CN-121684540-B - Intelligent dispatching method for public transport network based on machine learning model

CN121684540BCN 121684540 BCN121684540 BCN 121684540BCN-121684540-B

Abstract

The invention discloses a bus network intelligent scheduling method based on a machine learning model, which belongs to the technical field of data processing and comprises the following steps of receiving network sensing data, generating a network running situation snapshot when detecting that the network sensing data has abnormal evolution trend, transmitting the network running situation snapshot to a scheduling information coordination processing unit to generate each scheduling information entry, calling each scheduling information entry to carry out overlapping detection, outputting an overlapping relation matrix, dividing the overlapping relation matrix into a normal scheduling information set and each conflict scheduling information set, acquiring machine learning model training parameters of each conflict scheduling information entry, analyzing to obtain model historical stability indexes and model generalized credibility indexes of the machine learning model, and screening according to the model historical stability indexes to obtain each conflict scheduling execution instruction so as to execute bus network intelligent scheduling, thereby solving the technical problems that different scheduling information has deviation and inconsistency in the state cognition level in the prior art.

Inventors

WANG YU
ZHU HONG
CAI SHUZHOU

Assignees

厦门磁北科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260210

Claims (10)

1. The intelligent dispatching method of the public transport network based on the machine learning model is characterized by comprising the following steps of: Receiving network perception data, generating a network operation situation snapshot when detecting that the network perception data has an abnormal evolution trend, and transmitting the network operation situation snapshot to a scheduling information coordination processing unit to generate each scheduling information entry; Each scheduling information entry is called to carry out overlapping detection, an overlapping relation matrix is output, a normal scheduling information set and each conflict scheduling information set are obtained through division, and each conflict scheduling information entry with a certain overlapping relation is included in the conflict scheduling information set; The method comprises the steps that machine learning model training parameters of conflict scheduling information entries are obtained, the machine learning model is used for generating scheduling information entries according to bus network operation situation data, model historical stability indexes and model generalized reliability indexes of the machine learning model are obtained through analysis, conflict scheduling execution instructions are obtained through screening according to the model historical stability indexes and model generalized reliability parameters of the machine learning model, the model historical stability parameters comprise output distribution entropy, maximum posterior probability, local prediction gradient norms and hidden space average distance, and the model generalized reliability parameters comprise historical prediction consistency rate, scheduling proposal rollback rate, prediction symbol inversion density and distribution deviation response slope; And managing the executed conflict dispatching execution instructions by dispatching validation identifiers, and controlling the incorporation and exclusion of dispatching information entries in the subsequent dispatching period based on the dispatching validation identifiers so as to complete intelligent dispatching of the public transportation network.
2. The intelligent dispatching method of public transportation network based on machine learning model of claim 1, wherein the abnormal evolution trend is detected by the following specific detection method: In a preset scheduling period, carrying out state field evolution analysis on the wire network sensing data subjected to time alignment treatment, and evaluating the change amplitude and the change direction of each wire network state field between adjacent scheduling periods to obtain a change amplitude result and a change direction judging result of each wire network state field; When the change amplitude result of at least one wire network state field exceeds the corresponding preset state stability interval and the change direction judging result is forward evolution or oscillation evolution, and/or the wire network state fields of a plurality of wires present cooperative abnormal offset in the same scheduling period and the change direction judging result is forward evolution or oscillation evolution, judging that the wire network sensing data has abnormal evolution trend.
3. The intelligent dispatching method of public transportation network based on machine learning model of claim 1, wherein the method for generating the network operation situation snapshot comprises the following steps: In a preset scheduling period, receiving network sensing data from different functional modules, and mapping the network sensing data to a unified scheduling time window according to time stamps; performing time alignment processing on the mapped net sensing data, and eliminating historical data and look-ahead data which do not belong to the current scheduling period; carrying out state abstraction processing on the multi-source wire network perceived data after time alignment, and extracting unified wire network state fields, wherein the wire network state fields comprise a departure interval field, a section running density field, a site load change field and a capacity configuration state field; and constructing a net running state basic data set based on the unified net state field, and solidifying the net running state basic data set into a net running situation snapshot corresponding to the scheduling period.
4. The intelligent dispatching method of public transportation network based on machine learning model of claim 1, wherein the output overlap relation matrix comprises the following specific steps: Performing resource identification dimension analysis on each scheduling information entry, wherein the resource identification dimension comprises an action object dimension, a line dimension and a time window dimension; constructing an action relation mapping table among the scheduling information entries based on the action object dimension; When any two pieces of scheduling information entries have intersection on at least one action object dimension, marking as overlapping relation; and expressing the overlapping relation among all the dispatching information entries in a matrix form to obtain an overlapping relation matrix.
5. The intelligent scheduling method of public transportation network based on machine learning model of claim 1, wherein the method for obtaining normal scheduling information set and each conflict scheduling information set comprises the following specific steps: Based on the overlapping relation matrix, carrying out association analysis on the scheduling information entries, and dividing the scheduling information entries which do not form overlapping relation with other scheduling information entries into a normal scheduling information set; merging the scheduling information entries with at least one overlapping relation to each other into the same conflict scheduling information set, thereby obtaining each conflict scheduling information set; the scheduling information entry in each conflict scheduling information set of each conflict scheduling information set has a common conflict action object.
6. The intelligent dispatching method of public transportation network based on machine learning model of claim 1, wherein the method for obtaining model historical stability index of machine learning model comprises the following steps: Acquiring model historical stability parameters of a machine learning model corresponding to each scheduling information entry in a historical scheduling period; Acquiring a preset model history stability benchmark set, performing comparison analysis with model history stability parameters to obtain each model history stability processing value, introducing corresponding weights based on each model history stability processing value, and performing multiplication processing and superposition to obtain a model history stability index for representing the prediction stability degree of the machine learning model; the model historical stability benchmark set comprises an output distribution entropy benchmark, a maximum posterior probability benchmark, a local prediction gradient norm benchmark and a hidden space average distance benchmark.
7. The intelligent dispatching method of public transportation network based on machine learning model of claim 1, wherein the model generalizes reliability index, and the specific acquisition method is as follows: Obtaining model generalization credibility parameters of machine learning models corresponding to the scheduling information entries in a historical scheduling period; Obtaining a model generalization credibility benchmark set preset in a database, comparing and analyzing the model generalization credibility benchmark set with model generalization credibility parameters to obtain model generalization credibility processing values, introducing corresponding weighting factors into the model generalization credibility processing values to carry out multiplication processing and then superposing the model generalization credibility processing values to obtain model generalization credibility indexes for representing the consistency of machine learning model cross-scheduling-period prediction; The model generalization reliability benchmark set includes a historical prediction compliance benchmark, a scheduling proposal backoff benchmark, a prediction sign inversion density benchmark, and a distribution deviation response slope benchmark.
8. The intelligent dispatching method of the public transportation network based on the machine learning model of claim 1, wherein the method for obtaining each conflict dispatching execution instruction is as follows: Respectively acquiring corresponding model historical stability indexes and model generalized credibility indexes for each scheduling information entry in each conflict scheduling information set; comparing the model historical stability index and the model generalized credibility index with corresponding thresholds, screening scheduling information entries above the corresponding thresholds, and taking the scheduling information entries as candidate scheduling information entries, wherein the corresponding thresholds comprise a model historical stability threshold and a model generalized credibility threshold; when no candidate scheduling information entry exists in the conflict scheduling information set, any scheduling information entry in the conflict scheduling information set is not executed; when a plurality of candidate scheduling information entries still exist in the conflict scheduling information set, calculating a comprehensive confidence value for each candidate scheduling information entry; Selecting the dispatching information entry with the highest comprehensive confidence value as a conflict dispatching execution instruction of the conflict dispatching information set, counting the conflict dispatching execution instruction corresponding to each conflict dispatching information set and the dispatching information entry of the normal dispatching information set, jointly serving as each conflict dispatching execution instruction, and outputting the intelligent dispatching for the public transportation network.
9. The intelligent dispatching method of public transportation network based on machine learning model of claim 8, wherein the method for calculating comprehensive confidence value comprises the following steps: Performing difference processing based on the model historical stability index and the model historical stability threshold to obtain a model historical stability margin, performing difference processing on the model generalization credibility index and the model generalization credibility threshold to obtain a model generalization credibility margin, combining the model historical stability margin and the model generalization credibility margin, performing ratio processing with a preset reference margin set to obtain margin processing values, introducing the margin processing values into a corresponding margin weight set to perform multiplication processing, and performing superposition processing to obtain a comprehensive confidence value for representing the execution reliability of the scheduling information vocabulary entry; The reference margin set comprises a model historical stability margin reference value and a model generalization credibility margin reference value; The set of margin weights includes a model historical stability benchmark margin weight and a model generalization reliability benchmark margin weight.
10. The intelligent scheduling method of public transportation network based on machine learning model of claim 1, wherein the control processing is carried out on the merging and the exclusion of the scheduling information entries in the subsequent scheduling period, and the specific method is as follows: Performing scheduling validation identification management on each conflict scheduling execution instruction, marking each executed conflict scheduling execution instruction as a scheduling validation entry, and marking each conflict scheduling execution instruction which is not executed as a scheduling non-validation entry; After generating a line network operation situation snapshot and entering a corresponding scheduling period, establishing a scheduling coordination freezing constraint, and controlling the incorporation and exclusion of scheduling information entries in a subsequent scheduling period; The scheduling coordination freezing constraint comprises that each scheduling information entry is generated based on the same network operation situation snapshot, and the entries marked as scheduling effective entries do not participate in repeated overlapping detection.

Description

Intelligent dispatching method for public transport network based on machine learning model Technical Field The invention relates to the technical field of data processing, in particular to an intelligent dispatching method of a public transportation network based on a machine learning model. Background With the continuous expansion of the scale of the urban public transportation network and the dynamic change of the travel demand structure, the running state of the public transportation network presents obvious time variability and coupling, so that the intelligent dispatching of the public transportation network is required to be realized based on multidimensional running data. In the prior art, the departure interval, the running rhythm and the regional capacity configuration of a bus are generally optimized and adjusted by constructing a data-driven dispatching model, for example, passenger flow data, station information data, arrival-departure data and vehicle departure data are collected, a waiting time model and an interpolation model are constructed to describe the waiting time of passengers and the running speed among stations in different time periods, and then the waiting time of the passengers, the speed of a station section and the planned departure interval are jointly trained based on a multi-agent reinforcement learning algorithm, so that the departure time of each vehicle in each line is obtained, and the dispatching optimization of the bus is realized. Or judging the time slot state, when the time slot state is identified as the peak time slot, acquiring the running speed of the bus in each area, constructing congestion coefficients to generate a plurality of groups of scheduling schemes, carrying out simulation test on each scheduling scheme by means of a digital twin model of the vehicle scheduling, and constructing an improvement index according to the change of the congestion coefficients, thereby screening the optimal scheduling scheme with highest feasibility from the plurality of scheduling schemes. The public transportation intelligent scheduling method based on multi-agent reinforcement learning is disclosed in China patent publication No. CN116307448B, and comprises the steps of constructing a waiting time model and an interpolation model based on passenger flow data, station information data, arrival-departure data and vehicle departure data, obtaining waiting time of each passenger based on the waiting time model, obtaining speeds of different time periods between stations based on the interpolation model, training the waiting time of each passenger and the speeds and planned departure intervals of different time periods between stations based on a multi-agent reinforcement learning algorithm, obtaining final departure intervals, further obtaining departure time of each vehicle of each line, and realizing public transportation vehicle scheduling. The above technology is found to have at least the following technical problems: In the prior art, intelligent dispatching of a public transportation network is generally realized by modeling and analyzing multi-source data such as passenger flow characteristics, running speed, regional state and the like, and generating dispatching decisions by means of a reinforcement learning model, a regional state model or a digital twin model, however, in the technical scheme, a coordinated checking process between a dispatching information generating process and dispatching information generally presents a serial coupling processing mode on a system architecture, when a plurality of dispatching information are generated in a concentrated manner in a short time, the dispatching information is influenced by factors such as inconsistent multi-source data acquisition time stamps, delayed model reasoning and decision calculation, feedback lag of dispatching execution results and the like, the actual running state of the public transportation network is objectively changed by the dispatching information which is generated and executed in advance, and the follow-up dispatching information is still generated and judged based on state data or historical prediction results at the previous moment, so that different dispatching information has deviation and inconsistency on a state cognition level. Disclosure of Invention In order to solve the technical problem that in the prior art, due to the fact that the updating of the running state of the wire network is asynchronous with the generation of the scheduling decision, the deviation and the inconsistency of different scheduling information exist on the state cognition level, the embodiment of the invention provides a bus network intelligent scheduling method based on a machine learning model. The technical proposal is as follows: A bus network intelligent scheduling method based on a machine learning model comprises the steps of receiving network perception data, generating a network running situation snap