Battery life prediction method and system with lossless reserved characteristics

Abstract

The invention provides a battery life prediction method and system for lossless feature retention, and belongs to the technical field of lithium ion battery life prediction based on deep learning, and the method comprises the steps: obtaining the parameter data of a to-be-predicted battery; the parameter data comprises charging data, discharging data, temperature data and internal resistance data; and processing the obtained parameter data of the to-be-predicted battery by using a pre-trained prediction model to obtain a battery life prediction result. According to the method, aging information contained in various data is fully mined and utilized, and comprehensive and in-depth analysis is carried out; through screening and dimensionality reduction processing of a large amount of data features, on the premise that key feature information is reserved, feature dimensionality is remarkably reduced, complexity and operation time of the model are reduced, precision of life prediction is improved, and reliability and efficiency of a prediction result are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium-ion battery life prediction based on deep learning, and in particular to a battery life prediction method and system with lossless feature retention. Background Art

[0002] As the global energy crisis intensifies, the popularization of clean energy has become the key to global development, and lithium-ion batteries have become the first choice in energy storage technology due to their high energy density, long life and high safety. However, lithium batteries will inevitably age during use, resulting in a gradual decline in performance and even potential safety hazards. Therefore, it is particularly important to accurately predict battery life in order to perform timely maintenance and improve battery operating efficiency and safety.

[0003] Traditional battery life prediction methods rely on a large amount of cycle data, which usually requires thousands of charge and discharge cycles, a long time period, and high costs. To solve this problem, using a small amount of data from the early cycles of the battery to predict life has become a key research direction. This method can not only significantly shorten the prediction time and reduce the investment in material and human resources, but also play an important role in improving safety and economic benefits, providing a more efficient management method for the practical application of lithium batteries.

[0004] At present, the early life prediction methods of batteries are mainly divided into two categories: model-based methods and data-driven methods. Compared with model-based methods, data-driven methods do not rely on an in-depth understanding of the complex aging mechanism of batteries, but obtain hidden aging feature information through large-scale data mining. The Chinese invention patent application with publication number CN118549839A proposes a technology that uses a relative position matrix to convert raw time series data into an image matrix, and uses a residual network to automatically extract hidden feature information of different depths, so as to achieve accurate prediction of battery life using a small amount of life data. The Chinese invention patent application with publication number CN 114662784A extracts 30 early aging features related to voltage, temperature, charging time, capacity and incremental capacity curves from the first 100 cycles, and uses a fast search clustering method to screen the features to achieve battery life prediction. However, the existing model has high multi-feature dimensions, large computational complexity, long training time, a lot of redundant information in the data set, and a complex model, which is not conducive to improving the training and prediction efficiency of the model. Summary of the invention

[0005] The object of the present invention is to provide a battery life prediction method and system with lossless feature preservation to solve at least one technical problem existing in the above-mentioned background technology.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a method for predicting battery life without loss of features, comprising:

[0008] Acquire parameter data of the battery to be predicted; the parameter data includes charging data, discharging data, temperature data and internal resistance data;

[0009] The parameter data of the battery to be predicted is processed using a pre-trained prediction model to obtain a battery life prediction result; wherein, the training of the prediction model includes: cleaning and preprocessing the original data of the battery; the original battery data includes charging data, discharging data, temperature data and internal resistance data; performing aging characteristic analysis on the processed data to screen out data that can be used for feature extraction; performing feature extraction guided by domain knowledge from the early cycle data of the battery to construct a complete high-dimensional feature data set; analyzing the correlation between the features and the battery life to select high-quality medium-dimensional features; using deep learning to continue to reduce the dimensionality of the features to generate low-dimensional features and reduce the complexity of the model; constructing a machine learning model, using a high-dimensional feature data set for training, and realizing rapid prediction of the early life of the battery by fitting the mapping relationship between the recoded features and the life to obtain a trained prediction model.

[0010] Furthermore, the early cycle data are the first 50 cycle data, and the proposed features are based on the differences in parameters or sequences between the 50th cycle and the nth cycle, and the statistics of the differences are calculated, where the statistics include mean, variance, maximum value, minimum value, kurtosis and skewness.

[0011] Furthermore, the feature and life span correlation analysis method is the Pearson correlation coefficient, which is calculated as follows:

[0012]

[0013] Among them, r represents the correlation coefficient, X and Y are the characteristic parameter vector and the predicted target vector, namely the final life EOL, respectively, and n represents the number of elements in the two sets of data.

[0014] Furthermore, the high-quality medium-dimensional features are screened with a Pearson correlation coefficient of 0.6 as a threshold.

[0015] Furthermore, the feature dimensionality reduction method adopts a deep variational autoencoder (DVAE), and the feature set to be fused is represented by x={x 1 ,x 2 ,…,x m}∈R m×n m represents the number of battery samples, and n represents the feature dimension. The encoder consists of 4 fully connected layers, and the number of neurons decreases layer by layer. The process of the encoder encoding the original feature data and generating the mean and variance is expressed as follows:

[0016] x encoder '=f 1 (W 1 x+b 1 )

[0017] x encoder ” = f 2 (W 2 x encoder '+b 2 )

[0018] x encoder ”'=f 3 (W 3 x encoder ”+b 3 )

[0019]

[0020] x encoder 、x encoder ” and x encoder ”' respectively represent the results of the second, third and fourth layers of the neural network obtained by X operation, μ z and σ z Represent the mean and standard deviation of the first generation, W 1 , W 2 , W 3 , and Represents the weight matrix of each layer of the neural network, b 1 、b 2 , and Represents the bias vector of each layer of the neural network, f 1 、f 2 and f 3 are the activation functions from the input layer to the third layer of the neural network, and Generate activation functions for mean and variance for the fourth layer respectively;

[0021] Sample a random vector ε from the standard normal distribution N(0,1), and calculate the sampled latent vector z={z 1 ,z 2 ,…,z m}∈R m×l , where l is the dimension of the feature fusion result;

[0022] The latent vector z is processed by the decoder to restore the data to the original input.

[0023] Furthermore, the decoder is composed of three layers of fully connected neural networks. The number of neurons from the input layer to the output layer increases layer by layer, which is symmetrical with the encoder. The decoder operation process is expressed as follows:

[0024] x decoder =f 4 (W 4 z+b 4 )

[0025] x decoder '=f 5 (W 5 x decoder +b 5 )

[0026] x decoder ”'=f 6 (W 6 x decoder ”+b 6 )

[0027]

[0028] x decoder 、x decoder ” and x decoder ”' represents the result of the decoder's first to third layers of neural networks obtained after calculation. Represents the result of the third layer, i.e. the output layer; W 4 , W 5 , W 6 and W 7 represents the weight matrix of each layer of the decoder neural network, b 4 、b 5 、b 6 and b 7 represents the bias vector of each layer of the decoder neural network; f 4 、f 5 and f 6 Represents the activation functions of each layer of the decoder.

[0029] In a second aspect, the present invention provides a battery life prediction system with lossless feature retention, comprising:

[0030] An acquisition module, used to acquire parameter data of the battery to be predicted; the parameter data includes charging data, discharging data, temperature data and internal resistance data;

[0031] The processing module is used to use a pre-trained prediction model to process the acquired parameter data of the battery to be predicted to obtain a battery life prediction result; wherein, training the prediction model includes: cleaning and preprocessing the original data of the battery; the original battery data includes charging data, discharging data, temperature data and internal resistance data; performing aging characteristic analysis on the processed data to screen out data that can be used for feature extraction; performing feature extraction guided by domain knowledge from the early cycle data of the battery to build a complete high-dimensional feature data set; analyzing the correlation between features and battery life to select high-quality medium-dimensional features; using deep learning to continue to reduce the dimensionality of features, generate low-dimensional features, and reduce model complexity; building a machine learning model, using a high-dimensional feature data set for training, and realizing rapid prediction of the early life of the battery by fitting the mapping relationship between the recoded features and the life, and obtaining a trained prediction model.

[0032] In a third aspect, the present invention provides a non-transitory computer-readable storage medium, which is used to store computer instructions. When the computer instructions are executed by a processor, a battery life prediction method with lossless feature retention as described in the first aspect is implemented.

[0033] In a fourth aspect, the present invention provides a computer device comprising a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute a battery life prediction method with lossless feature retention as described in the first aspect.

[0034] In a fifth aspect, the present invention provides an electronic device, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to execute instructions for implementing a battery life prediction method with lossless feature retention as described in the first aspect.

[0035] The beneficial effects of the present invention are as follows: the aging information contained in various types of data is fully mined and utilized for comprehensive and in-depth analysis; by screening and dimensionality reduction processing of a large number of data features, while retaining key feature information, the feature dimensions are significantly reduced, the complexity and running time of the model are reduced, and the accuracy of life prediction is improved, ensuring the reliability and efficiency of the prediction results.

[0036] Additional advantages of the present invention will be more clearly given in the following description or learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0038] Figure 1 This is an overall flow chart of a method for predicting battery life without loss of features according to an embodiment of the present invention.

[0039] Figure 2 The figure is a graph showing the voltage changing with time during the constant current constant voltage (CCCV) charging stage according to an embodiment of the present invention.

[0040] Figure 3 This is a scatter plot of the correlation between some features described in an embodiment of the present invention and lifespan.

[0041] Figure 4 This is a network block diagram of a deep variational autoencoder (DVAE) described in an embodiment of the present invention.

[0042] Figure 5 This is a diagram of early life prediction results described in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below by the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0044] It should be understood by those skilled in the art that unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which this invention belongs.

[0045] It should also be understood that terms, such as those defined in commonly used dictionaries, should be understood to have a meaning consistent with that in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless as defined herein.

[0046] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements and / or groups thereof.

[0047] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. Different embodiments or examples described in this specification and features of different embodiments or examples may be combined and combined by those skilled in the art without contradiction.

[0048] To facilitate understanding of the present invention, the present invention is further explained below with reference to specific embodiments in conjunction with the accompanying drawings, and the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0049] Those skilled in the art should understand that the drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily necessary for implementing the present invention.

[0050] Example 1

[0051] In this embodiment 1, a battery life prediction system with lossless feature retention is first provided, including: an acquisition module for acquiring parameter data of a battery to be predicted; the parameter data includes charging data, discharging data, temperature data and internal resistance data. A processing module is used to process the acquired parameter data of the battery to be predicted using a pre-trained prediction model to obtain a battery life prediction result; wherein, training the prediction model includes: cleaning and pre-processing the original data of the battery; the original data of the battery includes charging data, discharging data, temperature data and internal resistance data; performing aging characteristic analysis on the processed data to screen out data that can be used for feature extraction; performing feature extraction guided by domain knowledge from the early cycle data of the battery to construct a complete high-dimensional feature data set; analyzing the correlation between features and battery life to select high-quality medium-dimensional features; using deep learning to continue to reduce the dimension of features, generate low-dimensional features, and reduce model complexity; constructing a machine learning model, using a high-dimensional feature data set for training, and realizing rapid prediction of the early life of the battery by fitting the mapping relationship between re-encoded features and life, and obtaining a trained prediction model.

[0052] In this embodiment, the above system is used to implement a method for predicting battery life without loss of feature retention, including: using an acquisition module to acquire parameter data of the battery to be predicted; the parameter data includes charging data, discharging data, temperature data and internal resistance data. Using a processing module to process the acquired parameter data of the battery to be predicted using a pre-trained prediction model to obtain a battery life prediction result; wherein, training the prediction model includes: cleaning and pre-processing the original data of the battery; the original data of the battery includes charging data, discharging data, temperature data and internal resistance data; performing aging characteristic analysis on the processed data to filter out data that can be used for feature extraction; performing feature extraction guided by domain knowledge from the early cycle data of the battery to construct a complete high-dimensional feature data set; analyzing the correlation between features and battery life to select high-quality medium-dimensional features; using deep learning to continue to reduce the dimension of features, generate low-dimensional features, and reduce model complexity; constructing a machine learning model, using a high-dimensional feature data set for training, and realizing rapid prediction of the early life of the battery by fitting the mapping relationship between the re-encoded features and the life, and obtaining a trained prediction model.

[0053] Among them, the early cycle data are the first 50 cycle data, and the proposed features are based on the differences in parameters or sequences between the 50th cycle and the nth cycle, and the statistics of the differences are calculated, wherein the statistics include mean, variance, maximum value, minimum value, kurtosis and skewness.

[0054] The feature and life span correlation analysis method is the Pearson correlation coefficient, which is calculated as follows:

[0055]

[0056] Among them, r represents the correlation coefficient, X and Y are the characteristic parameter vector and the predicted target vector, namely the final life EOL, respectively, and n represents the number of elements in the two sets of data.

[0057] The high-quality medium-dimensional features are screened with a Pearson correlation coefficient of 0.6 as a threshold.

[0058] The feature dimensionality reduction method adopts the deep variational autoencoder (DVAE), and the feature set to be fused is represented by x={x 1 ,x 2 ,…,x m}∈R m×n m represents the number of battery samples, and n represents the feature dimension. The encoder consists of 4 fully connected layers, and the number of neurons decreases layer by layer. The process of the encoder encoding the original feature data and generating the mean and variance is expressed as follows:

[0059] x encoder '=f 1 (W 1x+b 1 )

[0060] x encoder ” = f 2 (W 2 x encoder '+b 2 )

[0061] x encoder ”'=f 3 (W 3 x encoder ”+b 3 )

[0062]

[0063] x encoder 、x encoder ” and x encoder ”' respectively represent the results of the second, third and fourth layers of the neural network obtained by X operation, μ z and σ z Represent the mean and standard deviation of the first generation, W 1 , W 2 , W 3 , and Represents the weight matrix of each layer of the neural network, b 1 、b 2 , and Represents the bias vector of each layer of the neural network, f 1 、f 2 and f 3 are the activation functions from the input layer to the third layer of the neural network, and Generate activation functions for mean and variance for the fourth layer respectively;

[0064] Sample a random vector ε from the standard normal distribution N(0,1), and calculate the sampled latent vector z={z 1 ,z 2 ,…,z m}∈R m×l , where l is the dimension of the feature fusion result;

[0065] The latent vector z is processed by the decoder to restore the data to the original input.

[0066] The decoder consists of three layers of fully connected neural networks. The number of neurons from the input layer to the output layer increases layer by layer, which is symmetrical with the encoder. The decoder operation process is expressed as follows:

[0067] x decoder=f 4 (W 4 z+b 4 )

[0068] x decoder '=f 5 (W 5 x decoder +b 5 )

[0069] x decoder ”'=f 6 (W 6 x decoder ”+b 6 )

[0070]

[0071] x decoder 、x decoder ” and x decoder ”' represents the result of the decoder's first to third layers of neural networks obtained after calculation. Represents the result of the third layer, i.e. the output layer; W 4 , W 5 , W 6 and W 7 represents the weight matrix of each layer of the decoder neural network, b 4 、b 5 、b 6 and b 7 represents the bias vector of each layer of the decoder neural network; f 4 、f 5 and f 6 Represents the activation functions of each layer of the decoder.

[0072] Example 2

[0073] In view of the shortcomings of the prior art, a method for predicting battery life without loss of feature retention is provided in this embodiment. This method uses a small amount of data from the initial stage of the battery cycle and performs dimensionality reduction processing on the features through a dimensionality reduction algorithm of a deep neural network, effectively reducing the amount of input data for the model, significantly improving training efficiency, and reducing the complexity of the model, thereby achieving rapid and accurate prediction of battery life.

[0074] The method for predicting the early life of lithium-ion batteries based on the combination of domain knowledge and deep learning includes the following steps: cleaning and preprocessing the original data of the battery; analyzing the aging characteristics of the processed data and screening out the data that can be used for feature extraction; extracting features guided by domain knowledge from the early cycle data of the battery and building a complete high-dimensional feature data set; analyzing the correlation between the features and the battery life and selecting high-quality medium-dimensional features; continuing to reduce the dimensionality of the features to generate low-dimensional features and reduce the complexity of the model; building a machine learning model to achieve rapid prediction of the early life of the battery by fitting the mapping relationship between the recoded features and the life.

[0075] The raw battery data includes charging data, discharging data, temperature data and internal resistance data. The early cycle data is the first 50 cycle data. The proposed features are based on the difference in parameters or sequences between the 50th cycle and the nth cycle, and the statistics of the difference are calculated, where the statistics include mean, variance, maximum value, minimum value, kurtosis and skewness.

[0076] The feature and life span correlation analysis method is the Pearson correlation coefficient, which is calculated as follows:

[0077]

[0078] Where r represents the correlation coefficient, X and Y are the characteristic parameter vector and the predicted target vector, i.e., the final life EOL, respectively, and n represents the number of elements in the two sets of data.

[0079] The high-quality medium-dimensional features are screened with a Pearson correlation coefficient of 0.6 as a threshold.

[0080] The feature dimensionality reduction method adopts the deep variational autoencoder (DVAE), and the feature set to be fused is represented by x={x 1 ,x 2 ,…,x m}∈R m×n m represents the number of battery samples, and n represents the feature dimension. The encoder consists of 4 fully connected layers, and the number of neurons decreases layer by layer. The process of the encoder encoding the original feature data and generating the mean and variance is expressed as follows:

[0081] x encoder '=f 1 (W 1 x+b 1 )

[0082] x encoder ” = f 2 (W 2 x encoder '+b 2 )

[0083] x encoder ”'=f 3 (W 3 x encoder ”+b 3 )

[0084]

[0085] x encoder 、x encoder ” and x encoder ”' respectively represent the results of the second, third and fourth layers of the neural network obtained by X operation, μ z and σ z Represent the mean and standard deviation of the first generation, W 1 , W 2 , W 3 , and Represents the weight matrix of each layer of the neural network, b 1 、b 2 , and Represents the bias vector of each layer of the neural network, f 1 、f 2 and f 3 are the activation functions from the input layer to the third layer of the neural network, and Generate mean and variance activation functions for the fourth layer respectively.

[0086] Then, a random vector ε is sampled from the standard normal distribution N(0,1), and the sampled latent vector z={z 1 ,z 2 ,…,z m}∈R m×l , where l is the dimension of the feature fusion result. As shown in the following formula:

[0087] z=μ z +ε·σ z

[0088] The latent vector z is processed by the decoder to restore the data to the original input. The decoder is also composed of three layers of fully connected neural networks. The number of neurons from the input layer to the output layer increases layer by layer, which is symmetrical with the encoder. The decoder operation process can be expressed as follows:

[0089] x decoder =f 4 (W 4 z+b 4 )

[0090] x decoder '=f 5(W 5 x decoder +b 5 )

[0091] x decoder ”'=f 6 (W 6 x decoder ”+b 6 )

[0092]

[0093] x decoder 、x decoder ” and x decoder ”' represents the result of the decoder's first to third layers of neural networks obtained after calculation. Represents the result of the third layer, the output layer. 4 , W 5 , W 6 and W 7 represents the weight matrix of each layer of the decoder neural network, b 4 、b 5 、b 6 and b 7 Represents the bias vector of each layer of the decoder neural network. 4 、f 5 and f 6 Represents the activation functions of each layer of the decoder.

[0094] Among them, machine learning models include random forest regression, support vector regression, Gaussian process regression and artificial neural network.

[0095] The method described in this embodiment fully mines and utilizes the aging information contained in various types of data to conduct a comprehensive and in-depth analysis; by screening and dimensionality reduction processing of a large number of data features, while retaining key feature information, the feature dimension is significantly reduced, the complexity and running time of the model are reduced, and the accuracy of life prediction is improved, ensuring the reliability and efficiency of the prediction results.

[0096] Example 3

[0097] like Figure 1 As shown, the steps of a lithium-ion battery early life prediction method based on the combination of domain knowledge and deep learning in this embodiment 3 are as follows:

[0098] S1: Battery raw data cleaning and preprocessing

[0099] In this embodiment, the MIT data set is used as the research object, and 124 lithium iron phosphate (LFP) / graphite batteries are randomly selected. For some batteries, the curves have a fault phenomenon. In this example, such batteries are defined as abnormal battery cells and are eliminated, and finally only 110 cells are retained.

[0100] Taking the discharge voltage as the benchmark, the discharge voltage interval is uniformly set to 3.5V to 2V, and divided into 1000 sampling points with the same voltage interval. The discharge capacity, discharge time, discharge temperature and incremental capacity of the discharge process are resampled using linear interpolation method.

[0101] S2: Battery aging characteristics and multi-dimensional data analysis

[0102] Draw the curves of various data under different cycle conditions, and systematically analyze their evolution trends with the aging process. Specifically, in the constant current constant voltage (CCCV) charging stage, draw the voltage-time curve, such as Figure 2 As shown; in the discharge stage, the discharge capacity-time curve, the discharge time-voltage curve and the differential capacity (IC) curve are drawn; in terms of temperature data, the battery temperature-time curve and the discharge temperature-voltage curve are drawn; in terms of internal resistance data, the curve of the change of internal resistance with the number of cycles is drawn.

[0103] S3: Feature extraction and correlation screening based on cyclic data

[0104] Based on the charging process, the discharging process, the temperature data and the internal resistance data, a more comprehensive feature extraction is performed on the historical data of the first 50 cycles. The differences in parameters or sequences between the 50th cycle and the nth cycle and the statistics of the differences are calculated, where n is 40, 30, 20, 10, 2 and capslopemin, where capslopemin represents the cycle corresponding to the minimum slope of the capacity decay curve. The statistics are mean, variance, maximum, minimum, kurtosis and skewness. Assume that there is a set of series {x n}={x 1 ,x 2 ,…,x n The formula for solving the statistics of this set of series is as follows:

[0105]

[0106] x max =max{x 1 ,x 2 ,...,x n}

[0107] x min =min{x 1 ,x 2 ,...,xn}

[0108]

[0109] Among them, x mean 、x var 、x max 、x min 、x kur and x skew They represent the mean, variance, maximum value, minimum value, kurtosis and skewness of the series respectively.

[0110] The following describes each feature in detail:

[0111] Based on the characteristics of the charging process, F1-F27

[0112] F1 to F3 represent the difference between CC charging time, CV charging time and total charging time between 50 cycles and n cycles. F4 to F9 represent the statistics of total charging time from n cycles to 50 cycles. F10 to F27 represent the statistics of the latter difference series of CC charging time, CV charging time and total charging time from n cycles to 50 cycles.

[0113] Based on the characteristics of the discharge process, F28-F170

[0114] F28-F33 represents dt of 50 cycles and n cycles lin (V) Sequence difference statistics. F34-F39 represent dQ of 50 cycles and n cycles lin (V) The statistics of the difference between the sequences. F40-F45 represent the dt of 50 cycles and n cycles. lin (V) Difference in sequence statistics. F46 to F51 represent the dQ of 50 cycles and n cycles. lin (V) Difference in sequence statistics. F52-F87 represents the dt from n cycles to 50 cycles lin (V) Statistics of the post-difference sequence statistics. F88-F123 represent dQ from n cycles to 50 cycles lin(V) Statistics of the post-difference sequence statistics. F124 to F129 represent the statistics of the post-difference sequence of the peak values ​​of the IC curve from n cycles to 50 cycles. F130-F135 represent the statistics of the post-difference sequence of the voltage corresponding to the peak value of the IC curve from n cycles to 50 cycles. F136-F141 represent the statistics of the post-difference sequence of the integrated area of ​​the IC curve from n cycles to 50 cycles. F142-F144 represent the difference between the peak value, the voltage corresponding to the peak value, and the integrated area of ​​the IC curve between 50 cycles and n cycles. F145 represents the difference between the discharge time of 50 cycles and n cycles. F146 represents the difference between the discharge capacity of 50 cycles and n cycles. F147 represents the difference between the discharge capacity of 50 cycles and the maximum discharge capacity within 50 cycles. F148-F153 represent the statistics of the post-difference sequence of the discharge time from n cycles to 50 cycles. F154-F159 represent the statistics of the post-difference sequence of the discharge capacity from n cycles to 50 cycles. F160 to F170 represent the parameters of the model when the capacity decay curve from n cycles to 50 cycles is fitted to four common models: linear model, square root model, exponential model and double exponential model.

[0115] Temperature-based features, F171-F241

[0116] F171 and F172 represent the difference between the peak value of the charging temperature of 50 cycles and n cycles and the time corresponding to the peak value. F173 and F174 represent the difference between the peak value of the discharging temperature of 50 cycles and n cycles and the time corresponding to the peak value. F175 represents the difference between the integral area of ​​the temperature curve of 50 cycles and n cycles. F176 to F181 represent the difference between the statistics of the temperature change curve of 50 cycles and n cycles over time. F182 to F187 represent the dT curve of the discharge temperature versus voltage of 50 cycles and n cycles lin (V) The statistics of the difference of the sequence. F188-F205 represent the statistics of the minimum, maximum and mean of the temperature curve from n cycles to 50 cycles. F206-F241 represent the discharge temperature versus voltage curve dT from n cycles to 50 cycles lin (V) Statistics of the post-difference sequence statistics.

[0117] Based on the characteristics of internal resistance, F242-F254

[0118] F242 represents the difference between the internal resistance of 50 cycles and n cycles. F243-F248 represent the statistics of the internal resistance from n cycles to 50 cycles. F249-F254 represent the statistics of the latter difference sequence of the internal resistance from n cycles to 50 cycles.

[0119] A total of 254 complete feature sets were obtained, and the Pearson correlation coefficient was used to measure the strength of the relationship between the feature and life expectancy, as shown in the following formula:

[0120]

[0121] Where r represents the correlation coefficient, X and Y are the characteristic parameter vector and the predicted target vector, i.e., the final life EOL, respectively, and n represents the number of elements in the two sets of data.

[0122] The scatter plot of the correlation between some characteristics and life span is as follows Figure 3 As shown, features with correlation coefficients greater than 0.6 were screened out, and finally 76 features were screened out from 254 features.

[0123] S4: Lossless feature dimensionality reduction that preserves original information

[0124] This embodiment uses DVAE to reduce the dimension of features. DVAE is an unsupervised model that aims to reduce the dimension of raw data and learn the potential representation of raw data. Its network diagram is as follows: Figure 4 shown.

[0125] The 76 medium-dimensional feature sets to be fused are represented by x = {x 1 ,x 2 ,…,x m}∈R m×n m represents the number of battery samples, and n represents the feature dimension. The encoder consists of 4 fully connected layers, and the number of neurons decreases layer by layer. The process of the encoder encoding the original feature data and generating the mean and variance is expressed as follows:

[0126] x encoder '=f 1 (W 1 x+b 1 )

[0127] x encoder ” = f 2 (W 2 x encoder '+b 2 )

[0128] x encoder ”'=f 3 (W 3 x encoder ”+b 3 )

[0129]

[0130] x encoder 、x encoder ” and x encoder ”' respectively represent the results of the second, third and fourth layers of the neural network obtained by X operation, μ z and σ z Represent the mean and standard deviation of the first generation, W 1 , W2 , W 3 , and Represents the weight matrix of each layer of the neural network, b 1 、b 2 , and Represents the bias vector of each layer of the neural network, f 1 、f 2 and f 3 are the activation functions from the input layer to the third layer of the neural network, and Generate mean and variance activation functions for the fourth layer respectively.

[0131] Then, a random vector ε is sampled from the standard normal distribution N(0,1), and the sampled latent vector z={z 1 ,z 2 ,…,z m}∈R m×l , where l is the dimension of the feature fusion result. As shown in the following formula:

[0132] z=μ z +ε·σ z

[0133] The latent vector z is processed by the decoder to restore the data to the original input. The decoder is also composed of three layers of fully connected neural networks. The number of neurons from the input layer to the output layer increases layer by layer, which is symmetrical with the encoder. The decoder operation process can be expressed as follows:

[0134] x decoder =f 4 (W 4 z+b 4 )

[0135] x decoder '=f 5 (W 5 x decoder +b 5 )

[0136] x decoder ”'=f 6 (W 6 x decoder ”+b 6 )

[0137]

[0138] x decoder 、x decoder ” and x decoder”' represents the result of the decoder's first to third layers of neural networks obtained after calculation. Represents the result of the third layer, the output layer. 4 , W 5 , W 6 and W 7 represents the weight matrix of each layer of the decoder neural network, b 4 、b 5 、b 6 and b 7 Represents the bias vector of each layer of the decoder neural network. 4 、f 5 and f 6 Represents the activation functions of each layer of the decoder.

[0139] In order to make the DVAE output as close to the input as possible, reduce the reconstruction error, and ensure that the fusion result is as close to the standard normal distribution as possible, the following two equations can be used for constraints:

[0140]

[0141] Among them, Loss reconstruction is the reconstruction loss, Loss KL is the KL divergence loss. m is the number of battery samples, and l is the dimension of fusion features.

[0142] Finally, the 10-dimensional encoded features are obtained through dimensionality reduction, which contains most of the aging information in the 76-dimensional features, and only a small amount of information is lost in the encoding.

[0143] S5: Early battery life prediction based on machine learning model

[0144] The machine learning models including random forest regression, support vector regression, Gaussian process regression and artificial neural network are used to train the model with the 10-dimensional features obtained after recoding as input. The results are as follows Figure 5 The following indicators are introduced, namely, mean absolute error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE) and goodness of fit R 2 , the calculation method of the evaluation index is as follows:

[0145]

[0146]

[0147] Where N is the number of samples, y iis the true value, is the predicted value, is the mean.

[0148] Example 4

[0149] This embodiment 4 provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, a battery life prediction method for losslessly retaining features as described above is implemented. The method includes:

[0150] Acquire parameter data of the battery to be predicted; the parameter data includes charging data, discharging data, temperature data and internal resistance data;

[0151] The parameter data of the battery to be predicted is processed using a pre-trained prediction model to obtain a battery life prediction result; wherein, the training of the prediction model includes: cleaning and preprocessing the original data of the battery; the original battery data includes charging data, discharging data, temperature data and internal resistance data; performing aging characteristic analysis on the processed data to screen out data that can be used for feature extraction; performing feature extraction guided by domain knowledge from the early cycle data of the battery to construct a complete high-dimensional feature data set; analyzing the correlation between the features and the battery life to select high-quality medium-dimensional features; using deep learning to continue to reduce the dimensionality of the features to generate low-dimensional features and reduce the complexity of the model; constructing a machine learning model, using a high-dimensional feature data set for training, and realizing rapid prediction of the early life of the battery by fitting the mapping relationship between the recoded features and the life to obtain a trained prediction model.

[0152] Example 5

[0153] This embodiment 5 provides a computer device, including a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, the processor calls the program instructions to execute the above-mentioned method for predicting battery life without loss of feature retention, the method comprising:

[0154] Acquire parameter data of the battery to be predicted; the parameter data includes charging data, discharging data, temperature data and internal resistance data;

[0155] The parameter data of the battery to be predicted is processed using a pre-trained prediction model to obtain a battery life prediction result; wherein, the training of the prediction model includes: cleaning and preprocessing the original data of the battery; the original battery data includes charging data, discharging data, temperature data and internal resistance data; performing aging characteristic analysis on the processed data to screen out data that can be used for feature extraction; performing feature extraction guided by domain knowledge from the early cycle data of the battery to construct a complete high-dimensional feature data set; analyzing the correlation between the features and the battery life to select high-quality medium-dimensional features; using deep learning to continue to reduce the dimensionality of the features to generate low-dimensional features and reduce the complexity of the model; constructing a machine learning model, using a high-dimensional feature data set for training, and realizing rapid prediction of the early life of the battery by fitting the mapping relationship between the recoded features and the life to obtain a trained prediction model.

[0156] Example 6

[0157] This embodiment 6 provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to execute instructions for implementing a battery life prediction method for lossless feature retention as described above, the method including:

[0158] Acquire parameter data of the battery to be predicted; the parameter data includes charging data, discharging data, temperature data and internal resistance data;

[0159] The parameter data of the battery to be predicted is processed using a pre-trained prediction model to obtain a battery life prediction result; wherein, the training of the prediction model includes: cleaning and preprocessing the original data of the battery; the original battery data includes charging data, discharging data, temperature data and internal resistance data; performing aging characteristic analysis on the processed data to screen out data that can be used for feature extraction; performing feature extraction guided by domain knowledge from the early cycle data of the battery to construct a complete high-dimensional feature data set; analyzing the correlation between the features and the battery life to select high-quality medium-dimensional features; using deep learning to continue to reduce the dimensionality of the features to generate low-dimensional features and reduce the complexity of the model; constructing a machine learning model, using a high-dimensional feature data set for training, and realizing rapid prediction of the early life of the battery by fitting the mapping relationship between the recoded features and the life to obtain a trained prediction model.

[0160] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, 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 disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0161] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0162] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate 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 A function specified in one or more boxes.

[0163] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0164] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative work on the basis of the technical solution disclosed in the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting battery life without loss of features, characterized in that: include: Acquire parameter data of the battery to be predicted; the parameter data includes charging data, discharging data, temperature data and internal resistance data; The parameter data of the battery to be predicted is processed using a pre-trained prediction model to obtain a battery life prediction result; wherein, the training of the prediction model includes: cleaning and preprocessing the original data of the battery; the original battery data includes charging data, discharging data, temperature data and internal resistance data; performing aging characteristic analysis on the processed data to screen out data that can be used for feature extraction; performing feature extraction guided by domain knowledge from the early cycle data of the battery to construct a complete high-dimensional feature data set; analyzing the correlation between the features and the battery life to select high-quality medium-dimensional features; using deep learning to continue to reduce the dimensionality of the features to generate low-dimensional features and reduce the complexity of the model; constructing a machine learning model, using a high-dimensional feature data set for training, and realizing rapid prediction of the early life of the battery by fitting the mapping relationship between the recoded features and the life to obtain a trained prediction model.

2. The method for predicting battery life without loss of features according to claim 1, characterized in that: The early cycle data are the first 50 cycle data, and the proposed features are based on the differences in parameters or sequences between the 50th cycle and the nth cycle, and the statistics of the differences are calculated, where the statistics include mean, variance, maximum value, minimum value, kurtosis and skewness.

3. The method for predicting battery life without loss of features according to claim 1, characterized in that: The feature and life span correlation analysis method is the Pearson correlation coefficient, which is calculated as follows: Among them, r represents the correlation coefficient, X and Y are the characteristic parameter vector and the predicted target vector, namely the final life EOL, respectively, and n represents the number of elements in the two sets of data.

4. The method for predicting battery life without loss of features according to claim 1, characterized in that: The high-quality medium-dimensional features are screened with a Pearson correlation coefficient of 0.6 as a threshold.

5. The method for predicting battery life without loss of features according to claim 1, characterized in that: The feature dimensionality reduction method adopts the deep variational autoencoder (DVAE), and the feature set to be fused is x={x1,x2,…,x m }∈R m×n m represents the number of battery samples, and n represents the feature dimension. The encoder consists of 4 fully connected layers, and the number of neurons decreases layer by layer. The process of the encoder encoding the original feature data and generating the mean and variance is expressed as follows: x encoder '=f1(W1x+b1) x encoder ”=f2(W2x encoder '+b2) x encoder ”'=f3(W3x encoder ”+b3) x encoder 、x encoder ” and x encoder ”' respectively represent the results of the second, third and fourth layers of the neural network obtained by X operation, μ z and σ z Represent the mean and standard deviation of the generated values, W1, W2, W3, and Represents the weight matrix of each layer of the neural network, b1, b2, and Represents the bias vector of each layer of the neural network. f1, f2 and f3 are the activation functions from the input layer to the third layer of the neural network. and Generate activation functions for mean and variance for the fourth layer respectively; Sample a random vector ε from the standard normal distribution N(0,1), and calculate the sampled latent vector z={z1,z2,…,z m }∈R m×l , where l is the dimension of the feature fusion result; The latent vector z is processed by the decoder to restore the data to the original input.

6. The method for predicting battery life without loss of features according to claim 5, characterized in that: The decoder consists of three layers of fully connected neural networks. The number of neurons from the input layer to the output layer increases layer by layer, which is symmetrical with the encoder. The decoder operation process is expressed as follows: x decoder =f4(W4z+b4) x decoder '=f5(W5x decoder +b5) x decoder ”'=f6(W6x decoder ”+b6) x decoder 、x decoder ” and x decoder ”' represents the result of the decoder's first to third layers of neural networks obtained after calculation. represents the result of the third layer, i.e. the output layer; W4, W5, W6 and W7 represent the weight matrices of the neural networks of each layer of the decoder; b4, b5, b6 and b7 represent the bias vectors of the neural networks of each layer of the decoder; f4, f5 and f6 represent the activation functions of each layer of the decoder.

7. A battery life prediction system with lossless feature preservation, characterized in that: include: An acquisition module, used to acquire parameter data of the battery to be predicted; the parameter data includes charging data, discharging data, temperature data and internal resistance data; The processing module is used to use a pre-trained prediction model to process the acquired parameter data of the battery to be predicted to obtain a battery life prediction result; wherein, training the prediction model includes: cleaning and preprocessing the original data of the battery; the original battery data includes charging data, discharging data, temperature data and internal resistance data; performing aging characteristic analysis on the processed data to screen out data that can be used for feature extraction; performing feature extraction guided by domain knowledge from the early cycle data of the battery to build a complete high-dimensional feature data set; analyzing the correlation between features and battery life to select high-quality medium-dimensional features; using deep learning to continue to reduce the dimensionality of features, generate low-dimensional features, and reduce model complexity; building a machine learning model, using a high-dimensional feature data set for training, and realizing rapid prediction of the early life of the battery by fitting the mapping relationship between the recoded features and the life, and obtaining a trained prediction model.

8. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, a battery life prediction method with lossless feature retention as described in any one of claims 1-6 is implemented.

9. A computer device, characterized in that: It includes a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute a battery life prediction method with lossless feature retention as described in any one of claims 1-6.

10. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to execute instructions for implementing a battery life prediction method with lossless feature retention as described in any one of claims 1-6.

Images