A method and system for optimizing a dynamic power compensation path of an electric vehicle based on a nonlinear charging function

Abstract

The application provides a kind of electric vehicle dynamic power supplement path optimization method and system based on nonlinear charging function, it is related to electric vehicle power supplement technical field, the application is generated by collecting target area geomagnetic fingerprint map data and preprocessing, and the geomagnetic reference fingerprint library containing geographical position and geomagnetic field intensity vector mapping;Receive crowdsourcing vehicle sensor original reading, and obtain standardized geomagnetic vector data after correction processing;With reference library as the initial parameter of federated learning, the parameter is uploaded after local training of crowdsourcing vehicle, and the global geomagnetic fingerprint library is obtained by cloud aggregation optimization;The target vehicle geomagnetic data is matched and positioned with global library multidimension;Again based on battery nonlinear charging function and accurate position, fuse power, temperature and space topology information, plan energy optimal dynamic power supplement path, can be positioned by geomagnetism and optimized fingerprint library by federated learning, combine battery nonlinear charging function and multiple factors, realize the energy optimal dynamic power supplement path planning of electric vehicle.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of electric vehicle power supplementing, and in particular to an electric vehicle dynamic power supplementing path optimization method and system based on a nonlinear charging function. BACKGROUND

[0002] In a GPS signal-free environment such as a tunnel or an underground parking lot, an electric vehicle faces the problems of lack of accurate positioning and difficulty in dynamic power supplementing guidance, forming a special technical requirement. In such a scenario, the vehicle needs to obtain centimeter-level position coordinates to avoid obstacles and plan a path, and needs to dynamically match the optimal power supplementing node in combination with the battery state; and the power supplementing path planning needs to take into account the spatial topological complexity and charging energy loss to ensure efficient and safe power supplementing guidance in a signal blind area, meeting the user's demand for power supplementing convenience and reliable range.

[0003] The mainstream scheme for this technical requirement at present is a fusion positioning power supplementing guidance scheme based on inertial navigation and Bluetooth beacons. The scheme collects vehicle motion parameters by a vehicle-mounted inertial measurement unit (IMU) to perform positioning calculation, and deploys a Bluetooth beacon node to construct a local positioning network to periodically correct the inertial navigation result; the power supplementing path planning calculates the shortest distance path based on the preset power supplementing station position and the vehicle remaining power, and guides the vehicle to the power supplementing.

[0004] The existing scheme has the following significant defects: first, the positioning accuracy is easily affected by the accumulated error, the drift error of inertial navigation increases with time, and the Bluetooth beacon signal is easily disturbed by environmental shielding, so the correction frequency and accuracy are difficult to meet the demand of complex scenarios; second, the power supplementing path planning does not combine the nonlinear charging characteristics of the battery, only takes distance as the optimization target, ignores the energy loss difference in different charging stages, and easily leads to low actual power supplementing efficiency; third, it relies on fixed beacon deployment, has poor adaptability to dynamically changing scene topologies, and cannot realize dynamic adjustment of the path. SUMMARY

[0005] The application aims to provide an electric vehicle dynamic power supplementing path optimization method and system based on a nonlinear charging function to solve the problems of insufficient positioning accuracy, poor adaptability to scenarios and failure to combine the nonlinear characteristics of the battery in the existing scheme in a GPS signal-free environment.

[0006] To solve the above technical problems, in a first aspect, the application provides an electric vehicle dynamic power supplementing path optimization method based on a nonlinear charging function, comprising:

[0007] Collecting geomagnetic fingerprint map data of a target area, preprocessing the geomagnetic fingerprint map data, and generating a geomagnetic reference fingerprint library containing a geographic position and geomagnetic field intensity vector mapping relationship;

[0008] Receiving sensor raw readings uploaded by crowd-sourced vehicles, performing signal correction processing on the sensor raw readings, generating standardized geomagnetic vector data through temperature drift compensation and hard iron interference correction;

[0009] Taking the geomagnetic reference fingerprint library as initial model parameters of federated learning, each crowd-sourced vehicle client performs local model training based on the standardized geomagnetic vector data and uploads model parameters, and the cloud server aggregates and optimizes distributed model parameters to obtain an incrementally updated global geomagnetic fingerprint library;

[0010] Performing multi-dimensional pattern matching between geomagnetic vector data collected by a target vehicle and the global geomagnetic fingerprint library, and generating accurate position coordinates of the target vehicle through a sliding window matching mechanism and a probability weighted positioning calculation;

[0011] Performing dynamic charging path planning based on a nonlinear charging function of a battery and the accurate position coordinates of the target vehicle, and generating an energy-optimal dynamic charging path by fusing battery state of charge, environmental temperature, and spatial topology information in the global geomagnetic fingerprint library.

[0012] Optionally, performing dynamic charging path planning based on a nonlinear charging function of a battery and the accurate position coordinates of the target vehicle, and generating an energy-optimal dynamic charging path by fusing battery state of charge, environmental temperature, and spatial topology information in the global geomagnetic fingerprint library, comprising:

[0013] Obtaining a current battery state of charge and a current environmental temperature of the target vehicle;

[0014] Calculating expected charging efficiency of the target vehicle at different charging sites according to a nonlinear charging function of a battery, the current battery state of charge, and the current environmental temperature;

[0015] Extracting spatial topology information containing road network information and charging site location information from the global geomagnetic fingerprint library, and constructing a path planning graph based on the spatial topology information, wherein nodes in the path planning graph include road intersections and charging sites, and edges represent road segments;

[0016] Determining a vehicle travel energy consumption model, and calculating energy consumption weights of each edge in the path planning graph according to the vehicle travel energy consumption model;

[0017] Taking the accurate position coordinates of the target vehicle as a starting point, combining the expected charging efficiency and the energy consumption weights, searching for a travel path with minimum energy consumption in the path planning graph, and generating an energy-optimal dynamic charging path.

[0018] Optionally, the determining the vehicle driving energy consumption model, and calculating the energy consumption weight of each edge in the path planning graph according to the vehicle driving energy consumption model, comprises:

[0019] establishing an energy consumption calculation function based on vehicle speed, road slope and vehicle mass;

[0020] obtaining road attribute data of each edge in the path planning graph, the road attribute data comprising road length, road slope and average driving speed of the vehicle;

[0021] calculating the energy consumption value required by the vehicle to pass through each road section according to the energy consumption calculation function and the road attribute data;

[0022] assigning the calculated energy consumption value to the corresponding edge in the path planning graph as the energy consumption weight.

[0023] Optionally, the geomagnetic reference fingerprint library is used as the initial model parameter of federated learning, each crowdsourcing vehicle client performs local model training based on the standardized geomagnetic vector data and uploads the model parameter, the distributed model parameters are aggregated and optimized through the cloud server to obtain an incrementally updated global geomagnetic fingerprint library, which comprises:

[0024] convert the geomagnetic reference fingerprint library into the initial model parameter of federated learning, and distribute the initial model parameter to the crowdsourcing vehicle clients;

[0025] each crowdsourcing vehicle client receives the initial model parameter, performs local training on the initial model parameter based on the locally stored standardized geomagnetic vector data, obtains the updated local model parameter, and uploads the updated local model parameter to the cloud server;

[0026] receive the local model parameters uploaded by each client through the cloud server, aggregate the distributed model parameters using a federated averaging algorithm, and generate aggregated global model parameters;

[0027] optimize the aggregated global model parameters, and convert the optimized global model parameters into an incrementally updated global geomagnetic fingerprint library.

[0028] Optionally, the receiving the local model parameters uploaded by each client through the cloud server, aggregating the distributed model parameters using a federated averaging algorithm, and generating aggregated global model parameters, comprises:

[0029] receive the local model parameters uploaded by each crowdsourcing vehicle client through the cloud server, record the training data amount corresponding to each client, and calculate the corresponding aggregation weight according to the training data amount, the client with a larger training data amount being given a higher aggregation weight;

[0030] fuse all local model parameters based on the aggregated weights to obtain initial aggregated parameters, and calculate a difference degree between the initial aggregated parameters and each local model parameter;

[0031] determine whether the difference degree is less than a preset difference threshold, and if the difference degree is less than the difference threshold, take the initial aggregated parameters as the aggregated global model parameters;

[0032] if the difference degree is greater than or equal to the difference threshold, adjust the aggregated weights of each client according to the difference degree, and perform weighted averaging and difference degree determination again based on the adjusted weights until the difference degree is less than the difference threshold, and take the aggregated parameters obtained at this time as the aggregated global model parameters.

[0033] Optionally, the multi-dimensional pattern matching of the geomagnetic vector data collected by the target vehicle with the global geomagnetic fingerprint library is performed, and the accurate position coordinates of the target vehicle are generated through a sliding window matching mechanism and a probability weighted positioning calculation, including:

[0034] constructing a geomagnetic vector sequence according to the collection time sequence of the geomagnetic vector data collected by the target vehicle;

[0035] dividing the geomagnetic vector sequence into a plurality of sliding window data segments according to a preset window length, calculating a multi-dimensional similarity between each sliding window data segment and the geomagnetic vector data of each geographical location point in the global geomagnetic fingerprint library, and the multi-dimensional similarity includes a Euclidean distance similarity and a multi-angle direction comparison similarity;

[0036] Through the sliding window matching mechanism, a plurality of candidate geographical location points with the highest similarity are selected for each sliding window data segment;

[0037] Based on the multi-dimensional similarity values of the candidate geographical location points, the probability weight of each candidate geographical location point is calculated;

[0038] According to the probability weights and coordinate values of all candidate geographical location points, a probability weighted average calculation is performed to generate the accurate position coordinates of the target vehicle.

[0039] Optionally, the sensor raw readings uploaded by the crowd-sourced vehicles are received, and the sensor raw readings are subjected to signal correction processing to generate standardized geomagnetic vector data through temperature drift compensation and hard iron interference correction, including:

[0040] receiving sensor raw readings uploaded by crowd-sourced vehicles, the sensor raw readings including geomagnetic intensity readings and corresponding ambient temperature readings;

[0041] The sensor original readings are subjected to signal quality evaluation, and valid readings meeting preset quality threshold are screened out;

[0042] The valid readings are subjected to temperature drift compensation processing, the environment temperature readings of the valid readings are used to query the pre-stored temperature-magnetic relationship curve, the compensation value corresponding to the magnetic intensity reading is calculated, and the compensation value is applied to the magnetic intensity reading;

[0043] The magnetic intensity reading after temperature compensation is subjected to hard-iron interference correction processing, the pre-stored vehicle hard-iron interference parameters are read, the hard-iron interference vector is calculated according to the vehicle hard-iron interference, and the hard-iron interference vector is subtracted from the magnetic intensity reading;

[0044] The corrected magnetic intensity reading is subjected to dimension unification processing, and is converted to a magnetic vector representation in a standard physical dimension;

[0045] The processed magnetic vector data is organized in time sequence, and the standardized magnetic vector data is generated.

[0046] In a second aspect, the application provides a dynamic power supplement path optimization system for electric vehicles based on a nonlinear charging function, comprising:

[0047] A collection module is configured to collect geomagnetic fingerprint map data of a target area, pre-process the geomagnetic fingerprint map data, and generate a geomagnetic reference fingerprint library containing a mapping relationship between geographic location and geomagnetic field intensity vector;

[0048] A processing module is configured to receive sensor original readings uploaded by crowd-sourcing vehicles, perform signal correction processing on the sensor original readings, and generate standardized magnetic vector data through temperature drift compensation and hard-iron interference correction;

[0049] An optimization module is configured to use the geomagnetic reference fingerprint library as initial model parameters for federated learning, perform local model training based on the standardized magnetic vector data by each crowd-sourcing vehicle client and upload model parameters, aggregate and optimize distributed model parameters through a cloud server to obtain an incrementally updated global geomagnetic fingerprint library;

[0050] A matching module is configured to perform multi-dimensional pattern matching between geomagnetic vector data collected by a target vehicle and the global geomagnetic fingerprint library, generate accurate position coordinates of the target vehicle through a sliding window matching mechanism and a probability weighted positioning calculation;

[0051] A generation module is configured to perform dynamic power supplement path planning based on a nonlinear charging function of a battery and accurate position coordinates of the target vehicle, and generate an energy-optimal dynamic power supplement path by fusing battery state of charge, environment temperature, and spatial topology information in the global geomagnetic fingerprint library.

[0052] In a third aspect, the present application provides an electronic device, comprising:

[0053] a memory for storing a computer program;

[0054] a processor for executing the computer program to implement the steps of the method for optimizing a dynamic charging path of an electric vehicle based on a nonlinear charging function according to the first aspect.

[0055] In a fourth aspect, the present application provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program, when executed by a processor, can implement the steps of the method for optimizing a dynamic charging path of an electric vehicle based on a nonlinear charging function according to the first aspect.

[0056] The method for optimizing a dynamic charging path of an electric vehicle based on a nonlinear charging function provided by the present application collects target area geomagnetic fingerprint map data and pre-processes to generate a geomagnetic benchmark fingerprint library, combines standardized geomagnetic vector data corrected from raw sensor readings of crowd-sourced vehicles, and realizes incremental updating of a global geomagnetic fingerprint library through federated learning with the benchmark fingerprint library as initial parameters. Then, the method generates accurate position coordinates of a target vehicle through multi-dimensional pattern matching and probability weighted calculation. Finally, the method plans a dynamic charging path based on the battery nonlinear charging function and the accurate position, and fuses battery power, environmental temperature, and spatial topology information, effectively solving the problem of insufficient positioning accuracy in a GPS signal-free environment, and realizing energy-optimal charging guidance that takes into account both battery charging characteristics and scene topology, thereby improving the convenience and energy efficiency of electric vehicle charging in a complex signal blind area.

[0057] Further, the current battery power state and environmental temperature of a target vehicle are obtained, the expected charging efficiency of different charging stations is calculated according to the battery nonlinear charging function, the spatial topology information of road networks and charging station positions is extracted from the global geomagnetic fingerprint library to construct a path planning graph containing nodes and edges, a vehicle travel energy consumption model is determined and the energy consumption weight of each edge is calculated, and finally the accurate position of the target vehicle is taken as the starting point, the expected charging efficiency and the energy consumption weight are combined, and the travel path with the minimum energy consumption is searched in the path planning graph to generate an energy-optimal dynamic charging path. This method accurately calculates the expected charging efficiency of different charging stations and the energy consumption weight of each path segment, constructs a planning graph combined with spatial topology information, and searches for an optimal path, further refining the planning logic of the energy-optimal charging path, so that the charging path not only takes into account the nonlinear charging characteristics of the battery, but also accurately matches the road energy consumption, improving the scientific nature and energy utilization efficiency of electric vehicle charging path planning in a GPS signal-free environment. BRIEF DESCRIPTION OF DRAWINGS

[0058] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced. Obviously, the accompanying drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0059] Figure 1 A flowchart of a dynamic power compensation path optimization method for electric vehicles based on a nonlinear charging function provided by an embodiment of the present application;

[0060] Figure 2 A specific implementation flowchart of a dynamic power compensation path optimization method for electric vehicles based on a nonlinear charging function provided by an embodiment of the present application;

[0061] Figure 3 A specific implementation scenario diagram of a dynamic power compensation path optimization method for electric vehicles based on a nonlinear charging function provided by an embodiment of the present application;

[0062] Figure 4 A structural diagram of a dynamic power compensation path optimization system for electric vehicles based on a nonlinear charging function provided by an embodiment of the present application. DETAILED DESCRIPTION

[0063] In the positioning and power compensation guidance of electric vehicles in GPS signal-free environments such as tunnels and underground parking lots, the existing fusion scheme based on inertial navigation and Bluetooth beacons has obvious defects: the positioning accuracy is easily affected by the cumulative drift error of inertial navigation and the shielding interference of Bluetooth beacons, and it is difficult to meet the needs of complex scenarios; the power compensation path planning only takes distance as the optimization target, without considering the nonlinear charging characteristics of the battery, resulting in low power compensation efficiency; and it relies on fixed beacon deployment, which has poor adaptability to dynamic scene topology and cannot realize path adjustment. These problems jointly restrict the convenience and energy efficiency of power compensation for electric vehicles in GPS signal-free environments.

[0064] To solve the above problems, the application provides a dynamic power supplement path optimization method for electric vehicles based on a nonlinear charging function. The core of the method is to realize accurate positioning by combining geomagnetic fingerprint positioning with federated learning, and to plan an energy optimal path by fusing battery characteristics. Specifically, the method first collects geomagnetic data to generate a benchmark fingerprint library, updates the global fingerprint library through federated learning by combining the corrected geomagnetic data of crowd-sourced vehicles, generates an accurate vehicle position through multi-dimensional matching, and finally plans a path based on the nonlinear charging function of the battery, fusing the power, environmental temperature and spatial topology information. The method replaces inertial navigation and Bluetooth beacons with geomagnetic fingerprint positioning to avoid cumulative errors and signal interference. The nonlinear characteristics of the battery are incorporated into the path planning to break through the single optimization target of distance. Incremental updating of the global fingerprint library improves scene adaptability and fundamentally solves the positioning, planning and adaptability problems of existing technologies, improving the convenience and energy efficiency of power supplement in complex signal blind areas.

[0065] To make the person skilled in the art better understand the application scheme, the application will be further described in detail below in combination with the drawings and specific embodiments. Obviously, the described embodiments are only part of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the application.

[0066] The core of the application is to provide a dynamic power supplement path optimization method for electric vehicles based on a nonlinear charging function. The flowchart of a specific embodiment of the method is shown in Figure 1 The method comprises:

[0067] S101, collect geomagnetic fingerprint map data of a target area, preprocess the geomagnetic fingerprint map data, and generate a geomagnetic benchmark fingerprint library containing a mapping relationship between geographic positions and geomagnetic field intensity vectors;

[0068] In the above scheme, the geomagnetic fingerprint map data refers to the data collected in a specific area where electric vehicle positioning and power supplement guidance are needed, reflecting the distribution characteristics of the geomagnetic field in the area, including the intensity, direction and other vector information of the geomagnetic field at different geographic positions. The geomagnetic benchmark fingerprint library is a database storing one-to-one correspondence between each geographic position in the target area and the corresponding geomagnetic field intensity vector, providing a basic reference for subsequent vehicle positioning.

[0069] In the embodiments of the present application, first, the target area is divided into a plurality of geographical position grids by predefined rules, for example, the target area is uniformly divided into a plurality of square grids with a fixed side length of 2 meters x 2 meters, and each grid has a unique identification number. Then, a collection vehicle equipped with a geomagnetic sensor is used to drive along a predetermined path in the target area, and the geographical position coordinates and geomagnetic intensity readings of the collection vehicle are recorded synchronously, for example, the predetermined path is designed as a "Z" shape, which ensures that the driving track can cover all the divided grids, and the collection vehicle records the current geographical position coordinates and the geomagnetic intensity readings detected by the geomagnetic sensor every 0.5 meters of driving.

[0070] Secondly, the recorded geographical position coordinates are converted into a unified coordinate system, and the coordinate data obtained at different collection times are converted into the same plane rectangular coordinate system; at the same time, consistency detection is performed on the magnetic intensity readings, for example, the normal reading range is set as , and the readings exceeding the range are marked as abnormal and removed. Then, the detected geomagnetic intensity readings are grouped according to the geographical position grids, and the average value of all readings in each grid is calculated as the reference geomagnetic vector of the corresponding grid.

[0071] Finally, the center coordinates of each grid and the corresponding reference geomagnetic vector are established in a mapping relationship and stored as a geomagnetic reference fingerprint library, for example, the center coordinates (X1, Y1) of grid A and the reference geomagnetic vector correspond to each other, and are recorded in a table form and saved in a database.

[0072] In actual application, in the target area of A underground parking lot, a portable geomagnetic collection device is used by the staff to arrange collection points in the parking lot at intervals of 3 meters x 3 meters, 8 times of geomagnetic field intensity vector data are collected at each collection point, and the position coordinates of the point are recorded; then, the original collected data are preprocessed, the average value of 8 times of data of each collection point is calculated by using the mean filtering method to eliminate accidental interference, the abnormal values far exceeding the data of the surrounding collection points are deleted, and finally the position coordinates of each collection point are corresponded to the processed average geomagnetic field intensity vector, and the geomagnetic reference fingerprint library of A underground parking lot is generated.

[0073] The overall scheme of S101 above, by dividing the geographical position grids according to the rules, makes the geomagnetic data collection more systematic and orderly; the collection along the predetermined path ensures the completeness of data coverage; the coordinate unification and removal of abnormal readings improve the accuracy and consistency of the data; the calculation of the reference geomagnetic vector according to the grid and the establishment of the mapping relationship with the center coordinates generate the geomagnetic reference fingerprint library, which can provide structured and reliable reference for subsequent vehicle positioning, and help to improve the accuracy and efficiency of positioning in the environment without GPS signal.

[0074] S102, receiving sensor raw readings uploaded by crowd-sourced vehicles, performing signal correction processing on the sensor raw readings, generating standardized geomagnetic vector data through temperature drift compensation and hard-iron interference correction;

[0075] Optionally, step S102 can specifically include the following steps:

[0076] S10221, receiving sensor raw readings uploaded by crowd-sourced vehicles, the sensor raw readings including geomagnetic intensity readings and corresponding ambient temperature readings;

[0077] S1022, performing signal quality evaluation on the sensor raw readings, and screening out valid readings meeting a preset quality threshold;

[0078] S1023, performing temperature drift compensation processing on the valid readings, querying a pre-stored temperature-geomagnetic relationship curve according to the ambient temperature readings of the valid readings, calculating compensation values of the corresponding geomagnetic intensity readings, and applying the compensation values to the geomagnetic intensity readings;

[0079] S1024, performing hard-iron interference correction processing on the geomagnetic intensity readings after temperature compensation, reading pre-stored vehicle hard-iron interference parameters, calculating a hard-iron interference vector according to the vehicle hard-iron interference, and subtracting the hard-iron interference vector from the geomagnetic intensity readings;

[0080] S1025, performing dimension unification processing on the corrected geomagnetic intensity readings, and converting them to geomagnetic vector representation under a standard physical dimension;

[0081] S1026, organizing the processed geomagnetic vector data in time sequence, and generating standardized geomagnetic vector data.

[0082] In the above scheme, the crowd-sourced vehicles refer to ordinary electric vehicles participating in geomagnetic data collection tasks, including equipped geomagnetic sensors and positioning devices, including position sensing and data collection functions during vehicle driving, and can be used to provide geomagnetic related raw data. The sensor raw readings refer to the untreated sensing data uploaded by the crowd-sourced vehicles, including geomagnetic intensity readings and corresponding ambient temperature readings, including geomagnetic field size information and environmental cold and hot state during collection, and can be used as basic data for subsequent signal correction processing. The standardized geomagnetic vector data refer to the geomagnetic data after signal correction processing, including temperature drift compensation, hard-iron interference correction and dimension unification processing, including standardized data sequence organized in time sequence, and can be used for local model training in federated learning.

[0083] In the embodiments of the present application, first, the sensor original readings uploaded by the crowd-sourcing vehicle are received through step S1021. The crowd-sourcing vehicle collects and uploads the geomagnetic intensity readings detected by the geomagnetic sensor and the ambient temperature readings at the time of collection in a preset period during driving, and meanwhile, the collection time stamp information is reserved for subsequent time sorting.

[0084] Then, the signal quality evaluation is performed through step S1022. The uploaded sensor original readings are judged for reliability by using a signal-to-noise ratio analysis algorithm. A signal-to-noise ratio greater than 30 is pre-set as a quality threshold. The ratio of the signal strength to the noise strength of each original reading is calculated. The readings with a ratio lower than 30 are determined as invalid data and directly eliminated. Only the valid readings with a signal-to-noise ratio meeting the requirements are reserved. This step can preliminarily filter out low-quality data generated by device interference, sudden jitter, etc., and provide a reliable basis for subsequent correction processing.

[0085] Then, the temperature drift compensation processing is performed on the screened valid readings through step S1023. Based on the temperature-geomagnetic relationship curve determined by a large number of experiments in advance, the curve records the drift law of the geomagnetic readings at different ambient temperatures. For each valid reading, its ambient temperature reading is extracted and matched and queried on the curve. For example, when the ambient temperature is 28℃, the drift compensation value of the corresponding geomagnetic intensity is found to be 0.0001 from the curve. Then, this compensation value is superimposed on the geomagnetic intensity data of the valid reading, and the correction of the geomagnetic reading deviation caused by temperature change is completed.

[0086] After that, the hard-iron interference correction is performed through step S1024. Before participating in data collection, each crowd-sourcing vehicle will measure its hard-iron interference parameters, i.e., the fixed magnetic field interference data generated by the metal parts of the vehicle, through calibration experiments in advance. After the system reads the pre-stored parameters corresponding to the vehicle, the hard-iron interference vector is calculated based on the vector synthesis principle. Then, the geomagnetic intensity reading after the temperature compensation in step S1023 is taken as the basis, and the hard-iron interference vector is subtracted therefrom, so as to eliminate the fixed interference of the metal of the vehicle on the geomagnetic signal and further improve the accuracy of the geomagnetic data.

[0087] After the interference correction is completed, the dimensional unification processing is performed through step S1025. Since the geomagnetic sensors equipped in crowd-sourcing vehicles of different brands or models may use different measurement units, in order to ensure the consistency of the data, all the corrected geomagnetic intensity readings are uniformly converted to the standard physical dimension. The conversion is completed through a unit conversion formula, so that the subsequent data processing can be based on the unified measurement standard.

[0088] ​Finally, the step S1026 is performed to arrange all the geomagnetic vector data after temperature compensation, hard iron correction and dimension unification in the order from early to late according to the collection time stamp recorded in the step S1021 to form standardized geomagnetic vector data, which completely retains the time variation characteristics of the geomagnetic signal in the driving process of the crowdsourcing vehicle and can be directly used for the local model training in the subsequent federated learning.

[0089] In actual application, the B brand electric vehicles participating in crowdsourcing around the target area upload sensor original readings including geomagnetic intensity readings and environmental temperature readings every 20 seconds when driving; the system first performs signal quality evaluation on these original readings to eliminate invalid readings with a signal-to-noise ratio less than 28; for the remaining valid readings, a compensation value is calculated and applied according to the environmental temperature reading in the temperature geomagnetic relationship curve, for example, the compensation value is when the environmental temperature is 30℃; then the hard iron interference parameter of the B brand electric vehicle is read to calculate the hard iron interference vector as ; the vector is subtracted from the geomagnetic intensity reading after temperature compensation; then the corrected geomagnetic reading is converted into nano Tesla unit; finally, the processed geomagnetic vector data is arranged in the order of collection time to generate standardized geomagnetic vector data.

[0090] The overall scheme of the above S102 receives the sensor original readings uploaded by the crowdsourcing vehicle and sequentially performs signal quality evaluation, temperature drift compensation, hard iron interference correction, dimension unification and time sequence organization, effectively removes the interference and error in the original data, improves the accuracy and consistency of the geomagnetic data, and generates a standardized geomagnetic vector data sequence to provide a high-quality data basis for subsequent local model training, thereby guaranteeing the reliability of the subsequent model training result.

[0091] S103, taking the geomagnetic reference fingerprint library as the initial model parameter of federated learning, each crowdsourcing vehicle client performs local model training based on the standardized geomagnetic vector data and uploads the model parameter, and the cloud server aggregates and optimizes the distributed model parameters to obtain an incrementally updated global geomagnetic fingerprint library;

[0092] Optionally, the step S103 can specifically include the following steps:

[0093] S1031, converting the geomagnetic reference fingerprint library into the initial model parameter of federated learning and distributing the initial model parameter to the crowdsourcing vehicle client;

[0094] ​S1032, each crowdsourcing vehicle client receives the initial model parameters, locally trains the initial model parameters based on locally stored standardized geomagnetic vector data, obtains updated local model parameters, and uploads the updated local model parameters to the cloud server;

[0095] S1033, receiving the local model parameters uploaded by each client through the cloud server, aggregating the distributed model parameters by using a federated average algorithm, and generating aggregated global model parameters;

[0096] S1034, optimizing the aggregated global model parameters, and converting the optimized global model parameters into an incrementally updated global geomagnetic fingerprint library.

[0097] S1033, receiving the local model parameters uploaded by each client through the cloud server, recording the training data amount corresponding to each client, and calculating the corresponding aggregation weight according to the training data amount. The client with larger training data amount is given higher aggregation weight. Based on the aggregation weight, the weighted average method is used to fuse all local model parameters to obtain initial aggregation parameters, and the difference degree between the initial aggregation parameters and each local model parameter is calculated. It is judged whether the difference degree is less than a preset difference threshold. If it is less than the difference threshold, the initial aggregation parameters are taken as the aggregated global model parameters. If it is greater than or equal to the difference threshold, the aggregation weight of each client is adjusted according to the difference degree, and the weighted average and difference degree judgment are performed again based on the adjusted weight, until the difference degree is less than the difference threshold. The aggregation parameters obtained at this time are taken as the aggregated global model parameters.

[0098] In the above scheme, the initial model parameter refers to the basic parameter converted from the geomagnetic reference fingerprint library according to the requirement of the federated learning framework, contains the mapping relationship data of geographic position and geomagnetic vector, and is used for initial model training of the crowd-sourcing vehicle client. The crowd-sourcing vehicle client refers to a computing terminal participating in federated learning carried on the crowd-sourcing vehicle, and has data storage and model training functions. The local model parameter is an updated parameter obtained by training the crowd-sourcing vehicle client based on the local standardized geomagnetic vector data, and reflects the characteristics of the local geomagnetic data. The federated averaging algorithm is an algorithm for aggregating distributed model parameters, and fuses the parameters of each client by weighted averaging or the like. The aggregation weight is a weight value allocated according to the data size of the crowd-sourcing vehicle client, and the larger the data size, the higher the weight. The difference degree refers to the deviation degree between the initial aggregation parameter and each local model parameter, and the preset difference threshold is a standard value for judging whether the aggregation parameter is qualified. The global model parameter is a unified model parameter aggregated and optimized by the cloud server, and the incremental updated global geomagnetic fingerprint library is a database updated with the mapping relationship of the geomagnetic vector by converting the optimized global model parameter.

[0099] In the embodiment of the present application, first, the geomagnetic reference fingerprint library is converted into the initial model parameter of federated learning through step S1031, specifically, the mapping relationship between the center coordinates of each geographic position grid in the fingerprint library and the reference geomagnetic vector is standardized by converting the coordinates and vector data into a fixed-dimensional vector form according to the input format required by the federated learning model, forming an initial parameter set that can be directly used for model training; then, the initial model parameters are distributed to all crowd-sourcing vehicle clients participating in this federated learning task through a secure network communication channel, ensuring that each client can obtain consistent basic training parameters. For example, if the geomagnetic reference fingerprint library contains mapping data of 500 grids, the (X, Y) center coordinates of each grid and the geomagnetic vector are combined into a 5-dimensional vector, and the 500 vectors are packaged as initial model parameters and sent to the clients of 15 crowd-sourcing vehicles.

[0100] Then, after receiving the initial model parameters, each crowd-sourcing vehicle client loads the initial model parameters into the local model training framework in step S1032, and uses the standardized geomagnetic vector data stored locally as training samples to iteratively train the initial model parameters by using the gradient descent algorithm. The error between the model prediction output and the local actual geomagnetic data is calculated, and the parameters are gradually adjusted to reduce the error. After a preset number of iterations, the updated local model parameters that adapt to the local geomagnetic characteristics are obtained. After the training is completed, the client compresses and packages the updated parameters, and sends them to the cloud server through an encrypted upload channel, while attaching the data size information used for this training.

[0101] ​Next, the cloud server through step S1033 first receives all the local model parameters uploaded by the crowdsourcing vehicle clients and the corresponding training data amount, and then calculates the aggregation weight of each client according to the training data amount. The calculation method is that the data amount of a single client is divided by the total data amount of all clients. The larger the data amount of a client, the higher the weight. Then, the federated average algorithm is used to multiply the local model parameters of each client by the corresponding aggregation weight, and then add all the results to obtain the initial aggregation parameter. Then, the Euclidean distance is calculated to measure the difference between the initial aggregation parameter and each local model parameter, and compared with the preset difference threshold. If all the differences are less than the threshold, the initial aggregation parameter is directly used as the aggregated global model parameter. If there is a difference exceeding the threshold, such as a difference of 0.08 for a client, the aggregation weight of the client is appropriately reduced, the weight of the client with a small difference is increased, and the weighted average calculation is performed again. The difference judgment process is repeated until all the differences meet the threshold requirement. For example, the data amounts of 3 clients are 1000, 800 and 600 respectively, the total data amount is 2400, and the corresponding weights are 、 、 After weighted average to obtain the initial aggregation parameter, if the differences are 0.03, 0.04 and 0.02 respectively, all of which are less than the threshold 0.05, the global model parameter is determined.

[0102] Finally, through step S1034, the aggregated global model parameter is optimized. Usually, L2 regularization algorithm is used to constrain the parameter smoothing, so as to avoid the deviation of the parameter due to excessive fitting of local data. After optimization, the global model parameter is converted according to the format of the geomagnetic benchmark fingerprint library, that is, the vector form parameter is restored to the mapping relationship between the center coordinates of each geographical position grid and the corresponding geomagnetic vector. For the grid mapping relationship with new addition or large parameter adjustment, it is marked as an update item. Finally, the incremental updated global geomagnetic fingerprint library is formed. Compared with the original benchmark library, it contains more new geomagnetic feature data collected by the crowdsourcing vehicles, and the mapping relationship is more suitable for the actual environment. For example, after conversion of the optimized global model parameter, 20 grid mapping data not covered by the original benchmark library are added, and the geomagnetic vector values of 30 original grids are corrected, forming the updated global geomagnetic fingerprint library.

[0103] In practical applications, in the geomagnetic fingerprint library updating task of the underground parking lot in the target area, the geomagnetic reference fingerprint library containing 600 geographical position grid mapping relationships of the garage is first converted into initial model parameters in vector form according to the federated learning model format, and is distributed to 12 C brand electric vehicle clients participating in crowdsourcing; after each client receives the initial parameters, the initial parameters are loaded into the local training framework, and 1800 standardized geomagnetic vector data stored for nearly 3 hours are used to train the local model parameters through gradient descent algorithm for 50 times, and the compressed and encrypted local model parameters are uploaded to the cloud with data volume information; after the cloud receives the local model parameters, the aggregation weights of each client are calculated according to the training data of 1000, 800 and 600 of 4, 5 and 3 vehicles respectively, and the total data volume is 10200, and the corresponding weights are 、 、 The initial aggregation parameters are obtained by weighted averaging using the federated averaging algorithm, and the difference between the initial aggregation parameters and each local parameter is less than the preset threshold 0.06, which is directly used as the global model parameter; then the global parameter is optimized by using the L2 regularization algorithm, and the mapping relationship between the geographical position grid center coordinates and the geomagnetic vector is converted in reverse, 30 new mapping data of garage corner grids are added, and the geomagnetic vector values of 45 original grids are corrected, to form the global geomagnetic fingerprint library of the B underground garage incremental update.

[0104] The above-mentioned overall scheme of S103 realizes the incremental update of the global geomagnetic fingerprint library by using the federated learning method, taking the geomagnetic reference fingerprint library as the initial parameter, combining the local training of the crowdsourcing vehicle client and the parameter aggregation optimization of the cloud, which not only utilizes the richness of the crowdsourcing data to improve the coverage and accuracy of the fingerprint library, but also avoids the security risk of centralized data storage, and at the same time ensures the reliability of the global parameter by dynamically adjusting the aggregation weight, so that the updated global geomagnetic fingerprint library can better adapt to the geomagnetic environment changes in the target area, and provide more accurate reference for subsequent vehicle positioning.

[0105] S104, performing multi-dimensional pattern matching between the geomagnetic vector data collected by the target vehicle and the global geomagnetic fingerprint library, generating the accurate position coordinates of the target vehicle through the sliding window matching mechanism and the probability weighted positioning calculation;

[0106] Optionally, step S104 can specifically include the following steps:

[0107] S1041, constructing a geomagnetic vector sequence according to the collection time sequence of the geomagnetic vector data collected by the target vehicle;

[0108] S1042, divide the geomagnetic vector sequence into a plurality of sliding window data segments according to a preset window length, and calculate a multi-dimensional similarity between each sliding window data segment and the geomagnetic vector data of each geographical location point in the global geomagnetic fingerprint library, the multi-dimensional similarity including a Euclidean distance similarity and a multi-angle direction comparison similarity;

[0109] S1043, through a sliding window matching mechanism, filter out several candidate geographical location points with the highest similarity for each sliding window data segment;

[0110] S1044, based on the multi-dimensional similarity values of the candidate geographical location points, calculate the probability weight of each candidate geographical location point;

[0111] S1045, according to the probability weight and the coordinate value of all candidate geographical location points, perform a probability weighted average calculation to generate the accurate position coordinates of the target vehicle.

[0112] In the above scheme, the sliding window data segment refers to a small piece of data obtained by truncating the target vehicle geomagnetic vector sequence according to a preset window length, containing a fixed number or time span of geomagnetic vector information, including a plurality of continuously collected geomagnetic data points, which can be used for matching and comparison with the global geomagnetic fingerprint library. The multi-dimensional similarity refers to an index for comprehensively measuring the similarity degree of two groups of geomagnetic data, including a Euclidean distance similarity and a multi-angle direction comparison similarity, including the closeness of geomagnetic vector numerical value and the consistency of direction, which is used to select geographical location points similar to the sliding window data segment. The candidate geographical location point refers to the several geographical locations with the highest similarity matched by each sliding window data segment, containing the top similar place information in the global geomagnetic fingerprint library, which can be used as the alternative reference for the target vehicle position. The probability weight refers to a numerical value representing the position credibility allocated according to the multi-dimensional similarity value of the candidate geographical location point, which can be used for weighted calculation of the candidate point coordinates. The accurate position coordinates refer to the actual position of the target vehicle obtained by weighted average of the candidate geographical location point coordinates and the corresponding probability weight, containing the coordinate result after weighted calculation, including the specific plane coordinate value, which can realize the precise positioning of the vehicle in the GPS signal-free environment.

[0113] In the embodiments of the present application, first, the geomagnetic sensor on the target vehicle collects geomagnetic vector data at a preset time interval in step S1041, and synchronously records the time stamp of each collection during the collection process. Then, the geomagnetic vector data containing the geomagnetic field strength and direction information are arranged in order from early to late according to the collection time stamp, forming a continuous geomagnetic vector sequence. For example, if the target vehicle continuously collects 20 groups of geomagnetic vector data during driving, the data collected at the first time is placed at the beginning of the sequence, and the data collected at the 20th time is placed at the end, to construct an ordered geomagnetic vector sequence containing 20 data points.

[0114] Then, step S1042 sets a preset window length, and a constructed geomagnetic vector sequence is intercepted according to the length, starting from the first data point of the sequence, a data segment is intercepted each time by sliding back 1 data point until the entire sequence is traversed, and a plurality of sliding window data segments are obtained; subsequently, for each sliding window data segment, multidimensional similarity calculation is performed with the geomagnetic vector data of all geographic location points in the global geomagnetic fingerprint library, wherein the Euclidean distance similarity is obtained by calculating the square sum of the difference values of each component of the two groups of data and then taking the square root, and the smaller the value, the closer the size; the multi-angle direction comparison similarity is obtained by calculating the included angle between the two groups of vectors, and the smaller the included angle, the more consistent the direction; and the multidimensional similarity is obtained by comprehensively considering the two indexes.

[0115] Then, through step S1043, a sliding window matching mechanism is used to process the matching results of each sliding window data segment, each data segment is sorted from high to low according to the multidimensional similarity of each geographic location point in the global geomagnetic fingerprint library, and according to a preset number of candidates, a plurality of geographic location points at the top of the sorting are selected as candidate geographic location points of the sliding window data segment. For example, after a certain sliding window data segment is matched with 1000 location points in the fingerprint library, the top 3 location points are selected according to the similarity, and the 3 points are the candidate geographic location points of the window.

[0116] Then, step S1044 is performed, and the probability weight of each candidate geographic location point is calculated based on the multidimensional similarity value corresponding to the candidate geographic location point. The multidimensional similarity values of all candidate geographic location points under the same sliding window are added to obtain a total similarity, and then the similarity value of each candidate point is divided by the total similarity to obtain the probability weight of the candidate point. The higher the weight value, the higher the credibility of the candidate point as the actual location of the target vehicle. For example, the similarity values of 3 candidate points of a window are 0.9, 0.7 and 0.4 respectively, the total similarity is 2.0, and the probability weights of the 3 candidate points are 0.45, 0.35 and 0.2 respectively. 、 、 .

[0117] Finally, through step S1045, the X coordinate of each candidate geographic location point is multiplied by its corresponding probability weight, and all the results are added to obtain the X coordinate of the accurate position. Similarly, the Y coordinate of each candidate point is multiplied by its probability weight and summed to obtain the Y coordinate of the accurate position. The combination of the two is the accurate position coordinate of the target vehicle. For example, the coordinates of three candidate points are (10, 20), (12, 22), and (11, 19), and the corresponding weights are 0.45, 0.35, and 0.2. The accurate X coordinate is 10*0.45+12*0.35+11*0.2, and the accurate Y coordinate is 20*0.45+22*0.35+19*0.2. After calculation, the final accurate position coordinate is obtained.

[0118] In actual application, in a certain underground hub parking lot, the geomagnetic sensor of the target vehicle of brand C collects geomagnetic vector data every 5 seconds. After continuously collecting 25 groups, a geomagnetic vector sequence is constructed according to the order of collection time. According to the preset window length of 5 data points, the sequence is slidingly intercepted. Each time the data points are slid by 1, 21 sliding window data segments are obtained. The Euclidean distance similarity and multi-angle direction comparison similarity of each data segment with each geographic location point in the global geomagnetic fingerprint library are calculated. The three candidate geographic location points with the highest comprehensive similarity are selected for each sliding window data segment. For the multi-dimensional similarity values of the candidate points of a certain window, 0.8, 0.7, and 0.6, the corresponding probability weights are calculated as 、 、 . Finally, the coordinates of all candidate geographic location points are weighted and averaged with the corresponding probability weights. The X and Y coordinates of each candidate point are multiplied by the weight and summed to generate the accurate position coordinate of the target vehicle in the B underground hub parking lot.

[0119] The overall scheme of S104 above constructs a sequence of geomagnetic data collected by the target vehicle and divides sliding windows. The candidate points are selected by matching and combining multi-dimensional similarity. The accurate position coordinate is generated by probability weighted average calculation, which avoids the accidental error of single data point matching. The sliding window mechanism improves the continuity and stability of positioning. Multi-dimensional similarity and probability weighting make the positioning result more in line with the actual situation, effectively realizing the accurate positioning of the target vehicle in the GPS signal-free environment and providing a reliable position basis for subsequent dynamic power compensation path planning.

[0120] S105, based on the nonlinear charging function of the battery and the accurate position coordinate of the target vehicle, dynamic power compensation path planning is performed. By fusing the battery power state, the environmental temperature, and the spatial topology information in the global geomagnetic fingerprint library, an energy-optimal dynamic power compensation path is generated.

[0121] Optionally, step S105 can specifically include the following steps:

[0122] S1051, obtaining the current battery state of charge and the current ambient temperature of the target vehicle;

[0123] S1052, calculating the expected charging efficiency of the target vehicle at different charging stations according to the nonlinear charging function of the battery, the current battery state of charge and the current ambient temperature;

[0124] S1053, extracting spatial topology information containing road network information and charging station location information from the global geomagnetic fingerprint library, and constructing a path planning graph based on the spatial topology information, wherein the nodes in the path planning graph include road intersections and charging stations, and the edges represent road segments;

[0125] S1054, determining a vehicle travel energy consumption model, and calculating the energy consumption weight of each edge in the path planning graph according to the vehicle travel energy consumption model;

[0126] S1055, taking the precise position coordinates of the target vehicle as the starting point, combining the expected charging efficiency and the energy consumption weight, searching for a travel path with the minimum energy consumption in the path planning graph, and generating a dynamic power supplement path with the optimal energy consumption.

[0127] In the step S1054, the process specifically includes the following steps: establishing an energy consumption calculation function based on vehicle speed, road slope and vehicle mass; obtaining road attribute data of each edge in the path planning graph, wherein the road attribute data includes road length, road slope and average vehicle travel speed; calculating the energy consumption value required by the vehicle to pass through each road segment according to the energy consumption calculation function and the road attribute data; and assigning the calculated energy consumption value to the corresponding edge in the path planning graph as the energy consumption weight.

[0128] In the above scheme, the nonlinear charging function refers to a function relationship reflecting the change of battery charging efficiency with state of charge and ambient temperature, contains the information of efficiency difference in different charging stages, includes the corresponding law of power, temperature and efficiency, and can be used to calculate the expected charging efficiency of the charging station. The current battery state of charge refers to the proportion of the current remaining power of the target vehicle battery, contains the battery energy storage condition, and can reflect the vehicle power supply demand. The current ambient temperature refers to the degree of cold and heat of the environment where the target vehicle is located, contains temperature value information, and will affect the battery charging efficiency. The expected charging efficiency refers to the expected power conversion efficiency of the target vehicle when charging at different charging stations, contains the relationship between charging speed and energy consumption, and can be used to select a better charging station. The spatial topology information refers to the road network and charging station location association information extracted from the global geomagnetic fingerprint library, contains the road connection relationship and station distribution, and can be used to build a path planning basis. The path planning graph is a graphical path model taking road intersections and charging stations as nodes and road segments as edges, contains node coordinates and edge attributes, and can be used for path search. The vehicle travel energy consumption model refers to a model for calculating the energy consumption of the vehicle during travel, contains factors such as speed, slope and mass, and can be used to evaluate the energy consumption of road segments. The energy consumption weight refers to the energy consumption value corresponding to each road segment in the path planning graph, contains road segment energy consumption information, and can be used as a basis for path optimization. The energy-optimal dynamic power-up path refers to the power-up travel route with the minimum total energy consumption after considering charging efficiency and travel energy consumption, contains the optimal path from the starting point to the charging station, and can achieve efficient power-up.

[0129] In the embodiments of the present application, as shown in Figure 2 Step S1051, the battery management system (BMS) carried by the target vehicle reads the current remaining power percentage of the battery. This system continuously monitors the charging and discharging state of the battery and calculates the power percentage, and at the same time, collects temperature data of the location where the vehicle is located through the vehicle-mounted environmental temperature sensor, and the collection frequency is synchronized with the positioning data. Finally, the current battery state of charge and the current ambient temperature, two key parameters, are obtained.

[0130] Next, the nonlinear charging function of the battery is pre-stored in step S1052. This function is fitted from a large amount of experimental data as where η is the charging efficiency, SOC is the state of charge of the battery, and T is the ambient temperature, which can reflect the efficiency variation law under different working conditions. The current battery state of charge 35%, i.e. SOC=35 and the current ambient temperature 25℃ obtained in step S1051 are substituted into the function to calculate the expected charging efficiency of each charging station in the target area of the target vehicle, for example, the expected charging efficiency of charging station A is calculated as If not, adjust the reasonable coefficient, set the function as Substitute it to get , and the function coefficient is fine-tuned to obtain = 0.68.

[0131] Then, step S1053 is implemented to extract spatial topology information from the incrementally updated global geomagnetic fingerprint database, which contains the connection relationship of the road network in the target area, the length of each road, and the specific coordinate positions of all charging stations; a path planning graph is constructed based on these information, the road intersections and each charging station are taken as nodes in the graph, the actual road segments connecting the nodes are taken as edges in the graph, and each edge is labeled with corresponding road identification information to form a visual path planning base model.

[0132] Then, step S1054 is performed to first establish a vehicle driving energy consumption model where E is the energy consumption value, k is the energy consumption coefficient taking 0.001, m is the vehicle mass taking 1500 kg, v is the average driving speed, s is the road length, is the road slope; then data is obtained from the road attribute database, including the road length, the road slope, and the average driving speed of the vehicle on the road segment, and these data are substituted into the vehicle driving energy consumption model to calculate the energy consumption value required by the vehicle to pass through each road segment. For example, the road segment vehicle C-charging pile A is 200 meters long, the slope is 1.8°, and the average speed is 10 km / h, and the road segment vehicle C-charging pile B is 150 meters long, the slope is 2.5°, and the average speed is 8 km / h. Substituting the data into the model to calculate: the energy consumption of vehicle C-charging pile A is units, and the energy consumption of vehicle C-charging pile B is units. These values are the energy consumption weights of the corresponding edges in the path planning graph.

[0133] Finally, step S1055 is performed to take the accurate position coordinate of the target vehicle as the starting point, combine the expected charging efficiency calculated in step S1052 and the energy consumption weight obtained in step S1054, and use the improved Dijkstra algorithm to search for a path in the path planning graph. This algorithm will comprehensively consider the total energy consumption of “driving energy consumption + charging energy consumption”, calculate the total energy consumption of all possible paths from the starting point to each charging station, and the calculation formula is as follows: and select the path with the minimum total energy consumption as the energy-optimal dynamic power supplement path. For example, the charging efficiency of the path vehicle C→charging pile A is 0.7325, the driving energy consumption is 94.2, and the total energy consumption is calculated to be units, the charging efficiency of the path vehicle C→charging pile B is 0.68, the driving energy consumption is 78.48, and the total energy consumption is calculated to be units, and finally the path vehicle C→charging pile A with the minimum total energy consumption is selected as the energy-optimal dynamic power supplement path.

[0134] In a practical application, within the parking lot of an underground logistics park, a target vehicle of brand B obtains a current battery charge level of 30% through its onboard battery management system and a current ambient temperature of 27°C through an ambient temperature sensor. These two parameters are then substituted into a pre-stored nonlinear battery charging function. ,in Using charging efficiency as the benchmark, SOC as the state of charge, and T as the temperature, the expected charging efficiency of charging station D in the park is calculated to be 0.84, and the expected charging efficiency of charging station E is 0.79. The park's road network connectivity, road lengths, and the location information of stations D and E are extracted from the global geomagnetic fingerprint database. A path planning graph is constructed using road intersections F and G and D and E as nodes and road segments as edges. A vehicle driving energy consumption model is then established. Where k=0.001, m=1600kg, the road segment FD is 180 meters long, with a slope of 2° and an average speed of 9km / h, and FE is 220 meters long, with a slope of 1.5° and an average speed of 8km / h. The calculated energy consumption of FD is 88.1 units, and the energy consumption of FE is 73.1 units. Taking the precise intersection point F of the target vehicle as the starting point, the improved Dijkstra algorithm is used to calculate the total energy consumption. The total energy consumption from F to D is... The unit is the total energy consumption from F to E. The unit ultimately generates the optimal dynamic energy replenishment path from F to E.

[0135] The overall solution of S105 described above combines the nonlinear charging function of the battery with the precise location of the target vehicle, and integrates the battery level, ambient temperature and spatial topology information to dynamically plan the charging path. It takes into account the differences in charging efficiency of different charging stations and the energy consumption during driving, avoiding the limitations of optimizing only distance. The generated energy-optimal dynamic charging path can effectively reduce the total energy consumption of electric vehicles during the charging process, improve the efficiency and economy of charging, and adapt to the needs of scenarios without GPS signal, ensuring the rationality and practicality of charging guidance.

[0136] The following is a complete example for steps 101-105, such as Figure 3 As shown, when implementing dynamic power replenishment path planning in the parking lot of Underground Commercial Center A, the parking lot was first divided into multiple geographic location grids according to a predefined rule of 3 meters × 3 meters. A data collection vehicle equipped with a geomagnetic sensor traveled along a predetermined "U"-shaped path, simultaneously recording the vehicle's geographic location coordinates and geomagnetic intensity readings. Subsequently, the coordinates were converted to a unified Cartesian coordinate system, and the consistency of the magnetic intensity readings was checked, removing grids with readings lower than [the specified value]. or higher The outliers are removed, and the average value of each group of readings is calculated as the reference geomagnetic vector by grid grouping. Finally, the coordinates of the center of each grid are mapped with the corresponding reference geomagnetic vector, and stored as the geomagnetic reference fingerprint library.

[0137] Then, 10 C brand crowd-sourced vehicles upload sensor raw readings at a frequency of every 20 seconds. The system first performs signal quality evaluation on the raw data, and filters out valid readings with a signal-to-noise ratio greater than 30. Then, according to the environmental temperature of the valid readings, the pre-stored temperature geomagnetic relationship curve is queried for temperature drift compensation. The vehicle's pre-stored hard iron interference parameters are read to calculate the interference vector and subtract it from the geomagnetic readings. After that, the corrected data is uniformly converted to nano Tesla units, and the standardized geomagnetic vector data sequence is generated by organizing the time sequence.

[0138] Subsequently, the geomagnetic reference fingerprint library is converted into the initial model parameters of federated learning, and the mapping of grid coordinates and geomagnetic vectors is converted into vector form and distributed to the 10 crowd-sourced vehicle clients. After receiving the initial parameters, each client iteratively trains the local model parameters 50 times using the standardized data stored locally for nearly 2 hours through the gradient descent algorithm, and uploads the updated local model parameters with data volume information to the cloud. The cloud records the data volume of each client, such as 1000 for 3 vehicles, 800 for 5 vehicles, and 600 for 2 vehicles. After calculating the aggregation weights, the initial aggregation parameters are obtained by weighted averaging using the federated averaging algorithm. After difference judgment, if the difference is less than the threshold value 0.12, the global model parameters are obtained. After L2 regularization optimization, the global geomagnetic fingerprint library is updated, and the geomagnetic mapping data of the newly added passageway in the parking lot is supplemented.

[0139] Then, the geomagnetic sensor of the D brand target vehicle collects data every 5 seconds, and constructs a geomagnetic vector sequence in time order after continuously collecting 20 groups. According to the preset window length of 5, it is divided into 16 sliding window data segments, and the Euclidean distance similarity and multi-angle direction comparison similarity of each data segment with each geographical location point in the global fingerprint library are calculated. Each window selects the three candidate geographical location points with the highest comprehensive similarity. According to the probability weights 、 、 , the candidate point coordinates and corresponding weights are weighted and averaged to generate the accurate position coordinates of the target vehicle in the parking lot, i.e. the intersection point E.

[0140] Finally, the target vehicle obtains the current battery power state 28% and environmental temperature 26°C through the vehicle-mounted system, and substitutes them into the nonlinear charging function to calculate the expected charging efficiency of the charging station F in the parking lot as 0.85 and the charging station G as 0.80. The road network and station location information are extracted from the global fingerprint library, and the path planning graph is constructed with the intersection points E, H and the stations F, G as nodes and the road segments as edges. The energy consumption model Wherein k=0.001, m=1500kg, the road section E-F is 190 meters long, the slope is 1.6°, and the speed is 9km / h, E-G is 210 meters long, the slope is 2.2°, and the speed is 8km / h, the calculation of E-F energy consumption is 75.4 units, and the calculation of E-G energy consumption is 88.1 units; taking the intersection E as the starting point, the improved Dijkstra algorithm is used to calculate the total energy consumption E-F as units, and the total energy consumption E-G is units, and finally the energy-optimal dynamic charging path E→F is generated.

[0141] Figure 4 A specific implementation structure diagram of a kind of dynamic charging path optimization system of electric vehicle based on nonlinear charging function provided by the embodiment of the application, refer to Figure 4 The system can include:

[0142] The acquisition module 41 is used to collect the geomagnetic fingerprint map data of the target area, pre-process the geomagnetic fingerprint map data, and generate a geomagnetic reference fingerprint library containing the mapping relationship between geographic location and geomagnetic field intensity vector;

[0143] The processing module 42 is used to receive the sensor raw readings uploaded by the crowd-sourcing vehicles, perform signal correction processing on the sensor raw readings, generate standardized geomagnetic vector data through temperature drift compensation and hard iron interference correction;

[0144] The optimization module 43 is used to take the geomagnetic reference fingerprint library as the initial model parameter of federated learning, perform local model training based on the standardized geomagnetic vector data by each crowd-sourcing vehicle client and upload the model parameter, aggregate and optimize the distributed model parameter through the cloud server to obtain the incrementally updated global geomagnetic fingerprint library;

[0145] The matching module 44 is used to perform multi-dimensional pattern matching between the geomagnetic vector data collected by the target vehicle and the global geomagnetic fingerprint library, generate the accurate position coordinates of the target vehicle through the sliding window matching mechanism and the probability weighted positioning calculation;

[0146] The generation module 45 is used to perform dynamic charging path planning based on the nonlinear charging function of the battery and the accurate position coordinates of the target vehicle, generate the energy-optimal dynamic charging path by fusing the battery state of charge, the environmental temperature and the spatial topology information in the global geomagnetic fingerprint library.

[0147] The system for optimizing the dynamic power supplement path of the electric vehicle based on the nonlinear charging function according to the embodiments of the present application is used to realize the method for optimizing the dynamic power supplement path of the electric vehicle based on the nonlinear charging function, and the specific embodiments of the system for optimizing the dynamic power supplement path of the electric vehicle based on the nonlinear charging function can refer to the embodiments of the method for optimizing the dynamic power supplement path of the electric vehicle based on the nonlinear charging function, which will not be repeated here.

[0148] The present application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the methods for optimizing the dynamic power supplement path of the electric vehicle based on the nonlinear charging function.

[0149] The present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of any of the methods for optimizing the dynamic power supplement path of the electric vehicle based on the nonlinear charging function.

[0150] In an exemplary embodiment, the computer readable storage medium can include, but is not limited to, a U disk, a read-only memory, a random access memory, a mobile hard disk, a magnetic disk or an optical disk, and various media that can store computer programs.

[0151] The embodiments of the present application also provide a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps of any of the methods for optimizing the dynamic power supplement path of the electric vehicle based on the nonlinear charging function.

[0152] The skilled person can further realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been described in general terms in the above description. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0153] The above describes in detail the method and system for optimizing the dynamic power supplement path of the electric vehicle based on the nonlinear charging function provided by the present application. The principles and implementation modes of the present application are described by applying specific examples, and the above description of the examples is only used to help understand the method of the present application and its core idea. It should be pointed out that, for those skilled in the art, some improvements and modifications can be made to the present application without departing from the principles of the present application, and these improvements and modifications also fall within the protection scope of the present application.

Claims

1. A method for optimizing the dynamic charging path of electric vehicles based on a nonlinear charging function, characterized in that, include: Collect geomagnetic fingerprint map data of the target area, preprocess the geomagnetic fingerprint map data, and generate a geomagnetic reference fingerprint database containing the vector mapping relationship between geographical location and geomagnetic field intensity. Receive raw sensor readings uploaded by crowdsourced vehicles, perform signal correction processing on the raw sensor readings, and generate standardized geomagnetic vector data through temperature drift compensation and hard iron interference correction; Using the geomagnetic benchmark fingerprint database as the initial model parameters for federated learning, each crowdsourced vehicle client trains its local model based on the standardized geomagnetic vector data and uploads the model parameters. The distributed model parameters are aggregated and optimized by the cloud server to obtain an incrementally updated global geomagnetic fingerprint database. The geomagnetic vector data collected from the target vehicle is matched with the global geomagnetic fingerprint database in multiple dimensions. The precise location coordinates of the target vehicle are generated by a sliding window matching mechanism and probability-weighted positioning calculation. Dynamic charging path planning is performed based on the nonlinear charging function of the battery and the precise location coordinates of the target vehicle. By integrating the battery charge status, ambient temperature and spatial topology information in the global geomagnetic fingerprint database, the energy-optimal dynamic charging path is generated. The dynamic charging path planning is performed using the battery-based nonlinear charging function and the precise location coordinates of the target vehicle. By fusing battery state of charge, ambient temperature, and spatial topology information from the global geomagnetic fingerprint database, an energy-optimal dynamic charging path is generated, including: Obtain the target vehicle's current battery charge status and current ambient temperature; Based on the nonlinear charging function of the battery, the current battery state of charge, and the current ambient temperature, the expected charging efficiency of the target vehicle at different charging stations is calculated. Spatial topology information containing road network information and charging station location information is extracted from the global geomagnetic fingerprint database, and a path planning map is constructed based on the spatial topology information. The nodes in the path planning map include road intersections and charging stations, and the edges represent road segments. Determine the vehicle driving energy consumption model, and calculate the energy consumption weight of each edge in the path planning graph based on the vehicle driving energy consumption model; Starting from the precise location coordinates of the target vehicle, and combining the expected charging efficiency and the energy consumption weight, the driving path with the minimum energy consumption is searched in the path planning map to generate the dynamic charging path with optimal energy.

2. The method according to claim 1, characterized in that, The process of determining the vehicle driving energy consumption model, and calculating the energy consumption weight of each edge in the path planning graph based on the vehicle driving energy consumption model, includes: Establish an energy consumption calculation function based on vehicle speed, road gradient, and vehicle mass; Obtain road attribute data for each edge in the path planning graph. The road attribute data includes road length, road gradient, and average vehicle speed. Based on the energy consumption calculation function and road attribute data, calculate the energy consumption required for the vehicle to pass through each road segment; The calculated energy consumption value is assigned to the corresponding edge in the path planning graph as the energy consumption weight.

3. The method according to claim 1, characterized in that, The process involves using the geomagnetic benchmark fingerprint database as the initial model parameters for federated learning. Each crowdsourced vehicle client trains its local model based on the standardized geomagnetic vector data and uploads the model parameters. The distributed model parameters are then aggregated and optimized via a cloud server to obtain an incrementally updated global geomagnetic fingerprint database, including: The geomagnetic reference fingerprint database is converted into initial model parameters for federated learning, and the initial model parameters are distributed to the crowdsourced vehicle client. Each crowdsourced vehicle client receives the initial model parameters, performs local training on the initial model parameters based on locally stored standardized geomagnetic vector data, obtains updated local model parameters, and uploads the updated local model parameters to the cloud server; The cloud server receives local model parameters uploaded by each client, and aggregates the distributed model parameters using a federated averaging algorithm to generate aggregated global model parameters. The aggregated global model parameters are optimized, and the optimized global model parameters are converted into an incrementally updated global geomagnetic fingerprint database.

4. The method according to claim 3, characterized in that, The process involves receiving local model parameters uploaded by each client through the cloud server, aggregating the distributed model parameters using a federated averaging algorithm, and generating aggregated global model parameters, including: The cloud server receives local model parameters uploaded by each crowdsourced vehicle client, records the amount of training data for each client, and calculates the corresponding aggregation weight based on the amount of training data. The client with the larger amount of training data is assigned a higher aggregation weight. Based on the aggregate weights, all local model parameters are fused using a weighted average method to obtain initial aggregate parameters, and the difference between the initial aggregate parameters and each local model parameter is calculated. Determine whether the difference is less than a preset difference threshold. If it is less than the difference threshold, then use the initial aggregation parameters as the global model parameters after aggregation. If the difference is greater than or equal to the difference threshold, the aggregation weight of each client is adjusted according to the difference degree, and the weighted average and difference degree are re-judged based on the adjusted weight until the difference degree is less than the difference threshold. The aggregation parameters obtained at this time are used as the global model parameters after aggregation.

5. The method according to claim 1, characterized in that, The step of performing multi-dimensional pattern matching between the geomagnetic vector data collected from the target vehicle and the global geomagnetic fingerprint database, and generating the precise location coordinates of the target vehicle through a sliding window matching mechanism and probability-weighted positioning calculation, includes: The geomagnetic vector data collected from the target vehicle are used to construct a geomagnetic vector sequence in chronological order of collection time. The geomagnetic vector sequence is divided into multiple sliding window data segments according to a preset window length. The multi-dimensional similarity between each sliding window data segment and the geomagnetic vector data of each geographical location in the global geomagnetic fingerprint database is calculated. The multi-dimensional similarity includes Euclidean distance similarity and multi-angle direction comparison similarity. By using a sliding window matching mechanism, several candidate geographic locations with the highest similarity are selected for each sliding window data segment; Based on the multi-dimensional similarity values ​​of the candidate geographic locations, the probability weight of each candidate geographic location is calculated; Based on the probability weights and coordinate values ​​of all candidate geographical locations, a probability-weighted average is calculated to generate the precise location coordinates of the target vehicle.

6. The method according to claim 1, characterized in that, The process of receiving raw sensor readings uploaded by crowdsourced vehicles, performing signal correction processing on the raw sensor readings, and generating standardized geomagnetic vector data through temperature drift compensation and hard iron interference correction includes: Receive raw sensor readings uploaded by crowdsourced vehicles, including geomagnetic intensity readings and corresponding ambient temperature readings; The original readings of the sensor are evaluated for signal quality, and valid readings that meet the preset quality threshold are selected. Temperature drift compensation processing is performed on the effective readings. Based on the ambient temperature reading of the effective readings, a pre-stored temperature-geomagnetic relationship curve is queried, the compensation value of the corresponding geomagnetic intensity reading is calculated, and the compensation value is applied to the geomagnetic intensity reading. The temperature-compensated geomagnetic intensity readings are subjected to hard iron interference correction processing. The pre-stored vehicle hard iron interference parameters are read, the hard iron interference vector is calculated based on the vehicle hard iron interference, and the hard iron interference vector is subtracted from the geomagnetic intensity readings. The corrected geomagnetic intensity readings are dimensionally standardized and converted to geomagnetic vector representation under standard physical dimensions. The processed geomagnetic vector data is organized according to time series to generate standardized geomagnetic vector data.

7. A dynamic charging path optimization system for electric vehicles based on a nonlinear charging function, characterized in that, include: The acquisition module is used to acquire geomagnetic fingerprint map data of the target area, preprocess the geomagnetic fingerprint map data, and generate a geomagnetic reference fingerprint database containing the vector mapping relationship between geographical location and geomagnetic field intensity. The processing module is used to receive the raw sensor readings uploaded by the crowdsourced vehicles, perform signal correction processing on the raw sensor readings, and generate standardized geomagnetic vector data through temperature drift compensation and hard iron interference correction. The optimization module is used to use the geomagnetic benchmark fingerprint database as the initial model parameters for federated learning. Each crowdsourced vehicle client trains a local model based on the standardized geomagnetic vector data and uploads the model parameters. The distributed model parameters are aggregated and optimized through the cloud server to obtain an incrementally updated global geomagnetic fingerprint database. The matching module is used to perform multi-dimensional pattern matching between the geomagnetic vector data collected from the target vehicle and the global geomagnetic fingerprint database, and generate the precise location coordinates of the target vehicle through a sliding window matching mechanism and probability-weighted positioning calculation. The generation module is used to perform dynamic charging path planning based on the nonlinear charging function of the battery and the precise location coordinates of the target vehicle. By integrating the battery charge status, ambient temperature and spatial topology information in the global geomagnetic fingerprint database, it generates the dynamic charging path with optimal energy. The dynamic charging path planning is performed using the battery-based nonlinear charging function and the precise location coordinates of the target vehicle. By fusing battery state of charge, ambient temperature, and spatial topology information from the global geomagnetic fingerprint database, an energy-optimal dynamic charging path is generated, including: Obtain the target vehicle's current battery charge status and current ambient temperature; Based on the nonlinear charging function of the battery, the current battery state of charge, and the current ambient temperature, the expected charging efficiency of the target vehicle at different charging stations is calculated. Spatial topology information containing road network information and charging station location information is extracted from the global geomagnetic fingerprint database, and a path planning map is constructed based on the spatial topology information. The nodes in the path planning map include road intersections and charging stations, and the edges represent road segments. Determine the vehicle driving energy consumption model, and calculate the energy consumption weight of each edge in the path planning graph based on the vehicle driving energy consumption model; Starting from the precise location coordinates of the target vehicle, and combining the expected charging efficiency and the energy consumption weight, the driving path with the minimum energy consumption is searched in the path planning map to generate the dynamic charging path with optimal energy.

8. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the electric vehicle dynamic charging path optimization method based on a nonlinear charging function as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the electric vehicle dynamic charging path optimization method based on a nonlinear charging function as described in any one of claims 1 to 6.

Images