Method, system and storage medium for intelligent auxiliary dispatching of urban rail transit

Abstract

The present application relates to the field of rail transit technology, and discloses a method, system and storage medium for intelligent auxiliary dispatching of urban rail transit. The method includes: performing fluctuation analysis on train operation data to obtain fluctuation characteristics, time delay characteristics and trend characteristics of train delays, and constructing a train delay model; performing spatial distribution analysis on passenger flow data to obtain a crowd density map, performing mobility analysis on the crowd density map, and determining passenger flow concentration points; performing correlation analysis on train delay patterns and passenger flow concentration points, and determining the key pressure points among the 9 passenger flow concentration points that cause the train delay pattern; constructing multiple optimization parameters, predicting the crowd density at the key pressure points, obtaining density distribution values, and generating the best dispatching plan based on the density distribution values, optimization parameters and corresponding dispatching priorities. The present invention generates dispatching decisions based on the passenger flow prediction results, thereby improving passenger travel efficiency.

Description

Technical Field

[0001] The present application relates to the field of rail transit technology, and in particular to a method, system and storage medium for intelligent auxiliary scheduling of urban rail transit. Background Art

[0002] The traditional rail transit dispatching method mainly relies on manual dispatching instructions, and the train operation is monitored and directed through the control center. Although this dispatching method ensures the normal operation of rail transit to a certain extent, it is still prone to congestion and delays during peak hours and special circumstances due to its lack of flexibility and real-time performance.

[0003] In order to improve the accuracy and flexibility of scheduling, in recent years, more and more urban rail transit has begun to introduce intelligent auxiliary scheduling technology based on passenger flow prediction. For example, the Chinese patent document with publication number CN109552370A discloses a method for generating a train operation diagram based on passenger flow. The method obtains the original passenger flow data, processes the first passenger flow data according to the first passenger flow data of the previous prediction period of the current prediction period, and the preset model, obtains the optimal estimate of the current prediction period, and then generates the train operation diagram according to the preset constraints and the optimal estimate of the prediction period. The method adopts an automatic generation algorithm to improve the efficiency and quality of the operation diagram timetable generation, and makes the operation diagram more in line with passenger flow requirements. For example, a Chinese patent document with publication number CN108216311A discloses a rail transit scheduling method based on big data and the Internet of Things. The method predicts the passenger flow of trains on a certain rail line at different times of the day under a historical big data environment, and determines the departure intervals at different times of the day based on this. The departure intervals are used as basic data, and the current passenger flow of the train is monitored in real time through the Internet of Things. The current passenger situation of the rail line is collected, and the departure interval after correction of the predicted passenger flow at the next moment is obtained. Ultimately, the departure interval can be dynamically corrected according to the actual passenger flow of the train at any moment, making the train operation more efficient.

[0004] However, the existing dispatching system has poor linkage between passenger flow forecast results and dispatching decisions, and dispatching is often untimely and dispatching plans are poorly generated, which increases the labor intensity of central dispatchers and reduces passenger travel efficiency. Summary of the invention

[0005] In order to solve the problems existing in the above-mentioned background technology, the present application provides a method, system and storage medium for intelligent auxiliary scheduling of urban rail transit.

[0006] In order to achieve the above-mentioned invention object, the present invention proposes an intelligent auxiliary dispatching method for urban rail transit, comprising:

[0007] Acquire train operation data and passenger flow data of each station, wherein the train operation data includes the delay time of the train at each station;

[0008] Performing fluctuation analysis on the train operation data to obtain fluctuation characteristics, time delay characteristics and trend characteristics of train delays, and constructing a train delay model based on the fluctuation characteristics, the time delay characteristics and the trend characteristics;

[0009] Performing spatial distribution analysis on the passenger flow data to obtain a crowd density map, and performing mobility analysis on the crowd density map to determine passenger flow concentration points;

[0010] Performing correlation analysis on the train delay pattern and the passenger flow concentration point to determine the key pressure point in the passenger flow concentration point that causes the train delay pattern;

[0011] Construct multiple optimization parameters, predict the crowd density of the key pressure points of each station, obtain the density distribution value, and generate the best scheduling plan for each line based on the density distribution value, the optimization parameters and the corresponding scheduling priority.

[0012] Further, predicting the density distribution value includes the following steps:

[0013] Determine a forecast time period, extract historical passenger flow data of the same time period as the forecast time period from the passenger flow data, generate a first time series corresponding to each day based on the historical passenger flow data, extract the first time series N days before the forecast date as a second time series, combine all the second time series into a current sequence group, select a historical sequence group with the highest similarity to the current sequence group from the first time series, and compare the current sequence group with the historical sequence group to obtain a floating value;

[0014] The first time series after the historical series group is acquired as a fourth time series, and the fourth time series is corrected based on the floating value to obtain the density distribution value of the predicted date.

[0015] Further, correcting the fourth time series comprises the following steps:

[0016] Calculate the first average value of each first time series in the historical sequence group, calculate the second average value of each second time series in the current sequence group, calculate the ratio of the first average value to the second average value at corresponding positions, take the summed average value of the ratios as the floating value, and complete the correction process after each value in the fourth time series is expanded by the floating value.

[0017] Furthermore, determining the passenger flow concentration point includes the following steps:

[0018] Divide the station into multiple spatial areas, construct a spatial topology map based on the spatial areas, the spatial topology map includes nodes and edges, the nodes represent the spatial areas, and the edges represent the connections between the spatial areas, obtain the average crowd density of each node to generate the crowd density map, and define the node whose average crowd density is greater than a first threshold as an initial pressure point;

[0019] Perform liquidity analysis on the initial pressure point to obtain the liquidity of other nodes and each of the initial pressure points, calculate the clustering of nodes based on the liquidity, screen the nodes with clustering greater than a second threshold as focus points, calculate the degree centrality, closeness centrality, and betweenness centrality of the focus points, calculate the pressure value of the focus point based on the clustering and centrality values, and define the initial pressure point and the focus point with the pressure value greater than a third threshold as the passenger flow concentration point.

[0020] Further, determining the key pressure point comprises the following steps:

[0021] A training set is constructed based on the data of the passenger flow concentration points and the train delay pattern including the delay time, a delay prediction model is constructed based on the training set, a contribution analysis is performed on each of the passenger flow concentration points based on the delay prediction model, the SHAP value of each of the passenger flow concentration points is obtained, and the passenger flow concentration point with a SHAP value greater than a fourth threshold is positioned as the key pressure point.

[0022] Further, calculating the liquidity comprises the following steps:

[0023] Setting a plurality of sub-time periods, obtaining the number of flow passengers from each of the initial pressure points toward each node in each of the sub-time periods, and the original number of passengers at the initial pressure point in each of the sub-time periods, constructing flow sequence data and original sequence data based on the number of flow passengers and the original number of passengers, analyzing the flow sequence data and the original sequence data based on the Bayesian algorithm, and obtaining a flow transfer coefficient;

[0024] Perform flow velocity analysis and flow pressure analysis on the initial pressure point to obtain the flow efficiency factor and spatial congestion factor of the initial pressure point and each node, and calculate the fluidity between the initial pressure point and each node based on the flow efficiency factor, the spatial congestion factor and the flow transfer coefficient.

[0025] Furthermore, generating the optimal scheduling solution includes the following steps:

[0026] Based on the optimization parameters and the density distribution values, multiple objective functions are constructed, the objective functions include minimizing train delays, minimizing passenger waiting time and the density distribution value limits of the key pressure points, individual codes are constructed, and the individual codes include train running intervals, stop sites, stop times, and running speeds. A fitness function is constructed based on the objective function, and the corresponding objective function weights are assigned according to the scheduling priorities of the optimization parameters. The fitness function is solved based on a genetic algorithm to obtain the optimal scheduling solution.

[0027] Furthermore, constructing the train delay model includes the following steps:

[0028] The fluctuation characteristics include fluctuation range and fluctuation severity, and the delay characteristics include average delay time and delay frequency;

[0029] Based on the delay time, a delay time series for each station is constructed, wherein the fluctuation range is the difference between the maximum and minimum values ​​of the delay time series, the fluctuation severity is the standard deviation of the delay time series, the average delay time is the average value of the delay time series, and the delay frequency is the ratio of the number of delays in the delay time series to the total amount of data. A sliding window calculation is performed on the delay time series based on a sliding window of a preset scale to identify the change trend of the delay time;

[0030] The fluctuation characteristics and the time delay characteristics are used as clustering characteristics, a clustering algorithm is used to perform clustering analysis, and a plurality of train delay modes are divided based on the clustering results.

[0031] The present application provides an intelligent auxiliary dispatching system for urban rail transit, which is used to implement the above-mentioned intelligent auxiliary dispatching method for urban rail transit. The system includes:

[0032] A collection module, which obtains train operation data and passenger flow data of each station, wherein the train operation data includes the delay time of the train at each station;

[0033] A calculation module performs fluctuation analysis on the train operation data to obtain fluctuation characteristics, time delay characteristics and trend characteristics of train delays, and constructs a train delay model based on the fluctuation characteristics, the time delay characteristics and the trend characteristics;

[0034] An analysis module performs spatial distribution analysis on the passenger flow data to obtain a crowd density map, performs mobility analysis on the crowd density map to determine passenger flow concentration points, performs correlation analysis on the train delay pattern and the passenger flow concentration points, and determines the key pressure points in the passenger flow concentration points that cause the train delay pattern

[0035] The optimization module constructs multiple optimization parameters, predicts the crowd density of the key pressure points of each station, obtains the density distribution value, and generates the best scheduling plan for each line based on the density distribution value, the optimization parameters and the corresponding scheduling priority.

[0036] The present application provides a computer-readable storage medium having instructions stored thereon, wherein the instructions, when executed by a processor, implement the above-mentioned method for intelligent assisted scheduling of urban rail transit.

[0037] The present invention uses the delay time of each station as the analysis target, and combines historical data for big data analysis to determine multiple train delay patterns. After that, the passenger flow data is analyzed to determine the passenger flow concentration point, and combined with the dynamic changes of the passenger flow concentration point, the key pressure points that cause various train delay patterns are accurately identified, reducing the amount of calculation required for subsequent generation of scheduling plans. Finally, the intelligent optimization method is used to generate the optimal scheduling plan in advance in combination with the predicted passenger flow data for each key pressure point, thereby reducing manual intervention and improving the level of automation. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0039] Figure 1 A schematic diagram of the steps of an intelligent auxiliary dispatching method for urban rail transit in this application;

[0040] Figure 2 The present invention discloses a structural diagram of an intelligent auxiliary dispatching system for urban rail transit. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention 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 invention and are not intended to limit the present invention.

[0042] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script.

[0043] like Figure 1 As shown, a method for intelligent auxiliary dispatching of urban rail transit includes:

[0044] S1: Obtain the train operation data and passenger flow data of each station. The train operation data includes the delay time of the train at each station.

[0045] First, obtain the train operation data and passenger flow data of the station. The train operation data includes the actual arrival time and actual departure time of each train. Then, combine the preset train operation diagram in the system to obtain the ideal arrival time and ideal departure time of each train. By comparing the above actual departure time and ideal departure time, the first delay time of each train can be determined. For example, if the ideal departure time is 8:00 and the actual departure time is 8:01, the first delay time is 1min. Then, by comparing the actual arrival time and the ideal arrival time, the first delay time is corrected. If the ideal arrival time is 7:58 and the actual arrival time is 7:59, the second delay time of the station after correction is 0min, indicating that the station did not cause the train delay, and its delay was caused by the previous station.

[0046] Passenger flow data can be obtained by using cameras distributed in the subway station to capture station images of the corresponding area, and then using the CNN convolutional neural network to analyze and track the image, so as to obtain the number of people in the image area and the flow of people between areas, and the number of people in each area per minute is used as passenger flow data.

[0047] S2: Perform fluctuation analysis on train operation data to obtain the fluctuation characteristics, time delay characteristics and trend characteristics of train delays, and construct a train delay model based on the fluctuation characteristics, time delay characteristics and trend characteristics.

[0048] In this embodiment, the train operation data is subjected to fluctuation analysis with a time span of 7:00 to 9:00 every day. The specific analysis process is introduced later. The fluctuation characteristics, delay characteristics and trend characteristics are obtained after the analysis. The train operation data of multiple days are then clustered with the above characteristics, and a variety of delay modes are divided according to the clustering results. For example, according to the clustering results, it is found that station A is in delay mode 1 during the morning rush hour on weekdays. In delay mode 1, the average delay time of trains passing through the station is 30s, and the delay probability is 30%.

[0049] S3: Perform spatial distribution analysis on passenger flow data to obtain a crowd density map, perform mobility analysis on the crowd density map, and determine passenger flow concentration points.

[0050] When generating a crowd density map, the station is first divided into multiple spatial areas, and then the crowd density of each area is determined by calculating the ratio of the number of people to the area of ​​the spatial area. Then, the crowd density map of each time period is constructed according to the location of each spatial area and the corresponding crowd density. Finally, the crowd density map is analyzed for mobility according to the flow of people and density changes in each spatial area in different time periods, so as to determine the passenger flow concentration point. For example, the passenger flow concentration points are spatial area A and spatial area B, spatial area A is located in the transfer channel, and spatial area B is located in the train waiting area.

[0051] S4: Conduct correlation analysis on train delay patterns and passenger flow concentration points to determine the key pressure points in passenger flow concentration points that cause train delay patterns.

[0052] Through analysis, it was found that since passengers can enter space area B through multiple channels, the crowd density in space area B during the morning peak period increased greatly, which in turn caused the slow boarding and alighting speed in this area after the train arrived at the station, and the train could not close the door within the specified time, resulting in train departure delays, and ultimately causing the trains at this station to be in a specific train delay mode. Therefore, space area B was set as the key pressure point.

[0053] S5: Construct multiple optimization parameters, predict the crowd density at the key pressure points of each station, obtain the density distribution value, and generate the optimal scheduling plan for each line based on the density distribution value, optimization parameters and corresponding scheduling priorities.

[0054] In this embodiment, the optimization parameters include the train speed, running interval, stop site, stop time, carriage crowding and crowd density at key pressure points. In order to reduce the computational complexity of the subsequent generation of the scheduling plan, the present invention only predicts the crowd density at the key pressure points of each station, thereby generating the density distribution value of the key pressure points in each station, and then constructs the objective function based on the optimization parameters and the density distribution value, and uses the genetic algorithm to solve the objective function, and generates the best scheduling plan based on the result. In other embodiments, the solution algorithm can also use the particle swarm algorithm and the simulated annealing algorithm. For example, the best scheduling plan generated is to increase the train speed or add more trains.

[0055] The present invention uses the delay time of each station as the analysis target, and combines historical data for big data analysis to determine multiple train delay modes. After that, the passenger flow data is analyzed to determine the passenger flow concentration point, and combined with the dynamic changes of the passenger flow concentration point, the key pressure points that cause various delay modes of the train are accurately identified, reducing the amount of calculation for subsequent generation of scheduling plans. Finally, the intelligent optimization method is used to generate the optimal scheduling plan in combination with the real-time predicted passenger flow data. The present invention not only greatly improves the scheduling accuracy, but also reduces manual intervention and improves the level of automation.

[0056] It is particularly noteworthy that the present invention generates scheduling decisions based on passenger flow prediction results, thereby improving passenger travel efficiency.

[0057] In this embodiment, predicting the density distribution value includes the following steps:

[0058] Determine the forecast time period, extract historical passenger flow data of the same time period as the forecast time period from the passenger flow data, generate the first time series corresponding to each day based on the historical passenger flow data, extract the first time series N days before the forecast date as the second time series, combine all the second time series into the current sequence group, screen out the historical sequence group with the highest similarity to the current sequence group in the first time series, and compare the current sequence group with the historical sequence group to obtain a floating value.

[0059] If you want to predict the density distribution value of a spatial area in the prediction time period, for example, from 7:00 to 9:00 on the current day (April 10), extract the historical passenger flow data of the spatial area between 7:00 and 9:00 every day from the passenger flow data. The historical passenger flow data is the crowd density in the spatial area. If the historical passenger flow data is updated every 5 minutes, there will be 24 historical passenger flow data between 7:00 and 9:00 every day. The 24 historical passenger flow data are sorted into the first time series in chronological order. If the passenger flow data includes data from the past 60 days, 60 first time series will be generated. Here, N is set to 10, and the first time series of the ten days before April 10 are extracted from the first time series as the second time series. The 10 second time series are stored as the current sequence group.

[0060] Since the peak time of passenger flow is different every day, the present invention uses the DTW algorithm to calculate the similarity between the second time series and each of the remaining first time series. And select the historical sequence group with the highest similarity to the first sequence group. Here, the first time series included in the historical sequence group also have continuous appearance times. For example, the first time series on March 1 and March 10 are similar to the second time series as a whole, so the first time series between March 1 and March 10 are set as the historical sequence group.

[0061] The first time series after the historical series group is obtained as the fourth time series, and the density distribution value of the predicted date is obtained by correcting the fourth time series based on the floating value.

[0062] Specifically, correcting the fourth time series includes the following steps:

[0063] Calculate the first average value of each first time series in the historical sequence group, the second average value of each second time series in the current sequence group, calculate the ratio of the first average value and the second average value at corresponding positions, take the summed average value of the ratio as the floating value, and complete the correction process after expanding the floating value for each value in the fourth time series.

[0064] If the first average value of the first time series 1 in the historical sequence group is 300, and the second average value of the second time series 1 in the current sequence group is 280, then the ratio is 280 / 300=0.93. Similarly, continue to calculate the ratio of the corresponding average values ​​of the first time series 2 and the second time series 2, and calculate the average value of all ratios. Finally, the first time series [100, 200, 150] on March 11 is obtained, and the floating value is 1.05. The fourth time series obtained after correction is [105, 210, 157.5], and the fourth time series is used as the density distribution value predicted from 7:00 to 9:00 on April 10.

[0065] In this embodiment, determining the passenger flow concentration point includes the following steps:

[0066] The station is divided into multiple spatial areas, and a spatial topology map is constructed based on the spatial areas. The spatial topology map includes nodes and edges. Nodes represent spatial areas, and edges represent connections between spatial areas. The average crowd density of each node is obtained to generate a crowd density map, and the nodes with an average crowd density greater than a first threshold are defined as initial pressure points.

[0067] Perform liquidity analysis on the initial pressure point, obtain the liquidity of other nodes and each initial pressure point, calculate the clustering of nodes based on liquidity, select nodes with clustering greater than the second threshold as focus points, calculate the degree centrality, closeness centrality, and betweenness centrality of the focus points, calculate the pressure value of the focus point based on the clustering and centrality values, and define the initial pressure point and the focus point with a pressure value greater than the third threshold as passenger flow concentration points.

[0068] First, grid division is performed according to the station plan, so as to divide the station into multiple spatial areas. Then, a spatial topology diagram is constructed according to the connectivity relationship between the spatial areas. In the spatial topology diagram, the spatial areas are used as nodes. If there is connectivity between two spatial areas, an edge is set between the two to connect them. For example, if there is a transfer channel between spatial area A and spatial area B, an edge is set between the nodes representing spatial area A and spatial area B. When calculating the average crowd density of each node here, 5 or 10 days can be selected, and the first crowd density within 5 minutes during the peak period (2 hours) every day is calculated. Then, the first crowd density during the peak period is averaged to obtain the second crowd density, and the second crowd density within five days is averaged as the average crowd density in this embodiment. The first threshold is, for example, 1.5 people / square meter, and the node with an average crowd density greater than the first threshold is defined as the initial pressure point.

[0069] Since the people in the station are constantly moving, some nodes may be missed by only comparing the average crowd density with the fixed first threshold. This is because when there are more people in a certain space, the crowd will flow to the surrounding space area, resulting in a higher crowd density in the surrounding space area in a short time. Therefore, after the present invention determines the initial pressure point, it determines the mobility of the initial pressure point and other nodes according to the flow of people. The mobility in the present invention is one-way mobility. The greater the mobility between the initial pressure point and node A, the more likely the crowd in the initial pressure point is to flow to the space area corresponding to node A. The following formula can be used to calculate the aggregation of each node, , where α is the clustering of the nodes, N is the number of initial pressure points, and Gn is the mobility between the node to be calculated and the nth initial pressure point; after the focus points are screened out based on the second threshold, the degree centrality, closeness centrality, and betweenness centrality of each focus point are calculated.

[0070] Degree centrality is the number of edges connected to a node. The higher the degree centrality, the higher the connectivity of the spatial area and the greater the possibility of crowd gathering. Closeness centrality is the average distance between a node and other nodes. When a node is in the middle of the shortest path between other nodes, the node can be regarded as an intermediary node. Betweenness centrality is the statistical proportion of a node that appears as an intermediary node in the shortest path between all node pairs. Its betweenness centrality is high, and the crowd can stay there to get on the bus during the flow. Therefore, the larger the above centrality, the more likely it is that the crowd will stay at the corresponding point of interest. Finally, construct a pressure value calculation function. The calculation function is, for example, , where β is the pressure value, α is the clustering of the node, De is the degree centrality, Cl is the closeness centrality, and Be is the betweenness centrality. All the initial pressure points screened out above and the focus points whose pressure values ​​are greater than the third threshold are taken as the passenger flow concentration points.

[0071] In this embodiment, determining the focus pressure point of the passenger flow concentration point includes the following steps:

[0072] A training set is constructed based on the data of passenger flow concentration points and the train delay pattern including the delay time. A delay prediction model is constructed based on the training set. A contribution analysis is performed on each passenger flow concentration point based on the delay prediction model to obtain the SHAP value of each passenger flow concentration point. The passenger flow concentration point with a SHAP value greater than the fourth threshold is positioned as a key pressure point.

[0073] Specifically, a delay prediction model is constructed based on the random forest algorithm. The random forest algorithm uses the first crowd density of each passenger flow concentration point within the interval between two train runs as the input feature, and the delay time of each train at the station as the output feature. The delay prediction model is constructed by collecting passenger flow and train data for three months. After the construction is completed, the contribution of the delay prediction model is analyzed to obtain the SHAP value of each passenger flow concentration point. The SHAP value can quantify the contribution of each input feature to the prediction result, and is suitable for analyzing which passenger flow concentration points have a greater impact on the delay time.

[0074] For example, for passenger flow concentration points A, B, and C, their corresponding SHAP values ​​are determined by analysis to be 0.35, 0.2, and 0.6, respectively. The higher the SHAP value, the greater the contribution of the passenger flow concentration point to the train delay time. Here, the fourth threshold is set to 0.4, and the passenger flow concentration points with SHAP values ​​greater than the fourth threshold are positioned as key pressure points. Therefore, passenger flow concentration point C is defined as a key pressure point, indicating that when the crowd density at passenger flow concentration point C is large, it is easy to cause train delays at this station.

[0075] The calculation of liquidity in this embodiment includes the following steps:

[0076] Set multiple sub-time periods, obtain the number of flowing passengers from each initial pressure point toward each node in each sub-time period, and the original number of passengers at the initial pressure point in each sub-time period, construct flow sequence data and original sequence data based on the number of flowing passengers and the original number of passengers, analyze the flow sequence data and the original sequence data based on the Bayesian algorithm, and obtain the flow transfer coefficient.

[0077] The flow velocity analysis and flow pressure analysis are performed on the initial pressure point to obtain the flow efficiency factor and spatial congestion factor of the initial pressure point and each node. The fluidity between the initial pressure point and each node is calculated based on the flow efficiency factor, spatial congestion factor and flow transfer coefficient.

[0078] First, the peak period is divided into multiple sub-time periods, and the passenger flow data in each sub-time period over multiple days is obtained. For example, the number of passengers flowing from the initial pressure point A to node B in sub-time periods 1-3 of a certain day is 100, 200, and 300 respectively. The flow sequence data from the initial pressure point A to node B is [100, 200, 300]. The transfer coefficient is preset between 0 and 1. The flow data is defined to conform to the normal distribution. The likelihood function of the Bayesian algorithm is set to ,in, is the passenger flow data from the initial pressure point A to the node B in the i-th sub-time period, is the transfer coefficient, is the original number of passengers at the initial pressure point A in the i-th sub-period, is the variance of the flow sequence data, and finally the posterior probability distribution is calculated by the Markov chain Monte Carlo method, and the maximum posterior estimate is extracted from the posterior distribution as the flow transfer coefficient. For example, after calculation, the flow transfer coefficient from the initial pressure point A to the node B is 0.8.

[0079] This embodiment performs flow velocity analysis and flow pressure analysis on the initial pressure point based on the following formula. The flow efficiency factor and spatial congestion factor can be calculated using the following formula: , , and are the flow efficiency factor and spatial congestion factor between the initial pressure point A and node B, is the corridor width between the initial pressure point A and node B, is the maximum walking speed in the corridor between the initial pressure point A and node B, is the maximum crowd density in the corridor between the initial pressure point A and the node B. The maximum walking speed and the maximum crowd density can be extracted from the passenger flow data through statistical analysis. Finally, the flow transfer coefficient, flow efficiency factor and spatial congestion factor are weighted and summed to calculate the mobility between the initial pressure point and the node. For example, the weighted summation formula is: Where L is the mobility between the initial pressure point A and the node B, The flow transfer coefficient from initial pressure point A to node B.

[0080] In this embodiment, generating the best scheduling solution includes the following steps:

[0081] Based on the optimization parameters and density distribution values, multiple objective functions are constructed. The objective functions include minimizing train delays, minimizing passenger waiting time and density distribution value limits on key pressure points. Individual codes are constructed. Individual codes include train running intervals, stop sites, stop times, and running speeds. A fitness function is constructed based on the objective function, and the corresponding objective function weights are assigned according to the scheduling priority of the optimization parameters. The fitness function is solved based on the genetic algorithm to obtain the optimal scheduling solution.

[0082] The goal of minimizing train delays is to reduce train operation delays and avoid delays caused by excessive passenger flow pressure. The goal of minimizing passenger waiting time is to reduce the time passengers wait at stations, especially during high passenger flow periods. The density distribution value limit of key pressure points is to avoid excessive crowding in certain areas and ensure the safety of passengers at stations. Before using genetic algorithms for solving, individuals must first be encoded. Here, individuals are defined as a complete train scheduling plan. The genes of each individual represent different scheduling decisions, including train running intervals, stop sites, stop times, and running speeds. A completed individual represents a complete scheduling plan. The fitness function constructed is ,in, , and They are train delay penalty, passenger waiting time penalty and density distribution value penalty respectively, and r1, r2 and r3 are the corresponding weights respectively.

[0083] The population in the genetic algorithm includes multiple individuals, which is a collection of multiple scheduling solutions. Finally, the genetic algorithm performs population initialization, crossover, mutation and other operations to perform iterative calculations, and the optimal individual finally obtained is used as the optimal scheduling solution.

[0084] In particular, during high passenger flow periods, the priority of passenger flow balance scheduling is more important, and the weight of r3 can be increased. In scenarios where train punctuality is highly required, the weight of r1 can be increased.

[0085] In this embodiment, constructing the train operation mode includes the following steps:

[0086] Fluctuation characteristics include fluctuation range and fluctuation severity, and delay characteristics include average delay time and delay frequency.

[0087] Based on the delay time, a delay time series is constructed for each station. The fluctuation range is the difference between the maximum and minimum values ​​of the delay time series, the fluctuation severity is the standard deviation of the delay time series, the average delay time is the average value of the delay time series, and the delay frequency is the ratio of the number of delays in the delay time series to the total amount of data. A sliding window calculation is performed on the delay time series based on a sliding window of a preset scale to identify the changing trend of the delay time.

[0088] Taking the fluctuation characteristics and delay characteristics as clustering features, clustering analysis is performed using clustering algorithm, and various train delay modes are divided based on the clustering results.

[0089] For example, the train delay time at station A is 20S, 30S, 10S, 40S, and 30S respectively. The fluctuation range of the delay time is 30S, the fluctuation intensity is 10.4S, the average delay time is 26S, and the delay frequency is 100%. If the sliding window is set to 3, the first window is [20,30,10], the second window is [30,10,40], and the third window is [10,40,30]. The corresponding standard deviations are 8.16, 12.47, and 12.47 respectively, indicating that the delay time gradually increases and then remains stable. The delay change trend is encoded as [1,0].

[0090] This embodiment uses the DBSCAN algorithm for clustering. The specific clustering process is a prior art and will not be introduced here.

[0091] like Figure 2 As shown, the present application provides an intelligent auxiliary dispatching system for urban rail transit, which is used to implement the above-mentioned intelligent auxiliary dispatching method for urban rail transit, and the system includes:

[0092] The acquisition module obtains the train operation data and passenger flow data of each station. The train operation data includes the delay time of the train at each station.

[0093] The calculation module performs fluctuation analysis on the train operation data to obtain the fluctuation characteristics, time delay characteristics and trend characteristics of train delays, and constructs a train delay model based on the fluctuation characteristics, time delay characteristics and trend characteristics.

[0094] The analysis module performs spatial distribution analysis on passenger flow data to obtain a crowd density map, performs mobility analysis on the crowd density map to determine passenger flow concentration points, performs correlation analysis on train delay patterns and passenger flow concentration points, and determines the key pressure points in passenger flow concentration points that cause train delay patterns.

[0095] The optimization module constructs multiple optimization parameters, predicts the crowd density at the key pressure points of each station, obtains the density distribution value, and generates the best scheduling plan for each line based on the density distribution value, optimization parameters and corresponding scheduling priorities.

[0096] The present application provides a computer-readable storage medium having instructions stored thereon, wherein when the instructions are executed by a processor, the above-mentioned intelligent auxiliary scheduling method for urban rail transit is implemented.

[0097] It should be understood that the various technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the various technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0098] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for intelligent auxiliary dispatching of urban rail transit, characterized in that: Acquire train operation data and passenger flow data of each station, wherein the train operation data includes the delay time of the train at each station; Performing fluctuation analysis on the train operation data to obtain fluctuation characteristics, time delay characteristics and trend characteristics of train delays, and constructing a train delay model based on the fluctuation characteristics, the time delay characteristics and the trend characteristics; Performing spatial distribution analysis on the passenger flow data to obtain a crowd density map, and performing mobility analysis on the crowd density map to determine passenger flow concentration points; Performing correlation analysis on the train delay pattern and the passenger flow concentration point to determine the key pressure point in the passenger flow concentration point that causes the train delay pattern; Constructing multiple optimization parameters, predicting the crowd density at the key pressure points of each station, obtaining density distribution values, and generating the best scheduling plan for each line based on the density distribution values, the optimization parameters and the corresponding scheduling priorities; Determining the critical pressure points comprises the following steps: constructing a training set based on the data of the passenger flow concentration point and the train delay pattern including the delay time, constructing a delay prediction model based on the training set, performing contribution analysis on each of the passenger flow concentration points based on the delay prediction model, obtaining a SHAP value of each of the passenger flow concentration points, and locating the passenger flow concentration point with a SHAP value greater than a fourth threshold as the key pressure point; Constructing the train delay model includes the following steps: The fluctuation characteristics include fluctuation range and fluctuation severity, and the delay characteristics include average delay time and delay frequency; Based on the delay time, a delay time series for each station is constructed, wherein the fluctuation range is the difference between the maximum and minimum values ​​of the delay time series, the fluctuation severity is the standard deviation of the delay time series, the average delay time is the average value of the delay time series, and the delay frequency is the ratio of the number of delays in the delay time series to the total amount of data. A sliding window calculation is performed on the delay time series based on a sliding window of a preset scale to identify the change trend of the delay time; The fluctuation characteristics and the time delay characteristics are used as clustering characteristics, a clustering algorithm is used to perform clustering analysis, and a plurality of train delay modes are divided based on the clustering results.

2. The method according to claim 1, characterized in that Predicting the density distribution value comprises the following steps: Determine a forecast time period, extract historical passenger flow data of the same time period as the forecast time period from the passenger flow data, generate a first time series corresponding to each day based on the historical passenger flow data, extract the first time series N days before the forecast date as a second time series, combine all the second time series into a current sequence group, select a historical sequence group with the highest similarity to the current sequence group from the first time series, and compare the current sequence group with the historical sequence group to obtain a floating value; The first time series after the historical series group is acquired as a fourth time series, and the fourth time series is corrected based on the floating value to obtain the density distribution value of the predicted date.

3. The method according to claim 2, characterized in that Correcting the fourth time series comprises the following steps: Calculate the first average value of each first time series in the historical sequence group, calculate the second average value of each second time series in the current sequence group, calculate the ratio of the first average value to the second average value at corresponding positions, take the summed average value of the ratios as the floating value, and complete the correction process after each value in the fourth time series is expanded by the floating value.

4. The method according to claim 1, characterized in that: Determining the passenger flow concentration point includes the following steps: Divide the station into multiple spatial areas, construct a spatial topology map based on the spatial areas, the spatial topology map includes nodes and edges, the nodes represent the spatial areas, and the edges represent the connections between the spatial areas, obtain the average crowd density of each node to generate the crowd density map, and define the node whose average crowd density is greater than a first threshold as an initial pressure point; Perform liquidity analysis on the initial pressure point to obtain the liquidity of other nodes and each of the initial pressure points, calculate the clustering of nodes based on the liquidity, screen the nodes with clustering greater than a second threshold as focus points, calculate the degree centrality, closeness centrality, and betweenness centrality of the focus points, calculate the pressure value of the focus point based on the clustering and centrality values, and define the initial pressure point and the focus point with the pressure value greater than a third threshold as the passenger flow concentration point.

5. The method according to claim 4, characterized in that Calculating the liquidity includes the following steps: Setting a plurality of sub-time periods, obtaining the number of flow passengers from each of the initial pressure points toward each node in each of the sub-time periods, and the original number of passengers at the initial pressure point in each of the sub-time periods, constructing flow sequence data and original sequence data based on the number of flow passengers and the original number of passengers, analyzing the flow sequence data and the original sequence data based on the Bayesian algorithm, and obtaining a flow transfer coefficient; Perform flow velocity analysis and flow pressure analysis on the initial pressure point to obtain the flow efficiency factor and spatial congestion factor of the initial pressure point and each node, and calculate the fluidity between the initial pressure point and each node based on the flow efficiency factor, the spatial congestion factor and the flow transfer coefficient.

6. The method according to claim 1, characterized in that Generating the optimal scheduling solution includes the following steps: Based on the optimization parameters and the density distribution values, multiple objective functions are constructed, the objective functions include minimizing train delays, minimizing passenger waiting time and the density distribution value limits of the key pressure points, individual codes are constructed, and the individual codes include train running intervals, stop sites, stop times, and running speeds. A fitness function is constructed based on the objective function, and the corresponding objective function weights are assigned according to the scheduling priorities of the optimization parameters. The fitness function is solved based on a genetic algorithm to obtain the optimal scheduling solution.

7. An intelligent auxiliary dispatching system for urban rail transit, used to implement the intelligent auxiliary dispatching method for urban rail transit as described in any one of claims 1 to 6, characterized in that: A collection module, which obtains train operation data and passenger flow data of each station, wherein the train operation data includes the delay time of the train at each station; a calculation module, performing fluctuation analysis on the train operation data to obtain fluctuation characteristics, time delay characteristics and trend characteristics of train delays, constructing a train delay mode based on the fluctuation characteristics, time delay characteristics and trend characteristics, wherein the fluctuation characteristics include fluctuation range and fluctuation severity, the time delay characteristics include average delay time and delay frequency, constructing a delay time series for each station based on the delay time, the fluctuation range is the difference between the maximum and minimum values ​​of the delay time series, the fluctuation severity is the standard deviation of the delay time series, the average delay time is the average value of the delay time series, the delay frequency is the ratio of the number of delays in the delay time series to the total amount of data, performing sliding window calculation on the delay time series based on a sliding window of a preset scale, identifying a changing trend of the delay time, taking the fluctuation characteristics and the time delay characteristics as clustering characteristics, using a clustering algorithm to perform clustering analysis, and dividing a plurality of the train delay modes based on the clustering results; An analysis module is provided for performing spatial distribution analysis on the passenger flow data to obtain a crowd density map, performing mobility analysis on the crowd density map to determine passenger flow concentration points, performing correlation analysis on the train delay pattern and the passenger flow concentration points, and determining key pressure points in the passenger flow concentration points that cause the train delay pattern, wherein a training set is constructed based on the data of the passenger flow concentration points and the train delay pattern including the delay time, a delay prediction model is constructed based on the training set, a contribution analysis is performed on each passenger flow concentration point based on the delay prediction model, a SHAP value of each passenger flow concentration point is obtained, and the passenger flow concentration point having the SHAP value greater than a fourth threshold is positioned as the key pressure point; The optimization module constructs multiple optimization parameters, predicts the crowd density of the key pressure points of each station, obtains the density distribution value, and generates the best scheduling plan for each line based on the density distribution value, the optimization parameters and the corresponding scheduling priority.

8. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instruction is executed by the processor, the method for intelligent auxiliary scheduling of urban rail transit as described in any one of claims 1 to 6 is implemented.

Images