer, the movement trajectory of the user when using the cloud computer, etc., and the time data information includes but is not limited to: the time period information for generating the user data, the time information when the user logs into the cloud computer, the time information when the user uses the cloud computer, the current time information, etc. Specifically, the present disclosed embodiment does not limit the content of the spatial data information and the time data information.

Step 302: Arrange the user stay state vectors according to the time data information to obtain the user stay state vector sequence.

In an embodiment of the present disclosure, a user stay state vector may represent a user's location at a certain time, and the user stay state vector sequence is data obtained by arranging corresponding user stay state vectors according to time. For example, the target user data is data from 8:00 to 12:00, and the user stay state vector is a single position of the user corresponding to each preset time period from 8:00 to 12:00, and the preset time period is a custom setting (such as 10 seconds, 1 minute, etc.). The arrangement of the user stay state vector is to arrange the user stay state vectors within 8:00 to 12:00 in chronological order to obtain a user stay state vector sequence, and the user stay state vector sequence may represent a single trajectory point of the user containing space and time information.

In an implementable manner of the embodiment of the present disclosure, as a refinement of the above step 101, regarding the conversion of the user data vector, the embodiment of the present disclosure provides a schematic diagram of the conversion process of the user data vector, as shown in FIG4, including:

Step 401: Perform data conversion processing on the geographic data in the target user data to obtain a geographic data vector.

In the embodiments of the present disclosure, since geographic data includes but is not limited to: raster data, vector data with built-in spatial information such as distance intervals, etc., which are data used to represent geographic location, the geographic data can be directly converted to obtain corresponding geographic data vectors. The geographic data vectors include but are not limited to: raster geographic indicator data vectors, distance interval data vectors, etc. The geographic data vectors can be used to represent information about the user's geographic location.

Step 402: Perform data conversion processing on other data in the target user data by using a preset data conversion algorithm to obtain other data vectors, wherein the other data is all data in the target user data except the geographic data.

In the embodiments of the present disclosure, the preset data conversion algorithm is a conversion algorithm with custom settings, for example: an improved GloVe method based on natural language processing technology, etc. The other data include but are not limited to: the total number of cloud computer users, cloud computer user spatial information, time information, semantic information, user unique ID, etc. Specifically, the embodiments of the present disclosure do not limit the content of the preset data conversion algorithm and the other data.

Since other data in the target user, such as time information, semantic information, user unique ID, etc., are non-vector data and cannot be directly converted into vector data, it is necessary to use a preset data conversion algorithm for conversion. Regarding the conversion process, for example, the improved GloVe method based on natural language processing technology is used to convert non-vector data such as time information, user unique ID, semantic information, etc., through co-occurrence matrix, singular value decomposition, semantic weighting, principal component analysis and other processes to corresponding time information vectors, user unique ID vectors, voice information vectors, etc. The other data vectors include but are not limited to: time information vectors, user unique ID vectors, voice information vectors, etc.

Step 403: merge the geographic data vector and the other data vectors to obtain the user data vector.

In the embodiment of the present disclosure, the merging process of the geographic data vector and other data vectors refers to merging all vectors in the geographic data vector and all vectors in the other data vectors to obtain the user data vector, where the user data vector represents a set of vectors corresponding to all data in the target user data.

In an implementable manner of an embodiment of the present disclosure, as a refinement of the above step 102, regarding the first inference processing of the user stay state vector sequence and the user data vector, an embodiment of the present disclosure provides a flow chart of a first inference processing, as shown in FIG5 , including:

Step 501, activate the user stay state vector sequence and the user data vector through a first preset activation function to obtain first activation input data, wherein the first activation input data includes the time data information, and the time data information at least includes the time period information for generating the user data.

In the embodiment of the present disclosure, the user stay state vector sequence and the user data vector need to be merged first to obtain input data. The input vector can represent the spatiotemporal position state in the user's current time data information. The input data may be expressed in a form including but not limited to a vector form, a matrix form, etc. The input data may be expressed as: ,in, Represents input data, represents the current time step, that is, the time data information, It indicates the location of the user's stay point at the current time step, that is, the user's location information in the time data information.

The first preset activation function is a custom-selected activation function, such as a Sigmoid function, etc. Specifically, the embodiment of the present disclosure does not limit the first preset activation function.

For ease of understanding, the first preset activation function is described below using the Sigmoid function as an example. Specifically, the Sigmoid function can be expressed by formula (2):

Formula (2)

At this time, the first activation input data can be expressed as .

Step 502, after data calculation and processing is performed according to the first activation input data, the first historical hidden state data, the first weight coefficient and the bias value, activation is performed through the first preset activation function to obtain the first forgetting gate data, wherein the first historical hidden state data is the hidden state data calculated according to the user data generated in the adjacent time period before the time period information, and the first weight coefficient and the bias value are used to control the data forgetting degree of the first activation input data.

In the disclosed embodiment, the first forget gate data is the forward forget gate state obtained after the first activation input data is calculated by the forward forget gate. The forward forget gate determines which data in the first activation input data should be discarded or retained from the first activation input data. The previous hidden state and the current first activation input data are checked through the forward forget gate, and a numerical value between 0 and 1 is output to each data in the first activation input data. 1 represents "completely retain this data", and 0 represents "completely discard this data". If the output of the forward forget gate is 1, the corresponding data in the first activation input data will remain unchanged; if the output is 0, the corresponding data will be forgotten, and if the output is between 0 and 1, the corresponding data will be adjusted according to the output value (retain part of the data).

Specifically, the calculation process of the first forget gate data can be performed using but not limited to formula (3):

Formula (3)

in, represents the forward forget gate state, i.e., the first forget gate data, represents the forward hidden layer state value at time t-1, i.e., the first historical hidden state data, and They respectively represent the weight coefficient and bias value of the forward forget gate, namely the first weight coefficient and bias value.

It should be noted that the first weight coefficient and bias value are trainable data, and the initial first weight coefficient and bias value are randomly generated data, and the first weight coefficient and bias value will be trained synchronously with the training of the prediction model. Therefore, in the pre-trained prediction model, the first weight coefficient and bias value are also trained data and can be directly called for use.

Step 503, after data calculation and processing according to the first activation input data, the first historical hidden state data, and the second weight coefficient and bias value, activation is performed through the first preset activation function to obtain the first input gate data, and the second weight coefficient and bias value are used to control the degree of acceptance of new data of the first activation input data.

In the disclosed embodiment, the first input gate data is the forward input gate state obtained after the first activation input data is calculated by the forward input gate. The forward input gate is responsible for updating the hidden state and determining what new data will be stored. Specifically, the calculation process of the first input gate data can be performed using, but not limited to, formula (4):

Formula (4)

in, represents the state of the forward input gate, i.e., the first input gate data, and The weight coefficient and bias value representing the forward input gate are the second weight coefficient and bias value.

For the second weight coefficient and the bias value, please refer to the description of the first weight coefficient and the bias value in the above step 502, so they will not be repeated here.

Step 504, after data calculation and processing are performed based on the first activation input data, the first historical hidden state data, and the third weight coefficient and bias value, activation is performed through a second preset activation function to obtain first candidate memory data, and the third weight coefficient and bias value are used to perform data adjustment on the first activation input data.

In the disclosed embodiment, the first candidate memory data is the state of the positive candidate memory unit obtained after the first activation input data is calculated by the forward candidate memory unit. The positive candidate unit is a candidate value generated before the first activation input is updated. This candidate value contains state information for updating, but this state information for updating has not been confirmed by the gate controller of the LSTM. This candidate value will then be determined by a combination of the forward input gate and the forward forget gate to determine whether to actually update the first activation input data. The positive candidate memory unit is a vector whose elements are the product of the inputs from the previous hidden state to the current time step.

The second preset activation function is a custom-selected activation function, such as a tanh function, etc. Specifically, the embodiment of the present disclosure does not limit the second preset activation function.

Specifically, the calculation process of the first candidate memory data may be performed using, but not limited to, formula (5):

Formula (5)

in, represents the state of the positive candidate memory unit, i.e., the first candidate memory data, and The weight coefficient and bias value representing the positive candidate memory unit are the third weight coefficient and bias value.

For the third weight coefficient and the bias value, please refer to the description of the first weight coefficient and the bias value in the above step 502, so they will not be repeated here.

Step 505: Perform data calculation and processing based on the first forget gate data, the first input gate data, the first candidate memory data, and the first historical updated memory data to obtain first updated memory data, where the first historical updated memory data is updated memory data calculated based on user data generated in an adjacent time period before the time period information.

In the disclosed embodiment, the first updated memory data is the forward updated memory unit state obtained after the first activated input data is calculated by the forward updated memory unit. The forward updated memory unit integrates the state results of the forward forget gate, the forward input gate and the forward candidate memory unit. The forward forget gate determines what kind of data needs to be discarded, the forward input gate determines what kind of new data needs to be added, and the positive candidate memory unit provides new state information. The combination of these three types of data and the historical memory unit state update forms the forward memory unit state. At the same time, the forward memory unit state also needs to be passed to the next time step as the historical memory unit state.

Specifically, the calculation process of the first updated memory data may be performed using, but not limited to, formula (6):

Formula (6)

in, Indicates the state of the memory unit is updated in the forward direction, i.e., the first updated memory data, represents the state of the forward updated memory unit at the previous time step, i.e., the first historical updated memory data, Indicates the forward forget gate state (first forget gate data), Indicates the state of the positive input gate (first input gate data), Represents the state of the positive candidate memory unit (the first candidate memory data), Represents element-wise multiplication.

Step 506, after data calculation and processing according to the first activation input data, the first historical hidden state data, and the fourth weight coefficient and bias value, activation is performed through the first preset activation function to obtain the first output gate data, and the fourth weight coefficient and bias value are used to control the output state generation degree of the first activation input data.

In the disclosed embodiment, the first output gate data is the forward output gate state obtained after the first activated input data is calculated by the forward output gate. The forward output gate determines the data that the next hidden state should contain. By controlling the hidden state part actually to be output, the hidden state only contains the data that needs attention, thereby reducing the redundant data in the output hidden state and interfering with the final prediction result.

Specifically, the calculation process of the first output gate data may be performed using, but not limited to, formula (7):

Formula (7)

in, represents the forward output gate state, i.e., the first output gate data, and The weight coefficient and bias value representing the forward output gate are the fourth weight coefficient and bias value.

Regarding the fourth weight coefficient and the bias value, reference may be made to the description of the first weight coefficient and the bias value in the above step 502, so they will not be described in detail here.

Step 507: Perform data calculation processing according to the first updated memory data and the first output gate data to obtain the first hidden state data.

In the disclosed embodiment, the first hidden state data is the internal memory state of the LSTM structure in the prediction model at the output stage of the network. The internal memory state is determined by the information accumulated and transmitted by the network when processing sequence data. The first hidden state data can carry information of all previous time steps and participate in the calculation of subsequent time steps. Therefore, the calculation of the first hidden state data is the key to the LSTM structure in the prediction model being able to process and predict sequence data.

Specifically, the calculation process of the first hidden state data may be performed using, but not limited to, formula (8):

Formula (8)

in, represents the first hidden state data, Indicates a merge calculation process. When calculating the first hidden state data, the first updated memory data needs to be activated by the second activation function and then calculated with the first output gate data to obtain the first hidden state data.

In an implementable manner of an embodiment of the present disclosure, as a refinement of the above step 102, regarding the second inference processing of the user stay state vector sequence and the user data vector, an embodiment of the present disclosure provides a flow chart of a second inference processing, as shown in FIG6, including:

Step 601, activate the user stay state vector sequence and the user data vector through a third preset activation function to obtain second activation input data, wherein the second activation input data includes the time data information, and the time data information at least includes the time period information for generating the user data.

In the embodiment of the present disclosure, similar to the first inference process described above, it is also necessary to merge the user stay state vector sequence with the user data vector to obtain input data. For details about the input data, please refer to the description in the above step 501, so it will not be repeated here.

The third preset activation function is a custom-selected activation function, such as a Sigmoid function, etc. Specifically, the third preset activation function is not limited in the embodiment of the present disclosure.

For ease of understanding, the subsequent third preset activation function is also explained using the Sigmoid function as an example. Therefore, the second activation input data can be expressed as: ,in, Enter data for the second activation.

Step 602, after data calculation and processing are performed according to the second activation input data, the second historical hidden state data, and the fifth weight coefficient and the bias value, activation is performed through the third preset activation function to obtain second forgetting gate data, wherein the second historical hidden state data is hidden state data calculated based on user data generated in an adjacent time period after the time period information, and the fifth weight coefficient and the bias value are used to control the data forgetting degree of the second activation input data.

In the embodiment of the present disclosure, regarding the second forget gate data and the reverse forget gate, please refer to the description of the first forget gate data and the forward forget gate in the above step 502, so they will not be repeated here.

Specifically, the calculation process of the second forget gate data can be performed using but not limited to formula (9):

Formula (9)

in, Represents the reverse forget gate state, i.e., the second forget gate data, represents the activation function Sigmoid, represents the reverse hidden layer state value at time t+1, i.e., the second historical hidden state data, and The weight coefficient and bias value representing the reverse forget gate are the fifth weight coefficient and bias value.

For the fifth weight coefficient and the bias value, reference may be made to the description of the first weight coefficient and the bias value in the above step 502, so they will not be described in detail here.

Step 603, after data calculation and processing according to the second activation input data, the second historical hidden state data, and the sixth weight coefficient and bias value, activation is performed through the third preset activation function to obtain the second input gate data, and the sixth weight coefficient and bias value are used to control the degree of acceptance of new data of the second activation input data.

In the embodiment of the present disclosure, regarding the second input gate data and the reverse input gate, please refer to the description regarding the first input gate data and the forward input gate in the above step 503, so they will not be described one by one here.

Specifically, the calculation process of the second input gate data may be performed using, but not limited to, formula (10):

Formula (10)

in, Indicates the reverse input gate state, i.e., the second input gate data, and The weight coefficient and bias value representing the reverse input gate are the sixth weight coefficient and bias value.

Regarding the sixth weight coefficient and the bias value, reference may be made to the description of the first weight coefficient and the bias value in the above step 502, so they will not be described in detail here.

Step 604, after data calculation and processing are performed based on the second activation input data, the second historical hidden state data, and the seventh weight coefficient and bias value, activation is performed through the fourth preset activation function to obtain the second candidate memory data, and the seventh weight coefficient and bias value are used to perform data adjustment on the second activation input data.

In the embodiment of the present disclosure, regarding the second candidate memory data and the reverse candidate memory unit, please refer to the description of the first candidate memory data and the forward candidate memory unit in the above step 504, so they are not repeated here.

The fourth preset activation function is an activation function selected by user, such as a tanh function, etc. Specifically, the fourth preset activation function is not limited in the embodiment of the present disclosure.

The calculation process of the second candidate memory data may be performed using, but not limited to, formula (11):

Formula (11)

in, represents the reverse candidate memory cell state, i.e., the second candidate memory data, and The weight coefficient and bias value representing the reverse candidate memory unit are the seventh weight coefficient and bias value.

Step 605: Perform data calculation and processing based on the second forget gate data, the second input gate data, the second candidate memory data and the second historical updated memory data to obtain second updated memory data, where the second historical updated memory data is updated memory data calculated based on user data generated in an adjacent time period after the time period information.

In the embodiment of the present disclosure, regarding the second updating memory data and the reverse updating memory unit, please refer to the description regarding the first updating memory data and the forward updating memory unit in the above step 505, so it will not be repeated here.

Specifically, the calculation process of the second candidate memory data may be performed using, but not limited to, formula (12):

Formula (12)

in, Indicates the reverse update of the memory unit state, i.e., the second candidate memory data, Indicates the reverse update memory unit state of the previous time step, i.e., the second historical update memory data, Represents the reverse forget gate state (second forget gate data), Indicates the reverse input gate state (second input gate data), Represents the reverse candidate memory cell state (second candidate memory data), Represents element-wise multiplication.

Step 606, after data calculation and processing according to the second activation input data, the second historical hidden state data, and the eighth weight coefficient and bias value, activation is performed through the third preset activation function to obtain second output gate data, and the eighth weight coefficient and bias value are used to control the degree of output state generation of the second activation input data.

In the embodiment of the present disclosure, regarding the second output gate data and the reverse output gate, please refer to the description regarding the first output gate data and the forward output gate in the above step 506, so they will not be described one by one here.

Specifically, the calculation process of the second output gate data may be performed using, but not limited to, formula (13):

Formula (13)

in, Indicates the reverse output gate state, i.e., the second output gate data, and The weight coefficient and bias value representing the reverse output gate are the eighth weight coefficient and bias value.

Step 607: Perform data calculation processing according to the second updated memory data and the second output gate data to obtain the second hidden state data.

In the embodiment of the present disclosure, regarding the second hidden state data, please refer to the description regarding the first hidden state data in the above step 507, so it will not be described in detail here.

Specifically, the calculation process of the second hidden state data may be performed using, but not limited to, formula (14):

Formula (14)

in, represents the second hidden state data, Indicates a merge calculation process. When calculating the second hidden state data, the second updated memory data needs to be activated by the fourth activation function and then calculated with the second output gate data to obtain the second hidden state data.

In one implementable method of the embodiment of the present disclosure, the third hidden state data is intermediate information for processing and predicting sequence data, and is expressed in the form of vectors, matrices, etc. It cannot be directly used as the result of distribution prediction, and data conversion processing is required for the third hidden state data. Therefore, in order to obtain the result of distribution prediction, it can be implemented by but not limited to the following methods: input the third hidden state data into a preset fully connected structure for full connection processing to obtain fully connected data; perform data activation processing on the fully connected data through the fifth preset activation function to obtain probability distribution data, and use the probability distribution data as the distribution prediction result.

In the embodiment of the present disclosure, the preset fully connected structure is a fully connected layer of custom selection, which is used to map the third hidden state data to the output space, and the fifth preset activation function is a function of custom selection, such as: softmax function, linear activation function, etc., which is used to convert the linear output mapped to the output space into a probability distribution, and output the predicted probability of the corresponding category to generate the final output result.

In summary, the embodiments of the present disclosure can achieve the following effects:

1. The disclosed embodiment converts user data into vectors, inputs the converted vectors into the prediction model in positive and reverse order for forward and reverse calculation, and converts the two calculation results into the final distribution prediction result. This eliminates the need for high-precision user data and simultaneously focuses on the user's spatial location information and temporal feature information, thereby improving the accuracy of the prediction result of the user distribution.

2. The disclosed embodiment uses an improved GloVe method based on natural language processing technology to effectively solve the problem that static text data cannot be represented by vectors. The improved GloVe method based on natural language processing technology is used to process the data in the target user data in a parallel manner of co-occurrence matrix and training learning, and convert the data into vectors, which can improve the efficiency of vector conversion.

3. The prediction model adopted by the disclosed embodiment includes a bidirectional LSTM recurrent neuron and a softmax function. The target user data set is inferred from the forward and reverse directions through the bidirectional LSTM recurrent neuron, so that the context expression is more compact, the characteristics in the time dimension are more significant, the time dimension can be better measured, and the distribution fluctuation of cloud computer users in the time dimension under different time granularities can be obtained, which can provide a visualization effect for the accurate promotion of cloud computer products.

Corresponding to the above-mentioned prediction method of spatiotemporal distribution, the present invention also proposes a prediction device of spatiotemporal distribution. Since the device embodiment of the present invention corresponds to the above-mentioned method embodiment, the details not disclosed in the device embodiment can be referred to the above-mentioned method embodiment, and will not be repeated in the present invention.

FIG. 7 is a schematic diagram of the structure of a prediction device for spatiotemporal distribution provided by an embodiment of the present disclosure. As shown in FIG. 7 , the device comprises:

The filtering unit 71 is used to filter the collected original user data to obtain target user data;

The processing unit 72 is used to perform data construction processing on the target user data to obtain a user stay state vector sequence corresponding to the target user data, and to perform data conversion processing on the target user data to obtain a user data vector corresponding to the target user data;

An input unit 73 is used to input the user stay state vector sequence and the user data vector into a pre-trained prediction model for a first inference process to obtain first hidden state data corresponding to the target user data, and to perform a second inference process on the user stay state vector sequence and the user data vector based on the pre-trained prediction model to obtain second hidden state data corresponding to the target user data;

The conversion unit 74 is used to obtain third hidden state data corresponding to the target user data according to the first hidden state data and the second hidden state data, and perform data conversion processing on the third hidden state data to obtain a corresponding distribution prediction result.

The prediction device for spatiotemporal distribution provided by the present disclosure filters the collected original user data to obtain target user data, performs data construction processing on the target user data to obtain a user stay state vector sequence corresponding to the target user data, and performs data conversion processing on the target user data to obtain a user data vector corresponding to the target user data; inputs the user stay state vector sequence and the user data vector into a pre-trained prediction model for a first inference processing to obtain first hidden state data corresponding to the target user data, and performs a second inference processing on the user stay state vector sequence and the user data vector based on the pre-trained prediction model to obtain second hidden state data corresponding to the target user data; obtains third hidden state data corresponding to the target user data based on the first hidden state data and the second hidden state data, and performs data conversion processing on the third hidden state data to obtain a corresponding distribution prediction result. Compared with the related art, the embodiment of the present disclosure converts user data into vectors, inputs the converted vectors into the prediction model in forward and reverse order for forward and reverse calculation processing, and converts the two calculation results into final distribution prediction results. It does not require high-precision user data and pays attention to the user's spatial location information and time feature information at the same time, thereby improving the accuracy of the prediction results of user distribution.

Furthermore, in a possible implementation of the embodiment of the present disclosure, as shown in FIG8 , the processing unit 72 includes:

The conversion module 721 is used to perform data conversion on the target user data according to the spatial data information and the time data information carried in the target user data to obtain a user stay state vector;

The arrangement module 722 is used to arrange the user stay state vector according to the time data information to obtain the user stay state vector sequence.

Furthermore, in a possible implementation of the embodiment of the present disclosure, as shown in FIG8 , the processing unit 72 further includes:

The conversion module 723 is used to perform data conversion processing on the geographic data in the target user data to obtain a geographic data vector;

The conversion module 723 is further used to perform data conversion processing on other data in the target user data by using a preset data conversion algorithm to obtain other data vectors, wherein the other data is all data in the target user data except the geographic data;

The merging module 724 is used to merge the geographic data vector and the other data vector to obtain the user data vector.

Further, in a possible implementation of the embodiment of the present disclosure, as shown in FIG8 , the input unit 73 includes:

A first activation module 731 is configured to activate the user stay state vector sequence and the user data vector by using a first preset activation function to obtain first activation input data, where the first activation input data includes the time data information, and the time data information includes at least time period information for generating the user data;

A first calculation module 732 is used to perform data calculation processing according to the first activation input data, the first historical hidden state data, and the first weight coefficient and the bias value, and then activate through the first preset activation function to obtain first forget gate data, wherein the first historical hidden state data is hidden state data calculated according to user data generated in an adjacent time period before the time period information, and the first weight coefficient and the bias value are used to control the data forgetting degree of the first activation input data;

The first calculation module 732 is further used to, after performing data calculation processing according to the first activation input data, the first historical hidden state data, and the second weight coefficient and the bias value, activate through the first preset activation function to obtain the first input gate data, and the second weight coefficient and the bias value are used to control the new data acceptance degree of the first activation input data;

The first calculation module 732 is further used to, after performing data calculation processing according to the first activation input data, the first historical hidden state data, and the third weight coefficient and the bias value, activate through a second preset activation function to obtain first candidate memory data, and the third weight coefficient and the bias value are used to perform data adjustment on the first activation input data;

The first calculation module 732 is further used to perform data calculation processing according to the first forget gate data, the first input gate data, the first candidate memory data and the first historical update memory data to obtain first update memory data, where the first historical update memory data is update memory data calculated according to user data generated in an adjacent time period before the time period information;

The first calculation module 732 is further used to, after performing data calculation processing according to the first activation input data, the first historical hidden state data, and the fourth weight coefficient and the bias value, activate through the first preset activation function to obtain the first output gate data, and the fourth weight coefficient and the bias value are used to control the output state generation degree of the first activation input data;

The first calculation module 732 is further used to perform data calculation processing according to the first updated memory data and the first output gate data to obtain the first hidden state data.

Furthermore, in a possible implementation of the embodiment of the present disclosure, as shown in FIG8 , the input unit 73 further includes:

A second activation module 733 is used to activate the user stay state vector sequence and the user data vector through a third preset activation function to obtain second activation input data, where the second activation input data includes the time data information, and the time data information at least includes time period information for generating the user data;

A second calculation module 734 is used to perform data calculation processing according to the second activation input data, the second historical hidden state data, and the fifth weight coefficient and the bias value, and then activate through the third preset activation function to obtain second forget gate data, wherein the second historical hidden state data is hidden state data calculated according to user data generated in an adjacent time period after the time period information, and the fifth weight coefficient and the bias value are used to control the data forgetting degree of the second activation input data;

The second calculation module 734 is further used to, after performing data calculation processing according to the second activation input data, the second historical hidden state data, and the sixth weight coefficient and the bias value, activate through the third preset activation function to obtain second input gate data, and the sixth weight coefficient and the bias value are used to control the new data acceptance degree of the second activation input data;

The second calculation module 734 is further used to, after performing data calculation processing according to the second activation input data, the second historical hidden state data, and the seventh weight coefficient and the bias value, activate through a fourth preset activation function to obtain second candidate memory data, and the seventh weight coefficient and the bias value are used to perform data adjustment on the second activation input data;

The second calculation module 734 is further used to perform data calculation processing according to the second forget gate data, the second input gate data, the second candidate memory data and the second historical update memory data to obtain second updated memory data, where the second historical update memory data is updated memory data calculated according to user data generated in an adjacent time period after the time period information;

The second calculation module 734 is further used to, after performing data calculation processing according to the second activation input data, the second historical hidden state data, and the eighth weight coefficient and the bias value, activate through the third preset activation function to obtain second output gate data, and the eighth weight coefficient and the bias value are used to control the output state generation degree of the second activation input data;

The second calculation module 734 is further used to perform data calculation processing according to the second updated memory data and the second output gate data to obtain the second hidden state data.

Furthermore, in a possible implementation of the embodiment of the present disclosure, as shown in FIG8 , the conversion unit 74 includes:

A processing module 741 is used to input the third hidden state data into a preset fully connected structure for full connection processing to obtain fully connected data;

The activation module 742 is used to perform data activation processing on the fully connected data through a fifth preset activation function to obtain probability distribution data, and use the probability distribution data as the distribution prediction result.

It should be noted that the above explanation of the method embodiment is also applicable to the device of the embodiment of the present disclosure, and the principle is the same, which is no longer limited in the embodiment of the present disclosure.

According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

FIG9 shows a schematic block diagram of an example electronic device 900 that can be used to implement an embodiment of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and/or required herein.

As shown in FIG9 , the device 900 includes a computing unit 901, which can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 902 or a computer program loaded from a storage unit 908 to a RAM (Random Access Memory) 903. Various programs and data required for the operation of the device 900 can also be stored in the RAM 903. The computing unit 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. An I/O (Input/Output) interface 905 is also connected to the bus 904.

A number of components in the device 900 are connected to the I/O interface 905, including: an input unit 906, such as a keyboard, a mouse, etc.; an output unit 907, such as various types of displays, speakers, etc.; a storage unit 908, such as a disk, an optical disk, etc.; and a communication unit 909, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 909 allows the device 900 to exchange information/data with other devices through a computer network such as the Internet and/or various telecommunication networks.

The computing unit 901 may be a variety of general and/or special processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, CPU (Central Processing Unit), GPU (Graphic Processing Units), various dedicated AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, etc. The computing unit 901 performs the various methods and processes described above, such as the prediction method of spatiotemporal distribution. For example, in some embodiments, the prediction method of spatiotemporal distribution may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 908. In some embodiments, part or all of the computer program may be loaded and/or installed on the device 900 via the ROM 902 and/or the communication unit 909. When the computer program is loaded into the RAM 903 and executed by the computing unit 901, one or more steps of the method described above may be performed. Alternatively, in other embodiments, the computing unit 901 may be configured to execute the aforementioned spatiotemporal distribution prediction method in any other appropriate manner (for example, by means of firmware).

Various embodiments of the systems and techniques described above herein may be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application Specific Standard Products), SOCs (System On Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: being implemented in one or more computer programs that may be executed and/or interpreted on a programmable system including at least one programmable processor that may be a dedicated or general-purpose programmable processor that may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, implements the functions/operations specified in the flow chart and/or block diagram. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or apparatus. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include an electrical connection based on one or more wires, a portable computer disk, a hard disk, RAM, ROM, EPROM (Electrically Programmable Read-Only-Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage device, magnetic storage device, or any suitable combination of the foregoing.

To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

The systems and techniques described herein may be implemented in a computing system that includes a backend component (e.g., as a data server), or a computing system that includes a middleware component (e.g., an application server), or a computing system that includes a frontend component (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: LAN (Local Area Network), WAN (Wide Area Network), the Internet, and blockchain networks.

A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

It should be noted that artificial intelligence is a discipline that studies how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), and includes both hardware-level and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, as well as machine learning/deep learning, big data processing technology, knowledge graph technology, and other major directions.

It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims (9)

1. A method for predicting spatiotemporal distribution, comprising: Filtering the collected original user data to obtain target user data, performing data construction processing on the target user data to obtain a user stay state vector sequence corresponding to the target user data, and performing data conversion processing on the target user data to obtain a user data vector corresponding to the target user data; Inputting the user stay state vector sequence and the user data vector into a pre-trained prediction model for a first inference process to obtain first hidden state data corresponding to the target user data, and performing a second inference process on the user stay state vector sequence and the user data vector based on the pre-trained prediction model to obtain second hidden state data corresponding to the target user data; Obtaining third hidden state data corresponding to the target user data according to the first hidden state data and the second hidden state data, and performing data conversion processing on the third hidden state data to obtain a corresponding distribution prediction result; The performing data construction processing on the target user data to obtain a user stay state vector sequence corresponding to the target user data includes: According to the spatial data information and the temporal data information carried in the target user data, the target user data is converted into data to obtain a user stay state vector; The user stay state vectors are arranged according to the time data information to obtain the user stay state vector sequence. 2. The method according to claim 1, wherein the step of performing data conversion processing on the target user data to obtain a user data vector corresponding to the target user data comprises: Performing data conversion processing on the geographic data in the target user data to obtain a geographic data vector; Performing data conversion processing on other data in the target user data by using a preset data conversion algorithm to obtain other data vectors, wherein the other data is all data in the target user data except the geographic data; The geographic data vector and the other data vectors are combined to obtain the user data vector. 3. The method according to claim 1, characterized in that the step of inputting the user stay state vector sequence and the user data vector into a pre-trained prediction model for first inference processing to obtain the first hidden state data corresponding to the target user data comprises: Activate the user stay state vector sequence and the user data vector through a first preset activation function to obtain first activation input data, where the first activation input data includes the time data information, and the time data information includes at least time period information for generating the user data; After data calculation processing is performed according to the first activation input data, the first historical hidden state data, and the first weight coefficient and the bias value, activation is performed through the first preset activation function to obtain first forget gate data, wherein the first historical hidden state data is hidden state data calculated according to user data generated in an adjacent time period before the time period information, and the first weight coefficient and the bias value are used to control the data forgetting degree of the first activation input data; After data calculation processing is performed according to the first activation input data, the first historical hidden state data, and the second weight coefficient and the bias value, activation is performed through the first preset activation function to obtain first input gate data, and the second weight coefficient and the bias value are used to control the new data acceptance degree of the first activation input data; After data calculation processing is performed according to the first activation input data, the first historical hidden state data, and the third weight coefficient and the bias value, activation is performed through a second preset activation function to obtain first candidate memory data, and the third weight coefficient and the bias value are used to perform data adjustment on the first activation input data; Performing data calculation processing according to the first forget gate data, the first input gate data, the first candidate memory data, and the first historical update memory data to obtain first update memory data, where the first historical update memory data is update memory data calculated according to user data generated in an adjacent time period before the time period information; After data calculation processing is performed according to the first activation input data, the first historical hidden state data, and a fourth weight coefficient and a bias value, activation is performed through the first preset activation function to obtain first output gate data, and the fourth weight coefficient and the bias value are used to control the degree of output state generation of the first activation input data; Data calculation and processing are performed according to the first updated memory data and the first output gate data to obtain the first hidden state data. 4. The method according to claim 1, characterized in that the performing a second inference process on the user stay state vector sequence and the user data vector based on the pre-trained prediction model to obtain the second hidden state data corresponding to the target user data comprises: Activate the user stay state vector sequence and the user data vector by a third preset activation function to obtain second activation input data, where the second activation input data includes the time data information, and the time data information includes at least time period information for generating the user data; After data calculation processing is performed according to the second activation input data, the second historical hidden state data, and the fifth weight coefficient and the bias value, activation is performed through the third preset activation function to obtain second forget gate data, wherein the second historical hidden state data is hidden state data calculated according to user data generated in an adjacent time period after the time period information, and the fifth weight coefficient and the bias value are used to control the data forgetting degree of the second activation input data; After data calculation processing is performed according to the second activation input data, the second historical hidden state data, and the sixth weight coefficient and the bias value, activation is performed through the third preset activation function to obtain second input gate data, and the sixth weight coefficient and the bias value are used to control the new data acceptance degree of the second activation input data; After data calculation processing is performed according to the second activation input data, the second historical hidden state data, and the seventh weight coefficient and the bias value, activation is performed through a fourth preset activation function to obtain second candidate memory data, and the seventh weight coefficient and the bias value are used to perform data adjustment on the second activation input data; Performing data calculation processing according to the second forget gate data, the second input gate data, the second candidate memory data, and the second historical update memory data to obtain second update memory data, where the second historical update memory data is update memory data calculated according to user data generated in an adjacent time period after the time period information; After data calculation processing is performed according to the second activation input data, the second historical hidden state data, and the eighth weight coefficient and the bias value, activation is performed through the third preset activation function to obtain second output gate data, and the eighth weight coefficient and the bias value are used to control the degree of output state generation of the second activation input data; Data calculation processing is performed according to the second updated memory data and the second output gate data to obtain the second hidden state data. 5. The method according to claim 1, characterized in that the step of performing data conversion processing on the third hidden state data to obtain a corresponding distribution prediction result comprises: Inputting the third hidden state data into a preset fully connected structure for full connection processing to obtain fully connected data; The fully connected data is subjected to data activation processing by using a fifth preset activation function to obtain probability distribution data, and the probability distribution data is used as the distribution prediction result. 6. A prediction device for spatiotemporal distribution, comprising: A filtering unit, used to filter the collected original user data to obtain target user data; a processing unit, configured to perform data construction processing on the target user data to obtain a user stay state vector sequence corresponding to the target user data, and to perform data conversion processing on the target user data to obtain a user data vector corresponding to the target user data; an input unit, configured to input the user stay state vector sequence and the user data vector into a pre-trained prediction model for a first inference process to obtain first hidden state data corresponding to the target user data, and to perform a second inference process on the user stay state vector sequence and the user data vector based on the pre-trained prediction model to obtain second hidden state data corresponding to the target user data; a conversion unit, configured to obtain third hidden state data corresponding to the target user data according to the first hidden state data and the second hidden state data, and perform data conversion processing on the third hidden state data to obtain a corresponding distribution prediction result; Wherein, the processing unit includes: A conversion module, configured to perform data conversion on the target user data according to the spatial data information and the temporal data information carried in the target user data, so as to obtain a user stay state vector; The arrangement module is used to arrange the user stay state vector according to the time data information to obtain the user stay state vector sequence. 7. An electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 5. 8. A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 5. 9. A computer program product, characterized in that it comprises a computer program, and when the computer program is executed by a processor, it implements the method according to any one of claims 1 to 5.