Multi-line metro passenger flow collaborative prediction and management system based on digital information transmission

Abstract

The application discloses a multi-line metro passenger flow cooperative prediction and management system based on digital information transmission, comprising: a signal field strength entropy sensing module, which generates a large passenger flow early warning signal by using the global synchronous decline of the spatial distribution entropy of wireless communication quality parameters; a train control information frame analysis module, which analyzes the train door action time sequence and operation parameters, and outputs the cross-line transfer passenger flow through a micro-scale mapping model; a dynamic guidance decision module, which generates a preferential price according to the comparison between the transfer passenger flow and the transport capacity, and pushes the guidance information through a differentiated transmission priority; a closed-loop feedback self-calibration module, which collects passenger response behavior and actual paths, and returns feedback signals to adjust the parameters of each module. The four modules are linked through the same digital information transmission infrastructure to form a sensing, tracking, guidance, and feedback closed loop. The application realizes advanced warning, precise tracking without privacy dependence, high response rate guidance, and system continuous evolution, and completely reuses the existing communication and signal facilities of the metro.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation and subway passenger flow management technology, and particularly relates to a multi-line subway passenger flow collaborative prediction and management system based on digital information transmission. Background Technology

[0002] Currently, passenger flow prediction and management systems in multi-line networked urban subway operations primarily rely on passenger card swiping data collected by the Automatic Fare Collection (AFC) system. Short-term passenger flow prediction is then performed using time series analysis, graph neural networks, or deep learning models. Some systems incorporate mobile phone signaling data, Wi-Fi probe connection counts, or train weighing data as supplementary sensing sources. For cross-line transfer tracking, existing technologies mostly employ OD (Original Departure / Outtake) back-calculation based on individual card swipe records or base station handover sequence matching based on mobile phone signaling. For passenger flow guidance, common solutions include pushing congestion alerts and route suggestions through station information display screens, onboard broadcasts, or mobile apps. In recent years, some research has attempted to utilize operational data from Automatic Train Operation (ATO) systems to assist in assessing train congestion levels. These technologies have been applied to some extent in subway operation management, forming a technical system primarily based on data collection, model prediction, and information dissemination.

[0003] However, the existing technologies still have the following shortcomings: First, AFC card swiping data has a natural delay of several minutes, making it impossible to anticipate upcoming large passenger flows, resulting in warnings lagging behind the speed of passenger flow formation. Second, transfer tracking based on individual identification data involves passenger privacy, and mobile phone signaling suffers from location drift and intermittent issues within tunnels, making it difficult to meet the requirements for refined prediction. Third, existing guidance methods are mostly one-way broadcasts or static suggestions, lacking dynamic optimization of passenger response rates, and unable to achieve real-time game-theoretic guidance tailored to each individual. Fourth, the prediction system and management system are independent of each other, lacking a closed-loop feedback mechanism, and cannot use actual guidance results to correct the prediction model and control strategies, causing system performance to degrade with extended operating time. Fifth, existing solutions generally neglect valuable data sources such as signal quality parameters and train control information frames contained in the subway's digital information transmission infrastructure, resulting in a waste of sensing resources. Summary of the Invention

[0004] To overcome the aforementioned shortcomings of existing technologies, this invention provides a multi-line subway passenger flow collaborative prediction and management system based on digital information transmission, which solves the problems of AFC perception delay, transfer tracking relying on privacy data, low guidance efficiency, lack of closed-loop self-calibration, and failure to utilize transmission infrastructure data in existing technologies.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A multi-line subway passenger flow collaborative prediction and management system based on digital information transmission includes: The signal field strength entropy sensing module is used to collect wireless communication quality parameters in subway platforms and tunnels in real time, calculate the spatial distribution entropy of signal quality in the platform area, and generate a large passenger flow warning signal when the spatial distribution entropy is detected to be synchronously decreasing across the entire area. The train control information frame parsing module is connected to the signal field strength entropy sensing module via a data bus or communication interface. It is used to parse the control information frame of the train that is about to arrive at the station after receiving the large passenger flow warning signal, extract the door action sequence and train operation parameters, and output the cross-line passenger flow through a micro-scale mapping model. The dynamic guidance decision module is connected to the train control information frame parsing module via a data bus or communication interface. It is used to generate a guidance discount price based on the comparison results of the cross-line transfer passenger flow and the real-time capacity saturation of each line, and push guidance information containing guidance path and discount price to the passenger terminal through the digital information transmission channel. The closed-loop feedback self-calibration module is connected to the signal field entropy sensing module, the train control information frame parsing module, and the dynamic guidance decision module via a data bus or communication interface, respectively. It is used to collect passenger terminal response behavior to guidance information and passenger actual travel path data, form feedback signals, and send the feedback signals back to the signal field entropy sensing module to adjust its judgment conditions for generating large passenger flow warning signals, back to the train control information frame parsing module to update the parameters of the micro-scale mapping model, and back to the dynamic guidance decision module to optimize the generation strategy of the guidance discount price. The signal field strength entropy sensing module, train control information frame parsing module, dynamic guidance decision-making module, and closed-loop feedback self-calibration module achieve data interaction through the same digital information transmission infrastructure, forming a closed loop of perception, tracking, guidance, and feedback.

[0006] Preferably, the conditions for the signal field strength entropy sensing module to determine the synchronous decrease across the entire area include: the decrease in spatial distribution entropy exceeds a first threshold within a preset time window, and the statistical standard deviation of the signal quality degradation rate of each monitoring grid within the station is less than a second threshold; the preset time window is 5 to 20 seconds, the first threshold is 0.2 to 0.5, and the second threshold is 0.03 to 0.08; and the large passenger flow warning signal is generated only when the wireless communication quality parameters of at least two different operator networks meet the conditions.

[0007] Preferably, the control information frames acquired by the train control information frame parsing module include train position report frames, door control frames, train automatic driving operation curve frames, and traction and braking status frames; the door action sequence includes door opening duration, door closing duration, and the number of times the door is opened and closed; the input features of the microscale mapping model also include the fluctuation amplitude of the rate of change of acceleration during the braking phase and the coefficient of variation of the braking force maintaining current during the parking period; the microscale mapping model is a machine learning model trained through supervised learning, and the machine learning model is updated through online incremental learning during operation.

[0008] Preferably, when the dynamic guidance decision module pushes guidance information through the digital information transmission channel, it allocates different transmission priorities based on the passenger's historical response rate and current location, so that the transmission delay of different passenger terminals receiving the guidance information is differentiated; the allocation of transmission priorities is achieved through 5G service quality flow identification or Wi-Fi air interface scheduling, wherein high priority queues are allocated to passenger terminals with historical response rates higher than a configurable response threshold and currently located in the transfer decision area.

[0009] Preferably, the dynamic guidance decision module further includes a bilateral auction submodule, which is used to use the guidance discount price as an auction offer in each guidance time window, and the acceptance or rejection decision returned by the passenger terminal as the bidding response. When the number of passengers accepting the offer exceeds the guidance quota allowed by the remaining capacity of the alternative route, the discount quota is allocated according to the passenger's historical credit score or the order of response arrival time.

[0010] Preferably, the closed-loop feedback self-calibration module adopts an online incremental learning algorithm. After each train stop event, the error between the actual number of passengers getting on and off the train and the predicted number of passengers getting on and off the train output by the microscale mapping model is used as the loss, and the model parameters are updated by stochastic gradient descent. When multiple consecutive large passenger flow warning signals are determined to be false alarms by the subsequent train information frame parsing results, the closed-loop feedback self-calibration module automatically raises the first threshold or the second threshold.

[0011] Preferably, it also includes a cross-line transfer passenger flow tracking submodule, which is connected to the train control information frame parsing module via a data bus or communication interface. This submodule is used to parse the control information frames of arriving trains on each line within the transfer station to obtain the number of passengers boarding and alighting on each line. Based on the flow conservation constraint and the prior historical transfer ratio, and combined with the real-time signal distribution information output by the signal field entropy sensing module, the transfer matrix is ​​solved to output the real-time transfer passenger flow between each line.

[0012] Preferably, the signal field strength entropy sensing module, train control information frame parsing module, and dynamic guidance decision-making module are deployed in the metro network control center, edge computing nodes, or cloud servers; the digital information transmission infrastructure includes the metro-dedicated wireless communication network, 5G public network slicing, and existing transmission channels of the train control signal system.

[0013] Preferably, the passenger's actual travel route data collected by the closed-loop feedback self-calibration module is obtained through passenger terminal application positioning, automatic fare collection system recording, or mobile phone signaling trajectory, and is de-identified during collection, including deleting mobile phone number and device identifier.

[0014] Preferably, the preset time window is 10 seconds, the first threshold is 0.3, and the second threshold is 0.05.

[0015] The technical effects and advantages of this invention, a multi-line subway passenger flow collaborative prediction and management system based on digital information transmission, are as follows: 1. This invention uses a signal field strength entropy sensing module to monitor wireless communication quality parameters in real time on platforms and within tunnels. It generates large passenger flow early warning signals using a synchronized decline condition across the entire area, enabling it to detect passenger flow trends earlier than data from automatic fare collection systems, thus providing a critical decision-making window of several minutes for operations management and passenger guidance. Simultaneously, by verifying the signal consistency between at least two different operator networks, it effectively suppresses false alarms caused by single-operator network failures, improving the reliability of the early warning system.

[0016] 2. This invention utilizes a train control information frame parsing module to extract door action timing, train automatic driving operation curves, and traction and braking feedback curves from existing signaling systems. Through a micro-scale mapping model, it outputs the number of passengers boarding, alighting, and transferring between lines, completely eliminating the need for individual identification data such as passenger card swipe records or mobile phone signaling. This achieves real-time and accurate counting of transfer volume while protecting passenger privacy. The micro-scale mapping model is pre-trained using supervised learning and continuously updated using online incremental learning, enabling it to adapt to dynamic changes in passenger flow patterns.

[0017] 3. The invention's dynamic guidance decision-making module dynamically generates induced discount prices based on a comparison of predicted transfer passenger flow and real-time capacity saturation of each route. This induced information is then pushed to passenger terminals through differentiated transmission priorities. High-priority queues are allocated to passengers with high historical response rates and located in the transfer decision-making area. This leverages low-latency transmission to create a "first-come, first-served" game mentality. Combined with a two-sided auction mechanism, limited discount slots are allocated based on credit score or response time, significantly improving passenger acceptance of the induced information and the actual conversion rate of transfer behavior.

[0018] 4. The closed-loop feedback self-calibration module of this invention collects the response behavior and actual travel paths of passenger terminals, forming feedback signals that are respectively sent back to the signal field entropy sensing module to adjust the early warning judgment threshold, back to the train control information frame parsing module to update the microscale mapping model parameters, and back to the dynamic guidance decision module to optimize the pricing strategy. Through online incremental learning algorithms and automatic threshold adjustment driven by false alarms, the system can adapt to passenger flow pattern drift, changes in communication environment, and evolution of passenger behavior in long-term operation. The prediction accuracy and guidance effect continuously improve with the accumulation of operating data, avoiding the problem of performance degradation over time in traditional open-loop systems.

[0019] 5. The cross-line transfer passenger flow tracking submodule of this invention analyzes the control information frames of arriving trains on each line within the transfer station to obtain the number of passengers boarding and alighting on each line. Based on flow conservation constraints, historical transfer ratio priors, and real-time signal distribution information, it solves the transfer matrix to achieve accurate perception of real-time transfer flow direction between multiple lines. The dynamic guidance decision module then performs cascaded guidance, directing passengers originally planning to transfer from one congested line to another to an empty line, thus achieving a dynamic and balanced distribution of capacity at the network level.

[0020] 6. This invention fully reuses the existing digital information transmission infrastructure of the subway (including dedicated wireless communication networks, 5G public network slices, and train control signal system transmission channels), eliminating the need for additional dedicated passenger flow sensors. Each module can be deployed in the network control center, edge computing nodes, or cloud servers, supporting both centralized and distributed deployment schemes to adapt to the operation and management needs of subway networks of different sizes. Passenger actual travel route data can be obtained through various methods such as application positioning, automatic fare collection system records, or mobile phone signaling trajectories, and is de-identified during collection, balancing data availability with privacy protection compliance requirements. Attached Figure Description

[0021] Figure 1 This is a system architecture diagram of the multi-line subway passenger flow collaborative prediction and management system based on digital information transmission proposed in this invention; Figure 2 This is a flowchart of the signal field strength entropy perception and determination process of the multi-line subway passenger flow collaborative prediction and management system based on digital information transmission proposed in this invention. Figure 3 This invention relates to a train frame analysis and transfer passenger flow tracking map for a multi-line metro passenger flow collaborative prediction and management system based on digital information transmission. Figure 4 This invention presents a dynamic guidance and bilateral auction diagram for a multi-line subway passenger flow collaborative prediction and management system based on digital information transmission. Figure 5This invention presents a closed-loop feedback self-calibration diagram for a multi-line subway passenger flow collaborative prediction and management system based on digital information transmission. Detailed Implementation

[0022] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0023] It should be noted that, in this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms include, contain, or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, the presence of additional identical elements in the process, method, article, or apparatus that includes elements is not excluded.

[0024] refer to Figures 1-5 This invention discloses a multi-line subway passenger flow collaborative prediction and management system based on digital information transmission, belonging to the field of intelligent transportation technology. The system includes: a signal field strength entropy sensing module, used to collect wireless communication quality parameters in subway platforms and tunnels in real time, calculate the spatial distribution entropy of signal quality, and generate a large passenger flow warning signal when a synchronous decline across the entire area is detected; a train control information frame parsing module, used to parse the control information frames of trains about to arrive after receiving the warning signal, and output cross-line transfer passenger flow through a micro-scale mapping model; a dynamic guidance decision module, used to generate induced preferential prices based on the comparison results of transfer passenger flow and the real-time capacity saturation of each line, and push induced information to passenger terminals through differentiated transmission priorities; and a closed-loop feedback self-calibration module, used to collect the response behavior and actual travel paths of passenger terminals, form feedback signals, and send them back to each module to adjust their operating parameters; the four modules achieve data interaction through the same digital information transmission infrastructure, forming a closed loop of perception, tracking, guidance, and feedback. This invention solves the problems of AFC perception delay, reliance on privacy data for transfer tracking, low guidance efficiency, lack of closed-loop self-calibration, and failure to utilize transmission infrastructure data in the prior art. It achieves advanced early warning, accurate tracking without individual identification, high response rate guidance, and continuous system evolution. Moreover, it fully reuses the existing communication and signaling system of the subway and has good engineering economics.

[0025] Example 1 This embodiment provides a multi-line subway passenger flow collaborative prediction and management system based on digital information transmission, which is used for early warning and guidance of large passenger flows at transfer hub stations.

[0026] Purpose of implementation: During the morning rush hour at transfer hubs, this system can be used to anticipate the upcoming large passenger flow, accurately count the number of transfer passengers, and use dynamic pricing to guide passengers to switch to alternative routes, thereby alleviating overcrowding on the target routes.

[0027] Implementation System: This embodiment is deployed at the "Central Square Station," a transfer station between Metro Lines A and B in a certain city. The system includes: Signal field strength entropy sensing module: connects to China Mobile and China Telecom 5G base stations and Wi-Fi probes within the station. The station area is divided into 20 monitoring grids (8 meters on each side).

[0028] Train control information frame parsing module: connected to the signal field strength entropy sensing module via data bus and connected to the metro signaling system interface (compliant with IEEE1474.1CBTC standard).

[0029] Dynamic guidance decision-making module: It is connected to the train control information frame parsing module through the data bus, and communicates with the passenger APP backend through 5G network slicing.

[0030] Closed-loop feedback self-calibration module: It is connected to the above three modules through the data bus and receives data from the automatic fare collection system (AFC) and APP logs.

[0031] Implementation steps: Signal Entropy Sensing: The signal field strength entropy sensing module collects parameters such as RSRP, SINR, and access failure rate of 5G base stations in each grid at a frequency of 10Hz, and calculates the spatial distribution entropy of signal quality H(t) in real time. During the morning peak at 8:05:00, H(t) decreased from 0.85 to 0.52 within 10 seconds (a decrease of 0.33), and the statistical standard deviation of the signal quality degradation rate in each grid was 0.04. Simultaneously, the signals of China Mobile and China Telecom both experienced synchronous degradation. Based on this, the module determined it to be a synchronous degradation across the entire area and generated a large passenger flow warning signal.

[0032] Train information frame parsing: After the warning signal is triggered, the train control information frame parsing module locks onto the next train about to arrive (train number A102, expected arrival time 8:05:30), and obtains the train's door control frame and ATO (Automatic Train Operation) curve frame through the signal system interface. The parsing reveals: door opening duration 4.2 seconds, door closing duration 3.8 seconds, number of multiple door openings / closings 1; acceleration change rate fluctuation amplitude during braking phase 0.25 m / s². 3The coefficient of variation of the braking force maintaining current is 0.18. Inputting these features into a microscale mapping model (gradient boosting tree, pre-trained with 30 days of historical AFC data) outputs 145 passengers alighting and 198 passengers boarding. Simultaneously, parsing the arrival train information frame on Line B reveals 89 passengers boarding and 76 passengers alighting on Line B. The cross-line transfer passenger flow tracking submodule (connected to the frame parsing module via a data bus) solves the transfer matrix based on flow conservation constraints and historical transfer ratio priors, combined with signal entropy distribution assistance, outputting 72 passengers transferring from Line A to Line B.

[0033] Dynamic Induction and Bilateral Auction: The dynamic induction decision module obtains the remaining capacity of Line A (only 50 people, 130% load factor) and the capacity of Line B (80 people), determining that at least 22 people need to be induced to switch from Line A to Line B. An induction discount price of 3.5 yuan is generated (base price 2 yuan, adjusted upwards based on congestion). Transmission priority is allocated using 5G QoS flow identifiers: 200 passengers with a historical response rate >40% and located at the transfer channel entrance are allocated URLLC slices with a delay of 8-10ms; other passengers are allocated default slices with a delay of 50-80ms. 180 high-priority passengers receive the offer, and 60 click "accept"; the system updates the remaining capacity to 22 people, reducing the discount to 2.8 yuan; subsequently, 25 low-priority passengers accept, and the first 22 people ranked by response time receive the discount. Ultimately, 22 people are successfully induced to switch to Line B.

[0034] Closed-loop feedback self-calibration: After the train departs the station, the closed-loop feedback self-calibration module receives the actual number of passengers boarding (201) and disembarking (140) recorded by the AFC system. The prediction error is calculated as: +3 passengers boarding, -5 passengers disembarking. Stochastic gradient descent (learning rate 0.01) is used to update the microscale mapping model parameters. This warning signal was verified as genuine high passenger flow, therefore the entropy threshold is not adjusted. Simultaneously, the actual travel paths of the 22 passengers who received guidance (via AFC exit records) are verified; 2 passengers were found not to have followed the guided paths, resulting in a reduction of their credit scores for future pricing optimization.

[0035] Implementation results: The entire process, from signal entropy warning to completion of guidance, took 55 seconds and was completed entirely within the train's station window. The guidance acceptance rate was approximately 32% (22 / 70), the load factor on Line A decreased from 130% to 118%, congestion was alleviated by 9%, and the average additional waiting time for passengers was reduced by approximately 2 minutes.

[0036] Example 2 This embodiment provides a multi-line subway passenger flow collaborative prediction and management system based on digital information transmission, which is used for early warning and flow restriction of large passenger flow on a single line without transfer stations.

[0037] Purpose of implementation: At single-line stations without transfer facilities (such as stadium stations), this system can be used to anticipate large passenger flows as people leave the venue and guide passengers to wait for their trains at different locations or choose alternative modes of transportation through information push notifications, thereby avoiding severe overcrowding on trains.

[0038] Implementation System: Similar to Example 1, but the dynamic guidance decision-making module is only configured to push notification information (without alternative route discounts). The module is deployed at "Stadium Station" on Line C.

[0039] Implementation steps: The signal field strength entropy sensing module detected a synchronous decline in the platform signal quality across the entire domain: within a 12-second time window, H(t) decreased by 0.41, the grid standard deviation was 0.03, and the signals from both operators deteriorated synchronously, generating a large passenger flow warning signal.

[0040] The train control information frame parsing module analyzes the arriving train and outputs a predicted number of passengers boarding of 650 (train's rated passenger capacity is 400), indicating severe overcrowding.

[0041] Since there were no alternative routes, the dynamic guidance decision-making module instead pushed a message to passengers' apps within the platform stating, "This station is saturated; we suggest waiting for the next train or choosing another mode of transportation," along with a 2 yuan coupon for their next ride. Push notification priority was allocated by platform area: areas near the turnstiles received high priority (5GURLLC slices), while outer areas received the default priority.

[0042] The closed-loop feedback self-calibration module receives actual AFC data: 620 people actually on board, with an error of -30 people. Since the prediction overestimated in three consecutive similar scenarios, the module automatically raises the first threshold from 0.3 to 0.35 to reduce the false trigger rate.

[0043] Implementation results: The system successfully issued an early warning of a large passenger flow after the event, sending notifications to approximately 500 passengers. Some passengers opted for connecting buses or waited for subsequent empty trains, and the actual number of passengers boarding the train was controlled at 620 (55% over capacity). This prevented an accident where the train could not close its doors due to severe overcrowding. At the same time, by raising the threshold, the system reduced false alarms in similar scenarios in the future.

[0044] Example 3 This embodiment provides a multi-line subway passenger flow collaborative prediction and management system based on digital information transmission, which is used for differentiated operator determination when 5G signal coverage is uneven.

[0045] Purpose of implementation: When there are single-operator signal blind spots in some sections of the subway line, the dual-operator judgment mechanism is used to avoid false large passenger flow warnings caused by network failures and to achieve multi-path redundant push.

[0046] Implementation System: Similar to Example 1, but the signal field entropy sensing module collects independent signal indicators from the three major operators: China Mobile, China Unicom, and China Telecom. The dynamic guidance decision-making module supports selecting transmission channels based on the operator.

[0047] Implementation steps: In a certain tunnel section, the signal field strength entropy sensing module detected a synchronous decrease in China Mobile signal across the entire area (H(t) decrease of 0.38, grid standard deviation of 0.04), while China Unicom and China Telecom signals were normal (decreases of 0.05 and 0.07 respectively). Because the condition of "at least two different operator networks meeting the synchronous decrease condition across the entire area" was not met, the module determined it to be a mobile network fault and did not generate a large passenger flow warning signal.

[0048] Subsequent actual passenger flow showed no abnormalities (AFC records showed that the number of passengers entering and leaving the station was the same as usual), verifying the effectiveness of false alarm suppression.

[0049] During periods of normal signal strength, the dynamic guidance decision-making module dynamically allocates push channels based on the real-time signal quality of each operator: China Unicom users are pushed through Unicom 5G slices, and China Mobile users are pushed through Wi-Fi probes (due to the instability of mobile base station signals), ensuring that the reception delay of guidance information for all passenger terminals is less than 50ms.

[0050] Implementation results: No false warnings were generated during the single-carrier network failure, avoiding ineffective responses from operators; multi-path redundant push enabled the delivery rate of induced information to reach 99.5%, which is about 15% higher than that of a single channel.

[0051] Example 4 This embodiment provides a multi-line subway passenger flow collaborative prediction and management system based on digital information transmission, which uses neural networks and online updates for microscale mapping models.

[0052] Purpose of implementation: In the absence of historical training data for newly opened routes, neural network models and online incremental learning are used to quickly adapt to changes in passenger flow, enabling accurate predictions from scratch.

[0053] Implementation System: Similar to Example 1, but the microscale mapping model in the train control information frame parsing module adopts a three-layer MLP neural network (8 nodes in the input layer, 16 nodes in the hidden layer, and 2 nodes in the output layer). The closed-loop feedback self-calibration module is configured to trigger a batch update every 10 events.

[0054] Implementation steps: On the first day of operation of the new line, the microscale mapping model used neural network parameters that had been pre-trained on similar lines (similar line types and station spacing) as initial values.

[0055] During the first week of operation, after each train stop event, the closed-loop feedback self-calibration module compared the actual number of passengers getting on and off the train at AFC with the model's predicted value, accumulating the error gradient. A batch update was performed every 10 events (batchsize=10, learning rate 0.005).

[0056] Two weeks later, the average absolute error of the model's predictions decreased from the initial ±12 people to ±5 people, reaching a practical level.

[0057] When special passenger flow patterns occur during holidays (such as large-scale events held around the station), online incremental learning can quickly adjust the weights within 1-2 trains, so that the prediction error returns to the normal range within 30 minutes.

[0058] Implementation results: The new route can be put into use without long-term data accumulation, achieving the accuracy that traditional models take three months to reach in just two weeks. The adaptation time for sudden changes in passenger flow during holidays is shortened to less than 30 minutes, while traditional models require several days.

[0059] Example 5 This embodiment provides a multi-line subway passenger flow collaborative prediction and management system based on digital information transmission, which is used for multi-line collaborative transfer tracking and cascading guidance.

[0060] Purpose of implementation: In a three-line transfer hub (lines X, Y, and Z), this system is used to achieve dynamic balance among multiple lines, guiding passengers who originally planned to transfer from the congested line X to the equally congested line Y to the less crowded line Z, thus avoiding excessive congestion in a single transfer channel.

[0061] Implementation System: Similar to Example 1, but with the added capability of parallel analysis for three lines. The cross-line transfer passenger flow tracking submodule supports solving the transfer matrix (3×3 matrix) for the three lines.

[0062] Implementation steps: The signal field entropy sensing module detected a synchronous decrease across the entire X-ray station and generated an early warning.

[0063] The train control information frame parsing module analyzes the arriving train on line X and outputs the number of passengers disembarking as 280. The cross-line transfer passenger flow tracking submodule, combining historical transfer ratio priors and signal distribution assistance, outputs the estimated number of passengers transferring to line Y as 160, transferring to line Z as 80, and exiting the station as 40.

[0064] The dynamic guidance decision-making module obtains the remaining capacity of 30 people on line Y and 120 people on line Z, and calculates that 130 people need to be guided to switch from line Y to line Z.

[0065] The system induces the process in two stages: Phase 1 (30 seconds before train arrival): A message was sent to 160 passengers who planned to transfer to Line Y, offering a discount of 4 yuan for transferring to Line Z. 32 high-priority passengers (at the transfer entrance area) accepted the message.

[0066] Phase Two (Train Stopping): The train control information frame parsing module updates the actual number of people transferring to Line Y to 128. The dynamic guidance decision-making module updates the remaining guidance demand to 98 people and pushes a "discount of 3 yuan". An additional 45 people are induced to accept.

[0067] After the train departed the station, the closed-loop feedback self-calibration module verified through AFC records that 77 people were successfully induced (32 in the first phase + 45 in the second phase). The load factor of line Y decreased from 140% to 105%, and line Z increased from 60% to 85%. The feedback data was used to update the historical transfer ratio prior (the prior probability of transferring to line Y was reduced from 0.5 to 0.45).

[0068] Implementation results: Dynamic balance was achieved among the three lines, congestion on line Y was reduced by 35%, capacity utilization on line Z was increased by 25%, and the average queuing time at the transfer channel was reduced from 8 minutes to 3 minutes.

[0069] Comparative Example 1 This comparison provides a traditional open-loop system with no feedback self-calibration.

[0070] Purpose of implementation: By comparing it with the present invention, the key role of the closed-loop feedback self-calibration module in the long-term performance of the system is demonstrated.

[0071] The structure of a comparative system: It only includes the signal field entropy sensing module, the train control information frame parsing module, and the static guidance decision module. Among them: The signal field strength entropy sensing module uses a fixed entropy threshold (0.3) that cannot be adjusted.

[0072] The parameters of the microscale mapping model in the train control information frame parsing module are fixed and are not updated online.

[0073] The dynamic guidance decision-making module adopts static rules: when the predicted congestion is >120%, it broadcasts "recommendation to take a detour" to all passengers, with no differential delay and no bilateral auction.

[0074] The system has no closed-loop feedback self-calibration module, and the system parameters (entropy threshold, model parameters, pricing strategy) never change.

[0075] Run the comparative experiment: At the same transfer station in Example 1, the comparative system and the present invention (Example 1) were run in parallel for 30 days, and key performance indicators were statistically analyzed: Accuracy of large passenger flow warning: 92% (13 out of 14 real large passenger flow warnings were correct, with only 1 missed warning); the comparison rate was 78% (only 16 out of 21 warnings corresponded to real large passenger flow, with 5 false alarms).

[0076] Average absolute error of passenger flow forecast: This invention decreased from ±8 people in the first week to ±5 people in the fourth week; the comparative example was ±15 people in the first week, and increased to ±22 people in the fourth week due to seasonal changes in passenger flow patterns (students returning to school, shopping mall promotions).

[0077] The acceptance rate of the guiding information was stable at 30%-35% in this invention; the comparative rate dropped from 12% in the first week to 8% in the fourth week (due to passengers becoming increasingly desensitized to repeated, invalid broadcasts).

[0078] System performance changes after 30 days of operation: The prediction error of the present invention decreased by 40% compared with the first week; the prediction error of the comparative example increased by approximately 47% compared with the first week.

[0079] False alarm rate: 8% for this invention (only 1 signal anomaly caused by extreme weather); 22% for the comparative invention (5 false alarms).

[0080] Results analysis: The comparative model, lacking a closed-loop feedback self-calibration module, had a fixed entropy threshold, making it unable to adapt to seasonal passenger flow changes or network fluctuations, leading to an increased false alarm rate; the micro-scale mapping model could not learn online, and prediction errors accumulated and worsened over time; static broadcast guidance lacked personalized pricing and priority push, resulting in a continuous decline in passenger response rates. In contrast, this invention, through a closed-loop feedback self-calibration module, achieves adaptive parameter adjustment and continuous model evolution, resulting in improved performance over long-term operation, fully demonstrating the technological advancement of this invention.

[0081] The above five embodiments and one comparative example fully demonstrate the technical effects of the present invention: Early detection capability: The signal entropy provides an early warning 3-5 minutes earlier than AFC data, winning a critical time window for dynamic guidance.

[0082] Accurate tracking capability: The frame analysis model can count passenger transfers without individual identification data, with an error of less than ±8%.

[0083] Highly effective persuasion capabilities: Differentiated delayed push notifications and bilateral auctions enable persuasion acceptance rates to reach three times that of traditional broadcasting.

[0084] Closed-loop evolution capability: The self-calibration module ensures that the system does not decay during long-term operation, with a false alarm rate of less than 10%.

[0085] Engineering economics: It fully reuses existing communication and signal systems, without the need for additional sensor deployment.

[0086] Compared to Examples 1-5 and Comparative Example 1, in terms of perception and tracking, the signal field strength entropy perception modules in Examples 1-5 all adopt a warning mechanism of "full-domain synchronous decrease + dual-operator judgment," and are combined with the dynamic threshold adjustment of the closed-loop feedback self-calibration module, so that the accuracy of large passenger flow warnings is stable at over 90%. For example, in Examples 1 and 2, the system can issue warnings 3-5 minutes before AFC data is generated, and the false alarm rate is less than 10%. In contrast, the open-loop system in Comparative Example 1 uses a fixed entropy threshold, which cannot distinguish between single-operator network failures and actual passenger flow surges. It has 5 false alarms in 30 days of operation, and the warning accuracy is only 78%. In terms of passenger transfer flow tracking, the train control information frame parsing modules in Examples 1, 4, and 5 all adopt a microscale mapping model with online incremental learning and updating. The prediction error gradually converges from the initial ±8-12 people to within ±5 people, and can quickly adapt to sudden changes in passenger flow such as holidays. In contrast, the model parameters in Comparative Example 1 remained constant. However, with seasonal changes in passenger flow patterns, the prediction error worsened from ±15 people in the first week to ±22 people in the fourth week, resulting in a lack of reliable data for subsequent guidance decisions. The closed-loop feedback self-calibration module in this example periodically corrected the model parameters using real AFC data, but the comparative example lacked this step, leading to a continuous accumulation of errors.

[0087] In terms of induction effectiveness, the dynamic induction decision-making modules in Examples 1, 3, and 5 employ differentiated transmission priorities and a bilateral auction mechanism. Taking Example 1 as an example, high-priority passengers (those with high historical response rates and located in the transfer decision area) receive accurate quotes within 10 milliseconds, achieving an induction acceptance rate of 32%, nearly three times that of the comparative static broadcast (11%). The two-stage cascaded induction in Example 5 further boosts the acceptance rate to over 35%. The bilateral auction rule (allocating limited discount slots based on response time or credit score) creates a sense of urgency, significantly improving passengers' decision-making speed. In contrast, the comparative example 1 uses indiscriminate static broadcasting, causing passengers to gradually become "immune" to repetitive information, with the acceptance rate dropping from 12% in the first week to 8% in the fourth week. Regarding long-term stability, the closed-loop feedback self-calibration module in the examples continuously optimizes the entropy threshold and pricing strategy. In Example 2, after automatically raising the threshold due to continuous false alarms, the accuracy of subsequent warnings significantly improves. In Example 4, online incremental learning reduces the model prediction error by 40% after 30 days of operation. In contrast, the performance of Comparative Example 1 deteriorated over time, with an increased false alarm rate and decreased guidance effectiveness, making it completely unable to cope with the slow drift of passenger flow patterns. In summary, this invention, through a closed-loop linkage of perception, tracking, guidance, and self-calibration, significantly outperforms traditional open-loop systems in terms of early warning accuracy, tracking accuracy, guidance efficiency, and system robustness, verifying the irreplaceable synergistic effect of the four modules.

[0088] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of protection of the claims.

[0089] In conclusion, the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-line subway passenger flow collaborative prediction and management system based on digital information transmission, characterized in that, include: The signal field strength entropy sensing module is used to collect wireless communication quality parameters in subway platforms and tunnels in real time, calculate the spatial distribution entropy of signal quality in the platform area, and generate a large passenger flow warning signal when the spatial distribution entropy is detected to be synchronously decreasing across the entire area. The train control information frame parsing module is connected to the signal field strength entropy sensing module via a data bus or communication interface. It is used to parse the control information frame of the train that is about to arrive at the station after receiving the large passenger flow warning signal, extract the door action sequence and train operation parameters, and output the cross-line passenger flow through a micro-scale mapping model. The dynamic guidance decision module is connected to the train control information frame parsing module via a data bus or communication interface. It is used to generate a guidance discount price based on the comparison results of the cross-line transfer passenger flow and the real-time capacity saturation of each line, and push guidance information containing guidance path and discount price to the passenger terminal through the digital information transmission channel. The closed-loop feedback self-calibration module is connected to the signal field entropy sensing module, the train control information frame parsing module, and the dynamic guidance decision module via a data bus or communication interface, respectively. It is used to collect passenger terminal response behavior to guidance information and passenger actual travel path data, form feedback signals, and send the feedback signals back to the signal field entropy sensing module to adjust its judgment conditions for generating large passenger flow warning signals, back to the train control information frame parsing module to update the parameters of the micro-scale mapping model, and back to the dynamic guidance decision module to optimize the generation strategy of the guidance discount price. The signal field strength entropy sensing module, train control information frame parsing module, dynamic guidance decision-making module, and closed-loop feedback self-calibration module achieve data interaction through the same digital information transmission infrastructure, forming a closed loop of perception, tracking, guidance, and feedback.

2. The multi-line subway passenger flow collaborative prediction and management system based on digital information transmission as described in claim 1, characterized in that, The conditions for the signal field strength entropy sensing module to determine a synchronous decrease across the entire area include: the decrease in spatial distribution entropy exceeds a first threshold within a preset time window, and the statistical standard deviation of the signal quality degradation rate of each monitoring grid within the station is less than a second threshold; the preset time window is 5 to 20 seconds, the first threshold is 0.2 to 0.5, and the second threshold is 0.03 to 0.08; and the large passenger flow warning signal is generated only when the wireless communication quality parameters of at least two different operator networks meet the conditions.

3. The multi-line subway passenger flow collaborative prediction and management system based on digital information transmission as described in claim 1, characterized in that, The control information frames acquired by the train control information frame parsing module include train position report frames, door control frames, train automatic driving operation curve frames, and traction and braking status frames; the door action sequence includes door opening duration, door closing duration, and the number of times the door is opened and closed; the input features of the microscale mapping model also include the fluctuation amplitude of the rate of change of acceleration during the braking phase and the coefficient of variation of the braking force maintaining current during the parking period; the microscale mapping model is a machine learning model trained through supervised learning, and the machine learning model is updated through online incremental learning during operation.

4. The multi-line subway passenger flow collaborative prediction and management system based on digital information transmission as described in claim 1, characterized in that, When the dynamic guidance decision module pushes guidance information through the digital information transmission channel, it allocates different transmission priorities based on the passenger's historical response rate and current location, so that the transmission delay of the guidance information received by different passenger terminals is differentiated. The allocation of transmission priorities is achieved through 5G service quality flow identification or Wi-Fi air interface scheduling, wherein high priority queues are allocated to passenger terminals whose historical response rate is higher than a configurable response threshold and who are currently located in the transfer decision area.

5. The multi-line subway passenger flow collaborative prediction and management system based on digital information transmission as described in claim 1, characterized in that, The dynamic guidance decision-making module also includes a bilateral auction submodule, which is used to use the guidance discount price as an auction offer in each guidance time window, and the acceptance or rejection decision returned by the passenger terminal as the bidding response. When the number of passengers accepting the offer exceeds the guidance quota allowed by the remaining capacity of the alternative route, the discount quota is allocated according to the passenger's historical credit score or the order of response arrival time.

6. The multi-line subway passenger flow collaborative prediction and management system based on digital information transmission as described in claim 2, characterized in that, The closed-loop feedback self-calibration module adopts an online incremental learning algorithm. After each train stop event, the error between the actual number of passengers getting on and off the train and the predicted number of passengers getting on and off the train output by the micro-scale mapping model is used as the loss, and the model parameters are updated by stochastic gradient descent. When multiple consecutive large passenger flow warning signals are determined to be false alarms by the subsequent train information frame parsing results, the closed-loop feedback self-calibration module automatically raises the first threshold or the second threshold.

7. The multi-line subway passenger flow collaborative prediction and management system based on digital information transmission as described in claim 1, characterized in that, It also includes a cross-line transfer passenger flow tracking submodule, which is connected to the train control information frame parsing module via a data bus or communication interface. This submodule is used to parse the control information frames of arriving trains on each line within the transfer station to obtain the number of passengers boarding and alighting on each line. Based on the flow conservation constraint and the prior historical transfer ratio, it combines the real-time signal distribution information output by the signal field entropy sensing module to solve the transfer matrix and output the real-time transfer passenger flow between each line.

8. The multi-line subway passenger flow collaborative prediction and management system based on digital information transmission as described in claim 1, characterized in that, The signal field strength entropy sensing module, train control information frame parsing module, and dynamic guidance decision-making module are deployed in the metro network control center, edge computing nodes, or cloud servers; the digital information transmission infrastructure includes the metro-dedicated wireless communication network, 5G public network slicing, and existing transmission channels of the train control signal system.

9. The multi-line subway passenger flow collaborative prediction and management system based on digital information transmission as described in claim 1, characterized in that, The closed-loop feedback self-calibration module collects passenger actual travel route data through passenger terminal application positioning, automatic fare collection system records, or mobile phone signaling trajectory acquisition, and performs de-identification processing during collection, which includes deleting mobile phone numbers and device identifiers.

10. The multi-line subway passenger flow collaborative prediction and management system based on digital information transmission as described in claim 2, characterized in that, The preset time window is 10 seconds, the first threshold is 0.3, and the second threshold is 0.05.

Images