Public transport automatic scheduling method based on multi-model mutual learning

Abstract

The invention discloses a multi-model mutual learning public transport automatic scheduling method, and relates to the technical field of public transport, and the method comprises the steps: obtaining the real-time arrival, departure and station dwell time of each bus at each station in each line and the historical arrival, departure and station dwell time; acquiring real-time passenger flow and historical passenger flow of the station; on the basis of the collected data, all scheduling events occurring in actual operation are analyzed in real time, and the scheduling events comprise a vehicle fault event, a large passenger flow event and a vehicle delay event; generating a corresponding scheduling scheme for each scheduling event; and the dispatcher makes a final dispatching decision according to each dispatching event and the corresponding dispatching scheme. According to the method, analysis models and processing models of different scheduling events are established, characteristics and results of different models are integrated, the most suitable model is dynamically selected according to real-time scheduling requirements, simulation prediction and scheduling decision making are achieved, and finally intelligent and automatic scheduling of buses is achieved.

Description

Technical Field

[0001] The present invention relates to the field of public transportation technology, and in particular to a multi-model mutual learning public transportation automation scheduling method. Background Art

[0002] Currently, many cities' public transportation systems have introduced intelligent dispatching systems to optimize dispatching plans by real-time monitoring of vehicle locations, passenger flow and other information. However, most current dispatching products still require a lot of manual intervention by dispatchers, with limited automation, and fail to fully tap the potential of intelligent technology. Summary of the invention

[0003] In order to overcome the defects in the above-mentioned prior art, the present invention provides a public transportation automated scheduling method with multi-model mutual learning, establishes analysis models and processing models for different scheduling events, integrates the characteristics and results of different models, and dynamically selects the most suitable model according to real-time scheduling needs to realize simulation prediction and scheduling decision-making, and finally realizes intelligent and automated scheduling of public buses.

[0004] To achieve the above object, the present invention adopts the following technical solutions, including: A multi-model mutual learning method for public transportation automatic scheduling, comprising: Data collection: obtain the real-time entry, exit, and stop time of each bus and each trip at each station, as well as the historical entry, exit, and stop time; obtain the real-time and historical passenger flow at the station; Dispatch event analysis: Based on the collected data, various dispatch events that occur in actual operations are analyzed in real time, including vehicle failure events, large passenger flow events, and vehicle delay events; Scheduling event processing: Generate corresponding scheduling plans for each scheduling event.

[0005] Preferably, the analysis of the vehicle failure event is specifically as follows: If the speed of the vehicle decreases by more than the set range and the position of the vehicle does not change for a certain period of time, it is determined as a vehicle fault event; otherwise, it is not determined as a vehicle fault event; Analyze the impact of vehicle failure events: the average total passenger flow from the vehicle failure location to the terminal at the same time in history is taken as the passenger flow P0 affected by the vehicle failure event; the delay of the vehicle failure event on the next vehicle’s travel time is T g =P0×λ; where λ is the average boarding and alighting time for each passenger.

[0006] Preferably, the analysis of large passenger flow events is specifically as follows: The passenger flow of each station in the future is predicted. If the predicted passenger flow P1 in the future is greater than k1 times the historical passenger flow P2 in the same period, it is judged as a large passenger flow event; otherwise, it is not judged as a large passenger flow event. <k1<1.1; Analyze the impact of large passenger flow events: the passenger flow affected by large passenger flow events is P3 = P1-P2; the delay of large passenger flow on the next vehicle's travel time is T k = P3×λ; where λ is the average boarding and alighting time for each passenger; Large passenger flow situations are classified into levels: the larger the passenger flow P3 affected by the large passenger flow event, the higher the level, indicating that the large passenger flow situation is more urgent.

[0007] Preferably, the passenger flow of the station in the future is predicted based on the historical passenger flow of the station, combined with the real-time passenger flow of the station and the passenger flow affected by the vehicle failure event.

[0008] Preferably, the analysis of vehicle delay events is specifically as follows: Calculate the predicted travel time T of the vehicle from the current location to the terminal p for: ; Among them, the station closest to the current location of the vehicle is the mth station, 1 ≤ m ≤ n, and there are n stations in total; T m is the travel time of the vehicle from the current location to the nearest station, i.e. the mth station; T i,i+1 is the travel time of the vehicle from the i-th station to the i+1-th station; D i is the stop time of the vehicle at the i-th station; T g T is the delay in the driving time of this vehicle caused by the vehicle failure event of the previous vehicle; k is the delay of the vehicle's travel time caused by the large passenger flow event; α and β are the weights of the vehicle failure event and the large passenger flow event respectively; Based on the predicted travel time T from the current location of the vehicle to the terminal p , get the predicted arrival time of the vehicle; The predicted arrival time is compared with the planned arrival time of the vehicle operation plan. If the predicted arrival time is later than the planned arrival time, it is determined as a vehicle delay event; otherwise, it is not determined as a vehicle delay event; Vehicle delays are classified into levels: the number of remaining stops from the current position of the delayed vehicle to the terminal is n-m+1, and the ratio of the remaining stops to the total number of stops is (n-m+1) / n; the greater the ratio of the remaining stops to the total number of stops, the higher the level, indicating that the vehicle delay is more urgent.

[0009] Preferably, the travel time between stations and the stop time at each station are predicted based on the historical entry, exit and stop times.

[0010] Preferably, the optimal scheduling scheme is generated according to the analysis results of each scheduling event, as follows: In the event of a vehicle failure, according to the status of the other vehicles on the same line as the failed vehicle, the departure time of the other vehicles is adjusted based on the original operation plan, and a new operation plan is generated with the goal of adjusting the number of vehicles to the minimum; In case of large passenger flow events, a dispatch plan for adding vehicles is generated based on the location of the large passenger flow station and the predicted passenger flow in the future to meet the actual passenger flow demand; In the event of vehicle delays, a scheduling plan of stopping the vehicle at the station or taking a detour is generated based on the actual operating conditions and the predicted arrival time to meet the vehicle's operating plan.

[0011] Preferably, the scheduling plans of various scheduling events are collectively presented in the form of cards on the scheduling interface, and the urgency of each scheduling event is displayed.

[0012] Preferably, it also includes a dispatch decision: the dispatcher makes a final dispatch decision based on each dispatch event and the corresponding dispatch plan, that is, the dispatcher confirms whether to adopt the dispatch plan or adopt it after adjusting the dispatch plan.

[0013] The present invention also provides a multi-model mutual learning public transportation automation dispatching system, which is applicable to the above-mentioned multi-model mutual learning public transportation automation dispatching method, and the system includes: a data acquisition module, a dispatching event analysis module, a dispatching event processing module, and a dispatching decision module; The data collection module obtains the following data: obtain the real-time entry, exit, and stop time of each bus and each trip at each station, as well as the historical entry, exit, and stop time; obtain the real-time and historical passenger flow at the station; The dispatch event analysis module includes a vehicle failure analysis model, a large passenger flow prediction model and a vehicle delay prediction model, which are respectively used for real-time analysis of vehicle failure events, large passenger flow events and vehicle delay events occurring in actual operation; the vehicle failure analysis model, the large passenger flow prediction model and the vehicle delay prediction model respectively input the analysis results into the dispatch event processing module to generate corresponding dispatch plans; among them, the large passenger flow prediction model analyzes large passenger flow events in combination with the analysis results of the vehicle failure analysis model; the vehicle delay prediction model analyzes vehicle delay events in combination with the analysis results of the vehicle failure analysis model and the analysis results of the large passenger flow prediction model; The scheduling event processing module includes vehicle failure processing model, large passenger flow processing model and vehicle delay processing model, which are used to generate scheduling plans for different scheduling events; The scheduling decision module is used by the dispatcher to view each scheduling event and the corresponding scheduling plan, and make the final scheduling decision.

[0014] The advantages of the present invention are: (1) The present invention establishes analysis models and processing models for different dispatching events by utilizing advanced information technology, big data analysis and other means, integrates the characteristics and results of different models, dynamically selects the most suitable model according to real-time dispatching needs, realizes simulation prediction and dispatching decision-making, and finally realizes intelligent and automated dispatching of public buses.

[0015] (2) The present invention forms a data closed loop among vehicle fault analysis, large passenger flow prediction and vehicle delay prediction (e.g., vehicle fault analysis results are used for large passenger flow prediction and vehicle delay prediction), realizes cross-validation and correction in a dynamic environment, and data feedback between multiple models, thereby improving the accuracy of complex event correlation analysis and the overall analysis accuracy.

[0016] (3) The present invention not only calculates the current delay for vehicle failure events, but also predicts chain reactions through historical passenger flow data at the same location, thereby preventing local problems from causing systemic paralysis.

[0017] (4) The present invention introduces weight coefficients of vehicle failure events and large passenger flow events in vehicle delay analysis, optimizes weight distribution through historical data learning, and enhances the adaptability of the model to complex scenarios.

[0018] (5) The present invention classifies the urgency of large passenger flow events and vehicle delay events, and combines quantitative indicators such as passenger flow increment and delay time to provide a priority basis for subsequent scheduling strategies, which is conducive to the precise allocation of resources according to the degree of urgency.

[0019] (6) In response to vehicle failure events, the present invention generates a new operation plan with the goal of minimizing the number of vehicle adjustments, thereby reducing operating costs and reducing scheduling disturbances. It adopts a strategy of adding vehicles for large passenger flows and a strategy of leaving stations or taking detours for delays, thereby balancing passenger waiting time and the overall efficiency of the line.

[0020] (7) After the present invention generates a scheduling plan, each scheduling event and the urgency of the event are displayed in the form of cards on the scheduling interface, allowing manual adjustment methods to take into account both algorithm efficiency and dispatcher experience.

[0021] (8) The public transportation automated dispatching method of the present invention can improve operational efficiency, reduce operating costs, and enhance passenger experience.

[0022] (9) The public transportation automation dispatching system of the present invention adopts a modular design (data collection, analysis, processing, and decision-making), has strong scalability, and supports the rapid access of new event types. The model mutual learning framework reserves interfaces for the integration of more AI algorithms in the future.

[0023] (10) The present invention constructs an efficient and low-cost public transportation automation and intelligent dispatching method and dispatching system through multi-model collaborative analysis, hierarchical response mechanism and human-machine collaborative decision-making, providing an innovative solution to the real-time and complexity problems of urban public transportation. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Schematic diagram of a public transportation automated scheduling system with multi-model mutual learning.

[0025] Figure 2 Flowchart of a public transportation automation scheduling method based on multi-model mutual learning. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] Example 1 Depend on Figure 1 As shown, a public transportation automatic dispatching system with multi-model mutual learning includes: a data acquisition module, a dispatching event analysis module, a dispatching event processing module, and a dispatching decision module.

[0028] The data collection module obtains the following data: obtains the operation plan of the line vehicles; obtains the real-time entry, exit and stop time of each line and each vehicle at each station, as well as the entry, exit and stop time over a period of history; obtains the real-time passenger flow (number of waiting people) and historical passenger flow at the station; and collects data on uncertain factors such as weather, emergencies, and diversions.

[0029] The dispatch event analysis module is used to monitor and analyze various dispatch events that occur in actual operations in real time, including operation plan change events caused by vehicle failure (or vehicle accident), large passenger flow, and vehicle delays, referred to as vehicle failure events, large passenger flow events, and vehicle delay events.

[0030] A plurality of models are constructed in the scheduling event analysis module, which are respectively used to analyze each scheduling event, and the analysis results of each scheduling event are input into the scheduling event processing module to generate a corresponding scheduling plan.

[0031] Specifically, the scheduling event analysis module includes a vehicle failure analysis model, a large passenger flow prediction model and a vehicle delay prediction model.

[0032] The vehicle fault analysis model uses certain rules to determine whether the vehicle has failed or had an accident, and thus determine whether the vehicle can continue to operate. If the vehicle speed suddenly drops and stagnates for a long time, that is, the speed drop exceeds the set range and the vehicle position does not change for a certain period of time, the vehicle is judged to have failed or had an accident. Specifically, the on-board equipment uploads the real-time speed of the vehicle every 10 seconds. During the non-entry and exit period, the vehicle speed suddenly drops significantly, for example, from 30 km / h to less than 5 km / h, and finally drops to 0 km / h.

[0033] The vehicle failure analysis model takes the location, time and current weather of the vehicle failure as input, determines and outputs the impact of the vehicle failure event, including the passenger flow and road traffic conditions affected by the vehicle failure event; among them, the road traffic conditions affected by the event specifically refer to the delay in the travel time of the next vehicle caused by the event.

[0034] The average total passenger flow from the vehicle failure location to the terminal station under the same historical factors and at the same time is taken as the passenger flow P0 affected by the vehicle failure, and the delay caused by the vehicle failure on the next vehicle's travel time is calculated as T g = P0×λ; where λ is the average boarding and alighting time for each passenger, which is determined by the historical stop time and the number of boarding and alighting passengers.

[0035] For example, at 7:35 during the peak hour, the upbound vehicle V of line A breaks down at station i. The average total passenger flow from station i to the upbound terminal under the same historical factors and time is P0. The passenger flow affected by this vehicle failure event is P0, and the delay of the next vehicle's travel time due to this vehicle failure event is T. g = P0 / λ.

[0036] The vehicle failure analysis model uses the impact of vehicle failure events as an input variable for subsequent vehicle delay prediction and large passenger flow prediction, and inputs them into the vehicle delay prediction model and the large passenger flow prediction model respectively; at the same time, the impact of vehicle failure events is also input into the dispatching event processing module to generate a dispatching plan, and the dispatcher will subsequently confirm the event and plan.

[0037] The large passenger flow prediction model is based on the historical passenger flow data of each station, historical weather, weekdays or holidays, seasons and other factors, combined with the number of waiting people collected in real time at the station (real-time passenger flow at the station), and passenger flow affected by vehicle failure events, etc., and uses a general passenger flow prediction algorithm to predict the passenger flow in the future (30 minutes). The predicted passenger flow P1 in the future is compared with the historical passenger flow P2 with the same factors and the same period. If the predicted passenger flow P1 is greater than 1.1 times the historical passenger flow P2 with the same factors and the same period, it is judged as a large passenger flow event and the large passenger flow situation is classified into levels.

[0038] The levels of large passenger flow situations are divided into the following categories: If the predicted passenger flow P1 is greater than 1.1 times and less than or equal to 1.2 times the historical passenger flow P2 with the same factors and time period, the level of the large passenger flow is determined to be level one; if the predicted passenger flow P1 is greater than 1.2 times and less than or equal to 1.5 times the historical passenger flow P2 with the same factors and time period, the level of the large passenger flow is determined to be level two; if the predicted passenger flow P1 is greater than 1.5 times the historical passenger flow P2 with the same factors and time period, the level of the large passenger flow is determined to be level three; the higher the level, the more urgent the large passenger flow situation is.

[0039] The large passenger flow prediction model takes the large passenger flow location (station), time, current weather, predicted passenger flow and level as input, determines and outputs the impact results of large passenger flow events, including the passenger flow and road traffic conditions affected by the large passenger flow events.

[0040] If the level of the large passenger flow is level 1, the passenger flow affected by the large passenger flow event is P3 = P1-P2, and the delay of the large passenger flow event on the next vehicle's travel time is T k = P3×λ; where λ is the average boarding and alighting time for each passenger, which is determined by the historical stop time and the number of boarding and alighting passengers.

[0041] If the level of the large passenger flow is level 2 or level 3, the passenger flow affected by the large passenger flow event is P3 = P1-P2. However, at this time, the large passenger flow event will delay the travel time of the next vehicle by T k It needs to be determined based on the processing results after the dispatch decision, and based on the dispatch strategy selected by the dispatcher, the delay in the next vehicle's travel time will be determined after returning.

[0042] For example, at 7:15 am during the peak hour, a large passenger flow occurs at station j on the uplink of line A, and the passenger flow affected by the large passenger flow event is P3, and the level of the large passenger flow is level 1. Then the delay of the large passenger flow event on the travel time of the next vehicle is T k = P3×λ.

[0043] The large passenger flow prediction model uses the impact of large passenger flow events as an input variable for subsequent vehicle delay prediction and inputs it into the vehicle delay prediction model; at the same time, it also inputs the impact of large passenger flow events into the scheduling event processing module to generate a scheduling plan, and the dispatcher subsequently confirms the event and plan.

[0044] The vehicle delay prediction model is used to predict the travel time of a vehicle from any location after the start of operation to the final stop. The travel time of a vehicle from any location after the start of operation to the final stop is composed of the stop time at each intermediate stop, the travel time from the current location to the nearest stop, and the travel time between the intermediate stops, plus the influencing variables output by other models.

[0045] The vehicle delay prediction model calculates the predicted travel time T of the vehicle from the current location to the terminal p for: ; Among them, the station closest to the current position of the vehicle is the mth station, 1 ≤ m ≤ n, there are n stations in total, and the last station is the nth station; T m is the travel time of the vehicle from the current location to the nearest station, i.e. the mth station; T i,i+1 is the travel time of the vehicle from the i-th station to the i+1-th station; D i is the stop time of the vehicle at the i-th station; T g The delay in the driving time of this vehicle caused by the failure of the previous vehicle; T g is the delay of the vehicle’s travel time caused by large passenger flow, T g and T k are the input variables of the vehicle failure analysis model and the large passenger flow prediction model respectively; α and β are the weights of vehicle failure events and large passenger flow events respectively, and this weight is dynamically adjusted according to the stability and accuracy of the historical output results of other models after the scheduling decision module is processed; during peak hours, the results of vehicle failure events have a greater impact on vehicle delay prediction, so the weight of vehicle failure events during peak hours will be increased; during off-peak hours, large passenger flow events have a greater impact on vehicle delay prediction, so the weight of large passenger flow events during off-peak hours will be increased.

[0046] In this embodiment, the vehicle delay prediction model classifies the entry and exit data based on historical GPS data, weather, working days or holidays, seasons and other factors. By combining the above-mentioned influencing factors, with a time granularity of 15 minutes, the travel time data between each station in each time period under different historical factors and the stop time data of each station can be obtained, including the mean and standard deviation. Based on the weather, working days or holidays, seasons and other conditions on the forecast day, the data under the same historical factors are selected as the prediction basis, and the general time series prediction algorithm is used to predict the travel time between each station and the stop time of each station in each time period under different factors.

[0047] According to the above formula, the predicted arrival time of the vehicle can be obtained based on the predicted travel time of the vehicle from the current location to the terminal.

[0048] The vehicle delay prediction model compares the predicted arrival time of the vehicle with the planned arrival time of the operation plan. If the predicted arrival time is later than the planned arrival time, the vehicle is considered to be delayed. The vehicle delay situation is then graded according to the current location of the delayed vehicle.

[0049] The levels of vehicle delay are as follows: The number of remaining stops from the current position of the delayed vehicle to the final stop is n-m+1; the ratio of the remaining stops to the total number of stops is (n-m+1) / n. If the ratio of the remaining stops to the total number of stops (n-m+1) / n) is greater than 40%, the vehicle delay is level one; if the ratio of the remaining stops to the total number of stops (n-m+1) / n is greater than 20% and less than or equal to 40%, the vehicle delay is level two; if the ratio of the remaining stops to the total number of stops (n-m+1) / n is less than or equal to 20%, the vehicle delay is level three; the higher the level, the more urgent the vehicle delay.

[0050] The vehicle delay prediction model inputs the predicted arrival time of delayed vehicles and the corresponding situation level into the scheduling event processing module to generate a scheduling plan.

[0051] The dispatch event processing module has different processing models (based on general machine learning models) for different dispatch events, including vehicle fault processing models, large passenger flow processing models, and vehicle delay processing models, which are used to generate dispatch plans for different dispatch events.

[0052] In the case of vehicle failure, since the vehicle cannot continue to operate as planned due to failure or accident in actual operation, the original operation plan needs to be adjusted in real time to meet the subsequent operation needs. Specifically, according to the status of the remaining vehicles on the same line as the faulty vehicle, including the status of the currently operating vehicles and the status of the vehicles that can be operated in the future, the departure time of the remaining vehicles is adjusted based on the original operation plan, with the goal of adjusting the number of vehicles to the minimum, and taking into account the departure intervals and turnaround time of different peak sections, etc., a new operation plan with reduced trains is generated from the global level.

[0053] In response to large passenger flow events, a scheduling plan for adding vehicles is generated based on the location of large passenger flow stations and the predicted passenger flow in the future to meet actual passenger flow demand.

[0054] In case of vehicle delays, a station-skipping or detour scheduling plan is generated based on the actual operation situation and predicted arrival time to meet the vehicle operation plan.

[0055] Among them, station cancellation refers to the temporary cancellation of stops at certain stations in the public transportation system in order to speed up operations or deal with special circumstances. Detour refers to the selection of an alternative route to reach the destination in order to avoid congestion, accidents or other obstacles in traffic route planning. Adding vehicles refers to adding additional vehicles or flights in the public transportation system to meet passenger demand or alleviate congestion during peak hours.

[0056] In this embodiment, the scheduling event processing module establishes multiple different processing models according to specific scheduling requirements, and can dynamically select the most suitable model for processing according to real-time scheduling requirements and resource status to generate a scheduling plan corresponding to the scheduling event.

[0057] The dispatch decision module presents each dispatch event and the corresponding dispatch plan in the form of cards on the dispatch interface, and uses different colors to distinguish the urgency of each dispatch event, so that dispatchers can view each dispatch event and the corresponding dispatch plan and make the final dispatch decision.

[0058] For automated dispatching, the dispatcher mainly plays a supervisory role. Since the system has a high degree of automation, the dispatcher only needs to pay attention to the dispatching events that exist and the dispatching plans that are generated. After entering each card, the dispatcher can observe the results of the dispatching plan and decide whether to adopt the dispatching plan based on the actual situation, or adjust the generated dispatching plan before adopting it, thereby forming the final dispatching decision.

[0059] Example 2 Depend on Figure 2 As shown, based on a multi-model mutual learning public transportation automation scheduling system provided in the above embodiment 1, the specific process of the public transportation automation scheduling method is as follows: S1, data collection: obtain the real-time entry, exit, and stop time and historical entry, exit, and stop time of each line, each vehicle, and each trip at each station; obtain the real-time and historical passenger flow of the station; and collect data on uncertain factors such as weather data, emergencies, and diversions.

[0060] S2, Dispatch Event Analysis: Based on the collected data, real-time analysis of various dispatch events that occur in actual operations, including vehicle failure events, large passenger flow events, and vehicle delay events; S3, scheduling event processing: generating corresponding scheduling plans for each scheduling event; S4, Dispatch Decision: The dispatcher makes the final dispatch decision based on each dispatch event and the corresponding dispatch plan, that is, the dispatcher confirms whether to adopt the dispatch plan or to adopt it after adjusting the dispatch plan.

[0061] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-model mutual learning method for public transportation automation scheduling, characterized in that: include: Data collection: Get the real-time entry, exit, and stop time of each bus and each trip at each station, as well as the historical entry, exit, and stop time; Get the real-time and historical passenger flow of the site; Dispatch event analysis: Based on the collected data, various dispatch events that occur in actual operations are analyzed in real time, including vehicle failure events, large passenger flow events, and vehicle delay events; Scheduling event processing: Generate corresponding scheduling plans for each scheduling event.

2. The public transportation automation scheduling method based on multi-model mutual learning according to claim 1 is characterized in that: The analysis of vehicle failure events is as follows: If the speed of the vehicle decreases by more than the set range and the position of the vehicle does not change for a certain period of time, it is determined as a vehicle fault event; otherwise, it is not determined as a vehicle fault event; Analyze the impact of vehicle failure events: take the average total passenger flow from the vehicle failure location to the terminal at the same time in history as the passenger flow P0 affected by the vehicle failure event; The delay of the failure event on the next vehicle's travel time is T g = P0×λ; where λ is the average boarding and alighting time for each passenger.

3. The public transportation automation scheduling method based on multi-model mutual learning according to claim 1 is characterized in that: The analysis of large passenger flow events is as follows: The passenger flow of each station in the future is predicted. If the predicted passenger flow P1 in the future is greater than k1 times the historical passenger flow P2 in the same period, it is determined to be a large passenger flow event; otherwise, it is not determined to be a large passenger flow event; where 1< k1; Analyze the impact of large passenger flow events: the passenger flow affected by large passenger flow events is P3 = P1-P2; the delay of large passenger flow on the next vehicle's travel time is T k = P3×λ; where λ is the average boarding and alighting time for each passenger; Large passenger flow situations are classified into levels: the larger the passenger flow P3 affected by the large passenger flow event, the higher the level, indicating that the large passenger flow situation is more urgent.

4. The public transportation automation scheduling method based on multi-model mutual learning according to claim 3 is characterized in that: Based on the historical passenger flow of the station, combined with the real-time passenger flow of the station and the passenger flow affected by the vehicle failure event, the passenger flow of the station in the future is predicted.

5. The public transportation automation scheduling method based on multi-model mutual learning according to claim 1 is characterized in that: The analysis of vehicle delay events is as follows: Calculate the predicted travel time T of the vehicle from the current location to the terminal p for: ； Among them, the station closest to the current location of the vehicle is the mth station, 1 ≤ m ≤ n, and there are n stations in total; T m is the travel time of the vehicle from the current location to the nearest station, i.e. the mth station; T i,i+1 is the travel time of the vehicle from the i-th station to the i+1-th station; D i is the stop time of the vehicle at the i-th station; T g T is the delay in the driving time of this vehicle caused by the vehicle failure event of the previous vehicle; k is the delay of the vehicle's travel time caused by the large passenger flow event; α and β are the weights of the vehicle failure event and the large passenger flow event respectively; Based on the predicted travel time T from the current location of the vehicle to the terminal p , get the predicted arrival time of the vehicle; The predicted arrival time is compared with the planned arrival time of the vehicle operation plan. If the predicted arrival time is later than the planned arrival time, it is determined as a vehicle delay event; otherwise, it is not determined as a vehicle delay event; Vehicle delays are classified into levels: the number of remaining stops from the current position of the delayed vehicle to the terminal is n-m+1, and the ratio of the remaining stops to the total number of stops is (n-m+1) / n; the greater the ratio of the remaining stops to the total number of stops, the higher the level, indicating that the vehicle delay is more urgent.

6. The public transportation automation scheduling method based on multi-model mutual learning according to claim 5 is characterized in that: Based on the historical entry, exit and stop times, the travel time between stations and the stop time at each station are predicted.

7. The public transportation automation scheduling method based on multi-model mutual learning according to claim 1 is characterized in that: According to the analysis results of each scheduling event, the optimal scheduling plan is generated, as follows: In the event of a vehicle failure, according to the status of the other vehicles on the same line as the failed vehicle, the departure time of the other vehicles is adjusted based on the original operation plan, and a new operation plan is generated with the goal of adjusting the number of vehicles to the minimum; In case of large passenger flow events, a dispatch plan for adding vehicles is generated based on the location of the large passenger flow station and the predicted passenger flow in the future to meet the actual passenger flow demand; In case of vehicle delays, a scheduling plan of stopping the vehicle at the station or taking a detour is generated based on the actual operation situation and the predicted arrival time to meet the vehicle's operation plan.

8. The public transportation automation scheduling method based on multi-model mutual learning according to claim 1 is characterized in that: The scheduling plans for each scheduling event are centrally presented in the form of cards on the scheduling interface, and the urgency of each scheduling event is displayed.

9. The public transportation automation scheduling method based on multi-model mutual learning according to claim 1 is characterized in that: It also includes dispatching decisions: the dispatcher makes the final dispatching decision based on each dispatching event and the corresponding dispatching plan, that is, the dispatcher confirms whether to adopt the dispatching plan or to adjust the dispatching plan before adopting it.

10. A public transportation automation dispatching system based on multi-model mutual learning, characterized in that: A method for public transportation automation scheduling based on multi-model mutual learning according to any one of claims 1 to 9, the system comprising: a data acquisition module, a scheduling event analysis module, a scheduling event processing module, and a scheduling decision module; The data collection module obtains the following data: the real-time and historical entry, exit and stop times of each bus and each trip at each station; the real-time and historical passenger flow at the station; The dispatch event analysis module includes a vehicle failure analysis model, a large passenger flow prediction model and a vehicle delay prediction model, which are respectively used for real-time analysis of vehicle failure events, large passenger flow events and vehicle delay events occurring in actual operation; the vehicle failure analysis model, the large passenger flow prediction model and the vehicle delay prediction model respectively input the analysis results into the dispatch event processing module to generate corresponding dispatch plans; among them, the large passenger flow prediction model analyzes large passenger flow events in combination with the analysis results of the vehicle failure analysis model; the vehicle delay prediction model analyzes vehicle delay events in combination with the analysis results of the vehicle failure analysis model and the analysis results of the large passenger flow prediction model; The scheduling event processing module includes vehicle failure processing model, large passenger flow processing model and vehicle delay processing model, which are used to generate scheduling plans for different scheduling events; The scheduling decision module is used by the dispatcher to view each scheduling event and the corresponding scheduling plan, and make the final scheduling decision.

Images