CN-121998165-A - Rail transit vehicle multi-target scheduling method based on V2X architecture

Abstract

The invention discloses a rail transit vehicle multi-target scheduling method based on a V2X framework, which comprises the following steps of S1 vehicle-road-cloud cooperative state acquisition, S2 scene passenger flow prediction based on multi-source data, S3 overall multi-target scheduling optimization facing prediction requirements, S4 grouping and ungrouping scheduling based on prediction and real-time passenger flow, S5 operation feedback and self-adaptive parameter adjustment, and self-adaptive adjustment mechanism based on operation evaluation indexes, wherein the prediction model structure or super parameters of S2, the target weight of S3 and the passenger flow threshold, the congestion degree threshold and the safety inter-vehicle distance coefficient of S4 are respectively adjusted according to different deviation modes, so that all links of prediction, scheduling and grouping are cooperatively converged in a long-term operation process, and systematic performance degradation of a traditional fixed parameter model after the passenger flow structure is changed is avoided. The invention improves the dispatching refinement level, can realize the scene modeling of the passenger flow, improves the prediction accuracy of the passenger flow in a complex scene, and has robustness and adaptability in long-term operation.

Inventors

• YUE HUIJUN
• FENG FAN

Assignees

• 北京工业大学

Dates

Publication Date

20260508

Application Date

20251225

Claims (5)

1. 1. The rail transit vehicle multi-target scheduling method based on the V2X architecture is characterized by comprising the following steps of: s1, acquiring a vehicle-road-cloud cooperative state; The method comprises the steps of setting a GPS/BDS antenna, an RTK antenna, a 5G antenna, a V2V antenna, a vehicle-mounted reader, an environment sensing sensor, an inertial measurement unit IMU, a vehicle-mounted unit and a data processing unit on the vehicle side by means of a V2X architecture, arranging a track electronic tag, an RTK base station and a 5G base station device on the road side, fusing satellite positioning, RTK difference, IMU and trackside tag information by the vehicle data processing unit, periodically estimating the high-precision position, speed, acceleration, attitude, inter-vehicle distance and track occupation state of the vehicle, transmitting the running state to a cloud scheduling platform through a 5G communication network by the vehicle-mounted unit according to a preset reporting period, and simultaneously accessing external environment data and running data by the cloud scheduling platform to construct a full-line continuous and high-precision running state diagram and a data warehouse, and providing a unified data basis for follow-up passenger flow prediction and scheduling optimization; s2, predicting the scene passenger flow based on the multi-source data; On the basis of a data warehouse constructed by the S1, a cloud scheduling platform fuses external factors, historical operation data and real-time operation data, a typical scene is divided by a combination of site level, date type, time period and weather/special events, a corresponding passenger flow time sequence is extracted from the data warehouse to form a scene passenger flow time sequence, EMD is decomposed in an empirical mode to the passenger flow time sequence under each typical scene to obtain intrinsic mode components of different time scales, then the EMD characteristics are clustered by K-Means clustering improved by a genetic algorithm, a passenger flow mode label is automatically extracted, the passenger flow time sequence, the passenger flow mode label, date type and weather type auxiliary characteristics are used as input, an LSTM prediction model is constructed, predicted passenger flows of each site or section in a plurality of time windows in the future are output, the S2 is the predicted passenger flow with passenger flow mode label structured scene information, and is directly input into a multi-target scheduling model of the S3 for adjusting weight and constraint, and when the predicted error under a certain typical scene continuously exceeds a preset threshold value in a plurality of evaluation periods, retraining or super-parameter under the scene is triggered to be adjusted to ensure the effectiveness of the passenger flow model; S3, global multi-objective scheduling optimization facing to prediction requirements; Based on the future passenger flow prediction of each station or section obtained in the step S2, the cloud dispatching platform combines the line topology structure, the number and grouping capacity of vehicles, the platform and foldback capacity and the safety interval constraint, and provides a unified evaluation standard for comparing the advantages and disadvantages of the schemes according to different possible dispatching schemes in the same prediction time domain, so as to output an optimal dispatching scheme; s4, grouping and ungrouping scheduling based on prediction and real-time passenger flow; in the step, in the actual running process of the train, each candidate window is subjected to local self-adaptive judgment to determine whether to actually execute grouping or ungrouping operation; S5, running feedback and self-adaptive parameter adjustment; After the scheduling schemes determined in the step S3 and the step S4 are executed, the cloud scheduling platform calculates average waiting time, section crowding degree, unit customer kilometer energy consumption, positive point rate, grouping/ungrouping times and failure rate operation indexes according to preset evaluation periods and compares the operation indexes with respective target intervals; The cloud scheduling platform carries out scene configuration on the weights of the multi-objective cost function in the step S3 and the passenger flow threshold and the crowding threshold in the step S4 according to the passenger flow mode labels output by the step S2, and improves the energy consumption objective weight and the passenger flow threshold properly in the night low-valley mode so as to reduce unnecessary grouping behaviors and idle running.
2. 2. The multi-objective scheduling method for rail transit vehicles based on the V2X architecture according to claim 1, wherein in S3, a multi-objective cost function J is constructed, and the following indexes are considered: Average waiting time, time in transit and time in late condition of passengers; Traction energy consumption and braking energy recovery energy consumption indexes; average level, peak value and congestion duration of the degree of congestion of carriage; Cost class indexes of unit kilometer operation cost and vehicle turnover efficiency; the comprehensive target write is: ; Wherein, the Is an adjustable weight; Under the condition of the cost function and related constraint conditions, the space-time prediction passenger flow given by the S2 is used as input, optimization algorithms such as an ant colony algorithm and the like are adopted to search the departure frequency, the minimum tracking interval, the vehicle operation path and the stop plan decision variables in the future prediction time domain, candidate scheduling schemes are continuously generated in the iteration process, each scheme is evaluated by using the cost function J, schemes with better comprehensive performance are gradually reserved, a group of scheduling results in a time-energy consumption-crowdedness-cost compromise mode are finally obtained, the scheduling results form a global reference operation diagram and a vehicle operation scheme facing the prediction requirements, the departure frequency and the minimum tracking interval of each line and each section in different time windows, the operation path, the departure time and the stop plan of each vehicle are formed, and grouping or ungrouping is carried out in what site or section and in what time window, namely the candidate grouping/ungrouping window and the priority thereof.
3. 3. The V2X architecture-based rail transit vehicle multi-objective scheduling method according to claim 1, wherein in S4, in each candidate grouping/ungrouping window, the cloud scheduling platform comprehensively considers the following conditions: Predicting the passenger flow condition, namely judging whether the passenger flow of the section exceeds a preset passenger flow threshold value or is continuously lower than the threshold value in a period of time according to the prediction results of S2 on a plurality of current and subsequent time windows; Judging whether the current and short-term crowdedness of the candidate vehicle is close to or exceeds a preset crowdedness threshold according to the real-time passenger carrying information of the vehicle; Safety conditions, namely, calculating the safety inter-vehicle distance according to the speed and the inter-vehicle distance of the vehicle provided by S1 When the actual distance between vehicles is not less than And when the signal system allows related operation, the safety condition is considered to be met; when the conditions are met together, the system confirms that the grouping or ungrouping operation is carried out in the candidate window, a control instruction is issued to related vehicles through V2X, speed synchronization, coupling engagement or unlocking and operation mode switching among the vehicles are achieved, and if the key conditions are not met, the vehicles continue to execute according to the single vehicle operation plan/the existing grouping plan given by S3.
4. 4. The V2X architecture-based rail transit vehicle multi-objective scheduling method according to claim 3, wherein S4 makes a local decision on whether to execute each candidate grouping/ungrouping window by using three types of information of prediction passenger flow-real-time crowding-speed related safety inter-vehicle distance in the global reference operation diagram framework given in S3, and realizes the adaptive adjustment of the operation performance of the section level on the premise of not damaging the global operation diagram overall structure.
5. 5. The V2X architecture-based rail transit vehicle multi-objective scheduling method according to claim 1, wherein the rule pair in S5 comprises: 1) Weight of multi-objective cost function in S3 ; 2) S4, passenger flow threshold, congestion degree threshold and safety inter-vehicle distance coefficient ; 3) S2, partial structure or super parameter of the passenger flow prediction model.

Description

Rail transit vehicle multi-target scheduling method based on V2X architecture Technical Field The invention relates to the technical field of rail transit, in particular to a rail transit Vehicle multi-target scheduling method based on a V2X (Vehicle-to-evanescence) architecture, which is suitable for a novel medium-low traffic urban rail transit line with independent road rights and supporting unmanned and variable marshalling operation. Background The medium-low traffic track traffic system has low construction cost and flexible line layout, and is an important component for relieving urban congestion and perfecting public traffic network. The conventional system which uses a train as a basic operation unit and uses a fixed operation diagram to start at fixed points gradually exposes the following problems: 1. Insufficient state sensing accuracy In the existing system, a considerable part of lines still mainly depend on the trackside signal equipment and the track circuit partition to detect and control the train, the length of the detection section is usually tens to hundreds of meters, the dispatching center can only obtain discrete position information such as 'whether the section is occupied' and the like, and the position, the speed and the acceleration of the train are difficult to continuously describe. Under the working conditions of high-density running, temporary train adding and starting, obvious passenger flow fluctuation and the like, the coarse granularity perception is difficult to support the fine scheduling of the inter-vehicle distance, the speed curve and the grouping behavior, and the safety redundancy is often realized only by increasing the interval, so that the utilization rate of the transportation energy is lower. 2. Underutilization of data and limited ability to predict passenger flow The passenger flow prediction mainly uses historical average values and experience rules, a system modeling and dynamic prediction mechanism is lacked for multi-source factors such as site grades, date types, time periods, weather, large activities and the like, passenger flow changes are difficult to reflect in time under the scenes such as holiday passenger flow surge, extreme weather or temporary activities and the like, and the passenger flow changes are easy to cause delay of peak time capacity allocation and low vehicle utilization rate in valley time. 3. Global scheduling target single The traditional train running diagram and scheduling scheme mainly uses time indexes such as quasi-point rate, minimum tracking interval and the like, and takes departure time, tracking interval and retracing capacity as main constraints. There is a lack of a mechanism for uniformly incorporating passenger travel time, energy consumption, crowdedness and operation cost into the same optimization framework for trade-off. When passenger flow fluctuates greatly or the transport capacity is intense, it is difficult to make a dynamic balance between "quasi point rate-energy consumption-passenger experience-operation cost". 4. Multimode operation and limited marshalling scheduling capability When the local congestion or idle running needs to be relieved through grouping or ungrouping, the existing system is manually completed in a vehicle section or a foldback station according to a preset scheme, and a grouping scheme and an operation diagram are relatively static. The section-level grouping and ungrouping mechanism based on real-time running state and predicted passenger flow automatic triggering is lacking, the self-adaptive switching of the bicycle running and the grouping running is difficult to realize, and the advantages of variable grouping are not fully exerted. 5. Lack of operational feedback and adaptive scheduling capabilities The existing passenger flow prediction and scheduling method mostly adopts offline modeling and fixed parameter configuration, real-time operation data is mainly used for post statistics evaluation, and the lack of an online feedback and parameter correction mechanism easily causes gradual aging of a model, mismatching of a scheduling strategy and an actual operation environment, and the problems of scheduling performance reduction and difficulty in continuous optimization of energy consumption occur in a complex scene. In summary, the prior art has not organically cooperated with high-precision state sensing, scene passenger flow prediction, multi-objective scheduling optimization, variable marshalling scheduling based on safety constraint and operation feedback self-adaptive mechanism under a unified V2X architecture, so that a closed-loop multi-objective scheduling system of prediction-optimization-execution-feedback-re-optimization is constructed, and the comprehensive potential of the novel medium-low traffic track traffic system in the aspects of safety, energy utilization, energy consumption control and passenger experience is difficult to fully release. Disclosure of Invent