A management system and method for highway dynamic toll collection

Abstract

The application discloses a highway dynamic charging management system and method, and relates to the technical field of intelligent traffic management; the application sets different charging standards to stimulate drivers to adjust travel, realizes active adjustment of traffic flow, fundamentally guides users to select travel modes according to real-time road conditions and cost benefits, changes a traditional passive traffic distribution mode, optimizes highway network flow distribution, effectively relieves traffic congestion, and improves overall road network operation efficiency and service quality; the application uses a flow prediction model to perform parallel prediction operation, obtains multi-target optimization results of a highway dynamic charging scheme, including an optimal income target scheme, an optimal traffic flow target scheme, a road network balance target scheme and a minimum congestion time target scheme, provides multiple charging scheme selections for administrators, and meets different management requirements.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent traffic management, and particularly relates to a management system and method for dynamic toll collection on expressways. BACKGROUND

[0002] With the acceleration of urbanization and the continuous growth of car ownership, traffic congestion has become a common problem faced by major cities around the world. Traffic congestion not only leads to a significant increase in travel time, reducing people's quality of life, but also causes energy waste and environmental pollution, posing a serious threat to the sustainable development of cities. In this context, intelligent traffic management technology has emerged and gradually become a key means to solve traffic problems.

[0003] In order to alleviate traffic congestion on expressways, some areas have begun to implement dynamic toll collection strategies, i.e., adjusting toll standards according to different time periods, road sections, and traffic flow conditions to guide drivers to avoid congestion periods and road sections. However, existing dynamic toll collection technologies have the following problems and deficiencies:

[0004] Most current dynamic toll collection schemes are based on experience or simple statistical methods, lacking comprehensive consideration of factors such as traffic flow, driver behavior, and road network structure. This rough toll scheme formulation method is difficult to achieve precise control of traffic flow and cannot achieve the expected congestion relief effect.

[0005] In addition, existing dynamic toll collection technologies usually only focus on a single target, such as increasing toll revenue or alleviating traffic congestion, while ignoring other important targets such as road network balance and minimum congestion time. In actual traffic management, these targets are often interrelated and interdependent, and simply pursuing a certain target may worsen other targets.

[0006] In view of the above problems, it is necessary to propose a management system and method for dynamic toll collection on expressways. SUMMARY

[0007] The purpose of the present application is to solve the problems in the background art and propose a management system and method for dynamic toll collection on expressways.

[0008] In order to achieve the above purpose, the present application adopts the following technical solutions:

[0009] The present application provides, in a first aspect, a management system for dynamic toll collection on expressways, comprising a data acquisition module, a road network modeling module, a data analysis and traffic prediction module, a toll strategy matching and adjustment module, and a signal matching and execution module.

[0010] The data collection module collects the expressway dynamic data on the expressway in real time, including traffic flow data, toll collection data, road condition data, weather environment data, and the path selection preference data of private cars, shared cars and public transport passengers facing differential charges. The collected expressway dynamic data is sent to the data analysis and flow prediction module.

[0011] The traffic flow data is collected, including the daily average vehicle speed, the daily average vehicle flow, the daily average vehicle density, the congestion length ratio and the daily average accident rate of each road section.

[0012] The toll collection data is collected, including the basic toll and the toll per kilometer of each road section.

[0013] The road condition data is collected, including the maintenance length ratio, the average flatness and the accumulated water section length ratio of each road section.

[0014] The weather environment data is collected, including the average temperature, the average wind speed and the daily average rainfall of each road section.

[0015] The path selection preference data of private cars, shared cars and public transport passengers facing differential charges is collected, including the private car ratio, the shared car ratio and the public transport vehicle ratio of each road section.

[0016] The road network modeling module obtains the topological structure of each road section in the entire road network, and obtains the road length of each road section according to the topological structure. According to the topological structure of each road section and the expressway dynamic data, mathematical modeling is carried out, and the road length, road condition data and weather environment data of each road section are converted into a road network environment data matrix; the traffic flow data and the path selection preference data of private cars, shared cars and public transport passengers facing differential charges are converted into a road network road condition response data matrix; and the toll collection data is converted into a price control data matrix.

[0017] In the road network environment data matrix, the road network road condition response data matrix and the price control data matrix, each row represents a road section, and each column represents an expressway dynamic data. The generated road network environment data matrix, road network road condition response data matrix and price control data matrix are sent to the data analysis and flow prediction module.

[0018] The data analysis and flow prediction module extracts the correlation features and mathematical regression features between the road network environment data matrix, the road network road condition response data matrix and the price control data matrix through a machine learning algorithm based on a multi-layer perception, to obtain a flow prediction model for predicting traffic flow, vehicle speed and vehicle type through expressway dynamic data.

[0019] As a preferred mode of the present application, a machine learning model based on a multi-layer perceptron is established, including an input layer, hidden layers and an output layer.

[0020] The input layer has a preset even number of nodes, wherein the first half of the nodes receive the road network environment data matrix, and the second half of the nodes receive the price control data matrix.

[0021] The hidden layer has a preset number of layers, and the input of the previous layer is the output of the next layer. Each node in the hidden layer processes the output result of the previous node through convolution operation, adds a bias term and an activation function, and increases the expression ability of the model through the activation function.

[0022] The output layer is used to predict the prediction result of the road network condition response data matrix from the current time to a preset number of days.

[0023] As a preferred mode of the present application, the machine learning model based on the multi-layer perceptron is trained to determine the specific values of the weight matrix and the bias term in the output layer and each hidden layer, and the specific process is as follows:

[0024] The loss function is defined, and the error between the prediction result of the road network condition response data matrix and the true result is described by calculating the average value of the sum of the squares of the differences between the element values in the output layer and the corresponding true results, to obtain the loss value.

[0025] A preset number of training samples, i.e., road condition response data matrices at each time obtained according to true data, and their corresponding prediction results at each time are collected. For each training sample, forward propagation calculation is performed in turn, and the gradient of the loss function with respect to the weight matrix of the output layer and each hidden layer is calculated through the back propagation algorithm based on the loss function:

[0026] In the back propagation calculation process, the weight matrix and the bias term are updated by the stochastic gradient descent method SGD, and in the updating process, the first, second, third and fourth learning rates are used as the influence coefficients of the loss function with respect to the hidden layer weight matrix, the hidden layer weight matrix bias term, the output layer weight matrix and the output layer bias term, to control the update of the weight matrix and the bias term.

[0027] The weight matrix and the bias term are updated and iterated by the stochastic gradient descent method SGD until the termination condition is reached: the specific value of the loss function is less than a preset threshold. At this time, the final updated specific values of the hidden layer weight matrix, the hidden layer weight matrix bias term, the output layer weight matrix and the output layer bias term are recorded, and are brought back to the corresponding number of layers and node positions in the hidden layer and the output layer to obtain the final traffic prediction model.

[0028] The toll policy matching and adjusting module carries out parallel prediction operation through the traffic prediction model, obtains multi-objective optimization results representing the dynamic toll scheme of the expressway, including an optimal revenue target scheme, an optimal traffic flow target scheme, a road network balance target scheme and a minimum congestion time target scheme, and recommends the dynamic toll scheme of the expressway according to the multi-objective optimization results.

[0029] As a preferred mode of the present application, the parallel prediction operation is carried out based on the traffic prediction model, the traffic prediction model is copied several times, and each copy is sequentially numbered.

[0030] The upper and lower limits of the basic fee and the unit kilometer toll of each road section are defined, a random array uniquely corresponding to the copied traffic prediction model is randomly generated between the upper and lower limits of the basic fee and the unit kilometer toll of each road section, and a price control data matrix representing the randomly generated simulation value of the toll is obtained.

[0031] The randomly generated price control data matrix and the collected road network environment data matrix historical data are combined to form a data set for inputting the input layer of the traffic prediction model, and all data sets are respectively input into the traffic prediction model uniquely corresponding thereto, and the prediction results of each traffic prediction model are obtained.

[0032] Elements are extracted from the prediction results of each traffic prediction model, and the average value of the daily average speed, the average value of the traffic flow, the average value of the daily average vehicle density, the average value of the congestion length ratio, the average value of the daily average accident rate, the average value of the private car ratio, the average value of the shared car ratio and the average value of the public transport vehicle ratio of each road section in the road network road condition response data matrix are calculated.

[0033] As a preferred mode of the present application, the multi-objective optimization results, including the optimal revenue target scheme, the optimal traffic flow target scheme, the road network balance target scheme and the minimum congestion time target scheme, are found in the final results of each prediction through the target function, and the specific process is as follows:

[0034] The optimal revenue target scheme target function is defined: the product of the average value of the daily average speed of each road section and the unit kilometer toll is added to the basic fee of each road section to obtain result one, result one is multiplied by the average value of the traffic flow to obtain result two, and the traffic prediction model number satisfying the condition that the sum of result two of each road section is maximum is solved.

[0035] The optimal traffic flow target scheme target function is defined according to the average value of the daily average speed, the average value of the congestion length ratio and the average value of the daily average accident rate output by each traffic prediction model.

[0036] Solve the traffic prediction model number that satisfies the condition: the result of the optimal traffic flow target scheme target function is maximum.

[0037] Define the road network equilibrium target scheme target function:

[0038] Solve the result four of the square root of the daily average vehicle speed of each road section with respect to the road section; solve the result five of the square root of the difference between the average value of the traffic flow and the length of the road section with respect to the road section; solve the result six of the square root of the average value of each daily average vehicle density with respect to the road section; solve the traffic prediction model number that satisfies the condition: the sum of the result four, the result five and the result six of each road section is minimum;

[0039] Define the minimum congestion time target scheme target function:

[0040] Calculate the result seven of the sum of the average value of the congestion length proportion of each road section and the average value of the daily average accident rate; solve the traffic prediction model number that satisfies the condition: the sum of the result seven with respect to the road section is minimum.

[0041] Get the prediction result numbers that satisfy the optimal revenue target scheme target function, the optimal traffic flow target scheme target function, the road network equilibrium target scheme target function and the minimum congestion time target scheme target function respectively, denoted as the first number, the second number, the third number and the fourth number.

[0042] The signal matching and execution module sends the multi-objective optimization result representing the dynamic tolling scheme of the expressway to the administrator, recommends the scheme, and obtains the expressway dynamic tolling scheme selected by the administrator. According to the expressway dynamic tolling scheme selected by the administrator, dynamic tolling adjustment is implemented. Verify the consistency of the price control data matrix corresponding to each multi-objective optimization result, and according to the consistency verification result, feedback correction is performed on the operating parameters in the traffic prediction model, and the prediction result of the traffic prediction model is optimized.

[0043] The optimal revenue target scheme, the optimal traffic flow target scheme, the road network equilibrium target scheme and the minimum congestion time target scheme are sent to the administrator, and the corresponding prediction results, i.e. the road network traffic response data matrix output by the traffic prediction model numbered as the first number, the second number, the third number and the fourth number, are sent to the administrator, and dynamic tolling adjustment is implemented according to the expressway dynamic tolling scheme selected by the administrator.

[0044] As a preferred mode of the present application, the consistency of each multi-objective optimization result corresponding to the price control data matrix is verified, the input price control data matrix in the traffic prediction model numbered as the first number, the second number, the third number and the fourth number is obtained, and the consistency coefficient between each price control data matrix is obtained through the sum of the absolute value of the difference of each same position element between each price control data matrix, when there is a consistency coefficient less than a preset threshold, it is considered that there is a similar price control data matrix between the best benefit target scheme, the best traffic flow target scheme, the road network equilibrium target scheme and the minimum congestion time target scheme, it is determined that the traffic prediction model training is insufficient, and the multi-objective optimization result is insufficient in discrimination.

[0045] If it is identified that the multi-objective optimization result is insufficient in discrimination, the data analysis and traffic prediction module is commanded to retrain the traffic prediction model training, and the training parameters: the first, second, third and fourth learning rates are reduced to half of the initial value.

[0046] The present application provides a highway dynamic charging management method in the second aspect, comprising the following steps:

[0047] Step one, data acquisition and formatting processing;

[0048] Real-time acquisition of highway dynamic data on the highway, including traffic flow data, fee collection data, road condition data, weather environment data and path selection preference data of private car, shared car and public transportation passengers facing differential charging. Obtain the topological structure of each road section in the entire road network, and obtain the length of each road section according to the topological structure. According to the topological structure of each road section and the highway dynamic data, mathematical modeling is carried out, and the length of each road section, the road condition data and the weather environment data are converted into a road network environment data matrix; the traffic flow data and the path selection preference data of private car, shared car and public transportation passengers facing differential charging are converted into a road network traffic response data matrix; the fee collection data is converted into a price control data matrix.

[0049] Step two, machine learning modeling and training;

[0050] Feature extraction and regression fitting are carried out through a multi-layer perception machine learning algorithm, the correlation features and mathematical regression features between the road network environment data matrix, the road network traffic response data matrix and the price control data matrix are extracted, and the traffic prediction model for predicting traffic flow, vehicle speed and vehicle type through highway dynamic data is obtained. The future traffic flow trend and the selection trend of vehicle types are predicted through the traffic prediction model, and the prediction result data is obtained, which provides basic data for traffic flow optimization and charging mechanism change decision.

[0051] Step three, multi-objective prediction based on machine learning;

[0052] Optimization operation is carried out through the traffic prediction model, and multi-objective optimization results representing the dynamic tolling scheme of the expressway are obtained through parallel prediction operation and objective functions, including the best revenue target scheme, the best traffic flow target scheme, the road network balance target scheme and the minimum congestion time target scheme, and the dynamic tolling scheme of the expressway is recommended according to the multi-objective optimization results.

[0053] Step four, prediction result verification and execution;

[0054] The multi-objective optimization results representing the dynamic tolling scheme of the expressway are sent to the administrator for scheme recommendation, and the dynamic tolling scheme of the expressway selected by the administrator is obtained.

[0055] Step five, machine learning parameter feedback adjustment;

[0056] In the implementation process of the dynamic tolling adjustment, the consistency of the actual generated traffic flow data and the path selection preference data of private cars, shared cars and public transportation passengers facing different charges with the prediction results of the dynamic tolling scheme of the expressway selected by the administrator is verified.

[0057] The beneficial effects of the present application are:

[0058] 1. The present application stimulates drivers to adjust travel by setting different charging standards, actively adjusts traffic flow, fundamentally guides users to choose travel mode according to real-time road conditions and cost-effectiveness, changes the traditional "passive" traffic distribution mode, optimizes the traffic distribution of the expressway network, effectively relieves traffic congestion, and improves the overall road network operation efficiency and service quality;

[0059] 2. The present application uses the traffic prediction model to perform parallel prediction operation to obtain multi-objective optimization results of the dynamic tolling scheme of the expressway, including the best revenue target scheme, the best traffic flow target scheme, the road network balance target scheme and the minimum congestion time target scheme, to provide administrators with multiple charging scheme options to meet different management needs;

[0060] 3. In the implementation process of the dynamic tolling adjustment, the consistency of the actual data and the prediction results is verified, the running parameters of the traffic prediction model are feedback corrected according to the verification results, when the multi-objective optimization results are found to be insufficient in differentiation, the model is retrained and the training parameters are adjusted, so as to optimize the prediction results of the traffic prediction model. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 A system architecture diagram of a highway dynamic toll management system provided in an embodiment of the present application is shown in FIG. 1.

[0062] Figure 2 A multi-layer perceptron-based machine learning model topology diagram provided in an embodiment of the present application is shown in FIG. 2.

[0063] Figure 3 A method flowchart of a highway dynamic toll management method provided in an embodiment of the present application is shown in FIG. 3. DETAILED DESCRIPTION

[0064] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0065] In addition, if the present application embodiments involve "first", "second" and the like, the "first", "second" and the like are only for description purposes, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features with "first", "second" can explicitly or implicitly include at least one of the features. In addition, the meaning of "and / or" appearing throughout the text includes three parallel solutions. For example, "A and / or B" includes A solution or B solution, or A and B solutions. In addition, the technical solutions of each embodiment can be combined with each other, but it must be based on the realization of those skilled in the art. When the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, and is not within the scope of protection claimed by the present application.

[0066] Please refer to Figure 1 As shown in FIG. 1, a highway dynamic toll management system includes a data acquisition module, a road network modeling module, a data analysis and traffic prediction module, a toll strategy matching and adjustment module, and a signal matching and execution module.

[0067] The data acquisition module is responsible for accessing sensors, API interfaces, cameras, and GPS devices, and collecting highway dynamic data on the highway in real time, including traffic flow data, toll collection data, road condition data, weather environment data, and collecting path selection preference data of private cars, shared cars, and public transportation passengers facing differential tolls. The collected highway dynamic data is sent to the data analysis and traffic prediction module.

[0068] Traffic flow data is collected, including traffic flow data of each road segment average daily vehicle speed average daily vehicle flow average daily vehicle density congestion length ratio and average daily accident rate ;

[0069] collecting toll data, including the basic fee and the toll per kilometer of each road section i.

[0070] collecting road condition data, including the maintenance length ratio , average flatness and the length ratio of waterlogged road sections of each road section i.

[0071] collecting weather environment data, including the average temperature , average wind speed and average daily rainfall of each road section .

[0072] collecting private car, shared car and public transportation passenger path selection preference data, including the private car ratio , shared car ratio and public transportation vehicle ratio of each road section .

[0073] wherein, is the road section number, ; is the total number of road sections in the road network; wherein is the data collection time, in days.

[0074] The road network modeling module obtains the topological structure of each road section in the entire road network, and obtains the length of each road section according to the topological structure. According to the topological structure of each road section and the dynamic data of the highway, mathematical modeling is carried out to convert the length of each road section, road condition data and weather environment data into a road network environment data matrix; convert the traffic flow data and the path selection preference data of private cars, shared cars and public transportation passengers facing differential charges into a road network condition response data matrix; and convert the toll data into a price control data matrix.

[0075] Obtain the topological structure of each road section in the entire road network, generate a two-dimensional model of each road section in the road network, and obtain the length of each road section in the two-dimensional model.

[0076] Furthermore, mathematical modeling is performed based on the topology of each road segment and the dynamic data of the highway. The specific process is as follows:

[0077] Generate a road network environment data matrix:

[0078] ,

[0079] Generate a road network traffic response data matrix:

[0080] ,

[0081] Generate a price control data matrix:

[0082] ,

[0083] In the road network environment data matrix, road network traffic condition response data matrix, and price control data matrix, each row represents a road segment. Each column represents a specific dynamic data point for a highway.

[0084] The generated road network environment data matrix, road network traffic condition response data matrix, and price control data matrix are sent to the data analysis and traffic forecasting module.

[0085] The data analysis and traffic prediction module uses a multilayer perceptron-based machine learning algorithm to extract features and perform regression fitting. It extracts the correlation features and mathematical regression features between the road network environment data matrix, the road network traffic condition response data matrix, and the price control data matrix, and obtains a traffic prediction model that predicts traffic flow, vehicle speed, and vehicle type based on highway dynamic data.

[0086] Please see Figure 2 As shown, a machine learning model based on a multilayer perceptron is established, including an input layer, a hidden layer, and an output layer.

[0087] Among them, the input layer has Each node is used to receive data from the previous node. The datasets of road network environment data matrix and price control data matrix collected daily: and ,in This is a preset model backtracking constant, representing the number of days in the historical data window that needs to be backtracked for the predicted road network traffic response data matrix.

[0088] The hidden layer has a preset number of layers. Layers, each layer With a preset number There are nodes, among which In the hidden layers, the input of the previous layer is the output of the next layer. Each node in the hidden layer processes the output of the previous node through convolution operations, adding bias terms, and activation functions, and the activation functions further enhance the expressive power of the model.

[0089] The formula for calculating hidden layers is:

[0090]

[0091] in This is the index of the hidden layer number. ; For the first The first hidden layer The output values ​​of each node, where This is the activation function. For the first Layer node number index, ; For the first The total number of nodes contained in the hidden layer; where For the first Layer The output of each node, and The first Layer The weight matrix and bias terms of each node.

[0092] The output layer has a preset number of layers. Each node is used to predict the time from the current moment. Beginning, until Prediction results of the road network traffic response data matrix for Tianhou:

[0093]

[0094] The output formula of the output layer is:

[0095] in For the node index of the output layer, Corresponding to the prediction results ;in For the first Layer The output of each node, among which and The output layer is respectively the first The weight matrix and bias terms of each node.

[0096] Furthermore, the machine learning model based on the multilayer perceptron is trained to determine the specific values ​​of the weight matrix and bias terms in the output layer and each hidden layer. The specific process is as follows:

[0097] Define the loss function to describe the error between the predicted result and the true result of the road network traffic response data matrix:

[0098]

[0099] wherein is the column number of the road network traffic response data matrix, .

[0100] wherein is the element value of the i-th row and the j-th column in the road network traffic response data matrix output by the output layer on the t-th day, wherein is the true result corresponding to the element value

[0101] Collect a preset number of training samples, i.e. the road traffic response data matrix at each time according to the true data, and obtain the prediction result at each time corresponding thereto. For each training sample, forward propagation calculation is performed in turn, and the gradient of the loss function with respect to the weight matrix and the bias term of the output layer and each hidden layer is calculated by the back propagation algorithm: The gradient calculation formula of the weight matrix and the bias term of the hidden layer:

[0102]

[0103] The gradient calculation formula of the weight matrix and the bias term of the output layer:

[0104]

[0105]

[0106] Update the weight matrix and the bias term by the stochastic gradient descent method SGD, and control the update of the weight matrix and the bias term by the learning rate: the first learning rate , the second learning rate , the third learning rate and the fourth learning rate in the update process, and the formula is:

[0107] The gradient update formula of the weight matrix and the bias term of the hidden layer:

[0108] ,

[0109] The gradient update formula of the weight matrix and the bias term of the output layer:

[0110] ,​​​​​​​

[0111] wherein , , and are the updated specific values of the hidden layer weight matrix, the hidden layer weight matrix bias term, the output layer weight matrix and the output layer bias term respectively; wherein , , and are the original values of the hidden layer weight matrix, the hidden layer weight matrix bias term, the output layer weight matrix and the output layer bias term respectively before updating.

[0112] The updating iteration operation of the weight matrix and the bias term is carried out by the stochastic gradient descent method (SGD), until the termination condition is reached: the specific value of the loss function is less than the preset threshold At this time, the final updated specific values of the hidden layer weight matrix, the hidden layer weight matrix bias term, the output layer weight matrix and the output layer bias term are recorded, and are brought back to the corresponding number of layers and node positions in the hidden layer and the output layer to obtain the final traffic prediction model.

[0113] It should be noted that in modern traffic management, road tolling as an effective traffic demand management tool can not only improve economic efficiency, but also optimize the use efficiency of roads. The basic principle is to encourage drivers to make adjustments when traveling by setting different toll standards, thereby achieving active adjustment of traffic flow. This approach can fundamentally guide users to make travel choices based on real-time traffic conditions and cost-effectiveness, changing the traditional "passive" traffic distribution mode to a more flexible and efficient "active" control. Through this approach, not only can the traffic distribution of the expressway network be optimized, but also traffic congestion can be effectively alleviated, and the overall road network operation efficiency and service quality can be improved.

[0114] It should be further noted that the traffic prediction model aims to learn the rules between road tolling conditions and traffic flow from historical data through the multilayer perceptron algorithm, and make predictions of future traffic based on these rules. The input layer receives road network environment and price control data from the previous k1 days, which is processed through multiple hidden layer convolution operations and activation functions, and finally the road network traffic in the next few days is predicted through the output layer.

[0115] The charging strategy matching and adjusting module carries out parallel prediction operation through the traffic prediction model to obtain multi-objective optimization results representing the dynamic charging scheme of the expressway, including the best revenue target scheme, the best traffic flow target scheme, the road network equilibrium target scheme and the minimum congestion time target scheme, and recommends the dynamic charging scheme of the expressway according to the multi-objective optimization results.

[0116] The traffic prediction model is copied for parallel prediction operation parts, and are numbered as ;

[0117] The upper and lower limits of the basic fee and the unit kilometer charge of each road section are defined, and the random generation is generated between the upper and lower limits of the basic fee and the unit kilometer of each road section ;

[0118] ,

[0119] wherein = , , , is the random generated simulation value of the basic fee of each road section i in the i-th array; wherein = , , is the random generated simulation value of the unit kilometer charge of each road section i in the i-th array;

[0120] The total of randomly generated price control data matrices are recorded.

[0121] The randomly generated price control data matrix and the road network environment data matrix collected in the previous day are respectively composed into sets of data sets for group input traffic prediction model input layer, including:

[0122] The first group: and ;

[0123] The second group: and ;

[0124] The third group: and ;

[0125]

[0126] The Pth group: and​​ ;

[0127] Among them , ,..., The price control data matrix contained in each of them , ,..., The quantity is .

[0128] Group 1 to Group 2 Each dataset was copied according to its respective identifier. In each traffic prediction model, simulations are performed in continuous... The combination of basic fees and per-kilometer charges for the Tianzhong route: In road network environment data: Traffic flow prediction models are obtained from the road network traffic condition response data generated under the given conditions. The prediction results.

[0129] From various traffic prediction models Extract elements from the prediction results and calculate the prediction results for each iteration. In total Each road segment in the road network traffic response data matrix of the day Average daily vehicle speed Average daily traffic volume Average daily traffic density Average of the proportion of congestion length Average daily accident rate The average percentage of private cars The average percentage of car-sharing and the average proportion of public transport vehicles .

[0130] Furthermore, the objective function is used in each prediction. The final result The process involves finding multi-objective optimization results, including the optimal revenue target scheme, the optimal traffic flow target scheme, the road network equilibrium target scheme, and the minimum congestion time target scheme. The specific process is as follows:

[0131] Define the objective function of the optimal revenue target scheme: multiply the average daily vehicle speed of each road segment by the toll per unit kilometer, and add the basic cost of each road segment to obtain result one. Multiply result one by the average traffic flow to obtain result two. Solve for the traffic flow prediction model number that satisfies the condition: maximize the sum of result two for each road segment.

[0132] An average value of the daily average vehicle speed output according to each traffic flow prediction model p An average value of the congestion length proportion And an average value of the daily average accident rate Define the optimal traffic flow target scheme target function:

[0133]

[0134] Solve the condition: let the optimal traffic flow target scheme target function result: The flow prediction model number p when the result is maximum;

[0135] Define the road network equilibrium target scheme target function:

[0136] Solve the square root of the sum of the daily average vehicle speed of each road section with respect to the road section to obtain result four; solve the square root of the difference between the average value of the traffic flow of each road section and the length of the road section with respect to the road section to obtain result five; solve the square root of the average value of each daily average vehicle density with respect to the road section to obtain result six; solve the condition: let the sum of result four, result five and result six of each road section be the smallest;

[0137] Define the minimum congestion time target scheme target function:

[0138] Calculate the sum of the average value of the congestion length proportion of each road section and the average value of the daily average accident rate to obtain result seven; solve the condition: let the sum of result seven with respect to the road section be the smallest.

[0139] Get the prediction result numbers respectively satisfying the optimal benefit target scheme target function, the optimal traffic flow target scheme target function, the road network equilibrium target scheme target function and the minimum congestion time target scheme target function, denoted as first number Second number Third number And the fourth number .

[0140] The signal matching and execution module sends the multi-objective optimization result representing the dynamic toll scheme of the expressway to the administrator, recommends the scheme, and obtains the dynamic toll scheme of the expressway selected by the administrator. According to the dynamic toll scheme of the expressway selected by the administrator, the dynamic toll adjustment is implemented. Verify the consistency of the price control data matrix corresponding to each multi-objective optimization result, and according to the consistency verification result, feedback correction is performed on the operating parameters in the traffic flow prediction model, and the prediction result of the traffic flow prediction model is optimized.

[0141] The optimal revenue target scheme, the optimal traffic flow target scheme, the road network equilibrium target scheme and the minimum congestion time target scheme are sent to the administrator, and the corresponding prediction results, i.e. the road network traffic response data matrix output by the traffic prediction model numbered 、 、 and are sent to the administrator, and dynamic toll adjustment is implemented according to the administrator's selected highway dynamic toll scheme.

[0142] Further, the consistency of the price control data matrix corresponding to each multi-objective optimization result is verified, and the input price control data matrix in the traffic prediction model numbered 、 、 and is obtained.

[0143] 、 、 and .

[0144] The consistency coefficient between each price control data matrix is obtained by calculating the sum of the absolute values of the differences of each element 、 、 and , including:

[0145] and consistency coefficient ;

[0146] and consistency coefficient ;

[0147] and consistency coefficient ;

[0148] and consistency coefficient ;

[0149] and consistency coefficient ;

[0150] and consistency coefficient ;

[0151] When the consistency coefficient is less than the preset threshold, it is considered that there is a similar price control data matrix between the optimal benefit target scheme, the optimal traffic flow target scheme, the road network equilibrium target scheme and the minimum congestion time target scheme, it is determined that the traffic flow prediction model training is insufficient, and the multi-objective optimization result distinction is insufficient.

[0152] If the multi-objective optimization result distinction is insufficient, the data analysis and traffic prediction module is commanded to retrain the traffic flow prediction model training, and the training parameters: the first learning rate , the second learning rate , the third learning rate and the fourth learning rate are reduced to half of the initial value.

[0153] Referring to FIG. 1, a management method of highway dynamic tolling is shown, which comprises the following steps: Figure 3

[0154] Step 1, data acquisition and formatting processing;

[0155] Real-time acquisition of highway dynamic data on the highway, including traffic flow data, fee collection data, road condition data, weather environment data, and collection of private car, shared car and public transportation passenger path selection preference data in the face of differential charging. Obtain the topological structure of each road section in the entire road network, and obtain the length of each road section according to the topological structure. According to the topological structure of each road section and the highway dynamic data, mathematical modeling is carried out, and the length of each road section, road condition data, weather environment data are converted into road network environment data matrix; the traffic flow data and the path selection preference data of private cars, shared cars and public transportation passengers in the face of differential charging are converted into road network traffic response data matrix; the fee collection data is converted into price control data matrix.

[0156] Step 2, machine learning modeling and training;

[0157] Feature extraction and regression fitting are carried out through a machine learning algorithm based on a multi-layer perception machine, and the correlation features and mathematical regression features between the road network environment data matrix, the road network traffic response data matrix and the price control data matrix are extracted, to obtain a traffic flow prediction model for predicting traffic flow, vehicle speed and vehicle type through highway dynamic data. Through the traffic flow prediction model, the future traffic flow trend and the selection trend of vehicle types are predicted, and the prediction result data is obtained, which provides basic data for traffic flow optimization and charging mechanism change decision.

[0158] Step 3, multi-objective prediction based on machine learning;

[0159] ​The optimization operation is carried out through the traffic flow prediction model, and the multi-objective optimization results representing the expressway dynamic charging scheme are obtained through parallel prediction operation and objective function, including the best income target scheme, the best traffic flow target scheme, the road network balance target scheme and the minimum congestion time target scheme, and the expressway dynamic charging scheme is recommended according to the multi-objective optimization results.

[0160] Step four, prediction result verification and execution;

[0161] The multi-objective optimization results representing the expressway dynamic charging scheme are sent to the administrator for scheme recommendation, and the expressway dynamic charging scheme selected by the administrator is obtained. The dynamic charging adjustment is implemented according to the expressway dynamic charging scheme selected by the administrator.

[0162] Step five, machine learning parameter feedback adjustment;

[0163] In the implementation process of the dynamic charging adjustment, the consistency of the actual generated traffic flow data and the path selection preference data of private cars, shared cars and public transportation passengers facing different charges with the prediction results of the expressway dynamic charging scheme selected by the administrator is verified. According to the consistency verification result, the running parameters in the traffic flow prediction model are feedback corrected, and the prediction results of the traffic flow prediction model are optimized.

Claims

1. A management system for dynamic toll collection on highways, characterized in that, include: The data acquisition module collects real-time dynamic data of highways on the expressway. The road network modeling module obtains the topology of each road segment in the entire road network, and obtains the distance length of each road segment based on the topology; it performs mathematical modeling based on the topology of each road segment and the dynamic data of the expressway to obtain the road network environment data matrix, the road network traffic condition response data matrix, and the price control data matrix. The data analysis and traffic prediction module extracts the correlation and mathematical regression features between the road network environment data matrix, the road network traffic condition response data matrix, and the price control data matrix through a machine learning algorithm based on a multilayer perceptron, and obtains the traffic prediction model. The toll strategy matching and adjustment module performs parallel prediction calculations through a traffic prediction model to obtain multi-objective optimization results representing the dynamic toll scheme of the expressway. The specific process of performing parallel prediction calculations using the traffic prediction model is as follows: Make several copies of the traffic prediction model and number them sequentially. The basic cost and the upper and lower limits of the charge per unit kilometer for each road segment are defined. A random array that uniquely corresponds to the replicated traffic prediction model is randomly generated between the basic cost and the upper and lower limits of the charge per unit kilometer for each road segment, resulting in a price control data matrix that represents the randomly generated simulated value of the charge. The randomly generated price control data matrix and the collected road network environment data matrix historical data are combined to form the input layer dataset for the traffic prediction model. All datasets are then input into their unique corresponding traffic prediction models to obtain the prediction results of each traffic prediction model. Elements are extracted from the prediction results of various traffic prediction models, and the average values ​​of the prediction results for each road segment in the road network traffic response data matrix are calculated as follows: average daily vehicle speed, average traffic volume, average daily driving density, average congestion length percentage, average daily accident rate, average private car percentage, average shared car percentage, and average public transportation vehicle percentage. The objective function is used to find multi-objective optimization results in the final results of each prediction, including the optimal revenue objective scheme, the optimal traffic flow objective scheme, the road network equilibrium objective scheme, and the minimum congestion time objective scheme.

2. The management system for dynamic toll collection on highways according to claim 1, characterized in that, It also includes a signal matching and execution module: The signal matching and execution module sends the generated multi-objective optimization results to the administrator, recommends a scheme, and obtains the dynamic toll collection scheme for highways selected by the administrator; it implements dynamic toll collection adjustments based on the dynamic toll collection scheme selected by the administrator; it verifies the consistency of the price control data matrix corresponding to each multi-objective optimization result, and corrects the operating parameters in the traffic prediction model based on the consistency verification results.

3. The management system for dynamic toll collection on highways according to claim 1, characterized in that, The collected dynamic data of highways includes; Traffic flow data includes the daily average vehicle speed, daily average traffic volume, daily average traffic density, congestion length percentage, and daily average accident rate for each road segment; The toll collection data includes the basic toll for each road section and the toll per unit kilometer; Road condition data, including the percentage of maintenance length, average smoothness, and percentage of waterlogged road sections for each road segment; Weather and environmental data, including average temperature, average wind speed, and average daily rainfall for each road section; Data on route selection preferences of private car, car-sharing, and public transport passengers facing different fares, including the proportion of private cars, car-sharing, and public transport vehicles on each road segment.

4. The management system for dynamic toll collection on highways according to claim 1, characterized in that: In the road network environment data matrix, road network traffic condition response data matrix, and price control data matrix, each row represents a road segment; each column represents dynamic data of a highway; the generated road network environment data matrix, road network traffic condition response data matrix, and price control data matrix are sent to the data analysis and traffic forecasting module.

5. The management system for dynamic toll collection on highways according to claim 1, characterized in that, The specific process of feature extraction and regression fitting using a multilayer perceptron-based machine learning algorithm is as follows: Establish a machine learning model based on a multilayer perceptron, including an input layer, a hidden layer, and an output layer; The input layer is used to receive historical data from the road network environment data matrix and the price control data matrix. The hidden layer has a preset number of layers, and the input of the previous layer is the output of the next layer. Each node in the hidden layer processes the output of the previous node through convolution operations, adding bias terms and activation functions, and increases the expressive power of the model through activation functions. The output layer is used to predict the road network traffic condition response data matrix from the current time until a preset time later. The machine learning model based on a multilayer perceptron is trained to determine the specific values ​​of the weight matrix and bias terms in the output layer and each hidden layer.

6. The management system for dynamic toll collection on highways according to claim 5, characterized in that: The specific process of training a machine learning model based on a multilayer perceptron is as follows: Define a loss function, which is the average of the summations of the squared differences between the element values ​​in the output traffic response data matrix and the corresponding true results, to describe the error between the predicted and true results of the traffic response data matrix, and thus obtain the loss value. Forward propagation is performed based on the loss value, and the gradient of the loss function with respect to the weight matrices of the output layer and each hidden layer is calculated using a backpropagation algorithm based on the loss function, thereby updating the weight matrix and bias terms. During the backpropagation calculation, the weight matrix and bias terms are updated using the stochastic gradient descent (SGD) method. The update speed is controlled by the first, second, third, and fourth learning rates during the update process. The weight matrix and bias term are updated and iterated until the termination condition is met: the specific value of the loss function is less than a preset threshold.

7. The management system for dynamic toll collection on highways according to claim 1, characterized in that, The specific process of finding the multi-objective optimization result in the final result of each prediction using the objective function is as follows: Define the objective function of the optimal profit target scheme: multiply the average daily average vehicle speed of each road segment by the toll per unit kilometer, and add the basic cost of each road segment to obtain result one. Multiply result one by the average traffic flow to obtain result two. Solve for the traffic flow prediction model number that satisfies the condition: maximize the sum of result two for each road segment. Define the objective function for the optimal traffic flow target scheme: The objective function of the optimal traffic flow target scheme is defined based on the average daily average vehicle speed, the average daily congestion length percentage, and the average daily accident rate output by each traffic flow prediction model. Find the traffic flow prediction model number that satisfies the condition: maximizes the objective function result of the optimal traffic flow target scheme; Define the objective function of the road network equilibrium target scheme: The summation of the square root of the daily average vehicle speed for each road segment yields result four; the summation of the square root of the difference between the average traffic flow and the road segment length for each road segment yields result five; the summation of the square root of the average daily traffic density for each road segment yields result six. Solve for the condition that the traffic flow prediction model number is minimized when the sum of results four, five, and six for each road segment is minimized; Define the objective function of the minimum congestion time target scheme: Calculate the sum of the average congestion length percentage of each road segment and the average daily accident rate to obtain result seven; solve for the condition that the traffic flow prediction model number is minimized when the sum of result seven is minimized with respect to the road segments. Obtain the prediction results that satisfy the objective functions of the optimal profit target scheme, the optimal traffic flow target scheme, the road network equilibrium target scheme, and the minimum congestion time target scheme, respectively, and denote them as the first number, the second number, the third number, and the fourth number.

8. The management system for dynamic toll collection on highways according to claim 2, characterized in that, The specific process for optimizing the prediction results of the traffic prediction model is as follows: The generated optimal revenue target scheme, optimal traffic flow target scheme, road network equilibrium target scheme, and minimum congestion time target scheme will be sent to the administrator, and the corresponding prediction results, namely the road network traffic condition response data matrix output by the traffic prediction models numbered first, second, third, and fourth, will be sent to the administrator. Obtain the price control data matrices corresponding to the first, second, third, and fourth numbers, and calculate the consistency coefficient between each price control data matrix. If any consistency coefficient is less than a preset threshold, it is determined that the traffic prediction model is not sufficiently trained and the output multi-objective optimization results have insufficient discrimination. The data analysis and traffic prediction module is then ordered to retrain the traffic prediction model and reduce the learning rates of the first, second, third, and fourth training parameters to half of their initial values.

9. A method for managing dynamic toll collection on highways, applied to a dynamic toll collection management system for highways as described in any one of claims 1-8, characterized in that, Includes the following steps: Step 1: Data Acquisition and Formatting Processing; Real-time collection of dynamic highway data and road length of each segment; mathematical modeling based on the topology of each segment and the dynamic highway data to obtain road network environment data matrix, road network traffic condition response data matrix and price control data matrix; Step 2: Machine learning modeling and training; By extracting the correlation and mathematical regression features between the road network environment data matrix, the road network traffic condition response data matrix, and the price control data matrix using a machine learning algorithm based on a multilayer perceptron, a traffic flow prediction model is obtained. Step 3: Multi-objective prediction based on machine learning; The optimization calculation is carried out by the traffic prediction model. The multi-objective optimization results representing the dynamic tolling scheme of the expressway are obtained by parallel prediction calculation and objective function. Based on the multi-objective optimization results, the dynamic tolling scheme of the expressway is recommended. Step 4: Verification and execution of prediction results; The generated multi-objective optimization results representing the dynamic toll collection scheme for expressways are sent to the administrator for scheme recommendation, and the dynamic toll collection scheme selected by the administrator is obtained; dynamic toll collection is adjusted according to the dynamic toll collection scheme selected by the administrator. Step 5: Adjusting machine learning parameters based on feedback; Verify the consistency of the multi-objective optimization results, and based on the verification results, make feedback corrections to the operating parameters in the traffic prediction model to optimize the prediction results of the traffic prediction model.

Images