Charging method and system for self-driving vehicle based on digital twinning

Abstract

The application discloses a kind of based on digital twin's unmanned vehicle charging method and system, and the present application relates to the technical field of digital twin, in twin simulation body, based on battery state tensor and the dynamic deduction of the planned travel path of unmanned vehicle generates power consumption envelope line;When power consumption envelope line reaches dynamic charging threshold, mark the corresponding candidate charging node, generate node matrix in combination with the prediction of concurrent bearing upper limit of charging pile group, improve the accuracy of node matrix.Based on the identification of the charging strategy to determine the corresponding non-constant power charging curve, the target charging node and the charging strategy are executed, and in the charging process of unmanned vehicle, the residual error between the charging interactive physical quantity and the deduction prediction corresponding to the twin simulation body is marked, the accuracy of the residual error is improved, and the online correction of the twin simulation body is realized.

Description

Technical Field

[0001] This invention relates to the technical field of digital twins, and more particularly to a charging method and system for autonomous vehicles based on digital twins. Background Technology

[0002] With the rapid development and commercialization of autonomous driving technology, the operating time and mileage of autonomous vehicles have increased significantly, placing extremely high demands on the timeliness, safety, and economy of their refueling. Since autonomous vehicles typically travel along planned routes and rely entirely on the system's autonomous decision-making.

[0003] In existing technologies, charging scheduling methods for autonomous vehicles are typically based on triggering charging replenishment requests using a static battery state of charge threshold, and then allocating charging to the nearest available charging station along the navigation path. These methods generally use a static, macroscopic SOC as a single scalar for both power prediction and charging triggering. This approach fails to consider the nonlinear aging of the battery throughout its lifespan, nor does it incorporate multi-physics coupling deduction based on dynamic environmental disturbances along the autonomous vehicle's planned driving path. This power prediction method, based on static scalars and empirical formulas, cannot generate a dynamically changing power consumption envelope that varies with operating conditions, leading to either delayed or premature charging threshold triggering, thus affecting the accuracy of the node matrix. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art. This invention provides a charging method and system for autonomous vehicles based on digital twins.

[0005] This invention provides a charging method for autonomous vehicles based on digital twins, comprising:

[0006] The system acquires the operational data of autonomous vehicles and the status data of charging modules, constructs a twin simulation in the digital twin space, and further generates the corresponding battery state tensor by combining the battery degradation characteristics of autonomous vehicles. In the twin simulation, the power consumption envelope is generated based on the battery state tensor and the dynamic deduction of the planned driving path of the autonomous vehicle; when the power consumption envelope reaches the dynamic charging threshold, the corresponding candidate charging node is marked, and a node matrix is ​​generated by combining the prediction of the concurrent carrying capacity of the charging pile group. Multi-objective optimization is performed on the power consumption envelope and node matrix, and the target charging node and corresponding charging strategy are determined by combining the optimization mechanism. Based on the identification of the charging strategy, the corresponding non-constant power charging curve is determined. The target charging node and charging strategy are sent out for execution. During the charging process of the autonomous vehicle, the physical quantities of the charging interaction are collected, and the residuals between the physical quantities of the charging interaction and the inferred prediction quantities corresponding to the twin simulation are marked to trigger the online correction of the twin simulation.

[0007] This invention provides a charging system for an autonomous vehicle based on a digital twin, which is applied to the aforementioned charging method for an autonomous vehicle based on a digital twin. The charging system for the autonomous vehicle based on a digital twin includes: The digital twin module is used to acquire the operating data of the autonomous vehicle and the status data of the charging module, and to build a twin simulation in the digital twin space. It further combines the battery degradation characteristics of the autonomous vehicle to generate the corresponding battery state tensor. The node matrix module is used to generate a power consumption envelope in the twin simulation based on the battery state tensor and the dynamic deduction of the planned driving path of the autonomous vehicle. When the power consumption envelope reaches the dynamic charging threshold, the corresponding candidate charging node is marked, and a node matrix is ​​generated by combining the prediction of the concurrent carrying capacity limit of the charging pile group. The charging strategy module is used to perform multi-objective optimization on the power consumption envelope and node matrix, and combine the optimization mechanism to determine the target charging node and the corresponding charging strategy. Based on the identification of the charging strategy, the corresponding non-constant power charging curve is determined. The online correction module is used to send out the target charging node and charging strategy for execution, and during the charging process of the autonomous vehicle, it collects the physical quantities of the charging interaction, marks the residual between the physical quantities of the charging interaction and the inferred prediction quantities corresponding to the twin simulation, so as to trigger the online correction of the twin simulation.

[0008] Compared with the prior art, the beneficial effects of the present invention are: (1) Obtain the operation data of the autonomous vehicle and the status data of the charging module, and construct a twin simulation in the digital twin space. Further combine the battery degradation characteristics of the autonomous vehicle to generate the corresponding battery state tensor. In the twin simulation, generate the power consumption envelope based on the battery state tensor and the dynamic deduction of the planned driving path of the autonomous vehicle. When the power consumption envelope reaches the dynamic charging threshold, mark the corresponding candidate charging node, and generate a node matrix based on the prediction of the concurrent carrying capacity limit of the charging pile group. The battery state tensor is introduced to further control the power consumption envelope and improve the accuracy of the node matrix.

[0009] (2) Perform multi-objective optimization on the power consumption envelope and node matrix, and determine the target charging node and corresponding charging strategy by combining the optimization mechanism. Based on the identification of the charging strategy, determine the corresponding non-constant power charging curve. Send the target charging node and charging strategy to be executed. During the charging process of the autonomous vehicle, collect the charging interaction physical quantity, mark the residual between the charging interaction physical quantity and the inferred prediction quantity corresponding to the twin simulation, so as to trigger the online correction of the twin simulation, further control the charging strategy, fully consider the charging interaction physical quantity and the inferred prediction quantity corresponding to the twin simulation, improve the accuracy of the residual, and realize the online correction of the twin simulation. Attached Figure Description

[0010] Figure 1 This is a schematic flowchart of a charging method for an autonomous vehicle based on digital twins according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating step S11 of the charging method for an autonomous vehicle based on digital twins in an embodiment of the present invention. Figure 3 This is a flowchart illustrating step S12 of the charging method for an autonomous vehicle based on digital twins in an embodiment of the present invention. Figure 4 This is a flowchart illustrating step S13 of the charging method for an autonomous vehicle based on digital twins in an embodiment of the present invention. Figure 5 This is a flowchart illustrating step S14 of the charging method for an autonomous vehicle based on digital twins in an embodiment of the present invention. Figure 6 This is a schematic diagram of the structural composition of a charging system for an autonomous vehicle based on digital twins, according to an embodiment of the present invention. Detailed Implementation

[0011] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0012] Please see Figures 1 to 6 A charging method for autonomous vehicles based on digital twins, applied to digital twin scenarios; the charging method for autonomous vehicles based on digital twins includes: Step S11: Obtain the operating data of the autonomous vehicle and the status data of the charging module, and construct a twin simulation in the digital twin space. Further combine the battery degradation characteristics of the autonomous vehicle to generate the corresponding battery state tensor. Step S12: In the twin simulation, a power consumption envelope is generated based on the battery state tensor and the dynamic deduction of the planned driving path of the autonomous vehicle; when the power consumption envelope reaches the dynamic charging threshold, the corresponding candidate charging node is marked, and a node matrix is ​​generated by combining the prediction of the concurrent carrying capacity limit of the charging pile group. Step S13: Perform multi-objective optimization on the power consumption envelope and node matrix, and determine the target charging node and corresponding charging strategy by combining the optimization mechanism. Based on the identification of the charging strategy, determine the corresponding non-constant power charging curve. Step S14: Send the target charging node and charging strategy to be executed, and during the charging process of the autonomous vehicle, collect the physical quantities of the charging interaction, mark the residual between the physical quantities of the charging interaction and the predicted quantities corresponding to the twin simulation, so as to trigger the online correction of the twin simulation.

[0013] refer to Figure 2 In step S11, the specific steps are as follows: S111: Real-time acquisition of multiple operational data of autonomous vehicles in different dimensions, and alignment with the status data of the charging module in the air, so as to synchronously input into the digital twin space. In the digital twin space, multiple operational data of autonomous vehicles and status data of charging modules are coupled in multiple physics fields to further construct a twin simulation body that represents the interaction characteristics of vehicle-network-environment. S112: Battery parameters are extracted based on the detection of the twin simulation. These battery parameters include electrochemical impedance spectroscopy and polarization potential. Furthermore, the battery degradation characteristics of autonomous vehicles are combined to construct a battery state tensor that integrates the spatiotemporal dimension and the degradation dimension in the twin simulation, so as to quantitatively characterize the nonlinear aging and dynamic charge-discharge boundary of the battery throughout its entire life cycle.

[0014] In the embodiments of this application, multiple operational data of the autonomous vehicle in different dimensions are collected in real time and aligned with the status data of the charging module in the air to be synchronously input into the digital twin space. In the digital twin space, multiple operational data of the autonomous vehicle and the status data of the charging module are coupled in a multi-physics field to further construct a twin simulation body that characterizes the interaction features of vehicle-network-environment.

[0015] At this time, multiple operational data of the autonomous vehicle in the dimensions of power drive, thermal management and spatial location are collected in real time. At the same time, the status data of battery cell voltage, current and module temperature in the charging module are acquired. Since there are inherent differences between the sampling frequency and transmission link of the vehicle's various sensors and controllers, there is a time delay distortion in the data arriving at the digital twin space. Therefore, a global unified clock mechanism based on Precise Time Protocol (PTP) is adopted to bind absolute timestamps to the data of each dimension. The low frequency data is mapped to the high frequency time reference through interpolation fitting algorithm to achieve microsecond-level time alignment of multi-source heterogeneous data and eliminate twin input artifacts caused by timing misalignment.

[0016] The aligned time-series data stream is spatially bound to the high-precision positioning data of the autonomous vehicle and mapped onto the 3D high-precision map road network topology in the digital twin space. By calculating the spatial correlation between the vehicle's real-time pose and the road network topology nodes, the vehicle operation data and charging module status data are accurately mounted onto the corresponding virtual spatial grid coordinates. This completes the projection of data from the physical time domain to the twin spatial domain, achieving absolute spatial alignment of the data on the spatiotemporal reference and ensuring that the boundary conditions of subsequent physical field calculations have accurate spatial location attributes.

[0017] The multidimensional data after spatial alignment is synchronously input into the digital twin space. Electrical parameters of the power drive dimension, thermal parameters of the thermal management dimension, and mechanical parameters of the spatial position dimension are extracted to reconstruct the boundary conditions of the electric-thermal-mechanical multi-physics coupling. At this time, the current and internal resistance parameters are used as the electric field heat source terms, the ambient temperature and air conditioning power are used as the thermal field boundary conditions, and the road slope and crosswind speed of the vehicle's driving trajectory are used as the mechanical load input. A joint solution matrix of the partial differential equations of each physical field is established in the twin space to realize the transient coupling of electric field heat generation and thermal field heat dissipation, as well as the cross influence of mechanical deformation on the internal contact resistance and thermal conductivity of the battery.

[0018] Based on the joint solution results of the above multi-physics coupling, a full-element twin simulation is constructed in the digital twin space. This twin simulation not only reproduces the internal micro-physical state of the autonomous vehicle and charging module, but also injects the real-time load fluctuation characteristics of the external power grid and the meteorological and geographical characteristics of the environment as dynamic constraints into the simulation. This enables the simulation to perceive the characteristics of the concurrent capacity of the external charging pile group fluctuating with the peak and valley of the power grid, as well as the characteristics of the dynamic weakening of the battery's chargeable and dischargeable boundary due to changes in ambient temperature. Thus, a digital twin simulation that can map and deduce the complex interaction state of "vehicle-grid-environment" in real time is formed.

[0019] The twin simulation refers to a virtual entity constructed in a digital twin space that can map and deduce the complex interaction processes between autonomous vehicles, charging modules, power grids, and the environment in real time. This twin simulation possesses the following capabilities: First, it receives real-time operational and status data from physical entities and reproduces the internal microscopic physical states of the vehicle and battery through a multiphysics coupling model. Second, it incorporates external dynamic constraints such as real-time power grid load fluctuations and meteorological and geographical features, enabling the simulation to perceive the concurrent capacity changes of external charging pile groups and the dynamic weakening effect of ambient temperature on the battery charging and discharging boundaries. Third, it outputs various prediction and optimization results, including battery state tensors, energy consumption envelopes, and non-constant power charging curves, and supports online correction based on actual charging interaction physical quantities. The twin simulation is the core computational carrier for realizing charging triggering, node optimization, and safe charging strategy generation in this invention.

[0020] Specifically, the autonomous vehicle is equipped with a liquid-cooled ternary lithium battery pack. The vehicle is performing an intercity highway freight mission in the early morning of winter, with an ambient temperature of 5°C and a level 7 crosswind. The vehicle is about to enter a highway service area for supercharging. During the operation of the autonomous vehicle, the system collects real-time operational data from multiple dimensions, including the drive motor torque output, the total voltage and individual cell voltage difference of the battery pack, and the power of the liquid cooling unit, as well as the status data of the temperature sensors in each module within the charging module. Given that the sampling periods of the motor controller and the battery management system are 10ms and 100ms respectively, the system uses a PTP-based global clock to bind absolute timestamps to the above data and uses a linear interpolation algorithm to fit and map the 100ms period battery data onto a 10ms reference time axis, achieving microsecond-level time alignment and eliminating the phase shift in current-voltage calculations caused by differences in sampling periods.

[0021] The system binds the time-aligned data stream with the high-precision pose data calculated by the autonomous vehicle through the fusion of RTK-GNSS and IMU, and projects it onto the three-dimensional topology of the high-speed road network in the digital twin space. By calculating the spatial correlation between the autonomous vehicle's latitude, longitude, elevation and the road network topology nodes, the system accurately attaches the vehicle's current spatial location attribute of "at an altitude of 250 meters on a long uphill section" to the virtual grid coordinates, completing the spatial alignment of the operation and status data from the physical time domain to the twin spatial domain, ensuring that the data has accurate three-dimensional spatial coordinates in the virtual space.

[0022] After data alignment, the system synchronously inputs it into the digital twin space. The system extracts the driving current of the autonomous vehicle as the electric field heat source term, extracts the ambient temperature of 5℃ and the power of the liquid cooling unit as the thermal field convection boundary conditions, and extracts the chassis mechanical vibration frequency caused by the slope of the long uphill road and the crosswind speed of level 7 as the mechanical load input. In the twin space, a joint solution matrix of electro-thermal-mechanical partial differential equations is established. The system calculates in real time the Joule heat generated by the high-frequency AC impedance inside the battery when the autonomous vehicle climbs a hill with high torque (electro-thermal coupling), and the micro-strain and contact resistance change of the battery connecting piece caused by the vibration of the chassis under crosswind and road excitation (mechanical-electrical coupling), thus completing the reconstruction of the multi-physics field coupled boundary conditions.

[0023] Based on the solution results of the above multiphysics coupling, the system constructs a full-element twin simulation of the autonomous vehicle in the digital twin space. This simulation not only maps in real time the microscopic state of the bottom module inside the autonomous vehicle's battery pack, which is caused by uneven crosswind cooling and a surge in polarization resistance, but also senses the power fluctuation characteristics of the supercharging station in the service area ahead due to increased heating load, as well as the weakening characteristics of the maximum output current of the liquid-cooled cables of the charging pile due to low ambient temperature by accessing external data. Thus, this twin simulation accurately represents the real physical state of the autonomous vehicle under the complex vehicle-network-environment interaction characteristics of "low temperature climate - high consumption during uphill climbing - limited grid load", providing a high-fidelity twin basis for the subsequent generation of battery state tensor and dynamic deduction of power consumption envelope.

[0024] Furthermore, battery parameters, including electrochemical impedance spectroscopy and polarization potential, are extracted based on the detection of the twin simulation. These parameters are then combined with the battery degradation characteristics of autonomous vehicles to construct a battery state tensor that integrates spatiotemporal and degradation dimensions in the twin simulation, thereby quantitatively characterizing the nonlinear aging and dynamic rechargeable / discharge boundary of the battery throughout its entire life cycle.

[0025] At this point, based on the constructed twin simulation, utilizing its multi-physics coupling characteristics, during the operation or standby of the autonomous vehicle, a broadband weak pulse excitation signal is injected through the electric field boundary conditions, or the current ripple of the power motor during gear shifting is intercepted as an equivalent excitation. The twin simulation performs a fast Fourier transform and frequency domain fitting on the voltage response generated by this excitation, extracts the electrochemical impedance spectrum that varies with frequency, and combines the current step transient response output by the electric field solution matrix to separate and calculate the ohmic internal resistance, charge transfer impedance and solid-phase diffusion impedance, thereby obtaining the polarization potential parameters that reflect the degree of interface dynamic polarization.

[0026] The extracted electrochemical impedance spectroscopy and polarization potential parameters are deeply correlated and mapped with the historical operation logs and charge-discharge cycle counts accumulated by the autonomous vehicle throughout its entire life cycle. At this point, the increment of charge transfer impedance and the shift of polarization potential are used as characterization indicators of interfacial active material loss, and the slope change of solid-phase diffusion impedance is used as a characterization indicator of lithium-ion insertion / extraction kinetics obstruction. Combined with the duration of historical extreme temperature exposure and the frequency of overcharge and over-discharge, the microscopic decay mechanisms such as lithium-ion consumption, SEI film thickening, and active material lattice distortion are quantified, thereby constructing a decay feature vector characterizing the nonlinear aging trajectory of the battery from manufacturing to the present moment.

[0027] The battery pack's spatial topology is used as the spatial dimension, the electrochemical impedance spectroscopy, polarization potential, and decay characteristic vector are used as the characteristic decay dimension, and the timestamp sequence of the entire life cycle is used as the time dimension. A battery state tensor is constructed in a twin simulation. The polarization potential and charge transfer impedance of each cell in this tensor are input into a pre-calibrated electrochemical aging mechanism model. The polarization potential is used to approximate the lithium plating critical threshold as the charging boundary constraint, and the impedance heat generation rate is used to approximate the thermal runaway critical temperature as the discharging boundary constraint. The maximum allowable charging current curve and the maximum allowable discharging power envelope, which fluctuate in real time with time and decay state, are dynamically solved and output, realizing the quantitative characterization of the dynamic chargeable and dischargeable boundaries.

[0028] The battery state tensor is a multidimensional data structure used to quantify the electrochemical state of a power battery for autonomous vehicles at different spatial locations, degradation levels, and time scales throughout its entire lifecycle. In this tensor, the spatial coordinates of each cell in the battery pack serve as the first dimension (spatial dimension); the electrochemical impedance spectroscopy (EIS) characteristic parameters, polarization potential, cumulative charge-discharge cycle count, and capacity decay rate of the cell at the current moment serve as the second dimension (characteristic decay dimension); and the time series of the battery from its first use to the current moment serves as the third dimension (time dimension). The EIS includes ohmic internal resistance, charge transfer impedance, and solid-state diffusion impedance, while the polarization potential characterizes the kinetic polarization degree of the positive and negative electrode interfaces. By constructing the battery state tensor, the system can dynamically describe the nonlinear aging state and rechargeable / discharge capability boundaries of each cell at different lifecycle stages, spatial locations, and operating times within a digital twin space.

[0029] Specifically, when the autonomous vehicle is cruising at high speed in a 5°C environment with a full load, the system performs online extraction of electrochemical parameters based on the constructed twin simulation. Since the autonomous vehicle is in the acceleration and overtaking phase, the power battery output current has high-frequency transient ripple. The twin simulation extracts this ripple as an equivalent wideband excitation and performs fast Fourier transform and frequency domain fitting on the voltage response signal synchronously acquired by the BMS to extract the electrochemical impedance spectrum of the liquid-cooled ternary lithium battery pack of the autonomous vehicle in real time. At the same time, by solving the electric field matrix inside the simulation, the voltage rebound characteristics at the moment of current step are captured, the ohmic internal resistance and charge transfer impedance are accurately separated, and the polarization potential of the positive and negative electrode interfaces of the battery under the current low temperature and high rate discharge condition is calculated. It is found that the polarization potential of the 4th module located on the windward bottom side of the battery pack has significantly increased to 0.12V.

[0030] The system extracts the abnormally high charge transfer impedance and polarization potential of the fourth module and performs a deep correlation mapping with the accumulated operation logs of the autonomous vehicle throughout its entire life cycle. By comparing historical data, it is confirmed that the impedance increase in this area is not a sudden fault, but rather due to the autonomous vehicle's long-term nighttime high-speed missions in the early stages, which caused the windward module to be in a local low-temperature operating state for a long time, resulting in continuous thickening of the SEI film and local microscopic lithium deposition. Combined with the vehicle's accumulated history of 1200 deep charge and deep discharge cycles, the system quantitatively evaluates that the loss rate of interfacial active material of the fourth module is 18% higher than that of other modules, thereby constructing a decay feature vector characterizing this local nonlinear aging trajectory.

[0031] Based on the above extraction results, the system constructs a battery state tensor that integrates spatiotemporal and degradation dimensions in the twin simulation. This tensor uses the spatial arrangement of the 96 cells of the autonomous vehicle as the spatial dimension, the impedance spectrum characteristics, polarization potential, and aging degradation vector of each cell as the characteristic degradation dimension, and the full life cycle timestamp as the time dimension. The system inputs the high polarization potential (0.12V) of the 4th module in this tensor and the low temperature boundary condition into the electrochemical aging mechanism model, and calculates that under 5℃ conditions, if a conventional 250kW ultra-high temperature battery is used... When charging at high power, the polarization potential of the fourth module will instantly exceed the critical threshold of 0.15V for lithium plating, triggering irreversible lithium dendrite growth. Therefore, the maximum allowable charging current of this module needs to be limited to 0.4C by tensor dynamic solution and quantization output, while other healthy modules can withstand 1.2C charging. This allows for precise quantization and characterization of the dynamic chargeable and dischargeable boundary of the autonomous vehicle battery pack under low-temperature degradation state in tensor form, which is "locally restricted but overall rechargeable". This provides an insurmountable safety constraint for the subsequent generation of non-constant power charging curves.

[0032] refer to Figure 3 In step S12, the specific steps are as follows: S121: Real-time monitoring of the twin simulation object to obtain the planned driving path of the autonomous vehicle. Based on the identification of the planned driving path of the autonomous vehicle, multiple key driving nodes are determined, and dynamic deduction is performed in combination with the battery state tensor and environmental disturbance factors to generate the corresponding power consumption envelope. S122: Mark the dynamic charging threshold based on the battery state tensor dynamic adjustment, compare the confidence lower bound of the power consumption envelope with the dynamic charging threshold, if the confidence lower bound of the power consumption envelope reaches the dynamic charging threshold, then the spatial coordinates of the time of reaching the threshold are reverse-mapped to the charging module and marked as a candidate charging node. Further, combine the upper limit of the charging pile group's group generation capacity and the node transformer load rate for dynamic prediction to generate a node matrix representing the spatiotemporal resource constraints and the grid capacity boundary.

[0033] In the embodiments of this application, the twin simulation is monitored in real time to obtain the planned driving path of the autonomous vehicle. Based on the identification of the planned driving path of the autonomous vehicle, multiple key driving nodes are determined, and dynamic deduction is performed in combination with battery state tensor and environmental disturbance factors to generate the corresponding power consumption envelope. This approach is compatible with the overall consideration of the planned driving path of the autonomous vehicle and ensures the accuracy of multiple key driving nodes.

[0034] At this point, the task scheduling layer of the real-time monitoring twin simulation object obtains the planned driving path issued by the autonomous vehicle, analyzes the high-precision map topology data and elevation data of the path, and identifies the topological feature points and geographical feature points in the path that cause sudden changes in driving conditions. Then, it extracts the starting points of uphill and downhill slopes with absolute slope values ​​exceeding a set threshold, the groups of curves with radii of curvature less than a set threshold, and the ramp entrances and toll stations where deceleration and lane changes must be performed, and determines them as key driving nodes, thereby discretizing the continuous planned driving path into continuous characteristic condition intervals defined by multiple key driving nodes.

[0035] For characteristic operating condition intervals between adjacent key driving nodes, the battery state tensor constructed within the twin simulation is used as the underlying constraint, and the environmental disturbance factor is used as the dynamic boundary to perform energy consumption extrapolation through multi-physics coupling. At this time, the decay dimension data representing the polarization potential and internal resistance increment in the battery state tensor are mapped to the current interval temperature conditions to correct the battery's charging and discharging efficiency and internal resistance heat generation rate. Simultaneously, the relative wind resistance is calculated by superimposing the wind speed vector with the vehicle driving vector, and the ambient temperature and light radiation are input into the thermodynamic solver to correct the energy consumption of the battery pack's thermal management system, thereby extrapolating the maximum energy consumption rate of each characteristic operating condition interval under the coupling of the battery's current decay state and the worst environmental disturbance.

[0036] The maximum energy consumption rate of each characteristic operating condition interval is used as the upper limit of the interval, and the minimum energy consumption rate of the vehicle's adaptive cruise under the most ideal environmental conditions is used as the lower limit of the interval. Considering the speed fluctuations caused by random traffic flow obstruction and the hysteresis effect of the battery thermal management system, Markov chain or Monte Carlo method is used to calculate the propagation and accumulation of uncertainties between adjacent intervals. According to the temporal position of key driving nodes in the planned driving path, the upper and lower limits of energy consumption derived from each interval are sequentially integrated and stitched in the two-dimensional coordinate system of state of charge-driving distance to generate an energy consumption envelope that characterizes the range of state of charge fluctuation when the autonomous vehicle arrives at each spatial node.

[0037] At this point, the power consumption envelope is a band curve characterizing the uncertainty range of the remaining state of charge of an autonomous vehicle as it travels along a planned path, varying with distance. Specifically, it consists of a confidence upper bound curve and a confidence lower bound curve. The confidence upper bound curve represents the trajectory of the highest remaining power the vehicle can achieve as a function of distance under relatively ideal environmental conditions and minimal environmental disturbances. The confidence lower bound curve represents the trajectory of the lowest remaining power the vehicle can achieve as a function of distance under the most adverse conditions, where environmental disturbances are coupled with the current battery degradation state.

[0038] Specifically, during the process of synchronizing task data from the autonomous vehicle to the digital twin space, the system monitors the twin simulation in real time and obtains the planned driving path issued by it to the highway service area ahead. The system then analyzes the high-precision map topology and elevation data of the path, accurately identifying a continuous long uphill section (node ​​K1) with a gradient of 4% and a length of 3 kilometers, as well as the merging ramp (node ​​K2) that follows. The system extracts nodes K1 and K2 and the current position as key driving nodes, thereby discretizing the autonomous vehicle's subsequent driving trajectory into three characteristic operating condition intervals: "flat road cruising - long uphill - ramp deceleration".

[0039] For the aforementioned characteristic operating conditions, the system uses the battery state tensor of the autonomous vehicle within the twin simulation as the underlying constraint, and performs range energy consumption extrapolation by combining the environmental disturbance factors of 5°C ambient temperature and level 7 crosswind. At this time, the system maps the dimension data representing the surge in polarization internal resistance and life-cycle decay of the fourth module in the battery state tensor to the low-temperature condition of 5°C, and calculates the decrease in discharge efficiency and surge in Joule heat caused by the significant increase in battery internal resistance. At the same time, after vector decomposition of the level 7 crosswind, it is found that in the long uphill section where the autonomous vehicle is driving towards node K1, the crosswind and the driving direction form a lateral wind resistance of nearly 90 degrees, which greatly increases the chassis driving resistance. Coupled with the auxiliary high power consumption caused by the start of the PTC heater of the battery liquid cooling unit due to the low temperature of 5°C, the system extrapolates that the maximum power consumption rate of the autonomous vehicle in this uphill section will be as high as 2.5 times that of a normal flat road.

[0040] The system performs uncertainty propagation calculations and envelope generation based on the extrapolation results of each interval. Due to the high internal resistance and heavy load superposition in long uphill sections, the battery temperature may fluctuate beyond the PTC start-stop threshold, causing a jump in auxiliary energy consumption. The system uses the Monte Carlo method to accumulate and calculate this thermodynamic hysteresis uncertainty. According to the spatial and temporal sequence of nodes K1 and K2, the system takes the ideal minimum energy consumption for flat road cruising as the lower bound and the maximum energy consumption under the worst coupling of uphill crosswind and polarization internal resistance as the upper bound. By integrating and stitching in the state of charge-distance coordinate system, a strip-shaped energy consumption envelope that expands with mileage is generated. The lower bound of this envelope shows that the SOC is 18% remaining when arriving at the service area, while the upper bound reaches the extremely low energy range of only 8%. This intuitively and rigorously represents the most severe risk of energy depletion that the autonomous vehicle may face before reaching the charging station under the current degradation state and adverse environmental disturbances, providing a preliminary extrapolation basis for the accurate triggering of the subsequent dynamic charging threshold.

[0041] Furthermore, a dynamic charging threshold based on battery state tensor dynamic adjustment is marked, and the confidence lower bound of the power consumption envelope is compared with the dynamic charging threshold. If the confidence lower bound of the power consumption envelope reaches the dynamic charging threshold, the spatial coordinates of the time of reaching the threshold are reverse-mapped to the charging module and marked as a candidate charging node. The upper limit of the charging pile group's capacity and the node transformer load rate are further combined for dynamic prediction to generate a node matrix representing the spatiotemporal resource constraints and the grid capacity boundary. The battery state tensor is introduced to further control the power consumption envelope and improve the accuracy of the node matrix.

[0042] At this point, the battery state tensor constructed in the twin simulation is extracted, and the decay dimension data representing the polarization potential and internal resistance increment in the tensor is mapped to the preset safety boundary constraint model. At this point, the polarization potential approaching the lithium plating critical threshold is used as the charging safety boundary, and the slope of the nonlinear aging trajectory throughout the entire life cycle is used as the life decay constraint. The minimum safe charging state of charge (SOC) and the maximum allowable charging rate under the current decay state are dynamically solved, and then the minimum redundant charge that the battery must retain in the current state to avoid irreversible damage is deduced. The vehicle-side SOC value corresponding to the minimum redundant charge is calibrated as the dynamic charging threshold that dynamically floats with the battery decay state.

[0043] The confidence lower bound of the power consumption envelope at the current moment is extracted in real time and compared with the adaptively calibrated dynamic charging threshold in space on a time-by-time basis. The confidence lower bound represents the lowest state of charge that the vehicle may reach under the coupling of extreme environmental disturbances and the worst operating conditions. When the value of the confidence lower bound is equal to or lower than the dynamic charging threshold, a charging demand is triggered, the timestamp of the arrival time is locked, and spatial reverse tracing is performed in the high-precision map topology of the twin simulation based on the timestamp to obtain the three-dimensional spatial coordinates of the autonomous vehicle at the arrival time. The three-dimensional spatial coordinates are then mapped inversely to the geographic information system of the charging module and marked as a candidate charging node.

[0044] For the charging pile cluster within the radiation range of the candidate charging node, the real-time active power and historical load curves of the transformer on the site side are obtained. Combined with meteorological data and holiday traffic flow characteristics, the basic power load for future periods is predicted. At the same time, based on the vehicle-to-grid interactive cloud platform, the reservation queue sequence and expected charging time of other new energy vehicles are obtained. The rated capacity of the transformer is subtracted from the predicted value of the basic power load and the predicted value of the charging load of other vehicles to calculate the remaining available capacity of the candidate charging node during the expected arrival time of the autonomous vehicle. Then, the concurrent carrying capacity limit of the charging pile cluster and the predicted value of the node transformer load rate are output.

[0045] Using candidate charging nodes as row vector elements, the temporal-spatial dimension is formed by the expected arrival time of vehicles and the spatial location deviation, the power grid capacity dimension is formed by the upper limit of concurrent carrying capacity of charging pile groups and the predicted value of node transformer load rate, and the physical resource dimension is formed by the number of idle charging piles and the maximum charging power of each candidate charging node. The above multi-dimensional parameters are matrix-reorganized to generate a node matrix representing the temporal-spatial resource constraints and the power grid capacity boundary. Each element set in this matrix uniquely maps the comprehensive available service capacity of a candidate charging node within a specific time window.

[0046] Specifically, during the autonomous vehicle's journey along the planned path, the system extracts the battery state tensor constructed from the twin simulation to perform adaptive calibration of the dynamic charging threshold. Since the battery state tensor shows that the polarization resistance of the fourth module surges at a low temperature of 5°C and exhibits a microscopic lithium plating trend, the system maps this polarization resistance increment to the safety boundary constraint model. It calculates that if the SOC is below 15%, the polarization potential of the fourth module will easily exceed the 0.15V lithium plating critical threshold during subsequent low-temperature high-rate discharge and pulse charging switching. To retain sufficient polarization buffer margin, the system dynamically increases the 10% static charging threshold set under normal conditions, calibrating the current dynamic charging threshold of the autonomous vehicle to an SOC of 20%.

[0047] The system extracts the confidence lower bound of the power consumption envelope in real time and compares it with the 20% dynamic charging threshold. Due to the superposition of level 7 crosswinds and long uphill conditions, the confidence lower bound of the envelope drops sharply. When the system extrapolates to a point 8 kilometers away from the next highway service area, the confidence lower bound drops to 19.5%, reaching the dynamic charging threshold. The system then locks the timestamp of this moment and performs spatial reverse tracing in the high-precision map topology of the twin simulation to obtain the three-dimensional spatial coordinates of the autonomous vehicle at the highway station K156+500 at that moment. This coordinate is then mapped back to the geographic information system of the charging module and marked as candidate charging node N1.

[0048] For the supercharging station in the highway service area pointed to by candidate charging node N1, the system performs dynamic prediction of the concurrent load capacity of the charging pile group and the transformer load rate. The system retrieves data from 10 250kW liquid-cooled supercharging piles at the supercharging station and finds that the rated capacity of the transformer is 2000kVA. Combined with the predicted value of 200kVA of the basic power load (lighting and HVAC) of the service area at 5:00 am, and the V2G cloud platform showing that there are currently 4 vehicles charging at an average power of 180kW, the system predicts that 15 minutes after the autonomous vehicle is expected to arrive, the transformer load rate will reach 75%, and the remaining available capacity can only support 2 supercharging piles to operate at full power concurrently. That is, the predicted value of the concurrent load capacity of the charging pile group is 2.

[0049] The system uses candidate charging node N1 as its core and combines surrounding alternative nodes to generate a node matrix representing spatiotemporal resource constraints and grid capacity limits. In this node matrix, the row vectors represent node N1, and the column vectors map to: "spatial-temporal dimension: 8 kilometers away from the autonomous vehicle, expected to arrive in 15 minutes", "grid capacity dimension: transformer load rate predicted at 75%, concurrent capacity limit of 2 units", and "physical resource dimension: 2 remaining idle 250kW liquid-cooled supercharging piles". Thus, the node matrix quantifies the spatiotemporal and grid-constrained state of the supercharging station at the moment of arrival of the autonomous vehicle with a rigorous mathematical structure, providing insurmountable physical and grid boundary constraints for subsequent multi-objective optimization to select target charging nodes and generate non-constant power charging curves.

[0050] refer to Figure 4 In step S13, the specific steps are as follows: S131: Mark the charging time window tolerance represented by the power consumption envelope, collect the node queuing loss and grid impact represented by the node matrix, and perform multi-objective Pareto optimization in combination with the charging time window tolerance, thereby determining the target charging node and its corresponding target charging strategy by combining the optimization mechanism based on non-dominated sorting. S132: Based on the traceability of the battery state tensor, the corresponding polarization tolerance and lithium plating boundary are determined and matched with the target charging strategy to generate a non-constant power charging curve to prevent lithium plating. The non-constant power charging curve is composed of multiple adaptive decay pulses and intermittent depolarization rest periods.

[0051] In the embodiments of this application, the charging time window tolerance represented by the power consumption envelope is marked, and the node queuing loss and grid impact represented by the node matrix are collected. Multi-objective Pareto optimization is performed in combination with the charging time window tolerance, thereby combining the optimization mechanism based on non-dominated sorting to determine the target charging node and its corresponding target charging strategy.

[0052] At this point, spatiotemporal analysis is performed on the generated power consumption envelope to extract the time nodes when the confidence upper and lower bounds of the power consumption envelope reach the dynamic charging threshold, respectively. The time difference between the two is calculated to characterize the charging time window tolerance. At this point, the time when the confidence upper bound reaches the dynamic charging threshold reflects the limit time that the vehicle can delay charging under the most ideal operating conditions, and the time when the confidence lower bound reaches the dynamic charging threshold reflects the hard cutoff time when the vehicle must complete charging access under the worst operating condition coupling. This time window tolerance quantifies the policy schedulable margin of the vehicle under the driving terminal time constraint.

[0053] The generated node matrix is ​​analyzed, and dimensional data representing spatiotemporal resource constraints and grid carrying capacity boundaries are extracted. For node queuing losses, the queue length at the expected arrival time and the average service time per vehicle are extracted, and the idle waiting time from vehicle access to actual physical power-on is calculated. For grid impacts, the predicted value of node transformer load rate and concurrent carrying capacity limit are extracted. Combined with the current peak and valley electricity price period and frequency adjustment requirements of the grid, the risk of transformer active power surge and reactive voltage fluctuation exceeding the limit caused by high-power charging access is quantitatively assessed, thereby forming a feature vector representing queuing time cost and grid safety cost.

[0054] With minimizing node queuing losses and grid impact as the objective functions and the charging time window tolerance as the global time constraint boundary, a multi-objective Pareto optimization model is constructed. Here, node queuing losses are mapped to the objective function of operating time cost, and grid impact is mapped to the objective function of grid safety penalty. At the same time, the charging time window tolerance is introduced as a hard constraint into the fitness evaluation system. That is, any candidate strategy that causes the actual charging completion time of the vehicle to exceed the confidence lower bound arrival time is regarded as an infeasible solution. Thus, within the feasible region defined by the time window tolerance, the Pareto non-dominated solution front, which is the inverse relationship between queuing losses and grid impact, is solved.

[0055] For the multi-objective Pareto optimization model, the system uses two conflicting objective functions as optimization objects. The first objective function is to minimize the node queuing loss, which is quantified in seconds and is specifically equal to the estimated waiting time between a vehicle arriving at a candidate charging node and actually starting charging. This waiting time is calculated by multiplying the queue length of the node within the current time window by the average service time per vehicle at that node.

[0056] The second objective function is to minimize the grid impact, which is quantified by a dimensionless penalty exponent. Specifically, it is equal to the ratio of the requested charging power of the vehicle at the moment of connection to the remaining available capacity of the transformer at that node, multiplied by the normalized value of the difference between the current transformer load rate and the preset upper limit of the load rate at that node. The higher the exponent, the greater the risk of impact on the grid.

[0057] The system also introduces a global time constraint boundary, namely the charging time window tolerance. This tolerance is defined as the time difference between the time when the upper confidence bound of the energy consumption envelope reaches the dynamic charging threshold and the time when the lower confidence bound reaches the dynamic charging threshold. In this optimization model, any candidate charging strategy that causes the vehicle to actually complete charging later than the time when the lower confidence bound reaches the dynamic charging threshold is considered an infeasible solution and will not participate in subsequent optimization comparisons.

[0058] In the multi-objective Pareto optimization model, a genetic optimization mechanism based on non-dominated sorting is introduced to determine the dominance relationship between individuals in the initial candidate population and to perform hierarchical sorting. At this time, the congestion distance of each individual in the population in the dual objective space of queuing loss and grid impact is calculated. Individuals with high non-dominated levels and large congestion distances are preferentially retained to ensure the diversity and convergence of solution distribution. Through iterative crossover and mutation operations, the true Pareto front is continuously approximated. Finally, based on the pre-set operational preference weights, such as prioritizing grid safety or prioritizing shortening operating time, a compromise solution is selected from the optimal frontier of non-dominated sorting, and the determined target charging node and its corresponding target charging strategy containing the predetermined arrival time and requested charging power are output.

[0059] For the genetic optimization mechanism based on non-dominated sorting, the system takes all candidate charging nodes that meet the time window constraints and their corresponding charging power request schemes as individuals in the initial population. Each individual contains three dimensions of information: the spatial location of the target charging node, the estimated arrival time, and the requested initial charging power. For any two individuals A and B in the population, the system determines their dominance relationship according to the following rules: if the value of individual A on the queuing loss target is less than or equal to the value of individual B on the same target, and the value of individual A on the grid impact target is also less than or equal to the value of individual B on the same target, and the value of individual A on at least one target is strictly less than that of individual B, then individual A is determined to dominate individual B.

[0060] Based on the aforementioned dominance relationships, the system performs hierarchical sorting of all individuals in the population. Specifically, the process is as follows: First, identify all individuals in the population that are not dominated by any other individual and mark them as the first non-dominated layer. Then, temporarily remove the individuals from the first non-dominated layer from the population, and again identify all individuals in the remaining population that are not dominated by any other individual and mark them as the second non-dominated layer. This process is repeated until all individuals in the population are assigned a non-dominated layer number. The individuals in the first non-dominated layer constitute the Pareto optimal frontier, representing the set of non-dominated solutions that achieve the optimal trade-off between queuing losses and grid impact.

[0061] To further filter among multiple optimal solutions in the first non-dominated layer regarding congestion distance, the system calculates the congestion distance for each individual. Specifically, for all individuals in the first non-dominated layer, the system sorts them according to the queuing loss target value from smallest to largest, setting the congestion distance of individuals at both ends of the sort to infinity. For each individual in the middle, its congestion distance is equal to the difference between its queuing loss target and the next adjacent individual divided by the maximum difference between the two individuals at that target, plus the difference between its grid impact target and the next adjacent individual divided by the maximum difference at that target. The larger the congestion distance, the sparser the individual's position in the target space, meaning that the solution has better diversity when balancing the two targets. In each generation of optimization, the system prioritizes retaining individuals with lower non-dominated layers (i.e., smaller layer numbers). When two individuals are in the same non-dominated layer, the system prioritizes retaining individuals with larger congestion distances. In this way, the system selects a representative set of solutions distributed on the Pareto front without relying on subjective weights.

[0062] Specifically, in the twin simulation of autonomous vehicles, the system performs spatiotemporal analysis on the generated power consumption envelope to extract the charging time window tolerance. Due to the continuous disturbance of the 5°C environment and level 7 crosswinds, the upper confidence bound of the power consumption envelope shows that the vehicle will reach the dynamic charging threshold of 20% after 60 minutes, while the lower confidence bound shows that the vehicle will reach this threshold in only 35 minutes under the worst conditions. Based on this, the system calculates that the charging time window tolerance is only 25 minutes. This narrow tolerance strictly defines that the autonomous vehicle must complete the physical connection to the charging pile within 35 minutes, otherwise it will face the risk of polarization lithium plating runaway under extremely low SOC.

[0063] The system analyzes the node matrix corresponding to the supercharging stations in the highway service area ahead, and collects node queuing losses and grid impact characteristics. The node matrix shows that if the autonomous vehicle selects the standard 250kW supercharging node N1, although there are currently 2 idle charging piles, the V2G cloud platform predicts that 3 other vehicles will arrive in sequence when the autonomous vehicle arrives, resulting in a queuing loss of up to 12 minutes. More seriously, at 5 a.m., the service area's heating load is at its peak, and the transformer load rate has reached 75%. If the autonomous vehicle directly requests 250kW full-power charging, it will generate a drastic change in active power, which is very likely to cause the reactive voltage fluctuation on the low-voltage side of the transformer to exceed the limit, causing the grid impact penalty index to soar.

[0064] Based on the above constraints, the system constructs a multi-objective Pareto optimization model and performs optimization. The system uses queuing loss and grid impact as dual objective functions, and the 25-minute charging time window tolerance as a hard constraint boundary. During the optimization process, the system finds that if the system detours 5 kilometers to the suboptimal candidate node N2, although the travel time increases by 4 minutes, this node is located in a lightly loaded grid area and does not require queuing, so the queuing loss is 0 and the grid impact is minimal. By iterating the scheme using a genetic algorithm based on non-dominated sorting, it is found that the strategy of node N2 is at the first non-dominated level in the dual objective space of queuing loss and grid impact, and satisfies the time window tolerance constraint, such as 4 minutes of detour + 0 minutes of queuing = 4 minutes, which is much smaller than the 25-minute tolerance.

[0065] Based on the preference weight of prioritizing grid security and overall operational efficiency, the system selects the compromise solution from the Pareto optimal frontier and determines the second-best candidate node N2 as the target charging node. At the same time, the system determines the corresponding target charging strategy as follows: the autonomous vehicle immediately changes lanes and drives towards node N2, expected to arrive in 12 minutes, and requests an initial charging power limit of 150kW to suppress grid impact. This strategy perfectly balances the autonomous vehicle's energy replenishment needs, queuing time costs, and the safety carrying capacity boundary of the local grid within the strict charging time window, laying the decision-making foundation for the subsequent generation of non-constant power charging curves.

[0066] Furthermore, the corresponding polarization tolerance and lithium plating boundary are determined by tracing the battery state tensor, and matched with the target charging strategy to generate a non-constant power charging curve to prevent lithium plating. The non-constant power charging curve consists of multiple adaptive decay pulses and intermittent depolarization rest periods, which are compatible with the overall consideration of the battery state tensor and ensure the accuracy of the corresponding polarization tolerance and lithium plating boundary.

[0067] At this point, the battery state tensor constructed in the twin simulation is subjected to reverse tracing in the time dimension and in-depth analysis in the feature dimension to extract the polarization internal resistance increment and charge transfer impedance under the current thermodynamic and kinetic coupling conditions. The above impedance parameters are input into the preset electrochemical mechanism model, and the lithium plating boundary voltage at which lithium ion deposition does not occur on the negative electrode surface under the current battery aging state and temperature field is solved with the negative electrode solid-liquid phase potential difference approaching zero volt as the absolute safety threshold. At the same time, the difference between the lithium plating boundary voltage and the current actual polarization potential is used as the safety margin to quantify and calculate the polarization tolerance under the current state. This polarization tolerance characterizes the maximum concentration polarization and electrochemical polarization superposition amplitude that the battery can withstand during charging.

[0068] The initial charging power limit requested in the determined target charging strategy and the node grid carrying capacity constraint are mapped into a charging current command, and this current command is dynamically matched with the solved polarization tolerance. At this time, the maximum transient charging current allowed by the current battery state is calculated based on the polarization tolerance, and the smaller value between its intersection with the current limit given by the target charging strategy is taken as the reference current for the initial constant current stage. At the same time, the initial duty cycle and pulse width of the adaptive decay pulse are set with the decay rate of the polarization tolerance as a constraint to ensure that the accumulated polarization potential is always under the envelope of the polarization tolerance during the energizing cycle of pulse charging.

[0069] Based on the initialized pulse parameters, the polarization potential evolution trajectory during the charging process is simulated in the twin simulation, and the rising slope and cumulative amplitude of the polarization potential during the pulse energization period are calculated in real time. When the simulation shows that the polarization potential approaches the safety margin threshold set by the polarization tolerance, the polarization approach judgment mechanism is triggered, the current output is forcibly cut off, and an intermittent depolarization resting period with zero current is adaptively interpolated. During the resting period, based on the concentration polarization diffusion time constant and the charge transfer relaxation time constant, the relaxation dissipation process of the lithium ion concentration gradient inside the solid phase is calculated. When the polarization potential falls back below the safety recovery threshold, the depolarization is determined to be complete, thereby adaptively determining the duration of the resting period.

[0070] The adaptive decay pulse's energizing period and intermittent depolarization resting period are sequentially arranged. As the depth of charge increases, the battery's state of charge (SOC) rises, leading to a reduction in the lithium intercalation channel and a narrowing of polarization tolerance. Based on a polarization approximation judgment mechanism, the system reduces the amplitude of the next cycle pulse current in real time and extends the duration of the depolarization resting period, forming a waveform sequence of multiple segments alternating between adaptive decay pulses and intermittent depolarization resting periods. This waveform sequence is then converted into a power-time mapping relationship to generate a non-constant power charging curve for the entire process. This curve, in principle, prevents the negative electrode potential from crossing the lithium plating boundary, achieving safe fast charging with lithium plating prevention.

[0071] Specifically, during the process of the autonomous vehicle driving towards the target charging node N2, the system traces and analyzes the battery state tensor within the twin simulation to solve for the polarization tolerance and lithium plating boundary. Since the tensor data shows that the charge transfer impedance of the fourth module of the autonomous vehicle battery pack surges under the coupling of low temperature (5℃) and long-term aging, the system inputs this impedance parameter into the electrochemical mechanism model and solves that under the operating condition of 5℃, the solid-liquid phase potential difference on the negative electrode surface of this module will approach the lithium plating boundary voltage of 0V when the charging current exceeds 0.5C. Thus, the system calculates that the fourth module currently has extremely low tolerance for charging polarization, with a polarization potential safety margin of only 0.03V, and even a slight large current will exceed the boundary and trigger lithium plating.

[0072] The system dynamically matches the initial request of "150kW charging limit: mapped to approximately 0.8C current" in the target charging strategy with the polarization tolerance. Given that the 0.8C current far exceeds the 0.5C polarization tolerance limit, the system performs a strategy degradation matching, clamping the initial pulse current amplitude to 0.5C. At the same time, in order to accelerate the energy replenishment efficiency, the system sets a charging mode based on the third law of Musk, which alternates between 0.5C constant current pulses and short rest periods, and initializes the pulse power-on time to 10 seconds.

[0073] Based on the initialization parameters, the system simulates the charging process in the twin simulation and performs adaptive interpolation for polarization approximation judgment and intermittent depolarization rest period. The simulation shows that after 6 seconds of 0.5C current application, the polarization potential of the fourth module rapidly rises and approaches the safety margin threshold of 0.03V. The system immediately triggers the polarization approximation judgment mechanism, cuts off the current output at the 7th second, and inserts the interpolation into the intermittent depolarization rest period. During the rest period, the system calculates based on the relaxation time constant of lithium ions in the graphite solid phase at low temperature that at least 5 seconds are required for the concentration polarization to dissipate below the safety recovery threshold. Therefore, the system adaptively determines the rest period of this cycle to be 5 seconds, and prepares to enter the next pulse cycle after the depolarization is determined to be complete.

[0074] The system alternates between power-on and rest periods to generate a non-constant power charging curve that prevents lithium plating. As the SOC of the autonomous vehicle gradually increases from 20% to 50%, the lithium intercalation channel gradually saturates, and the polarization tolerance further narrows. Based on polarization approach judgment, the system reduces the pulse current of the next cycle from 0.5C to 0.4C in real time, while extending the depolarization rest period from 5 seconds to 8 seconds. As a result, the system generates a non-constant power charging curve that smoothly transitions from multiple segments of "0.5C-6s power-on + 5s rest" to "0.4C-5s power-on + 8s rest". Under the premise of satisfying the overall power envelope of 150kW, this curve ensures that the polarization potential of the fourth module is always clamped above the lithium plating boundary during the entire charging process through micro-scale pulse depolarization and current decay, achieving the ultimate balance between safety and efficiency under low-temperature decay conditions.

[0075] refer to Figure 5 In step S14, the specific steps are as follows: S141: Package the target charging node and charging strategy as a whole and send them to the corresponding physical entity for execution. Monitor the charging process of the autonomous vehicle in real time, collect the terminal voltage and inflow current, and use the terminal voltage and inflow current as the physical quantities of the charging interaction. S142: Simultaneously acquire the predicted quantities derived by the twin simulation, and determine the corresponding residuals by comparing them with the physical quantities of the charging interaction; if the residuals exceed the preset evolution tolerance range, then use unscented Kalman filtering to correct the battery state tensor in the twin simulation online.

[0076] In the embodiments of this application, the target charging node and charging strategy are packaged as a whole and sent to the corresponding physical entity for execution. The charging process of the autonomous vehicle is monitored in real time, the terminal voltage and the inflow current are collected, and the terminal voltage and the inflow current are used as the physical quantities of charging interaction.

[0077] At this point, the determined target charging node coordinates, target charging strategy, and non-constant power charging curve for preventing lithium plating are structured and packaged to generate a composite charging control message containing physical addressing information, power timing instructions, and safety constraint boundaries. Then, the composite message is encapsulated into application layer data according to the vehicle-to-pile communication protocol, and the voltage / current closed-loop instruction set pointing to the vehicle-side battery management system (BMS) and the power output response curve pointing to the pile-side charging control system (CCS) are extracted respectively, realizing cross-domain collaborative mapping and encapsulation of a single strategy to the vehicle-to-pile dual-end execution logic.

[0078] The encapsulated composite charging control message is sent to the physical entity via a low-latency vehicle-road-cloud cooperative network, triggering instruction parsing and security verification between the vehicle-side BMS and the charging pile-side CCS. After the verification is successful, the physical entity performs a charging handshake interaction. The charging pile-side connects the contactor of the corresponding power module according to the target charging node information in the message. The vehicle-side BMS closes the charging relay according to the timing zero point of the non-constant power charging curve to establish a physical conductive circuit, thereby smoothly mapping the digital strategy derived from the twin space into a physical layer power transmission channel.

[0079] After the physical charging circuit is turned on, a real-time online monitoring mechanism for the charging process of autonomous vehicles is established. Through a high-precision sensor array on the vehicle and a power metering module on the charging pile, the terminal voltage and current of the power battery pack are collected synchronously at a set high-frequency sampling rate. At this time, the terminal voltage collection covers the total voltage of the battery pack and the voltage of key module cells, and the current collection covers the absolute value of the actual charging current under the current time sequence. After removing sensor measurement noise, the terminal voltage and current are aligned on the time axis according to a unified global clock stamp, which serves as the charging interaction physical quantity that reflects the real electrochemical reaction state.

[0080] The high-frequency collected charging interaction physical quantities are reconstructed in time sequence. Data is framed and features are extracted according to the control cycle of the non-constant power charging curve to filter out glitches and distortions caused by electromagnetic interference or switching transients. The reconstructed terminal voltage and inflow current time sequence data streams are transmitted back to the digital twin space in real time via the twin feedback channel. This data stream is not only used for local closed-loop protection of physical entities, but also serves as an observation benchmark for the real physical world, providing a real-time dynamic raw data foundation for the residual between the subsequent extraction and twin inference prediction quantities.

[0081] Specifically, after the autonomous vehicle enters the target charging node N2 and completes physical parking, the system performs structured packaging and cross-domain collaborative encapsulation of the strategy instructions. The system integrates the device identification code of the target charging node N2, the strategy of initially requesting a power limit of 150kW, and the non-constant power charging curve transitioning from "0.5C-6s power-on + 5s rest" to "0.4C attenuation" to generate a composite charging control message. According to the GB / T 27930 protocol, the system decomposes and encapsulates this message into two parallel instructions. One instruction sets the polarization voltage over-limit protection threshold of 0.03V to the vehicle-side BMS, and the other instruction issues a multi-segment pulse power output timing sequence to the charging pile-side CCS, realizing cross-domain collaborative mapping between the vehicle and the charging pile in the analytical dimension.

[0082] The system sends the encapsulated composite message to the physical entity via the 5G-V2X low-latency network, triggering the security verification and handshake between the vehicle and the charging pile. After parsing the message, the charging pile's CCS confirms that the charging gun No. 2 at node N2 is conducting and closes the power module contactor. The vehicle's BMS confirms that the timing zero point of the non-constant power curve is aligned and then closes the high-voltage relay, establishing a physical conductive circuit. This formally transforms the "0.5C-6s energization + 5s rest" digital strategy derived from twin space into electrical pulses in the physical power transmission channel.

[0083] After the charging circuit is turned on, the system establishes a real-time online monitoring mechanism and performs high-frequency acquisition of interactive physical quantities. The high-precision Hall sensor at the vehicle end and the metering module at the charging pile end simultaneously acquire the terminal voltage and inflow current of the autonomous vehicle battery pack during the pulse power-on period and the rest period at a sampling rate of 100Hz. At the 6th second of power-on, the sensor accurately acquires the inflow current of 0.5C, about 150A, and at the same time acquires the terminal voltage of the 4th module, which climbs to 3.92V under the polarization effect. The system aligns this terminal voltage and inflow current with the global clock stamp as the charging interactive physical quantity that truly reflects the polarization accumulation effect of the autonomous vehicle at a low temperature of 5℃.

[0084] The system performs time-series reconstruction and twin feedback channel construction on the collected interactive physical quantities. The system divides the 100Hz high-frequency current into data frames according to the 11-second (6s energization + 5s rest) control cycle of the non-constant power charging curve, filters out the current spikes at the moment of relay closing, and extracts the terminal voltage and current time-series waveforms that characterize the real electrochemical polarization. The reconstructed waveform stream is transmitted back to the digital twin space in real time through the twin feedback channel. This not only serves as the safety basis for the local BMS of the autonomous vehicle to determine whether energization is allowed in the next cycle, but also provides a raw data foundation that accurately maps the dynamics of the physical world for subsequent residual comparison with the predicted quantities derived from the twin simulation and realizes online correction.

[0085] Furthermore, the predicted quantities derived by the twin simulation are acquired simultaneously, and the corresponding residuals are determined by comparing them with the physical quantities of the charging interaction. If the residuals exceed the preset evolution tolerance range, the battery state tensor in the twin simulation is corrected online using unscented Kalman filtering, which further controls the charging strategy, fully considers the physical quantities of the charging interaction and the predicted quantities corresponding to the twin simulation, improves the accuracy of the residuals, and realizes the online correction of the twin simulation.

[0086] At this point, in the digital twin space, based on the issued non-constant power charging curve and the current battery state tensor, a multi-physics coupling model is run synchronously to deduce the charging process. The terminal voltage prediction sequence and the inflow current prediction sequence, which are isomorphic to the physical end acquisition indicators, are extracted as prediction quantities. At the same time, the charging interaction physical quantities composed of the real-time acquired terminal voltage and inflow current are used as observation quantities. Clock synchronization and dimension alignment based on a unified timestamp are performed on the prediction quantities and observation quantities to eliminate non-mechanistic errors caused by communication delay and sampling phase difference, and to reconstruct a comparison data pair between the virtual deduction and the physical reality at the same time section.

[0087] The reconstructed control data are subjected to time-step difference calculations. The actual terminal voltage of the observed quantity is subtracted from the predicted terminal voltage, and the actual inflow current is subtracted from the predicted inflow current to construct a residual vector characterizing the degree of model mismatch. For this residual vector, an evolution tolerance interval based on electrochemical mechanism and sensor measurement noise covariance calibration is introduced. This evolution tolerance interval dynamically fluctuates with the nonlinear changes of battery state of charge (SOC) and temperature. The norm of the residual vector is compared with the dynamic boundary of the evolution tolerance interval. If the residual norm is within the tolerance interval, it is determined to be normal measurement noise. If the residual norm exceeds the tolerance interval, it is determined that the twin simulation has experienced mechanism drift or parameter distortion, triggering an online correction mechanism.

[0088] Optionally, the evolution tolerance intervals are set separately for two charging interaction physical quantities: the terminal voltage tolerance interval and the inflow current tolerance interval. Both tolerance intervals adopt a symmetrical boundary form, that is, they are jointly formed by the upper and lower boundaries of the tolerance interval. For the terminal voltage tolerance interval, its half-width reference value is set to 0.5% of the rated voltage of the battery cell. For the inflow current tolerance interval, its half-width reference value is set to 2% of the current charging command current value. The above reference values ​​are empirical values ​​obtained by statistically analyzing the residual standard deviations of the twin simulation and the physical entity in multiple repeated charging experiments under the calibration conditions of 50% battery state of charge, 25 degrees Celsius battery temperature, and 0.5 times the charging current rate, and then rounding up to three times the standard deviation.

[0089] When the residual exceeds the evolution tolerance range, the "key decay dimensions, such as polarization internal resistance increment, polarization capacitance, and capacity decay rate" in the current battery state tensor are used as the system state vector. The pulse current of the non-constant power charging curve is used as the known input of the system. The residual between the terminal voltage and the inflow current is used as the observation information. An unscented Kalman filter (UKF) framework is constructed. A set of Sigma points are generated near the current state tensor through UT (unscented transformation). The Sigma points are substituted into the electrochemical mechanism model for nonlinear propagation. The state prediction covariance and the observation prediction covariance are calculated. Then, the Kalman gain matrix is ​​solved. This gain matrix quantifies the reverse correction weight of the residual information on the internal hidden decay state.

[0090] Based on the solved Kalman gain matrix, observation information, and state prediction covariance, the state update equation is executed to calculate the optimal estimated correction of the battery state tensor at the current moment. This correction is then superimposed on the corresponding decay dimension of the original battery state tensor to achieve adaptive online correction of micro-parameters such as polarization resistance and polarization capacitance within the twin simulation. Based on the corrected battery state tensor, the polarization tolerance and lithium plating boundary are resolved, and the amplitude of subsequent pulses and the duration of intermittent depolarization rest periods in the non-constant power charging curve are updated in real time, completing the closed loop of "physical observation - residual identification - state correction - strategy evolution".

[0091] Specifically, during the process of the autonomous vehicle executing the non-constant power charging curve of "0.5C-6s power-on + 5s rest", the system synchronously runs the electrochemical mechanism model in the digital twin space. Based on the tensor deduction of the current battery state, it is deduced that at the end of the 6th second of the 0.5C power-on, the predicted terminal voltage of the 4th module should be 3.92V. At the same time, the system synchronously collects the physical quantities of the charging interaction of the physical entity at a sampling rate of 100Hz, performs clock synchronization and dimension alignment between the predicted quantity and the physical quantity, and obtains the actual observed terminal voltage of the 4th module at the same time section as 3.96V, forming a comparison data pair between the virtual deduction and the physical reality.

[0092] The system performs difference calculations on the control data pair to construct a residual vector with a terminal voltage residual of 0.04V. The system introduces an evolution tolerance range that dynamically fluctuates with SOC and temperature. Under the current conditions of 5℃ and 20% SOC, the upper limit of the evolution tolerance range of the terminal voltage is set to 0.025V. The comparison shows that the residual norm of 0.04V has significantly exceeded the tolerance range of 0.025V. The system determines that the electrochemical parameters inside the twin simulation have drifted, and the original state tensor can no longer accurately map the polarization accumulation effect of the current physical battery, and immediately triggers the online correction mechanism.

[0093] After the correction mechanism is triggered, the system uses the polarization internal resistance and polarization capacitance, which represent the interface dynamics of the fourth module in the battery state tensor, as the state vector. With a 0.5C pulse current as the input and a 0.04V terminal voltage residual as the observation information, an unscented Kalman filter (UKF) framework is constructed. The Sigma point set is generated through unscented transformation. The point set is then substituted into a nonlinear equivalent circuit model containing the SEI film impedance and charge transfer impedance for propagation. It is calculated that the current residual is mainly caused by the sharp decrease in polarization capacitance. Then, the Kalman gain matrix pointing to the polarization capacitance and polarization internal dimension is solved, and the reverse correction weight of the 0.04V residual on the internal micro parameters is quantified.

[0094] The system performs online correction of the battery state tensor based on the Kalman gain matrix, reducing the polarization capacitance parameter of the fourth module by 12% and increasing the polarization internal resistance parameter by 8%, thus achieving adaptive correction of the microscopic electrochemical characteristics inside the twin simulation. Based on the corrected state tensor, the system re-solved and found that the polarization tolerance of the fourth module was further narrowed, and the planned 0.5C charging for 6 seconds in the next cycle would break the lithium plating boundary. The system then updated the non-constant power charging curve in real time, reducing the pulse amplitude of the next cycle from 0.5C to 0.45C and extending the depolarization rest period from 5 seconds to 7 seconds. In this way, through residual-driven closed-loop correction, it ensures that the entire charging process of the autonomous vehicle under low-temperature degradation is always constrained within the safety boundary, realizing the synchronous evolution of the twin and the physical entity.

[0095] Regarding the adjustment of the polarization capacitance parameter of module 4 by 12% and the polarization internal resistance parameter by 8%, the system calculates through unscented Kalman filtering that the polarization capacitance needs to be reduced by 12%. The physical meaning of this correction is that the polarization capacitance parameter value originally stored in the twin simulator was too high, failing to accurately reflect the decreased reactivity of the physical battery due to low-temperature aging and interface degradation. Correcting this parameter downwards by 12% means that the system identifies a 12% reduction in the charge transfer capability at the electrode interface compared to the original model estimate, thus enabling the twin simulator to more accurately predict the rate of increase of the polarization potential in subsequent charging simulations.

[0096] The system calculates through unscented Kalman filtering that the polarization resistance needs to be increased by 8%. The physical meaning of this correction is that the polarization resistance parameter originally stored in the twin simulator was too low, underestimating the actual resistance of the interfacial electrochemical reaction in the current physical battery under low-temperature aging conditions. Adjusting this parameter upwards by 8% means that the system's ability to identify charge transfer at the electrode interface increases by 8% compared to the original model estimate. This allows the twin simulator to more accurately calculate the polarization potential amplitude generated under the same current excitation in subsequent charging simulations.

[0097] Please see Figure 6 The digital twin-based autonomous vehicle charging system is applied to the aforementioned digital twin-based autonomous vehicle charging method; the digital twin-based autonomous vehicle charging system includes: The digital twin module 21 is used to acquire the operating data of the autonomous vehicle and the status data of the charging module, and to construct a twin simulation in the digital twin space. It further combines the battery degradation characteristics of the autonomous vehicle to generate the corresponding battery state tensor. The node matrix module 22 is used to generate a power consumption envelope in the twin simulation based on the battery state tensor and the dynamic deduction of the planned driving path of the autonomous vehicle; when the power consumption envelope reaches the dynamic charging threshold, the corresponding candidate charging node is marked, and a node matrix is ​​generated by combining the prediction of the concurrent carrying capacity of the charging pile group. The charging strategy module 23 is used to perform multi-objective optimization on the power consumption envelope and node matrix, and combine the optimization mechanism to determine the target charging node and the corresponding charging strategy. Based on the identification of the charging strategy, the corresponding non-constant power charging curve is determined. The online correction module 24 is used to send the target charging node and charging strategy for execution, and during the charging process of the autonomous vehicle, it collects the physical quantities of the charging interaction, marks the residual between the physical quantities of the charging interaction and the inferred prediction quantities corresponding to the twin simulation, so as to trigger the online correction of the twin simulation.

[0098] It should be noted that although multiple modules are mentioned in the detailed description above, this division is not mandatory; in fact, according to the embodiments of this disclosure, the features and functions of two or more modules or described above can be embodied in one module; conversely, the features and functions of one module described above can be further divided into multiple modules to be embodied.

[0099] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein; this application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein; the specification and embodiments are to be considered exemplary only.

[0100] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A charging method for an autonomous vehicle based on digital twins, characterized in that, include: The system acquires the operational data of autonomous vehicles and the status data of charging modules, constructs a twin simulation in the digital twin space, and further generates the corresponding battery state tensor by combining the battery degradation characteristics of autonomous vehicles. In the twin simulation, the power consumption envelope is generated based on the battery state tensor and the dynamic deduction of the planned driving path of the autonomous vehicle; when the power consumption envelope reaches the dynamic charging threshold, the corresponding candidate charging node is marked, and a node matrix is ​​generated by combining the prediction of the concurrent carrying capacity of the charging pile group. Multi-objective optimization is performed on the power consumption envelope and node matrix, and the target charging node and corresponding charging strategy are determined by combining the optimization mechanism. Based on the identification of the charging strategy, the corresponding non-constant power charging curve is determined. The target charging node and charging strategy are sent out for execution. During the charging process of the autonomous vehicle, the physical quantities of the charging interaction are collected, and the residuals between the physical quantities of the charging interaction and the inferred prediction quantities corresponding to the twin simulation are marked to trigger the online correction of the twin simulation.

2. The charging method for an autonomous vehicle based on digital twins according to claim 1, characterized in that, The process of acquiring the operating data of the autonomous vehicle and the status data of the charging module, constructing a twin simulation in the digital twin space, and further generating a corresponding battery state tensor by combining the battery degradation characteristics of the autonomous vehicle includes: Real-time acquisition of multiple operational data of autonomous vehicles in different dimensions is performed and aligned with the status data of the charging module in the air to be synchronously input into the digital twin space. In the digital twin space, multiple operational data of autonomous vehicles and status data of charging modules are coupled in a multi-physics field to further construct a twin simulation body that represents the interaction characteristics of vehicle-network-environment.

3. The charging method for an autonomous vehicle based on digital twins according to claim 2, characterized in that, The process of acquiring the operating data of the autonomous vehicle and the status data of the charging module, constructing a twin simulation in the digital twin space, and further generating a corresponding battery state tensor by combining the battery degradation characteristics of the autonomous vehicle, also includes: Battery parameters, including electrochemical impedance spectroscopy and polarization potential, are extracted based on the detection of the twin simulation. These parameters are further combined with the battery degradation characteristics of autonomous vehicles to construct a battery state tensor that integrates spatiotemporal and degradation dimensions in the twin simulation, so as to quantitatively characterize the nonlinear aging and dynamic charge / discharge boundary of the battery throughout its entire life cycle.

4. The charging method for an autonomous vehicle based on digital twins according to claim 1, characterized in that, In the twin simulation, the power consumption envelope is generated based on the battery state tensor and the dynamic deduction of the planned driving path of the autonomous vehicle. When the power consumption envelope reaches the dynamic charging threshold, the corresponding candidate charging nodes are marked. A node matrix is ​​generated based on the prediction of the charging pile group's concurrent capacity limit, including: The system monitors the twin simulation in real time to obtain the planned driving path of the autonomous vehicle. Based on the identification of the planned driving path of the autonomous vehicle, multiple key driving nodes are determined. The system then combines the battery state tensor and environmental disturbance factors to perform dynamic simulation, thereby generating the corresponding power consumption envelope.

5. The charging method for an autonomous vehicle based on digital twins according to claim 4, characterized in that, In the twin simulation, a power consumption envelope is generated based on the battery state tensor and the dynamic deduction of the planned driving path of the autonomous vehicle; when the power consumption envelope reaches the dynamic charging threshold, the corresponding candidate charging nodes are marked, and a node matrix is ​​generated by combining the prediction of the concurrent carrying capacity limit of the charging pile group, which also includes: The dynamic charging threshold based on the battery state tensor dynamic adjustment is marked, and the confidence lower bound of the power consumption envelope is compared with the dynamic charging threshold. If the confidence lower bound of the power consumption envelope reaches the dynamic charging threshold, the spatial coordinates of the time of reaching the threshold are reverse-mapped to the charging module and marked as a candidate charging node. Furthermore, the upper limit of the charging pile group's capacity and the node transformer load rate are combined for dynamic prediction to generate a node matrix representing the spatiotemporal resource constraints and the grid capacity boundary.

6. The charging method for an autonomous vehicle based on digital twins according to claim 1, characterized in that, The process of performing multi-objective optimization on the power consumption envelope and node matrix, and determining the target charging node and corresponding charging strategy using an optimization mechanism, and determining the corresponding non-constant power charging curve based on the identification of the charging strategy, includes: The charging time window tolerance, characterized by the energy consumption envelope, is marked, and the node queuing loss and grid impact, characterized by the node matrix, are collected. Multi-objective Pareto optimization is then performed in combination with the charging time window tolerance, thereby determining the target charging node and its corresponding target charging strategy by combining the optimization mechanism based on non-dominated sorting.

7. The charging method for an autonomous vehicle based on digital twins according to claim 6, characterized in that, The process of performing multi-objective optimization on the power consumption envelope and node matrix, and determining the target charging node and corresponding charging strategy based on the optimization mechanism, and determining the corresponding non-constant power charging curve based on the identification of the charging strategy, further includes: The corresponding polarization tolerance and lithium plating boundary are determined by tracing the battery state tensor, and matched with the target charging strategy to generate a non-constant power charging curve that prevents lithium plating. The non-constant power charging curve consists of multiple adaptive decay pulses and intermittent depolarization rest periods.

8. The charging method for an autonomous vehicle based on digital twins according to claim 1, characterized in that, The process of sending out the target charging node and charging strategy for execution, and collecting charging interaction physical quantities during the charging process of the autonomous vehicle, marking the residual between the charging interaction physical quantities and the inferred prediction quantities corresponding to the twin simulation, to trigger online correction of the twin simulation, includes: The target charging node and charging strategy are packaged together and sent to the corresponding physical entity for execution. The charging process of the autonomous vehicle is monitored in real time, and the terminal voltage and inflow current are collected and used as the physical quantities of the charging interaction.

9. The charging method for an autonomous vehicle based on digital twins according to claim 8, characterized in that, The process of sending out the target charging node and charging strategy for execution, and collecting charging interaction physical quantities during the charging process of the autonomous vehicle, marking the residual between the charging interaction physical quantities and the inferred prediction quantities corresponding to the twin simulation, so as to trigger the online correction of the twin simulation, also includes: The predicted quantities derived from the twin simulation are acquired synchronously, and the corresponding residuals are determined by comparing the physical quantities of the charging interaction. If the residuals exceed the preset evolution tolerance range, the battery state tensor in the twin simulation is corrected online using unscented Kalman filtering.

10. A charging system for an autonomous vehicle based on digital twins, characterized in that, The charging system for the digital twin-based autonomous vehicle is applied to the charging method for the digital twin-based autonomous vehicle as described in any one of claims 1-9; the charging system for the digital twin-based autonomous vehicle includes: The digital twin module is used to acquire the operating data of the autonomous vehicle and the status data of the charging module, and to build a twin simulation in the digital twin space. It further combines the battery degradation characteristics of the autonomous vehicle to generate the corresponding battery state tensor. The node matrix module is used to generate a power consumption envelope in the twin simulation based on the battery state tensor and the dynamic deduction of the planned driving path of the autonomous vehicle. When the power consumption envelope reaches the dynamic charging threshold, the corresponding candidate charging node is marked, and a node matrix is ​​generated by combining the prediction of the concurrent carrying capacity limit of the charging pile group. The charging strategy module is used to perform multi-objective optimization on the power consumption envelope and node matrix, and combine the optimization mechanism to determine the target charging node and the corresponding charging strategy. Based on the identification of the charging strategy, the corresponding non-constant power charging curve is determined. The online correction module is used to send out the target charging node and charging strategy for execution, and during the charging process of the autonomous vehicle, it collects the physical quantities of the charging interaction, marks the residual between the physical quantities of the charging interaction and the inferred prediction quantities corresponding to the twin simulation, so as to trigger the online correction of the twin simulation.

Images