A vehicle charging path planning method and system

Abstract

The application discloses a kind of vehicle charging path planning method and system.The method obtains vehicle state vector, charging node data and historical charging service data, predicts the available probability of each charging node, waiting time, serviceable time window and energy supplement capacity, and constructs candidate charging node set;Based on the candidate charging node set, a space-time extended charging map is constructed, and the edge cost is determined according to the distance cost, time cost, environmental risk cost, predicted energy consumption cost and energy supplement benefit;Under the constraints of remaining power safety, task time limit and charging node serviceable time window, search for candidate charging path, and determine the target charging path according to the comprehensive objective function;After execution, the execution feedback data is written back, which is used for state updating and model correction in the next planning cycle.The application can improve the accuracy, adaptability and reliability of charging path planning in dynamic environment.

Description

Technical Field

[0001] This invention relates to the field of vehicle route planning and intelligent charging scheduling technology, and more particularly to a vehicle charging route planning method and system. More specifically, this invention relates to a joint optimization method and system for charging routes under dynamic road environments, changes in node availability states, and vehicle energy constraints, capable of optimizing routes based on vehicle state vectors. Charging node availability prediction results and the spacetime extension charging diagram It enables candidate charging node selection, candidate path search, target charging path solution, and online replanning. Background Technology

[0002] With the increasing number of electric vehicles, unmanned delivery vehicles, inspection vehicles, special-purpose vehicles, and other vehicles that rely on onboard batteries for operation, these vehicles often need to plan reasonable driving and charging routes under limited remaining battery power to ensure timely completion of tasks and reduce the risk of power loss en route when performing tasks such as transportation, inspection, delivery, and emergency support. Therefore, how to generate safe, efficient, and executable charging routes for vehicles under the combined effects of road network, traffic conditions, environmental changes, and dynamic changes in charging resources has become a key technical issue in intelligent vehicle scheduling and operation control.

[0003] In existing technologies, common charging route planning schemes are typically based on static maps and fixed charging station information, employing shortest path algorithms, minimum time path algorithms, or simple remaining battery threshold determination strategies. When low battery is detected, the system searches for the nearest or least costly charging station from the current vehicle location and generates a driving route accordingly. While such methods can meet certain application requirements in static or minimally changing scenarios, they still have many shortcomings when facing real-world dynamic environments.

[0004] On the one hand, existing technologies typically treat charging nodes as statically available nodes, assuming that each charging node is always in a serviceable state during planning, or simply make a judgment based on the current idle information, failing to combine historical charging service data, current traffic activity levels, and time-varying patterns to assess the probability of node availability in the future planning time domain. Waiting time Service Window and node power supply capacity Unified forecasting is performed. As a result, when vehicles actually arrive at the target charging node, there are often situations such as excessively long queuing times, node failures, closed service windows, or insufficient available power, causing the original planned route to become invalid or its performance to drop significantly.

[0005] On the other hand, existing technologies typically use ordinary road maps for path searching, rarely explicitly incorporating the vehicle's time and battery status during its journey into the graph model simultaneously. While some existing solutions consider remaining battery constraints, they usually only perform post-hoc feasibility checks after the path is solved, or use a single threshold to determine whether the target charging station can be reached. This lacks a systematic model of the coupled relationship between "time, battery status, and node status," making it difficult to accurately express the true charging feasibility and path advantages / disadvantages when the vehicle arrives at different nodes at different times.

[0006] Furthermore, existing technologies primarily focus on minimizing distance or time when evaluating routes, rarely considering the impact of factors such as traffic congestion, weather, gradient, abnormal road conditions, and weak communication coverage on risk and cost. and predicted energy consumption The impact is also less considered, as is the cost of waiting for charging. Path deviation cost and the benefit items in path evaluation The combined effect of these factors means that the obtained path may not be the overall optimal path in complex environments.

[0007] Furthermore, existing path planning results are typically generated all at once during the planning phase, lacking a rolling time-domain-based approach. The existing online update mechanism is often unable to keep up with changes in road congestion, weather deterioration, increased instantaneous energy consumption of vehicles, changes in waiting time at target charging nodes, or a decrease in the availability probability of nodes when road congestion suddenly changes. or probability of power outage Triggering replanning can easily lead to increased vehicle detours, longer waiting times, and even insufficient battery power to reach the next available charging node.

[0008] In multi-vehicle scenarios, the above problems are even more pronounced. Multiple vehicles may choose the same charging node for charging at similar times. Without a unified time slot allocation and congestion coordination mechanism, this can lead to overload of local charging nodes, increased queues, and a decrease in overall operating efficiency. Traditional path planning methods do not adequately consider the resource competition problem among multiple vehicles and nodes, making it difficult to adapt to the increasingly complex needs of vehicle-road cooperation and group scheduling.

[0009] Therefore, existing technologies suffer from at least the following problems: insufficient prediction of the future state of charging nodes, insufficient coupled modeling of time and energy states, insufficient consideration of dynamic traffic and environmental impacts, insufficient joint optimization of charging benefits, waiting costs, and path deviation costs, and insufficient support for online replanning and multi-vehicle conflict coordination. To address these issues, it is necessary to propose a vehicle charging path planning method and system to improve path feasibility, energy security, charging efficiency, and task completion efficiency in dynamic environments. Summary of the Invention

[0010] The purpose of this invention is to provide a vehicle charging path planning method and system to solve the problems of static charging node states, insufficient time and power coupling modeling, single path evaluation dimension, insufficient replanning capability in dynamic environments, and insufficient multi-vehicle competition coordination capability in the prior art. This enables dynamic prediction of candidate charging nodes, safety constraint search of charging paths, comprehensive optimization solution of target paths, and online adaptive replanning of the operation process.

[0011] To achieve the above objectives, the present invention provides a vehicle charging path planning method, executed by a processor, comprising:

[0012] Acquire vehicle state vectors, charging node data, and historical charging service data; based on the charging node data, historical charging service data, and write-back execution feedback data, use a charging node availability prediction model to predict the availability probability, waiting time, service window, and node power supply capacity of each charging node, and construct a candidate charging node set; based on the candidate charging node set, road network data, traffic state data, and environmental data, construct a spatiotemporal extended charging map with physical location nodes, predicted time, and remaining battery power as node states, and determine the charging node status based on distance cost, time cost, environmental risk cost, predicted energy consumption cost, and benefit items in path evaluation. Edge cost; on the spatiotemporally extended charging map, candidate charging paths are searched using remaining power safety constraints, task time constraints, and charging node service window constraints to obtain a set of candidate charging paths; the target charging path is determined from the set of candidate charging paths based on the comprehensive objective function; after the vehicle executes the charging decision, the actual arrival time, actual waiting time, actual single replenishment power, actual path energy consumption, and actual service status of the charging node are collected as the execution feedback data, and the execution feedback data is written back to the vehicle state construction module and the charging node availability prediction model for state updates and model corrections in the next planning cycle.

[0013] To achieve the aforementioned objective, the present invention also provides a vehicle charging route planning system, which includes:

[0014] The vehicle state construction module is used to obtain the vehicle state vector and the execution feedback data for write-back.

[0015] A charging node availability prediction model is used to predict the availability probability, waiting time, service window and power supply capacity of each charging node based on charging node data, historical charging service data and the execution feedback data, and to construct a candidate charging node set.

[0016] The spatiotemporal mapping module is used to construct a spatiotemporal extended charging map based on the candidate charging node set, road network data, traffic status data, and environmental data, and to determine the edge cost based on distance cost, time cost, environmental risk cost, predicted energy consumption cost, and the benefit item in the path evaluation.

[0017] The path search module is used to search for candidate charging paths on the spatiotemporal extended charging map using remaining power safety constraints, task time constraints, and charging node service window constraints, to obtain a set of candidate charging paths.

[0018] The path determination module is used to determine the target charging path from the candidate charging path set based on the comprehensive objective function;

[0019] The feedback write-back module is used to collect the actual arrival time, actual waiting time, actual single replenishment of electricity, actual path energy consumption, and actual service status of the charging node as the execution feedback data after the vehicle executes the charging decision result. The execution feedback data is then written back to the vehicle status construction module and the charging node availability prediction model for status updates and model corrections in the next planning cycle.

[0020] Compared with the prior art, the present invention has at least the following beneficial effects:

[0021] This invention predicts the availability probability, waiting time, service window, and power supply capacity of charging nodes based on charging node data, historical charging service data, and execution feedback data. It then constructs a candidate charging node set, thereby improving the ability to represent the dynamic state of charging resources and reducing path planning failures caused by unavailable charging nodes, excessively long waiting times, or insufficient power supply capacity. Furthermore, by constructing a spatiotemporally extended charging graph with physical location nodes, predicted time, and remaining battery power as node states, and incorporating distance cost, time cost, environmental risk cost, predicted energy consumption cost, and benefits from path evaluation into the edge cost calculation, this invention enables joint modeling of the temporal evolution, energy evolution, and environmental risks during vehicle operation, thus improving path planning. The results are consistent with the actual operating status. By searching for candidate charging paths under the constraints of remaining power safety, task time limit, and charging node service window, and determining the target charging path based on the comprehensive objective function, it is possible to balance energy replenishment efficiency, path cost, and task completion effect while meeting safety and timeliness requirements. After the vehicle executes the charging decision, the actual arrival time, actual waiting time, actual single replenishment power, actual path energy consumption, and actual service status of the charging node are written back as execution feedback data to the vehicle state vector construction and charging node availability prediction model, thus forming a closed-loop mechanism of planning, execution, feedback, and model correction, improving the system's adaptability, prediction accuracy, and charging path planning reliability in dynamic environments. Attached Figure Description

[0022] Figure 1 This is a flowchart of the vehicle charging path planning method provided by the present invention. Detailed Implementation

[0023] The present invention will now be described in detail with reference to the accompanying drawings.

[0024] Figure 1 This is a flowchart of the vehicle charging path planning method provided by the present invention, such as... Figure 1 As shown, this invention provides a vehicle charging path planning method, which is executed by a processor calling program instructions from memory. The method is applicable to charging path planning scenarios in dynamic road environments, dynamic charging resource environments, and under conditions of limited power. By jointly modeling vehicle status, charging node status, road traffic status, environmental risk status, and energy status, it achieves candidate charging node screening, candidate charging path search, target charging path determination, online replanning, and execution feedback write-back, thereby forming a closed-loop process of planning, execution, feedback, and replanning.

[0025] In this embodiment, the vehicle can be a pure electric passenger vehicle, delivery vehicle, inspection vehicle, unmanned vehicle, park operation vehicle, or other vehicle that relies on battery power for operation; the charging node can be a fixed charging station, temporary charging point, mobile energy replenishment point, battery swapping point, or other node that can replenish the vehicle's power. The following description is based on steps S1 to S7.

[0026] S1. Data Acquisition and Vehicle Status Modeling

[0027] Specifically, it includes:

[0028] It acquires the vehicle's current location, target task point, remaining battery power, load information, historical energy consumption per unit mileage, road network data, traffic status data, terrain and environmental data, meteorological data, charging node data, and historical charging service data.

[0029] The vehicle's current location can be obtained from a satellite positioning module, inertial navigation module, odometer, or visual positioning module; the target task point can be provided by a dispatch platform, navigation system, or user input; the remaining battery power can be obtained from the battery management system; the load information can be obtained from the vehicle controller, weighing device, or task system; road network data includes at least road topology, road segment length, speed limit information, and road connectivity; traffic status data includes at least real-time vehicle speed, congestion level, accident information, and road closure information; terrain environment data includes at least slope, altitude, and curvature; meteorological data includes at least rainfall, temperature, wind speed, and visibility; charging node data includes at least node location, charging power, node capacity, and service type; and historical charging service data includes at least historical occupancy records, historical waiting times, historical fault records, and historical charging records.

[0030] After obtaining the above data, the construction time is established. Vehicle state vector:

[0031]

[0032] in, Indicates time The vehicle state vector, Indicates the vehicle's time Current location Indicates the vehicle's time The remaining battery power, Indicates the vehicle's time Load information, Indicates the vehicle's time Task time limit status, Indicates the vehicle's time Environmental observation information, This indicates the execution feedback information that has been written back at the start of the current planning cycle.

[0033] Preferably, environmental observation information It can be represented as:

[0034]

[0035] in, Indicates environmental observation information, Represents traffic condition observations. Indicates meteorological condition observations, Indicates road condition observations, Indicates slope environmental observations, This indicates the observation of communication coverage status.

[0036] Preferably, the execution feedback information Represented as:

[0037]

[0038] in, This indicates the execution feedback information. Indicates the actual arrival time of the previous execution cycle. This indicates the actual waiting time in the previous execution cycle. This indicates the actual single replenishment amount in the previous execution cycle. This indicates the actual path energy consumption in the previous execution cycle. Indicates the first Each charging node at time The actual service status.

[0039] In this embodiment, the acquired data can also be preprocessed, including time alignment, coordinate unification, outlier removal, missing value completion, and numerical normalization. For example, traffic status data and meteorological data are aligned according to a unified time slice; road node coordinates, vehicle coordinates, and charging node coordinates are mapped to the same coordinate system; abnormal energy consumption records are removed; and short-term missing observation data are completed using interpolation or nearest neighbor methods.

[0040] Furthermore, after executing the charging decision results in the previous planning cycle, the actual arrival time will be... Actual waiting time Actual single replenishment of electricity Actual path energy consumption and the actual service status of charging nodes Feedback data is written back to the vehicle state vector construction module for updating. , , and The corresponding state variables are used to generate the vehicle state vector for the next planning cycle.

[0041] This invention integrates the vehicle, battery, task, environment, and execution feedback into the vehicle state vector. This allows subsequent planning processes to be based on both the current observed state and the actual execution results of the previous execution cycle, thereby improving the completeness of the state representation and the accuracy of decision-making in the next planning cycle.

[0042] S2, Charging Node Availability Prediction

[0043] Specifically, it includes:

[0044] Based on charging node data, historical charging service data, and write-back execution feedback data, a charging node availability prediction model is constructed for the first... Each charging node at time Predict the state within the corresponding future planning timeframe:

[0045]

[0046] in, Indicates the first Each charging node at time The predicted output, The parameter is Charging node availability prediction model, Indicates the first Each charging node at time Input features, Indicates the first Each charging node at time The probability of prediction availability. Indicates the first Each charging node at time Predicted waiting time Indicates the first Each charging node at time The prediction service window, Indicates the first Each charging node at time The predicted node power supply capacity.

[0047] The input features This can be further expressed as:

[0048]

[0049] in, This represents the input feature vector. Indicates the first Geographical attributes of each charging node Indicates the first Service type characteristics of each charging node Indicates the first Each charging node at time Historical occupancy characteristics, Indicates the first Each charging node at time Historical waiting time characteristics, Indicates the first Each charging node at time Historical characteristics of energy supply capacity Indicates time The time period characteristics, Indicates time Environmental characteristics, Indicates the first The reachability characteristics of each charging node related to the current vehicle task Indicates the first The actual service status written back by each charging node in the previous execution cycle.

[0050] In a preferred embodiment, the charging node availability prediction model A multi-task neural network model is employed. Specifically, the input features are first... The input is a shared feature encoding network, from which the common representation of charging nodes is extracted. This representation is then connected to the availability probability prediction branch, waiting time prediction branch, service window prediction branch, and node power supply capacity prediction branch, respectively, to simultaneously output... , , and The shared feature coding network can employ one or more of the following: multi-layer fully connected network, gated recurrent network, temporal convolutional network, or Transformer coding structure.

[0051] The model is trained using supervised learning. A training sample set is constructed.

[0052]

[0053] in, This represents the training sample set for charging node prediction. Indicates the input sample. Indicates the first Each charging node at time The true label includes at least the true availability status, true waiting time, true service window, and true node power supply capacity.

[0054] During training, the model parameters are updated by minimizing the following loss function. :

[0055]

[0056] in, This represents the total loss of the charging node prediction model. This indicates that the loss can be predicted using probability. Indicates the waiting time to predict loss. This indicates the service window for predicting losses. This indicates the predicted loss of node power supply capacity. to This represents the loss weighting coefficient.

[0057] Preferably, the probability-predictable loss is:

[0058]

[0059] in, This indicates that the loss can be predicted using probability. Indicates the first A real, usable tag for each charging node. This indicates the probability that the corresponding prediction is available.

[0060] The loss due to predicted waiting time is:

[0061]

[0062] in, Indicates the waiting time to predict loss. Indicates the first The actual waiting time at each charging station. This indicates the corresponding predicted waiting time.

[0063] The serviceable window prediction loss is expressed as:

[0064]

[0065] in: To predict losses for available service windows; The number of charging node samples participating in the training; For the first The predicted service start time for each charging node; For the first The actual service availability start time of each charging node; For the first Predicted end-of-service time for each charging node; For the first The actual end-of-service time for each charging node.

[0066] The predicted loss in node power supply capacity is:

[0067]

[0068] in, This indicates the predicted loss of node power supply capacity. Indicates the first The actual node power supply capacity of each charging node. This indicates the power supply capacity of the corresponding predicted node.

[0069] In terms of model training methods, it is preferable to first perform offline training using historical charging service data, and then use the Adam optimizer to update parameters through backpropagation. The samples are divided into training, validation, and test sets. During training, Dropout, weight decay, early stopping, and learning rate decay strategies can be introduced to improve the model's generalization ability. During online runtime, the actual arrival time after the vehicle executes its charging decision is then recorded. Actual waiting time Actual single replenishment of electricity and the actual service status of charging nodes Additional samples are added to form incremental samples, and the model is fine-tuned online to achieve model correction.

[0070] After obtaining the prediction results, a set of candidate charging nodes is constructed:

[0071]

[0072] in, Indicates time The set of candidate charging nodes, Indicates the first One charging node Indicates the available probability threshold. This indicates the threshold value for the node's power supply capacity; Indicates the first Each charging node at time The probability of prediction availability; Indicates the first Each charging node at time Its predictive energy replenishment capability.

[0073] This invention feeds back the actual service status and actual waiting results of charging nodes to the charging node availability prediction module, and combines offline training with online fine-tuning of the model training method. This enables the prediction model to learn long-term statistical laws and adapt to short-term state fluctuations, thereby improving the prediction accuracy of the actual state of charging nodes.

[0074] S3, Construction of Spatiotemporal Extended Charging Map

[0075] Specifically, it includes:

[0076] Based on road network data, traffic condition data, terrain and environmental data, meteorological data, and a set of candidate charging nodes Construct a spatiotemporally extended charging map:

[0077]

[0078] in, Represents a spatiotemporal extended charging diagram. Represents a spatiotemporal extended node set. Represents the set of spatiotemporal extension edges.

[0079] Any node in the spatiotemporal extended charging graph is represented as:

[0080]

[0081] in, Indicates the first in the path A spatiotemporal node, Indicates the first The physical location nodes corresponding to each node. Indicates reaching the th The predicted time for each node, Indicates reaching the th Remaining power at each node.

[0082] Define any spatiotemporal extension edge The overall cost is:

[0083]

[0084] in, Representing an edge The overall cost, Representing an edge The cost of distance, Representing an edge The time cost, Representing an edge Environmental risks and costs Representing an edge The predicted energy consumption cost Representing an edge Benefits in path evaluation to This represents the corresponding weighting coefficient.

[0085] Among them, predicting energy consumption costs Calculate using the following formula:

[0086]

[0087] in, Representing an edge The predicted energy consumption cost to Indicates the energy consumption coefficient. Indicates the distance of a road segment. Indicates the slope factor. Indicates the road surface resistance factor. Indicates meteorological influencing factors, Indicates time Vehicle load information.

[0088] The predicted energy consumption model can be obtained through supervised learning training. A sample set is constructed.

[0089]

[0090] in, This represents the training sample set for the energy consumption model. Representing an edge The actual energy consumption.

[0091] If a linear model is used, then the least squares method, ridge regression, or LASSO method can be used to obtain the results. to If a nonlinear model is used, a residual network or multilayer perceptron can be constructed based on the above formula to compensate for errors in the linear energy consumption term. During training, real energy consumption is used. With predicted energy consumption The objective is to minimize the error between the two paths; the optimizer can be Adam or SGD. During online runtime, the actual path energy consumption will be written back after execution. Additional samples are added to form incremental samples, and the parameters of the energy consumption model are corrected.

[0092] Environmental risk costs Calculate using the following formula:

[0093]

[0094] in, Representing an edge Environmental risks and costs Indicates the risk of traffic congestion. Indicates weather risk. This indicates an abnormal road condition risk. Indicates the risk of passage due to slope gradient. This indicates the risk of weak communication coverage. to This represents the corresponding risk weight coefficient.

[0095] When composing the image, the normal driving edge corresponds to... Waiting for the corresponding edge Charging side corresponds Its size is determined by the power supply capacity of the charging node, the charging duration, and the charging method.

[0096] This invention trains an energy consumption model and writes back the actual path energy consumption for model correction, thereby enabling the edge cost in the spatiotemporally extended charging graph to continuously approximate the actual running cost, thus improving the accuracy of subsequent path search and target path determination.

[0097] S4. Candidate charging path search with safety constraints

[0098] Specifically, it includes:

[0099] In the spatiotemporal extended charging diagram The candidate charging path is searched under the constraints of remaining power safety, task time limit, and charging node service window.

[0100] When transitioning states along a path, the following conditions must be met:

[0101]

[0102] in, and These represent the paths to the destination. The node and the first Remaining power at each node Indicates from node Transfer to node Predicted energy consumption Indicates from node Transfer to node The benefit item in the path evaluation.

[0103] Simultaneously satisfy the following constraints:

[0104]

[0105] in, Indicates the safe power threshold. Indicates reaching the th The predicted time for each node, Indicates the maximum allowed completion time for the task. Indicates the first Each charging node at time The prediction can serve the time window.

[0106] In a preferred embodiment, a heuristic search algorithm with constrained pruning is used to perform path search. For candidate path prefixes... Calculate the evaluation function:

[0107]

[0108] in, Indicates path prefix The evaluation value, Indicates path prefix The cumulative actual cost, This represents the heuristic cost estimate from the current node to the target node.

[0109] in,

[0110]

[0111] in, This represents the cumulative actual cost. Represents a node With nodes The overall cost of the edges between them.

[0112] The cost of heuristic estimation satisfies:

[0113]

[0114] in, This represents a heuristic cost estimation. Indicates starting from the current node To the target node The predicted remaining distance, Indicates starting from the current node To the target node The predicted remaining time, Indicates starting from the current node To the target node Predicted remaining energy consumption to This represents the heuristic weighting coefficient.

[0115] All paths that satisfy the constraints constitute the candidate charging path set:

[0116]

[0117] in, Indicates time The set of candidate charging paths, This represents any candidate charging path.

[0118] The heuristic search algorithm itself does not need to be obtained through training, but in some embodiments, the heuristic weight coefficients... to Parameters can be optimized using historical path samples, such as by using grid search, Bayesian optimization, or genetic algorithms, to ensure that search efficiency and path quality meet preset requirements simultaneously.

[0119] This invention improves the quality of subsequent path selection by performing constrained search on the updated spatiotemporal extended charging graph and optimizing heuristic parameters, thereby more efficiently screening out candidate charging paths that meet safety and time limit requirements.

[0120] S5. Solving the comprehensive objective function and determining the target charging path

[0121] Specifically, it includes:

[0122] For the set of candidate charging paths Candidate charging paths in China Calculate the comprehensive objective function:

[0123]

[0124] in, Indicates candidate charging paths The comprehensive objective function value, Indicates the task's reward items. This represents the benefit items in path evaluation. Represents the time cost term. Indicates the risk cost item. Indicates the waiting cost term. This represents the path deviation cost term. to This represents the corresponding weighting coefficient.

[0125] In the path evaluation, the benefit term and the waiting cost term are represented as follows:

[0126]

[0127] as well as

[0128]

[0129] in, Representing a path The benefit items in the path evaluation Representing a path The waiting cost item, and This represents the internal weighting coefficient of the benefit item in path evaluation. Indicates the first Each charging node at time The predicted node power supply capacity, Indicates the first Each charging node at time Predicted waiting time Indicates the first Each charging node at time The congestion coefficient.

[0130] After the calculation is completed, the target charging path is determined according to the following formula:

[0131]

[0132] in, Indicates time The target charging path.

[0133] In a preferred embodiment, the weighting coefficient to , , and Parameters can be learned or optimized by using historical scheduling results samples. Specifically, Bayesian optimization, genetic algorithms, or reinforcement learning methods can be used to achieve the best overall performance in terms of task completion rate, average waiting time, and power failure rate.

[0134] This invention constructs a comprehensive objective function and learns or optimizes its parameters, thereby enabling the determination of the target charging path to take into account the benefits of task completion, path evaluation, waiting costs, time costs, and risk costs, thus improving the overall optimality of the final path.

[0135] S6, Scrolling Time Domain Online Replanning

[0136] Specifically, it includes:

[0137] As the vehicle follows the target charging path During operation, real-time information on changes in road conditions, environment, vehicle status, and charging node status is acquired and displayed in the rolling time domain. The internal update of charging node prediction results and spatiotemporal extended charging map is performed.

[0138] Online replanning is triggered when the performance of the updated path deteriorates or the risk of subsequent power outages increases. The specific triggering criteria are as follows:

[0139]

[0140] in, Indicates time The amount of path performance degradation, This represents the objective function value of the original target charging path. This represents the objective function value obtained by re-evaluating the execution path under the current new state. This indicates the current execution path after the update.

[0141] When satisfied

[0142]

[0143] Or satisfy

[0144]

[0145] At that time, online replanning is triggered.

[0146] in, Indicates the performance degradation threshold. This indicates the probability that the remaining power of subsequent nodes will fall below the safe power threshold. This indicates the threshold for the risk of power loss.

[0147] In one embodiment, rolling temporal online replanning is achieved by locally updating the spatiotemporally extended charging graph, and the updated edge cost is represented as:

[0148]

[0149] in, This represents the updated edge cost. This represents the edge cost before the update. This represents the increment of edge cost.

[0150] The edge cost increment satisfies:

[0151]

[0152] in, Indicates the edge cost increment. Representing an edge The change in risk cost, Representing an edge The predicted change in energy consumption Indicates the first Each charging node at time The predicted change in waiting time Indicates the first Each charging node at time The change in available probability, to This indicates that the weight coefficients are being updated.

[0153] To obtain a more accurate estimate of the probability of power outage risk, an energy consumption uncertainty model can be established and trained using historical samples of the "predicted energy consumption - actual energy consumption" deviation. This model can employ Gaussian regression, quantile regression, or deep ensemble networks to output the future energy consumption distribution or quantile intervals, which are then used to calculate... .

[0154] This invention introduces a rolling time-domain online replanning mechanism and trains a power outage risk probability estimation model, enabling the system to promptly identify path performance degradation and energy security risks, thereby improving the real-time adjustment capability and robustness of the planning results in dynamic environments.

[0155] S7. Output the target charging path and charging decision results, and perform feedback write-back.

[0156] Specifically, it includes:

[0157] After obtaining the target charging path, output the target charging path and the charging decision result:

[0158] ,

[0159] in, Indicates time The charging decision results Indicates the target charging node. Indicates the estimated time of arrival at the target charging node. Indicates the estimated waiting time. Indicates the charging method. This indicates the expected amount of electricity to be replenished in a single transaction.

[0160] The estimated amount of electricity to be replenished in a single instance is determined by the following formula:

[0161] ,

[0162] in, This indicates the target single replenishment of electricity. Indicates the target charging node at time... The predicted node power supply capacity, This indicates the predicted electricity demand from the current charging node to the next critical road segment or the next backup charging node. Indicates the safe reserve of electricity. Indicates time The current remaining battery power.

[0163] In this embodiment, the vehicle executes the charging decision result. Then, the following actual execution results were collected:

[0164] ,

[0165] in, Indicates time Execution feedback data, Indicates the actual arrival time. Indicates the actual waiting time. This indicates the actual amount of electricity replenished in a single transaction. Indicates actual path energy consumption. Indicates the first The service status of each charging node during actual operation.

[0166] Then, the execution feedback data The data is written back to the vehicle state vector construction module and the charging node availability prediction module. Specifically, it will... , and Used to update vehicle location status and remaining battery status for the next planning cycle. and It is used to update the historical service data of charging nodes and the input features of the prediction model, and writes all feedback data as new samples into the historical sample library for status updates and model corrections in the next planning cycle.

[0167] This creates the following closed-loop relationship:

[0168]

[0169] The closed-loop relationship is as follows: S1 completes vehicle state modeling, S2 completes charging node availability prediction, S3 completes spatiotemporal extended charging map construction, S4 completes candidate charging path search, S5 completes target charging path determination, S6 completes online replanning, and S7 completes execution feedback write-back and re-inputs the execution results into the next round S1.

[0170] In multi-vehicle scenarios, a multi-vehicle-multi-charging-node time slot allocation model can be further established:

[0171] ,

[0172] The objective function represents maximizing the overall scheduling benefits under conditions of multiple vehicles, multiple charging nodes, and multiple time slots. Indicates vehicle Is it at the moment? Assigned to a charging node Indicator variables, This represents the benefit item in the corresponding path evaluation. This indicates the corresponding waiting cost. This indicates the corresponding deviation cost. to Indicates the weighting coefficient. Indicates the number of vehicles. Indicates the number of charging nodes. It represents a discrete-time set.

[0173] And satisfy:

[0174] ,

[0175] ,

[0176] The first equation indicates that each vehicle is allocated to at most one charging node within a time slot allocation cycle, and the second equation indicates that the first equation indicates that the second equation indicates that the third ... Each charging node at time The number of vehicles accepted shall not exceed its service capacity. .

[0177] The multi-vehicle-multi-charging-node time slot allocation model can be solved using integer programming, Lagrange relaxation, greedy allocation, or heuristic scheduling algorithms. If a learning-based scheduling method is used, a graph neural network model or reinforcement learning strategy network can be trained based on historical fleet scheduling data to output a time slot allocation scheme.

[0178] This invention further collects actual execution results after outputting charging decision results, and writes back the actual arrival time, actual waiting time, actual single replenishment power, actual path energy consumption, and actual service status of charging nodes to the vehicle state vector construction module and the charging node availability prediction module. At the same time, it adopts a combination of offline training and online correction for the prediction model, energy consumption model, and scheduling model involved, so that the system can continuously use the real execution results to correct subsequent states, models, and plans, thereby forming a complete data closed loop and significantly improving the system's adaptability, prediction accuracy, and planning reliability in dynamic environments.

[0179] According to one embodiment, a vehicle charging route planning system is also provided, comprising:

[0180] The vehicle state construction module is used to obtain the vehicle state vector and the execution feedback data for write-back.

[0181] A charging node availability prediction model is used to predict the availability probability, waiting time, service window and power supply capacity of each charging node based on charging node data, historical charging service data and the execution feedback data, and to construct a candidate charging node set.

[0182] The spatiotemporal mapping module is used to construct a spatiotemporal extended charging map based on the candidate charging node set, road network data, traffic status data, and environmental data, and to determine the edge cost based on distance cost, time cost, environmental risk cost, predicted energy consumption cost, and the benefit item in the path evaluation.

[0183] The path search module is used to search for candidate charging paths on the spatiotemporal extended charging map using remaining power safety constraints, task time constraints, and charging node service window constraints, to obtain a set of candidate charging paths.

[0184] The path determination module is used to determine the target charging path from the candidate charging path set based on the comprehensive objective function;

[0185] The feedback write-back module is used to collect the actual arrival time, actual waiting time, actual single replenishment of electricity, actual path energy consumption, and actual service status of the charging node as the execution feedback data after the vehicle executes the charging decision result. The execution feedback data is then written back to the vehicle status construction module and the charging node availability prediction model for status updates and model corrections in the next planning cycle.

[0186] In the above embodiments, each step can be executed sequentially by the same processor or collaboratively by multiple functional modules; S1 to S7 can also be executed periodically to form a closed-loop update mechanism. The above embodiments are only used to illustrate the present invention and are not intended to limit the scope of protection of the present invention; equivalent substitutions, modifications and improvements made by those skilled in the art without departing from the concept of the present invention should all fall within the scope of protection of the present invention.

Claims

1. A vehicle charging path planning method, characterized in that, Executed by the processor, including: Acquire vehicle state vectors, charging node data, and historical charging service data; based on the charging node data, historical charging service data, and write-back execution feedback data, use a charging node availability prediction model to predict the availability probability, waiting time, service window, and node power supply capacity of each charging node, and construct a candidate charging node set; based on the candidate charging node set, road network data, traffic state data, and environmental data, construct a spatiotemporal extended charging map with physical location nodes, predicted time, and remaining battery power as node states, and determine the charging node status based on distance cost, time cost, environmental risk cost, predicted energy consumption cost, and benefit items in path evaluation. Edge cost; on the spatiotemporally extended charging map, candidate charging paths are searched using remaining power safety constraints, task time constraints, and charging node service window constraints to obtain a set of candidate charging paths; the target charging path is determined from the set of candidate charging paths based on the comprehensive objective function; after the vehicle executes the charging decision, the actual arrival time, actual waiting time, actual single replenishment power, actual path energy consumption, and actual service status of the charging node are collected as the execution feedback data, and the execution feedback data is written back to the vehicle state construction module and the charging node availability prediction model for state updates and model corrections in the next planning cycle.

2. The vehicle charging path planning method according to claim 1, characterized in that: The vehicle state vector is represented as follows: in, For a moment The vehicle state vector, For a moment The current location of the vehicle. For a moment The vehicle's remaining battery power. For a moment Vehicle load information, For a moment Task time limit status, For a moment Environmental observation information, For a moment The execution feedback information.

3. The vehicle charging path planning method according to claim 1, characterized in that: The prediction output of the charging node availability prediction model is expressed as follows: in, For the first Each charging node at time The predicted output, For parameters Charging node availability prediction model, For the first Each charging node at time Input features, For the first Each charging node at time The probability of prediction availability. For the first Each charging node at time Predicted waiting time For the first Each charging node at time The prediction service window, For the first Each charging node at time The predicted node power supply capacity; the candidate charging node set is represented as: ， in, For a moment The set of candidate charging nodes, For the first One charging node The available probability threshold, The threshold for the node's power supply capacity.

4. The vehicle charging path planning method according to claim 3, characterized in that: The charging node availability prediction model employs a multi-task neural network model, which is trained by minimizing the following loss function: ， in, The total loss of the charging node prediction model. To predict loss based on available probability, To predict losses based on waiting time, To predict losses during the service window, Predicting losses in node power supply capacity. to These are the loss weighting coefficients; The available probability prediction loss is expressed as: in, For the first A real, usable tag for each charging node. For the first The predicted availability probability of each charging node; the waiting time prediction loss is expressed as: in, For the first Predicted waiting time for each charging node For the first The actual waiting time for each charging node; The serviceable window prediction loss is expressed as: in: To predict losses for available service windows; The number of charging node samples participating in the training; For the first The predicted service start time for each charging node; For the first The actual service availability start time of each charging node; For the first Predicted end-of-service time for each charging node; For the first The actual end of service availability for each charging node; The predicted loss of node power supply capacity is expressed as: in, For the first Predicted node power supply capacity of each charging node. For the first The actual node power supply capacity of each charging node; It uses historical charging service data for offline training and incremental samples formed by the execution feedback data for online fine-tuning.

5. The vehicle charging path planning method according to claim 1, characterized in that: The spatiotemporal extended charging diagram is represented as follows: ， in, For the spatiotemporal expansion charging map, For a spatiotemporal extended node set, It is a set of spatiotemporally extended edges; The nodes in the spatiotemporal extended charging graph are represented as follows: ， in, For the first in the path A spatiotemporal node, For the first The physical location nodes corresponding to each node. To reach the The predicted time for each node, To reach the Remaining power at each node.

6. The vehicle charging path planning method according to claim 5, characterized in that: The edge cost is expressed as in, For the edge The overall cost, For the edge The cost of distance, For the edge The time cost, For the edge Environmental risks and costs For the edge The predicted energy consumption cost For the edge Benefits in path evaluation to These are the corresponding weighting coefficients; The predicted energy consumption cost is expressed as: in, to Energy consumption coefficient Distance of road segment For slope factor, For road surface resistance factor, Meteorological influencing factors, For a moment Vehicle load information; The environmental risk cost is expressed as: in, to For risk weighting coefficients, To mitigate the risk of traffic congestion, Due to weather risks, Due to abnormal road conditions, Due to the risk of crossing on the slope, To mitigate the risk of weak communication coverage, offline training is performed using historical path energy consumption data, and online correction is performed using the actual path energy consumption.

7. The vehicle charging path planning method according to claim 1, characterized in that: The following state transition relationships and constraints must be satisfied when searching for candidate charging paths: in, and These are the paths to reach the destination. The node and the first Remaining power at each node For the node Transfer to node Predicted energy consumption For the node Transfer to node Benefits in path evaluation; in, For safe power threshold, To reach the The predicted time for each node, The maximum allowed completion time for the task. For the first Each charging node at time The prediction can serve the time window.

8. The vehicle charging path planning method according to claim 7, characterized in that: Candidate charging path prefix Calculate the evaluation function in, path prefix The evaluation value, path prefix The cumulative actual cost, A heuristic cost estimate for the journey from the current node to the target node; in, For nodes With nodes The overall cost of the edges between them; in, To start from the current node To the target node The predicted remaining distance, To predict the remaining time, To predict remaining energy consumption, to These are heuristic weighting coefficients.

9. The vehicle charging path planning method according to claim 1, characterized in that: The comprehensive objective function is expressed as follows: in, Candidate charging paths The comprehensive objective function value, For task rewards, For the benefit items in path evaluation, For time cost, For risk cost item, For the waiting cost term, This is the path deviation cost term. to The weighting coefficients are used; and the target charging path is determined according to the following formula: in, For a moment The target charging path, This is a set of candidate charging paths.

10. A vehicle charging route planning system, characterized in that, include: The vehicle state construction module is used to obtain the vehicle state vector and the execution feedback data for write-back. A charging node availability prediction model is used to predict the availability probability, waiting time, service window and power supply capacity of each charging node based on charging node data, historical charging service data and the execution feedback data, and to construct a candidate charging node set. The spatiotemporal mapping module is used to construct a spatiotemporal extended charging map based on the candidate charging node set, road network data, traffic status data, and environmental data, and to determine the edge cost based on distance cost, time cost, environmental risk cost, predicted energy consumption cost, and the benefit item in the path evaluation. The path search module is used to search for candidate charging paths on the spatiotemporal extended charging map using remaining power safety constraints, task time constraints, and charging node service window constraints, to obtain a set of candidate charging paths. The path determination module is used to determine the target charging path from the candidate charging path set based on the comprehensive objective function; The feedback write-back module is used to collect the actual arrival time, actual waiting time, actual single replenishment of electricity, actual path energy consumption, and actual service status of the charging node as the execution feedback data after the vehicle executes the charging decision result. The execution feedback data is then written back to the vehicle status construction module and the charging node availability prediction model for status updates and model corrections in the next planning cycle.

Images