A freight electrifying station path intelligent planning method based on multi-source OD data

Abstract

The application belongs to the cross technical field of artificial intelligence and intelligent transportation system, and specifically discloses a freight electrification station path intelligent planning method based on multi-source OD data. The method comprises the following steps: constructing a multi-source OD data credible evidence and completion system based on a blockchain and a geographic weighted regression fusion cellular automaton; establishing a strategic layer stochastic robust optimization model with logistics cost, power grid load, construction investment and battery attenuation as multiple targets; designing a tactical layer flexible path scheduling algorithm combining an improved genetic algorithm and an ant colony mechanism; deploying an execution layer real-time response module based on digital twinning and SAC reinforcement learning; and integrating a three-dimensional collaborative perception and health management closed loop driven by the Internet of Things and digital twinning. Through the above technical scheme, the four-dimensional collaborative optimization of logistics efficiency, battery life, power grid stability and comprehensive cost is realized, and the robustness, response speed and operation benefit of the freight electrification network are improved.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of artificial intelligence and intelligent transportation systems, specifically relating to an intelligent route planning method for freight battery swapping stations based on multi-source OD data. Background Technology With the rapid evolution of the new energy freight logistics system, battery swapping, as a key path to alleviate range anxiety and improve operational efficiency, is gradually becoming an important infrastructure in the heavy-duty transportation sector. Traditional freight route planning methods often focus on single cost or time objectives, relying on static road networks and fixed energy consumption models, making it difficult to adapt to multi-dimensional coupled constraints such as dynamic traffic, grid load, and battery status. Especially in scenarios driven by large-scale, multi-source origin-destination (OD) data, freight demand exhibits high spatiotemporal heterogeneity and uncertainty, placing higher demands on the layout of battery swapping stations and route collaborative optimization.

[0002] However, existing route planning technologies are generally limited to the single disciplines of operations research or traffic engineering, lacking the ability to reliably fuse and dynamically correct multi-source origin-destination (OD) data. This results in high noise and incomplete coverage of input data, severely impacting the robustness of the planning results. Furthermore, the algorithms often employ deterministic optimization models, making it difficult to account for multiple stochastic factors such as vehicle-to-grid (V2G) interaction, battery health degradation, and real-time traffic disturbances. This hinders the organic linkage between strategic-level global planning, tactical-level flexible scheduling, and execution-level real-time response. In addition, the system architecture lacks a collaborative sensing mechanism supported by digital twins and the Internet of Things (IoT), causing the four-dimensional objectives of logistics efficiency, battery life, grid stability, and overall cost to be fragmented, making it difficult to achieve global optimization.

[0003] Therefore, there is an urgent need for a method for intelligent route planning of freight battery swapping stations based on multi-source OD data. Summary of the Invention

[0004] The purpose of this invention is to provide a method for intelligent route planning of freight swapping stations based on multi-source OD data, which can effectively solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this invention provides a method for intelligent route planning of freight battery swapping stations based on multi-source OD data, comprising the following steps: Construct a multi-source OD data fusion and trusted evidence storage system, collect multi-source freight origin and destination data, perform hash encryption and timestamp binding on the original OD data to achieve tamper-proof trusted evidence storage, and use a geographic weighted regression model combined with cellular automata mechanism to perform spatial correction and dynamic completion of missing or abnormal data. A strategic-level stochastic robust optimization model is established. Based on a trusted OD dataset, a global optimization function is constructed with multiple objectives, including logistics cost, power grid load balance, investment in battery swapping station construction, and battery health degradation. Stochastic disturbance variables corresponding to traffic flow, electricity price fluctuations, and battery performance degradation uncertainty are introduced. The robust optimal battery swapping station layout scheme is solved by scenario tree decomposition and dual cutting method. A tactical-level elastic path scheduling algorithm is designed. Based on the established layout of battery swapping stations, an improved genetic algorithm and an ant colony optimization mechanism are integrated. The vehicle range constraint, battery swapping waiting time, V2G reverse power supply capability, and real-time road network status are used as dynamic fitness factors to generate an elastic path set for multi-vehicle collaboration. The SAC reinforcement learning real-time response module is deployed at the execution layer. Based on the digital twin environment, a state vector containing vehicle location, remaining battery power, queue length of the battery swapping station ahead, real-time electricity price of the grid, and battery health index is constructed. The actions are spatially continuous path offset and battery swapping decision. The soft Actor-Critic algorithm is used to learn the optimal strategy online to achieve millisecond-level path fine-tuning and battery swapping timing determination. Integrating a three-dimensional collaborative perception and health management closed loop, the system collects battery temperature, internal resistance, charge-discharge cycle count, and battery swapping station equipment status in real time through an IoT sensor network. Combined with a digital twin, it constructs a battery health degradation prediction model and feeds the prediction results back to the strategic and tactical layers, forming a four-dimensional collaborative optimization closed loop covering logistics efficiency, battery life, grid stability, and overall cost.

[0006] Preferably, the hash encryption and timestamp binding of the original OD data are performed using blockchain technology. The blockchain adopts a consortium blockchain architecture, with nodes jointly maintained by logistics operators, power grid companies, and traffic management departments. The consensus mechanism is a practical Byzantine fault-tolerant algorithm, the block generation cycle is a preset time interval, and data integrity verification is achieved through a Merkle tree structure.

[0007] Preferably, the spatial weight function of the geographic weighted regression model adopts a Gaussian kernel function, the bandwidth parameter is dynamically adjusted through cross-validation, the neighborhood rule of the cellular automaton is defined as a specific grid size, and the state transition probability is jointly determined by the historical OD traffic density and the road network topology connectivity.

[0008] Preferably, the random disturbance variables include traffic speed fluctuation coefficient, time-of-use electricity price deviation rate and battery capacity decay rate, the probability distribution of which is obtained by fitting historical data, the scene tree is divided into multiple main scenes and sub-scenes, and the dual cutting method iterative convergence threshold is set to a preset precision.

[0009] Preferably, the crossover operator of the improved genetic algorithm adopts a path segment retention strategy, the mutation operation introduces local search perturbation, the ant colony pheromone update mechanism integrates the service capacity of the battery swapping station and the V2G adjustment margin, and the fitness function weight vector is determined by the analytic hierarchy process.

[0010] Preferably, the V2G reverse power supply capability is quantified as the upper limit of the vehicle's release capacity. This upper limit is constrained by the current battery state of charge, health index, and grid dispatch instructions. When the grid load exceeds a preset load threshold, the system automatically activates the V2G mode and prioritizes dispatching vehicles with high health indices to participate in peak shaving.

[0011] Preferably, the digital twin environment maps the physical road network and battery swapping station facilities at a predetermined ratio, the state vector dimension is a preset dimension, the action space is a two-dimensional continuous variable, and the temperature coefficient α of the SAC algorithm maintains the strategy entropy within a preset range through an adaptive adjustment mechanism.

[0012] Preferably, the calculation model of the battery health index integrates incremental capacity analysis and impedance spectrum characteristics. The input parameters include cumulative charge and discharge energy, daily average temperature difference and fast charging times. The output is a normalized health value. When the health value is less than the preset health threshold, the system automatically restricts the vehicle from participating in V2G and prioritizes maintenance.

[0013] Preferably, the IoT sensors include temperature sensors, voltage and current sensors, and vibration sensors. The sampling frequency is a preset frequency, and the data is uploaded to the edge computing node via the LoRaWAN protocol. The edge node synchronizes the battery health prediction result to the central cloud platform once at a predetermined time interval.

[0014] Preferably, in the four-dimensional collaborative optimization closed loop, the logistics efficiency index is the average delivery time, the battery life index is the average annual health degradation rate, the power grid stability index is the voltage fluctuation amplitude at the battery swapping station access point, and the comprehensive cost index is the total expenditure per ton-kilometer. The four indicators achieve multi-objective balance through Pareto front analysis, and the system performs a global re-optimization once according to a predetermined cycle.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. By combining blockchain with geographic weighted regression and cellular automata, the core problems of high noise, incomplete coverage, and susceptibility to tampering of multi-source OD data are solved. The data availability is significantly improved, the spatial correction error is less than the predetermined distance, and high-fidelity input is provided for upper-level planning, fundamentally improving the robustness and reliability of path planning.

[0016] 2. Create a three-level optimization system of strategy, tactics, and execution. The strategic layer uses a stochastic robust model to ensure the long-term stability of the layout, the tactical layer uses a hybrid intelligent algorithm to achieve elastic scheduling, and the execution layer uses SAC reinforcement learning to achieve millisecond-level response. The three are organically linked, which significantly improves the replanning efficiency when the system is dealing with disturbances such as traffic congestion, power grid failure or sudden vehicle failure, and shortens the path deviation recovery time to within the preset time limit.

[0017] 3. Breaking down the barriers between logistics, energy, transportation and manufacturing disciplines, it is the first to integrate battery health management, V2G grid interaction, battery swapping station layout and freight routes into the optimization framework. Actual tests show that, under the same transportation capacity, the overall operating cost is significantly reduced, the average annual battery degradation rate is significantly reduced, the peak-valley difference of the power grid is effectively narrowed, and the on-time delivery rate of logistics is greatly improved, realizing deep collaboration and value maximization of cross-domain resources.

[0018] 4. Based on high-fidelity digital twins and dense IoT sensing, a real-time feedback channel from the physical world to the decision-making model is constructed, enabling the system to have self-diagnosis, self-optimization and self-repair capabilities. The operation and maintenance manpower costs are significantly reduced, and the utilization rate of battery swapping station equipment is improved to a high level. This provides a scalable and replicable technical paradigm for the intelligent operation of large-scale new energy freight networks. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram illustrating the core principle framework of the three-level synergistic optimization system of strategy-tactics-execution in this invention; Figure 3 This is a logical flowchart of the multi-source OD data fusion and trusted evidence storage system in this invention; Figure 4 This is a schematic diagram of the joint decision-making framework of the strategic layer stochastic robust optimization model and the tactical layer elastic path scheduling algorithm in this invention; Figure 5 This is a schematic diagram of the interaction and data flow between the SAC reinforcement learning real-time response module in the execution layer and the digital twin environment in this invention; Figure 6 This is a schematic diagram illustrating the collaborative optimization principle of the three-dimensional collaborative perception and health management closed loop in this invention under four-dimensional objectives (logistics efficiency, battery life, power grid stability, and overall cost). Detailed Implementation

[0020] Example 1: Reference Figures 1 to 6This invention proposes a solution centered on multi-source OD data for freight transportation, breaking down disciplinary barriers and integrating technologies from multiple fields to construct an intelligent planning system. The data layer utilizes blockchain, GWR, and CA technologies to achieve trusted storage, spatial correction, and dynamic completion of OD data, solving the challenge of multi-source data fusion. The algorithm layer employs a three-tiered architecture of strategy, tactics, and execution, integrating stochastic robust optimization, improved genetic-ant colony algorithms, and SAC reinforcement learning, linking with the V2G system to achieve global optimization and real-time response. The collaboration layer leverages digital twins and IoT to construct a three-dimensional module, integrating battery health management technology.

[0021] Interdisciplinary integration achieves optimal performance across four dimensions: logistics, battery, power grid, and cost, significantly improving efficiency, reducing costs and risks, and generating unexpected synergistic effects. This approach is applied to an intelligent route planning method for freight battery swapping stations based on multi-source OD data.

[0022] In the above-mentioned intelligent route planning method for freight battery swapping stations based on multi-source OD data, step (1) constructs a multi-source OD data fusion and trusted evidence storage system. The specific implementation process is as follows: First, the system synchronously collects original freight origin-destination (OD) data from four heterogeneous data sources, including order-level origin and destination coordinates provided by the logistics platform, vehicle passage time and license plate recognition information recorded by traffic checkpoints, GPS trajectory point sequence uploaded by vehicle terminals, and battery swapping request logs fed back by the power grid dispatching system.

[0023] All raw data streams are connected to the central data lake through the API gateway. The data format is uniformly converted to the GeoJSON standard structure. Each OD record contains the following fields: starting latitude and longitude (accuracy 0.00001°), ending latitude and longitude, cargo weight (unit: tons), planned departure time (UTC timestamp), vehicle ID (16-bit hexadecimal code), and the identifier of the logistics company to which it belongs.

[0024] The system initiates the blockchain trusted evidence storage process. This blockchain adopts a consortium blockchain architecture, with three core institutions—a national logistics operator, a provincial power grid company, and a municipal transportation management bureau—each deploying full nodes. The nodes are interconnected through a dedicated network, and the consensus mechanism adopts the Practical Byzantine Fault Tolerance (PBFT) algorithm, setting the number of faulty nodes tolerable f=1, meaning that at least 3 normal nodes must reach a consensus before a block can be submitted.

[0025] Before each OD data record is written, a unique fingerprint value H is generated using the SHA-256 hash function, and then bound to a timestamp T accurate to the millisecond level to form a tuple.<H,T> This tuple is submitted to the blockchain network as the transaction payload. The block generation cycle is fixed at 30 seconds to ensure that the end-to-end latency of any OD data from collection to on-chain is less than 200 milliseconds.

[0026] Data integrity verification is implemented using a Merkle Tree structure: within each block, all transaction hashes are paired from bottom to top to calculate parent hashes, ultimately generating a root hash (Root_Hash) which is stored in the block header. When verifying the authenticity of an OD record, only the hash value of that OD record and its authentication path in the Merkle Tree need to be provided. Verification can be completed within a short time, improving efficiency by two orders of magnitude compared to traditional full comparison.

[0027] After credible evidence storage is completed, the system performs spatial correction and dynamic completion. This process integrates a Geographically Weighted Regression (GWR) model with a Cellular Automata (CA) mechanism. First, the urban road network is divided into a regular grid with sides of 500 meters. Each grid cell is considered a unit cell, and its state is defined as the historical daily average OD (Original Discharge) traffic density within the region (unit: vehicles / km²). The GWR model is used to interpolate and correct missing or anomalous OD points, and its spatial weighting function employs a Gaussian kernel function. : ; Points to be calibrated With a known reference point The Euclidean distance between them This is a bandwidth parameter. Bandwidth The model is dynamically adjusted through five-fold cross-validation to select the value that minimizes the sum of squared predicted residuals, typically ranging from 1.5 to 3.0 kilometers. For track breaks caused by GPS signal loss, the GWR model utilizes the spatial correlation of surrounding complete track points to output corrected latitude and longitude coordinates, with spatial correction errors controlled within 150 meters.

[0028] Subsequently, a cellular automaton mechanism dynamically completes the corrected OD data. The cellular neighborhood rule is defined as the Moore neighborhood, which includes the central cell itself and its eight surrounding neighboring cells, forming a 3×3 grid of nine cells. The state transition probability is determined by two factors: historical OD flow density. (Normalized to the [0,1] interval) and road network topological connectivity (Shortest path reachability index calculated based on Dijkstra's algorithm). Specifically, if a cell is in If the time state is 0 (no OD record), then it is in The probability of being activated (state changes to 1) at a given time. for: ; The weighting coefficient is set to 0.6, reflecting the dominant role of historical traffic in the completion process. This cellular automaton mechanism can effectively identify data gaps caused by temporary construction or sensor malfunctions, and dynamically fill in appropriate OD points based on surrounding activity levels, thus improving the accuracy of dynamic completion.

[0029] In the above-mentioned intelligent route planning method for freight swapping stations based on multi-source OD data, step (2), establishing a strategic-level stochastic robust optimization model, is implemented as follows: Based on the reliable OD dataset output in step (1), the system constructs a freight demand heatmap covering the entire region. The heatmap resolution is consistent with the aforementioned 500-meter grid, and the demand intensity Q of each grid unit is quantified by the corrected OD flow density. On this basis, a global optimization function F is defined, which contains four mutually coupled objective components: logistics cost. Power grid load balance Investment in the construction of battery swapping stations and battery health degradation .

[0030] Logistics costs It consists of three parts: total vehicle mileage cost (calculated at 2.5 yuan / km), battery swapping service fee (calculated at 0.8 yuan / kWh), and time penalty cost (50 yuan per hour of delay). Power grid load balance. The load balance is measured by calculating the reciprocal of the daily load variance at each battery swapping station access point; a smaller variance indicates a more balanced load. (Battery swapping station construction investment) Using an economies of scale model, the investment cost per station increases non-linearly with the improvement of service capacity, as shown in the formula: , Design service capacity (number of trains / day). Ten thousand yuan, Battery health degradation The annual average decline rate of the health index is used as the indicator. For details on the calculation of the health index, please refer to step (4).

[0031] To address the uncertainties of the real world, the model introduces three types of random disturbance variables: traffic speed fluctuation coefficient. (Following a log-normal distribution with a mean of 1.0 and a standard deviation of 0.15), time-of-use electricity price deviation rate (Following a normal distribution with a mean of 0 and a standard deviation of 0.08) and battery capacity decay rate (Follows a Beta distribution, shape parameter) These probability distributions were obtained by fitting 5 years of historical operating data, using the Kolmogorov-Smirnov test. A value greater than 0.05 indicates a good fit.

[0032] The solution process employs a combined strategy of scenario tree decomposition and dual slicing. First, the continuous probability space is discretized into a scenario tree, with 5 main scenarios (corresponding to high and low combinations of the three major disturbances: traffic, electricity prices, and batteries). Each main scenario is further subdivided into 3 sub-scenarios to capture tail risks, resulting in a total of 15 representative scenarios. Each scenario is assigned a probability weight. ,satisfy .

[0033] The dual cutting method solves the main problem (determining candidate locations for battery swapping stations) and subproblems (optimal path allocation under each scenario) in each iteration, generating the feasible region of the main problem constrained by the Benders cut plane. The iteration convergence threshold is set to 10. -4 The algorithm stops when the relative difference between the upper and lower bounds of two consecutive iterations is less than 0.01%. It can complete large-scale layout optimization of 1000 candidate nodes within 4 hours, outputting a robust and optimal battery swapping station layout scheme, including the geographical location, design service capabilities, and V2G interface configuration of each site.

[0034] In the above-mentioned intelligent route planning method for freight vehicle swapping stations based on multi-source OD data, step (3) involves designing a tactical-level elastic route scheduling algorithm. The specific implementation process is as follows: Based on the determined layout of swapping stations, the system generates an initial route plan for each freight vehicle. The algorithm integrates an improved genetic algorithm (GA) and an ant colony optimization (ACO) mechanism to form a hybrid intelligent scheduling engine.

[0035] During the population initialization phase, each chromosome is encoded as a sequence of battery swapping stations along the vehicle's route, for example, [starting point, S3, S7, ending point], where S3 and S7 are the battery swapping station numbers determined in step (2). Fitness function A combination of four dynamic factors: vehicle range constraints (Length of the route beyond the remaining driving range) Battery swapping waiting time (Estimated queuing time for the upcoming battery swapping station), V2G reverse power supply capability and real-time road network status (Road segment speed calculated based on floating car data). Fitness The calculation formula is: ; Weight vector The values ​​were determined using the Analytic Hierarchy Process (AHP) and, based on expert scoring and a consistency test (CR < 0.1), were typically [0.35, 0.25, 0.20, 0.20].

[0036] The improved genetic algorithm's operational operators are designed as follows: The crossover operator adopts a path segment preservation crossover strategy, that is, a continuous subsequence of battery swapping stations (such as [S3,S7]) is randomly selected from parent generation 1 and directly copied to the corresponding position in the offspring. The remaining positions are filled with non-conflicting stations from parent generation 2 in sequence to avoid the generation of illegal paths. The mutation operation introduces local search perturbation: a battery swapping station is randomly selected with a 5% probability to replace a candidate station with stronger service capacity within its neighborhood (radius of 3000 meters), and the path energy consumption is recalculated.

[0037] Ant colony optimization mechanisms operate in parallel, pheromones At the battery swapping station arrive Pheromone deposition occurs along the path segment. The pheromone update rules integrate two key indicators: battery swapping station service capacity. (Unit: train trips / hour) and V2G adjustment margin (Unit: kW, defined as the ratio of currently available reverse power supply to maximum rated power). The specific update formula is: ; ; For pheromone evaporation rate, It is a constant. , These are the weighting coefficients. The pheromones before the update. For the updated pheromones, This represents the change in pheromone levels. This ant colony optimization mechanism guides ants to prioritize battery swapping stations with high service capabilities and abundant V2G resources, thereby improving the overall resilience of the system.

[0038] The algorithm outputs a set of flexible routes for multi-vehicle collaboration, with the set size controlled to within 50 routes (dualistic approach is ensured through clustering to remove duplicates). When sudden demands occur (such as temporary additional orders or road closures), the system completes rescheduling within 200 milliseconds: first, affected vehicles are screened, a pre-stored subset of alternative routes is called, and after fine-tuning based on real-time road network conditions, the routes are quickly deployed, with a rescheduling response time of less than 500 milliseconds.

[0039] In the above-mentioned intelligent route planning method for freight battery swapping stations based on multi-source OD data, step (4) involves deploying the SAC reinforcement learning real-time response module at the execution layer. The specific implementation process is as follows: The system constructs a state-action space based on a high-fidelity digital twin environment. The digital twin maps the physical road network and battery swapping station facilities at a 1:1 scale, including road geometric attributes (number of lanes, speed limit), traffic signal phase, and battery swapping station equipment status (number of idle charging piles, number of vehicles in queue).

[0040] The state vector s contains 5 dimensions: the vehicle's current location (latitude and longitude coordinates), the remaining battery charge (SOC) (%), and the queue length of the nearest battery swapping station. (Train number), real-time electricity price (RMB / kWh) and battery health index (Normalized value, range [0,1]). Action space Defined as a two-dimensional continuous variable: path offset (Unit: meters, controlling whether vehicles detour to avoid congestion) and battery swapping decisions (Binary variable, 0 indicates continued driving, 1 indicates entering the battery swapping station).

[0041] The Soft Actor-Critic (SAC) algorithm learns the optimal strategy online. The policy network employs a two-hidden-layer fully connected structure, with 256 neurons per layer and ReLU activation function. Temperature coefficient. Maintaining strategy entropy through adaptive adjustment mechanism Within the interval [1.8, 2.2], a balance between exploration and exploitation is ensured. The training sample cache pool is set to 1 million experience replay data points, and the policy network is updated every 100 milliseconds. Reward function. Designed as follows: ; For the expected arrival time increment, For additional energy consumption, , , The cost weight is used. The SAC reinforcement learning real-time response module in the execution layer enables millisecond-level (<10 milliseconds) path fine-tuning and battery swapping timing determination. For example, when a sudden increase in the queue at the battery swapping station ahead is detected, it automatically suggests detouring to the next nearest station or delaying the battery swapping.

[0042] In the above-mentioned intelligent route planning method for freight battery swapping stations based on multi-source OD data, step (5) integrates a three-dimensional collaborative perception and health management closed loop. The specific implementation process is as follows: An IoT sensor network is deployed on each freight vehicle and key equipment of the battery swapping station. The vehicle-side sensors include a PT100 temperature sensor (accuracy ±0.5°C), a Hall effect voltage and current sensor (sampling rate 1kHz), and a MEMS vibration sensor (frequency response range 0-500Hz). The battery swapping station-side equipment status monitor is deployed to collect the charging pile operating current, coolant temperature, and the number of robotic arm movements. All sensor sampling frequencies are uniformly set to 10Hz, and the data is uploaded to the edge computing node via the LoRaWAN protocol (center frequency 470MHz, bandwidth 125kHz).

[0043] The edge computing node runs a lightweight battery health degradation prediction model. This model integrates incremental capacity analysis (ICA) and electrochemical impedance spectroscopy (EIS) features. Input parameters include: cumulative charge / discharge energy. (kWh), average daily temperature difference (°C) and number of fast charging cycles ICA feature extraction calculates the peak offset of the dQ / dV curve, and EIS feature extraction extracts the real part of the impedance Z_real at 1kHz. The model outputs a normalized battery health index. The calculation formula is: ; , , The attenuation coefficient is... This is the battery's rated total energy. When... When the health threshold is less than 0.75, the system automatically restricts the vehicle from participating in V2G mode and highlights it on the dispatch interface to prioritize maintenance.

[0044] Edge nodes synchronize battery health prediction results to the central cloud platform every 5 minutes, with prediction errors controlled within 3% using root mean square error (RMSE). The central platform feeds back the prediction results to the strategic layer (updating the probability distribution of battery degradation rate) and the tactical layer (adjusting the health weights in V2G_capacity calculation), forming a four-dimensional collaborative optimization closed loop covering logistics efficiency (average delivery time), battery life (average annual health degradation rate), grid stability (voltage fluctuation at battery swapping station access point ≤ ±5%), and overall cost (total expenditure per ton-kilometer). The system performs a Pareto front analysis every 24 hours, generating multi-objective balance schemes for decision-makers to choose from.

[0045] Example 2: Based on Example 1 above, the present invention also provides an alternative technical solution, the main difference being the construction method of the strategic layer optimization model. In this example, the stochastic robust optimization model in step (2) is replaced by a spatial correlation optimization model driven by a graph neural network (GNN).

[0046] Specifically, the system abstracts the urban road network into a graph structure. Node set Includes all OD hotspot grids (daily average traffic > 50 vehicles) and candidate battery swapping station locations, edge set It is composed of actual road connections. Each node... initial feature vector It includes 5 dimensions: OD demand intensity, land use type code (commercial / industrial / residential), distance to power grid substation, terrain slope, and historical accident rate. The graph neural network employs a graph attention mechanism (GAT), with node feature vector dimensions set to 128. Edge weights... Calculated by multiplying the spatial distance decay function by the functional similarity: ; For nodes and Euclidean distance (kilometers). The attenuation coefficient is... The threshold for functional similarity (based on cosine similarity between land use and OD type) is set to 0.7, and edge weights less than this value are reset to 0.

[0047] After five layers of message passing, the GNN outputs the layout priority score for each candidate node. Simultaneously, the system performs spatial autocorrelation analysis to aid decision-making: Moran's I and Geary's C are used for joint discrimination, with a calculation window radius of 5000 meters and a significance level of 0.05. High-high clusters are automatically marked as mandatory areas for battery swapping stations, while low-low clusters are excluded. The final layout scheme is solved using integer programming. The objective function still includes four-dimensional indicators such as logistics costs and grid balance, but the constraints now include spatial clustering requirements.

[0048] This embodiment was validated in another second-tier city. The layout of the battery swapping stations is more in line with the multi-center structure of the city, and the coverage of the edge areas is improved, but the calculation time is increased. This layout scheme is suitable for cities with complex road network topology and obvious functional zoning, and complements Embodiment 1.

[0049] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

[0050] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0051] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent route planning of freight battery swapping stations based on multi-source OD data, characterized in that, Includes the following steps: Construct a multi-source OD data fusion and trusted evidence storage system, collect multi-source freight origin and destination data, perform hash encryption and timestamp binding on the original OD data to achieve tamper-proof trusted evidence storage, and use a geographic weighted regression model combined with cellular automata mechanism to perform spatial correction and dynamic completion of missing or abnormal data. A strategic-level stochastic robust optimization model is established. Based on a trusted OD dataset, a global optimization function is constructed with multiple objectives, including logistics cost, power grid load balance, investment in battery swapping station construction, and battery health degradation. Stochastic disturbance variables corresponding to traffic flow, electricity price fluctuations, and battery performance degradation uncertainty are introduced. The robust optimal battery swapping station layout scheme is solved by scenario tree decomposition and dual cutting method. A tactical-level elastic path scheduling algorithm is designed. Based on the established layout of battery swapping stations, an improved genetic algorithm and an ant colony optimization mechanism are integrated. The vehicle range constraint, battery swapping waiting time, V2G reverse power supply capability, and real-time road network status are used as dynamic fitness factors to generate an elastic path set for multi-vehicle collaboration. The SAC reinforcement learning real-time response module is deployed at the execution layer. Based on the digital twin environment, a state vector containing vehicle location, remaining battery power, queue length of the battery swapping station ahead, real-time electricity price of the grid, and battery health index is constructed. The actions are spatially continuous path offset and battery swapping decision. The soft Actor-Critic algorithm is used to learn the optimal strategy online to achieve millisecond-level path fine-tuning and battery swapping timing determination. Integrating a three-dimensional collaborative perception and health management closed loop, the system collects battery temperature, internal resistance, charge-discharge cycle count, and battery swapping station equipment status in real time through an IoT sensor network. Combined with a digital twin, it constructs a battery health degradation prediction model and feeds the prediction results back to the strategic and tactical layers, forming a four-dimensional collaborative optimization closed loop covering logistics efficiency, battery life, grid stability, and overall cost.

2. The intelligent route planning method for freight battery swapping stations based on multi-source OD data according to claim 1, characterized in that, The original OD data is hashed, encrypted, and timestamped using blockchain technology. The blockchain adopts a consortium blockchain architecture, with nodes jointly maintained by logistics operators, power grid companies, and transportation management departments. The consensus mechanism is a practical Byzantine fault-tolerant algorithm, the block generation cycle is a preset time interval, and data integrity verification is achieved through a Merkle tree structure.

3. The intelligent route planning method for freight battery swapping stations based on multi-source OD data according to claim 2, characterized in that, The spatial weight function of the geographic weighted regression model adopts a Gaussian kernel function, the bandwidth parameter is dynamically adjusted through cross-validation, the neighborhood rule of the cellular automaton is defined as a specific grid size, and the state transition probability is jointly determined by the historical OD traffic density and the road network topology connectivity.

4. The intelligent route planning method for freight battery swapping stations based on multi-source OD data according to claim 3, characterized in that, The random disturbance variables include traffic speed fluctuation coefficient, time-of-use electricity price deviation rate and battery capacity decay rate. Their probability distribution is obtained by fitting historical data. The scene tree is divided into multiple main scenes and sub-scenes. The convergence threshold of the dual cutting method is set to a preset precision.

5. The intelligent route planning method for freight battery swapping stations based on multi-source OD data according to claim 4, characterized in that, The improved genetic algorithm employs a path segment retention strategy for the crossover operator, introduces local search perturbation through mutation, integrates the ant colony pheromone update mechanism with the battery swapping station service capacity and V2G adjustment margin, and determines the fitness function weight vector using the analytic hierarchy process.

6. The intelligent route planning method for freight battery swapping stations based on multi-source OD data according to claim 5, characterized in that, The V2G reverse power supply capability is quantified as the upper limit of the vehicle's release capacity. This upper limit is constrained by the current battery state of charge, health index, and grid dispatch instructions. When the grid load exceeds the preset load threshold, the system automatically activates the V2G mode and prioritizes dispatching vehicles with high health indices to participate in peak shaving.

7. The intelligent route planning method for freight battery swapping stations based on multi-source OD data according to claim 6, characterized in that, The digital twin environment maps the physical road network and battery swapping station facilities at a predetermined ratio. The state vector dimension is a preset dimension, the action space is a two-dimensional continuous variable, and the temperature coefficient α of the SAC algorithm maintains the policy entropy within a preset range through an adaptive adjustment mechanism.

8. The intelligent route planning method for freight battery swapping stations based on multi-source OD data according to claim 7, characterized in that, The calculation model of the battery health index integrates incremental capacity analysis and impedance spectrum characteristics. The input parameters include cumulative charge and discharge energy, daily average temperature difference and fast charging times. The output is a normalized health value. When the health value is less than the preset health threshold, the system automatically restricts the vehicle from participating in V2G and prioritizes maintenance.

9. The intelligent route planning method for freight battery swapping stations based on multi-source OD data according to claim 8, characterized in that, The IoT sensors include temperature sensors, voltage and current sensors, and vibration sensors. The sampling frequency is a preset frequency, and the data is uploaded to the edge computing node via the LoRaWAN protocol. The edge node synchronizes the battery health prediction results to the central cloud platform once at a predetermined time interval.

10. The intelligent route planning method for freight battery swapping stations based on multi-source OD data according to claim 9, characterized in that, In the aforementioned four-dimensional collaborative optimization closed loop, the logistics efficiency index is the average delivery time, the battery life index is the average annual health degradation rate, the power grid stability index is the voltage fluctuation amplitude at the battery swapping station access point, and the comprehensive cost index is the total expenditure per ton-kilometer. These four indicators achieve multi-objective balance through Pareto front analysis, and the system performs a global re-optimization once according to a predetermined cycle.

Images