Highway toll station exit traffic flow prediction method and system based on multi-source data fusion

Abstract

The present invention discloses a method and system for predicting the traffic flow at the exit of a highway toll station through multi-source data fusion. It acquires highway toll data and vehicle license plate recognition data, extracts the traffic flow characteristics of each single-source data, unifies the spatio-temporal dimensions of each single-source according to the spatio-temporal characteristics of the toll station exit to be predicted, constructs a prediction model based on toll data and a prediction model based on vehicle license plate recognition data, uses the outputs of the two single-source prediction models as the inputs of the multi-source data fusion prediction model, and obtains the predicted traffic flow based on the multi-source data fusion prediction model. The present invention combines toll data and vehicle license plate recognition data for multi-faceted detection, utilizes the complementarity and redundancy between multi-source data, avoids the limitations of a small detection range and few data features that may exist based on a single data source, and improves the accuracy of traffic prediction.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation, and particularly to a method for predicting the traffic flow at the exit of a highway toll station by fusing multi-source data, and also relates to a system for predicting the traffic flow at the exit of a highway toll station by fusing multi-source data. Background Art

[0002] Highways have closely connected cities, bringing more convenient travel conditions for travelers. People's demands for highway services and traffic guidance services are increasing continuously, and traffic management departments have put forward higher requirements for traffic control and the alleviation of traffic conditions. Traffic flow prediction is a process of analyzing traffic conditions on roads, mining traffic patterns, and predicting road traffic trends.

[0003] Toll stations are bottleneck nodes in the highway traffic network and are prone to traffic queuing problems. Timely traffic flow prediction can help traffic management departments make decisions in advance. Traffic management departments can implement preemptive countermeasures according to the prediction results, such as dispatching more toll collectors, etc., to cope with peak hours, maximizing the improvement of traffic conditions and reducing traffic delays. Making predictions on traffic flow through the collected data can effectively alleviate the congestion problem at the exit of the toll station, provide convenient and efficient travel services for travelers, and establish a good image of the highway.

[0004] Most of the existing traffic flow predictions at the exit of highway toll stations are based on a single data source for prediction. Single data may have limitations such as a small detection range and few data features. In highway big data, multi-source data has the characteristics of complete data types and a large amount of data. By utilizing the complementarity and redundancy between multi-source data, the accuracy of traffic prediction can be effectively improved through multi-source data fusion. Summary of the Invention

[0005] The object of the present invention is to provide a method and system for predicting the traffic flow at the exit of a highway toll station by fusing multi-source data. By fusing toll data and vehicle license plate recognition data to establish a prediction model, the traffic flow at the exit of the highway toll station is predicted, thereby improving the accuracy of traffic flow prediction and alleviating traffic congestion.

[0006] To achieve the above functions, the present invention designs a method for predicting the traffic flow at the exit of a highway toll station by fusing multi-source data, including the following steps S1 - S5:

[0007] Step S1: Obtain highway toll data and vehicle license plate recognition data, where the highway toll data includes data of each exit toll station and each entrance toll station on the highway;

[0008] Step S2: Extract single-source data flow features for the highway toll data and the vehicle license plate recognition data respectively;

[0009] Step S3: According to the spatio-temporal characteristics of the toll station to be predicted at the exit, unify the spatio-temporal dimensions of the traffic flow characteristics of each single-source data, and obtain the data of the entrance toll station with consistent spatio-temporal dimensions and the vehicle license plate recognition data with consistent spatio-temporal dimensions;

[0010] Step S4: Respectively construct a prediction model based on toll data and a prediction model based on vehicle license plate recognition data. The prediction model based on toll data takes the data of the entrance toll station with consistent spatio-temporal dimensions obtained in Step S3 as input and the predicted traffic flow of the exit toll station as output. The prediction model based on vehicle license plate recognition data takes the vehicle license plate recognition data with consistent spatio-temporal dimensions obtained in Step S3 as input and the predicted traffic flow of the exit toll station as output;

[0011] Step S5: For the predicted traffic flow of the exit toll station output by the prediction model based on toll data and the prediction model based on vehicle license plate recognition data, perform fusion based on a neural network to obtain a multi-source prediction result of the fused traffic flow of the exit toll station.

[0012] As a preferred technical solution of the present invention: The method for extracting the traffic flow characteristics of single-source data in Step S2 is as follows: Aggregate the data obtained in Step S1 according to a time dimension of 5 minutes. There are s pieces of data in total, and the prediction time series is The traffic flow sequence of the toll station to be predicted at the exit is

[0013] As a preferred technical solution of the present invention: The method for spatio-temporal consistency of vehicle license plate recognition data in Step S3 is as follows:

[0014] Screen the vehicle license plate recognition data according to the principle of matching time and space characteristics, and convert it into vehicle license plate recognition data with the same time period and the same interval by taking the same road section as a constraint;

[0015] The method for spatio-temporal consistency of the data of the entrance toll station is as follows:

[0016] Select a certain exit toll station as the target station D. According to the entrance toll station where the vehicle enters and the exit toll station where the vehicle exits, extract all the entrance toll station sequences {O1, O2,..., O n};

[0017] Calculate the vehicle travel time within the OD pair. The calculation formula is: Vehicle travel time = Exit time - Entrance time;

[0018] Calculate the average value of the vehicle travel time within the OD pair. The calculation formula is:

[0019]

[0020] In the formula, is the entrance toll station Oj The average travel time of all vehicles within time period k is O j The vehicle travel time of the m-th vehicle on section D within time period k, where M is the total number of vehicles;

[0021] Select twice the standard deviation of the sample mean as the threshold range to eliminate data outside the range, where:

[0022]

[0023] Calculate the sample mean Obtain the vehicle time surplus for the OD pair

[0024] Construct the association matrix between the target site and each entrance toll station. For a specific target site, at the time series when the entrance toll stations are {O1, O2, …, O n}}, the association matrix is:

[0025]

[0026] Indicates the traffic volume when the entrance toll station is O j , the target site is D, and the time point is ;

[0027] Select a certain time point in the time series to be predicted for the target site D whose fluctuation range is and represent the start time and end time of this time point;

[0028] The corresponding entrance toll station considering the time surplus The time fluctuation range is d j is the vehicle time surplus for the OD pair; Calculate the correlation coefficient between the target site D and the associated site as follows:

[0029]

[0030] In the formula, is the mean value of the traffic volumes of each entrance toll station, is the traffic volume of the target site D at the time point ; is the average traffic flow of the target site D at each time point; set the correlation threshold to 0.8, and filter out the entrance toll stations with correlation coefficients greater than the correlation threshold to complete the spatio-temporal consistency processing of the entrance toll station data.

[0031] As a preferred technical solution of the present invention: the prediction model based on toll data described in step S4 adopts a GCN model, and the GCN model is expressed as:

[0032]

[0033] X is the feature matrix, E is the identity matrix, D is the degree matrix, A is the adjacency matrix, W represents the weight, b represents the bias in the training process, Relu() is the activation function, and f(X,A) represents the graph convolution process;

[0034] The calculation process of each parameter in the TGCN model is as follows, f(A,X t ) represents the graph convolution process, h t-1 is the output at time t-1:

[0035] u t =σ(W u [f(A,X t ),h t-1 +b u )

[0036] r t =σ(W r [f(A,X t ),h t-1 +b r )

[0037] c t =tanh(W c [f(A,X t ),(r t *h t-1 )]+b c )

[0038] h t =u t *h t-1 +(1-u t )*c t

[0039] The output of the prediction model based on toll data is the predicted traffic flow of the exit toll station, and the output form is a traffic flow sequence aggregated according to the 5-minute time dimension;

[0040] The prediction model based on vehicle license plate recognition data uses an LSTM model, and the optimal hyperparameters of the LSTM model are obtained through a random search algorithm. The optimal hyperparameters of the model include the hidden layer size and the learning rate. The search range of the hidden layer size is [32, 512], and the search range of the learning rate is [0.0001, 0.5]. The output of the model is the predicted traffic flow of the exit toll station, in the form of a traffic flow sequence of the exit toll station according to a 5-minute time dimension.

[0041] As a preferred technical solution of the present invention: in step S5, a BP neural network is used to fuse the predicted traffic flow of the exit toll station output by the prediction model based on toll data and the prediction model based on vehicle license plate recognition data. The structure of the BP neural network is as follows:

[0042] j = m + n + a

[0043] In the formula: j, m, and n are the numbers of neurons in the hidden layer, input layer, and output layer respectively, and a is a constant.

[0044] The present invention also designs a multi-source data fusion-based highway toll station exit traffic flow prediction system, which includes a multi-source detector module, a feature extraction module, a spatio-temporal consistency module, a single-source data prediction module, and a multi-source data fusion prediction module to implement the multi-source data fusion-based highway toll station exit traffic flow prediction method:

[0045] The multi-source detector module obtains highway toll data and vehicle license plate recognition data; the feature extraction module extracts the traffic flow characteristics of each single-source data; the spatio-temporal consistency module unifies the spatio-temporal dimensions of each single source; the single-source data prediction module constructs a prediction model based on toll data and a prediction model based on vehicle license plate recognition data; the outputs of the two single-source prediction models are used as the inputs of the multi-source data fusion prediction module, and the predicted traffic flow of the exit toll station is obtained based on the multi-source data fusion prediction module.

[0046] Beneficial effects: Compared with the prior art, the advantages of the present invention include:

[0047] Compared with the highway toll station exit traffic flow prediction method based on a single data source, the method provided by the present invention utilizes the complementarity and redundancy between multi-source data to make up for the limitations of a small detection range and few data features existing in the single data source. In highway big data, multi-source data has the characteristics of complete data types and large data volume. Through multi-source data fusion, the accuracy of traffic prediction can be effectively improved. Brief Description of the Drawings

[0048] Figure 1 is a framework diagram of the multi-source data fusion-based highway toll station exit traffic flow prediction method provided by the embodiment of the present invention;

[0049] Figure 2 It is a diagram of the associated import screening steps provided according to an embodiment of the present invention;

[0050] Figure 3 It is a flowchart of travel time processing provided according to an embodiment of the present invention;

[0051] Figure 4 It is a schematic diagram of a highway toll station exit flow prediction system for multi-source data fusion provided according to an embodiment of the present invention. Specific embodiments

[0052] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and cannot be used to limit the protection scope of the present invention.

[0053] Refer to Figure 1 According to an embodiment of the present invention, a multi-source data fusion-based highway toll station exit flow prediction method includes the following steps S1-S5:

[0054] Step S1: Obtain highway toll data and vehicle license plate recognition data, where the highway toll data includes data of each exit toll station and each entrance toll station on the highway;

[0055] The entrance toll station data represents the transaction record information generated when a vehicle enters the toll station entrance, and the fields that need to be included are: entrance station number, entrance lane number, vehicle speed, entrance date and time, serial number.

[0056] The exit toll station data represents the transaction record information generated when passing through the toll station exit, and the fields that need to be included are: entrance station hexadecimal code, entrance lane number, entrance date and time, exit station number, exit lane number, exit time, serial number.

[0057] The vehicle license plate recognition data represents the license plate recognition data generated by the high-definition license plate recognition device for real-time license plate capture when the vehicle passes through the gantry, and the fields that need to be included are: license plate recognition serial number, recognized license plate, vehicle speed, recognized vehicle type, lane number, data generation time, mileage number.

[0058] Extract vehicle OD based on the toll data. Each OD is a two-dimensional vector, and its first component is the toll station entrance O, represented by the entrance station number; the second component is the toll station exit D, represented by the exit station number.

[0059] Step S2: Extract single-source data flow characteristics for the highway toll data and vehicle license plate recognition data respectively;

[0060] The method for extracting the single-source data flow characteristics in step S2 is: aggregate the data obtained in step S1 according to the time dimension of 5 minutes, there are s data in total, and the predicted time series is The flow sequence of the exit toll station to be predicted is:

[0061] Step S3: According to the spatiotemporal characteristics of the exit toll station to be predicted, the spatiotemporal dimensions of the traffic characteristics of each single-source data are unified to obtain the spatiotemporally consistent data of the entrance toll station and the spatiotemporally consistent vehicle license plate recognition data;

[0062] The method for temporal and spatial consistency of vehicle license plate recognition data in step S3 is as follows:

[0063] The vehicle license plate recognition data is screened according to the principle of time and space characteristic matching, and converted into vehicle license plate recognition data of the same time period and interval with the same road section as the constraint;

[0064] Reference Figure 2 , the temporal and spatial consistency method of the data of the entrance toll station is as follows:

[0065] Select a certain exit toll station as the target station D, and extract all the entry toll station sequences {O1, O2, …, O n};

[0066] Calculate the vehicle travel time within the OD pair, the calculation formula is: vehicle travel time = exit time - entry time;

[0067] Calculate the mean travel time of vehicles within the OD pair, and the calculation formula is:

[0068]

[0069] In the formula, Entrance toll booth O j The average travel time of all vehicles in time period k, O j The vehicle travel time of the mth vehicle on the D section in time period k, where M is the total number of vehicles;

[0070] Reference Figure 3 , select twice the standard deviation of the sample mean As the threshold range, remove Data outside the range, where:

[0071]

[0072] Calculate the sample mean Get the OD internal vehicle time surplus

[0073] Construct the association matrix between the target site and each entrance toll station. For a specific target site, at the time series when the entrance toll stations are {O1, O2, …, O n}, the association matrix is:

[0074]

[0075] Indicates the traffic volume when the entrance toll station is O j , the target site is D, and the time point is ;

[0076] Select a certain time point in the time series to be predicted for the target site D whose fluctuation range is and represent the start time and end time of this time point;

[0077] Since there is a certain distance in space between the target site and the corresponding entrance toll station, after a vehicle enters an entrance toll station, it takes some time to drive out from the target site, and there is a certain time lag. Therefore, considering the time lag, the corresponding entrance toll station The time fluctuation range is d j is the time lag of the vehicles within the OD pair; Calculate the correlation coefficient between the target site D and the associated sites as follows:

[0078]

[0079] In the formula, is the average value of the traffic volumes of each entrance toll station, is the traffic volume of the target site D at the time point ; is the average value of the traffic volumes of each time point of the target site D; Set the correlation threshold to 0.8, and screen out the entrance toll stations with correlation coefficients greater than the correlation threshold to complete the spatio-temporal consistency processing of the entrance toll station data.

[0080] Step S4: Construct a prediction model based on toll data and a prediction model based on vehicle license plate recognition data respectively. Among them, the prediction model based on toll data takes the spatio-temporally consistent entrance toll station data obtained in step S3 as input and the predicted traffic volume of the exit toll station as output. The prediction model based on vehicle license plate recognition data takes the spatio-temporally consistent vehicle license plate recognition data obtained in step S3 as input and the predicted traffic volume of the exit toll station as output;

[0081] Use the graph convolutional network (GCN) to capture the topological structure of the highway network and obtain its spatial dependence. Use the gated recurrent unit (GRU) to capture the dynamic changes of traffic information on the highway and obtain the temporal correlation. Construct a temporal graph convolutional network (TGCN) prediction model based on toll data to obtain the prediction results based on toll data.

[0082] The input of the model is the entrance toll station data after spatio-temporal homogenization processing in step S3, and the input form is a traffic flow sequence aggregated according to the 5-minute time dimension.

[0083] The temporal graph convolutional network model includes a GCN layer, a GRU layer, and a fully connected layer.

[0084] The GCN layer extracts the spatial features between regional blocks through spectral convolution. The GCN layer outputs a traffic flow matrix with spatial features to the GRU layer, and the GRU layer outputs a traffic flow matrix with temporal features. The obtained traffic flow matrix with temporal and spatial features is output through the fully connected layer to obtain the prediction result of the highway toll station exit traffic flow.

[0085] The prediction model GCN model described in step S4 is expressed as:

[0086]

[0087] X is the feature matrix, E is the identity matrix, D is the degree matrix, A is the adjacency matrix, W represents the weight, b represents the bias in the training process, Relu() is the activation function, and f(X,A) represents the graph convolution process;

[0088] The calculation process of each parameter in the TGCN model is as follows, f(A,X t ) represents the graph convolution process, h t-1 is the output at time t-1:

[0089] u t =σ(W u [f(A,X t ),h t-1 +b u )

[0090] r t =σ(W r [f(A,X t ),h t-1 +b r )

[0091] c t =tanh(W c [f(A,X t ),(rt *h t-1 )]+b c )

[0092] h t = u t *h t-1 +(1 - u t )*c t

[0093] The output of the prediction model based on toll data is the predicted traffic volume of the exit toll station, and the output form is a traffic volume sequence aggregated according to a 5-minute time dimension.

[0094] The evaluation metrics of the model are RMSE, MAE, and MAPE.

[0095] Construct an LSTM prediction model based on vehicle license plate recognition data, obtain the optimal hyperparameters of the LSTM model through a random search algorithm, and train the LSTM model with the optimal hyperparameters to obtain the prediction results based on vehicle license plate recognition data.

[0096] The input of the model is the vehicle license plate recognition data after spatio-temporal consistency processing in step S3.

[0097] The input form of the model is a traffic volume sequence of vehicle license plate recognition data aggregated according to a 5-minute time dimension. The optimal hyperparameters of the model include the hidden layer size and the learning rate. The search range of the hidden layer size is [32, 512], and the search range of the learning rate is [0.0001, 0.5]. The output of the model is the predicted traffic volume of the exit toll station, in the form of an exit toll station traffic volume sequence according to a 5-minute time dimension.

[0098] The output of the model is the predicted traffic volume at the toll station exit. The output form of the model is a highway toll station exit traffic volume sequence with one record every 5 minutes. The evaluation metrics of the model are RMSE, MAE, and MAPE.

[0099] Step S5: For the predicted traffic volume of the exit toll station output by the prediction model based on toll data and the prediction model based on vehicle license plate recognition data, perform fusion based on a neural network to obtain the multi-source prediction result of the fused exit toll station traffic volume.

[0100] In step S5, a BP neural network is adopted to fuse the predicted exit toll station flows output by the prediction model based on toll data and the prediction model based on vehicle license plate recognition data. The BP neural network has a three-layer structure including an input layer, a hidden layer, and an output layer. The neurons in the input layer are the two predicted results of the highway toll station exits under the same spatio-temporal conditions obtained through step S4. The form of the data input into the input layer is a sequence of highway toll station exit flows at 5-minute intervals.

[0101] The number of hidden layer nodes is initially determined according to the following formula, and then the number of hidden layer nodes is gradually increased to retrain the network:

[0102] j = m + n + a

[0103] In the formula: j, m, and n are the numbers of neurons in the hidden layer, input layer, and output layer respectively, and a is a constant.

[0104] The neurons in the output layer are the predicted results of the highway exit toll stations after being fused and predicted by the BP neural network. The form of the data output by the output layer is a sequence of highway toll station exit flows at 5-minute intervals.

[0105] The embodiment of the present invention also provides a highway toll station exit flow prediction system for multi-source data fusion. Referring to Figure 4 , it includes a multi-source detector module, a feature extraction module, a spatio-temporal consistency module, a single-source data prediction module, and a multi-source data fusion prediction module to implement the multi-source data fusion highway toll station exit flow prediction method described above:

[0106] The multi-source detector module acquires highway toll data and vehicle license plate recognition data; the feature extraction module extracts the flow features of each single-source data; the spatio-temporal consistency module unifies the spatio-temporal dimensions of each single source; the single-source data prediction module constructs a prediction model based on toll data and a prediction model based on vehicle license plate recognition data; the outputs of the two single-source prediction models are used as the inputs of the multi-source data fusion prediction module, and the predicted exit toll station flow is obtained based on the multi-source data fusion prediction module.

[0107] In the embodiment of the present invention, both the toll data and the vehicle license plate recognition data are based on the existing highway detection system. Therefore, there is no need to additionally establish a collection system during the process of predicting the highway toll station exit flow using the method and system provided by the embodiment of the present invention. So the method provided by the embodiment of the present invention has good portability for similar data sources.

[0108] The embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the scope of knowledge possessed by those of ordinary skill in the art.

Claims

1. A method for predicting the exit flow of highway toll stations by multi-source data fusion, characterized in that, It includes the following steps S1 - S5: Step S1: Obtain highway toll data and vehicle license plate recognition data, where the highway toll data includes data of each exit toll station and each entrance toll station on the highway; Step S2: Extract single-source data flow characteristics for the highway toll data and the vehicle license plate recognition data respectively; Step S3: According to the spatio-temporal characteristics of the to-be-predicted exit toll station, unify the spatio-temporal dimensions of each single-source data flow characteristic to obtain spatio-temporally unified data of the entrance toll station and spatio-temporally unified vehicle license plate recognition data; The spatio-temporal unification method of the vehicle license plate recognition data is as follows: Screen the vehicle license plate recognition data according to the principle of matching time and space characteristics, and convert it into vehicle license plate recognition data with the same time period and the same interval with the same section as the constraint; The spatio-temporal unification method of the data of the entrance toll station is as follows: Select a certain exit toll station as the target station D, and extract all the entrance toll station sequences {O1, O2, …, O n} with the target station as the exit toll station according to the entrance toll station where the vehicle enters and the exit toll station where the vehicle exits; Calculate the vehicle travel time within the OD pair, and its calculation formula is: vehicle travel time = exit time - entrance time; Calculate the average value of the vehicle travel time within the OD pair, and its calculation formula is: In the formula, is the entrance toll station O j is the average travel time of all vehicles within the time period k, is O j is the vehicle travel time of the m-th vehicle on section OD within the time period k, and M is the total number of vehicles; Select the two standard deviations of the sample mean as the threshold range, and exclude the data outside the range of where: Calculate the sample mean Obtain the time difference of in-vehicle for OD pairs Construct the association matrix between the target site and each entrance toll station. For a specific target site, at the time series the entrance toll stations are {O1, O2, …, O n}, the association matrix is: Indicates that the entrance toll station is O j , the destination station is D, and the time point is the traffic flow at that time; Select a certain time point in the time series to be predicted for the target site D Its fluctuation range is and represent the start time and end time of this time point; The corresponding entrance toll station considering the time surplus The time fluctuation range is d j is the time surplus of OD internal vehicles; calculate the correlation coefficient between the target station D and the associated stations as follows: Wherein, is the average value of the traffic flow of each entrance toll station, is the traffic flow of the target station D at the time point ; is the average value of the traffic flow of the target station D at each time point; Set the correlation threshold to 0.8, screen out the entrance toll stations with the correlation coefficient greater than the correlation threshold, and complete the spatio-temporal consistency processing of the entrance toll station data; Step S4: Build a prediction model based on toll data and a prediction model based on vehicle license plate recognition data respectively. The prediction model based on toll data takes the spatio-temporally unified data of the entrance toll station obtained in Step S3 as the input and the predicted exit toll station traffic flow as the output. The prediction model based on vehicle license plate recognition data takes the spatio-temporally unified vehicle license plate recognition data obtained in Step S3 as the input and the predicted exit toll station traffic flow as the output; Step S5: Based on the neural network, fuse the predicted exit toll station traffic flows output by the prediction model based on toll data and the prediction model based on vehicle license plate recognition data to obtain a multi-source prediction result of the fused exit toll station traffic flow.

2. The method for predicting the traffic flow at the exit of a highway toll station with multi-source data fusion according to claim 1, wherein The method for extracting single-source data traffic characteristics in step S2 is as follows: Aggregate the data obtained in step S1 according to a time dimension of 5 minutes. There are s pieces of data in total, and the prediction time series is The traffic flow sequence of the toll station to be predicted at the exit is 3. The method for predicting the exit flow of highway toll stations with multi-source data fusion according to claim 1, characterized in that The prediction model based on toll data described in Step S4 adopts a GCN model, and the GCN model is expressed as: X is the feature matrix, E is the identity matrix, D is the degree matrix, and A is the adjacency matrix, W represents the weight, b represents the bias during the training process, Relu() is the activation function, and f(X, A) represents the graph convolution process; The calculation process of each parameter in the TGCN model is as follows. f(A, X t ) represents the graph convolution process, and h t-1 is the output at time t - 1: u t = σ(W u [f(A, X t ), h t-1 + b u ) r t = σ(W r [f(A, X t ), h t-1 + b r ) c t = tanh(W c [f(A, X t ), (r t * h t-1 )] + b c ) h t = u t * h t-1 + (1 - u t ) * c t The output of the prediction model based on toll data is the predicted exit toll station traffic flow, and the output form is a traffic flow sequence aggregated according to the 5-minute time dimension; The prediction model based on vehicle license plate recognition data adopts an LSTM model, and the optimal hyperparameters of the LSTM model are obtained through a random search algorithm. The optimal hyperparameters of the model include the hidden layer size and the learning rate. The search range of the hidden layer size is [32, 512], and the search range of the learning rate is [0.0001, 0.5]. The output of the model is the predicted exit toll station traffic flow, in the form of an exit toll station traffic flow sequence according to the 5-minute time dimension.

4. The method for predicting the exit flow of highway toll stations with multi-source data fusion according to claim 1, characterized in that, In Step S5, a BP neural network is adopted to fuse the predicted exit toll station traffic flows output by the prediction model based on toll data and the prediction model based on vehicle license plate recognition data. The BP neural network structure is as follows: j = m + n + a In the formula: j, m, and n are the numbers of neurons in the hidden layer, input layer, and output layer respectively, and a is a constant.

5. The highway toll station exit traffic flow prediction system for multi-source data fusion is characterized in that, It includes a multi-source detector module, a feature extraction module, a spatio-temporal consistency module, a single-source data prediction module, and a multi-source data fusion prediction module to implement the multi-source data fusion-based highway toll station exit traffic flow prediction method as described in any one of claims 1-4: The multi-source detector module obtains highway toll data and vehicle license plate recognition data; the feature extraction module extracts the traffic flow features of each single-source data; the spatio-temporal consistency module unifies the spatio-temporal dimensions of each single source; the single-source data prediction module constructs a prediction model based on toll data and a prediction model based on vehicle license plate recognition data; the outputs of the two single-source prediction models are used as the inputs of the multi-source data fusion prediction module, and the predicted exit toll station traffic flow is obtained based on the multi-source data fusion prediction module.

Images