CN115620524A - A traffic jam prediction method, system, device and storage medium - Google Patents

Abstract

The invention discloses a traffic jam prediction method, a system, equipment and a storage medium, wherein the method obtains traffic event data, jam event data and station flow data by obtaining a multi-source heterogeneous data set; extracting the space correlation characteristics among the congestion event data, the traffic event data and the traffic congestion to be predicted by adopting a convolutional neural network; extracting time dependency relationship characteristics of the station traffic processing data and the congestion through a characteristic extraction network; acquiring various factor data except traffic event data, congestion event data and station flow data, and performing multi-classification processing and single-hot coding processing on the various factor data to obtain various factor characteristics; and splicing the spatial correlation characteristic, the time dependency relationship characteristic and the multiple factor characteristics to obtain a space-time joint characteristic, and inputting the space-time joint characteristic into the multilayer perceptron model to obtain a prediction result of the traffic jam to be predicted. The invention can improve the accuracy of traffic jam prediction.

Description

A traffic jam prediction method, system, device and storage medium

technical field

The present invention relates to the technical field of traffic jam prediction, in particular to a traffic jam prediction method, system, device and storage medium.

Background technique

The problem of traffic congestion has always been a matter of great concern to citizens, and traffic congestion prediction is also an important research field of intelligent transportation systems. Combined with big data, the traffic congestion prediction model can effectively predict future traffic conditions based on road conditions, site traffic, and historical congestion data, so as to guide citizens to travel, detour, and off-peak travel. Existing research methods mainly include methods based on statistics, traditional machine learning methods and methods based on deep learning models. The methods based on statistics are mainly designed for small data sets, which are not suitable for processing complex dynamic data and cannot capture features. The relationship between them; traditional machine learning methods require manual feature extraction and cannot extract complex spatiotemporal features.

In my country, expressway sections with an average speed lower than 30km/h are regarded as congested sections. The data source provided by an APP indicates that when the driving speed is lower than 30km/h, the congestion data will be uploaded to the cloud, while when the driving speed is higher than 30km/h h, there is no record of relevant congestion data. Congestion events are more difficult to predict than other traffic flows (such as high-speed site traffic, high-speed section speed, etc.) due to the characteristics of discontinuous data, difficult calculation of congestion length, and difficulty in distinguishing between congestion events.

Contents of the invention

The present invention aims to solve at least one of the technical problems existing in the prior art. Therefore, the present invention proposes a traffic congestion prediction method, system, device and storage medium, which can improve the accuracy of traffic congestion prediction.

In a first aspect, an embodiment of the present invention provides a method for predicting traffic congestion, the method for predicting traffic congestion includes:

Obtain a multi-source heterogeneous data set, and obtain traffic event data, congestion event data and site flow data from the multi-source heterogeneous data set;

Using a convolutional neural network to extract spatial correlation features between the congestion event data, the traffic event data and the traffic congestion to be predicted;

Extracting the time-dependent relationship features of the site traffic data and congestion through a feature extraction network; wherein, the feature extraction network is made by fusing a gated neural unit and an attention mechanism;

Obtain multiple factor data from the multi-source heterogeneous data set except traffic event data, congestion event data and site traffic data, and perform multi-classification processing and one-hot encoding processing on the multiple factor data to obtain multiple a factor characteristic;

Splicing the spatial correlation feature, the temporal dependence feature and the multiple factor features to obtain the joint spatio-temporal feature, and inputting the joint spatio-temporal feature into the multi-layer perceptron model to obtain the predicted Traffic congestion prediction results.

Compared with the prior art, the first aspect of the present invention has the following beneficial effects:

This method obtains multi-source heterogeneous data sets, traffic event data, congestion event data and site flow data from multi-source heterogeneous data sets. Predict traffic congestion, thereby improving the accuracy of traffic congestion prediction; use convolutional neural network to extract spatial correlation features between congestion event data, traffic event data and traffic congestion to be predicted, and extract site traffic data and The time-dependent relationship features of congestion; among them, the feature extraction network is made by the fusion of gated neural units and attention mechanisms, and obtains a variety of data in addition to traffic event data, congestion event data and site traffic data from multi-source heterogeneous data sets. Factor data, and perform binary classification processing and one-hot encoding processing on multiple factor data to obtain multiple factor features, and extract features from different data sets in different ways, and can capture the relationship between features, improving the feature Effectiveness of extraction; splicing spatial correlation features, time-dependent relationship features, and multiple factor features to obtain joint spatio-temporal features, and input the joint spatio-temporal features into the multi-layer perceptron model to obtain the prediction result of traffic congestion to be predicted , predicting traffic congestion comprehensively through multiple types of features can improve the accuracy of traffic congestion prediction.

According to some embodiments of the present invention, the acquisition of traffic event data, congestion event data and site traffic data from the multi-source heterogeneous data set includes:

performing one-hot encoding processing on the traffic events and congestion events in the multi-source heterogeneous data set, obtaining the traffic event data and the congestion event data, and normalizing the site traffic in the multi-source heterogeneous data set Processing, obtaining the site traffic data.

According to some embodiments of the present invention, the use of a convolutional neural network to extract spatial correlation features between the congestion event data, the traffic event data, and the traffic congestion to be predicted includes:

Preset the first historical time step, and acquire the historical data of the first quantity of the congestion event data and the historical data of the traffic event data that are geographically adjacent within the preset first historical time step;

Splicing the historical data of the traffic event data and the historical data of the congestion event data to obtain a spliced data sequence;

The concatenated data sequence is input into the convolutional neural network to obtain spatial correlation features between the congestion event data, the traffic event data, and the traffic congestion to be predicted.

According to some embodiments of the present invention, the splicing data sequence is input into the convolutional neural network to obtain the spatial correlation between the congestion event data, the traffic event data and the traffic congestion to be predicted features, including:

The spliced data sequence is input into the convolutional neural network, and the spliced data sequence is processed by a convolutional layer and a pooling layer:

和

表示卷积层的输出，E表示所述拼接数据序列，

和

表 示权重矩阵，

、

、

和

表示偏差矩阵，ReLU表示激活函数，

表示最大函数 值，

和

表示池化层的输出，

表示卷积运算；in,

and

Represents the output of the convolutional layer, E represents the spliced data sequence,

and

represents the weight matrix,

,

,

and

Represents the bias matrix, ReLU represents the activation function,

represents the maximum function value,

and

Represents the output of the pooling layer,

Indicates the convolution operation;

输入至全连接层，获 得所述空间相关性特征，所述空间相关性特征表示为： After the convolutional layer and the pooling layer process the spliced data sequence, the

Input to the fully connected layer to obtain the spatial correlation feature, the spatial correlation feature is expressed as:

表示在t时刻的所述拥堵事件数据、所述交通事件数据和所述待预测的 交通拥堵之间的空间相关性特征，

表示权重矩阵，

表示偏差矩阵。 in,

Representing the spatial correlation feature between the congestion event data, the traffic event data and the traffic congestion to be predicted at time t,

represents the weight matrix,

represents the bias matrix.

According to some embodiments of the present invention, the extraction of time-dependent relationship features between the site traffic data and congestion through the feature extraction network includes:

Preset a second historical time step, and obtain a second quantity of inbound site traffic data and outbound site traffic data of the top geographical locations within the second historical time step;

Splicing the inbound site traffic data and the outbound site traffic data to obtain spliced site traffic data;

表示为： Input the splicing site traffic data into the gated neural unit to obtain a first vector, the first vector output by the gated neural unit in step t

Expressed as:

表示第t-1步的拼接站点流量数据，

表示第t步的拼接站点流量数 据，GRU表示门控神经单元； in,

Indicates the splicing site traffic data of step t-1,

Represents the splicing site traffic data of step t, and GRU represents the gated neural unit;

The first vector is input into the attention mechanism to obtain a second vector, and the calculation formula of the attention mechanism is:

表示在t时刻由所述门控神经单元输出向量

的注意力分布值，

和

表示权重系数，

表示偏差系数，

表示在j时刻由所述门控神经单元输出向量

的注意 力分布值，

表示注意力权重，i表示总时间； in,

Indicates the output vector by the gated neuron unit at time t

The attention distribution value of

and

Indicates the weight coefficient,

Indicates the coefficient of deviation,

Indicates that the vector output by the gated neural unit at time j

The attention distribution value of

Represents the attention weight, i represents the total time;

The vector output by the attention mechanism is calculated through a fully connected layer to obtain the time-dependent relationship feature, and the fully connected layer calculation formula is:

表示在t时刻的时间依赖关系特征，

表示权重矩阵，

表示偏差向 量，ReLU表示激活函数。 in,

Indicates the time-dependent relationship feature at time t,

represents the weight matrix,

Represents the bias vector, and ReLU represents the activation function.

According to some embodiments of the present invention, the multi-category processing and one-hot encoding processing are performed on the multiple factor data to obtain multiple factor features, including:

If the multiple factor data in the multi-source heterogeneous data set is a binary variable, then the multiple factor data is expressed as a binary 0-1 variable after multi-classification processing, and the binary factor data is obtained, and passed One-hot encoding maps the binary factor data to multiple factor features;

If the multi-factor data in the multi-source heterogeneous data set is a multi-category variable, one-hot encoding is used to map the multi-factor data into multi-factor features.

According to some embodiments of the present invention, the spatial correlation feature, the temporal dependency feature and the multiple factor features are concatenated to obtain a spatio-temporal joint feature, and the spatio-temporal joint feature is input to the multi-layer perception In the computer model, the prediction result of the traffic jam to be predicted is obtained, including:

The joint feature of time and space is input into the multi-layer perceptron model, and the traffic congestion prediction result is obtained through the calculation of the hidden layer and the output layer, wherein the calculation of the hidden layer includes:

表示在t时刻的所述空间相关性特征，

表示在t时刻的时间依赖关系特 征，

表示在t时刻的所述多种因素特征，

表示拼接函数，

表示所述时空 联合特征，

表示所述隐藏层输出的特征向量，

表示权重矩阵，

表示偏差矩阵， ReLU表示激活函数，

表示卷积运算； in,

Represents the spatial correlation feature at time t,

Indicates the time-dependent relationship feature at time t,

Represents the characteristics of the various factors at time t,

represents the concatenation function,

represents the spatio-temporal joint feature,

represents the feature vector output by the hidden layer,

represents the weight matrix,

Represents the bias matrix, ReLU represents the activation function,

Indicates the convolution operation;

The feature vector output by the hidden layer is input to the output layer, and the calculation of the output layer includes:

表示t+1时刻的所述待预测的交通拥堵的预测结果，

表示权重矩 阵，

表示偏差矩阵，

表示激活函数。 in,

Indicates the prediction result of the traffic jam to be predicted at time t+1,

represents the weight matrix,

Denotes the bias matrix,

represents the activation function.

In the second aspect, the embodiment of the present invention also provides a traffic congestion prediction system, the traffic congestion prediction system comprising:

A data acquisition unit, configured to acquire a multi-source heterogeneous data set, and acquire traffic event data, congestion event data and site flow data from the multi-source heterogeneous data set;

The first feature extraction unit is used to extract the spatial correlation features between the congestion event data, the traffic event data and the traffic congestion to be predicted by using a convolutional neural network;

The second feature extraction unit is used to extract the time-dependent relationship features between the site traffic data and congestion through a feature extraction network; wherein, the feature extraction network is made by fusing a gated neural unit and an attention mechanism;

The third feature extraction unit is used to obtain multiple factor data except traffic event data, congestion event data and site flow data from the multi-source heterogeneous data set, and perform multi-classification processing on the multiple factor data and one-hot encoding processing to obtain multiple factor features;

The prediction result acquisition unit is used to splice the spatial correlation feature, the time dependency feature and the multiple factor features to obtain the joint spatio-temporal feature, and input the joint spatio-temporal feature into the multi-layer perceptron model , to obtain the prediction result of the traffic congestion to be predicted.

In the third aspect, the embodiment of the present invention also provides a traffic jam prediction device, which includes at least one control processor and a memory for communicating with the at least one control processor; Instructions executed by a control processor, said instructions being executed by said at least one control processor, to enable said at least one control processor to perform a traffic congestion prediction method as described above.

In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to make a computer perform one of the above-mentioned Traffic congestion prediction method.

It can be understood that the beneficial effects of the above-mentioned second aspect to the fourth aspect compared with the related technology are the same as those of the above-mentioned first aspect compared with the related technology. Please refer to the relevant description in the above-mentioned first aspect. This will not be repeated here.

Description of drawings

The above and/or additional aspects and advantages of the present invention will become apparent and comprehensible from the description of the embodiments in conjunction with the following drawings, wherein:

Fig. 1 is a flow chart of a kind of traffic jam prediction method of an embodiment of the present invention;

Fig. 2 is a structural diagram of a traffic jam prediction system according to an embodiment of the present invention.

detailed description

Embodiments of the present invention are described in detail below, examples of which are shown in the drawings, wherein the same or similar reference numerals designate the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the figures are exemplary only for explaining the present invention and should not be construed as limiting the present invention.

In the description of the present invention, if the first, second, etc. are described only for the purpose of distinguishing technical features, it cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features or implying Indicates the sequence of the indicated technical features.

In the description of the present invention, it should be understood that when it comes to orientation descriptions, for example, the orientation or positional relationship indicated by up, down, etc. is based on the orientation or positional relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description , rather than indicating or implying that the device or element referred to must have a particular orientation, be constructed and operate in a particular orientation, and thus should not be construed as limiting the invention.

In the description of the present invention, it should be noted that, unless otherwise clearly defined, words such as setting, installation, and connection should be understood in a broad sense, and those skilled in the art can reasonably determine that the above words are included in the present invention in combination with the specific content of the technical solution. specific meaning.

First, analyze some nouns involved in this application:

Decision tree: It is a non-parametric supervised learning algorithm for classification and regression tasks. It is a hierarchical tree structure consisting of root nodes, branches, internal nodes, and leaf nodes.

Extreme Gradient Boosting Tree: is an ensemble learning algorithm based on decision trees.

Random Forest: It is a classifier that contains multiple decision trees, and its output category is determined by the mode of the category output by individual trees.

K-nearest neighbor algorithm: is a non-parametric statistical method for classification and regression.

Long short-term memory network: a variant of recurrent neural network, which is widely used in deep learning models for time series forecasting.

The problem of traffic congestion has always been a matter of great concern to citizens, and traffic congestion prediction is also an important research field of intelligent transportation systems. Combined with big data, the traffic congestion prediction model can effectively predict future traffic conditions based on road conditions, site traffic, and historical congestion data, so as to guide citizens to travel, detour, and off-peak travel. Existing research methods mainly include methods based on statistics, traditional machine learning methods and methods based on deep learning models. The methods based on statistics are mainly designed for small data sets, which are not suitable for processing complex dynamic data and cannot capture features. The relationship between them; traditional machine learning methods require manual feature extraction and cannot extract complex spatiotemporal features.

In my country, expressway sections with an average speed lower than 30km/h are regarded as congested sections. The data source provided by an APP indicates that when the driving speed is lower than 30km/h, the congestion data will be uploaded to the cloud, while when the driving speed is higher than 30km/h h, there is no record of relevant congestion data. Congestion events are more difficult to predict than other traffic flows (such as high-speed site traffic, high-speed section speed, etc.) due to the characteristics of discontinuous data, difficult calculation of congestion length, and difficulty in distinguishing between congestion events.

In order to solve the above-mentioned problems, the present invention acquires traffic event data, congestion event data and site traffic data from multi-source heterogeneous data sets by acquiring multi-source heterogeneous data sets, and considers traffic congestion from various aspects by acquiring multiple data sets. Forecasting can comprehensively predict traffic congestion, thereby improving the accuracy of traffic congestion prediction; using convolutional neural network to extract the spatial correlation features between congestion event data, traffic event data and traffic congestion to be predicted, through feature extraction network Extract the time-dependent relationship features between site traffic data and congestion; among them, the feature extraction network is made by the fusion of gated neural units and attention mechanisms, and obtains traffic event data, congestion event data and site traffic data from multi-source heterogeneous data sets Multiple factor data other than multi-factor data, and multi-category processing and one-hot encoding processing for multiple factor data, to obtain multiple factor features, feature extraction for different data sets in different ways, and can capture the relationship between features relationship, which improves the effectiveness of feature extraction; splicing spatial correlation features, time-dependent relationship features, and multiple factor features to obtain joint spatio-temporal features, and input the joint spatio-temporal features into the multi-layer perceptron model to obtain predictive The prediction results of traffic congestion can comprehensively predict traffic congestion through various types of features, which can improve the accuracy of traffic congestion prediction.

With reference to Fig. 1, the embodiment of the present invention provides a kind of traffic congestion prediction method, this traffic congestion prediction method includes but not limited to step S100 to step S500:

Step S100, obtaining multi-source heterogeneous data sets, and obtaining traffic event data, congestion event data and site flow data from the multi-source heterogeneous data sets;

Step S200, using a convolutional neural network to extract spatial correlation features between congestion event data, traffic event data, and traffic congestion to be predicted;

Step S300, extracting time-dependent relationship features between site traffic data and congestion through a feature extraction network; wherein, the feature extraction network is made by fusing a gated neural unit and an attention mechanism;

Step S400, obtain multiple factor data from the multi-source heterogeneous data set except traffic event data, congestion event data and site traffic data, and perform multi-classification processing and one-hot encoding processing on the multiple factor data to obtain multiple factor characteristics;

Step S500, splicing spatial correlation features, time-dependent relationship features and multiple factor features to obtain joint spatio-temporal features, and input the joint spatio-temporal features into the multi-layer perceptron model to obtain the prediction result of traffic congestion to be predicted.

In some embodiments, from step S100 to step S500, in order to predict traffic congestion from various aspects, comprehensively predict traffic congestion, thereby improving the accuracy of traffic congestion prediction, by obtaining multi-source heterogeneous data sets, from multiple sources Collect traffic event data, congestion event data, and site traffic data from heterogeneous datasets; in order to capture the relationship between features and improve the effectiveness of feature extraction, convolutional neural networks are used to extract congestion event data, traffic event data, and to-be-predicted data. The spatial correlation features between the traffic congestion, and the time-dependent relationship between site traffic data and congestion are extracted through the feature extraction network; among them, the feature extraction network is made by the fusion of the gated neural unit and the attention mechanism, from the multi-source heterogeneous Collect multiple factor data except traffic event data, congestion event data and site traffic data in the data set, and perform multi-classification processing and one-hot encoding processing on multiple factor data to obtain multiple factor characteristics; in order to improve traffic congestion prediction Accuracy, by splicing spatial correlation features, time-dependent relationship features and multiple factor features to obtain joint spatio-temporal features, and input the joint spatio-temporal features into the multi-layer perceptron model to obtain the prediction result of traffic congestion to be predicted .

In some embodiments, traffic event data, congestion event data and site flow data are obtained from multi-source heterogeneous data sets, including:

Perform one-hot encoding processing on traffic events and congestion events in multi-source heterogeneous data sets, obtain traffic event data and congestion event data, and normalize site traffic in multi-source heterogeneous data sets to obtain site traffic data.

In this embodiment, different methods are used to extract multiple types of data from multi-source heterogeneous data sets, and different extraction methods are used for different types of data, so as to effectively extract data.

In some embodiments, a convolutional neural network is used to extract spatial correlation features between congestion event data, traffic event data, and traffic congestion to be predicted, including:

Preset the first historical time step, and obtain the historical data of the first quantity of congestion event data and the historical data of the traffic event data that are geographically adjacent within the preset first historical time step;

Splicing the historical data of the traffic event data and the historical data of the congestion event data to obtain the spliced data sequence;

The concatenated data sequence is input into the convolutional neural network to obtain the spatial correlation features between the congestion event data, the traffic event data and the traffic congestion to be predicted.

It should be noted that the first historical time step and the first quantity in this embodiment may be changed according to actual conditions, and are not specifically limited in this embodiment.

In this embodiment, the convolutional neural network can effectively extract the spatial correlation features between the congestion event data and the traffic event data, and can capture the relationship between features, thereby improving the effectiveness of feature extraction.

In some embodiments, the spliced data sequence is input into the convolutional neural network to obtain the spatial correlation features between the congestion event data, the traffic event data and the traffic congestion to be predicted, including:

Input the spliced data sequence into the convolutional neural network, and process the spliced data sequence through the convolutional layer and the pooling layer:

和

表示卷积层的输出，E表示拼接数据序列，

和

表示权 重矩阵，

、

、

和

表示偏差矩阵，ReLU表示激活函数，

表示最大函数值，

和

表示池化层的输出，

表示卷积运算； in,

and

Represents the output of the convolutional layer, E represents the spliced data sequence,

and

represents the weight matrix,

,

,

and

Represents the bias matrix, ReLU represents the activation function,

represents the maximum function value,

and

Represents the output of the pooling layer,

Indicates the convolution operation;

输入至全连接层，获得空间相关性 特征，空间相关性特征表示为： After the concatenated data sequence is processed by the convolutional and pooling layers, the

Input to the fully connected layer to obtain the spatial correlation feature, the spatial correlation feature is expressed as:

表示在t时刻的拥堵事件数据、交通事件数据和待预测的交通拥堵之间 的空间相关性特征，

表示权重矩阵，

表示偏差矩阵。 in,

Represents the spatial correlation feature between the congestion event data, traffic event data and traffic congestion to be predicted at time t,

represents the weight matrix,

represents the bias matrix.

In some embodiments, the feature extraction network is used to extract the time-dependent relationship between site traffic data and congestion, including:

Preset the second historical time step, and obtain the second quantity of inbound site traffic data and outbound site traffic data with the highest geographical location within the second historical time step;

Splice inbound site traffic data and outbound site traffic data to obtain spliced site traffic data;

表示为： Input the spliced site traffic data into the gated neural unit to obtain the first vector, the first vector output by the gated neural unit at step t

Expressed as:

表示第t-1步的拼接站点流量数据，

表示第t步的拼接站点流量数 据，GRU表示门控神经单元； in,

Indicates the splicing site traffic data of step t-1,

Represents the splicing site traffic data of step t, and GRU represents the gated neural unit;

Input the first vector into the attention mechanism to obtain the second vector. The calculation formula of the attention mechanism is:

表示在t时刻由门控神经单元输出向量

的注意力分布值，

和

表示 权重系数，

表示偏差系数，

表示在j时刻由门控神经单元输出向量

的注意力分布值，

表示注意力权重，i表示总时间； in,

Indicates that the vector output by the gated neural unit at time t

The attention distribution value of

and

Indicates the weight coefficient,

Indicates the coefficient of deviation,

Indicates that the vector output by the gated neural unit at time j

The attention distribution value of

Represents the attention weight, i represents the total time;

The vector output by the attention mechanism is calculated through the fully connected layer to obtain time-dependent relationship features. The calculation formula of the fully connected layer is:

表示在t时刻的时间依赖关系特征，

表示权重矩阵，

表示偏差向 量，ReLU表示激活函数。 in,

Indicates the time-dependent relationship feature at time t,

represents the weight matrix,

Represents the bias vector, and ReLU represents the activation function.

It should be noted that the second historical time step and the second number in this embodiment may be changed according to actual conditions, and are not specifically limited in this embodiment.

In this embodiment, the use of the gated neural unit and the attention mechanism can effectively extract the time-dependent relationship features between site traffic data and congestion, and can capture the relationship between features, thereby improving the effectiveness of feature extraction.

In some embodiments, multi-category processing and one-hot encoding processing are performed on the multi-factor data to obtain multi-factor features, including:

If the multi-factor data in the multi-source heterogeneous data set is a binary variable, the multi-factor data is expressed as a binary 0-1 variable after multi-classification processing, and the binary classification factor data is obtained, and the binary classification is processed by one-hot encoding. Factor data is mapped to multiple factor features;

If the multi-factor data in the multi-source heterogeneous data set is a multi-category variable, one-hot encoding is used to map the multi-factor data into multi-factor features.

In this embodiment, the binary classification processing and the one-hot encoding processing can effectively extract the features of multiple factors, and can capture the relationship between features, thereby improving the effectiveness of feature extraction.

In some embodiments, spatial correlation features, time-dependent relationship features, and multiple factor features are spliced to obtain joint spatio-temporal features, and the joint spatio-temporal features are input into the multi-layer perceptron model to obtain the prediction of traffic congestion to be predicted Results, including:

The spatio-temporal joint features are input into the multi-layer perceptron model, and the traffic congestion prediction results are obtained through the calculation of the hidden layer and the output layer. The calculation of the hidden layer includes:

表示在t时刻的空间相关性特征，

表示在t时刻的时间依赖关系特征，

表示在t时刻的多种因素特征，

表示拼接函数，

表示时空联合特征，

表示隐藏层输出的特征向量，

表示权重矩阵，

表示偏差矩阵，ReLU表示激活函 数，

表示卷积运算； in,

Indicates the spatial correlation feature at time t,

Indicates the time-dependent relationship feature at time t,

Represents the characteristics of various factors at time t,

represents the concatenation function,

Represents the spatiotemporal joint feature,

Represents the feature vector output by the hidden layer,

represents the weight matrix,

Represents the bias matrix, ReLU represents the activation function,

Indicates the convolution operation;

The feature vector output by the hidden layer is input to the output layer, and the calculation of the output layer includes:

表示t+1时刻的待预测的交通拥堵的预测结果，

表示权重矩阵，

表示偏差矩阵，

表示激活函数。 in,

Indicates the prediction result of the traffic congestion to be predicted at time t+1,

represents the weight matrix,

Denotes the bias matrix,

represents the activation function.

In this embodiment, after splicing spatial correlation features, time-dependent relationship features, and multiple factor features, the traffic congestion situation can be comprehensively predicted through multiple types of features, which can improve the accuracy of traffic congestion prediction.

For the convenience of those skilled in the art to understand, a group of best embodiments are provided below:

1. Processing of multi-source heterogeneous datasets.

In this embodiment, according to the characteristics of multi-source heterogeneous data, the data is divided into two types of data: category data and continuous data. For different types of data in multi-source heterogeneous data sets, different processing methods are adopted, for example:

This embodiment combines the congestion event data returned by AutoNavi Maps (vehicle speed is lower than 30km/h), and combines the congestion event data with the stake number, so that the congestion event can be accurately located on the expressway. In this embodiment, the road section n is When congestion occurs within the time period [t, t+t1] (t1 takes 0.5h, that is, 30min), the congestion event data is recorded as 1, if the road section n is in the time period [t, t+t1] (t1 takes 0.5h, that is If there is no congestion event information within 30 minutes), the congestion event data will be recorded as 0. After the congestion event is processed, it is processed into the congestion event data of 0-1 category variable.

，其中M表示站点的总数，得到268个收 费站点的进出站的站点流量数据。其中站点流量的统计间隔为0.5小时，每个站点流量数据 均包含从2019年1月1日至2019年12月31日一年的收费站进出站的站点流量数据，流量单位 为辆每小时（即，辆/小时）。针对站点流量数据，本实施例采用归一化层对站点流量数据进 行处理，归一化层的目的为使用梯度下降的方法求解最优化问题时，归一化后可以加快梯 度下降的求解速度，即提升模型的收敛速度。站点流量经过归一化处理之后，每个样本集包 含17520条数据，按照6:2:2划分为模型的训练集、验证集和测试集。流量数据是连续变量就 是某个时间步有多少辆车，经过归一化处理之后变成了0到1里面的一个浮点数。 The data set of this example contains the entry and exit records of the toll stations of a provincial expressway network system. By cleaning and preprocessing the data set, the inbound and outbound traffic of all toll stations at each moment is counted and obtained. In this embodiment, the inbound and outbound traffic of site m at time t is expressed as

, where M represents the total number of stations, and the inbound and outbound traffic data of 268 toll stations are obtained. Among them, the statistical interval of station traffic is 0.5 hours, and each station traffic data includes the station traffic data of the toll gate in and out of the station from January 1, 2019 to December 31, 2019, and the flow unit is vehicle per hour ( i.e. vehicles/hour). For the site traffic data, this embodiment uses the normalization layer to process the site traffic data. The purpose of the normalization layer is to use the gradient descent method to solve the optimization problem. After normalization, the solution speed of the gradient descent can be accelerated. That is to improve the convergence speed of the model. After the site traffic is normalized, each sample set contains 17520 pieces of data, which are divided into the training set, verification set and test set of the model according to 6:2:2. Traffic data is a continuous variable, that is, how many vehicles there are at a certain time step, and after normalization, it becomes a floating point number between 0 and 1.

2. Spatial correlation feature extraction module.

Due to the complex factors affecting expressway network congestion, the congestion incidents in the expressway network are not only affected by the upstream sections of the expressway network, but also by the downstream sections of the expressway network. The highway network structure is intricate and complex, and the spatial influence is difficult to extract. Therefore, this embodiment captures the spatial correlation features of expressway network congestion through a convolutional neural network. Specifically:

，其中，

表示相邻的i号路段第 j个历史时间步的交通事件情况；构建地理位置最近的相邻N个路段的拥堵事件数据的历史 数据为

，其中，

表示相邻的i号路段第j个历史时间步 的拥堵情况。将交通事件数据的历史数据C与拥堵事件数据的历史数据D进行拼接操作之 后，得到高速路网相邻路段的交通事件与拥堵事件序列，用

进 行表示，将E输入到卷积神经网络中。其中，交通事件数据包括2019年全省各市区、各高速公 路的全量交通事件信息，包括事件类型(如车流量大、交通管制、突发交通事故、道路施工 等)、持续时间、起点高速公路桩号、终点高速公路桩号、发生时间数据。在本实施例中，第一 历史时间步长T取6，地理位置最近的路段数N取8。 The first historical time step used in this embodiment is T, and the historical data of the traffic event data of the nearest adjacent N road sections to construct the geographic location is

,in,

Represents the traffic event situation of the jth historical time step of the adjacent road section i; the historical data of the congestion event data of the nearest N adjacent road sections constructed is

,in,

Indicates the congestion situation of the jth historical time step of the adjacent road section i. After splicing the historical data C of the traffic event data and the historical data D of the congestion event data, the sequence of traffic events and congestion events on the adjacent sections of the expressway network is obtained.

To represent, input E into the convolutional neural network. Among them, the traffic event data includes the full amount of traffic event information in all urban areas and expressways of the province in 2019, including event types (such as heavy traffic, traffic control, sudden traffic accidents, road construction, etc.), duration, and starting point expressways Chainage, terminal expressway pile number, occurrence time data. In this embodiment, the first historical time step T is set to 6, and the number N of road sections closest to the geographic location is set to 8.

In this embodiment, a convolutional neural network framework consisting of two convolutional layers and two pooling layers is constructed. According to the characteristics of the highway network, both convolutional layers are designed as two-dimensional convolutions, and the pooling The method is selected as maximum pooling, and the activation function of the convolutional layer is selected as the ReLU activation function. The processing of convolutional layer and pooling layer can be expressed as follows:

。

.

输入到全连接层，以获得空间相关性特征，全连接层采用激活函数ReLU。卷 积神经网络在t时刻输出的空间相关性特征可表示为： After being processed by the convolutional layer and the pooling layer, the historical traffic events and congestion events are mapped into the hidden layer feature space, and then

Input to the fully connected layer to obtain spatial correlation features, and the fully connected layer uses the activation function ReLU. The spatial correlation features output by the convolutional neural network at time t can be expressed as:

。

.

The spatial correlation between the congestion events of adjacent road sections and the traffic congestion to be predicted can be captured by the convolutional neural network, and the relationship between the traffic events of the adjacent road sections and the traffic congestion to be predicted can be captured by the convolutional neural network. , so that the spatial correlation characteristics between the traffic congestion event data and the traffic congestion to be predicted can be obtained, and the spatial correlation characteristics between the traffic event data and the traffic congestion to be predicted can be obtained.

3. Time dependency feature extraction module.

Since the high-speed site traffic itself has a strong periodicity and time dependence, and the adjacent high-speed site traffic has a strong non-linear time correlation with the congestion of the high-speed section, this embodiment captures The temporal periodicity of site traffic data itself and the non-linear time correlation between capture congestion and site traffic data. Specifically:

，其中，

为站点m长度 为T的入站站点流量数据的历史数据，构建地理位置最近的排名靠前的M个高速站点的出站 站点流量数据为

，其中，

为站点m长度为T 的出站站点流量数据的历史数据。将入站站点流量数据的历史数据

和出站站点流量数 据的历史数据

进行拼接操作之后，得到拼接站点流量数据为

，将拼接站点流量数据输入到门控神经单元中，对需提取的特 征进行充分学习，以捕获时间依赖关系。在本实施例中，第二历史时间步长T取6，地理位置 最近的高速站点数M取10。门控神经单元的输出为第一向量，门控神经单元的输出在第t步 输出的第一向量

可以表示为： The second historical time step used in this embodiment is T, and the inbound site traffic data of the top M high-speed sites with the closest geographic location are constructed as

,in,

is the historical data of the inbound site traffic data of site m with a length of T, and construct the outbound site traffic data of the top M high-speed sites with the closest geographic location as

,in,

is the historical data of the outbound site traffic data of site m with length T. Historical data for inbound site traffic data

and historical data for outbound site traffic data

After the splicing operation, the spliced site traffic data is obtained as

, input the spliced site traffic data into the gated neural unit, and fully learn the features to be extracted to capture the temporal dependence. In this embodiment, the second historical time step T is 6, and the number M of the geographically closest express stations is 10. The output of the gated neuron unit is the first vector, and the output of the gated neuron unit is the first vector output at step t

It can be expressed as:

。

.

Subsequently, the first vector after the activation of the gated neural unit is input into the attention mechanism for summarization, and the weights corresponding to different feature vectors are calculated through weight distribution, and a better parameter matrix is continuously calculated and iterated. The calculation of the attention mechanism The way can be expressed as follows:

。

.

The output of the attention mechanism is calculated through a fully-connected layer to obtain time-dependent relationship features. The activation function of the fully-connected layer is ReLU, and the time-dependent relationship features output at time t are:

。

.

4. Multiple factor feature extraction module.

In this embodiment, an embedding layer and a fully connected layer are designed to extract data of various factors that affect expressway network congestion, such as time, holidays, and weather. For example:

For the traffic at each moment of the charging station, extract the hour of the day, the day of the week, whether it is a weekend, whether it is a holiday, whether the previous day is a holiday, and whether the next day is a holiday. And use the one-hot encoding mechanism to convert the time features of the corresponding moment into the embedding vector Other, including the hour of the day (24 features), the day of the week (7 features), whether it is a holiday (2 feature), whether the previous day is a holiday (2 features), whether the next day is a holiday (2 features), and the historical traffic event data of the current road segment (2 features), etc. In this embodiment, for the data of binary classification variables (such as whether it is a holiday or not), this embodiment expresses the data as binary classification 0-1 variables through multi-classification processing, obtains binary classification factor data, and uses one-hot encoding to convert the two Categorical factor data is mapped to a variety of factor features; for data of multi-category categorical variables, this embodiment uses a one-hot encoding method to map the data into multiple 0-1 binary features (that is, multiple factor features) to ensure The distance between different categories is the same, which facilitates better extraction of the relationship between features. In this embodiment, the historical time step is set to 6 (the past 3 hours), that is, the road congestion situation of a single time step in the future is predicted through 6 historical time steps (3 hours).

The processing of weather factors is as follows: first, one-hot encoding processing, and then input into the embedding layer together with other time factors for embedding operation, and then input the output of the embedding layer into the fully connected layer to obtain multiple factor features. The weather data in this embodiment includes weather data for the whole year of 2019, detailed to each urban area and every half hour, including weather conditions (such as cloudy, sunny, cloudy, light rain, sleet, etc.), temperature , wind force and direction data.

5. Predict traffic jams.

Input the spatio-temporal joint feature into the multi-layer perceptron model, and obtain the prediction result of the traffic congestion to be predicted through the calculation of the hidden layer and the output layer, wherein the calculation of the hidden layer includes:

表示在t时刻的多种因素特征。 in,

Represents the characteristics of various factors at time t.

The feature vector output by the hidden layer is input to the output layer, and the calculation of the output layer includes:

。

.

For better illustration, this embodiment has carried out following experiment:

1. Performance indicators

This example uses F1 score (F1 Score), positive sample classification precision rate (P) and positive sample classification recall rate (R) as the evaluation indicators of prediction model performance. These three indicators are widely used in the classification of samples with extremely uneven distribution. Condition. Among them, TP (True Positive) means true positive, FP (False Positive) means false positive, and FN (False Negative) means false negative. The calculation of the performance index is as follows:

。

.

2. Comparative experiment.

In order to evaluate the performance of the model, this embodiment divides the processed multi-source heterogeneous data set into a training set, a verification set and a test set according to a certain ratio. The training set is used for model training, and the verification set and test set are used for model evaluation. , this embodiment evaluates the effectiveness of the model by comparing it with other baseline models. This embodiment uses the baseline prediction methods such as decision tree, extreme gradient boosting tree, random forest, K-nearest neighbor algorithm, long short-term memory network and the technical solution of this embodiment to compare and evaluate the overall performance.

By comparison and evaluation, the prediction effect of the technical solution of this embodiment is the best, and the results are shown in Table 1, and three conclusions can be drawn:

(1) The technical solution of this embodiment is significantly superior to other methods in all indicators, especially in the indicators of F1 score, precision rate and recall rate, which have a very important guiding role for citizens' travel;

(2) Since the deep learning network can fully learn the nonlinear relationship between features, this type of model has a greater advantage in modeling data related to the expressway network;

(3) Since the baseline model cannot learn the spatial correlation of the expressway network, the technical solution of this embodiment can learn the spatial correlation through the convolutional neural network, so it has better performance in prediction performance.

Table 1

3. Ablation Experiment

In order to evaluate the effectiveness of different components in the model, this embodiment is verified by ablation experiments, that is, the performance of the model is evaluated by reducing some design components in the model, and the ablation experiments can reflect the effectiveness of various components in the model.

Through ablation experiments, the effectiveness of different components in the technical solution of this embodiment was evaluated, and the results are shown in Table 2, which shows three indicators, namely F1 score, precision rate and recall rate. The results show that deleting any component will affect the performance of the technical solution of this embodiment. Among them, when the spatial correlation feature extraction module is deleted, the F1 score, precision rate, and recall rate drop from 0.683, 0.685, and 0.683 to 0.488, 0.502, and 0.475 respectively; when the time-dependent feature extraction module is deleted, the F1 score, precision rate, and recall rate The rate decreased from 0.683, 0.685, 0.683 to 0.605, 0.623, 0.589; when the multi-factor feature extraction module was deleted, the F1 score, precision rate and recall rate decreased from 0.683, 0.685, 0.683 to 0.637, 0.633, 0.642, respectively.

This shows that: 1) Congestion events on the expressway network are significantly affected by the road network structure and the traffic conditions of adjacent road sections. 2) The impact of the spatial correlation of expressway network congestion is greater than the impact of traffic station flow; 3) Due to the spatial correlation of expressway network and the influence of expressway station flow than time factors (such as the first day of a week How many days, whether today is a holiday or not) is more directly affected, so when the first two components are ablated, the prediction accuracy and recall rate both drop greatly.

Table 2

With reference to Fig. 2, the embodiment of the present invention also provides a kind of traffic jam prediction system, this traffic jam prediction system comprises data acquisition unit 100, first feature extraction unit 200, second feature extraction unit 300, the 3rd feature extraction unit 400 and Prediction result acquisition unit 500, wherein:

The data acquisition unit 100 is configured to acquire multi-source heterogeneous data sets, and acquire traffic event data, congestion event data and site traffic data from the multi-source heterogeneous data sets;

The first feature extraction unit 200 is used to extract spatial correlation features between congestion event data, traffic event data and traffic congestion to be predicted by using a convolutional neural network;

The second feature extraction unit 300 is used to extract the time-dependent relationship features between site traffic data and congestion through a feature extraction network; wherein, the feature extraction network is made by fusing a gated neural unit and an attention mechanism;

The third feature extraction unit 400 is used to obtain multiple factor data other than traffic event data, congestion event data and site traffic data from a multi-source heterogeneous data set, and perform multi-classification processing and unique heat on the multiple factor data Coding processing to obtain multiple factor characteristics;

The prediction result acquisition unit 500 is used to splicing spatial correlation features, time-dependent relationship features, and multiple factor features to obtain joint spatio-temporal features, and input the joint spatio-temporal features into the multi-layer perceptron model to obtain traffic congestion to be predicted prediction results.

It should be noted that, since a traffic congestion prediction system in this embodiment is based on the same inventive concept as the above-mentioned traffic congestion prediction method, the corresponding content in the method embodiment is also applicable to the system embodiment, here will not be described in detail.

The embodiment of the present invention also provides a traffic jam prediction device, including: at least one control processor and a memory for communicating with the at least one control processor.

As a non-transitory computer-readable storage medium, memory can be used to store non-transitory software programs and non-transitory computer-executable programs. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory optionally includes memory located remotely from the processor, and these remote memories may be connected to the processor via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The non-transitory software programs and instructions required to realize the traffic jam prediction method of the above-mentioned embodiment are stored in the memory, and when executed by the processor, a kind of traffic jam prediction method in the above-mentioned embodiment is executed, for example, the above Steps S100 to S500 of the method in FIG. 1 are described.

The system embodiments described above are only illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

The embodiment of the present invention also provides a computer-readable storage medium, the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are executed by one or more control processors, so that the above-mentioned one or more The control processor executes a traffic congestion prediction method in the above method embodiment, for example, executes the functions from step S100 to step S500 of the method in FIG. 1 described above.

Those skilled in the art can understand that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware and an appropriate combination thereof. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit . Such software may be distributed on computer readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media). As known to those of ordinary skill in the art, the term computer storage media includes both volatile and nonvolatile media implemented in any method or technology for storage of information, such as computer readable instructions, data structures, program modules, or other data. permanent, removable and non-removable media. Computer storage media including, but not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cartridges, tape, magnetic disk storage or other magnetic storage devices, or Any other medium used to store desired information and which can be accessed by a computer. In addition, as is well known to those of ordinary skill in the art, communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media .

The embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned embodiments. Within the scope of knowledge of those of ordinary skill in the art, various modifications can be made without departing from the spirit of the present invention. Variety.

Claims (10)

1. a traffic jam prediction method, is characterized in that, described traffic jam prediction method comprises: Obtain a multi-source heterogeneous data set, and obtain traffic event data, congestion event data and site flow data from the multi-source heterogeneous data set; Using a convolutional neural network to extract spatial correlation features between the congestion event data, the traffic event data and the traffic congestion to be predicted; Extracting the time-dependent relationship features of the site traffic data and congestion through a feature extraction network; wherein, the feature extraction network is made by fusing a gated neural unit and an attention mechanism; Obtain multiple factor data from the multi-source heterogeneous data set except traffic event data, congestion event data and site traffic data, and perform multi-classification processing and one-hot encoding processing on the multiple factor data to obtain multiple a factor feature; Splicing the spatial correlation feature, the temporal dependence feature and the multiple factor features to obtain the joint spatio-temporal feature, and inputting the joint spatio-temporal feature into the multi-layer perceptron model to obtain the predicted Traffic congestion prediction results. 2. traffic congestion prediction method according to claim 1, is characterized in that, described acquisition traffic event data, congestion event data and site traffic data from described multi-source heterogeneous data set, comprises: performing one-hot encoding processing on the traffic events and congestion events in the multi-source heterogeneous data set, obtaining the traffic event data and the congestion event data, and normalizing the site traffic in the multi-source heterogeneous data set Processing, obtaining the site traffic data. 3. traffic jam prediction method according to claim 1, is characterized in that, described employing convolutional neural network extracts the spatial correlation feature between described traffic jam event data, described traffic event data and the traffic jam to be predicted ,include: Preset the first historical time step, and acquire the historical data of the first quantity of the congestion event data and the historical data of the traffic event data that are geographically adjacent within the preset first historical time step; Splicing the historical data of the traffic event data and the historical data of the congestion event data to obtain a spliced data sequence; The concatenated data sequence is input into the convolutional neural network to obtain spatial correlation features between the congestion event data, the traffic event data, and the traffic congestion to be predicted. 4. traffic jam prediction method according to claim 3, is characterized in that, described splicing data sequence is input in described convolutional neural network, obtains described congestion event data, described traffic event data and waiting time. Spatial correlation features between predicted traffic jams, including: The spliced data sequence is input into the convolutional neural network, and the spliced data sequence is processed by a convolutional layer and a pooling layer:

in,

and

Represents the output of the convolutional layer, E represents the spliced data sequence,

and

represents the weight matrix,

,

,

and

Represents the bias matrix, ReLU represents the activation function,

represents the maximum function value,

and

Represents the output of the pooling layer,

Indicates the convolution operation; After the convolutional layer and the pooling layer process the spliced data sequence, the

Input to the fully connected layer to obtain the spatial correlation feature, the spatial correlation feature is expressed as:

in,

Representing the spatial correlation feature between the congestion event data, the traffic event data and the traffic congestion to be predicted at time t,

represents the weight matrix,

represents the bias matrix. 5. traffic jam prediction method according to claim 1, is characterized in that, described site traffic data and the time-dependent relationship feature of congestion are extracted by feature extraction network, comprising: Preset a second historical time step, and obtain a second quantity of inbound site traffic data and outbound site traffic data of the top geographical locations within the second historical time step; Splicing the inbound site traffic data and the outbound site traffic data to obtain spliced site traffic data; Input the splicing site traffic data into the gated neural unit to obtain a first vector, the first vector output by the gated neural unit in step t

Expressed as:

in,

Indicates the splicing site traffic data of step t-1,

Represents the splicing site traffic data of step t, GRU represents the gated neural unit; The first vector is input into the attention mechanism to obtain a second vector, and the calculation formula of the attention mechanism is:

in,

Indicates the output vector by the gated neuron unit at time t

The attention distribution value of

and

Indicates the weight coefficient,

Indicates the coefficient of deviation,

Indicates that the vector output by the gated neural unit at time j

The attention distribution value of

Represents the attention weight, i represents the total time; The vector output by the attention mechanism is calculated through a fully connected layer to obtain the time-dependent relationship feature, and the fully connected layer calculation formula is:

in,

Indicates the time-dependent relationship feature at time t,

represents the weight matrix,

Represents the bias vector, and ReLU represents the activation function. 6. traffic congestion prediction method according to claim 1, is characterized in that, described multiple factor data is carried out multi-category processing and one-hot encoding process, obtains multiple factor features, comprising: If the multiple factor data in the multi-source heterogeneous data set is a binary variable, then the multiple factor data is expressed as a binary 0-1 variable after multi-classification processing, and the binary factor data is obtained, and passed One-hot encoding maps the binary factor data to multiple factor features; If the multi-factor data in the multi-source heterogeneous data set is a multi-category variable, one-hot encoding is used to map the multi-factor data into multi-factor features. 7. traffic congestion prediction method according to claim 1, is characterized in that, described spatial correlation feature, described temporal dependence feature and described multiple factor feature are spliced, obtain space-time joint feature, and The joint feature of time and space is input into the multi-layer perceptron model to obtain the prediction result of the traffic jam to be predicted, including: The joint feature of time and space is input into the multi-layer perceptron model, and the traffic congestion prediction result is obtained through the calculation of the hidden layer and the output layer, wherein the calculation of the hidden layer includes:

in,

Represents the spatial correlation feature at time t,

Indicates the time-dependent relationship feature at time t,

Represents the characteristics of the various factors at time t,

represents the concatenation function,

represents the spatio-temporal joint feature,

represents the feature vector output by the hidden layer,

represents the weight matrix,

Represents the bias matrix, ReLU represents the activation function,

Indicates the convolution operation; The feature vector output by the hidden layer is input to the output layer, and the calculation of the output layer includes:

in,

Indicates the prediction result of the traffic jam to be predicted at time t+1,

represents the weight matrix,

Denotes the bias matrix,

represents the activation function. 8. A traffic jam forecasting system, characterized in that, the traffic jam forecasting system comprises: A data acquisition unit, configured to acquire a multi-source heterogeneous data set, and acquire traffic event data, congestion event data and site flow data from the multi-source heterogeneous data set; The first feature extraction unit is used to extract the spatial correlation features between the congestion event data, the traffic event data and the traffic congestion to be predicted by using a convolutional neural network; The second feature extraction unit is used to extract the time-dependent relationship features between the site traffic data and congestion through a feature extraction network; wherein, the feature extraction network is made by fusing a gated neural unit and an attention mechanism; The third feature extraction unit is used to obtain multiple factor data except traffic event data, congestion event data and site flow data from the multi-source heterogeneous data set, and perform multi-classification processing on the multiple factor data and one-hot encoding processing to obtain multiple factor features; The prediction result acquisition unit is used to splice the spatial correlation feature, the time dependency feature and the multiple factor features to obtain the joint spatio-temporal feature, and input the joint spatio-temporal feature into the multi-layer perceptron model , to obtain the prediction result of the traffic congestion to be predicted. 9. A traffic jam prediction device, characterized in that it comprises at least one control processor and a memory for communication connection with the at least one control processor; the memory stores information that can be executed by the at least one control processor instructions, the instructions are executed by the at least one control processor, so that the at least one control processor can execute the traffic congestion prediction method according to any one of claims 1 to 7. 10. A computer-readable storage medium, characterized in that, the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to make a computer perform the operation described in any one of claims 1 to 7. traffic congestion prediction method.

Images