A comprehensive traffic assignment method based on mobile phone signaling

Abstract

The present invention discloses a comprehensive traffic distribution method based on mobile phone signaling, which uses mobile phone signaling data to solve the parameter estimation and parameter calibration problems of the comprehensive traffic random user equilibrium model, specifically including: mobile phone signaling preprocessing; extracting travel OD and travel chain data; building a comprehensive integrated traffic network and travel cost function; generating a travel path selection set; building a discrete choice model based on generalized selection; estimating parameters using the maximum likelihood estimation method; building a random user equilibrium model and using the MSWA algorithm to solve it; building and training a support vector regression model, if the accuracy does not meet the requirements, adjusting the parameters and retraining; combining the trained model with the STA_GA algorithm to solve the optimal parameters. The present invention can give full play to the advantages of mobile phone signaling data, improve the accuracy of traffic distribution of the comprehensive traffic random user equilibrium model, improve the efficiency of model parameter calibration through machine learning, and better provide a scientific basis for regional traffic planning and management.

Description

Technical Field

[0001] The invention belongs to the technical field of large-scale traffic travel planning, and in particular relates to a comprehensive traffic allocation method based on mobile phone signaling. Background Art

[0002] With the continuous development of the economy and society, my country's transportation infrastructure has developed rapidly and gradually improved. In order to better play the role of the transportation system, it is urgent to build a comprehensive transportation system. The comprehensive transportation model is the basis of the comprehensive transportation system. Through the transportation model, the user's travel situation can be obtained, providing data support for the construction of the comprehensive transportation system. However, due to the limitation of available data, most of the previous transportation models use the traditional four-stage method, which can only roughly show the travel situation of regional transportation and cannot provide clear guidance for comprehensive transportation planning and management. With the development of information technology and the advent of the big data era, traffic-related data can be better stored and utilized, such as GPS data, mobile phone signaling data, highway toll data, etc., which provides the possibility of improving the traditional comprehensive transportation allocation model.

[0003] With the popularization of smart phones, mobile phone signaling data has important application value in various fields. It can record the complete travel activity trajectory of users. It has the advantages of wide coverage, large user volume, low acquisition cost, and strong real-time performance. It is one of the important data types of traffic big data and can be used for regional traffic travel analysis.

[0004] However, the larger the amount of data, the more difficult it is to solve using the traditional traffic distribution model, and the time spent on solving will far exceed the acceptable range. The development of technologies such as parallel computing, machine learning, and heuristic algorithms can greatly improve the ability to process massive data and the efficiency of model solving, providing technical support for the application and development of traffic big data. Summary of the invention

[0005] The purpose of the present invention is to provide a comprehensive traffic distribution method based on mobile phone signaling in view of the problems existing in the above-mentioned prior art.

[0006] The technical solution to achieve the purpose of the present invention is: a comprehensive traffic distribution method based on mobile phone signaling, the method comprising:

[0007] Step 1, obtain residents’ travel OD data and travel chain data through mobile phone signaling data;

[0008] Step 2, generating a comprehensive integrated transportation network based on the existing road network;

[0009] Step 3, generating a travel path selection set based on the travel chain data of mobile phone signaling;

[0010] Step 4: construct a discrete choice model considering generalized choice;

[0011] Step 5, constructing a traffic assignment model;

[0012] Step 6, estimate the discrete choice model parameters using maximum likelihood estimation;

[0013] Step 7, calibrating parameters of the traffic assignment model based on machine learning;

[0014] Step 8: Based on the traffic assignment model, traffic assignment is performed according to traffic travel demand.

[0015] Furthermore, before executing step 1, the method further includes: preprocessing the mobile phone signaling data, specifically: cleaning the mobile phone signaling data to remove erroneous data, including duplicate data, ping-pong data and drift data.

[0016] Furthermore, the removing of erroneous data specifically includes:

[0017] The data is sorted according to the MDN code and time of the data. According to the location of the data, whether the distance between two adjacent data is less than the preset threshold is calculated. If so, the two are determined to be duplicate data and the latter data is deleted; according to the time and location information of the data, whether the speed of the two adjacent data is greater than the preset threshold is calculated. If so, it is determined to be ping-pong data and the latter data is deleted; it is determined whether there are base stations in the same MDN code data that appear more than the preset threshold number of times in a row. If so, it is determined to be drift data, and only the first and last data of the base station are retained, and the rest of the data is deleted.

[0018] Furthermore, in step 1, the resident travel OD data and travel chain data are obtained through mobile phone signaling data, specifically including:

[0019] Based on the pre-processed mobile phone signaling data, the user's residence point is determined according to the existing trajectory data, and then the OD data of the user's travel is identified by the residence point;

[0020] Extract the traveler codes based on OD data to obtain travel trajectory data;

[0021] Match the travel trajectory data with the GIS road network data to obtain the user's travel chain data.

[0022] Furthermore, step 2 of generating a comprehensive integrated transportation network based on the existing road network specifically includes:

[0023] Through virtual transfer sections, the hub nodes of different transportation modes are connected to form an integrated transportation network.

[0024] A road segment travel cost function is established for the integrated transportation network, and the travel cost of each road segment is calculated and used as the road segment weight.

[0025] Furthermore, the generation of the travel path selection set based on the travel chain data of the mobile phone signaling in step 3 specifically includes:

[0026] For each start-end point pair, i.e., OD point pair, the improved deviation path algorithm is used to obtain a travel path selection set including K possible travel paths, including:

[0027] Step 3-1, construct an empty shortest path set P and a candidate path set B;

[0028] Step 3-2, solve the shortest path between the OD point pairs and put them into the shortest path set P, and set the number of paths to be solved k to 2;

[0029] Step 3-3, let i be the cardinality of the shortest path set P, that is, the number of shortest paths in the set P, and replace the i-th shortest path p i Except for the end point v n The nodes are taken as deviation points v j , j = 1, 2, .., n-1, n+1 is the path p i The total number of nodes in ;

[0030] Step 3-4, for the deviation point v j , let the shortest path p i v j and the next node v j+1 The weight of the corresponding road segment d jq =∞, use astar algorithm to find the deviation point v j To the end v n The deviation path h j , and with the shortest path p i From the starting point v0 to v j Combine the paths and put them into the candidate path set B; repeat this step, traverse all deviation points, and obtain n candidate shortest paths;

[0031] Step 3-5, adding the path with the smallest weight in the candidate path set B to the shortest path set P, and deleting the path with the smallest weight from the candidate path set B;

[0032] Step 3-6, determine whether the number of paths in the shortest path set P is less than K. If so, set k=k+1 and go to step 3-3. If not, output the first K paths in the shortest path set P as the initial travel path selection set;

[0033] Step 3-7, compare the initial travel path selection set with the travel chain data extracted in step 1, and retain the paths with consistent data as the final travel path selection set of the OD pair.

[0034] Furthermore, in step 4, a discrete choice model considering generalized choice is constructed, and the Path-sizeLogit model, namely the PSL model, is adopted. According to the PSL model, the probability of the user selecting path k is:

[0035]

[0036] Among them, P k is the probability of selecting path k, V k =-μU k , U k and V k are the characteristic variables and system utility of path k, U k =x k γ+ε k , x k represents the cost influencing factor vector of path k, γ represents the coefficient vector of each influencing factor, and ε k represents an error term with a mean value of 0, μ is a positive parameter, and U l and V l are the characteristic variables and system utility of path l, K ij is the set of paths on OD pairs (i, j), any k∈K ij , β1 and β2 are parameters to be labeled, S k For the correction item:

[0037]

[0038] In the formula, l a is the cost of section a, L k is the cost of path k, Γ k is the set of road segments of path k. If road segment a is on path k, then δ aj =1, otherwise 0, C n is a set of paths, C n The minimum path cost on .

[0039] Furthermore, the traffic assignment model is constructed in step 5, specifically: a PSL-based random user equilibrium traffic assignment model is constructed based on the discrete choice model, and its specific form is:

[0040]

[0041]

[0042]

[0043]

[0044]

[0045] The first item in the model is the total travel cost of all road sections, and the second item is the error in the perceived travel cost; the constraints represent the flow conservation, the relationship between path flow and travel demand, the feasible domain of variables, and the calculation formula for section flow.

[0046] In the formula, x a is the flow rate on section a. If section a is on path k between OD pair (o, d), then otherwise t a (w) is the average cost required to travel when the traffic volume on road section a is w; is the traffic flow of path k between the OD pair (o, d); θ is a non-negative correction parameter.

[0047] Furthermore, in step 6, the discrete choice model parameters are estimated using the maximum likelihood estimation method. Due to the low accuracy of mobile phone signaling positioning, the selection set is grouped according to the main travel mode to generate a broad selection set, and then the maximum likelihood function is constructed for parameter estimation. Suppose the original precise selection set contains all paths k = 1, 2, 3..., K, and B is the broad selection set, C b (b=1, 2, 3, ..., B) represents a path selection group based on b types of transportation. Path k belongs to only one selection group. The probability of a user selecting a path selection group is the sum of the probabilities of selecting all paths included in the path group.

[0048] This step specifically includes: according to the trip chain data obtained in step 1 and the discrete choice model constructed in step 4, the parameters of the discrete choice model are estimated by using the maximum likelihood estimation method, wherein the likelihood function of the discrete choice model is constructed:

[0049]

[0050] Among them, S b is the set of routes of generalized category b, Z n'b is a dummy variable, which takes the value of 1 when user n1 chooses the generalized path selection group b, and 0 otherwise;

[0051] The user's travel chain data and road network data are loaded into the discrete choice model, and the maximum value of the likelihood function is calculated, thereby obtaining the parameter value of the discrete choice model.

[0052] Furthermore, the step 7 of calibrating the parameters of the traffic assignment model based on machine learning specifically includes:

[0053] Step 7-1, randomly generate traffic assignment model parameters: randomly generate several groups of candidate parameters near the empirical values ​​of the road segment travel cost function parameters;

[0054] Step 7-2, input each set of candidate parameters into the traffic assignment model constructed in step 5, and load the travel OD data into the model, and use the successive weighted average algorithm, i.e., the MSWA algorithm, to iteratively solve the traffic assignment model to obtain the traffic volume of the road section; the specific process of the MSWA algorithm includes:

[0055] (1) Let the number of iterations m = 1, the auxiliary variable γ0 = 0, and the initial flow rate of each section a

[0056] (2) Calculate the total formal cost of each path in the travel path selection set for each OD pair through the road segment travel cost function Calculate the allocated traffic volume of all paths in the travel choice set based on OD demand:

[0057]

[0058] In the formula, is the allocated flow of the kth path between OD pair (i, j), is the probability that the user chooses the kth path between the OD pair (i, j), q ij is the total demand between OD pairs (i, j);

[0059] Load the allocated flow of the path into the integrated transportation network and obtain the auxiliary flow of each section a after loading.

[0060] (3) Calculate the iteration step length, let the auxiliary variable β m =m d , γ m =γ m-1 +β m , d is a parameter to calculate the iteration step length

[0061] (4) Update road traffic: in are the flow rates of section a at the mth and m+1th iterations respectively;

[0062] (5) Determine the convergence conditions Whether it is satisfied, where ε' is the preset convergence threshold; if the convergence requirement is met, the algorithm terminates and outputs the section flow; otherwise, let m = m + 1 and go to (2);

[0063] Step 7-3, calculate the coupling degree z between the result of the traffic assignment model and the actual flow result:

[0064]

[0065] in, represents the flow distribution on section a, y a represents the actual flow on road segment a, and n' is the total number of road segments in the road network;

[0066] Traverse all candidate parameters and combine the parameters and corresponding coupling degrees as the training set for the subsequent machine learning model;

[0067] Step 7-4, build a machine learning model:

[0068] The mapping relationship between the parameters to be calibrated and the coupling degree of the road section flow is set as follows:

[0069] z=f(x)=w·x+b

[0070] Among them, x represents the value vector of the parameter to be calibrated, w represents the variable coefficient vector, and b represents the bias;

[0071] Let x i is the i-th group of candidate parameter vectors, i=1,2,…,q, q is the number of candidate parameter groups, z i is the road flow coupling degree corresponding to the i-th group of candidate parameters, the acceptable error is ε, and the slack variable ξ is introduced i and The SVR model is represented as:

[0072]

[0073] stz i -f(x i )=ε+ξ i ,i=1,2,…,q

[0074]

[0075] Where C is the regularization constant; l ε The loss function is an insensitive loss function:

[0076]

[0077] Construct the Lagrangian function:

[0078]

[0079] a i , μ i , As auxiliary variables, let the Lagrangian function be the variables w, b, ξ, ξ except the Lagrangian multiplier * The partial derivative of is zero, so:

[0080]

[0081] Convert the SVR model into its dual problem form:

[0082]

[0083]

[0084]

[0085]

[0086] Step 7-5, input the training set data obtained in step 7-2 into the SVR model and train it to obtain the best SVR model; input the traffic assignment model parameters into the SVR to output the coupling degree between the corresponding assigned flow and the actual flow;

[0087] Step 7-6, using the heuristic algorithm genetic algorithm to solve the parameters of the travel cost function with the highest coupling degree in the optimal SVR model, the specific steps are:

[0088] (1) Select a suitable coding scheme and randomly generate a population as the initial solution;

[0089] (2) For each individual in the population, calculate its fitness value;

[0090] (3) Select the individuals to be retained based on their fitness;

[0091] (4) Perform pairwise crossover on the selected individuals and add mutation operations to obtain the offspring population

[0092] (5) Whether the fitness difference of each individual meets certain conditions, if so, exit the algorithm; otherwise, go to step (2) to continue population reproduction and produce offspring until the optimal individual is obtained.

[0093] This completes the parameter calibration of the traffic assignment model.

[0094] Furthermore, in step 8, based on the traffic assignment model, traffic assignment is performed according to traffic travel demand, specifically including:

[0095] Input the existing travel OD data into the traffic assignment model calibrated in step 7, and use the MSWA algorithm to solve it to obtain the traffic assignment results on each road section.

[0096] Compared with the prior art, the present invention has the following significant advantages:

[0097] 1) Obtain users’ travel information through mobile phone signaling, match it with the integrated transportation network mainly composed of railways, highways and virtual transfer roads, and generate travel OD information and travel chain data. These data are used for choice set generation, parameter estimation of discrete choice models and parameter calibration of traffic assignment models, respectively, to fully tap the value of mobile phone signaling data. Mobile phone signaling data is large in volume and covers a wide range, making the model results closer to actual travel conditions.

[0098] 2) The generalized choice method is used to generate the generalized choice group to overcome the shortcoming that the mobile phone signaling cannot observe the user's specific travel route due to the low positioning accuracy. The generalized choice group contains some specific paths and can fit the user's choice from a higher dimension. And combined with the path size logit model (PSL), the problem of illogical selection probability in the MNL model on highly overlapping routes is solved.

[0099] 3) By constructing a machine learning model, instead of repeatedly solving the random user equilibrium model, the mapping relationship between the model parameters and the model allocation results and the actual traffic coupling degree can be obtained through a small number of samples, which greatly reduces the time for parameter calibration. At the same time, the genetic algorithm in the heuristic algorithm is used to search for the optimal parameters and calibrate the parameters, which improves the efficiency of parameter calibration. The random user equilibrium model after parameter calibration has higher accuracy.

[0100] The present invention is further described in detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0101] Figure 1 The present invention is a flow chart of the integrated traffic distribution method based on mobile phone signaling.

[0102] Figure 2 This is a schematic diagram of the comprehensive transportation network.

[0103] Figure 3 This is the flow chart of the deviation path algorithm.

[0104] Figure 4 This is the flow chart of the STA_GA algorithm. DETAILED DESCRIPTION

[0105] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0106] It should be noted that if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0107] In one embodiment, a comprehensive traffic distribution method based on mobile phone signaling is provided. By extracting the travel information of mobile phone signaling users, the travel OD and travel chain data are used in the random user equilibrium model, and the model is calibrated using the machine learning model and heuristic algorithm, thereby improving the accuracy and efficiency of regional comprehensive traffic network traffic distribution. Figure 1 , the method comprising:

[0108] Step S1, obtaining residents' travel OD data and travel chain data through mobile phone signaling data; before executing this step, it also includes: preprocessing the mobile phone signaling data, specifically: cleaning the mobile phone signaling data to remove erroneous data, including duplicate data, ping-pong data and drift data.

[0109] Here, removing the erroneous data specifically includes:

[0110] The data is sorted according to the MDN code and time of the data. According to the location of the data, whether the distance between two adjacent data is less than the preset threshold is calculated. If so, the two are determined to be duplicate data and the latter data is deleted; according to the time and location information of the data, whether the speed of the two adjacent data is greater than the preset threshold is calculated. If so, it is determined to be ping-pong data and the latter data is deleted; it is determined whether there are base stations in the same MDN code data that appear more than the preset threshold number of times in a row. If so, it is determined to be drift data, and only the first and last data of the base station are retained, and the rest of the data is deleted.

[0111] Step S1 specifically includes:

[0112] Based on the pre-processed mobile phone signaling data, the user's stay point is determined according to the existing trajectory data, and then the OD data of the user's travel is identified by the stay point;

[0113] Extract the traveler codes based on OD data to obtain travel trajectory data;

[0114] Match the travel trajectory data with the GIS road network data to obtain the user's travel chain data.

[0115] Step S2, generating an integrated transportation network based on the existing road network. Due to the low accuracy of mobile phone signaling positioning and the fact that the main modes of provincial travel are roads and railways, in this embodiment, the provincial road network and railway network are selected as the skeleton of the integrated transportation network, and the nodes and routes of the two road networks are stored in the form of a network graph, and virtual transfer sections are added to the adjacent nodes on the road network to connect the two road networks into an integrated transportation network, so that users can use multiple modes of transportation on the network. The connected network is as follows: Figure 2 As shown. The cost function of road segment travel is established for the transportation network, and its expression is:

[0116]

[0117] in, is the generalized travel cost of path p between OD pairs w. If road segment l belongs to path p, then b lp is 1, otherwise it is 0, c l is the generalized travel cost of road segment l:

[0118]

[0119] in, They represent travel time cost, travel monetary cost and travel comfort loss cost respectively, γ t , γ p , γ u They represent the weights of time, money and comfort respectively, and their sum is 1.

[0120] The specific calculation methods for various fees for highway sections are as follows:

[0121]

[0122]

[0123]

[0124] In the formula, is the formal time on the selected section l, v l is the traffic volume on road section l, is the traffic capacity on section l, α and β are parameters to be calibrated, and x l is the length of road section l, ρ is the toll per unit length, s h is the comfort loss of the car per unit time, η and λ are the monetary and time costs and the coefficient between the comfort loss and time, and their values ​​are 3.09 and 1.5 respectively.

[0125] For railway sections, the specific time and fare can be determined by the specific route operation and pricing, while the monetary cost and travel comfort loss cost are calculated as follows:

[0126]

[0127]

[0128] where p l is the fare for railway section l, s r is the comfort loss per unit time of the railway, is the maximum passenger capacity of railway section l. For the transfer section, the time taken to transfer from highway to railway is 30 minutes, and the time taken to transfer from railway to highway is 15 minutes. The monetary cost is set to 0, and the comfort loss is equal to the time cost.

[0129] Step S3: Generate a travel path selection set based on the travel chain data of the mobile phone signaling. Figure 3 , specifically including:

[0130] For each start-end point pair, i.e., OD point pair, the improved deviation path algorithm is used to obtain a travel path selection set including K possible travel paths, including:

[0131] Step S3-1, construct an empty shortest path set P and a candidate path set B;

[0132] Step S3-2, solving the shortest path between the OD point pairs and putting them into the shortest path set P, and setting the number of paths to be solved k to 2;

[0133] Step S3-3, let i be the cardinality of the shortest path set P, that is, the number of shortest paths in the set P, and take the i-th shortest path p i Except for the end point v n The nodes are taken as deviation points v j , j = 1, 2, .., n-1, n+1 is the path p i The total number of nodes in ;

[0134] Step S3-4, for the deviation point v j , let the shortest path p i v j and the next node v j+1 The weight of the corresponding road segment d jq =∞, use astar algorithm to find the deviation point v j To the end v n The deviation path h j , and with the shortest path p i From the starting point v0 to v jCombine the paths and put them into the candidate path set B; repeat this step, traverse all deviation points, and obtain n candidate shortest paths;

[0135] Step S3-5, adding the path with the smallest weight in the candidate path set B to the shortest path set P, and deleting the path with the smallest weight from the candidate path set B;

[0136] Step S3-6, determine whether the number of paths in the shortest path set P is less than K. If so, set k=k+1 and go to step S3-3. If not, output the first K paths in the shortest path set P as the initial travel path selection set;

[0137] Step S3-7, compare the initial travel path selection set with the travel chain data extracted in step S1, and retain the paths with consistent data as the final travel path selection set of the OD pair.

[0138] Step S4: construct a discrete choice model considering generalized choice. The Path-sizeLogit model, namely the PSL model, is used. According to the PSL model, the probability of a user choosing path k is:

[0139]

[0140] Among them, P k is the probability of selecting path k, V k =-μU k , U k and V k are the characteristic variables and system utility of path k, U k =x k γ+ε k , x k represents the cost influencing factor vector of path k, γ represents the coefficient vector of each influencing factor, and ε k represents an error term with a mean value of 0, μ is a positive parameter, and U l and V l are the characteristic variables and system utility of path l, K ij is the set of paths on OD pairs (i, j), any k∈K ij , β1 and β2 are parameters to be labeled, S k For the correction item:

[0141]

[0142] In the formula, l a is the cost of section a, L k is the cost of path k, Γ k is the set of road segments of path k. If road segment a is on path k, then δ aj=1, otherwise 0, C n is a set of paths, C n The minimum path cost on .

[0143] Step S5: construct a traffic assignment model. Specifically:

[0144] Based on the discrete choice model, a PSL-based stochastic user equilibrium traffic assignment model is constructed, and its specific form is:

[0145]

[0146]

[0147]

[0148]

[0149]

[0150] The first term in the model is the total travel cost of all road segments, and the second term is the error in perceived travel cost;

[0151] In the formula, x a is the flow rate on section a. If section a is on path k between OD pair (o, d), then otherwise t a (w) is the average cost required to travel when the traffic volume on road section a is w; is the traffic flow of path k between the OD pair (o, d); θ is a non-negative correction parameter.

[0152] Step S6, estimate the discrete choice model parameters using the maximum likelihood estimation method. Due to the low accuracy of mobile phone signaling positioning, the selection set is grouped according to the main travel mode to generate a broad selection set, and then the maximum likelihood function is constructed for parameter estimation. Suppose the original precise selection set contains all paths k = 1, 2, 3..., K, let B be the broad selection set, C b (b=1, 2, 3, ..., B) represents a path selection group based on b types of transportation, path k belongs to only one selection group, and the probability of a user selecting a path selection group is the sum of the probabilities of selecting all paths included in the path group. This step specifically includes: according to the travel chain data obtained in step S1 and the discrete choice model constructed in step S4, the maximum likelihood estimation method is used to estimate the parameters of the discrete choice model, wherein the likelihood function of the discrete choice model is constructed:

[0153]

[0154] Among them, S bis the set of routes of generalized category b, Z n'b is a dummy variable, which takes the value of 1 when user n1 chooses the generalized path selection group b, and 0 otherwise;

[0155] The user's travel chain data and road network data are loaded into the discrete choice model, and the maximum value of the likelihood function is calculated, thereby obtaining the parameter value of the discrete choice model.

[0156] Step S7, calibrating the parameters of the traffic assignment model based on machine learning, including:

[0157] Step S7-1, randomly generating traffic assignment model parameters: randomly generating several groups of candidate parameters near the empirical values ​​of the road segment travel cost function parameters;

[0158] Step S7-2, input each set of candidate parameters into the traffic assignment model constructed in step S5, and load the travel OD data into the model, and use the successive weighted average algorithm, i.e., the MSWA algorithm, to iteratively solve the traffic assignment model to obtain the traffic volume of the road section; the specific process of the MSWA algorithm includes:

[0159] (1) Let the number of iterations m = 1, the auxiliary variable γ0 = 0, and the initial flow rate of each section a

[0160] (2) Calculate the total formal cost of each path in the travel path selection set for each OD pair through the road segment travel cost function Calculate the allocated traffic volume of all paths in the travel choice set based on OD demand:

[0161]

[0162] In the formula, is the allocated flow of the kth path between OD pair (i, j), is the probability that the user chooses the kth path between the OD pair (i, j), q ij is the total demand between OD pairs (i, j);

[0163] Load the allocated flow of the path into the integrated transportation network and obtain the auxiliary flow of each section a after loading.

[0164] (3) Calculate the iteration step length, let the auxiliary variable β m =m d , γ m =γ m-1 +β m , d is a parameter to calculate the iteration step length

[0165] (4) Update road traffic: in are the flow rates of section a at the mth and m+1th iterations respectively;

[0166] (5) Determine the convergence conditions Whether it is satisfied, where ε' is the preset convergence threshold; if the convergence requirement is met, the algorithm terminates and outputs the section flow; otherwise, let m = m + 1 and go to (2);

[0167] Step S7-3, calculate the coupling degree z between the result of the traffic assignment model and the actual flow result:

[0168]

[0169] in, represents the flow distribution on section a, y a represents the actual flow on road segment a, and n' is the total number of road segments in the road network;

[0170] Traverse all candidate parameters and combine the parameters and corresponding coupling degrees as the training set for the subsequent machine learning model;

[0171] Step S7-4, building a machine learning model:

[0172] The mapping relationship between the parameters to be calibrated and the coupling degree of the road section flow is set as follows:

[0173] z=f(x)=w·x+b

[0174] Among them, x represents the value vector of the parameter to be calibrated, w represents the variable coefficient vector, and b represents the bias;

[0175] Let x i is the i-th group of candidate parameter vectors, i=1,2,…,q, q is the number of candidate parameter groups, z i is the road flow coupling degree corresponding to the i-th group of candidate parameters, the acceptable error is ε, and the slack variable ξ is introduced i and The SVR model is represented as:

[0176]

[0177] stz i -f(x i )=ε+ξ i ,i=1,2,…,q

[0178]

[0179] Where C is the regularization constant; l ε The loss function is an insensitive loss function:

[0180]

[0181] Construct the Lagrangian function:

[0182]

[0183] a i , μ i , As auxiliary variables, let the Lagrangian function be the variables w, b, ξ, ξ except the Lagrangian multiplier * The partial derivative of is zero, so:

[0184]

[0185] Convert the SVR model into its dual problem form:

[0186]

[0187]

[0188]

[0189]

[0190] Step S7-5, input the training set data obtained in step S7-2 into the SVR model and train it to obtain the best SVR model; input the traffic assignment model parameters into the SVR to output the coupling degree between the corresponding assigned flow and the actual flow;

[0191] Step S7-6, using a heuristic genetic algorithm (STA_GA) to solve the parameters of the travel cost function with the highest coupling degree in the optimal SVR model, the specific steps are:

[0192] (1) Select a suitable coding scheme and randomly generate a population as the initial solution;

[0193] (2) For each individual in the population, calculate its fitness value;

[0194] (3) Select the individuals to be retained based on their fitness;

[0195] (4) Perform pairwise crossover on the selected individuals and add mutation operations to obtain the offspring population

[0196] (5) Whether the fitness difference of each individual satisfies certain conditions, if so, exit the algorithm, otherwise go to step (2) to continue population reproduction and produce offspring until the optimal individual is obtained. The algorithm flow and coupling degree changes are as follows: Figure 4 shown.

[0197] This completes the parameter calibration of the traffic assignment model.

[0198] Step S8, traffic allocation is performed according to traffic travel demand, the existing travel OD data is input into the calibrated random user equilibrium model, and the MSWA algorithm is used to solve it to obtain the traffic allocation results on each road section.

[0199] In one embodiment, a comprehensive traffic distribution system based on mobile phone signaling is provided, the system comprising:

[0200] The first module is used to obtain residents’ travel OD data and travel chain data through mobile phone signaling data;

[0201] The second module is used to generate a comprehensive transportation network based on the existing road network;

[0202] The third module is used to generate a travel path selection set based on the travel chain data of mobile phone signaling;

[0203] The fourth module is used to build discrete choice models considering generalized choices;

[0204] The fifth module is used to build a traffic assignment model;

[0205] The sixth module is used to estimate the parameters of discrete choice models using maximum likelihood estimation;

[0206] A seventh module is used to calibrate parameters of the traffic assignment model based on machine learning;

[0207] The eighth module is used to perform traffic allocation according to traffic travel demand based on the traffic allocation model.

[0208] For the specific definition of the integrated traffic distribution system based on mobile phone signaling, please refer to the definition of the integrated traffic distribution method based on mobile phone signaling above, which will not be repeated here. Each module in the above-mentioned integrated traffic distribution system based on mobile phone signaling can be implemented in whole or in part by software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0209] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0210] Step 1, obtain residents’ travel OD data and travel chain data through mobile phone signaling data;

[0211] Step 2, generating a comprehensive integrated transportation network based on the existing road network;

[0212] Step 3, generating a travel path selection set based on the travel chain data of mobile phone signaling;

[0213] Step 4: construct a discrete choice model considering generalized choice;

[0214] Step 5, constructing a traffic assignment model;

[0215] Step 6, estimate the discrete choice model parameters using maximum likelihood estimation;

[0216] Step 7, calibrating parameters of the traffic assignment model based on machine learning;

[0217] Step 8: Based on the traffic assignment model, traffic assignment is performed according to traffic travel demand.

[0218] For the specific limitations of each step, please refer to the limitations of the comprehensive traffic allocation method based on mobile phone signaling in the above text, which will not be repeated here.

[0219] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0220] Step 1, obtain residents’ travel OD data and travel chain data through mobile phone signaling data;

[0221] Step 2, generating a comprehensive integrated transportation network based on the existing road network;

[0222] Step 3, generating a travel path selection set based on the travel chain data of mobile phone signaling;

[0223] Step 4: construct a discrete choice model considering generalized choice;

[0224] Step 5, constructing a traffic assignment model;

[0225] Step 6, estimate the discrete choice model parameters using maximum likelihood estimation;

[0226] Step 7, calibrating parameters of the traffic assignment model based on machine learning;

[0227] Step 8: Based on the traffic assignment model, traffic assignment is performed according to traffic travel demand.

[0228] For the specific limitations of each step, please refer to the limitations of the comprehensive traffic allocation method based on mobile phone signaling in the above text, which will not be repeated here.

[0229] The present invention can give full play to the advantages of mobile phone signaling data, improve the accuracy of traffic distribution of the comprehensive traffic random user equilibrium model, improve the efficiency of model parameter calibration through machine learning, and better provide a scientific basis for regional traffic planning and management.

[0230] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A comprehensive traffic distribution method based on mobile phone signaling, characterized in that: The method comprises: Step 1, obtain residents’ travel OD data and travel chain data through mobile phone signaling data; Step 2, generating a comprehensive integrated transportation network based on the existing road network; Step 3: Generate a travel path selection set based on the travel chain data of mobile phone signaling; specifically including: For each start-end point pair, i.e., OD point pair, the improved deviation path algorithm is used to obtain a travel path selection set including K possible travel paths, including: Step 3-1, construct an empty shortest path set P and a candidate path set B; Step 3-2, solve the shortest path between the OD point pairs and put them into the shortest path set P, and set the number of paths to be solved k to 2; Step 3-3, let i be the cardinality of the shortest path set P, that is, the number of shortest paths in the set P, and replace the i-th shortest path p i Except for the end point v n The nodes are taken as deviation points v j , j = 1, 2, .., n-1, n+1 is the path p i The total number of nodes in ; Step 3-4, for the deviation point v j , let the shortest path p i v j and the next node v j+1 The weight of the corresponding road segment d jq =∞, use astar algorithm to find the deviation point v j To the end v n The deviation path h j , and with the shortest path p i From the starting point v0 to v j Combine the paths and put them into the candidate path set B; repeat this step, traverse all deviation points, and obtain n candidate shortest paths; Step 3-5, adding the path with the smallest weight in the candidate path set B to the shortest path set P, and deleting the path with the smallest weight from the candidate path set B; Step 3-6, determine whether the number of paths in the shortest path set P is less than K. If so, set k=k+1 and go to step 3-3. If not, output the first K paths in the shortest path set P as the initial travel path selection set; Step 3-7, compare the initial travel path selection set with the travel chain data extracted in step 1, and retain the paths with consistent data as the final travel path selection set of OD pairs; Step 4: construct a discrete choice model considering generalized choice; Step 5, constructing a traffic assignment model; Step 6, estimate the discrete choice model parameters using maximum likelihood estimation; Step 7, calibrating parameters of the traffic assignment model based on machine learning; specifically including: Step 7-1, randomly generate traffic assignment model parameters: randomly generate several groups of candidate parameters near the empirical values ​​of the road segment travel cost function parameters; Step 7-2, input each set of candidate parameters into the traffic assignment model constructed in step 5, and load the travel OD data into the model, and use the successive weighted average algorithm, i.e., the MSWA algorithm, to iteratively solve the traffic assignment model to obtain the traffic volume of the road section; the specific process of the MSWA algorithm includes: (1) Let the number of iterations m = 1, the auxiliary variable γ0 = 0, and the initial flow rate of each section a (2) Calculate the total formal cost of each path in the travel path selection set for each OD pair through the road segment travel cost function Calculate the allocated traffic volume of all paths in the travel choice set based on OD demand: In the formula, is the allocated flow of the kth path between the OD pair (i, j), is the probability that the user chooses the kth path between the OD pair (i, j), q ij is the total demand between OD pairs (i, j); Load the allocated flow of the path into the integrated transportation network and obtain the auxiliary flow of each section a after loading. (3) Calculate the iteration step length, let the auxiliary variable β m =m d , γ m =γ m-1 +β m , d is a parameter to calculate the iteration step length (4) Update road traffic: in are the flow rates of section a at the mth and m+1th iterations respectively; (5) Determine the convergence conditions Whether it is satisfied, where ε′ is the preset convergence threshold; if the convergence requirement is met, the algorithm terminates and outputs the section flow; otherwise, let m=m+1 and go to (2); Step 7-3, calculate the coupling degree z between the result of the traffic assignment model and the actual flow result: in, represents the flow distribution on section a, y a represents the actual flow on road segment a, and n′ is the total number of road segments in the road network; Traverse all candidate parameters and combine the parameters and corresponding coupling degrees as the training set for the subsequent machine learning model; Step 7-4, build a machine learning model: The mapping relationship between the parameters to be calibrated and the coupling degree of the road section flow is set as follows: z=f(x)=w·x+b Among them, x represents the value vector of the parameter to be calibrated, w represents the variable coefficient vector, and b represents the bias; Let x i is the i-th group of candidate parameter vectors, i = 1, 2, ..., q, q is the number of candidate parameter groups, z i is the road flow coupling degree corresponding to the i-th group of candidate parameters, the acceptable error is ε, and the slack variable ξ is introduced i and The SVR model is represented as: s.t.z i -f(x i )=ε+ξ i ,i=1,2,…,q Where C is the regularization constant; l ε Is an insensitive loss function: Among them, e is the function independent variable; Construct the Lagrangian function: As auxiliary variables, let the Lagrangian function be the variables w, b, ξ, ξ except the Lagrangian multiplier * The partial derivative of is zero, so: Convert the SVR model into its dual problem form: Step 7-5, inputting the training set data obtained in step 7-2 into the SVR model and training it to obtain the best SVR model; Step 7-6, using a heuristic genetic algorithm to solve the parameters of the travel cost function with the highest coupling degree in the optimal SVR model, thereby completing the parameter calibration of the traffic assignment model; Step 8: Based on the traffic assignment model, traffic assignment is performed according to traffic travel demand.

2. The integrated traffic distribution method based on mobile phone signaling according to claim 1 is characterized in that: Before executing step 1, the method further includes: preprocessing the mobile phone signaling data, specifically: cleaning the mobile phone signaling data to remove erroneous data, including duplicate data, ping-pong data and drift data.

3. The integrated traffic distribution method based on mobile phone signaling according to claim 2 is characterized in that: The removing of erroneous data specifically includes: The data is sorted according to the MDN code and time of the data. According to the location of the data, whether the distance between two adjacent data is less than the preset threshold is calculated. If so, the two are determined to be duplicate data and the latter data is deleted; according to the time and location information of the data, whether the speed of the two adjacent data is greater than the preset threshold is calculated. If so, it is determined to be ping-pong data and the latter data is deleted; it is determined whether there are base stations in the same MDN code data that appear more than the preset threshold number of times in a row. If so, it is determined to be drift data, and only the first and last data of the base station are retained, and the rest of the data is deleted.

4. The integrated traffic distribution method based on mobile phone signaling according to claim 2 is characterized in that: In step 1, the resident travel OD data and travel chain data are obtained through mobile phone signaling data, including: Based on the pre-processed mobile phone signaling data, the user's stay point is determined according to the existing trajectory data, and then the OD data of the user's travel is identified by the stay point; Extract the traveler codes based on OD data to obtain travel trajectory data; Match the travel trajectory data with the GIS road network data to obtain the user's travel chain data.

5. The integrated traffic distribution method based on mobile phone signaling according to claim 1 is characterized in that: Step 2 generates a comprehensive transportation network based on the existing road network, specifically including: Through virtual transfer sections, the hub nodes of different transportation modes are connected to form an integrated transportation network. A road segment travel cost function is established for the integrated transportation network, and the travel cost of each road segment is calculated and used as the road segment weight.

6. The integrated traffic distribution method based on mobile phone signaling according to claim 1 is characterized in that: Step 4 constructs a discrete choice model considering generalized choice, using the Path-size Logit model, namely the PSL model. According to the PSL model, the probability of a user choosing path k is: Among them, P k is the probability of selecting path k, V k =-μU k , U k and V k are the characteristic variables and system utility of path k, U k =x k γ+ε k , x k represents the cost influencing factor vector of path k, γ represents the coefficient vector of each influencing factor, and ε k represents an error term with a mean value of 0, μ is a positive parameter, and U l and V l are the characteristic variables and system utility of path l, K ij is the set of paths on the OD pair (i, j), any k∈K ij , β1 and β2 are parameters to be labeled, S k For the correction item: In the formula, l a is the cost of section a, L k is the cost of path k, Γ k is the set of road segments of path k. If road segment a is on path k, then δ aj =1, otherwise 0, C n is a set of paths, C n The minimum path cost on .

7. The integrated traffic distribution method based on mobile phone signaling according to claim 6 is characterized in that: The traffic assignment model described in step 5 is constructed as follows: a PSL-based random user equilibrium traffic assignment model is constructed based on the discrete choice model, and its specific form is: st The first term in the model is the total travel cost of all road segments, and the second term is the error in perceived travel cost; In the formula, x a is the flow rate on section a. If section a is on path k between the OD pair (o, d), then otherwise t a (w) is the average cost required to travel when the traffic volume on road section a is w; is the traffic flow of path k between the OD pair (o, d); θ is a non-negative correction parameter.

8. The integrated traffic distribution method based on mobile phone signaling according to claim 1 is characterized in that: The method of estimating the parameters of the discrete choice model by using the maximum likelihood estimation method in step 6 specifically includes: estimating the parameters of the discrete choice model by using the maximum likelihood estimation method based on the trip chain data obtained in step 1 and the discrete choice model constructed in step 4, wherein the likelihood function of the discrete choice model is constructed: Among them, S b is the set of routes of generalized category b, Z n′b is a dummy variable, which takes the value of 1 when user n1 chooses the generalized path selection group b, and 0 otherwise; The user's travel chain data and road network data are loaded into the discrete choice model, and the maximum value of the likelihood function is calculated, thereby obtaining the parameter value of the discrete choice model.

Images