CN-121999617-A - AI-based vehicle scheduling and electricity changing path collaborative optimization method and system

CN121999617ACN 121999617 ACN121999617 ACN 121999617ACN-121999617-A

Abstract

The invention discloses a vehicle dispatching and electricity conversion path collaborative optimization method and system based on AI, wherein the method comprises the following steps of collecting multi-source heterogeneous data, including vehicle states, traffic road conditions and demand thermodynamic diagrams, identifying vehicle silting points, shortage points and electricity conversion demand points, abstracting the silting points, the shortage points and the electricity conversion demand points into space-time network nodes in a unified mode, constructing a mixed integer planning model aiming at minimizing operation and maintenance total cost and maximizing operation and benefit, inputting a real-time city state into a deep reinforcement learning model, extracting topological characteristics of road network and task points by using a graph neural network, dynamically calculating weights of all task nodes through an attention mechanism, outputting an optimal task sequence, converting the optimal task sequence into a work order instruction, transmitting the work order instruction to a mobile terminal, and feeding back a monitoring execution result to a credit evaluation system so as to update the optimization model. The invention can reduce the total driving mileage and labor cost and improve the comprehensive availability of the vehicle.

Inventors

FANG LI
LI YONG
CHEN HUAMING
LIU FANGZHEN

Assignees

福信富通科技股份有限公司

Dates

Publication Date: 20260508
Application Date: 20260325

Claims (10)

1. The AI-based vehicle scheduling and electricity changing path collaborative optimization method is characterized by comprising the following steps: Collecting multi-source heterogeneous data in real time, wherein the multi-source heterogeneous data comprise vehicle states, traffic road conditions and demand thermodynamic diagrams, and identifying vehicle siltation points, shortage points and power conversion demand points; unifying and abstracting the siltation points, the shortage points and the electricity change demand points into space-time network nodes, and constructing a mixed integer programming model which aims at minimizing the total operation and maintenance cost and maximizing the operation benefit; inputting the real-time city state constructed based on the multi-source heterogeneous data into a deep reinforcement learning model, extracting topological features of road networks and task points by using a graph neural network, dynamically calculating the weight of each task node through an attention mechanism, and outputting an optimal task sequence comprising scheduling and power conversion mixed operation; And converting the optimal task sequence into a work order instruction, transmitting the work order instruction to a mobile terminal, and monitoring and executing results and feeding back the results to a credit evaluation system to update an optimization model.
2. The method of claim 1, wherein the objective function of the mixed integer programming model is configured to minimize a weighted sum of total operating cost and maximize operating benefit; The method comprises the steps of minimizing the running cost of an operation and maintenance person from a current node to a next node, minimizing the time cost of the operation and maintenance person for executing a task and minimizing the processed cost of a battery at a task point; The constraint conditions of the mixed integer programming model at least comprise vehicle flow conservation constraint, task allocation uniqueness constraint, operation and maintenance vehicle carrier capacity constraint, single task maximum battery processing capacity constraint and task execution time window constraint.
3. The method of claim 2, wherein the connection edges of the spatio-temporal network nodes have dynamic weights: Acquiring the congestion level and the average running speed of each road section in real time; And dynamically updating the transit time cost among the network nodes according to the average running speed, and adjusting the weight of the corresponding edge of one road section when the congestion level of the road section exceeds a preset threshold value.
4. The method according to claim 1, wherein in the step of extracting topology features of road network and task points by using the graph neural network, node feature vectors and edge feature vectors of the graph structure are constructed: The node characteristic vector comprises geographic position coordinates of the node, real-time states of the current vehicle number and the space number, and a demand fluctuation rate calculated based on historical data; The edge feature vector comprises Euclidean distance and path distance between nodes, connection strength calculated based on historical flow and real-time traffic congestion index; the graph neural network generates a space embedded vector of each task node by aggregating the characteristic information of the neighbor nodes and combining the edge characteristic vectors.
5. The method according to claim 1, wherein the weight calculation method of each task node is as follows: constructing a query-bond-value structure, and calculating the attention score by using a Softmax function; adopting a multi-head attention mechanism to pay attention to different subspace characteristics respectively, and generating a final context vector through linear transformation after output splicing of each head; the context vector is used to indicate that the model is preferentially focused on power-change demand points with high urgency or high-priority dispatch fouling points at the current time step.
6. The method of claim 1, wherein the reward function of the deep reinforcement learning model comprises an efficiency reward, a quality reward, a cost reward, and a constraint penalty term: The efficiency rewards are inversely related to the total time spent on task completion; The quality rewards are positively correlated with the vehicle supply and demand balance degree and the battery electric quantity level after the task point processing; The cost rewards are inversely related to the total driving mileage of the operation and maintenance vehicle; the constraint penalty term gives a negative feedback value when the generated action violates a time window or capacity constraint.
7. The method of claim 1, wherein the action space of the deep reinforcement learning model is configured as a hybrid action space comprising a discrete action dimension, a continuous action dimension, and a hybrid scheduling dimension; The discrete action dimension is used for task allocation decision and is defined as an N multiplied by M matrix, wherein N is the number of task nodes, M is the number of operation and maintenance vehicles and is used for indicating a specific vehicle to respond to a request of a specific node; the continuous action dimension is used for path planning decision and outputting a specific path coordinate point sequence; The hybrid scheduling dimension is used for battery scheduling decisions while outputting the number of batteries that need to be replaced and a suggested battery remaining capacity threshold.
8. The method of claim 1, further comprising a spatiotemporal alignment and fusion process for multi-source heterogeneous data: In the time dimension, aligning the data with different frequencies to a unified time slice by adopting a weighted minimum absolute deviation method; in the space dimension, searching and matching the satellite positioning coordinates of the vehicle to road network nodes by using a K-dimensional tree data structure; And carrying out feature level fusion on the aligned time features and space features through the multi-layer perceptron, and taking the time features and the space features as input states for decision solving.
9. The method of claim 1, wherein the optimal task sequence is configured as a collaborative path: on the way of the operation and maintenance vehicle executing a dispatch task that transports the vehicle from a siltation point to a shortage point, can be assigned a shutdown demand point; and during stopping, the operation and maintenance vehicle uses the loaded full-charge battery to replace the low-power vehicle at the power-change demand point, and then continues to run to finish the delivery of the dispatching vehicle, so that various tasks are finished in series in a single trip.
10. AI-based vehicle scheduling and power-change path co-optimization system, characterized in that it is adapted to implement the method according to any of claims 1-9, comprising: The multi-source data perception module is used for acquiring multi-source heterogeneous data in real time, wherein the data comprise vehicle states, traffic road conditions and demand thermodynamic diagrams, and identifying vehicle siltation points, shortage points and power conversion demand points according to the data; The AI collaborative decision engine is used for abstracting the siltation points, the shortage points and the power conversion demand points into space-time network nodes uniformly, constructing a mixed integer programming model which aims at minimizing the total operation and maintenance cost and maximizing the operation benefit, inputting the real-time city state into the deep reinforcement learning model, extracting the topological characteristics of the road network and the task points by using the graph neural network, and dynamically calculating the weight of each task node by using the attention mechanism so as to output an optimal task sequence comprising the scheduling and power conversion mixed operation; and the scheme execution and feedback module is used for converting the optimal task sequence into a work order instruction to be issued to the mobile terminal, and monitoring and feeding back an execution result to the credit evaluation system to update the optimization model.

Description

AI-based vehicle scheduling and electricity changing path collaborative optimization method and system Technical Field The invention belongs to the technical field of intelligent urban traffic management and shared traveling, and particularly relates to a vehicle scheduling and electricity changing path collaborative optimization method and system based on AI. Background With the popularization of the shared electric bicycle, the shared electric bicycle is used as an important supplement for urban short-distance travel, and greatly facilitates the life of citizens. However, operation of the shared electric bicycle faces two major core challenges, namely, uneven space-time distribution of the vehicle, such as accumulation of vehicles at subway stations in early peaks and vehicle shortage in business circles, caused by tidal effect, and limited battery endurance of the vehicle, and periodic battery replacement or charging is needed to ensure usability. To maintain the normal operation of the system, the operating enterprises need to frequently perform vehicle scheduling and power-changing operations. Currently, the prior art is mainly focused on single-dimensional optimization. For example, chinese patent application publication No. CN114862115A discloses a method of dispatching a shared vehicle by analyzing vehicle flow data, constructing a beta distribution model of the vehicle's ride-out probability, and randomly sampling based on the predicted beta distribution to determine a target parking fence that requires an increase in dispatching amount. The scheme focuses on solving the problem of demand prediction of where the vehicle is tuned by using a statistical model, and optimizing the spatial configuration of the vehicle through probability distribution. For another example, chinese patent application publication No. CN120911851a discloses an intelligent scheduling method for shared electric bicycles, which uses a prediction model to obtain a site supply and demand difference value, calculates a scheduling cost matrix, and optimizes a scheduling policy based on a preset scheduling principle and an objective function. The scheme introduces a cost matrix and a prediction model, quantifies the scheduling cost to a certain extent, and achieves the response to the supply-demand difference. In a practical urban complex operation and maintenance scenario, the scheme still has the following limitations: In the existing scheme, vehicle scheduling is mostly regarded as an independent optimization target, the power change requirement is not brought into the same decision frame, the operation tasks are split, and the operation and maintenance vehicle path is repeated due to the lack of the characteristic of cooperative optimization, so that the idle running rate is high. In the face of real-time change of urban traffic road conditions, sudden weather conditions or instantaneous order surge, the traditional operation optimization algorithm based on fixed threshold or off-line calculation often has response lag, and real-time self-adaptive adjustment is difficult to carry out in the executing process. Therefore, a solution capable of breaking the operation barriers of scheduling and power conversion, deeply sensing urban dynamics by using an advanced artificial intelligence algorithm and generating a collaborative optimization path in real time is urgently needed, so as to achieve the dual goals of operation, maintenance, cost reduction, efficiency improvement and service quality improvement. Disclosure of Invention The invention provides a vehicle scheduling and electricity changing path collaborative optimization method and system based on AI, which are used for constructing urban space-time states by fusing multi-source heterogeneous data in real time, abstracting vehicle siltation, shortage and electricity changing requirements into network nodes in a unified way, and generating a collaborative path sequence considering scheduling and electricity changing by utilizing a deep reinforcement learning model fusing a graph neural network and an attention mechanism, so as to solve the problems that resource waste is caused by mutual splitting of scheduling and electricity changing operation in the prior art, a static decision mechanism cannot adapt to urban dynamic complex environments, and global operation and maintenance cost is difficult to realize due to local optimization. In order to solve the technical problems, on the one hand, the invention provides an AI-based vehicle scheduling and electricity changing path collaborative optimization method, which comprises the following steps: collecting multi-source heterogeneous data in real time, wherein the multi-source heterogeneous data comprise vehicle states, traffic road conditions and demand thermodynamic diagrams, and identifying vehicle siltation points, shortage points and power conversion demand points according to the multi-source heterogeneous data; unifying an