CN112434872A - Hotel yield prediction method, system, equipment and storage medium - Google Patents

Abstract

The present invention provides a method, system, equipment and storage medium for predicting hotel output. The method includes the steps of: acquiring output data of multiple hotels, where the output data includes historical output data, hotel attribute feature data and current reservation progress data. ; build a quantile loss function based on the quantile regression model; build an initial deep learning network model; based on the quantile loss function and the output data of the multiple hotels, train the initial deep learning network model, respectively obtaining A target network model corresponding to each hotel; based on the target network model, the output of multiple hotels in a preset time period in the future is respectively predicted, and the output value of each hotel in the preset time period is output; The output of the hotel is forecast at the same time, and the accuracy of the hotel output forecast is improved.

Description

Hotel yield forecasting method, system, equipment and storage medium

technical field

The invention relates to the technical field of artificial intelligence, in particular to a method, system, equipment and storage medium for predicting hotel yield.

Background technique

In the prior art, a traditional time series model or a neural network model LSTM (Long Short-Term Memory, artificial neural network for long short-term memory) is generally used to predict the output of a hotel (ie, the sales volume of resources such as hotel rooms). However, the traditional time series model is difficult to comprehensively consider the influence of multiple exogenous variables on the final yield, and its expressive ability is limited. In the process of hotel output forecasting, in addition to time series features, it also involves exogenous variables such as reservation schedule features, hotel competitive circle features, hotel attribute features, and date features. Therefore, the prediction accuracy of the traditional time series model will be significantly worse at this time. However, the neural network model LSTM has limited expressive ability in the scenario of multi-hotel prediction, and cannot be used in the scenario of multi-hotel prediction at the same time.

SUMMARY OF THE INVENTION

In view of the problems in the prior art, the purpose of the present invention is to provide a method, system, equipment and storage medium for predicting the output of hotels, which can not only realize the simultaneous prediction of the output of multiple hotels, but also improve the accuracy of the hotel output prediction.

To achieve the above object, the present invention provides a method for predicting the output of a hotel, the method comprising the following steps:

Obtain production data for multiple hotels;

Construct quantile loss function based on quantile regression model;

Build the initial deep learning network model;

Based on the quantile loss function and the output data of the multiple hotels, the initial deep learning network model is trained, and the target network models corresponding to each hotel are obtained respectively;

Based on the target network model, the output of a plurality of hotels in a preset time period in the future is respectively predicted, and the output value of each hotel in the preset time period is output.

Optionally, the steps of training the initial deep learning network model based on the quantile loss function and the output data of the multiple hotels, respectively obtaining target network models corresponding to each hotel include:

Carrying out a vectorization operation on the output data of each hotel to obtain a plurality of feature vectors belonging to a hotel;

Perform low-order feature crossover and feature high-order crossover on the feature vector, respectively, to obtain a first predicted value and a second predicted value;

Weighted summation is performed on the first predicted value and the second predicted value to obtain an initial predicted result, and a normalization operation is performed on the initial predicted result using a preset activation function to obtain a predicted probability value;

The predicted probability value and the preset real result value are used as the input of the quantile loss function to obtain the target network model.

Optionally, the initial deep learning network model includes a DNN model, and the step of performing low-order feature crossover and feature high-order crossover on the feature vector, respectively, to obtain the first predicted value and the second predicted value includes: :

Choose two eigenvectors from all the eigenvectors to combine to form a plurality of eigenvector sets, and each of the eigenvector sets contains two eigenvectors;

Perform an inner product operation on all the feature vector sets respectively to obtain a first predicted value;

Using all the feature vectors as the input of the DNN model, a second predicted value is obtained.

Optionally, the target network model includes quantiles, and the quantiles in the target network models corresponding to different hotels are different.

Optionally, the steps of training the initial deep learning network model based on the quantile loss function and the output data of the multiple hotels, respectively obtaining target network models corresponding to each hotel include:

For each hotel, a plurality of preset quantiles are used as the input of the quantile loss function respectively, and a plurality of different second network models belonging to the same hotel are generated by training, and each of the second network models corresponds to 1. the preset quantile;

Based on the output data of the past N days as the test set, the multiple different second network models are tested respectively, and the second network model with the smallest difference between the obtained predicted value and the actual output value is used as the test set. The target network model corresponding to the hotel; the production data of the past N days includes the real production value.

Optionally, the steps of training the initial deep learning network model based on the quantile loss function and the output data of the multiple hotels, respectively obtaining target network models corresponding to each hotel include:

According to the features contained in the output data of each hotel, a feature ID corresponding to each feature is generated;

Generate a feature dictionary file according to the feature and the feature ID;

Store the feature dictionary file in the target network model.

Optionally, each piece of the output data includes a feature and a feature value corresponding to the feature; the initial deep learning network model has a plurality of parameters; each of the features matches one of the parameters; and each of the feature matching—the feature vector;

The step of performing inner product operations on all the feature vector sets respectively to obtain the first predicted value includes: performing inner product operations on each feature vector set respectively according to a low-order cross function;

The low-order cross function is:

in,

y ₁ represents the first predicted value, w ₀ represents the first parameter of the initial deep learning network model, w _i represents the ith parameter of the initial deep learning network model, m represents each piece of the output data The total number of features, x _i represents the eigenvalue corresponding to the ith feature in each piece of the yield data, <v _i , v _j > represents the inner product of the feature vectors v _i and v _j , v _i represents the yield of each item The eigenvector corresponding to the ith feature in the data, v _j represents the eigenvector corresponding to the jth feature in each piece of the output data, x _j represents the eigenvalue corresponding to the jth feature in each piece of the output data, v _{i, f} represent the f-th element in the feature vector v _i , v _{j, f} represent the f-th element in the feature vector v _j , and k represents the total number of elements in the feature vector v _i .

Optionally, the quantile loss function is:

Wherein, n represents the total quantity of the yield data, q represents the preset quantile, y′ _p represents the predicted value of the yield data of the pth item, _yp represents the actual yield value of the yield data of the pth item, L represents the total loss value.

Optionally, the method further includes the steps:

updating the production data of the plurality of hotels based on the hotel production data of the current period;

Based on the updated yield data, the initial deep learning network model is retrained to obtain the updated target network model and the updated quantile.

Optionally, the output data includes historical output data, hotel attribute feature data, and current reservation progress data; the hotel attribute feature data includes hotel star rating, room type level, and data on the city where the hotel is located, and the output data also includes historical reservations. Progress data, current cancellation progress, competitive circle characteristics of the hotel, and date characteristics.

The present invention also provides a hotel yield forecasting system for realizing the above hotel yield forecasting method, the system comprising:

The output data acquisition module is used to obtain the output data of multiple hotels, and the output data includes historical output data, hotel attribute feature data and current reservation progress data;

A loss function building module, which builds a quantile loss function based on a quantile regression model;

A model building module for building an initial deep learning network model;

The model training module, based on the quantile loss function and the output data of the multiple hotels, trains the initial deep learning network model, and obtains the target network model corresponding to each hotel respectively;

The output prediction module, based on the target network model, respectively predicts the output of a plurality of hotels within a preset time period in the future, and outputs the output value of each hotel within the preset time period.

The present invention also provides a hotel yield forecasting device, comprising:

processor;

a memory in which executable instructions for the processor are stored;

Wherein, the processor is configured to execute the steps of any one of the above-mentioned hotel yield forecasting methods by executing the executable instructions.

The present invention also provides a computer-readable storage medium for storing a program, when the program is executed by a processor, the steps of any one of the above-mentioned hotel yield prediction methods are implemented.

Compared with the prior art, the present invention has the following advantages and outstanding effects:

The hotel output forecasting method, system, equipment and storage medium provided by the invention can comprehensively consider the influence of multiple exogenous variables on the final output, improve the hotel output forecasting accuracy; The hotels used different quantile generation models, and under the premise of achieving simultaneous forecasting of multiple hotels, the accuracy of the output forecast for each hotel during parallel forecasting was guaranteed.

Description of drawings

Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments with reference to the following drawings.

1 is a schematic diagram of a method for predicting hotel output disclosed in an embodiment of the present invention;

FIG. 2 is a schematic flowchart of step S40 in an embodiment of the present invention;

FIG. 3 is a schematic flowchart of step S402 in an embodiment of the present invention;

FIG. 4 is a schematic flowchart of step S404 in an embodiment of the present invention;

FIG. 5 is a schematic flowchart of step S40 in another embodiment of the present invention;

6 is a schematic diagram of a method for predicting hotel output disclosed by another embodiment of the present invention;

7 is a schematic structural diagram of a hotel yield prediction system disclosed in an embodiment of the present invention;

8 is a schematic structural diagram of a hotel yield forecasting device disclosed in an embodiment of the present invention;

FIG. 9 is a schematic structural diagram of a computer-readable storage medium disclosed in an embodiment of the present invention.

Detailed ways

Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be embodied in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of example embodiments to those skilled in the art. The same reference numerals in the drawings denote the same or similar structures, and thus their repeated descriptions will be omitted.

As shown in FIG. 1 , an embodiment of the present invention discloses a method for predicting hotel output, and the method includes the following steps:

S10: Obtain output data of multiple hotels, where the output data may include historical output data, hotel attribute feature data, and current reservation progress data. The above-mentioned hotel attribute characteristic data may include hotel star rating, room type level, and data of the city where the hotel is located, and the above-mentioned output data may also include historical reservation progress data, current cancellation progress, characteristics of the competition circle and date characteristics of the hotel. This application does not limit the above-mentioned production data and hotel attribute feature data, for example, the above-mentioned production data may also include hotel review features.

S20, constructing a quantile loss function based on the quantile regression model. Among them, each quantile loss function includes a corresponding preset quantile, each hotel corresponds to a preset quantile, and the preset quantiles in the quantile loss functions of different hotels may be different, or can be the same. Those skilled in the art can make settings as required.

In this embodiment, the above-mentioned quantile loss function is:

Among them, n represents the total number of the above-mentioned production data, that is, the total number of production data. q represents a preset quantile, such as 0.45, 0.5 or 0.6. y' _p represents the predicted value of the p-th production data, and y _p represents the actual production value of the p-th production data. The yield data contains the actual yield value corresponding to this piece of data. L represents the total loss.

作为第p条产量数据的损失值计算总损失值。如果预测值小于真实产量值，也即y′_p<y_p，则执行

作为第p条产量数据的损失值计算总损失值。∑ _p:y′p≥yp (y′ _p -y _p ) means that for the p-th production data, if the predicted value is greater than or equal to the actual production value, that is, y′ _p ≥ y _p , execute

The total loss value is calculated as the loss value for the p-th yield data. If the predicted value is less than the true yield value, i.e. y′ _p <y _p , then execute

The total loss value is calculated as the loss value for the p-th yield data.

S30, constructing an initial deep learning network model. In this embodiment, the initial deep learning network model includes a DNN (Deep Neural Networks, deep neural network) model. The initial deep learning network model may be a DeepFM deep learning model. This application does not limit this. The above DeepFM model includes two parts: FM (factor Machine, factorization machine) and DNN model.

S40, based on the quantile loss function and the output data of the multiple hotels, train the initial deep learning network model, and obtain target network models corresponding to each hotel respectively. Specifically, as shown in Figure 2, step S40 includes:

S401, performing a vectorization operation on the above-mentioned output data of each hotel to obtain a plurality of feature vectors belonging to a hotel. In this embodiment, each piece of the above-mentioned production data includes a feature and a feature value corresponding to the above-mentioned feature. For example, if the output data includes hotel star data, each piece of hotel star data includes a star feature and a feature value corresponding to the feature.

The above initial deep learning network model has multiple parameters. Each feature matches one of the above parameters. And each feature matches one of the above feature vectors. Performing the vectorization operation on the output data is to perform the above-mentioned vectorization operation on the features in the output data to obtain feature vectors corresponding to the features. Among them, the above-mentioned vectorization operation can be realized by using the embedding technology.

S402: Perform low-order feature crossover and feature high-order crossover on all feature vectors, respectively, to obtain a first predicted value and a second predicted value, respectively. Specifically, as shown in Figure 3, step S402 includes:

S4021 , combining any two feature vectors from all the aforementioned feature vectors to form multiple feature vector sets, each of the aforementioned feature vector sets includes two feature vectors.

S4022: Perform an inner product operation on the feature vectors in each feature vector set to obtain a first predicted value. Specifically, in this step, an inner product operation is respectively performed on the feature vectors in each feature vector set according to the low-order cross function.

The above low-order crossover function is:

in,

y ₁ represents the above-mentioned first predicted value. w ₀ represents the first parameter of the above initial deep learning network model. w _i represents the i-th parameter of the above initial deep learning network model. m represents the total number of features for each piece of the above yield data. _xi represents the eigenvalue corresponding to the ith feature in each of the above-mentioned production data. <v _i ,v _j > represents the inner product of the feature vectors v _i and v _j . v _i and v _j represent two eigenvectors in a set of eigenvectors.

v _i represents the feature vector corresponding to the i-th feature in each of the above-mentioned production data. v _j represents the feature vector corresponding to the j-th feature in each of the above-mentioned production data. x _j represents the eigenvalue corresponding to the j-th feature in each of the above-mentioned production data, vi _{, f} represent the f-th element in the feature vector v _i , v _{j, f} represent the f-th element in the feature vector v _j , _k represents the total number of elements in the feature vector vi.

And S4023, all the above-mentioned feature vectors are used as the input of the above-mentioned DNN model to obtain a second predicted value. For the calculation process of the DNN model, the existing technology can be used, and details are not repeated here.

S403: Perform a weighted summation on the first predicted value and the second predicted value to obtain an initial prediction result, and perform a normalization operation on the initial prediction result using a preset activation function to obtain a predicted probability value. Wherein, the weight in the above-mentioned weighted summation can be set as required, which is not limited in this application. The above-mentioned preset activation function may use a softmax activation function.

And S404 , using the predicted probability value and the preset real result value as the input of the quantile loss function to obtain a target network model. Specifically, the above quantile loss function can be used to calculate the loss to minimize the loss; at this time, the corresponding value of each parameter in the initial deep learning network model can be obtained, and the feature parameter is set to this value, that is, the target network can be obtained. Model.

In this embodiment, the target network model includes a quantile, and the quantile is one of the preset quantiles. The quantiles in the target network model corresponding to different hotels are different. In this way, reasonable quantiles can be set according to the attribute characteristics and other data of different hotels, and the prediction accuracy can be improved, so as to improve the prediction accuracy of each hotel.

and S50 , predicting the output of a plurality of hotels in a preset time period in the future based on the target network model, and outputting the output value of each hotel in the preset time period. For example, the above-mentioned preset time period may be 6 days, that is, the output value of each hotel in the next 6 days is predicted respectively. In this way, the parallel forecasting capability for multiple hotels can be realized, which can save the time required to forecast the output of multiple hotels, and improve the efficiency of the forecasting system for the output forecasting in the multi-hotel scenario.

In another embodiment of the present application, on the basis of the above-mentioned embodiment, as shown in FIG. 4 , step S404 includes:

S4041 , for each hotel, use a plurality of preset quantiles as the input of the above-mentioned quantile loss function, and train to generate a plurality of different second network models belonging to the same hotel. Each of the second network models corresponds to the predetermined quantile.

S4042, based on the above-mentioned production data of the past N days as the test set, test the above-mentioned multiple different second network models respectively, and use the above-mentioned second network model with the smallest difference between the obtained predicted value and the actual production value as the above-mentioned hotel The corresponding target network model. The above-mentioned production data for the above-mentioned past N days includes the above-mentioned actual production value.

In this way, for each hotel, multiple preset second network models can be obtained by using multiple preset quantiles, so as to prevent the model from only selecting one preset quantile to generate a model and cause the model to converge too unilaterally, it can avoid abnormal data values to the model. It has a significant impact, improves the fault tolerance rate of the model, and makes the generated model more accurate. Moreover, using the most recent production data for testing can ensure the accuracy of the obtained model parameters, which can make the final prediction result more accurate.

In another embodiment of the present application, on the basis of the above-mentioned embodiment, as shown in FIG. 5 , in step S40, step S402 further includes: according to the characteristics contained in the above-mentioned production data of each hotel, generating each hotel The feature ID corresponding to the feature.

Step S403 further includes: generating a feature dictionary file according to the aforementioned feature and the aforementioned feature ID.

Step S404 further includes: storing the above-mentioned feature dictionary file in the above-mentioned target network model.

This can avoid the problem that the prior art must first generate a dictionary file according to all the features in the development, and the dictionary file cannot be updated in time if new features are generated in the future, resulting in an increase in the development cost. The present application does not need to develop the features separately. The mapping relationship reduces engineering development costs.

In another embodiment of the present application, on the basis of the above-mentioned embodiment, as shown in FIG. 6 , the above-mentioned method for predicting the output of the hotel may further include the steps:

S60: Update the output data of the above-mentioned multiple hotels based on the hotel output data of the current period.

S70, based on the updated output data, train the above-mentioned initial deep learning network model again to obtain the updated target network model and the above-mentioned updated quantile.

In this way, the most suitable quantile can be found and updated according to the latest prediction effect, and the production database can be enriched, which can better describe the distribution of data, which is conducive to ensuring the timeliness of historical production data and improving the next production. accuracy in forecasting.

It should be noted that all the above-mentioned embodiments of the present application can be freely combined, and the technical solutions obtained after the combination are also within the protection scope of the present application.

As shown in FIG. 7 , an embodiment of the present invention also discloses a hotel yield prediction system 7, which includes:

The output data acquisition module 71 is used to obtain the output data of a plurality of hotels, and the above-mentioned output data includes historical output data, hotel attribute characteristic data and current reservation progress data;

a loss function construction module 72, which constructs a quantile loss function based on the quantile regression model;

Model building module 73, used to build an initial deep learning network model;

The model training module 74, based on the above-mentioned quantile loss function and the output data of the above-mentioned multiple hotels, trains the above-mentioned initial deep learning network model, and obtains the target network model corresponding to each hotel respectively;

The output forecasting module 75 respectively predicts the output of a plurality of hotels within a preset time period in the future based on the target network model, and outputs the output value of each hotel within the preset time period.

It can be understood that the hotel production forecasting system of the present invention also includes other existing functional modules that support the operation of the hotel production forecasting system. The hotel yield forecasting system shown in FIG. 7 is only an example, and should not impose any limitations on the functions and scope of use of the embodiments of the present invention.

The hotel output forecasting system in this embodiment is used to implement the above-mentioned method for hotel output forecasting. Therefore, for the specific implementation steps of the hotel output forecasting system, reference may be made to the above description of the method for hotel output forecasting, which will not be repeated here.

The embodiment of the present invention also discloses a hotel yield prediction device, including a processor and a memory, wherein the memory stores executable instructions of the processor; the processor is configured to execute the above method for hotel yield prediction by executing the executable instructions. A step of. FIG. 8 is a schematic structural diagram of the hotel yield forecasting device disclosed in the present invention. The electronic device 600 according to this embodiment of the present invention is described below with reference to FIG. 8 . The electronic device 600 shown in FIG. 8 is only an example, and should not impose any limitation on the function and scope of use of the embodiments of the present invention.

As shown in FIG. 8, electronic device 600 takes the form of a general-purpose computing device. Components of the electronic device 600 may include, but are not limited to, at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different platform components (including the storage unit 620 and the processing unit 610), a display unit 640, and the like.

Wherein, the storage unit stores program codes, which can be executed by the processing unit 610, so that the processing unit 610 executes the steps according to various exemplary embodiments of the present invention described in the above section of the hotel yield prediction method of this specification. For example, the processing unit 610 may perform the steps shown in FIG. 1 .

The storage unit 620 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 6201 and/or a cache storage unit 6202 , and may further include a read only storage unit (ROM) 6203 .

The storage unit 620 may also include a program/utility 6204 having a set (at least one) of program modules 6205 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, An implementation of a network environment may be included in each or some combination of these examples.

The bus 630 may be representative of one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local area using any of a variety of bus structures bus.

The electronic device 600 may also communicate with one or more external devices 700 (eg, keyboards, pointing devices, Bluetooth devices, etc.), with one or more devices that enable a user to interact with the electronic device 600, and/or with Any device (eg, router, modem, etc.) that enables the electronic device 600 to communicate with one or more other computing devices. Such communication may occur through input/output (I/O) interface 650 . Also, the electronic device 600 may communicate with one or more networks (eg, a local area network (LAN), a wide area network (WAN), and/or a public network such as the Internet) through a network adapter 660 . Network adapter 660 may communicate with other modules of electronic device 600 through bus 630 . It should be appreciated that, although not shown, other hardware and/or software modules may be used in conjunction with electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives And data backup storage platform, etc.

The invention also discloses a computer-readable storage medium for storing a program, when the program is executed, the steps in the above-mentioned hotel yield prediction method are implemented. In some possible implementations, various aspects of the present invention can also be implemented in the form of a program product, which includes program code, when the program product runs on a terminal device, the program code is used to cause the terminal device to execute the above-mentioned description in this specification. The steps according to various exemplary embodiments of the present invention are described in the method for forecasting hotel yield.

As shown above, when the program of the computer-readable storage medium of this embodiment is executed, the influence of multiple exogenous variables on the final output can be comprehensively considered, thereby improving the prediction accuracy of hotel output; and it has the parallel prediction ability for multiple hotels. , it can use different quantile generation models according to different hotels, and under the premise of realizing multi-hotel simultaneous prediction, it ensures the accuracy of output prediction for each hotel during parallel prediction.

FIG. 9 is a schematic structural diagram of a computer-readable storage medium of the present invention. Referring to FIG. 9, a program product 800 for implementing the above method according to an embodiment of the present invention is described, which can adopt a portable compact disk read only memory (CD-ROM) and include program codes, and can be used in a terminal device, For example running on a personal computer. However, the program product of the present invention is not limited thereto, and in this document, a readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or a combination of any of the above. More specific examples (non-exhaustive list) of readable storage media include: electrical connections with one or more wires, portable disks, hard disks, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM or flash memory), optical fiber, portable compact disk read only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

A computer-readable storage medium may include a data signal propagated in baseband or as part of a carrier wave with readable program code embodied thereon. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A readable storage medium can also be any readable medium other than a readable storage medium that can transmit, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. Program code embodied on a readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Program code for carrying out operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages—such as Java, C++, etc., as well as conventional procedural programming Language - such as the "C" language or similar programming language. The program code may execute entirely on the user computing device, partly on the user device, as a stand-alone software package, partly on the user computing device and partly on a remote computing device, or entirely on the remote computing device or server execute on. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any kind of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (eg, using an Internet service provider business via an Internet connection).

The hotel output forecasting method, system, device and storage medium provided by the embodiments of the present invention can comprehensively consider the influence of multiple exogenous variables on the final output, thereby improving the accuracy of hotel output forecasting; Handle cold start scenarios well;

On the other hand, the present application has the capability of parallel forecasting for multiple hotels, and can use different quantile generation models according to different hotels. Under the premise of realizing simultaneous forecasting of multiple hotels, the output of each hotel during parallel forecasting is guaranteed. prediction accuracy.

The above content is a further detailed description of the present invention in combination with specific preferred embodiments, and it cannot be considered that the specific implementation of the present invention is limited to these descriptions. For those of ordinary skill in the technical field of the present invention, without departing from the concept of the present invention, some simple deductions or substitutions can be made, which should be regarded as belonging to the protection scope of the present invention.

Claims (13)

1. a hotel yield forecasting method, is characterized in that, comprises the following steps: Obtain production data for multiple hotels; Construct quantile loss function based on quantile regression model; Build the initial deep learning network model; Based on the quantile loss function and the output data of the multiple hotels, the initial deep learning network model is trained, and the target network models corresponding to each hotel are obtained respectively; Based on the target network model, the output of a plurality of hotels in a preset time period in the future is respectively predicted, and the output value of each hotel in the preset time period is output. 2. The method for predicting hotel output according to claim 1, wherein the initial deep learning network model is trained based on the quantile loss function and the output data of the multiple hotels, and each The steps of the target network model corresponding to the hotel include: Carrying out a vectorization operation on the output data of each hotel to obtain a plurality of feature vectors belonging to a hotel; Perform low-order feature crossover and feature high-order crossover on the feature vector, respectively, to obtain a first predicted value and a second predicted value; Weighted summation is performed on the first predicted value and the second predicted value to obtain an initial predicted result, and a normalization operation is performed on the initial predicted result using a preset activation function to obtain a predicted probability value; The predicted probability value and the preset real result value are used as the input of the quantile loss function to obtain the target network model. 3. hotel output forecasting method as claimed in claim 2, is characterized in that, in described initial deep learning network model, comprises DNN model, described feature vector is carried out respectively low-order cross of feature and high-order cross of feature, The steps of obtaining the first predicted value and the second predicted value respectively include: Choose two eigenvectors from all the eigenvectors to combine to form a plurality of eigenvector sets, and each of the eigenvector sets contains two eigenvectors; Perform an inner product operation on all the feature vector sets respectively to obtain a first predicted value; Using all the feature vectors as the input of the DNN model, a second predicted value is obtained. 4 . The method for predicting hotel output according to claim 1 , wherein the target network model includes quantiles, and the target network models corresponding to different hotels have different quantiles. 5 . 5. The method for predicting hotel output according to claim 1, wherein the initial deep learning network model is trained based on the quantile loss function and the output data of the multiple hotels, and each The steps of the target network model corresponding to the hotel include: For each hotel, a plurality of preset quantiles are used as the input of the quantile loss function respectively, and a plurality of different second network models belonging to the same hotel are generated by training, and each of the second network models corresponds to 1. the preset quantile; Based on the output data of the past N days as the test set, the multiple different second network models are tested respectively, and the second network model with the smallest difference between the obtained predicted value and the actual output value is used as the test set. The target network model corresponding to the hotel; the production data of the past N days includes the real production value. 6. The hotel yield prediction method according to claim 1, wherein the initial deep learning network model is trained based on the quantile loss function and the yield data of the multiple hotels, and each The steps of the target network model corresponding to the hotel include: According to the features contained in the output data of each hotel, a feature ID corresponding to each feature is generated; Generate a feature dictionary file according to the feature and the feature ID; Store the feature dictionary file in the target network model. 7. The method for predicting hotel yield as claimed in claim 3, wherein each piece of said yield data comprises a feature and a feature value corresponding to said feature; said initial deep learning network model has a plurality of parameters; each the feature matches one of the parameters; and each of the features matches one of the feature vectors; The step of performing inner product operations on all the feature vector sets respectively to obtain the first predicted value includes: performing inner product operations on each feature vector set respectively according to a low-order cross function; The low-order cross function is:

in,

y ₁ represents the first predicted value, w ₀ represents the first parameter of the initial deep learning network model, w _i represents the ith parameter of the initial deep learning network model, m represents each piece of the output data The total number of features, x _i represents the eigenvalue corresponding to the ith feature in each piece of the yield data, <v _i , v _j > represents the inner product of the feature vectors v _i and v _j , v _i represents the yield of each item The eigenvector corresponding to the ith feature in the data, v _j represents the eigenvector corresponding to the jth feature in each piece of the output data, x _j represents the eigenvalue corresponding to the jth feature in each piece of the output data, v _{i, f} represent the f-th element in the feature vector v _i , v _{j, f} represent the f-th element in the feature vector v _j , and k represents the total number of elements in the feature vector v _i . 8. hotel yield forecasting method as claimed in claim 1, is characterized in that, described quantile loss function is:

Wherein, n represents the total quantity of the yield data, q represents the preset quantile, y _p ^′ represents the predicted value of the yield data of the p-th item, y _p represents the actual yield value of the yield data of the p-th item, L represents the total loss value. 9. hotel yield forecasting method as claimed in claim 4, is characterized in that, described method also comprises the step: updating the production data of the plurality of hotels based on the hotel production data of the current period; Based on the updated yield data, the initial deep learning network model is retrained to obtain the updated target network model and the updated quantile. 10. The method for predicting hotel output according to claim 1, wherein the output data includes historical output data, hotel attribute feature data and current reservation progress data; the hotel attribute feature data includes hotel star rating, room type rating and data of the city where the hotel is located, the output data also includes historical booking progress data, current cancellation progress, characteristics of the competitive circle where the hotel is located, and date characteristics. 11. A hotel output forecasting system for realizing the hotel output forecasting method as claimed in claim 1, wherein the system comprises: The output data acquisition module is used to obtain the output data of multiple hotels, and the output data includes historical output data, hotel attribute feature data and current reservation progress data; A loss function building module, which builds a quantile loss function based on a quantile regression model; A model building module for building an initial deep learning network model; The model training module, based on the quantile loss function and the output data of the multiple hotels, trains the initial deep learning network model, and obtains the target network model corresponding to each hotel respectively; The output prediction module, based on the target network model, respectively predicts the output of a plurality of hotels within a preset time period in the future, and outputs the output value of each hotel within the preset time period. 12. A hotel yield forecasting device, characterized in that it comprises: processor; a memory in which executable instructions for the processor are stored; Wherein, the processor is configured to perform the steps of the method for predicting hotel yield according to any one of claims 1 to 10 by executing the executable instructions. 13. A computer-readable storage medium for storing a program, wherein when the program is executed by a processor, the program implements the steps of the method for predicting hotel yield according to any one of claims 1 to 10.

Images