A method and system for predicting the remaining driving range of lithium battery vehicles online

Abstract

This invention belongs to the field of lithium-ion battery vehicle technology, and relates to an online method and system for predicting the remaining driving range of lithium-ion battery vehicles, including the following steps: collecting vehicle data during a preset process, and predicting the remaining driving range of vehicles based on the Spearman coefficient ρ between various data items. s The process involves selecting variables strongly correlated with remaining driving range; using a least squares regression algorithm to clean the vehicle data corresponding to these strongly correlated variables, removing outliers; inputting the cleaned vehicle data into a bidirectional stacked gated recurrent network to selectively record and discard vehicle data during driving, and mining temporal dependencies online; and then inputting the vehicle data and temporal dependencies into a Kolmogorov-Arnold network to achieve nonlinear mapping, extracting intermediate sequence features from the vehicle data, and outputting the remaining driving range. This accurate prediction of remaining driving range allows drivers to plan their trips, optimize charging methods and times, alleviate range anxiety, and improve the driving experience.

Description

Technical Field

[0001] This invention relates to an online method and system for predicting the remaining driving range of lithium-ion battery vehicles, belonging to the field of lithium-ion battery vehicle technology. Background Technology

[0002] Lithium-ion battery vehicles offer advantages such as environmental friendliness, high energy conversion efficiency, quiet operation, and low operating costs. However, if the remaining driving range of a lithium-ion battery vehicle cannot be provided, users may worry about the vehicle's range and lack confidence to continue driving for extended periods, resulting in the actual driving range being lower than the nominal range. This causes range anxiety and prevents users from fully utilizing the vehicle's performance. Furthermore, during the driving process of an electric vehicle, the remaining driving range exhibits a complex and non-linear decreasing trend depending on battery status and driving environment. The remaining driving range of an electric vehicle is affected by multiple factors, including driving speed and vehicle power, which are influenced by terrain, traffic congestion, and user driving behavior, all of which affect the accuracy of range prediction. Summary of the Invention

[0003] To address the aforementioned issues, the purpose of this invention is to provide an online method and system for predicting the remaining driving range of lithium battery vehicles. This system can accurately predict the remaining driving range, allowing drivers to rationally plan their trips, optimize charging methods and times, alleviate range anxiety, and improve the driving experience.

[0004] To achieve the above objectives, the present invention proposes the following technical solution: an online method for predicting the remaining driving range of a lithium battery vehicle, comprising the following steps: collecting vehicle data during a preset process, and determining the remaining driving range based on the Spearman coefficient ρ between various items of the vehicle data. s The process involves selecting variables strongly correlated with remaining mileage; using a least squares regression algorithm to clean the vehicle data corresponding to these strongly correlated variables and remove outliers; inputting the cleaned vehicle data into a bidirectional stacked gated recurrent network to selectively record and discard vehicle data during the driving process, thereby mining temporal dependencies online; and then inputting the vehicle data and temporal dependencies into a Kolmogorov-Arnold network to achieve nonlinear mapping, extract intermediate sequence features from the vehicle data, and output the remaining mileage.

[0005] Furthermore, the preset process is the process of the current vehicle battery state of charge (SOC) decreasing by 1%; the vehicle data includes battery status information and driving environment information, the battery status information includes voltage, current and battery SOC, and the driving environment information includes ambient temperature, total vehicle mileage and vehicle speed.

[0006] Furthermore, the method for cleaning the vehicle data corresponding to the strongly correlated variables is as follows: delete vehicle driving process data with power consumption below a preset value; obtain the calculated capacity of the on-board battery in each usage cycle, and delete vehicle data corresponding to calculated capacity higher than the factory nominal capacity; generate a linear fitting equation for capacity change over time according to the least squares regression algorithm, obtain the fitted capacity corresponding to the usage cycle based on the linear fitting equation, calculate the difference between the calculated capacity and the fitted capacity, and delete vehicle data where the difference is greater than a threshold; delete abnormal battery capacity data and corresponding vehicle data to complete the data cleaning.

[0007] Furthermore, the calculation method for the on-board battery capacity is as follows: The energy released by the on-board battery within a sampling period is obtained by multiplying the battery's power output by the sampling period. This process is repeated for each sampling period during the driving process to obtain the energy released by the on-board battery for all sampling periods. The energy released by the battery in all sampling periods is then discretely summed and divided by the depth of discharge, where the depth of discharge refers to the reduction in the battery's state of charge (SOC). For example, if the SOC is 90% when the vehicle starts and 40% at the end of the trip, then the depth of discharge is 90% - 40% = 50%, thus obtaining the calculated capacity of the on-board battery during the driving process.

[0008] Furthermore, the slope of the linear fitting equation for the capacity changing over time is:

[0009]

[0010] Where b is the slope, Cap i It refers to the calculated capacity of the vehicle's battery, Cap. mean This is the average calculated capacity of the vehicle's battery; Num i It is the number of travel stages, Num mean It is the average number of driving processes.

[0011] Furthermore, in the bidirectional stacked gated cyclic network, a bidirectional stacked structure is established with the gated cyclic network as the basic unit. The current state of each unit is simultaneously input into the forward and reverse directional sequences. After data feature extraction, the hidden states of the sequences in the two directions are fused to simultaneously mine the forward and reverse propagation information in the data feature sequences.

[0012] Furthermore, in the Kolmogorov-Arnold network, nodes are responsible for addition operations and do not contain nonlinear activation functions. The nonlinear activation function is moved to the edge of the Kolmogorov-Arnold network and calculated using spline functions that can be trained and updated. The calculation result is used as a component of the weight function. By adjusting the activation function at the network edge, the Kolmogorov-Arnold network represents the complex continuous function of predicting the remaining driving range of the vehicle as the sum of several basic elementary functions. By nesting and stacking multiple basic elementary functions, a multi-layer nonlinear neural network is constructed to map the intermediate feature sequence of vehicle data and predict the remaining driving range of the vehicle.

[0013] Furthermore, the complex continuous function f(x) is expressed as:

[0014]

[0015] Where n represents the number of activation functions available for training in each layer of the neural network, φ (l,i,j) It is the residual activation function used to replace the weight parameters of the KAN, where L is the number of layers in the Kolmogorov-Arnold network (KAN). The residual activation function is a univariate continuous function obtained by a linear combination of basis functions and spline functions.

[0016] Furthermore, the accuracy evaluation index R of the output vehicle's remaining driving range... 2 for:

[0017]

[0018] Among them, rdr i Fore represents the actual remaining mileage of the vehicle. i The remaining driving distance is predicted using the model, where n is the number of predictions made during the vehicle's journey. It is the average of n actual remaining driving distances.

[0019] This invention also discloses an online lithium battery vehicle remaining driving range prediction system, comprising: a data acquisition module for acquiring vehicle data during a preset process, and based on the Spearman coefficient ρ between various items of the vehicle data. sThe system selects variables strongly correlated with the remaining driving mileage; a data cleaning module cleans the vehicle data corresponding to the strongly correlated variables using a least squares regression algorithm to remove outliers; a GRU network output module inputs the cleaned vehicle data into a bidirectional stacked gated recurrent network to selectively record and forget vehicle data during the driving process, and mines temporal dependencies online; a KAN network output module inputs the vehicle data and temporal dependencies into a Kolmogorov-Arnold network to achieve nonlinear mapping, extract intermediate sequence features of the vehicle data, and output the remaining driving mileage.

[0020] The technical solution of the present invention has at least the following technical effects or advantages:

[0021] 1. The algorithm model for online prediction of the remaining driving range of lithium battery vehicles designed in this invention does not need to rely too much on existing historical information. After the model is trained, it only needs the data of the current vehicle battery SOC decreasing by 1% to predict the remaining driving range of the vehicle.

[0022] 2. Based on Spearman correlation analysis, this invention can select relevant data that are strongly correlated with the remaining driving mileage from multiple categories of data. Through data cleaning, outlier data is removed and key data features are screened, which can effectively improve the convergence speed of model training.

[0023] 3. This invention combines the advantages of bidirectional stacked GRU and KAN modules. It uses the GRU network to mine relevant time-series information and then uses KAN regression to predict the remaining driving range, thereby improving the accuracy of range prediction and realizing online prediction and display of the remaining range of the vehicle battery management system.

[0024] 4. This invention can predict the remaining driving range, help users to reasonably arrange their trips, plan charging methods and times, alleviate range anxiety, and improve the driving experience; it can also help car manufacturers develop more refined vehicle energy management strategies and improve the energy storage utilization rate of on-board batteries. Attached Figure Description

[0025] Figure 1 This is a flowchart of an online lithium battery vehicle remaining driving range prediction method according to an embodiment of the present invention;

[0026] Figure 2 This is a schematic diagram of the structure of a gated loop unit in one embodiment of the present invention;

[0027] Figure 3 This is a schematic diagram of the structure of a bidirectional stacked gated loop network in one embodiment of the present invention;

[0028] Figure 4This is a schematic diagram of the structure of the Kolmogorov-Arnold network in one embodiment of the present invention;

[0029] Figure 5 This is a schematic diagram of the remaining driving mileage prediction result in one embodiment of the present invention. Detailed Implementation

[0030] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention is described in detail through specific embodiments. However, it should be understood that the specific embodiments are provided only for a better understanding of the present invention and should not be construed as limiting the present invention. In the description of the present invention, it should be understood that the terminology used is for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0031] To address the problems of inaccurate prediction of remaining driving range in existing technologies, which hinders drivers' ability to effectively plan their trips and leads to range anxiety, this invention proposes an online method and system for predicting the remaining driving range of lithium-ion battery vehicles. It utilizes the Kolmogorov-Arnold theorem and a bidirectional stacked gated recurrent network (GRU) to predict the vehicle's remaining driving range online. First, it collects vehicle data during the 1% decrease in the battery's state of charge (SOC). This data is then cleaned, and relevant time-series information is mined using a bidirectional stacked gated recurrent network (GRU). Finally, a nonlinear regression prediction is implemented based on a Kolmogorov-Arnold network (KAN), outputting the remaining driving range online. This accurate prediction of remaining driving range allows drivers to rationally plan their trips, optimize charging methods and times, alleviate range anxiety, and improve the driving experience. The invention will be described in detail below with reference to the accompanying drawings.

[0032] Example 1

[0033] This embodiment discloses an online method for predicting the remaining driving range of lithium battery vehicles, such as... Figure 1 As shown, it includes the following steps:

[0034] S1 collects vehicle data during the preset process, and calculates the Spearman coefficient ρ between the various vehicle data items. s We selected variables that are strongly correlated with the remaining driving mileage.

[0035] The preset process is the decrease of the current state of charge (SOC) of the vehicle battery by 1%. In this embodiment, 30 sets of vehicle data during this 1% SOC decrease process are used as the basic input unit. Each set of vehicle data includes two types of data: battery status information and driving environment information. The battery status information includes voltage, current, and SOC, reflecting the impact of the vehicle's power consumption and remaining battery capacity on the driving range. The driving environment information includes ambient temperature, total vehicle mileage, and vehicle speed, reflecting the impact of battery capacity decay and vehicle driving resistance on the efficiency of converting battery energy into vehicle kinetic energy.

[0036] Spearman coefficients are used to calculate the correlation between in-vehicle data and the vehicle's driving environment within the basic input unit, analyzing the impact of in-vehicle data on the vehicle's remaining driving range. The correlation between the vehicle data input to the model and the remaining driving range differs; therefore, Spearman coefficients are needed to mine the influence of relative rank position on the monotonic relationship of the in-vehicle data, analyzing the monotonic relationships between various variables within the vehicle dataset to adjust the weights of the activation function within the deep learning model. Specifically, the Spearman coefficient ρ... s The calculation formula is:

[0037]

[0038] Among them, R x R is the rank of the x-th vehicle data point after sorting the original sequence; y It is the rank of the y-th vehicle data after sorting the original sequence; Cov(R) x ,R y ) represents the covariance of the positional rank. It is the standard deviation of the position rank variable x. ρ is the standard deviation of the positional rank variable y. s The value ranges between -1 and 1, and the larger the absolute value, the stronger the correlation.

[0039] The input data for this deep learning model is in key-value pair format, including timestamps as keys and sensor data and remaining driving range as data. Each timestamp is unique and corresponds to the vehicle data at the current moment. In this embodiment, data strongly correlated with remaining driving range are selected, including SOC, voltage, battery maximum peak current, and vehicle acceleration. The values ​​of these data decrease as mileage increases and battery power is depleted.

[0040] In this embodiment, the vehicle mileage ranges from 0 to 150,000 kilometers, and the data collection period does not exceed six months. The collected vehicle data is shown in Table 1.

[0041] S2 uses the least squares regression algorithm to clean the vehicle data corresponding to strongly correlated variables and remove outliers.

[0042] When filtering vehicle driving data, due to limitations in battery sensor accuracy and sampling frequency, outliers in battery capacity may occur. Therefore, data cleaning is required for vehicle data corresponding to strongly correlated variables. The specific method is as follows:

[0043] S2.1 Delete vehicle driving data where power consumption is lower than the preset value.

[0044] The vehicle driving process is segmented based on the battery's state of charge (SOC). As the vehicle's mileage and the number of charge-discharge cycles increase, the battery's usable capacity gradually decreases. When predicting the battery capacity at different mileages, since shallow discharge processes cannot fully reflect the battery's capacity information, data segments of vehicle driving processes with at least 35% power consumption are selected.

[0045] S2.2 Obtain the calculated capacity of the vehicle battery in each adoption cycle, and delete vehicle data corresponding to calculated capacities higher than the factory nominal capacity.

[0046] The calculation method for the on-board battery capacity is as follows: During each segment of vehicle operation, the on-board battery's State of Charge (SOC) decreases monotonically. The energy released by the battery within a sampling period is obtained by multiplying the battery's power output by the sampling period (i.e., integrating the battery power over time). This process is repeated for each sampling period during the driving process to obtain the energy released by the battery for all sampling periods. The energy released by the battery in all sampling periods is then discretely summed and divided by the depth of discharge to obtain the calculated capacity of the on-board battery during the driving process. The depth of discharge refers to the reduction in the battery's SOC. For example, if the SOC is 90% when the vehicle starts and 40% at the end of the trip, then the depth of discharge is 90% - 40% = 50%.

[0047] Computational Capacity i The calculation formula is:

[0048]

[0049] Among them, I j For current, V J For voltage, ΔT j It is the battery's output power multiplied by the sampling period. SOC is the battery's state of charge. i -SOC j This is the depth of discharge.

[0050] Table 1. Collected Vehicle Data

[0051]

[0052] S2.3 Based on the least squares regression algorithm, a linear fitting equation for the capacity change over time is generated. Based on the linear fitting equation, the fitted capacity corresponding to the adopted period is obtained. The difference between the calculated capacity and the fitted capacity is calculated, and vehicle data with a difference greater than a threshold is deleted. Abnormal battery capacity data and corresponding vehicle data are deleted, thus completing the data cleaning.

[0053] A least squares regression algorithm was used to establish a regression fitting model for the historical capacity sequence of vehicle batteries. To screen outliers, the battery capacity time series decay process was approximated as a linear model, and the linear fitting equation of the original capacity data was obtained. Based on the linear fitting equation, the fitted capacity corresponding to the adopted period was obtained. The difference between the calculated capacity and the fitted capacity was calculated, and vehicle data with a difference greater than a threshold were deleted. Abnormal battery capacity data and their corresponding vehicle data were also deleted, completing the data cleaning process.

[0054] In this embodiment, the method for judging the anomaly is as follows: based on the linear fitting equation, if the difference obtained exceeds 35% of the calculated value, then the battery capacity calculation is considered abnormal, and the corresponding vehicle data and the corresponding capacity calculation value need to be discarded.

[0055] The slope of the linear fitting equation for the capacity changing over time is:

[0056]

[0057] Where b is the slope, Cap i It refers to the calculated capacity of the vehicle's battery, Cap. mean This is the average calculated capacity of the vehicle's battery; Num i It is the number of travel stages, Num mean It is the average number of driving processes.

[0058] S3 inputs the cleaned vehicle data into a bidirectional stacked gated recurrent network to selectively record and discard vehicle data during vehicle operation, thereby mining temporal dependencies online.

[0059] In this embodiment, the deep learning model consists of two cascaded network models, arranged in the following order according to the data flow process: a bidirectional stacked gated recurrent network (GRU) and a Kolmogorov-Arnold network (KAN), enabling online capture of data dependencies and prediction of the vehicle's remaining mileage. Before using it for predicting remaining mileage in real-world scenarios, this deep learning model needs to be trained and optimized. A bidirectional stacked model is obtained by combining the GRU to selectively record and discard vehicle information during the driving process, and to mine the temporal dependencies of this vehicle information online. Then, the KAN module is used to implement nonlinear mapping, extract intermediate sequence features, and output the vehicle's remaining mileage. By calculating the error between the predicted and actual remaining mileage values, backpropagating the error, and iteratively updating the model's internal weights, the optimal deep learning model is obtained.

[0060] The construction method of bidirectional stacked gated recurrent network is as follows:

[0061] First, establish a single gated recurrent node (GRU), such as Figure 2 As shown, data flow is controlled through update and reset gates to extract key features and dependencies of battery capacity degradation. Update gate z t The information to be retained in the current state is determined from two aspects: one is the information from the current candidate input x. t The decision to retain information is based on two factors: firstly, the information to be retained, and secondly, the output h from the hidden state of the previous time step. t-1 Select key information as shown in the following formula:

[0062] z t =σ(W z ·x t +U z ·h t-1 +b z )#

[0063] W z and U z These are the weight matrices for the current state input and the previous state, respectively, b. z These are bias vectors, and their parameters are continuously updated through deep learning training of a neural network; σ is the activation function used to map battery and driving environment information to remaining driving range, enabling the non-linear calculation of the update gate. The calculation process for the reset gate is similar to that of the update gate, as shown in the following equation:

[0064] r t =σ(W r ·x t +U r ·h t-1 +b r )#

[0065] Among them, Wr and U r It is the weight matrix of the reset gate, b z It is the bias vector.

[0066] The candidate state of the model at the current time step is calculated as follows:

[0067]

[0068] Among them, W h and U h Both are internal weight matrices of the model obtained through training, b h It is the bias vector, representing a linear transformation on the current input; the reset gate r t The output h of the hidden state at the previous time step t-1 Multiplication represents the feature extraction of historical state information; finally, the linearly transformed vector is fed into the activation function tanh, nonlinearly mapping the battery-related features to the interval between -1 and 1. The hidden state output of a single GRU is shown in the following equation:

[0069]

[0070] The hidden state output of each unit is affected by both the current state and the historical state, and the reset gate r t The closer the hidden state is to 0, the more it is affected by the current state; the closer the update gate is to 1, the more it is affected by the historical state of the previous time step. Through the gating mechanism, the GRU can simultaneously consider the impact of the current driving environment and existing battery information on the vehicle's remaining driving range, and calculate the weight matrix and bias information of each node through iterative training of the neural network.

[0071] like Figure 3 As shown, in the bidirectional stacked gated recurrent network (GRU), a bidirectional stacked structure is established using GRU as the basic unit. The current state of each unit is simultaneously input into both forward and reverse sequences. After data feature extraction, key features of the vehicle data are extracted, and the hidden states of the sequences from both directions are fused. The forward and backward propagation information in the data feature sequences is simultaneously mined to analyze the impact of sequence information on the remaining driving range. At each time step during training, the bidirectional stacked GRU fuses the outputs from both directions and the current state information to calculate the hidden state as the single-step output. By stacking two GRUs from different directions together, detailed forward and reverse sequence features can be captured, improving the accuracy of model feature extraction and information analysis.

[0072] S4 inputs vehicle data and temporal dependencies into the Kolmogorov-Arnold network to achieve nonlinear mapping, extract intermediate sequence features of vehicle data, and output the remaining driving mileage of the vehicle.

[0073] like Figure 4 As shown, in the Kolmogorov-Arnold network, nodes are responsible for addition operations and do not contain nonlinear activation functions. Instead, nonlinear activation functions are moved to the edges of the Kolmogorov-Arnold network and calculated using spline functions that can be trained and updated. The calculation results are used as components of the weight function. By adjusting the activation functions at the network edges, the Kolmogorov-Arnold network represents the complex continuous function predicting the vehicle's remaining mileage as the sum of several basic elementary functions. By nesting and stacking multiple basic elementary functions, a multi-layer nonlinear neural network is constructed, mapping the intermediate feature sequence of vehicle data to predict the vehicle's remaining mileage.

[0074] In the Kolmogorov-Arnold network, a multivariate continuous bounded function is first decomposed into the sum of continuous univariate functions, as shown in the following equation:

[0075]

[0076] By adjusting the activation functions of the neural network edges, the complex continuous function predicting the remaining driving range of a vehicle is represented as a sum of several basic elementary functions, thus constructing a multi-layer nonlinear neural network. The activation function of each edge is adjusted in stages according to the layer, achieving feature mapping. The complex continuous function f(x) is expressed as:

[0077]

[0078] Where n represents the number of activation functions available for training in each layer of the neural network, φ (l,i,j) It is the residual activation function used to replace the weight parameters of KAN, where L is the number of layers in the Kolmogorov-Arnold network (KAN). The residual activation function is a univariate continuous function obtained by a linear combination of basis functions and spline functions.

[0079] The model is trained to implement error backpropagation and iterative weight updates. Input data is forward-propagated through the network to obtain the output prediction. A 120-epoch cycle is set during model training. In each epoch, the output error is calculated and then backpropagated to the internal nodes and non-linear activation functions to adjust the weights. The error between the predicted remaining mileage output and the actual mileage label is calculated in each training epoch. The network activation function and node positions are updated through the backpropagation algorithm to reduce the model's prediction error and optimize the mileage prediction results.

[0080] The accuracy evaluation index R of outputting the vehicle's remaining driving range 2 for:

[0081]

[0082] Among them, rdr i Fore represents the actual remaining mileage of the vehicle. i The remaining driving distance is predicted using the model, where n is the number of predictions made during the vehicle's journey. This is the average of n actual remaining driving mileages. The predicted remaining driving mileage is as follows: Figure 5 As shown.

[0083] Example 2

[0084] Based on the same inventive concept, this embodiment discloses an online lithium battery vehicle remaining driving range prediction system, including:

[0085] The data acquisition module is used to collect vehicle data during a preset process, and to determine the relationship between different data items based on the Spearman coefficient ρ. s Select variables that are strongly correlated with the remaining driving mileage;

[0086] The data cleaning module is used to clean vehicle data corresponding to strongly correlated variables and remove outliers using the least squares regression algorithm.

[0087] The GRU network output module is used to input the cleaned vehicle data into a bidirectional stacked gated cyclic network for selectively recording and discarding vehicle data during vehicle operation, and to mine temporal dependencies online.

[0088] The KAN network output module is used to input vehicle data and temporal dependencies into the Kolmogorov-Arnold network to achieve nonlinear mapping, extract intermediate sequence features of vehicle data, and output the remaining driving range of the vehicle.

[0089] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0090] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0091] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0092] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific embodiments of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention. The above content is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the protection scope of the claims.

Claims

1. A method for predicting the remaining driving range of an online lithium-ion battery vehicle, characterized in that, Includes the following steps: Vehicle data is collected during the preset process, and the Spearman coefficients between various vehicle data items are used to... Select variables that are strongly correlated with the remaining driving mileage; The vehicle data corresponding to the strongly correlated variables were cleaned and outliers were removed using the least squares regression algorithm. The cleaned vehicle data is input into a bidirectional stacked gated recurrent network to selectively record and forget vehicle data during the vehicle's journey, and to mine temporal dependencies online. The vehicle data and temporal dependencies are input into the Kolmogorov-Arnold network to achieve nonlinear mapping, extract intermediate sequence features of the vehicle data, and output the remaining driving mileage of the vehicle. The preset process is the process of the current vehicle battery state of charge (SOC) decreasing by 1%; the vehicle data includes battery status information and driving environment information. The battery status information includes voltage, current and battery SOC, and the driving environment information includes ambient temperature, total vehicle mileage and vehicle speed.

2. The online lithium battery vehicle remaining driving range prediction method as described in claim 1, characterized in that, The method for cleaning vehicle data corresponding to the strongly correlated variables is as follows: delete vehicle driving process data with power consumption below a preset value; obtain the calculated capacity of the on-board battery in each usage cycle, and delete vehicle data corresponding to calculated capacity higher than the factory nominal capacity; generate a linear fitting equation for capacity changing with time based on the least squares regression algorithm, obtain the fitted capacity corresponding to the usage cycle based on the linear fitting equation, calculate the difference between the calculated capacity and the fitted capacity, and delete vehicle data where the difference is greater than a threshold; delete abnormal battery capacity data and corresponding vehicle data to complete data cleaning.

3. The online lithium battery vehicle remaining driving range prediction method as described in claim 2, characterized in that, The calculation method for the on-board battery capacity is as follows: the energy released by the on-board battery within the sampling period is obtained by multiplying the power released by the on-board battery by the sampling period. The energy released by the on-board battery in all sampling periods is obtained by iterating through each sampling period during the driving process. The energy released by the battery in all sampling periods is discretely summed and then divided by the depth of discharge to obtain the calculated capacity of the on-board battery during the driving process.

4. The online lithium battery vehicle remaining driving range prediction method as described in claim 2, characterized in that, The slope of the linear fitting equation for the capacity changing over time is: Where b is the slope. This is the calculated capacity of the vehicle's battery. It is the average calculated capacity of the vehicle battery; It is the number of driving processes. It is the average number of driving processes.

5. The online lithium battery vehicle remaining driving range prediction method as described in claim 1, characterized in that, In the bidirectional stacked gated cyclic network, a bidirectional stacked structure is established with the gated cyclic network as the basic unit. The current state of each unit is simultaneously input into the forward and reverse sequences. After data feature extraction, the hidden states of the two sequences are fused to simultaneously mine the forward and reverse propagation information in the data feature sequence.

6. The online lithium battery vehicle remaining driving range prediction method as described in claim 1, characterized in that, In the Kolmogorov-Arnold network, nodes are responsible for addition operations and do not contain nonlinear activation functions. Instead, the nonlinear activation function is moved to the edge of the Kolmogorov-Arnold network and calculated using spline functions that can be trained and updated. The calculation result is used as a component of the weight function. By adjusting the activation function at the network edges, the Kolmogorov-Arnold network transforms the complex continuous function of predicting the remaining driving range of a vehicle into a more efficient and effective system. It can be represented as the sum of several basic elementary functions; by nesting and stacking multiple basic elementary functions, a multi-layer nonlinear neural network is constructed to map the intermediate feature sequence of vehicle data and predict the remaining driving mileage of the vehicle.

7. The online lithium battery vehicle remaining driving range prediction method as described in claim 6, characterized in that, The complex continuous function Represented as: Where n represents the number of activation functions available for training in each layer of the neural network. It is a residual activation function used to replace the weight parameters of KAN, where L is the number of layers in the Kolmogorov-Arnold network. The residual activation function is a univariate continuous function obtained by a linear combination of basis functions and spline functions.

8. The online lithium battery vehicle remaining driving range prediction method as described in claim 1, characterized in that, Accuracy evaluation index for outputting the vehicle's remaining driving range for: in, This represents the actual remaining mileage of the vehicle. The remaining driving distance is predicted using the model, where n is the number of predictions made during the vehicle's journey. It is the average of n actual remaining driving distances.

9. An online lithium battery vehicle remaining driving range prediction system, characterized in that, include: The data acquisition module is used to collect vehicle data during a preset process and to analyze the Spearman coefficients between various vehicle data items. Select variables that are strongly correlated with the remaining driving mileage; The data cleaning module is used to clean the vehicle data corresponding to the strongly correlated variables by using the least squares regression algorithm to remove outliers. The GRU network output module is used to input the cleaned vehicle data into a bidirectional stacked gated cyclic network for selectively recording and discarding vehicle data during vehicle operation, and to mine temporal dependencies online. The KAN network output module is used to input the vehicle data and temporal dependencies into the Kolmogorov-Arnold network to achieve nonlinear mapping, extract intermediate sequence features of the vehicle data, and output the remaining driving mileage of the vehicle. The preset process is the process of the current vehicle battery state of charge (SOC) decreasing by 1%; the vehicle data includes battery status information and driving environment information. The battery status information includes voltage, current and battery SOC, and the driving environment information includes ambient temperature, total vehicle mileage and vehicle speed.

Images