Effective path generation method based on clustering back-inference and section passenger flow estimation method

Abstract

The present invention relates to an effective path generation method based on clustering back-stepping and a cross-sectional passenger flow estimation method. The effective path generation method comprises the following steps: obtaining relevant data and performing pre-processing; determining the passenger boarding time, the passenger waiting time and the in-and-out travel time by statistics; obtaining the train travel time by comparing the urban rail train timetable, calculating the k_dist value of each data point in the passenger travel time data set D and drawing the k_dist curve of the passenger travel time data set D, and selecting a suitable neighborhood density parameter; selecting an object p that has not been traversed from the passenger travel time data set D, generating relevant cluster results by density connection until all objects in D are traversed, determining the density coefficient and screening out k_dist from the path set L. route valid paths; determine potential transfer nodes, and splice each path segment into a valid path set L. The present invention can realize the accurate generation of valid paths for urban rail transit travel and provide a basis for efficient and safe urban rail transit intelligent data.

Description

Technical Field

[0001] The present invention relates to a method for generating an effective path based on clustering reverse deduction, and also relates to a method for estimating a section passenger flow based on generating an effective path based on clustering reverse deduction, belonging to the technical field of rail transportation. Background Art

[0002] In recent years, my country's urban rail transit has developed rapidly, and the scale of the line network has continued to increase, showing a trend of complexity and networking. The passenger volume of rail transit is also increasing day by day. In order to alleviate the pressure brought by the huge passenger flow, it is necessary to carry out reasonable operation planning and capacity allocation of the line network, effectively allocate system resources, and adjust the operation strategy in time to ensure that rail transit can complete the transportation service tasks safely and efficiently. Therefore, accurate generation of effective path sets is conducive to improving the scientificity and accuracy of operation management, and can also provide an important basis for the decision-making of operation managers.

[0003] As urban rail transit networks gradually enter a networked operating environment, intelligent technologies such as big data provide a foundation for the development of "seamless transfers" in urban rail networks. They can accurately and efficiently analyze the travel characteristics and patterns of passengers in rail transit systems under complex network conditions, and become a key link in achieving efficient and safe urban rail transit intelligent data platform management. Cross-sectional passenger flow is an important indicator of capacity deployment and ticket clearing in the operation and management of rail transit systems, and accurate effective path generation is the basis for cross-sectional passenger flow calculation. Under network conditions, the urban rail transit network is highly complex and passengers have a variety of transfer options, resulting in many factors affecting the effective path, which brings challenges to the analysis of influencing factors and the determination of effective paths. Summary of the invention

[0004] The primary purpose of the present invention is to overcome the problems existing in the prior art and provide an effective path generation method based on clustering back-inference, which can realize the accurate generation of effective paths for urban rail transit travel, facilitate the efficient analysis of the travel characteristics and laws of passengers in the rail transit system under complex networking conditions, and provide a basis for efficient and safe urban rail transit intelligent data.

[0005] In order to solve the above technical problems, the present invention provides an effective path generation method based on clustering back-inference, which establishes a passenger travel simulation model based on the accurately generated effective path of urban rail transit travel, obtains a variety of information including cross-sectional passenger flow, and provides guidance for the capacity deployment of the urban rail system.

[0006] The effective path generation method based on clustering back-inference of the present invention comprises the following steps in sequence:

[0007] S1, obtaining relevant data and preprocessing it, the preprocessed relevant data includes AFC data set, train timetable data, line network start and end time information and line network connectivity;

[0008] S2, under the premise of arriving at the same platform, assume that the time it takes for passengers to enter and exit the station is the same, and that passengers will exit the station immediately after arriving at the terminal instead of staying; on this basis, the passenger's boarding time, passenger waiting time, and entry and exit travel time are statistically determined;

[0009] S3, compare the urban rail train schedule to obtain the train travel time, calculate the k_dist value of each data point in the passenger travel time dataset D and draw the k_dist curve of the passenger travel time dataset D, and select the appropriate neighborhood density parameter k custering ;

[0010] S4, select an object p that has not been traversed from the passenger travel time dataset D, and search for the time interval between entering and exiting the station in (N p -eps,N p +eps) range, determine the density parameters eps and minPts; where eps represents the neighborhood radius and minPts represents the neighborhood threshold;

[0011] S5, generate relevant cluster results by density connection until all objects in D are traversed, determine the density coefficient density and select k valid paths from the path set L. route Paths;

[0012] S6, determine potential transfer nodes, use Dijkstra algorithm to solve the feasible path from the starting point p to the end point q, solve the path and update it through the definition of the effective path by Dial algorithm, extract the transfer stations in each line according to the relationship between the "edges" between the station topology nodes, and store them in the transfer node set Transit; the specific steps are:

[0013] S601, calculate the travel cost of the extended path, assuming that p and q are two points in the path, and the dist(x1, x2) function represents the minimum travel cost between the two points x1 and x2 calculated by the Dijkstra algorithm in the road network structure diagram, and the minimum travel cost is based on the travel time, the number of transfers and the full load rate of the carriage;

[0014] S602, determine the validity of the extended path; according to the Dial algorithm's requirements for valid paths, during the process of path expansion, each node expansion must be far away from the starting point and close to the end point, that is, satisfy the valid path constraint; then for the valid path L in the valid path set L n , (i,j) is L n For the edge (i≠p,j≠q), the nodes i and j of the edge (i,j) satisfy:

[0015]

[0016] When the above valid path conditions are met, the extension of the path is considered valid and saved, and then the transfer node set Transit in the path is considered. If there is no node in Transit, the current extended path is considered to be the only valid path between the current travel pair.

[0017] S7, if Transit contains nodes, traverse the nodes in Transit, assuming that there is a transfer node in the transfer node set Transit as t i , from t i Starting from point q, the end point is calculated again using the Dijkstra algorithm from t i The path to the end point q is formed by concatenating each path segment and putting them into the path set L.

[0018] Furthermore, the train schedule data specifically includes: train departure schedule shed_depart, arrival schedule shed_inbound, then shed_depart i Indicates the departure times of all trains at station i throughout the day, shed_inbound i represents the arrival time of all trains at station i throughout the day, where the departure time of the kth train is shed_depart i (k), the arrival time of the kth train is shed_inbound i (k).

[0019] Furthermore, S2 specifically includes the following steps:

[0020] S201, determine the arrival time difference vector. A passenger travel record in the AFC data set includes: the starting station is station i, the terminal station is station j, and the passenger entry time at station i is T i , the exit time of passengers at station j is T j , when the passenger's trip is a non-transfer trip, it is assumed that the starting time of the passenger's ride is T start , the end time of the ride is T end , search for the train schedule arriving at station j shed_inbound j , shed_inbound j The length is L shed ; Define the arrival time difference vector The collection is as follows:

[0021]

[0022] S202, determine the passenger's travel train schedule, delete All elements less than 0 in the arrival time difference vector c are sorted from small to large. L×1 (j) is rearranged, so is the time when the passenger arrives at station j;

[0023] when satisfy When k is the train number that the passenger takes to arrive at station j, the waiting time and the time to enter the station can be determined. At this time, T start shed_depart is the time when train k leaves station i i (k), T end shed_inbound is the time when train k arrives at station j j (k);

[0024] S203, determine the passenger's boarding time, the time for entering and exiting each station, and the waiting time at the platform: the passenger's boarding time is T i -T j ;

[0025] Passengers’ walking time in and out of station i T_walk i satisfy:

[0026] For the waiting time T_wait at station i i Satisfy: T_wait i =shed_depart i (k)-T i .

[0027] Furthermore, S3 specifically includes the following steps:

[0028] S301, determining the k_dist value of each passenger travel record in the data set D; k represents the train number;

[0029] Define the function Dist to satisfy: Dist(i, j) = abs(D[i] - D[j]);

[0030] Define the row vector x_dist L×1 Satisfies: x_dist L×1 (i)={Dist(i,j),1≤j≤L};

[0031] In order from largest to smallest, the row vector x_dist L×1 Rearrange the elements in the sorted k_dist to get the row vector x_sorted n×1 , then, for the i-th data, its k_dist value is: k_dist(i) = x_sorted n×1 [nk];

[0032] S302, determine the k_dist graph segmentation point q;

[0033] k custering The value range is (1, L], k custering The value of should be able to distinguish between dense and sparse areas in the data set, with k custering Get the sorted k_dist value matrix x_sorted for the benchmark n×1 , with index value x as variable, plot its value with respect to x_sorted n×1 [x] curve, x_sorted n×1 Each point in (x) is a scattered point. Connecting the points gives the curve of the function f(x), where f(x) satisfies:

[0034]

[0035] In the formula, ceiling(x) means that the variable x is rounded up, and floor(x) means that the variable x is rounded down;

[0036] For the function f(x), there exists a split point q that satisfies:

[0037] S303, determine k custering The value of: Select all possible k in [1, L] custering Value, calculated q is saved as row vector x q , whose length is L q , then k custering The reasonable value of should be:

[0038]

[0039] Furthermore, S4 specifically includes the following steps:

[0040] S401, determine the value of the neighborhood radius eps: for eps, it satisfies:

[0041] For each point in the data set D, search for core points in its eps neighborhood and calculate the number. Assume that the eps neighborhood corresponding to the i-th data point in the data set is eps n×1 (i), then the neighborhood radius eps corresponding to the i-th data point takes the value eps i ,

[0042] eps n×1 (i)={Dist(i,j)<eps,1≤j≤L};

[0043] S402, determine the value of the neighborhood threshold minPts:

[0044] Another object of the present invention is to overcome the problems existing in the prior art and provide a method for estimating cross-sectional passenger flow based on clustering reverse effective path generation, comprising the following steps:

[0045] T1. Draw a network topology map based on the actual connectivity of the network, and clarify the various attributes of the urban rail transit network, including network start time, network closing time, network peak start time, network peak end time, and train timetable;

[0046] T2. Based on AFC passenger travel data, the passenger flow is input into the line network in time-divided units with specific passenger travel information as the unit. Passengers start waiting for the train from the moment they enter the network. When the time specified in the train schedule arrives, they end their waiting and start path selection. If the current train is fully loaded, it is considered that the passengers at the current stop station will not enter the passenger flow and continue to wait. If the current train is not fully loaded but cannot carry all the waiting passengers at the current stop station, it is considered that the passengers board the train in the order starting from the waiting time, and the remaining passengers who cannot board the train continue to wait.

[0047] T3. After the passenger data enters the network, the possible transfer behavior of the passenger is determined at each transfer station according to the valid path set. Assuming that the transfer station connects stations A and B, if only station A appears in the valid path set corresponding to the passenger, it is determined that the passenger will go to station A; if both stations A and B appear in the valid path set corresponding to the passenger, the probability of the passenger going to each station at the transfer station is determined based on the path travel time using the Logit probability distribution model. After allocating the passengers by the method of average probability distribution or maximum probability distribution, the simulation process continues;

[0048] T4. According to the simulation time granularity, the line network operation time period is divided. The time it takes for passengers to arrive at each section is the time it takes to reach the section before the destination plus half of the travel time to the last stop. When a vehicle arrives at a section, all passengers on the vehicle will be counted into the section passenger flow of the section during that period.

[0049] Compared with the prior art, the present invention has achieved the following beneficial effects: (1) extracting passenger travel characteristics and transfer-related parameters from existing passenger travel data, and on this basis providing a parameter adaptive model for generating a set of valid paths between specific travel pairs.

[0050] (2) A cross-sectional passenger flow estimation method based on the network simulation concept is used to set relevant parameters for different network lines, such as network opening and closing times, train arrival schedules and other data to set up the simulation model, and provide historical or predicted cross-sectional passenger flow data, thereby obtaining more passenger travel data.

[0051] (3) By using the non-aggregate method of cross-section simulation, the whole process of passenger tracking and the tracing of passenger flow at a specific section can be realized during the cross-section passenger flow simulation, thereby achieving the rational control and allocation of passenger flow and rail transit capacity at all sections of the entire network.

[0052] (4) Compared with the traditional effective path generation algorithm, this algorithm analyzes the AFC ticket data based on the temporal and spatial characteristics of the travel path, and establishes an adaptive density clustering model of travel data based on the density clustering method. After clustering, the effective path is searched based on the average time of each cluster, thereby achieving a more accurate generation of an effective path set.

[0053] (5) Traditional cross-sectional passenger flow is mostly based on numerical regression and analysis, lacking analysis of individual passenger travel tendencies. The generation of cross-sectional passenger flow in the present invention is based on simulation ideas. From the moment passengers enter the station, they move according to the network attributes. This method is adaptable and reliable, and can record the simulated passenger movement process and obtain more travel information including cross-sectional passenger flow, providing more accurate guidance for ticket clearing and capacity deployment, thereby reducing the network operating costs and improving economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a general flow chart of the effective path generation method based on clustering back-inference in the present invention;

[0055] Figure 2 It is a simplified flow chart of the clustering back-inference method in the effective path generation method based on clustering back-inference of the present invention;

[0056] Figure 3 It is a simplified flow chart of a generation method in the effective path generation method based on clustering back-inference of the present invention;

[0057] Figure 4 It is a flow chart of the method for estimating the cross-section passenger flow of the present invention;

[0058] Figure 5 is the k corresponding to the set of passenger travel time in the present invention custering -eps graph and k custering -minPts graph;

[0059] Figure 6 It is a line graph of the passenger travel time density from Zhongquzhuang to Xiaogang Station of the present invention;

[0060] Figure 7 This is a cross-sectional passenger flow estimation curve diagram of the present invention. DETAILED DESCRIPTION

[0061] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below with reference to specific diagrams.

[0062] like Figures 1 to 3 As shown, the effective path generation method based on clustering back-inference of the present invention comprises the following steps in sequence:

[0063] S1, obtaining relevant data and preprocessing it, the preprocessed relevant data includes AFC data set, train timetable data, line network start and end time information and line network connectivity;

[0064] Train schedule data specifically includes: train departure schedule shed_depart, arrival schedule shed_inbound, then shed_depart i Indicates the departure times of all trains at station i throughout the day, shed_inbound i represents the arrival time of all trains at station i throughout the day, where the departure time of the kth train is shed_depart i (k), the arrival time of the kth train is shed_inbound i (k).

[0065] S2, under the premise of arriving at the same platform, assume that the time it takes for passengers to enter and exit the station is the same, and that passengers will exit the station immediately after arriving at the terminal instead of staying; on this basis, the passenger's boarding time, passenger waiting time, and entry and exit travel time are statistically determined;

[0066] S2 specifically includes the following steps:

[0067] S201, determine the arrival time difference vector. A passenger travel record in the AFC data set includes: the starting station is station i, the terminal station is station j, and the passenger entry time at station i is T i , the exit time of passengers at station j is T j , when the passenger's trip is a non-transfer trip, it is assumed that the starting time of the passenger's ride is T start , the end time of the ride is T end , search for the train schedule arriving at station j shed_inbound j , shed_inbound j The length is L shed ; Define the arrival time difference vector The collection is as follows:

[0068]

[0069] S202, determine the passenger's travel train schedule, delete All elements less than 0 in the arrival time difference vector c are sorted from small to large. L×1 (j) is rearranged, so is the time when the passenger arrives at station j;

[0070] when satisfy When k is the train number that the passenger takes to arrive at station j, the waiting time and the time to enter the station can be determined. At this time, T start shed_depart is the time when train k leaves station i i (k), T end shed_inbound is the time when train k arrives at station j j (k);

[0071] S203, determine the passenger's boarding time, the time for entering and exiting each station, and the waiting time at the platform: the passenger's boarding time is T i -T j ;

[0072] Passengers’ walking time in and out of station i T_walk i satisfy:

[0073] For the waiting time T_wait at station i i Satisfy: T_wait i =shed_depart i (k)-T i .

[0074] S3, compare the urban rail train schedule to obtain the train travel time, calculate the k_dist value of each data point in the passenger travel time dataset D and draw the k_dist curve of the passenger travel time dataset D, and select the appropriate neighborhood density parameter k custering ;

[0075] S3 specifically includes the following steps:

[0076] S301, determining the k_dist value of each passenger travel record in the data set D;

[0077] Define the function Dist to satisfy: Dist(i, j) = abs(D[i] - D[j]);

[0078] Define the row vector x_dist L×1 Satisfies: x_dist L×1 (i)={Dist(i,j),1≤j≤L};

[0079] In order from largest to smallest, the row vector x_dist L×1Rearrange the elements in the sorted k_dist to get the row vector x_sorted n×1 , then, for the i-th data, its k_dist value is: k_dist(i) = x_sorted n×1 [nk];

[0080] S302, determine the k_dist graph segmentation point q;

[0081] k custering The value range is (1, L], k custering The value of should be able to distinguish between dense and sparse areas in the data set, with k custering Get the sorted k_dist value matrix x_sorted for the benchmark n×1 , with index value x as variable, plot its value with respect to x_sorted n×1 [x] curve, x_sorted n×1 Each point in (x) is a scattered point. Connecting the points gives the curve of the function f(x), where f(x) satisfies:

[0082]

[0083] For the function f(x), there exists a split point q that satisfies:

[0084] S303, determine k custering The value of: Select all possible k in [1, L] custering Value, calculated q is saved as row vector x q , whose length is L q , then k custering The reasonable value of should be:

[0085]

[0086] S4, select an object p that has not been traversed from the passenger travel time dataset D, and search for the time interval between entering and exiting the station in (N p -eps,N p +eps) range, determine the density parameters eps and minPts;

[0087] S4 specifically includes the following steps:

[0088]

[0089] For each point in the data set D, search for core points in its eps neighborhood and calculate the number. Assume that the eps neighborhood corresponding to the i-th data point in the data set is eps n×1(i), then the domain radius eps corresponding to the i-th data point takes the value eps i ,

[0090] eps n×1 (i)={Dist(i,j)<eps,1≤j≤L};

[0091] S402, determine the value of the neighborhood threshold minPts:

[0092] S5, generate relevant cluster results by density connection until all objects in D are traversed, determine the density coefficient density and select k valid paths from the path set L. route Paths;

[0093] S6, determine potential transfer nodes, use Dijkstra algorithm to solve the feasible path from the starting point p to the end point q, solve the path and update it through the definition of the effective path by Dial algorithm, extract the transfer stations searched in each line according to the relationship between the "edges" between the station topology nodes, and store them in the transfer node set Transit; S6 specifically includes the following steps:

[0094] S601, calculate the travel cost of the extended path, assuming that p and q are two points in the path, and the dist(x1, x2) function represents the minimum travel cost between the two points x1 and x2 calculated by the Dijkstra algorithm in the road network structure diagram, and the minimum travel cost is based on the travel time, the number of transfers and the full load rate of the carriage;

[0095] S602, determine the validity of the extended path; according to the Dial algorithm's requirements for valid paths, during the process of path expansion, each node expansion must be far away from the starting point and close to the end point, that is, satisfy the valid path constraint; then for the valid path L in the valid path set L n , (i,j) is L n For the edge (i≠p,j≠q), the nodes i and j of the edge (i,j) satisfy:

[0096]

[0097] When the above valid path conditions are met, the extension of the path is considered valid and saved, and then the transfer node set Transit in the path is considered. If there is no node in Transit, the current extended path is considered to be the only valid path between the current travel pair.

[0098] S7, if Transit contains nodes, traverse the nodes in Transit, assuming that there is a transfer node in the transfer node set Transit as t i , from ti Starting from point q, the end point is calculated again using the Dijkstra algorithm from t i The path to the end point q is formed by concatenating each path segment and putting them into the valid path set L.

[0099] like Figure 4 As shown, the cross-sectional passenger flow estimation method based on clustering reverse effective path generation of the present invention comprises the following steps:

[0100] T1. Draw a network topology map based on the actual connectivity of the network, and clarify the various attributes of the urban rail transit network, including network start time, network closing time, network peak start time, network peak end time, and train timetable;

[0101] T2. Based on AFC passenger travel data, the passenger flow is input into the line network in time-divided units with specific passenger travel information as the unit. Passengers start waiting for the train from the moment they enter the network. When the time specified in the train schedule arrives, they end their waiting and start path selection. If the current train is fully loaded, it is considered that the passengers at the current stop station will not enter the passenger flow and continue to wait. If the current train is not fully loaded but cannot carry all the waiting passengers at the current stop station, it is considered that the passengers board the train in the order starting from the waiting time, and the remaining passengers who cannot board the train continue to wait.

[0102] T3. After the passenger data enters the network, the possible transfer behavior of the passenger is determined at each transfer station according to the valid path set. Assuming that the transfer station connects stations A and B, if only station A appears in the valid path set corresponding to the passenger, it is determined that the passenger will go to station A; if both stations A and B appear in the valid path set corresponding to the passenger, the probability of the passenger going to each station at the transfer station is determined based on the path travel time using the Logit probability distribution model. After allocating the passengers by the method of average probability distribution or maximum probability distribution, the simulation process continues;

[0103] T4. According to the simulation time granularity, the line network operation time period is divided. The time it takes for passengers to arrive at each section is the time it takes to reach the section before the destination plus half of the travel time to the last stop. When a vehicle arrives at a section, all passengers on the vehicle will be counted into the section passenger flow of the section during that period.

[0104] The process of the effective path generation method based on clustering back-stepping is explained by taking the Guangzhou Metro Quzhuang-Xiaogang section as an example. Quzhuang Station is a transfer station for Line 5 and Line 6, and Xiaogang Station is a regular station for Line 8. The passenger flow input is the subway network data, timetable data and passenger information data for June 2015, including the subway AFC card swiping data for a week from June 1 to June 7 and the corresponding 2015 network planning data. The network route map is as follows Figure 2As shown. The earliest operating time of each line is 6:00, and the latest operating time is 23:30. The passenger travel data with the entry station as Quzhuang and the exit station as Xiaogang are screened, and a total of 1138 corresponding data sets are searched. After step S1, the entry and exit travel time of each station are obtained. According to the entry time and travel time, the travel time is calculated to obtain the Quzhuang-Xiaogang passenger travel time set.

[0105] According to the passenger entry and exit time and train travel schedule, after step S2, the passenger entry and exit travel time, waiting time and boarding time are calculated. The corresponding k custering -eps graph and k custering -minPts graph, such as Figure 5 As shown. custering -eps and k custering -minPts curves are relatively stable in [1,120], and the union [1,120] is taken as k custering The value range of k_dist is calculated, and the corresponding k-dist segmentation point q is calculated, where k custering The value of satisfies:

[0106]

[0107] Calculate k custering =19, take the k-dist value corresponding to the k-dist graph split point q as the expected eps value, and take the k value at this time as the expected minPts value. Take the average of eps and minPts under all k values ​​as the clustering parameter input. MinPts must be an integer, and round it off. Finally, eps is 63.97 and minPts is 23.

[0108] The travel time density of passengers from Quzhuang to Xiaogang Station is as follows: Figure 6 As shown in the figure, the length of passenger travel time is mostly concentrated between 1250s and 2400s, a few passengers travel between 2500s and 2750s, and the passenger travel time density in the rest of the time periods shows a flat peak feature. The data set contains 3 clusters, among which cluster I contains the largest amount of data, with a peak value of 4.282 and an average value of 1.984. The density parameter density I =4. The data volume of cluster II and cluster III is relatively small, with peak values ​​of 0.757 and 0.461, and average values ​​of 0.462 and 0.246, respectively. The density parameter density II =1, density III =1, k route The value is rounded to 6. Use this parameter value as input.

[0109] By comparing and analyzing the proposed density clustering effective path analysis, Dial algorithm and BFS algorithm under different constraints, taking the results of the path survey as the actual value, a total of 13 potential effective paths are generated, as shown in Table 1, where the cumulative inter-station running time does not include the time spent on entering the station, waiting for the train, vehicle parking, inter-station running and exiting the station. Path 1-Path 5 corresponds to Path 1-5 in the table, which is considered to be the confirmed effective travel path in the path survey results, called the effective path set, and Path E1-E8 represents 8 paths that were not selected by passengers in the questionnaire survey, called the potential path set.

[0110] Valid path set generation results

[0111]

[0112] Table 1

[0113] The adjusted Rand coefficient ARI is a performance evaluation indicator of the clustering model, which reflects the effect and accuracy of the clustering results and the estimated data set. Its value is between [0,1]. The closer the ARI is to 1, the closer the data set is to the true value. Using the adjusted Rand coefficient ARI as the evaluation standard, the Dial algorithm generates 9 paths in total, and the path generation results are [1,2,3,4,5,E1,E2,E3,E4], of which 4 additional paths are generated, and the Rand coefficient RI is 0.625; the BFS algorithm uses the number of transfers as a constraint. When the number of transfers is limited to 2, the path generation results are [1,2,3,E5], of which 1 additional path is generated, and the Rand coefficient RI is 0.5. When the number of transfers is limited to 3, the path generation results are [1,2,3,4,5,E1,E4,E5], of which 3 additional paths are generated, and the Rand coefficient RI is 0.556. Using the density clustering effective path method, the path generation result is [1,2,3,4,5,E1], the additional path is 1, and the Rand coefficient is 0.833. The comparison results are shown in the table. The adjusted Rand coefficient shows that the effective path generated by the density clustering method used in this paper has an advantage in accuracy compared to the traditional path generation method. The path generation results of each algorithm are shown in Table 2.

[0114] Comparison of algorithm results with travel survey data

[0115]

[0116] Table 2

[0117] Taking the Xilang-Kengkou section of Guangzhou Metro as an example, the process of the section passenger flow estimation method based on clustering reverse effective path generation is explained. Xilang Station is a transfer station for Line 1 and Guangfo Line, and Kengkou Station is a regular station on Line 1. The passenger flow input is the Guangzhou Metro line network data, all-day subway AFC card swiping data, timetable data and passenger information data on April 13, 2015. The simulation period is all day.

[0118] A total of 65,929 valid paths were obtained with the whole network data as input, and a total of 9,318 relevant paths of the Xilang-Kengkou section were intercepted. The section passenger flow was statistically analyzed with a time granularity of 15 minutes. Compared with the Guangzhou Metro clearing algorithm, the Spearman correlation was 0.943, and the two methods had a good correlation. The morning peak of the section passenger flow prediction of the two methods was both at 480 minutes (i.e. 8:00-8:15); the evening peak was both at 1110 minutes (i.e. 18:30-18:45). Based on the Guangzhou Metro section clearing data, the section passenger flow prediction of the two methods was based on the 15-minute time granularity of 15 minutes. The morning and evening peak values ​​of bimodal passenger flow are studied. The morning peak passenger flow estimated by Guangzhou Metro section clearing data is 2893 people / 15 minutes, and the morning peak passenger flow estimated by section passenger flow estimation method is 2641 people / 15 minutes, with a relative accuracy of 91.3%; the evening peak passenger flow estimated by Guangzhou Metro section clearing data is 1332 people / 15 minutes, and the evening peak passenger flow estimated by section passenger flow estimation method is 1207 people / 15 minutes, with a relative accuracy of 90.6%; the relative accuracy of all-day data is 84.4%. Guangzhou clearing data and method prediction results are all estimated data, so it is believed that the estimation results of this method are of practical significance. The section passenger flow estimation curve is shown in the figure below: Figure 7 shown.

[0119] This method adopts a non-aggregate computing approach and can achieve reverse tracking of section passenger flow. Taking the section from Quzhuang to the Zoo as an example, this section is located in the center of the network and has a complex passenger flow composition. The stations where the passenger flow from Quzhuang to the Zoo comes from throughout the day are shown in Table 3.

[0120] Tracing the source of passenger flow from Quzhuang to the zoo

[0121]

[0122]

[0123] Table 3

[0124] Compared with the traditional cross-sectional passenger flow estimation method, the present invention can achieve detailed positioning of passenger flow and individual station-to-station running trajectories, track the potential running trajectories of passengers between stations, or further achieve efficient regulation of cross-sectional passenger flow across the entire network and accurate allocation of transportation capacity. Obviously, based on the idea of ​​clustering reverse inference, other different forms of changes or modifications can be made to the present invention, such as pedestrian waiting time and travel time density clustering. It is impossible to list all implementation methods here. All obvious changes or modifications derived from the technical solution of the present invention are still within the scope of protection of the present invention.

[0125] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A method for generating an effective path based on clustering back-inference, characterized in that: The following steps are included in sequence: S1, obtaining relevant data and preprocessing it, the preprocessed relevant data includes AFC data set, train timetable data, line network start and end time information and line network connectivity; S2: Under the premise of arriving at the same platform, the time taken for passengers to enter and exit the station is the same, and passengers will exit the station immediately after arriving at the terminal instead of staying; on this basis, the passenger's boarding time, passenger waiting time, and entry and exit travel time are statistically determined; S3, compare the urban rail train schedule to obtain the train travel time, calculate the k_dist value of each data point in the passenger travel time dataset D and draw the k_dist curve of the passenger travel time dataset D, select the neighborhood density parameter k custering ; S4, select an object p that has not been traversed from the passenger travel time dataset D, and search for the time interval between entering and exiting the station in (N p -eps,N p +eps) range, determine the density parameters eps and minPts; where eps represents the neighborhood radius and minPts represents the neighborhood threshold; S5, generate relevant cluster results by density connection until all objects in D are traversed, determine the density coefficient density and select k valid paths from the path set. route paths, forming a valid path set L; S6, determine potential transfer nodes, use Dijkstra algorithm to solve the feasible path from the starting point p to the end point q, solve the path and update it through the definition of the effective path by Dial algorithm, extract the transfer stations in each line according to the relationship between the "edges" between the station topology nodes, and store them in the transfer node set Transit; the specific steps are: S601, calculate the travel cost of the extended path, assuming that p and q are two points in the path, and the dist(x1, x2) function represents the minimum travel cost between the two points x1 and x2 calculated by the Dijkstra algorithm in the road network structure diagram, and the minimum travel cost is based on the travel time, the number of transfers and the full load rate of the carriage; S602, determine the validity of the extended path; according to the Dial algorithm's requirements for valid paths, during the path extension process, each node extension must be far away from the starting point and close to the end point, that is, satisfy the valid path constraint; then for the valid path L in the valid path set L n , (i,j) is L n For the edge (i≠p,j≠q), the nodes i and j of the edge (i,j) satisfy: When the above valid path conditions are met, the extension of the path is considered valid and saved, and then the transfer node set Transit in the path is considered. If there is no node in Transit, the current extended path is considered to be the only valid path between the current travel pair; S7, if Transit contains nodes, traverse the nodes in Transit, and there is a transfer node in the transfer node set Transit that is t i , from t i Starting from point q, the end point is calculated again using the Dijkstra algorithm from t i The path to the end point q is formed by concatenating each path segment and putting them into the path set L.

2. The effective path generation method based on clustering back-inference according to claim 1 is characterized in that: The train schedule data specifically includes: train departure schedule shed_depart, arrival schedule shed_inbound, then shed_depart i Indicates the departure times of all trains at station i throughout the day, shed_inbound i represents the arrival time of all trains at station i throughout the day, where the departure time of the kth train is shed_depart i (k), the arrival time of the kth train is shed_inbound i (k).

3. The effective path generation method based on clustering back-inference according to claim 1 is characterized in that: S2 specifically includes the following steps: S201, determine the arrival time difference vector. A passenger travel record in the AFC data set includes: the starting station is station i, the terminal station is station j, and the passenger entry time at station i is T i , the exit time of passengers at station j is T j When the passenger's trip is a non-transfer trip, the starting time of the passenger's ride is T start , the end time of the ride is T end , search for the train schedule arriving at station j shed_inbound j , shed_inbound j The length is L shed ; Define the arrival time difference vector The collection is as follows: S202, determine the passenger's travel train schedule, delete All elements less than 0 in the arrival time difference vector c are sorted from small to large. L×1 (j) is rearranged, so is the time when the passenger arrives at station j; when satisfy When k is the train number that the passenger takes to arrive at station j, the waiting time and the time to enter the station are determined. At this time, T start shed_depart is the time when train k leaves station i i (k), T end shed_inbound is the time when train k arrives at station j j (k); S203, determine the passenger's boarding time, the time for entering and exiting each station, and the waiting time at the platform: the passenger's boarding time is T i -T j ; Passengers’ walking time in and out of station i T_walk i satisfy: For the waiting time T_wait at station i i Satisfy: T_wait i =shed_depart i (k)-T i .

4. The effective path generation method based on clustering back-inference according to claim 1 is characterized in that: S3 specifically includes the following steps: S301, determining the k_dist value of each passenger travel record in the data set D; k represents the train number; Define the function Dist to satisfy: Dist(i, j) = abs(D[i] - D[j]); Define the row vector x_dist L×1 Satisfies: x_dist L×1 (i)={Dist(i,j),1≤j≤L}; In order from largest to smallest, the row vector x_dist L×1 Rearrange the elements in the sorted k_dist to get the row vector x_sorted n×1 , then, for the i-th data, its k_dist value is: k_dist(i) = x_sorted n×1 [nk]; S302, determine the k_dist graph segmentation point q; The value range of kcustering is (1, L], k custering The value of should be able to distinguish between dense and sparse areas in the data set, with k custering Get the sorted k_dist value matrix x_sorted for the benchmark n×1 , with index value x as variable, plot its value with respect to x_sorted n×1 [x] curve, x_sorted n×1 Each point in [x] is a scattered point. Connecting the points gives the curve of the function f(x), where f(x) satisfies: In the formula, ceiling(x) means that the variable x is rounded up, and floor(x) means that the variable x is rounded down; For the function f(x), there exists a split point q that satisfies: S303, determine k custering The value of: Select all k in [1,L] custering Value, calculated q is saved as row vector x q , whose length is L q , then k custering The value of should be:

5. The effective path generation method based on clustering back-inference according to claim 4 is characterized in that: S4 specifically includes the following steps: S401, determine the value of the neighborhood radius eps: for eps, it satisfies: For each point in the data set D, search for core points in its eps neighborhood and calculate the number. Assume that the eps neighborhood corresponding to the i-th data point in the data set is eps n×1 (i), then the neighborhood radius eps corresponding to the i-th data point takes the value eps i , eps n×1 (i)={Dist(i,j)<eps,1≤j≤L}; S402, determine the value of the neighborhood threshold minPts:

6. A method for estimating cross-sectional passenger flow based on the effective path generation method based on clustering back-calculation according to claim 1, characterized in that: The steps include: T1. Draw a network topology map based on the actual connectivity of the network, and clarify the various attributes of the urban rail transit network, including network start time, network closing time, network peak start time, network peak end time, and train timetable; T2. Based on AFC passenger travel data, the passenger flow is input into the line network in time-divided units based on passenger travel information. Passengers start waiting for the train when they enter the network. When the time specified in the train schedule arrives, they end the waiting process and start the route selection process. If the current train is fully loaded, it is considered that the passengers at the current stop station will not enter the passenger flow and continue to wait. If the current train is not fully loaded but cannot carry all the waiting passengers at the current stop station, it is considered that the passengers board the train in the order starting from the waiting time, and the remaining passengers who cannot board the train continue to wait. T3. After the passenger data enters the network, the passenger transfer behavior is determined at each transfer station based on the valid path set. The transfer station connects two stations A and B. If only A appears in the valid path set corresponding to the passenger, it is determined that the passenger will go to station A. If both stations A and B appear in the valid path set corresponding to the passenger, the Logit probability distribution model is used to determine the probability of the passenger going to each station at the transfer station based on the path travel time. After allocating the passengers by the method of average probability distribution or maximum probability distribution, the simulation process continues; T4. According to the simulation time granularity, the line network operation time period is divided. The time it takes for passengers to arrive at each section is the time it takes to reach the section before the destination plus half of the travel time to the last stop. When a vehicle arrives at a section, all passengers on the vehicle will be counted into the section passenger flow of the section during that period.

Images