Method for analyzing tourist rail transit passenger flow characteristics based on mobile phone signaling data

Abstract

The application discloses a tourist rail transit passenger flow feature analysis method based on mobile phone signaling data, which comprises the following steps: S1, collecting and cleaning multi-operator mobile phone signaling data to obtain cleaned mobile phone signaling data; S2, spatially registering and reconstructing trajectories according to user identifiers and time stamps to form user trajectory sequences; S3, counting passenger flow indexes, constructing a spatial weight matrix and calculating a spatial autocorrelation factor; S4, introducing the spatial factor into a hidden Markov model and training parameters; S5, completing missing trajectories to generate complete trip state sequences; and S6, analyzing passenger flow indexes, identifying aggregation zones, abnormal zones and diffusion paths and outputting monitoring information. The spatial autocorrelation factor is introduced into the hidden Markov model to model spatial dependence and complete states of rail transit passenger flow based on mobile phone signaling data, so that accurate identification and monitoring of passenger flow distribution, hot regions and abnormal events are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of rail transit passenger flow analysis, and particularly relates to a tourist rail transit passenger flow feature analysis method based on mobile phone signaling data. BACKGROUND

[0002] In the field of existing rail transit passenger flow analysis and state inference, the commonly used method is to statistically analyze passenger flow, stay number and transfer ratio and other indicators based on mobile phone signaling data, card swiping records or video monitoring information, and to model and infer rail transit travel states by combining with a hidden Markov model. This kind of method usually relies on the integrity of trajectory data, and when there are missing or discontinuities in the user trajectory, the state recognition accuracy of the model will obviously decrease. At the same time, in the calculation process of state transition probability of the traditional hidden Markov model, it is often assumed that each spatial unit is independent of each other, lacking effective depiction of the spatial adjacency and passenger flow spatial dependence in the rail transit network, resulting in certain deviation in passenger flow distribution and hotspot identification.

[0003] On the other hand, in the prior art, the identification of spatial aggregation and abnormal passenger flow mostly adopts post hoc statistical analysis or clustering algorithm, lacking the ability to directly integrate spatial autocorrelation factors into the state inference and missing completion process. This makes the model unable to fully utilize information such as spatial weight matrix for accurate inference when facing complex spatial structures, temporary abnormal events or passenger flow diffusion phenomena, thereby affecting the timeliness and accuracy of rail transit passenger flow dynamic monitoring and abnormal detection. The existing method lacks systematic processing in aspects such as multi-operator data fusion, trajectory spatial registration and state completion under multiple time scales, resulting in insufficient stability and explainability of passenger flow spatial distribution analysis results.

[0004] Therefore, how to provide a tourist rail transit passenger flow feature analysis method based on mobile phone signaling data is a problem that those skilled in the art need to solve. SUMMARY

[0005] One object of the present application is to provide a tourist rail transit passenger flow feature analysis method based on mobile phone signaling data. The present application fully utilizes multi-operator mobile phone signaling data, rail transit line geographic information and spatial autocorrelation analysis technology, realizes accurate completion of missing trajectories and high-precision inference of rail travel state sequences by trajectory spatial registration, passenger flow index statistics, spatial weight matrix construction and direct introduction of spatial autocorrelation factors into the state transition probability calculation of the hidden Markov model. The present application not only can accurately identify spatial aggregation zones, abnormal passenger flow zones and diffusion paths, but also can dynamically monitor rail transit passenger flow distribution and hotspot changes, having the advantages of high timeliness, strong spatial dependence modeling capability and high passenger flow abnormality detection accuracy.

[0006] The method for analyzing tourist rail transit passenger flow characteristics based on mobile phone signaling data according to the embodiment of the present application comprises the following steps:

[0007] S1, collecting mobile phone signaling original data of multiple operators, pre-processing the mobile phone signaling original data to obtain cleaned mobile phone signaling data;

[0008] S2, based on the cleaned mobile phone signaling data, combining with rail transit line geographic information, performing spatial registration on signaling points, and reconstructing trajectories according to user unique identifiers and time stamps to form user trajectory sequences;

[0009] S3, based on the user trajectory sequences, preliminarily counting the in-out passenger flow, the number of stay and the transfer ratio of each rail transit station in each time period, and constructing a spatial weight matrix according to the physical connection relationship and the adjacency relationship of the rail transit stations to calculate a spatial autocorrelation factor;

[0010] S4, introducing the spatial autocorrelation factor into the calculation of the state transition probability of the hidden Markov model to establish a spatial dependence improved hidden Markov model, and training the initial probability, the spatial enhanced state transition probability and the observation probability parameters of the hidden Markov model;

[0011] S5, using the hidden Markov model to complete the state of the user trajectory sequence with missing or discontinuous in the user trajectory sequence to output the complete rail travel state sequence of the user and the time label of each state;

[0012] S6, based on the complete rail travel state sequence, counting the passenger flow indicators of each spatial unit according to the rail transit stations and the time periods, applying spatial autocorrelation analysis to the passenger flow indicators of each spatial unit, identifying spatial agglomeration areas, abnormal passenger flow areas and spatial diffusion paths, and outputting rail transit passenger flow distribution, hot area and abnormal event monitoring information.

[0013] Optionally, the mobile phone signaling original data specifically comprises a user unique identifier, a base station location identifier, a switching event type, an event occurrence time stamp and related network signaling parameters.

[0014] Optionally, the pre-processing of the mobile phone signaling original data specifically comprises outlier rejection, data deduplication, invalid signaling point filtering and time and space consistency verification.

[0015] Optionally, S2 specifically comprises:

[0016] S21, from the obtained cleaned mobile phone signaling data, extracting the user unique identifier of the signaling point, the occurrence time of the signaling point and the geographic position information of the signaling point for each signaling data;

[0017] S22, compare the geographic location information of each signaling point with the location information of each rail transit station in the rail transit line geographic information, and assign each signaling point to the nearest rail transit station;

[0018] S23, merge all signaling points belonging to the same user together according to the user unique identifier, and sort all signaling points of the user in chronological order to form a time sequence trajectory point set of the user in the rail transit network;

[0019] S24, for each user's time sequence trajectory point set, in combination with the physical connection relationship between rail transit stations, check the trajectory sequence, identify and remove abnormal jump points and abnormal trajectory segments that do not conform to the actual operation rules of rail transit in time or space, and retain continuous, real and effective rail transit travel trajectories;

[0020] S25, output all user trajectory sequences in the effective rail transit network formed after screening to obtain a user trajectory sequence set of all users in the rail transit network.

[0021] Optionally, the S3 specifically comprises:

[0022] S31, group all user trajectory sequences in the effective rail transit network of the output user according to rail transit stations and time periods, and count the number of passengers entering, leaving, staying and transferring in each station in each time period;

[0023] S32, according to the number of passengers entering, leaving, staying and transferring, preliminarily count the passenger flow, staying number and transfer ratio of each rail transit station in each time period to obtain the passenger flow index of each rail transit station in each time period;

[0024] S33, according to the physical structure information of the rail transit network, identify the adjacent stations of each rail transit station, construct the spatial connection relationship between the rail transit stations, and form a spatial weight matrix;

[0025] S34, in combination with the passenger flow index of each rail transit station in each time period and the spatial weight matrix, calculate the spatial autocorrelation factor of each spatial unit, specifically:

[0026] standardize the passenger flow index of each rail transit station in the same time period to obtain a passenger flow standardized vector, and eliminate the influence of passenger flow level difference between different stations;

[0027] In constructing the spatial weight matrix, in addition to considering the physical adjacency relationship of the stations, a dynamic weight adjustment mechanism based on real-time passenger flow similarity is introduced, that is, for non-physical adjacent stations with high passenger flow correlation, additional spatial weights are assigned according to the historical passenger flow correlation, so that the spatial weight matrix reflects both physical adjacency and passenger flow behavior adjacency;

[0028] Based on the standardized passenger flow indicators and the adjusted spatial weight matrix, a multi-scale local spatial autocorrelation factor calculation method is used to calculate the spatial autocorrelation factor in the directly adjacent stations, the passenger flow behavior similar stations and their combined neighborhoods for each spatial unit, and output the spatial autocorrelation analysis results of each spatial unit under different scales and different neighborhood structures.

[0029] S35, output the analysis results of the passenger flow indicators and the spatial autocorrelation factors of each rail transit station in each time period.

[0030] Optionally, the S4 specifically comprises:

[0031] S41, obtain the spatial autocorrelation factor and the user trajectory related data as the input of the construction of the hidden Markov model;

[0032] S42, construct an improved hidden Markov model including a multi-chain switching module, a state adaptive module and a multi-layer modeling module, introduce the spatial autocorrelation factor into the state transition probability calculation, take the spatial autocorrelation factor between any two state corresponding spatial units as an adjustment parameter, and dynamically adjust the state transition probability according to the value of the spatial autocorrelation factor;

[0033] S43, the multi-chain switching module is composed of a chain management unit, a feature analysis unit and a switching control unit, the input is a to-be-completed user trajectory segment and a behavior feature, the processing is to select or fuse multiple state chains to infer the trajectory state according to different trajectory types, and the output is a multi-chain fused trajectory state sequence;

[0034] S44, the state adaptive module is composed of a complexity detection unit, a spatial optimization unit and a structure adjustment unit, the input is a multi-chain fused trajectory state sequence, the processing is to dynamically increase or decrease the number of states and adjust the state transition structure according to the trajectory complexity and the observation distribution, and the output is an optimized state set and a transition matrix;

[0035] S45, the multi-layer modeling module is composed of a time window unit, a layer coordination unit and a nested modeling unit, the input is an optimized state set and a transition matrix, the trajectory data is segmented and organized according to different time scales, the trajectory state and the transition probability under each time scale are coordinated and fed back through hierarchical analysis and state aggregation, and the output is a multi-layer state transition probability and a multi-scale state sequence;

[0036] S46, parameter joint training is performed on the hidden Markov model, a Baum-Welch algorithm is used to optimize state transition probability, initial probability and observation probability parameters, and final hidden Markov model parameters are output;

[0037] S47, the hidden Markov model after structure innovation and parameter optimization is output, and has the abilities of spatial correlation dynamic adjustment, multi-chain fusion, state space self-adaptation and multi-scale nested expression;

[0038] S48, the hidden Markov model is applied to perform user trajectory state completion, spatio-temporal behavior feature analysis and passenger flow state sequence output.

[0039] Optionally, the S5 specifically comprises:

[0040] S51, the hidden Markov model is taken as an input model, and input is a user trajectory segment with missing or discontinuous in a user trajectory sequence;

[0041] S52, for each input user trajectory segment, an optimal state path is inferred for the missing or discontinuous segment by using a forward-backward algorithm according to state order and spatial unit features in the trajectory segment;

[0042] S53, in the state path inference process, the optimal state of each step is calculated by using state transition probability, initial probability and observation probability parameters and combining dynamic adjustment of the spatial autocorrelation factor;

[0043] S54, for each completed state sequence, a corresponding time label is assigned to generate a complete user orbit travel state sequence and a time label thereof;

[0044] S55, the completion process of all users is batch processed, and a complete orbit travel state sequence of all users and a time label set corresponding to each state are output.

[0045] Optionally, the S6 specifically comprises:

[0046] S61, all the complete orbit travel state sequences of users and time labels thereof are taken as input data, and user states are grouped and counted according to rail transit stations and time periods;

[0047] S62, for each rail transit station in each time period, the number of people entering the station, the number of people leaving the station, the number of people staying in the station and the number of people transferring are counted to generate a passenger flow index data set;

[0048] S63, the passenger flow index data set is taken as input to perform spatial autocorrelation analysis, spatial autocorrelation coefficients of each spatial unit in each time period are calculated based on a spatial weight matrix, and passenger flow distribution correlation between spatial units is analyzed;

[0049] S64, according to the spatial autocorrelation coefficient result, identify the spatial cluster area, abnormal passenger flow area and spatial diffusion path, and generate rail transit passenger flow distribution and hotspot area data in combination with passenger flow index;

[0050] S65, output the spatial autocorrelation analysis result of each spatial unit in different time periods, rail transit passenger flow distribution, hotspot area information and abnormal event monitoring information, provide decision data support for rail transit operation management and passenger flow scheduling.

[0051] The beneficial effects of the present application are:

[0052] The present application introduces a spatial dependence enhanced hidden Markov model, deeply fuses multi-operator mobile phone signaling data and rail transit line geographic information, realizes accurate spatial dependence modeling in track registration, spatial weight matrix construction and spatial autocorrelation analysis, and effectively overcomes the deficiency that the traditional track completion method cannot fully utilize spatial relationship information.

[0053] The present application can significantly improve the continuity and accuracy of state inference in missing or intermittent track completion, and realizes real-time identification of spatial cluster area, abnormal passenger flow area and diffusion path in passenger flow distribution monitoring, thereby providing more timely and accurate decision basis for rail transit operation management. Compared with the prior art, the present application not only improves the spatial accuracy and time response speed of passenger flow analysis, but also has good expansibility and adaptability, and can be widely applied to various scales and types of rail transit networks, and enhances the reliability and scientificity of passenger flow monitoring and abnormal event early warning. BRIEF DESCRIPTION OF DRAWINGS

[0054] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, and are used to explain the present application together with embodiments of the present application, and do not constitute a limitation on the present application. In the drawings:

[0055] Figure 1 Flow chart of the tourist rail transit passenger flow feature analysis method based on mobile phone signaling data proposed by the present application;

[0056] Figure 2 Structure diagram of the hidden Markov model introducing a spatial autocorrelation factor of the tourist rail transit passenger flow feature analysis method based on mobile phone signaling data proposed by the present application. DETAILED DESCRIPTION

[0057] The present application will now be further described in detail in conjunction with the drawings. These drawings are all simplified schematic diagrams, and only schematically show the basic structure of the present application, and therefore only show the components related to the present application.

[0058] REFERENCE Figure 1 and Figure 2The method for analyzing the passenger flow characteristics of tourist rail transit based on mobile phone signaling data comprises the following steps:

[0059] S1, collecting mobile phone signaling raw data of multiple operators, preprocessing the mobile phone signaling raw data, and obtaining cleaned mobile phone signaling data;

[0060] S2, based on the cleaned mobile phone signaling data, combining with the geographic information of the rail transit line, spatially registering the signaling points, and reconstructing the trajectories according to the unique user identifier and the time stamp to form a user trajectory sequence;

[0061] S3, based on the user trajectory sequence, preliminarily counting the in-out passenger flow, the number of stay and the transfer ratio of each rail transit station in each time period, and constructing a spatial weight matrix according to the physical connection relationship and the adjacency relationship of the rail transit stations, and calculating the spatial autocorrelation factor;

[0062] S4, introducing the spatial autocorrelation factor into the state transition probability calculation of the hidden Markov model, establishing a spatial dependence improved hidden Markov model, and training the initial probability, the spatial enhanced state transition probability and the observation probability parameters of the hidden Markov model;

[0063] S5, using the hidden Markov model to complete the state of the user trajectory sequence with missing or discontinuous in the user trajectory sequence, and outputting the complete rail travel state sequence of the user and the time label of each state;

[0064] S6, based on the complete rail travel state sequence, counting the passenger flow indicators of each spatial unit according to the rail transit stations and the time periods, applying spatial autocorrelation analysis to the passenger flow indicators of each spatial unit, identifying the spatial aggregation area, the abnormal passenger flow area and the spatial diffusion path, and outputting the rail transit passenger flow distribution, the hotspot area and the abnormal event monitoring information.

[0065] In the embodiment, the mobile phone signaling raw data specifically includes a user unique identifier, a base station location identifier, a handover event type, an event occurrence time stamp and related network signaling parameters.

[0066] In the embodiment, the preprocessing of the mobile phone signaling raw data specifically includes outlier rejection, data deduplication, invalid signaling point filtering and time and space consistency verification.

[0067] In the embodiment, S2 specifically includes:

[0068] S21, from the obtained cleaned mobile phone signaling data, extracting the user unique identifier of the signaling point, the occurrence time of the signaling point and the geographic location information of the signaling point for each signaling data;

[0069] S22, compare the geographic location information of each signaling point with the location information of each rail transit station in the rail transit line geographic information, and assign each signaling point to the nearest rail transit station;

[0070] S23, merge all signaling points belonging to the same user together according to the user unique identifier, and sort all signaling points of the user in chronological order to form a time sequence trajectory point set of the user in the rail transit network;

[0071] S24, for each user's time sequence trajectory point set, in combination with the physical connection relationship between rail transit stations, check the trajectory sequence, identify and remove abnormal jump points and abnormal trajectory segments that do not conform to the actual operation rules of rail transit in time or space, and retain continuous, real and effective rail transit travel trajectories;

[0072] S25, output all user trajectory sequences in the effective rail transit network formed after screening to obtain a user trajectory sequence set of all users in the rail transit network.

[0073] In the embodiment, S3 specifically includes:

[0074] S31, group all user trajectory sequences in the effective rail transit network of the output user according to rail transit stations and time periods, and count the number of passengers entering, leaving, staying and transferring in each station in each time period;

[0075] S32, according to the number of passengers entering, leaving, staying and transferring obtained by statistics, preliminarily count the passenger flow, staying number and transfer ratio of each rail transit station in each time period to obtain the passenger flow index of each rail transit station in each time period;

[0076] S33, according to the physical structure information of the rail transit network, identify the adjacent stations of each rail transit station, construct the spatial connection relationship between the rail transit stations, and form a spatial weight matrix;

[0077] S34, in combination with the passenger flow index of each rail transit station in each time period and the spatial weight matrix, calculate the spatial autocorrelation factor of each spatial unit, specifically:

[0078] Standardize the passenger flow index of each rail transit station in the same time period to obtain a passenger flow standardized vector, and eliminate the influence of passenger flow level difference between different stations;

[0079] In constructing the spatial weight matrix, in addition to considering the physical adjacency relationship of the stations, a dynamic weight adjustment mechanism based on real-time passenger flow similarity is introduced, that is, for non-physical adjacent stations with high passenger flow correlation, additional spatial weights are assigned according to the historical passenger flow correlation, so that the spatial weight matrix reflects both physical adjacency and passenger flow behavior adjacency;

[0080] Based on the standardized passenger flow indicators and the adjusted spatial weight matrix, a multi-scale local spatial autocorrelation factor calculation method is used to calculate the spatial autocorrelation factor in the directly adjacent stations, the passenger flow behavior similar stations and their combined neighborhoods for each spatial unit, and output the spatial autocorrelation analysis results of each spatial unit under different scales and different neighborhood structures.

[0081] S35, output the analysis results of the passenger flow indicators and the spatial autocorrelation factors of each rail transit station in each time period.

[0082] In the embodiment, the S4 specifically includes:

[0083] S41, obtain the spatial autocorrelation factor and the user trajectory related data as the input of the construction of the hidden Markov model;

[0084] S42, construct an improved hidden Markov model including a multi-chain switching module, a state adaptive module and a multi-layer modeling module, introduce the spatial autocorrelation factor into the state transition probability calculation, take the spatial autocorrelation factor between any two state corresponding spatial units as an adjustment parameter, and dynamically adjust the state transition probability according to the value of the spatial autocorrelation factor;

[0085] S43, the multi-chain switching module is composed of a chain management unit, a feature analysis unit and a switching control unit, the input is a to-be-completed user trajectory segment and a behavior feature, the processing is to select or fuse multiple state chains to infer the trajectory state for different trajectory types, and the output is a multi-chain fused trajectory state sequence;

[0086] S44, the state adaptive module is composed of a complexity detection unit, a spatial optimization unit and a structure adjustment unit, the input is a multi-chain fused trajectory state sequence, the processing is to dynamically increase or decrease the number of states and adjust the state transition structure according to the trajectory complexity and the observation distribution, and the output is an optimized state set and a transition matrix;

[0087] S45, the multi-layer modeling module is composed of a time window unit, a layer coordination unit and a nested modeling unit, the input is an optimized state set and a transition matrix, the trajectory data is segmented and organized according to different time scales, the trajectory state and the transition probability under each time scale are coordinated and fed back through hierarchical analysis and state aggregation, and the output is a multi-layer state transition probability and a multi-scale state sequence;

[0088] S46, the parameters of the hidden Markov model are jointly trained, the Baum-Welch algorithm is used to optimize the state transition probability, the initial probability and the observation probability parameters, and the final hidden Markov model parameters are outputted.

[0089] Under the condition of the current hidden Markov model parameters and the observation sequence, the forward and backward probability recursive method is used to solve the conditional probability distribution of each time step in different hidden states and the joint probability distribution of state transition between adjacent time steps.

[0090] According to the state probability and state transition joint probability obtained in the expectation calculation stage, the initial state probability vector, the state transition probability matrix and the observation probability distribution function of the hidden Markov model are jointly re-estimated. The spatial autocorrelation factor is introduced as a weighted adjustment term in the update of the state transition probability, which enhances the ability of the spatial proximity to describe the transition relationship.

[0091] Under the condition of ensuring that the data likelihood value is monotonically increasing, the expectation calculation and parameter re-estimation process are alternately performed until the parameter variation is less than the preset convergence threshold or the maximum iteration number is reached, and the final hidden Markov model parameter set after convergence is outputted.

[0092] S47, the hidden Markov model after structure innovation and parameter optimization is outputted, which has the abilities of spatial correlation dynamic adjustment, multi-chain fusion, state space adaptation and multi-scale nested expression.

[0093] S48, the hidden Markov model is applied to perform user trajectory state completion, spatio-temporal behavior feature analysis and passenger flow state sequence output.

[0094] In the embodiment, the S5 specifically includes:

[0095] S51, the hidden Markov model is taken as an input model, and the input is a user trajectory segment with missing or discontinuous in the user trajectory sequence;

[0096] S52, for each input user trajectory segment, the forward-backward algorithm is used to perform optimal state path inference on the missing or discontinuous segment according to the state order and spatial unit features in the trajectory segment, and the forward-backward algorithm is used to perform optimal state path inference on the missing or discontinuous segment, and the forward-backward algorithm is used to perform optimal state path inference on the missing or discontinuous segment.

[0097] Based on the known parameter set of the hidden Markov model, the cumulative probability value of each time step in each possible state is recursively calculated from the starting state of the trajectory segment in time sequence, and the state transition process is weighted and corrected combined with the spatial unit feature vector, so as to reflect the influence of spatial dependence on path inference;

[0098] Starting from the end state of the trajectory segment, the conditional probability distribution of the transition to the observation state at each time step of the missing segment is calculated in reverse recursion and associated with the corresponding spatial unit characteristics to ensure that the state backtracking process is consistent with the spatial proximity characteristics.

[0099] The forward recursion probability and the backward backtracking probability are normalized and fused at each time step of the missing segment to obtain the posterior probability distribution of each candidate state. The state with the highest posterior probability is selected as the optimal state label, thereby forming a complete state path for the missing or discontinuous segment and realizing the continuous reconstruction of the user trajectory.

[0100] S53. In the state path inference process, the optimal state at each step is calculated by using the state transition probability, initial probability and observation probability parameters, combined with the dynamic adjustment of the spatial autocorrelation factor. Specifically, the calculation of the optimal state at each step means that at each time step, the state transition probability, initial probability and observation probability parameters of the Hidden Markov Model are used as the basis. The cumulative probability of the state at the previous time step is weighted with the transition probability of each candidate state at the current time step. The transition probability is then dynamically adjusted by combining the spatial autocorrelation factor of the corresponding spatial unit at that time step. The adjusted transition probability is multiplied with the current observation probability and normalized to obtain the posterior probability of each candidate state. Finally, the state with the largest posterior probability value is selected as the optimal state at that time step.

[0101] S54. For each completed state sequence, assign a corresponding time label to generate a complete user track travel state sequence and its time label.

[0102] S55. Perform batch processing on the completion process for all users, and output the complete track travel status sequence for all users and the set of time tags corresponding to each status.

[0103] In this embodiment, S6 specifically includes:

[0104] S61. Using the complete sequence of all users’ rail travel statuses and their time labels as input data, group and statistically analyze the user statuses according to rail transit stations and time periods.

[0105] S62. For each rail transit station, count the number of people entering the station, exiting the station, staying in the station, and transferring for each time period, and generate a passenger flow index dataset.

[0106] S63. Using the passenger flow index dataset as input, perform spatial autocorrelation analysis, calculate the spatial autocorrelation coefficient of each spatial unit in each time period based on the spatial weight matrix, and analyze the correlation of passenger flow distribution among spatial units.

[0107] S64, according to the spatial autocorrelation coefficient result, identify the spatial aggregation area, abnormal passenger flow area and spatial diffusion path, and generate rail transit passenger flow distribution and hotspot area data in combination with passenger flow indicators;

[0108] S65, output the spatial autocorrelation analysis result of each spatial unit in different time periods, rail transit passenger flow distribution, hotspot area information and abnormal event monitoring information, provide decision data support for rail transit operation management and passenger flow scheduling.

[0109] Embodiment 1:

[0110] In order to verify the feasibility of the application in implementation, the application is applied to holiday peak passenger flow monitoring, missing trajectory completion and abnormal early warning scene of a large urban rail transit hub and its adjacent stations. The hub station connects two main lines and is an important gathering and distributing place for commuting peak and holiday tourism peak. The passenger flow characteristics are obviously affected by adjacent stations. The traditional monitoring method only relies on the entrance and exit data of the station itself, and lacks consideration of the spatial adjacent relationship, resulting in late warning when encountering sudden passenger flow and low utilization rate of scheduling resources.

[0111] In the data collection stage, continuous seven-day mobile phone signaling data is obtained from multiple operators, covering the target station and its four adjacent stations. The collection frequency is 15 minutes once, and the total amount of original data is about 120 million. After removing abnormal positioning points and completing broken trajectory through data preprocessing, the trajectory reconstruction data set is formed, and the effective trajectory is about 950,000 per day. According to the trajectory data, the station's time-periodic entrance, exit, transfer ratio and stay time are calculated, and the spatial weight matrix based on physical connection relationship and adjacency relationship is constructed, and the spatial autocorrelation factor between stations is calculated.

[0112] The spatial autocorrelation factor is introduced into the state transition probability calculation of the hidden Markov model to complete the user trajectory with missing or discontinuous, and the complete rail travel state sequence and its time label are obtained. Based on the complete state sequence, the passenger flow indicators of each spatial unit are counted according to time period, and spatial autocorrelation analysis is carried out to identify spatial aggregation area and abnormal passenger flow area. When it is detected that the passenger flow correlation between the hub station and its east adjacent station significantly increases at 14:30 of a day (Moran index increases from 0.45 to 0.78), and the entrance quantity of the hub station is 31.5% higher than the predicted value, the system sends an early warning at 14:35, the operator completes the additional manpower and opens additional gates at 14:45, and then the average queuing time within 30 minutes decreases by 20.8% compared with that before the warning.

[0113] The verification results show that the method has an average advance of 21 minutes in abnormal passenger flow detection, an increase of 13 minutes over the original method, an abnormal recognition accuracy of 93.2%, an increase from 84.6%, and a false positive rate of 4.3%, a decrease from 8.1%. In particular, in the multi-station linkage type anomaly, the spatial dependence analysis significantly reduces the scheduling misjudgment caused by single-station fluctuations, enabling the operator to more efficiently allocate resources and service resources.

[0114] Table 1 Comparison of passenger flow monitoring and prediction deviation between hub station and adjacent station in seven-day period

[0115]

[0116]

[0117] From the data in Table 1, the sample covers multiple key periods in seven days and three types of stations (hub station, adjacent station A, and adjacent station B). Except for individual periods, the actual inbound volume is generally higher than the predicted value, with a deviation rate falling within the 3%-9% interval, and the corresponding Moran index is between 0.50 and 0.66, showing moderate spatial positive correlation but not triggering an early warning, indicating that these periods are normal fluctuations: for example, on the first day of the morning peak, the deviation rate of the hub station is only 3.52%, and the Moran index is 0.50; on the fourth day of the evening peak, the deviation is 8.55%, and the Moran index is 0.64; on the sixth and seventh days, the deviations are 9.35% and 9.16%, respectively, and the Moran indexes are 0.66 and 0.65, respectively, without triggering an early warning and with a "congestion index improvement rate" of "--", indicating that the system determines that it is in a controllable state and does not require linkage treatment.

[0118] On the second day from 14:30 to 14:45, a significant anomaly occurred, with the hub station deviation rate soaring to 31.50%, and the adjacent station A simultaneously experiencing a high deviation of 27.02%, both with a Moran index of 0.78, indicating strong spatial clustering. This combination of "high deviation + high Moran index" triggered a simultaneous early warning at both stations, and after treatment, the congestion index improvement rates were observed to be 20.8% and 19.3%, respectively. This phenomenon indicates that after embedding the spatial autocorrelation factor into the improved hidden Markov structure, the invention can identify multi-station linkage diffusion-type passenger flow shocks, rather than just single-station isolated increases; the treatment effect after the early warning can be quantitatively reflected in a short time window, verifying the effectiveness of the method in practical operations.

[0119] From the perspective of mechanism correlation, although the period without triggering early warning has positive spatial correlation, the correlation and deviation amplitude are at a medium level, the system does not give action suggestions, and the false positives and over-reactions are reduced; when the spatial correlation and deviation are amplified synchronously, the system timely outputs the linkage early warning and brings about about 20% congestion improvement, which shows that the threshold strategy with the double conditions of "deviation amplitude + spatial autocorrelation" can better depict the real risk scenario. Combined with these results, it can be inferred that the triggering threshold can be further refined into the joint criterion of "deviation rate threshold x Morant index threshold", and can be adaptively adjusted according to the site type and period, so as to keep a low false positive rate while having a larger advance, and further improve the quality of passenger flow control during the holiday and activity period.

[0120] The above merely describes preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can make equivalent replacements or changes to the technical solutions and inventive concepts of the present application within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for analyzing passenger flow characteristics of tourism rail transit based on mobile phone signaling data, characterized in that, Includes the following steps: S1. Collect raw mobile signaling data from multiple operators, preprocess the raw mobile signaling data, and obtain cleaned mobile signaling data. S2. Based on the cleaned mobile phone signaling data and combined with the geographical information of the rail transit line, the signaling points are spatially registered, and the trajectory is reconstructed according to the user's unique identifier and timestamp to form a user trajectory sequence. S3. Based on user trajectory sequences, preliminary statistics are made on the passenger flow, number of people staying and transfer ratio of each rail transit station in each time period. Based on the physical connection and adjacency relationship of rail transit stations, a spatial weight matrix is ​​constructed and the spatial autocorrelation factor is calculated. S4. Introduce the spatial autocorrelation factor into the calculation of the state transition probability of the Hidden Markov Model, establish a spatially dependent improved Hidden Markov Model, and train the initial probability, spatially enhanced state transition probability, and observation probability parameters of the Hidden Markov Model. S5. Using a Hidden Markov Model, state completion is performed on user trajectory sequences that have missing or discontinuous states, and the complete user trajectory travel state sequence and time label of each state are output. S6. Based on the complete rail travel status sequence, the passenger flow indicators of each spatial unit are statistically analyzed according to rail transit stations and time periods. Spatial autocorrelation analysis is applied to the passenger flow indicators of each spatial unit to identify spatial agglomeration areas, abnormal passenger flow areas and spatial diffusion paths, and output rail transit passenger flow distribution, hot spot areas and abnormal event monitoring information.

2. The method for analyzing passenger flow characteristics of tourism rail transit based on mobile phone signaling data according to claim 1, characterized in that, The raw mobile signaling data specifically includes the user's unique identifier, base station location identifier, handover event type, event timestamp, and related network signaling parameters.

3. The method for analyzing passenger flow characteristics of tourism rail transit based on mobile phone signaling data according to claim 1, characterized in that, The preprocessing of the raw mobile signaling data specifically includes outlier removal, data deduplication, invalid signaling point filtering, and time and space consistency verification.

4. The method for analyzing passenger flow characteristics of rail transit based on mobile phone signaling data according to claim 1, characterized in that, S2 specifically includes: S21. From the cleaned mobile phone signaling data, extract the user's unique identifier, the time of occurrence of the signaling point, and the geographical location information of the signaling point for each signaling data. S22. Compare the geographical location information of each signaling point with the location information of each rail transit station in the geographical information of the rail transit line, and assign each signaling point to the nearest rail transit station. S23. Using the user's unique identifier as an index, all signaling points belonging to the same user are merged together, and all signaling points of the user are sorted in chronological order to form a time-series trajectory point set of the user in the rail transit network. S24. For each user's time-series trajectory point set, combined with the physical connection relationship between rail transit stations, check the trajectory sequence, identify and remove abnormal jump points and abnormal trajectory segments that do not conform to the actual operation rules of rail transit in time or space, and retain continuous, real and valid rail transit travel trajectories. S25. Output the valid user trajectory sequences within the rail transit network formed after screening and processing of all users to obtain the set of user trajectory sequences for all users within the rail transit network.

5. The method for analyzing passenger flow characteristics of rail transit based on mobile phone signaling data according to claim 1, characterized in that, S3 specifically includes: S31. For the set of user trajectory sequences within the valid rail transit network of all users, group them according to rail transit stations and time periods, and count the number of people entering the station, exiting the station, staying in the station, and transferring at each station in each time period. S32. Based on the statistics of the number of people entering the station, the number of people exiting the station, the number of people staying, and the number of people transferring, conduct preliminary statistics on the passenger flow, number of people staying, and transfer ratio of each rail transit station in each time period to obtain the passenger flow index of each rail transit station in each time period. S33. Based on the physical structure information of the rail transit network, identify the adjacent stations of each rail transit station, construct the spatial connection relationship between rail transit stations, and form a spatial weight matrix; S34. Combining the passenger flow indicators and spatial weight matrix of each rail transit station in each time period, calculate the spatial autocorrelation factor of each spatial unit, specifically as follows: Standardize the passenger flow indicators of each rail transit station within the same time period to obtain a standardized passenger flow vector, thereby eliminating the impact of differences in passenger flow levels between different stations. When constructing the spatial weight matrix, in addition to considering the physical adjacency of the stations, a dynamic weight adjustment mechanism based on real-time passenger flow similarity is introduced. That is, for non-physical adjacent stations with highly correlated passenger flow indicators, additional spatial weights are allocated according to the historical passenger flow correlation, so that the spatial weight matrix reflects both physical adjacency and passenger flow behavioral adjacency. Based on standardized passenger flow indicators and adjusted spatial weight matrix, a multi-scale local spatial autocorrelation factor calculation method is adopted to calculate the spatial autocorrelation factor of each spatial unit in the neighborhood of directly adjacent stations, stations with similar passenger flow behavior, and their combined neighborhoods. The spatial autocorrelation analysis results of each spatial unit under different scales and different neighborhood structures are output. S35. Output the analysis results including passenger flow indicators and spatial autocorrelation factors for each rail transit station in each time period.

6. The method for analyzing passenger flow characteristics of tourism rail transit based on mobile phone signaling data according to claim 1, characterized in that, S4 specifically includes: S41. Obtain spatial autocorrelation factors and user trajectory-related data as inputs for constructing the Hidden Markov Model; S42. Construct an improved Hidden Markov Model that includes a multi-chain switching module, a state adaptive module, and a multi-layer modeling module. Introduce the spatial autocorrelation factor into the calculation of the state transition probability. Use the spatial autocorrelation factor between any two spatial units corresponding to the state as an adjustment parameter. Dynamically adjust the state transition probability based on the value of the spatial autocorrelation factor. S43. The multi-chain switching module consists of a chain management unit, a feature analysis unit, and a switching control unit. The input is the user trajectory segment to be completed and the behavioral features. The processing is to select or fuse multiple state chains for different trajectory types to infer the trajectory state, and the output is the trajectory state sequence fused by the multi-chain. S44. The state adaptive module consists of a complexity detection unit, a space optimization unit, and a structure adjustment unit. The input is a multi-chain fused trajectory state sequence. The processing is to dynamically increase or decrease the number of states and adjust the state transition structure according to the trajectory complexity and observation distribution. The output is the optimized state set and transition matrix. S45. The multi-layer modeling module consists of time window units, layer coordination units, and nested modeling units. The input is the optimized state set and transition matrix. The trajectory data is segmented and organized according to different time scales. Through hierarchical analysis and state aggregation, the trajectory state and transition probability at each time scale are coordinated and fed back. The output is multi-layer state transition probability and multi-scale state sequence. S46. Jointly train the parameters of the Hidden Markov Model by using the Baum-Welch algorithm to optimize the state transition probability, initial probability and observation probability parameters, and output the final Hidden Markov Model parameters. S47. Output the Hidden Markov Model after structural innovation and parameter optimization, which has the ability to dynamically adjust spatial correlation, multi-chain fusion, state space adaptation and multi-scale nested expression. S48. Apply Hidden Markov Model to complete user trajectory status, analyze spatiotemporal behavioral features, and output passenger flow status sequence.

7. The method for analyzing passenger flow characteristics of tourism rail transit based on mobile phone signaling data according to claim 1, characterized in that, S5 specifically includes: S51. Use the Hidden Markov Model as the input model. The input is a user trajectory segment that is missing or discontinuous in the user trajectory sequence. S52. For each input user trajectory segment, based on the state order and spatial unit characteristics in the trajectory segment, the forward-backward algorithm is used to infer the optimal state path for missing or discontinuous segments. S53. In the process of state path inference, the optimal state at each step is calculated by using the state transition probability, initial probability and observation probability parameters, combined with the dynamic adjustment of the spatial autocorrelation factor. S54. For each completed state sequence, assign a corresponding time label to generate a complete user track travel state sequence and its time label. S55. Perform batch processing on the completion process for all users, and output the complete track travel status sequence for all users and the set of time tags corresponding to each status.

8. The method for analyzing passenger flow characteristics of rail transit based on mobile phone signaling data according to claim 1, characterized in that, S6 specifically includes: S61. Using the complete sequence of all users’ rail travel statuses and their time labels as input data, group and statistically analyze the user statuses according to rail transit stations and time periods. S62. For each rail transit station, count the number of people entering the station, exiting the station, staying in the station, and transferring for each time period, and generate a passenger flow index dataset. S63. Using the passenger flow index dataset as input, perform spatial autocorrelation analysis, calculate the spatial autocorrelation coefficient of each spatial unit in each time period based on the spatial weight matrix, and analyze the correlation of passenger flow distribution among spatial units. S64. Based on the spatial autocorrelation coefficient results, identify spatial clustering areas, abnormal passenger flow areas, and spatial diffusion paths, and combine them with passenger flow indicators to generate data on rail transit passenger flow distribution and hotspot areas. S65. Output the spatial autocorrelation analysis results of each spatial unit at different time periods, the distribution of rail transit passenger flow, information on hotspot areas and monitoring information of abnormal events, to provide decision-making data support for rail transit operation management and passenger flow scheduling.

Images