Power battery health state estimation method, system and equipment based on end-cloud collaboration

Abstract

The invention discloses a power battery health state estimation method, system and device based on end-cloud collaboration, and relates to the technical field of power batteries, and the method comprises the steps: carrying out the preprocessing of the circulation aging data of a target power battery at a historical moment, carrying out feature extraction on the preprocessed circulating aging data at the historical moment by adopting a local linear embedding method, and inputting local linear embedding features at the historical moment into a cloud side battery health state estimation model to obtain an estimated value of the cloud side battery health state at the current moment; and updating model parameters of the end-side battery health state estimation model and an end-side battery health state estimation value at the current moment based on the cloud-side battery health state estimation value at the current moment and the end-side battery health state estimation model at the previous moment by adopting a double-unscented Kalman filter algorithm. And obtaining a final battery health state estimation value at the current moment. According to the invention, the estimation precision and real-time performance of the power battery health state are improved.

Description

Technical Field

[0001] This application relates to the field of power battery technology, and in particular to a method, system and device for estimating the health status of power batteries based on edge-cloud collaboration. Background Technology

[0002] As the primary power source for new energy vehicles, power batteries are crucial for ensuring vehicle safety and driving range. Lithium-ion batteries, due to their high energy density, long cycle life, and reasonable price, are currently the main choice for power batteries in new energy vehicles. To ensure the safety of lithium-ion battery systems, accurate and timely monitoring of the battery's State of Health (SOH) is essential. SOH reflects the battery's remaining lifespan and is directly related to its safety and performance, crucial for ensuring that lithium-ion batteries operate within a safe and efficient operating range. SOH estimation can be broadly categorized into direct measurement methods, model-based methods, and data-driven methods.

[0003] Direct measurement methods calculate battery capacity or test battery internal resistance by integrating ampere-hours under experimental conditions, thereby directly obtaining the battery's state of equilibrium (SOH). However, direct measurement methods can usually only be implemented in a laboratory under specific operating conditions and experiments, and are difficult to apply in dynamic, complex, and real-time-critical vehicle environments.

[0004] Model-based methods estimate state-of-the-art (SOH) by constructing equivalent circuit models and electrochemical mechanism models, exhibiting strong interpretability. However, the complex side reaction mechanisms and multi-field coupling characteristics within the battery make constructing high-precision models exceptionally difficult. Furthermore, while model-based methods can achieve high-precision SOH estimation under specific battery types and operating conditions, they struggle to achieve strong generalization of SOH estimation under different external conditions. This necessitates the design of complex parameter adjustment methods to adapt to varying operating conditions, temperatures, and material systems.

[0005] Data-driven approaches eliminate the reliance on complex battery mechanics. Using algorithms like machine learning and deep learning, they directly perform nonlinear fitting of the battery's State of Health (SOH) based on battery data and feature extraction, making it easier to achieve high-precision SOH estimation. Furthermore, external conditions such as temperature and operating conditions are directly represented as network input data, facilitating stronger generalization of SOH estimation under different external conditions. However, data-driven methods are entirely data-dependent, resulting in poor interpretability and extremely high requirements for data volume and quality. Additionally, data-driven models typically consume significant computational resources, making them difficult to implement in automotive BMS systems with limited computing power.

[0006] Therefore, it is necessary to provide a power battery health state estimation method based on edge-cloud collaboration to solve the above problems. Summary of the Invention

[0007] The purpose of this application is to provide a method, system, and device for estimating the health status of power batteries based on edge-cloud collaboration, so as to improve the estimation accuracy and real-time performance of the health status of power batteries.

[0008] To achieve the above objectives, this application provides the following solution:

[0009] Firstly, this application provides a method for estimating the health status of a power battery based on edge-cloud collaboration, the method comprising:

[0010] Obtain the cyclic aging data of the target power battery at historical moments; the target power battery is the power battery to be estimated.

[0011] Preprocess the cyclic aging data of the target power battery at historical time points to obtain the preprocessed cyclic aging data of the target power battery at historical time points.

[0012] The local linear embedding method is used to extract features from the preprocessed cyclic aging data of the target power battery at historical time points, so as to obtain the local linear embedding features of the target power battery at historical time points.

[0013] The local linear embedding features of the target power battery at historical moments are input into the cloud-side battery health status estimation model to obtain the estimated value of the cloud-side battery health status of the target power battery at the current moment; the cloud-side battery health status estimation model is obtained by training the Pyraformer model using the training set;

[0014] A dual unscented Kalman filter algorithm is employed. Based on the estimated cloud-side battery health status of the target power battery at the current moment and the estimated end-side battery health status of the target power battery at the previous moment, the model parameters of the end-side battery health status estimation model and the estimated end-side battery health status of the target power battery at the current moment are updated to obtain the final battery health status estimate of the target power battery at the current moment. The end-side battery health status estimation model of the target power battery at the previous moment is constructed based on the number of charge-discharge cycles of the target power battery at different stages of battery aging and the true values ​​of battery health status at historical moments.

[0015] In one embodiment, the cyclic aging data of the target power battery at historical time includes: voltage data of the target power battery at historical time and current data and time data of the target power battery at historical time.

[0016] The target power battery's historical cycle aging data is preprocessed to obtain preprocessed cycle aging data at historical time points, including:

[0017] Outliers are removed from the voltage and current data of the target power battery at historical time points to obtain the voltage and current data of the target power battery at historical time points after outlier removal.

[0018] At a fixed sampling interval, the voltage data and current data of the target power battery at historical time after removing outliers are resampled to obtain the preprocessed voltage data and preprocessed current data of the target power battery at historical time.

[0019] The pre-processed voltage data, pre-processed current data, and time data of the target power battery at historical time points are used as the pre-processed cycle aging data of the target power battery at historical time points.

[0020] In one embodiment, a local linear embedding method is used to extract features from the preprocessed cyclic aging data of the target power battery at historical time points, obtaining the local linear embedding features of the target power battery at historical time points, including:

[0021] Based on the preprocessed voltage data, preprocessed current data, and time data of the target power battery at historical time points, the incremental capacity of continuous voltage steps is calculated to obtain the IC curve of the target power battery at historical time points; the IC curve is plotted with voltage as the horizontal axis and incremental capacity as the vertical axis.

[0022] The local linear embedding method is used to extract features from the IC curves of the target power battery at historical time points, thereby obtaining the local linear embedding features of the target power battery at historical time points.

[0023] In one embodiment, the Pyraformer model includes: an input layer, a pyramid structure layer, a cross-scale attention layer, a multi-scale feature fusion layer, and an output layer;

[0024] The input layer is used to perform feature projection and position encoding on the local linear embedding features of the target power battery at historical time, so as to obtain the projection and encoded features of the target power battery at historical time.

[0025] The pyramid structure layer is used to downsample the projection and encoding features of the target power battery at historical moments to obtain a multi-scale coarse-grained sequence of the target power battery at historical moments.

[0026] The cross-scale attention layer is used to obtain the self-attention features of the target power battery at historical moments based on the multi-scale coarse-grained sequence of the target power battery at historical moments.

[0027] The multi-scale feature fusion layer is used to upsample and align, stitch features and compress and project the self-attention features of the target power battery at historical moments to obtain the fused features of the target power battery at historical moments.

[0028] The output layer is used to obtain an estimate of the cloud-side battery health status of the target power battery at the current moment based on the fusion characteristics of the target power battery at historical moments.

[0029] In one embodiment, the training process of the cloud-side battery health status estimation model specifically includes:

[0030] Construct a training set; the training set includes: the local linear embedding features of the sample power battery at historical moments and the true values ​​of the cloud-side battery health status of the sample power battery at the corresponding historical moments;

[0031] Build the Pyraformer model;

[0032] The Pyraformer model is trained by taking the local linear embedding features of the sample power battery at historical time as input and the estimated true value of the cloud-side battery health status of the sample power battery at historical time as output. Training is stopped when the loss function reaches the minimum value or the number of training rounds reaches the maximum value, thus obtaining the cloud-side battery health status estimation model.

[0033] In one embodiment, the different stages of battery aging include: an initial rapid aging stage, a linear aging stage, and a capacity-depleted accelerated aging stage.

[0034] The process of constructing the target power battery's end-side battery health state estimation model at the previous moment includes:

[0035] When the target power battery is in the initial rapid aging stage, the end-side battery health status estimation model of the target power battery at the previous moment is constructed based on the power law model, the charge and discharge cycle number of the target power battery, and the true value of the battery health status at historical moments.

[0036] When the target power battery is in the linear aging stage, the end-side battery health status estimation model of the target power battery at the previous moment is constructed based on the linear model, the number of charge-discharge cycles of the target power battery, and the true value of the battery health status at historical moments.

[0037] When the target power battery is in the stage of accelerated aging with a sharp drop in capacity, the end-side battery health status estimation model of the target power battery at the previous moment is constructed based on the exponential model, the number of charge-discharge cycles of the target power battery, and the actual values ​​of the battery health status at historical moments.

[0038] In one embodiment, when the target power battery is in the initial rapid aging stage, the expression for the end-side battery health state estimation model of the target power battery at the previous moment is:

[0039]

[0040] Among them, SOH r is the estimated value of the end-side battery health status of the target power battery corresponding to the current charge-discharge cycle number; r is the current charge-discharge cycle number; a1 and n1 are both model parameters of the end-side battery health status estimation model of the target power battery at the previous moment;

[0041] When the target power battery is in the linear aging stage, the expression for the end-side battery health state estimation model of the target power battery at the previous moment is:

[0042] SOH r =SOH(N1)-a2·(r-N1);

[0043] Where N1 is the number of charge-discharge cycles corresponding to the end of the initial rapid aging stage; SOH(N1) is the estimated final battery health status corresponding to the end of the initial rapid aging stage; a2 is the model parameter of the end-side battery health status estimation model of the target power battery at the previous moment.

[0044] When the target power battery is in the stage of accelerated aging and capacity reduction, the expression for the end-side battery health state estimation model of the target power battery at the previous moment is:

[0045] SOH r = a×exp(b×r)+c×exp(d×r);

[0046] Where a, b, c and d are model parameters of the end-side battery health state estimation model of the target power battery at the previous moment.

[0047] In one embodiment, a dual unscented Kalman filter algorithm is employed. Based on the estimated cloud-side battery health status of the target power battery at the current moment and the end-side battery health status estimation model of the target power battery at the previous moment, the model parameters of the end-side battery health status estimation model of the target power battery at the current moment and the estimated end-side battery health status of the target power battery at the current moment are updated to obtain the final battery health status estimate of the target power battery at the current moment, including:

[0048] When updating the model parameters of the target power battery's end-side battery health state estimation model at the current time:

[0049] The dual unscented Kalman filter algorithm is used to update the model parameters of the target power battery's end-side battery health status estimation model at the current time based on the estimated value of the target power battery's cloud-side battery health status at the current time and the model parameters of the target power battery's end-side battery health status estimation model at the previous time, so as to obtain the model parameters of the target power battery's end-side battery health status estimation model at the current time.

[0050] When updating the estimated value of the end-side battery health state of the target power battery at the current time using the end-side battery health state estimation model:

[0051] The dual unscented Kalman filter algorithm is used to update the estimated value of the target power battery's end-side battery health status at the current time based on the estimated value of the cloud-side battery health status at the current time and the model parameters of the estimated model of the target power battery's end-side battery health status at the current time, so as to obtain the estimated value of the target power battery's end-side battery health status at the current time.

[0052] Secondly, this application provides a power battery health state estimation system based on edge-cloud collaboration. The edge-cloud collaboration-based power battery health state estimation system is used to implement the aforementioned edge-cloud collaboration-based power battery health state estimation method. The edge-cloud collaboration-based power battery health state estimation system includes:

[0053] The data acquisition unit is used to acquire the cyclic aging data of the target power battery at historical moments; the target power battery is the power battery to be estimated.

[0054] The preprocessing unit is used to preprocess the cycle aging data of the target power battery at historical time points to obtain the preprocessed cycle aging data of the target power battery at historical time points.

[0055] The feature extraction unit is used to extract features from the preprocessed cyclic aging data of the target power battery at historical time using the local linear embedding method, so as to obtain the local linear embedding features of the target power battery at historical time.

[0056] The cloud-side battery health status estimation unit is used to input the local linear embedding features of the target power battery at historical moments into the cloud-side battery health status estimation model to obtain the estimated value of the cloud-side battery health status of the target power battery at the current moment; the cloud-side battery health status estimation model is obtained by training the Pyraformer model using the training set;

[0057] The edge-cloud collaborative battery health state estimation unit uses a dual unscented Kalman filter algorithm to update the model parameters and estimated values ​​of the target power battery's end-side battery health state estimation model at the current moment, based on the cloud-side battery health state estimate of the target power battery at the current moment and the edge-side battery health state estimation model of the target power battery at the previous moment, to obtain the final battery health state estimate of the target power battery at the current moment. The edge-side battery health state estimation model of the target power battery at the previous moment was constructed based on the number of charge-discharge cycles of the target power battery at different stages of battery aging and the true values ​​of the battery health state at historical moments.

[0058] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for estimating the health status of a power battery based on edge-cloud collaboration.

[0059] According to the specific embodiments provided in this application, this application has the following technical effects:

[0060] This application discloses a power battery state of health estimation method, system, and device based on end-cloud collaboration. By deploying a cloud-side battery state of health estimation model with high estimation accuracy but high computational cost on the cloud side, and deploying an end-side battery state of health estimation model with high real-time performance and low computational cost on the end side, the cloud-side battery state of health estimation model and the end-side battery state of health estimation model are then fused on the end side based on the end-cloud collaboration algorithm, thereby achieving high-precision and high-real-time SOH estimation, improving the estimation accuracy and real-time performance of SOH. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 A schematic diagram of the process for estimating the health status of a power battery based on edge-cloud collaboration, provided in an embodiment of this application;

[0063] Figure 2 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0064] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0065] The purpose of this application is to resolve the contradiction between estimation accuracy and real-time performance when estimating the state of health (SOH) of a battery. High-precision SOH estimation is typically based on complex models; however, the computationally limited resources of the vehicle environment cannot withstand the high computational costs, severely impacting the real-time performance of the SOH estimation. Conversely, simple models or empirical methods with low computational cost are often insufficient for achieving high-precision SOH estimation. The main objective of this application is to provide a power battery SOH estimation method that is highly practical, accurate, and real-time. This method can be used to estimate SOH in vehicle battery management systems and has broad application prospects in new energy vehicles.

[0066] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, this application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0067] In one exemplary embodiment, such as Figure 1 As shown, a method for estimating the health status of a power battery based on edge-cloud collaboration is provided, including the following steps: Wherein:

[0068] Step S1: Obtain the cyclic aging data of the target power battery at historical moments; the target power battery is the power battery to be estimated.

[0069] Specifically, the cyclic aging data of power batteries at historical moments are obtained through various means, including data from battery manufacturers, data stored on cloud platforms, data collected by the end-side BMS, and publicly available datasets.

[0070] Taking a four-cell lithium iron phosphate (LFP) battery as the target power battery as an example, the process of obtaining the battery cycle aging data is as follows: Testing was conducted using a horizontal cylindrical fixture on an Arbin LBT potentiostat. Charge-discharge aging cycles were performed in a forced convection temperature chamber set to 30°C. The battery's rated capacity was 1.1 Ah, and its nominal voltage was 3.3V. During charging, the battery was initially charged at a 5C rate. When the state of charge (SOC) reached 20%, the rate was increased to 4C until the SOC reached 80%. After a 5-second rest period, constant current and constant voltage charging was performed at a 1C rate. During discharging, a constant current discharge was performed at a 4C rate until the battery reached its lower cutoff voltage, followed by a 5-second rest period. The cycle aging experiment was terminated when the battery's state of harm (SOH) reached 80%. The final obtained cycle aging data includes the test time, current, voltage, temperature, and SOH label data for the four-cell cycle aging experiment.

[0071] Step S2: Preprocess the cyclic aging data of the target power battery at historical time points to obtain the preprocessed cyclic aging data of the target power battery at historical time points.

[0072] As an optional implementation, in step S2, the cyclic aging data of the target power battery at historical times includes: voltage data of the target power battery at historical times, current data of the target power battery at historical times, and time data. Therefore, step S2 includes:

[0073] Step S21: Remove outliers from the voltage data and current data of the target power battery at historical time points to obtain the voltage data and current data of the target power battery at historical time points after outlier removal.

[0074] Step S22: At a fixed sampling interval, resample the voltage data and current data of the target power battery at historical time after removing outliers to obtain the preprocessed voltage data and current data of the target power battery at historical time.

[0075] Step S23: The preprocessed voltage data, preprocessed current data, and time data of the target power battery at historical time are used as the preprocessed cycle aging data of the target power battery at historical time.

[0076] Step S3: Using the local linear embedding method, feature extraction is performed on the preprocessed cyclic aging data of the target power battery at historical time points to obtain the local linear embedding features of the target power battery at historical time points.

[0077] As an optional implementation, step S3 includes:

[0078] Step S31: Based on the preprocessed voltage data, preprocessed current data, and time data of the target power battery at historical time, calculate the incremental capacity of continuous voltage steps to obtain the IC curve of the target power battery at historical time; the IC curve is plotted with voltage as the horizontal axis and incremental capacity as the vertical axis.

[0079] Specifically, an incremental capacity (IC) curve is generated based on the incremental capacity at continuous voltage steps in lithium battery cycle aging data. In the IC curve, the voltage plateau is transformed into a clearly visible peak, which has a strong correlation with the microscopic phase transitions during battery aging. The IC curve is used to characterize battery aging. The formula for calculating the incremental capacity in the IC curve is as follows:

[0080]

[0081] Where dQ0 is the incremental capacity, dV is the voltage step size, I is the current, and dt is the time interval.

[0082] Step S32: The Locally Linear Embedding (LLE) method is used to extract features from the IC curves of the target power battery at historical time points to obtain the locally linearly embedded features of the target power battery at historical time points.

[0083] Specifically, the LLE method is used to reduce the dimensionality of IC data for feature extraction. By maintaining the linear relationship of data in the local neighborhood, high-dimensional data is mapped to a low-dimensional space. Since its weight calculation is based on neighboring data, the influence of noise is effectively suppressed by neighborhood smoothing, and it has good noise robustness, which can greatly reduce the cumbersome data processing process. At the same time, compared with analyzing the peak value and peak position changes of IC curve during battery aging, the LLE method can make full use of the information of the entire IC curve.

[0084] The principle of feature extraction using the LLE method is as follows:

[0085] 1) Treat the IC data in each loop as a set of high-dimensional data points, determine the size of its local neighborhood for each high-dimensional data point, and represent it as a linear combination of neighborhood points. Solve for the weights by minimizing the reconstruction error, and represent the IC data as high-dimensional data g. m :

[0086]

[0087] ∑ j∈N(m) W mj=1 (4)

[0088] Among them, g m For the m-th high-dimensional data point, dV m Let g be the voltage step size for the m-th high-dimensional data point; j Let m represent the index of the high-dimensional data point, and j represent the index of the neighboring points of the high-dimensional data point; N is the neighborhood size, and W is the index of the neighboring points of the high-dimensional data point. mj This is the weight matrix.

[0089] 2) The above-mentioned weight problem is solved in the following way:

[0090] First, calculate the local covariance matrix, where each element of the matrix is ​​calculated as follows:

[0091]

[0092] in, For each element in the covariance matrix corresponding to the m-th high-dimensional data point, It is the j1-th neighboring point; Let j1 be the j2th neighbor point, and let j2 be the position of the element in the matrix.

[0093] Next, the constrained optimization problem is solved using the Lagrange multiplier method:

[0094]

[0095] Among them, C m Let be the covariance matrix corresponding to the m-th high-dimensional data point, k0 be the index of the element in the matrix, and p and q be the element indices in the Lagrange multiplier method, respectively. Finally, the weight matrix W for each high-dimensional data point is obtained. ij .

[0096] Then, in the low-dimensional space, the weight matrix is ​​input, keeping the weights unchanged, and the low-dimensional coordinates are solved. The objective is:

[0097]

[0098] Among them, h m h represents the target's low-dimensional coordinates. j Given the low-dimensional coordinates of the j-th neighboring point, the solution to this optimization problem is as follows:

[0099] Construct the matrix:

[0100]

[0101] Where M is the constructed matrix and I is the identity matrix.

[0102] Perform eigenvalue decomposition:

[0103] Mv=λv (9)

[0104] Where λ is the eigenvalue and v is the eigenvector.

[0105] The feature vector corresponding to the smallest non-zero feature value in the target dimension is taken as the final extracted feature Y (i.e., local linear embedding feature):

[0106] Y = [v2, v3, ..., v d+1 (10)

[0107] Where d is the target dimension, v d+1 Let be the eigenvector corresponding to the smallest non-zero eigenvalue of the target dimension.

[0108] Step S4: Input the local linear embedding features of the target power battery at historical time into the cloud-side battery health status estimation model to obtain the estimated value of the cloud-side battery health status of the target power battery at the current time; the cloud-side battery health status estimation model is obtained by training the Pyraformer model using the training set.

[0109] As an optional implementation, in step S4, the Pyraformer model includes: an input layer, a pyramid structure layer, a cross-scale attention layer, a multi-scale feature fusion layer, and an output layer.

[0110] The input layer is used to perform feature projection and position encoding on the local linear embedding features of the target power battery at historical time, so as to obtain the projection and encoded features of the target power battery at historical time.

[0111] Specifically, before feature projection, the local linear embedding features of the target power battery at historical moments are integrated into a time series format, i.e., a three-dimensional tensor format is constructed. Specifically, a sliding window is used to truncate the local linear embedding features according to a certain step size. The length of the sliding window is the length of the time series. Each truncated two-dimensional matrix is ​​a time series. The obtained two-dimensional slices are constructed into a three-dimensional tensor according to a certain batch size. A three-dimensional tensor is regarded as a batch. During training, training is performed on a batch basis, thereby effectively improving the training efficiency of the network.

[0112] Feature projection is performed on the locally linearly embedded features in the input time-series format, and their positions are encoded. Feature projection is based on a linear transformation:

[0113] In e =Y·W e +b e (11)

[0114] Among them, In eFor the projected feature projection; W e and b e These represent the weights and biases of the input layer, respectively; e refers to the variables in this formula, which are all variables in the input layer processing.

[0115] Positional coding uses trigonometric functions to inject positional information into the time series, generating a unique code for each vector at each position. This code is then added to the input time series data. The trigonometric function positional coding can be described as follows:

[0116]

[0117] Among them, PE (pos,2z) Position encoding for the even-numbered position; PE (pos,2z+1) The position code for the odd-numbered position; POS represents the position, d model Let z represent the feature dimension and z represent the index of the data point. The final output H is obtained. e (That is, the projection and coding characteristics of the target power battery at a historical moment) are:

[0118]

[0119] Where PE is the position code and L is the sequence length.

[0120] The pyramid structure layer is used to downsample the projection and encoding features of the target power battery at historical moments, so as to obtain a multi-scale coarse-grained sequence of the target power battery at historical moments.

[0121] Specifically, a pyramid structure layer is used to divide the final output H of the input layer. e Progressive downsampling is performed to generate multi-scale coarse-grained sequences, forming a hierarchical pyramid structure. This reduces sequence resolution while preserving key temporal features, providing multi-granular input for subsequent cross-scale attention. The hierarchical downsampling operation generates coarse-grained sequences through sliding window aggregation, with each layer compressing the output of the previous layer. Specifically:

[0122] First, the final output H of the input layer e The input is fed into the pyramid structure layer, and the entire input sequence is divided into multiple segments according to a certain time series length to obtain the segmented sequence:

[0123]

[0124] in, For the segmented sequence, z f Here are the indices of the split sequence, and `split` represents the splitting operation. Then, average pooling downsampling is performed to obtain the sampled sequence:

[0125]

[0126] in, The sampled sequence, is the sequence to be sampled, s is the downsampling factor, representing the reduction in length of each layer; q0 is the sequence position index in the current layer; k is the pyramid level index.

[0127] The output of this layer is obtained through linear projection:

[0128]

[0129] Among them, H (k) W is the output of the pyramid structure layer obtained after projection. (k) and b (k) These are the weights and the biases, respectively.

[0130] The cross-scale attention layer is used to obtain the self-attention features of the target power battery at historical moments based on the multi-scale coarse-grained sequence of the target power battery at historical moments.

[0131] Specifically, in the cross-scale attention layer, a multi-scale attention mechanism is applied to achieve cross-level information exchange, enabling the model to flexibly handle data relationships across different time scales. Nodes only connect to local neighbors within the same and adjacent layers, resulting in lower computational complexity. Data undergoes self-attention mechanism calculation via a pyramid attention mechanism:

[0132]

[0133] Q = H (k) ·W Q (19)

[0134] Where Q, K, and V are the query matrix, key matrix, and value matrix, respectively, and d k For the key dimension, W Q Let Q be the weight matrix, and let Q represent the variables in the cross-scale attention layer.

[0135] Obtain the final output Z of this layer (k) (i.e., the self-attention characteristics of the target power battery at historical moments):

[0136] Z (k) =LayerNorm(Attention(Q,K,V)+H (k) (20)

[0137] LayerNorm represents layer normalization.

[0138] The multi-scale feature fusion layer is used to upsample and align, stitch together and compress and project the self-attention features of the target power battery at historical moments to obtain the fused features of the target power battery at historical moments.

[0139] Specifically, in the multi-scale feature fusion layer, feature representations from all scales are integrated to generate the final temporal code, as follows:

[0140] Upsampling alignment is performed using linear interpolation:

[0141]

[0142] Feature splicing:

[0143]

[0144] Compressed projection:

[0145] H out =H fuse ·W f +b f (twenty three)

[0146] Interpolate is a linear interpolation operation. The result is after linear interpolation. Conca represents the feature concatenation operation, and H... fuse H is the result after feature concatenation. out The result is after compressed projection, where k is the pyramid level index and W... f As the weight, b f As a bias, f indicates that the variables here are all parameters in the multi-scale fusion layer.

[0147] The output layer is used to obtain an estimate of the cloud-side battery health status of the target power battery at the current moment based on the fusion characteristics of the target power battery at historical moments.

[0148] Specifically, the output layer consists of a forward propagation network and a fully connected layer. The SOH estimation result is output by the final fully connected layer after passing through the forward propagation network.

[0149] Pyraformer t+1 =σ(W o ·H out +b o ) (twenty four)

[0150] Where σ is the activation function, typically the Sigmoid function is chosen in fully connected layers, and W... o and b o These represent the network weights and biases, respectively; 'o' indicates that the variable here is a variable from the output layer; Pyraformer t+1This is the estimated State of Health (SOH) of the cloud-side battery under iteration t+1. The estimation result will be compared with the true SOH label in the test set to evaluate the estimation accuracy of the cloud-side deep learning model.

[0151] Specifically, the cloud-side battery health status estimation model is built around the Pyraformer model, which better captures the global characteristics of long-sequence problems. Compared to traditional long-sequence models like the Transformer, it processes time-series data more efficiently. Its core mechanism is the pyramid attention mechanism, which introduces a coarse-scale construction module to progressively downsample the original time series into multi-scale coarse-grained sequences, forming a hierarchical pyramid structure. Then, through a self-attention mechanism, cross-level information exchange is achieved within the pyramid structure. This hierarchical structure captures the multi-scale dependencies of the time series, significantly reducing computational complexity and improving long-term prediction performance. Pyraformer reduces computation through hierarchical sparse connections; nodes are only connected to local neighbors within the same and adjacent layers, resulting in lower computational complexity while more effectively capturing long-term dependencies.

[0152] The cloud-side battery health status estimation model is built based on a deep learning model. While deep learning models offer high estimation accuracy, they are computationally expensive. Therefore, the model is deployed on the cloud for computation. The extracted features are used to construct a time series format, which is then input into a Pyraformer model for training and testing. The training process of the model is based on an optimization algorithm for automatic hyperparameter optimization. The optimization algorithm used is the Bayesian optimization algorithm, which uses a Gaussian process to model the objective function and finds the global optimum by iteratively updating the model.

[0153] As an optional implementation, step S4, the training process of the cloud-side battery health status estimation model, specifically includes:

[0154] Step S41, construct the training set; the training set includes: the local linear embedding features of the sample power battery at historical moments and the true values ​​of the cloud-side battery health status of the sample power battery at the corresponding historical moments.

[0155] The cloud-side battery health status estimation model is designed to estimate the state of health (SOH) at a future point in time based on historical or current data, thereby avoiding the impact of transmission latency on both the edge and cloud sides.

[0156] Step S42: Build the Pyraformer model.

[0157] Step S43: Using the local linear embedding features of the sample power battery at historical time as input and the estimated true value of the cloud-side battery health status of the sample power battery at historical time as output, train the Pyraformer model until the loss function reaches the minimum value or the training rounds reach the maximum value, then stop training to obtain the cloud-side battery health status estimation model.

[0158] Step S5: Using the Dual Unscented Kalman Filter (Dual-UKF) algorithm, based on the estimated cloud-side battery health status of the target power battery at the current moment and the estimated end-side battery health status of the target power battery at the previous moment, the model parameters of the estimated end-side battery health status of the target power battery at the current moment and the estimated end-side battery health status of the target power battery at the current moment are updated to obtain the final estimated battery health status of the target power battery at the current moment. The estimated end-side battery health status of the target power battery at the previous moment is constructed based on the number of charge-discharge cycles of the target power battery at different stages of battery aging and the true values ​​of battery health status at historical moments.

[0159] Specifically, high-precision and real-time SOH estimation is achieved through collaboration between the edge and cloud sides. The edge-cloud collaborative algorithm employs the Dual-UKF algorithm. KF is a filter used for linear dynamic system models, employing efficient recursive filtering based on Bayesian filtering algorithms to observe the system state from noisy measurements. UKF is a variant of KF, using unscented transformation to adapt the nonlinear system function to a standard Kalman system under linear assumptions. Dual-UKF is a nonlinear state estimation algorithm based on a dual estimation framework, capable of jointly estimating the system state and parameters, thereby decoupling the coupling relationship between state and parameters, achieving high-precision and robust nonlinear system modeling, and suitable for complex scenarios with time-varying parameters. In the alternating updates of Dual-UKF, the output of the state estimator serves as the input to the parameter estimator, and vice versa, forming a closed-loop optimization. The joint covariance online correction mechanism can calculate the joint uncertainty of state and parameters in real time using the Jacobian matrix.

[0160] An edge-cloud collaborative algorithm and an edge-side battery health state estimation model are deployed on the edge side. The edge-side battery health state estimation model has lower accuracy but lower computational cost and higher real-time performance, so it is deployed on the edge side for computation. The edge-cloud collaborative algorithm is deployed on the edge side, combining the models and estimation results from both the edge and cloud sides to perform online joint estimation of the system's state and parameters. This decouples the state and parameters, achieving high-precision and robust nonlinear system modeling, and realizing SOH estimation that balances high accuracy and high real-time performance. Finally, based on the characteristics and limitations of both the edge and cloud sides, the computational forms of the models on both sides are specifically designed.

[0161] The computational form of the edge-cloud model refers to the calculation of the cloud-side battery health state estimation model, which estimates the SOH at a future point in time based on current and historical data. This avoids the impact of transmission latency on both the edge and cloud sides. At the same time, during model training, the data is resampled at sparser time intervals to adapt to the different data sparsity on both the edge and cloud sides. The edge-side battery health state estimation model is discretized into the form of state equations to adapt to the edge-side computing environment, and high-real-time iterative calculations are performed on the edge based on real-time collected data.

[0162] As an optional implementation, in step S5, the different stages of battery aging include: the initial rapid aging stage, the linear aging stage, and the capacity-depleted accelerated aging stage.

[0163] Specifically, the on-device battery health state estimation model is built based on a segmented empirical model. This model effectively reflects the different aging characteristics of the battery at different stages of aging. The SOH estimation result is obtained by fitting the SOH and charge-discharge cycle number. This model has low computational cost and is mounted on the on-device side for high real-time SOH estimation. The model parameters will be updated online based on an on-device cloud collaborative algorithm, and the initial values ​​of the model parameters are fitted based on a large amount of experimental data.

[0164] Based on the different characteristics of different stages of battery aging, the battery aging process can be divided into three stages: Stage 1: Initial rapid aging stage; Stage 2: Linear aging stage; Stage 3: Accelerated aging stage with capacity drop. Therefore, a segmented model is used to construct an empirical model, including a power-law model, a linear model, and an exponential model.

[0165] The aging process of the battery is divided into intervals based on the number of charge-discharge cycles, into three stages: [0, N1], [N1, N2], and [N2, N3]. Based on the different material systems of the battery, for ternary material system batteries, the first segment point is selected when the state of oxygen (SOH) is about 92%, and the second segment point is selected when the SOH is about 80%. For lithium iron phosphate material system batteries, the first segment point is selected when the SOH is about 90%, and the second segment point is selected when the SOH is about 85%.

[0166] The process of constructing the target power battery's end-side battery health state estimation model at the previous moment includes:

[0167] When the target power battery is in the initial rapid aging stage, the end-side battery health status estimation model of the target power battery at the previous moment is constructed based on the power law model, the number of charge-discharge cycles of the target power battery, and the actual values ​​of the battery health status at historical moments.

[0168] When the target power battery is in the initial rapid aging stage, the expression for the end-side battery health state estimation model of the target power battery at the previous moment is:

[0169]

[0170] Among them, SOH r is the estimated value of the end-side battery health status of the target power battery under the current charge-discharge cycle number; r is the current charge-discharge cycle number; a1 and n1 are model parameters of the end-side battery health status estimation model of the target power battery at the previous moment, and their physical meaning is the aging intensity and rate in the early stage of battery aging.

[0171] The corresponding discretization form is:

[0172]

[0173] Among them, SOH r-1 This is an estimate of the terminal battery health status of the target power battery corresponding to the previous charge-discharge cycle number.

[0174] When the target power battery is in the linear aging stage, the end-side battery health status estimation model of the target power battery at the previous moment is constructed based on the linear model, the number of charge-discharge cycles of the target power battery, and the actual values ​​of the battery health status at historical moments.

[0175] When the target power battery is in the linear aging stage, the expression for the end-side battery health state estimation model of the target power battery at the previous moment is:

[0176] SOH r =SOH(N1)-a2·(r-N1) (27)

[0177] Where N1 is the number of charge-discharge cycles corresponding to the end point of the initial rapid aging stage (also known as the first segment point); SOH(N1) is the final SOH estimate of the entire system at the first segment point; a2 is the model parameter of the end-side battery health state estimation model of the target power battery at the previous moment, and its physical meaning is the mid-term linear decay slope, that is, the fixed reduction of SOH per cycle.

[0178] The corresponding discretization form is:

[0179] SOH r =SOH r-1 -a2 (28)

[0180] When the target power battery is in the stage of accelerated aging with a sharp drop in capacity, the end-side battery health status estimation model of the target power battery at the previous moment is constructed based on the exponential model, the number of charge-discharge cycles of the target power battery, and the actual values ​​of the battery health status at historical moments.

[0181] When the target power battery is in the stage of accelerated aging and capacity reduction, the expression for the end-side battery health state estimation model of the target power battery at the previous moment is:

[0182] SOH r = a×exp(b×r)+c×exp(d×r) (29)

[0183] Wherein, a, b, c, and d are model parameters of the end-side battery health state estimation model of the target power battery at the previous moment. Parameter a represents the health status of the battery in the initial state; parameter c represents the battery's degradation trend; parameters b and d represent the initial capacity degradation rate and the accelerated aging stage degradation rate of the battery, respectively.

[0184] The corresponding discretization form is:

[0185] SOH r =SOH r-1 +ab×e (r-1)×b +cd×e (r-1)×d (30)

[0186] As an optional implementation, step S5 includes:

[0187] Step S51, when updating the model parameters of the target power battery's end-side battery health state estimation model at the current time (i.e., parameter update):

[0188] The dual unscented Kalman filter algorithm is used to update the model parameters of the target power battery's end-side battery health status estimation model at the current time, based on the estimated value of the target power battery's cloud-side battery health status at the current time and the model parameters of the target power battery's end-side battery health status estimation model at the previous time. This yields the model parameters of the target power battery's end-side battery health status estimation model at the current time.

[0189] Specifically, when performing edge-cloud collaborative estimation, the edge and cloud sides are first constructed as a whole system. Based on state-space theory, the state-space equations of the system are jointly constructed based on the estimated value of the cloud-side battery health state and the edge-side battery health state estimation model. When updating the state (i.e., the SOH estimate) and parameters (i.e., the model parameters of the edge-side battery health state estimation model), two different state-space equations are constructed depending on the object being updated in the system.

[0190] When updating parameters, the model parameters of the terminal battery health state estimation model are iteratively updated based on the cloud-side battery health state estimate. The final SOH estimate of the system at the previous time step is treated as a state variable and taken as a known quantity. The model parameters of the terminal battery health state estimation model are used as estimates, and the cloud-side battery health state estimate is treated as a measurement value. The state-space equations of the system are constructed, where the state equations for different stages of battery aging are as follows:

[0191] Early rapid aging stage:

[0192] a1(t+1)=a1(t) (31)

[0193] n1(t+1)=n1(t) (32)

[0194] Linear aging stage:

[0195] a2(t+1)=a2(t) (33)

[0196] Capacity reduction accelerates aging phase:

[0197] A(t+1)=A(t) (34)

[0198] B(t+1)=B(t) (35)

[0199] C(t+1)=C(t) (36)

[0200] D(t+1)=D(t) (37)

[0201] Where t represents the number of iterations, and A, B, C, and D are the model parameters to be updated in the end-side battery health state estimation model. The time intervals and iteration numbers described in this application are in one-to-one correspondence; that is, the current time interval corresponds to the current iteration number, the previous time interval corresponds to the previous iteration number, the next time interval corresponds to the next iteration number, and so on. Further details will not be elaborated here.

[0202] The measurement equations for different stages of battery aging are as follows:

[0203] Early rapid aging stage:

[0204]

[0205] Linear aging stage:

[0206] Pyraformer t+1 =SOH(N1)-[a2(t+1)]×(t+1-N1) (39)

[0207] Capacity reduction accelerates aging phase:

[0208] Pyraformer t+1 =[A(t+1)]×e (t+1)×B(t+1) +[C(t+1)]×e (t+1)×D(t+1) (40)

[0209] Step S52, when updating the estimated value of the end-side battery health state of the target power battery at the current time (i.e., state update):

[0210] The dual unscented Kalman filter algorithm is used to update the estimated value of the target power battery's end-side battery health status at the current time based on the estimated value of the cloud-side battery health status at the current time and the model parameters of the estimated model of the target power battery's end-side battery health status at the current time, so as to obtain the estimated value of the target power battery's end-side battery health status at the current time.

[0211] Specifically, during state updates, the high-precision estimation results of the cloud-side battery health state estimation model (i.e., the estimated value of the cloud-side battery health state) and the high-real-time estimation results of the end-side battery health state estimation model are fused. The model parameters of the end-side battery health state estimation model at the current moment (i.e., the model parameters after parameter updates) of the target power battery are used as known quantities, and the end-side state (estimated value of SOH) is used as an estimate. The estimation results of the end-side battery health state estimation model are regarded as estimated values, and the estimated value of the cloud-side battery health state is regarded as measured values. The state-space equations of the system are constructed, and the state equations for different stages of battery aging are as follows:

[0212] Early rapid aging stage:

[0213]

[0214] Linear aging stage:

[0215] SOH t+1 =SOH t -a2(t+1) (42)

[0216] Capacity reduction accelerates aging phase:

[0217] SOH t+1 =SOH t +[A(t+1)]×[B(t+1)]×e [B(t+1)]×(t+1) +[C(t+1)]×[D(t+1)]×e [D(t +1)]×(t+1) (43)

[0218] The measurement equations for all three stages are:

[0219] Pyraformer t+1=SOH t+1 +v0 (44)

[0220] Where v0 represents the measurement noise.

[0221] The state-space equations constructed in the above steps are substituted into UKF for calculation, thereby fusing the estimation results from both the edge and cloud sides to achieve state update (i.e., update of the SOH estimation result), and obtain a higher-precision SOH estimation result (i.e., the final battery health state estimate of the target power battery at the current moment) that integrates the estimation results from both the edge and cloud models.

[0222] Furthermore, the above content describes the construction of two state-space equations when performing state updates and parameter updates, respectively. Next, the edge-cloud collaborative estimation process based on Dual-UKF will be described in detail.

[0223] First, according to state-space theory, the state-space equations constructed above can all be expressed in the following standard form:

[0224] x t =f(x) t-1 ,u t θ t )+w t (45)

[0225] y t =h(x t ,u t θ t )+v t (46)

[0226] Where, x t f(x) is the estimated state of health of the end-side battery at the current iteration number. t-1 ,u t θ t ) and h(x t ,u t θ t Let x be the function mapping function of the actual state-space equation. t-1 This is the final battery health state estimate under the previous iteration number; u t θ represents the system input at the current iteration number; θ represents the model parameter values ​​of the end-side battery health state estimation model at the current iteration number; w t The process noise at the current iteration number; v t This represents the observation noise at the current loop iteration number.

[0227] Based on the standard form of the above state-space equation, the edge-cloud collaborative process based on Dual-UKF is explained (i is the index of the Sigma point, θ refers to the variable used when updating the model parameters of the edge battery health state estimation model, and x refers to the variable used when updating the state. This will not be elaborated further).

[0228] The parameters of the end-side battery health state estimation model are updated, and the prior estimate of the SOH state is fixed. At this time, the state space equation is the state space equation constructed during the parameter update (Equation (31)-Equation (40). The same stage is paired with each other, and each parameter can construct a set of state space equations. Then, after the constructed state space equations are expressed in standard form, the Sigma points of the model parameter space of the end-side battery health state estimation model are generated based on the model parameters of the end-side battery health state estimation model at the previous time:

[0229]

[0230] in, Let Σ be the i-th Sigma point corresponding to the model parameters of the end-side battery health state estimation model. n represents the estimated values ​​of the model parameters for the terminal battery health state estimation model at the previous time step. θ λ is the dimension of the model parameters, λ is the scaling factor, and P is the model parameter dimension. θ,t-1 This is the model parameter estimation error covariance of the battery health status estimation model at the previous time step.

[0231] Based on prior estimates of the SOH state, observational predictions are made:

[0232]

[0233] in, For the observed predicted value, This is a priori estimation of the SOH state.

[0234] Then, the following statistics are calculated for updating the model parameters of the subsequent end-side battery health state estimation model:

[0235]

[0236] in, Let be the mean weight of the i-th Sigma point, where mean indicates that the weight is the mean weight. The mean weight of the first Sigma point; Let be the covariance weight of the i-th Sigma point. Here is the covariance weight for the first Sigma point, where c indicates that the weight is a covariance weight. P is the weighted observation prediction; yy,tThe covariance is the observation-predicted covariance; yy indicates that this covariance is the observation-predicted covariance; P θy,t Let θy be the parameter-observation cross-covariance; θy represents that the covariance is the parameter-observation cross-covariance; R is the observation noise covariance; T is the transpose of the matrix; α is an adjustable parameter used to compensate for higher-order moment errors; β is an adjustable parameter used to introduce prior knowledge of the state distribution (such as Gaussian distribution).

[0237] Finally, the model parameters of the end-side battery health state estimation model are updated:

[0238]

[0239] in, The Kalman gain of the model parameters for the end-side battery health state estimation model. The updated model parameters are estimated, y t The estimated modulus of the cloud-side battery health status, i.e., Pyraformer. t+1 P θ,t The covariance of the model parameters for the updated end-side battery health state estimation model.

[0240] A state update is performed, where "state" refers to SOH. The state is estimated using the updated model parameters of the end-side battery health state estimation model. At this time, the state space equation is the state space equation constructed during the state update (Equation (41)-Equation (44), with those at the same stage paired up. Based on the estimated value of the overall battery health state at the previous moment, the state space Sigma point is generated:

[0241]

[0242] in, Let Sigma be the i-th state. For state prior estimation, n x For the state dimension, P x,t∣t-1 Let be the prior state covariance of the system.

[0243] Then, the model parameters of the injected end-side battery health state estimation model are updated to perform state prediction and observation prediction:

[0244]

[0245] in, This is the predicted state value.

[0246] Then, the statistics are calculated for state updates:

[0247]

[0248] in, P is the weighted mean of state predictions. xx,t Let xx represent the state prediction covariance; P xy,t Let xy represent the state-observation cross-covariance; Q x Let be the state noise covariance.

[0249] Finally, the state is updated using the statistical values ​​obtained above:

[0250]

[0251] in, For state Kalman gain, For the updated state estimate, P x,t This represents the updated state covariance.

[0252] After parameter and state updates are performed, a joint covariance correction is performed to compensate for the state covariance, resulting in the state-parameter joint covariance:

[0253]

[0254] Among them, P xθ,t For the joint covariance of state and parameters; F θ P is the Jacobian matrix of the state function with respect to the parameters. θ,t Q represents the covariance from the model parameter update process; xθ This refers to noise in the state-parameter process.

[0255] Finally, when validating the power battery health state estimation method based on edge-cloud collaboration in this application, cross-validation was used. Specifically, data from each battery in the experimental dataset was used as the test set, and data from the remaining batteries were used as the training set. The training set was used to train the initial model parameters for both the cloud-side and edge-side battery health state estimation models, while the test set was used to validate the model's performance. The final SOH estimated by the model was compared with the true SOH labels in the test set to evaluate the model's accuracy.

[0256] Technical effects:

[0257] 1) This application deploys a cloud-side battery health state estimation model with high estimation accuracy but high computational cost on the cloud side, and a terminal-side battery health state estimation model with high real-time performance and low computational cost on the terminal side. Then, on the terminal side, the cloud-side battery health state estimation model and the terminal-side battery health state estimation model are fused based on the terminal-cloud collaborative algorithm, thereby achieving high-precision and high-real-time SOH estimation. This solves the contradiction between estimation accuracy and real-time performance caused by the limited computing power of the vehicle BMS in practical applications.

[0258] 2) This application constructs a high-precision cloud-side battery health status estimation model based on the Pyraformer model. The Pyraformer model captures the multi-scale dependencies of time series through a hierarchical sparse connection structure, which significantly reduces computational complexity while improving long-term prediction performance. It solves the estimation accuracy problem of long-term dependency and the problem that traditional deep learning models have difficulty in grasping global trends. At the same time, it solves the problems of complex computation and high consumption of computing resources caused by long series, significantly improves model efficiency and practical application capability, and solves the problem of high efficiency in long-term time series prediction.

[0259] 3) The input features of the cloud-side battery health status estimation model in this application are extracted based on the IC curve and the LLE method, which makes full use of the information of the entire curve. At the same time, the LLE method has strong robustness to noise and can obtain high-quality features that are strongly correlated with battery aging based on a simpler data processing flow. This solves the problems of difficulty in making full use of information when extracting features based on traditional signal analysis methods, as well as the lengthy data processing flow and poor robustness caused by obtaining high-quality features.

[0260] 4) This application designs the calculation of the cloud-side battery health state estimation model as a form of future SOH estimation based on current data and historical data. During model training, the data is resampled at sparser intervals. On the edge side, a high real-time edge-side battery health state estimation model is constructed based on a segmented empirical model, and it is discretized into an iterative calculation form for calculation in the vehicle BMS. This solves the transmission delay problem on both the edge and cloud sides and the timing asynchrony problem caused by the different data sparsity.

[0261] 5) The end-side battery health status estimation model of this application is based on a segmented empirical model. The segmented model can effectively reflect the different aging characteristics of the battery at different stages of aging, solving the problem that due to the complexity of the internal aging mechanism of the battery, a single model is difficult to reflect the physical characteristics of the battery aging process throughout the entire process.

[0262] 6) This application constructs a model fusion algorithm based on Dual-UKF. Based on the state-parameter bidirectional coupling architecture, the alternating updates of state and parameters gradually approximate the real state and parameters. Through joint covariance correction, the coupling relationship between state and parameters is explicitly modeled, thereby realizing information interaction and decoupled estimation of state and parameters. This solves the problem of high-precision parameter update and state estimation of battery strongly coupled systems in complex scenarios where parameters are unknown or time-varying.

[0263] 7) Based on the high-precision estimation results of the cloud-side battery health state estimation model, this application simultaneously updates the model parameters and high real-time estimation results of the end-side battery health state estimation model. This solves the problems of high computational consumption of data-driven models, which makes it difficult to guarantee real-time performance, poor accuracy of high real-time SOH estimation results of end-side battery health state estimation models, and the problem of long-term generalization of model parameters as battery aging changes, while fixed parameters are difficult to adapt to different batteries and different operating conditions.

[0264] Based on the same inventive concept, this application also provides a power battery health state estimation system based on edge-cloud collaboration for implementing the aforementioned power battery health state estimation method. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more power battery health state estimation system embodiments provided below can be found in the limitations of the power battery health state estimation method based on edge-cloud collaboration described above, and will not be repeated here.

[0265] In one exemplary embodiment, a power battery health status estimation system based on edge-cloud collaboration is provided, comprising:

[0266] The data acquisition unit is used to acquire the cycle aging data of the target power battery at historical moments; the target power battery is the power battery to be estimated.

[0267] The preprocessing unit is used to preprocess the cycle aging data of the target power battery at historical time points to obtain the preprocessed cycle aging data of the target power battery at historical time points.

[0268] The feature extraction unit is used to extract features from the preprocessed cyclic aging data of the target power battery at historical time using the local linear embedding method, so as to obtain the local linear embedding features of the target power battery at historical time.

[0269] The cloud-side battery health status estimation unit is used to input the local linear embedding features of the target power battery at historical time into the cloud-side battery health status estimation model to obtain the estimated value of the cloud-side battery health status of the target power battery at the current time. The cloud-side battery health status estimation model is obtained by training the Pyraformer model using the training set.

[0270] The edge-cloud collaborative battery health state estimation unit uses a dual unscented Kalman filter algorithm to update the model parameters and estimated values ​​of the target power battery's end-side battery health state estimation model at the current moment, based on the cloud-side battery health state estimate of the target power battery at the current moment and the edge-side battery health state estimation model of the target power battery at the previous moment, to obtain the final battery health state estimate of the target power battery at the current moment. The edge-side battery health state estimation model of the target power battery at the previous moment was constructed based on the number of charge-discharge cycles of the target power battery at different stages of battery aging and the true values ​​of the battery health state at historical moments.

[0271] In one exemplary embodiment, a computer device is provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a power battery health state estimation method based on edge-cloud collaboration.

[0272] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 2 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a power battery health state estimation method based on edge-cloud collaboration.

[0273] Those skilled in the art will understand that Figure 2 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0274] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0275] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0276] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0277] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0278] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods, systems, and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for estimating the health status of a power battery based on edge-cloud collaboration, characterized in that, The power battery health status estimation method based on edge-cloud collaboration includes: Obtain the cyclic aging data of the target power battery at historical moments; the target power battery is the power battery to be estimated. Preprocess the cyclic aging data of the target power battery at historical time points to obtain the preprocessed cyclic aging data of the target power battery at historical time points. The local linear embedding method is used to extract features from the preprocessed cyclic aging data of the target power battery at historical time points, so as to obtain the local linear embedding features of the target power battery at historical time points. The local linear embedding features of the target power battery at historical moments are input into the cloud-side battery health status estimation model to obtain the estimated value of the cloud-side battery health status of the target power battery at the current moment; the cloud-side battery health status estimation model is obtained by training the Pyraformer model using the training set; A dual unscented Kalman filter algorithm is employed. Based on the estimated cloud-side battery health status of the target power battery at the current moment and the estimated end-side battery health status of the target power battery at the previous moment, the model parameters of the end-side battery health status estimation model and the estimated end-side battery health status of the target power battery at the current moment are updated to obtain the final battery health status estimate of the target power battery at the current moment. The end-side battery health status estimation model of the target power battery at the previous moment is constructed based on the number of charge-discharge cycles of the target power battery at different stages of battery aging and the true values ​​of battery health status at historical moments.

2. The method for estimating the health status of a power battery based on edge-cloud collaboration according to claim 1, characterized in that, The target power battery's cycle aging data at historical moments includes: voltage data, current data, and time data at historical moments. The target power battery's historical cycle aging data is preprocessed to obtain preprocessed cycle aging data at historical time points, including: Outliers are removed from the voltage and current data of the target power battery at historical time points to obtain the voltage and current data of the target power battery at historical time points after outlier removal. At a fixed sampling interval, the voltage data and current data of the target power battery at historical time after removing outliers are resampled to obtain the preprocessed voltage data and preprocessed current data of the target power battery at historical time. The pre-processed voltage data, pre-processed current data, and time data of the target power battery at historical time points are used as the pre-processed cycle aging data of the target power battery at historical time points.

3. The method for estimating the health status of a power battery based on edge-cloud collaboration according to claim 2, characterized in that, The local linear embedding method is used to extract features from the preprocessed cyclic aging data of the target power battery at historical time points, obtaining the local linear embedding features of the target power battery at historical time points, including: Based on the preprocessed voltage data, preprocessed current data, and time data of the target power battery at historical time points, the incremental capacity of continuous voltage steps is calculated to obtain the IC curve of the target power battery at historical time points; the IC curve is plotted with voltage as the horizontal axis and incremental capacity as the vertical axis. The local linear embedding method is used to extract features from the IC curves of the target power battery at historical time points, thereby obtaining the local linear embedding features of the target power battery at historical time points.

4. The method for estimating the health status of a power battery based on edge-cloud collaboration according to claim 3, characterized in that, The Pyraformer model consists of: an input layer, a pyramid structure layer, a cross-scale attention layer, a multi-scale feature fusion layer, and an output layer. The input layer is used to perform feature projection and position encoding on the local linear embedding features of the target power battery at historical time, so as to obtain the projection and encoded features of the target power battery at historical time. The pyramid structure layer is used to downsample the projection and encoding features of the target power battery at historical moments to obtain a multi-scale coarse-grained sequence of the target power battery at historical moments. The cross-scale attention layer is used to obtain the self-attention features of the target power battery at historical moments based on the multi-scale coarse-grained sequence of the target power battery at historical moments. The multi-scale feature fusion layer is used to upsample and align, stitch features and compress and project the self-attention features of the target power battery at historical moments to obtain the fused features of the target power battery at historical moments. The output layer is used to obtain an estimate of the cloud-side battery health status of the target power battery at the current moment based on the fusion characteristics of the target power battery at historical moments.

5. The method for estimating the health status of a power battery based on edge-cloud collaboration according to claim 4, characterized in that, The training process of the cloud-side battery health status estimation model specifically includes: Construct a training set; the training set includes: the local linear embedding features of the sample power battery at historical moments and the true values ​​of the cloud-side battery health status of the sample power battery at the corresponding historical moments; Build the Pyraformer model; The Pyraformer model is trained by taking the local linear embedding features of the sample power battery at historical time as input and the estimated true value of the cloud-side battery health status of the sample power battery at historical time as output. Training is stopped when the loss function reaches the minimum value or the number of training rounds reaches the maximum value, thus obtaining the cloud-side battery health status estimation model.

6. The method for estimating the health status of a power battery based on edge-cloud collaboration according to claim 5, characterized in that, The different stages of battery aging include: the initial rapid aging stage, the linear aging stage, and the capacity-depleted accelerated aging stage. The process of constructing the target power battery's end-side battery health state estimation model at the previous moment includes: When the target power battery is in the initial rapid aging stage, the end-side battery health status estimation model of the target power battery at the previous moment is constructed based on the power law model, the charge and discharge cycle number of the target power battery, and the true value of the battery health status at historical moments. When the target power battery is in the linear aging stage, the end-side battery health status estimation model of the target power battery at the previous moment is constructed based on the linear model, the number of charge-discharge cycles of the target power battery, and the true value of the battery health status at historical moments. When the target power battery is in the stage of accelerated aging with a sharp drop in capacity, the end-side battery health status estimation model of the target power battery at the previous moment is constructed based on the exponential model, the number of charge-discharge cycles of the target power battery, and the actual values ​​of the battery health status at historical moments.

7. The method for estimating the health status of a power battery based on edge-cloud collaboration according to claim 6, characterized in that, When the target power battery is in the initial rapid aging stage, the expression for the end-side battery health state estimation model of the target power battery at the previous moment is: Among them, SOH r is the estimated value of the end-side battery health status of the target power battery corresponding to the current charge-discharge cycle number; r is the current charge-discharge cycle number; a1 and n1 are both model parameters of the end-side battery health status estimation model of the target power battery at the previous moment; When the target power battery is in the linear aging stage, the expression for the end-side battery health state estimation model of the target power battery at the previous moment is: SOH r =SOH(N1)-a2·(r-N1); Where N1 is the number of charge-discharge cycles corresponding to the end of the initial rapid aging stage; SOH(N1) is the estimated final battery health status corresponding to the end of the initial rapid aging stage; a2 is the model parameter of the end-side battery health status estimation model of the target power battery at the previous moment. When the target power battery is in the stage of accelerated aging and capacity reduction, the expression for the end-side battery health state estimation model of the target power battery at the previous moment is: SOH r =a×exp(b×r)+c×exp(d×r); Where a, b, c and d are model parameters of the end-side battery health state estimation model of the target power battery at the previous moment.

8. The method for estimating the health status of a power battery based on edge-cloud collaboration according to claim 7, characterized in that, A dual unscented Kalman filter algorithm is employed. Based on the estimated cloud-side battery health status of the target power battery at the current moment and the estimated end-side battery health status of the target power battery at the previous moment, the model parameters of the end-side battery health status estimation model and the estimated end-side battery health status of the target power battery at the current moment are updated to obtain the final battery health status estimate of the target power battery at the current moment, including: When updating the model parameters of the target power battery's end-side battery health state estimation model at the current time: The dual unscented Kalman filter algorithm is used to update the model parameters of the target power battery's end-side battery health status estimation model at the current time based on the estimated value of the target power battery's cloud-side battery health status at the current time and the model parameters of the target power battery's end-side battery health status estimation model at the previous time, so as to obtain the model parameters of the target power battery's end-side battery health status estimation model at the current time. When updating the estimated value of the end-side battery health state of the target power battery at the current time using the end-side battery health state estimation model: The dual unscented Kalman filter algorithm is used to update the estimated value of the target power battery's end-side battery health status at the current time based on the estimated value of the cloud-side battery health status at the current time and the model parameters of the estimated model of the target power battery's end-side battery health status at the current time, so as to obtain the estimated value of the target power battery's end-side battery health status at the current time.

9. A power battery health status estimation system based on edge-cloud collaboration, characterized in that, The power battery health state estimation system based on edge-cloud collaboration is used to implement the power battery health state estimation method based on edge-cloud collaboration as described in any one of claims 1-8, wherein the power battery health state estimation system based on edge-cloud collaboration includes: The data acquisition unit is used to acquire the cyclic aging data of the target power battery at historical moments; the target power battery is the power battery to be estimated. The preprocessing unit is used to preprocess the cycle aging data of the target power battery at historical time points to obtain the preprocessed cycle aging data of the target power battery at historical time points. The feature extraction unit is used to extract features from the preprocessed cyclic aging data of the target power battery at historical time using the local linear embedding method, so as to obtain the local linear embedding features of the target power battery at historical time. The cloud-side battery health status estimation unit is used to input the local linear embedding features of the target power battery at historical moments into the cloud-side battery health status estimation model to obtain the estimated value of the cloud-side battery health status of the target power battery at the current moment; the cloud-side battery health status estimation model is obtained by training the Pyraformer model using the training set; The edge-cloud collaborative battery health state estimation unit uses a dual unscented Kalman filter algorithm to update the model parameters and estimated values ​​of the target power battery's end-side battery health state estimation model at the current moment, based on the cloud-side battery health state estimate of the target power battery at the current moment and the edge-side battery health state estimation model of the target power battery at the previous moment, to obtain the final battery health state estimate of the target power battery at the current moment. The edge-side battery health state estimation model of the target power battery at the previous moment was constructed based on the number of charge-discharge cycles of the target power battery at different stages of battery aging and the true values ​​of the battery health state at historical moments.

10. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the power battery health state estimation method based on end-to-cloud collaboration as described in any one of claims 1-8.

Images