Lithium battery health status prediction method and device

Abstract

The present disclosure discloses a method and device for predicting the health status of a lithium battery, which relates to the field of data processing technology. By preprocessing the second preset number of health factor data, the data quality is improved to obtain the preprocessed health factor data set, and the preprocessed health factor data set is processed in sequence based on the two convolutional layers, the bidirectional long short-term memory network, the preset attention mechanism, the fully connected layer and the preset encoder-decoder, so as to achieve accurate prediction of the health status of the target battery. In addition, by directly using the health factor data extracted from the historical operation data of the target battery to predict the health status of the target battery, the process of constructing the target battery parameter model is avoided, so that the prediction method of the health status of the lithium battery disclosed in the present disclosure has strong adaptability and simple operation.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing technology, and in particular to a method and device for predicting the health status of a lithium battery. Background Art

[0002] To address environmental issues and the fossil energy crisis, clean energy sources such as wind power, hydropower, and nuclear power are being urgently and vigorously developed. Energy storage and utilization are becoming increasingly important. Compared to other types of batteries, such as lead-acid batteries and nickel-cadmium batteries, lithium batteries offer higher energy density, lower self-discharge rates, and longer charge and discharge lifespans. They are widely used in various portable electronic devices, electric vehicles, and energy storage systems. However, over time and with battery use, batteries inevitably age, increasing internal resistance, decreasing available capacity, and degrading performance. This can lead to battery leakage, local short circuits, equipment failures, shutdowns, and even dangerous events such as overheating and explosions. Therefore, in order to avoid these dangerous events, predicting the health status of lithium batteries has become an urgent issue that needs to be addressed.

[0003] At present, the detection of the health status of lithium batteries is mainly achieved by constructing a lithium battery parameter model to predict the health status of lithium batteries. Although the above method can detect the health status of lithium batteries, the construction of the lithium battery parameter model often needs to consider the internal chemical reactions and aging principles of the lithium battery. As a result, the application of the above method requires operators to have strong professional knowledge reserves. In addition, the method of constructing a lithium battery parameter model to predict the health status of lithium batteries is not generalizable enough when applied to different types of batteries. Therefore, providing a highly adaptable and easy-to-operate lithium battery health status prediction method to achieve accurate prediction of the health status of lithium batteries is an urgent problem that needs to be solved. Summary of the Invention

[0004] The present disclosure provides a method and device for predicting the health status of a lithium battery, the main purpose of which is to provide a method for predicting the health status of a lithium battery that is highly adaptable and easy to operate, so as to achieve accurate prediction of the health status of a lithium battery.

[0005] According to a first aspect of the present disclosure, a method for predicting the health status of a lithium battery is provided, comprising:

[0006] Selecting a second preset number of health factor data from a first preset number of health factor data based on the Pearson correlation coefficient, wherein the first preset number is greater than the second preset number;

[0007] preprocessing the second preset amount of health factor data to obtain a preprocessed health factor data set;

[0008] Performing feature extraction on the preprocessed health factor dataset based on two convolutional layers to obtain a feature data sequence;

[0009] Performing a long-term dependency analysis on the feature data sequence based on a bidirectional long short-term memory network to obtain a target hidden state;

[0010] Performing weighted processing on the target hidden state based on a preset attention mechanism to obtain an indicator vector related to the current health state of the target battery;

[0011] Processing the indicator vector based on a preset linear function in a fully connected layer to obtain the maximum available capacity of the target battery;

[0012] The maximum available capacity is denoised and optimized based on a preset encoder-decoder to obtain an optimized maximum available capacity, and the current health state of the target battery is determined according to the optimized maximum available capacity.

[0013] Optionally, selecting a second preset number of health factor data from the first preset number of health factor data based on the Pearson correlation coefficient includes:

[0014] Calculating the Pearson correlation coefficients between the first preset number of health factor data and the historical health status of the target battery based on a correlation coefficient function to obtain the first preset number of Pearson correlation coefficients;

[0015] The second preset number of Pearson correlation coefficients are selected from the first preset number of Pearson correlation coefficients according to the order of absolute value from large to small, and the second preset number of health factor data corresponding to the second preset number of Pearson correlation coefficients are determined.

[0016] Optionally, preprocessing the second preset amount of health factor data to obtain a preprocessed health factor data set includes:

[0017] performing normalization processing on the second preset amount of health factor data based on a normalization function to obtain a normalized health factor data set;

[0018] Gaussian noise is added to the standardized health factor dataset based on a data enhancement function to obtain the preprocessed health factor dataset.

[0019] Optionally, the correlation coefficient function is expressed as:

[0020]

[0021] Wherein, a is the health factor data, b is the historical health status, is the average value of the health factor data, is the average value of the historical health status, A is the Pearson correlation coefficient, i is the first subscript variable, and n is the sequence length of the health factor data;

[0022] The formula of the normalization function is expressed as:

[0023]

[0024] Among them, X scaled is the target health factor data after normalization, X is the feature value selected in the target health factor data, and X min is the minimum eigenvalue in the target health factor data, X max is the minimum eigenvalue in the target health factor data, and the target health factor data is one of the second preset number of health factor data;

[0025] The formula of the data enhancement function is expressed as:

[0026] x n =x+σ·N(0,1)

[0027] Where x is the standardized health factor dataset, σ is the standard deviation of Gaussian noise, N(0,1) represents the standard normal distribution with mean 0 and variance 1, and x n represents the pre-processed health factor dataset.

[0028] Optionally, the two convolutional layers include a feature extraction function, and the formula of the feature extraction function is expressed as:

[0029]

[0030] Among them, H is the feature data sequence, K is the size of the convolution kernel, x n is the preprocessed health factor dataset, I is the second subscript variable, t is the time step, k is the weight value, and k(I) represents the weight value at I.

[0031] Optionally, the bidirectional long short-term memory network includes an input gate, a forget gate, an output gate, a memory unit update function and a hidden state update function;

[0032] The formula of the input gate is expressed as:

[0033] i t =σ1(W i ·[h t-1 ,x t ])

[0034] Among them, W iis the input gate weight matrix, i t is the input gate activation value, h t-1 is the target hidden state at time t-1, x t is the input data at the current time t, σ1(.) is the first activation function;

[0035] The formula of the forget gate is expressed as:

[0036] f t =σ2(W f ·[h t-1 ,x t ])

[0037] Among them, W f is the forget gate weight matrix, f t is the activation value of the forget gate, h t-1 is the target hidden state at time t-1, x t is the input data at the current time t, σ2(.) is the second activation function;

[0038] The formula of the output gate is expressed as:

[0039] o t =σ3(W o ·[h t-1 ,x t ])

[0040] Among them, W o is the forget gate weight matrix, o t is the activation value of the forget gate, h t-1 is the target hidden state at time t-1, x t is the input data at the current time t, σ3(.) is the third activation function;

[0041] The formula of the memory unit update function is expressed as:

[0042]

[0043] Among them, C t-1 is the state of the memory unit at time t-1, is the candidate memory cell state, f t c t-1 It is used to control the contribution of the memory cell state at time t-1 to the memory cell state at time t. For control C t The contribution of c t is the state of the memory unit at time t, f t is the activation value of the forget gate, i t is the input gate activation value.

[0044] The formula of the hidden state update function is expressed as:

[0045] h t =o t ·σ4(c t )

[0046] Among them, h t is the target hidden state at time t, o t is the activation value of the forget gate, c t is the state of the memory unit at time t, and σ4(.) is the fourth activation function.

[0047] Optionally, performing weighted processing on the target hidden state based on a preset attention mechanism to obtain an indicator vector related to the current health state of the target battery includes:

[0048] Calculating an attention score for the target hidden state based on an attention score function;

[0049] Calculate the weight value corresponding to the target hidden state according to the attention score based on a weight calculation function;

[0050] Calculating the indicator vector according to the attention score and the weight value based on a weighting function, the preset attention mechanism includes the attention score function, the weight calculation function and the weighting function;

[0051] The formula of the attention score function is expressed as:

[0052] α=tanh(W a ·h t )

[0053] Among them, W a is the attention weight matrix, h t is the target hidden state at time t, tanh(.) is the hyperbolic tangent function, and α is the attention score;

[0054] The formula of the weight calculation function is expressed as:

[0055] w=softmax(α)

[0056] Wherein, w is the weight value, α is the attention score, and softmax(.) is the fifth activation function;

[0057] The formula of the weighting function is expressed as:

[0058]

[0059] Where V is the indicator vector, t is the time step, w is the weight value, h tis the target hidden state at time t.

[0060] Optionally, the performing denoising optimization on the maximum available capacity based on a preset encoder-decoder to obtain the optimized maximum available capacity includes:

[0061] Performing nonlinear transformation processing on the maximum available capacity based on an encoder to obtain a nonlinear maximum available capacity;

[0062] performing a linear transformation process on the nonlinear maximum available capacity based on a decoder to obtain the optimized maximum available capacity, wherein the preset encoder-decoder comprises the encoder and the decoder;

[0063] The encoder formula is expressed as:

[0064] d1=ReLU(W·O f +b)

[0065] Wherein, W is the weight matrix of the encoder, b is the bias of the encoder, d1 is the maximum available capacity of the nonlinearity, O f The maximum available capacity, ReLU(.) is a nonlinear transformation function;

[0066] The decoder formula is expressed as:

[0067] O d =W′·d1+b′

[0068] Where W′ is the weight matrix of the decoder, b′ is the bias of the decoder, d1 is the maximum available capacity of the nonlinearity, O d This is the maximum available capacity after optimization.

[0069] Optionally, before selecting a second preset number of health factor data from the first preset number of health factor data based on the Pearson correlation coefficient, the method includes:

[0070] Acquiring historical operating data of the target battery during the discharge process;

[0071] The first preset amount of health factor data is extracted from the historical operating data.

[0072] According to a second aspect of the present disclosure, a device for predicting the health status of a lithium battery is provided, comprising:

[0073] a selection unit, configured to select a second preset number of health factor data from a first preset number of health factor data based on a Pearson correlation coefficient, wherein the first preset number is greater than the second preset number;

[0074] a processing unit, configured to preprocess the second preset amount of health factor data to obtain a preprocessed health factor data set;

[0075] an extraction unit, configured to perform feature extraction on the preprocessed health factor dataset based on two convolutional layers to obtain a feature data sequence;

[0076] An analysis unit, configured to perform a long-term dependency analysis on the feature data sequence based on a bidirectional long short-term memory network to obtain a target hidden state;

[0077] a weighting unit, configured to perform weighted processing on the target hidden state based on a preset attention mechanism to obtain an indicator vector related to the current health state of the target battery;

[0078] a calculation unit, configured to process the indicator vector based on a preset linear function in a fully connected layer to obtain a maximum available capacity of the target battery;

[0079] An optimization unit is configured to perform denoising optimization on the maximum available capacity based on a preset encoder-decoder to obtain an optimized maximum available capacity, and determine a current health state of the target battery according to the optimized maximum available capacity.

[0080] According to a third aspect of the present disclosure, there is provided an electronic device, including:

[0081] at least one processor; and

[0082] a memory communicatively connected to the at least one processor; wherein,

[0083] The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect.

[0084] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute the method described in the first aspect.

[0085] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the method as described in the first aspect above.

[0086] The present disclosure provides a method and device for predicting the health status of a lithium battery. The method and device select a second preset number of health factor data from a first preset number of health factor data based on the Pearson correlation coefficient, where the first preset number is greater than the second preset number; preprocess the second preset number of health factor data to obtain a preprocessed health factor data set; extract features from the preprocessed health factor data set based on two convolutional layers to obtain a feature data sequence; perform long-term dependency analysis on the feature data sequence based on a bidirectional long short-term memory network to obtain a target hidden state; perform weighted processing on the target hidden state based on a preset attention mechanism to obtain an indicator vector related to the current health status of the target battery; process the indicator vector based on a preset linear function in a fully connected layer to obtain the maximum available capacity of the target battery; perform denoising optimization on the maximum available capacity based on a preset encoder-decoder to obtain the optimized maximum available capacity, and determine the current health status of the target battery based on the optimized maximum available capacity. Compared with the related art, by preprocessing the second preset number of health factor data, the data quality is improved to obtain the preprocessed health factor data set, and the preprocessed health factor data set is processed in sequence based on the two convolutional layers, the bidirectional long short-term memory network, the preset attention mechanism, the fully connected layer and the preset encoder-decoder, thereby realizing the prediction of the target battery health status. In addition, by directly using the health factor data extracted from the historical operation data of the target battery to realize the prediction of the target battery health status, the process of constructing the target battery parameter model is avoided, so that the prediction method of the lithium battery health status disclosed in the present invention has strong adaptability and simple operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] The accompanying drawings, which constitute part of the present invention, are intended to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are intended to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0088] Figure 1 A schematic flow chart of a method for predicting the health status of a lithium battery provided in an embodiment of the present disclosure;

[0089] Figure 2 The embodiment of the present disclosure provides a capacity decay curve of a NASA lithium battery using experimental data;

[0090] Figure 3 A capacity decay curve diagram of a lithium battery using experimental data CALCE provided in an embodiment of the present disclosure;

[0091] Figure 4A correlation heat map of characteristic factors in a NASA dataset provided by an embodiment of the present disclosure;

[0092] Figure 5 A correlation heat map of characteristic factors in the dataset CALCE provided by an embodiment of the present disclosure;

[0093] Figure 6 A curve result and error diagram of a NASA data set predicted by a design model provided by the present disclosure in an embodiment of the present disclosure;

[0094] Figure 7 The curve results and error diagram of the CALCE data set predicted by the design model of the present disclosure provided in an embodiment of the present disclosure;

[0095] Figure 8 A schematic diagram of the structure of a device for predicting the health status of a lithium battery provided by an embodiment of the present disclosure;

[0096] Figure 9 A schematic block diagram of an exemplary electronic device 300 provided in accordance with an embodiment of the present disclosure. DETAILED DESCRIPTION

[0097] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present invention can be combined with each other.

[0098] The following detailed description is an exemplary description and is intended to provide further detailed description of the present invention. Unless otherwise indicated, all technical terms used in the present invention have the same meaning as those generally understood by those skilled in the art to which the present invention belongs. The terms used in the present invention are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention.

[0099] In addition, the terms "first," "second," and the like in the specification and claims of the present disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, such that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein.

[0100] The following describes a method and apparatus for predicting the health status of a lithium battery according to an embodiment of the present disclosure with reference to the accompanying drawings.

[0101] In order to at least provide a method for predicting the health status of a lithium battery that is highly adaptable and easy to operate, so as to achieve accurate prediction of the health status of a lithium battery, this embodiment provides a method for predicting the health status of a lithium battery.

[0102] Figure 1This is a flow chart of a method for predicting the health status of a lithium battery provided by an embodiment of the present disclosure. Figure 1 As shown, the method includes the following steps:

[0103] Step 101: selecting a second preset number of health factor data from a first preset number of health factor data based on a Pearson correlation coefficient, wherein the first preset number is greater than the second preset number;

[0104] In some embodiments, the value of the first preset number includes but is not limited to twelve, the value of the second preset number includes but is not limited to five, and the types of the health factor data of the first preset number include but are not limited to time, temperature, voltage, current, and internal resistance.

[0105] Step 102: preprocess the second preset amount of health factor data to obtain a preprocessed health factor data set;

[0106] Step 103: extracting features from the preprocessed health factor dataset based on two convolutional layers to obtain a feature data sequence;

[0107] Step 104: performing a long-term dependency analysis on the feature data sequence based on a bidirectional long short-term memory network to obtain a target hidden state;

[0108] Step 105: performing weighted processing on the target hidden state based on a preset attention mechanism to obtain an indicator vector related to the current health state of the target battery;

[0109] Step 106: Process the indicator vector based on a preset linear function in a fully connected layer to obtain the maximum available capacity of the target battery;

[0110] Step 107 : performing denoising optimization on the maximum available capacity based on a preset encoder-decoder to obtain an optimized maximum available capacity, and determining the current health status of the target battery according to the optimized maximum available capacity.

[0111] The present invention provides a method for predicting the health status of a lithium battery. The method selects a second preset number of health factor data from a first preset number of health factor data based on the Pearson correlation coefficient, wherein the first preset number is greater than the second preset number; preprocesses the second preset number of health factor data to obtain a preprocessed health factor data set; extracts features from the preprocessed health factor data set based on two convolutional layers to obtain a feature data sequence; performs long-term dependency analysis on the feature data sequence based on a bidirectional long short-term memory network to obtain a target hidden state; performs weighted processing on the target hidden state based on a preset attention mechanism to obtain an indicator vector related to the current health status of the target battery; processes the indicator vector based on a preset linear function in a fully connected layer to obtain the maximum available capacity of the target battery; performs denoising optimization on the maximum available capacity based on a preset encoder-decoder to obtain the optimized maximum available capacity, and determines the current health status of the target battery based on the optimized maximum available capacity. Compared with the related art, by preprocessing the second preset number of health factor data, the data quality is improved to obtain the preprocessed health factor data set, and the preprocessed health factor data set is processed in sequence based on the two convolutional layers, the bidirectional long short-term memory network, the preset attention mechanism, the fully connected layer and the preset encoder-decoder, thereby realizing the prediction of the target battery health status. In addition, by directly using the health factor data extracted from the historical operation data of the target battery to realize the prediction of the target battery health status, the process of constructing the target battery parameter model is avoided, so that the prediction method of the lithium battery health status disclosed in the present invention has strong adaptability and simple operation.

[0112] As a refinement of the above embodiment, before executing step 101 to select a second preset number of health factor data from a first preset number of health factor data based on the Pearson correlation coefficient, the method may also adopt but is not limited to the following implementation methods, for example: obtaining historical operating data of the target battery during the discharge process; extracting the first preset number of health factor data from the historical operating data.

[0113] In order to facilitate understanding of the above embodiment, this embodiment provides an exemplary description. This embodiment obtains historical operating data of four batteries from the NASA data set and the CALCE data set respectively. The experimental conditions for generating the data in the NASA data set are shown in Table 1, and the experimental conditions for generating the data in the CALCE data set are shown in Table 2. The four batteries in the NASA data set are B0005 (B5), B0006 (B6), B0007 (B7), and B0018 (B18). Each battery undergoes a charge and discharge process at 24°C. The nominal capacity is 2Ah. During the charging stage, the battery is charged to 4.2V at a constant current of 1.5A, and then charged at a constant voltage until the current drops to 20mA. During the discharging stage, the battery is discharged to the cut-off voltage at a constant current of 2A. This cycle is repeated, and its capacity decay curve is shown as follows. Figure 2 As shown, Figure 2 The embodiment of the present disclosure provides a capacity decay curve of a lithium battery using experimental data NASA. The four batteries in the CALCE data set are CS2_35 (C35), CS2_36 (C36), CS 37 (C37) and CS2_38 (C38). Each battery undergoes a charge and discharge process at an ambient temperature of 1°C. The nominal capacity is 1.1Ah. It is charged in a constant current 0.5A (CC) mode until the battery voltage reaches 4.2V, and then charged in a constant voltage (CV) mode until the charging current drops to 20mA. In the discharge stage, it is discharged in a constant current 1A (CC) mode until the battery voltage drops to 2.7V. Its capacity decay curve is as follows: Figure 3 As shown, Figure 3 A capacity decay curve diagram of a lithium battery using experimental data CALCE is provided in an embodiment of the present disclosure.

[0114] Table 1 shows the experimental conditions of the NASA dataset

[0115]

[0116] Table 2 shows the experimental conditions of the CALCE dataset.

[0117]

[0118] Health factor data was extracted from the historical operating data of eight lithium batteries in the NASA and CALCE datasets. The types of health factor data and their corresponding aging characteristics are shown in Table 3. The health factor data shown in Table 3 are multidimensional features, each with varying degrees of correlation with SOH. Because the extracted health factor data includes data that is less relevant to the battery's health status, prediction accuracy is reduced.

[0119] Table 3 Types of lithium battery health factor data and their aging performance

[0120]

[0121] As a refinement of an embodiment of the present disclosure, when executing step 101 to select a second preset number of health factor data from a first preset number of health factor data based on the Pearson correlation coefficient, the following implementation method may also be adopted but is not limited to, for example: calculating the Pearson correlation coefficients between the first preset number of health factor data and the historical health status of the target battery based on a correlation coefficient function to obtain the first preset number of Pearson correlation coefficients; selecting the second preset number of Pearson correlation coefficients from the first preset number of Pearson correlation coefficients according to the order of absolute value from large to small, and determining the second preset number of health factor data corresponding to the second preset number of Pearson correlation coefficients.

[0122] In some embodiments, the types of the first preset number of health factor data include but are not limited to the types of health factor data shown in Table 3. In order to select the health factor data most relevant to the historical health status of the target battery, the Pearson correlation coefficients between the first preset number of health factor data and the historical health status of the target battery are calculated respectively. The closer the absolute value of the calculated Pearson correlation coefficient is to 1, the more relevant the health factor data is to the health status of the target battery. Combined with the above exemplary description, the Pearson correlation coefficients between the health factor data of eight batteries shown in the NASA dataset and the CALCE dataset and the health status of their respective corresponding batteries are shown in Table 4. The correlation shown in Table 4 is made into a correlation heat map for clearer visualization, as shown in Table 4. Figure 4 、 Figure 5 As shown, Figure 4 A correlation heat map of characteristic factors in a NASA dataset provided by an embodiment of the present disclosure; Figure 5 An embodiment of the present disclosure provides a correlation heat map of characteristic factors in the data set CALCE, where SOH represents state of health.

[0123] Table 4 Pearson correlation coefficients between battery health factor data and SOH (top 5 are bolded)

[0124]

[0125] As a refinement of the above embodiment, when performing the preprocessing of the second preset number of health factor data as described in step 102 to obtain a preprocessed health factor data set, the following implementation methods can also be adopted but are not limited to, for example: standardizing the second preset number of health factor data based on a normalization function to obtain a standardized health factor data set; adding Gaussian noise to the standardized health factor data set based on a data enhancement function to obtain the preprocessed health factor data set.

[0126] As a refinement of the above embodiment, the formula of the correlation coefficient function is expressed as:

[0127]

[0128] Wherein, a is the health factor data, b is the historical health status, is the average value of the health factor data, is the average value of the historical health status, A is the Pearson correlation coefficient, i is the first subscript variable, and n is the sequence length of the health factor data. The Pearson correlation coefficient A is in (-1, 1). The larger its absolute value, the stronger the correlation. The closer it is to 1, the positive correlation is shown, and the closer it is to -1, the negative correlation is shown.

[0129] The formula of the normalization function is expressed as:

[0130]

[0131] Among them, X scaled is the target health factor data after normalization, X is the feature value selected in the target health factor data, and X min is the minimum eigenvalue in the target health factor data, X max is the minimum eigenvalue in the target health factor data, and the target health factor data is one of the second preset number of health factor data;

[0132] The formula of the data enhancement function is expressed as:

[0133] x n =x+σ·N(0,1)

[0134] Where x is the standardized health factor dataset, σ is the standard deviation of Gaussian noise, N(0,1) represents the standard normal distribution with mean 0 and variance 1, and x n Represents the preprocessed health factor dataset. Generate Gaussian noise with the same shape as x, then scale the noise by multiplying the standard deviation of the noise by the standard normal distribution and add it to x.

[0135] As a refinement of the above embodiment, the two convolutional layers include a feature extraction function, and the formula of the feature extraction function is expressed as:

[0136]

[0137] Among them, H is the feature data sequence, K is the size of the convolution kernel, x n is the preprocessed health factor dataset, I is the second subscript variable, t is the time step, k is the weight value, and k(I) represents the weight value at I. In the present disclosure, K is preferably 3. Two convolutional layers are defined in the present disclosure, and the input and output channels are defined as x→64 and 64→128, respectively.

[0138] As a refinement of the above embodiment, the bidirectional long short-term memory network includes an input gate, a forget gate, an output gate, a memory unit update function and a hidden state update function, the forget gate selects to delete weak relevant information in the lithium-ion battery charge and discharge data, and the output gate determines the output information;

[0139] The formula of the input gate is expressed as:

[0140] i t =σ1(W i ·[h t-1 ,x t ])

[0141] Among them, W i is the input gate weight matrix, i t is the input gate activation value, h t-1 is the target hidden state at time t-1, x t is the input data at the current time t, σ1(.) is the first activation function, i t Used to control the current input x t The impact on the state of the memory unit, the output value of the first activation function is between 0 and 1. When the value is close to 0, it means "off", and when it is close to 1, it means "on";

[0142] The formula of the forget gate is expressed as:

[0143] f t =σ2(W f ·[h t-1 ,x t ])

[0144] Among them, W f is the forget gate weight matrix, f t is the activation value of the forget gate, h t-1 is the target hidden state at time t-1, x t is the input data at the current time t, σ2(.) is the second activation function, f tUsed to control the influence of the previous memory unit state on the current memory unit state;

[0145] The formula of the output gate is expressed as:

[0146] o t =σ3(W o ·[h t-1 ,x t ])

[0147] Among them, W o is the forget gate weight matrix, o t is the activation value of the forget gate, h t-1 is the target hidden state at time t-1, x t is the input data at the current time t, σ3(.) is the third activation function, O t Used to determine the impact of the current memory cell state on the hidden state;

[0148] The formula of the memory unit update function is expressed as:

[0149]

[0150] Among them, C t-1 is the state of the memory unit at time t-1, is the candidate memory cell state, f t c t-1 It is used to control the contribution of the memory cell state at time t-1 to the memory cell state at time t. For control C t The contribution of c t is the state of the memory unit at time t, f t is the activation value of the forget gate, i t is the input gate activation value.

[0151] The formula of the hidden state update function is expressed as:

[0152] h t =o t ·σ4(c t )

[0153] Among them, h t is the target hidden state at time t, o t is the activation value of the forget gate, c t is the state of the memory unit at time t, σ4(.) is the fourth activation function, o t ·σ(c t ) represents the contribution of the memory cell state to the hidden state controlled by the output gate, and the fourth activation function is used to ensure that the output value is between 0 and 1.

[0154] In some embodiments, the bidirectional long short-term memory network (BiLSTM) feeds the lithium-ion battery charge and discharge data into the forward LSTM hidden layer and the reverse LSTM hidden layer respectively, and the forward LSTM hidden layer output and the reverse LSTM hidden layer output are combined to finally output the sequence {h 1, , h2……h t In this disclosure, the number of input layers is defined as 128 and the number of hidden layers is defined as 100.

[0155] As a refinement of the above embodiment, when performing weighted processing on the target hidden state based on the preset attention mechanism in step 105 to obtain an indicator vector related to the current health state of the target battery, the following implementation methods may also be adopted but are not limited to, for example: calculating the attention score of the target hidden state based on an attention score function; calculating the weight value corresponding to the target hidden state based on the attention score based on a weight calculation function; calculating the indicator vector based on the attention score and the weight value based on a weighting function, wherein the preset attention mechanism includes the attention score function, the weight calculation function, and the weighting function;

[0156] The formula of the attention score function is expressed as:

[0157] α=tanh(W a ·h t )

[0158] Among them, W a is the attention weight matrix, h t is the target hidden state at time t, tanh(.) is the hyperbolic tangent function, and α is the attention score;

[0159] The formula of the weight calculation function is expressed as:

[0160] w=softmax(α)

[0161] Wherein, w is the weight value, α is the attention score, and softmax(.) is the fifth activation function;

[0162] The formula of the weighting function is expressed as:

[0163]

[0164] Where V is the indicator vector, t is the time step, w is the weight value, h t is the target hidden state at time t.

[0165] In short, define a linear layer to calculate the attention score, and use the softmax function to calculate the attention weight. First, calculate the attention score at time t, and then calculate the weight corresponding to the attention score. The larger the calculated weight, the more important the hidden layer feature at that moment. The vector h at that moment is t , the greater the contribution to the prediction result, then by inputting vector h at time t t Perform weighted calculation vector V t .

[0166] As a refinement of the above embodiment, the formula of the preset linear function in step 106 is expressed as:

[0167] O f =W o ·C t +b o

[0168] Where W o is the weight of the fully connected layer, b o is the bias term, O f is the maximum available capacity.

[0169] As a refinement of the above embodiment, when performing denoising optimization on the maximum available capacity based on the preset encoder-decoder to obtain the optimized maximum available capacity in step 107, the following implementation methods may also be used, but are not limited to, for example: performing nonlinear transformation processing on the maximum available capacity based on the encoder to obtain the nonlinear maximum available capacity; and performing linear transformation processing on the nonlinear maximum available capacity based on the decoder to obtain the optimized maximum available capacity, wherein the preset encoder-decoder comprises the encoder and the decoder;

[0170] The encoder formula is expressed as:

[0171] d1=ReLU(W·O f +b)

[0172] Wherein, W is the weight matrix of the encoder, b is the bias of the encoder, d1 is the maximum available capacity of the nonlinearity, O f is the maximum available capacity, ReLU(.) is the nonlinear transformation function;

[0173] The decoder formula is expressed as:

[0174] O d =W′·d1+b′

[0175] Where W′ is the weight matrix of the decoder, b′ is the bias of the decoder, d1 is the maximum available capacity of the nonlinearity, O dThis is the maximum available capacity after optimization.

[0176] In some embodiments, the health status of the lithium battery is defined as follows:

[0177]

[0178] Among them C max and C norm are the maximum available capacity and nominal capacity of the battery, respectively. In most applications, the failure threshold is typically set at 70% of the initial capacity. Based on the aforementioned definition of lithium battery health status, the target battery's health status can be determined based on the optimized maximum available capacity.

[0179] The preset encoder-decoder described in the present disclosure is a denoising autoencoder (DAE), which is an unsupervised learning method for reconstructing input data from a low-dimensional representation while preserving as much information as possible. During the training process, the loss function of the DAE (such as the mean square error) measures the difference between the reconstructed output and the original noise-free input. By minimizing this loss, the DAE forces the model to learn a reconstruction method that effectively removes the noise in the input data. The loss function is calculated as:

[0180]

[0181] Where, L d is the loss of the denoising encoder, n is the total number of data, is the input data with noise, is the reconstructed output, i.e., the denoised data generated from the input data by the DAE. t is the time step, φ(·) is the loss function, and the reconstruction error is usually measured using metrics such as mean squared error (MSE). λ is the regularization parameter that controls the model complexity. is the Frobenius norm of the encoder weight matrix, indicating the size of the weight, is the Frobenius norm of the decoder weight matrix, which indicates the size of the weight and helps prevent overfitting. Through the above structure, the model can extract features from the input sequence and make predictions, while using the autoencoder to perform denoising.

[0182] The present disclosure predicts the health status of lithium batteries based on the hybrid DAE-CNN-BiLSTM-Attention model designed in this disclosure. To verify the performance of the model designed in this disclosure, this disclosure uses two commonly used indicators, mean absolute error (MAE) and root mean squared error (RMSE), to quantify the performance of the model in predicting the health status of batteries. The calculation formula is:

[0183]

[0184] Where n represents the length of feature data, T represents the length of training feature data, x t The predicted and actual values ​​of the lithium battery SOH at time t are respectively. MAE is the mean absolute error between the predicted and actual values, measuring the average difference between the predicted and actual values; RMSE is the square root of the mean of the squared errors between the predicted and actual values, providing the standard deviation of the error. Smaller values ​​indicate better model performance. We compared the LSTM, At-LSTM, CNN-BiLSTM, CNN-BiLSTM-At, and DAE-CNN-BiLSTM-At models and conducted ablation experiments. Table 5 shows the prediction performance results on the dataset.

[0185] Table 5. Prediction performance results on the dataset (the best prediction result is bold)

[0186]

[0187] Table 5 shows that, with the exception of the CS2_36 model (RMSE of 0.0230, slightly lower than the CNN-BiLSTM-At model), and the CS2_38 model (MAE of 0.0227, slightly lower than the CNN-BiLSTM-At model), the proposed model outperforms other models in all other metrics, demonstrating its robustness and stability. The CALCE model achieves significantly lower metrics than NASA's, with lower errors. The best evaluation result for this model was achieved on the CS2_35 battery, with a mean absolute error (MAE) of 0.0154 and a root mean square error (RMSE) of 0.0191. Compared to the undenoised model, the MAE and RMSE performance improved by 55.4% and 3.14%, respectively. This demonstrates that the denoising step is effective in improving model predictions and enhances model performance. The proposed model exhibits lower MAE and RMSE across all datasets, demonstrating its superior predictive performance. From the perspective of the optimal number of predictions, NASA's prediction performance is better than CALCE, with a score of 4:2, which means that there is room for improvement in the prediction performance of large data sets.

[0188] Predicting the state of health of lithium-ion batteries: Figure 6 and Figure 7 Provides visualization of prediction results and associated errors for NASA and CALCE datasets. Figure 6 The curve results and error diagram of the NASA data set predicted by the disclosed design model provided in the embodiment of the present disclosure are as follows: Figure 7The curve results and error diagram of the CALCE data set predicted by the disclosed design model provided in the embodiment of the present disclosure are as follows: Figure 6 In the , the SOH predictions for NASA batteries are very close to the actual battery health status, and all prediction errors remain within 5% even at peak anomaly points. This shows that despite the anomalies, the model can still accurately track the degradation process. Similarly, Figure 7 The CALCE battery state-of-health predictions and error trends are described. In this publication, 70% of the battery data was used for training, and the model predicts the entire degradation process. Compared to the NASA dataset, the CALCE dataset presents a more challenging prediction scenario due to its significantly larger data volume and the presence of more anomalous noise. Despite these challenges, the model's predictions remain close, demonstrating its robustness and generalization across diverse datasets.

[0189] In summary, the embodiments of the present disclosure can achieve the following effects:

[0190] 1. By preprocessing the second preset amount of health factor data, the data quality is improved to obtain the preprocessed health factor data set, and the preprocessed health factor data set is processed in sequence based on the two convolutional layers, the bidirectional long short-term memory network, the preset attention mechanism, the fully connected layer and the preset encoder-decoder, thereby realizing the prediction of the target battery health state. In addition, by directly using the health factor data extracted from the historical operation data of the target battery to realize the prediction of the target battery health state, the process of constructing the target battery parameter model is avoided, so that the prediction method of the lithium battery health state disclosed in the present invention has strong adaptability and simple operation.

[0191] 2. The prediction model designed in this paper integrates the advantages of CNN in extracting local features, BiLSTM in learning the temporal dependencies between the past and the present, Attention in allocating the most favorable weights for prediction, and self-encoding and decoding to restore the original data, thereby improving the accuracy of the model's prediction results.

[0192] Corresponding to the above-mentioned lithium battery health state prediction method, the present invention also provides a lithium battery health state prediction device. Since the device embodiment of the present invention corresponds to the above-mentioned method embodiment, any details not disclosed in the device embodiment can be referred to the above-mentioned method embodiment and will not be further described in this invention.

[0193] Figure 8 This is a schematic diagram of the structure of a device for predicting the health status of a lithium battery provided by an embodiment of the present disclosure, such as Figure 8 As shown, including:

[0194] a selection unit 21 for selecting a second preset number of health factor data from a first preset number of health factor data based on a Pearson correlation coefficient, wherein the first preset number is greater than the second preset number;

[0195] a processing unit 22, configured to preprocess the second preset amount of health factor data to obtain a preprocessed health factor data set;

[0196] An extraction unit 23 is configured to perform feature extraction on the preprocessed health factor dataset based on two convolutional layers to obtain a feature data sequence;

[0197] An analysis unit 24 is configured to perform a long-term dependency analysis on the feature data sequence based on a bidirectional long short-term memory network to obtain a target hidden state;

[0198] a weighting unit 25 for performing weighted processing on the target hidden state based on a preset attention mechanism to obtain an indicator vector related to the current health state of the target battery;

[0199] a calculation unit 26, configured to process the indicator vector based on a preset linear function in a fully connected layer to obtain a maximum available capacity of the target battery;

[0200] The optimization unit 27 is configured to perform denoising optimization on the maximum available capacity based on a preset encoder-decoder to obtain an optimized maximum available capacity, and determine the current health state of the target battery according to the optimized maximum available capacity.

[0201] The present disclosure provides a device for predicting the health status of a lithium battery. The device selects a second preset number of health factor data from a first preset number of health factor data based on the Pearson correlation coefficient, where the first preset number is greater than the second preset number; preprocesses the second preset number of health factor data to obtain a preprocessed health factor data set; extracts features from the preprocessed health factor data set based on two convolutional layers to obtain a feature data sequence; performs long-term dependency analysis on the feature data sequence based on a bidirectional long short-term memory network to obtain a target hidden state; performs weighted processing on the target hidden state based on a preset attention mechanism to obtain an indicator vector related to the current health status of the target battery; processes the indicator vector based on a preset linear function in a fully connected layer to obtain the maximum available capacity of the target battery; performs denoising optimization on the maximum available capacity based on a preset encoder-decoder to obtain the optimized maximum available capacity, and determines the current health status of the target battery based on the optimized maximum available capacity. Compared with the related art, by preprocessing the second preset number of health factor data, the data quality is improved to obtain the preprocessed health factor data set, and the preprocessed health factor data set is processed in sequence based on the two convolutional layers, the bidirectional long short-term memory network, the preset attention mechanism, the fully connected layer and the preset encoder-decoder, thereby realizing the prediction of the target battery health status. In addition, by directly using the health factor data extracted from the historical operation data of the target battery to realize the prediction of the target battery health status, the process of constructing the target battery parameter model is avoided, so that the prediction method of the lithium battery health status disclosed in the present invention has strong adaptability and simple operation.

[0202] It should be noted that the above explanation of the method embodiment is also applicable to the device of this embodiment, and the principles are the same, which is not limited in this embodiment.

[0203] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0204] Figure 9 A schematic block diagram of an example electronic device 300 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are provided as examples only and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0205] like Figure 9 As shown, the device 300 includes a computing unit 301, which can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 302 or a computer program loaded from a storage unit 308 into a RAM (Random Access Memory) 303. Various programs and data required for the operation of the device 300 can also be stored in the RAM 303. The computing unit 301, ROM 302, and RAM 303 are connected to each other via a bus 304. An I / O (Input / Output) interface 305 is also connected to the bus 304.

[0206] Various components in device 300 are connected to I / O interface 305, including: an input unit 306, such as a keyboard, mouse, etc.; an output unit 307, such as various types of displays, speakers, etc.; a storage unit 308, such as a magnetic disk, optical disk, etc.; and a communication unit 309, such as a network card, modem, wireless communication transceiver, etc. The communication unit 309 allows device 300 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0207] The computing unit 301 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, a CPU (Central Processing Unit), a GPU (Graphic Processing Unit), various specialized AI (Artificial Intelligence) computing chips, various computing units that run machine learning model algorithms, a DSP (Digital Signal Processor), and any suitable processor, controller, microcontroller, etc. The computing unit 301 performs the various methods and processes described above, such as the lithium battery state of health prediction method. For example, in some embodiments, the lithium battery state of health prediction method can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 308. In some embodiments, part or all of the computer program can be loaded and / or installed on the device 300 via the ROM 302 and / or the communication unit 309. When the computer program is loaded into the RAM 303 and executed by the computing unit 301, one or more steps of the method described above can be performed. Alternatively, in other embodiments, the computing unit 301 may be configured to execute the aforementioned lithium battery health status prediction method in any other appropriate manner (for example, by means of firmware).

[0208] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application-Specific Standard Products), SOCs (System on Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special-purpose or general-purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0209] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0210] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or apparatus. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or apparatus, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include an electrical connection based on one or more wires, a portable computer disk, a hard disk, RAM, ROM, EPROM (Electrically Programmable Read-Only-Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0211] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0212] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: LAN (Local Area Network), WAN (Wide Area Network), the Internet, and blockchain networks.

[0213] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact via a communication network. This client-server relationship is established by computer programs running on the respective computers, establishing a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host, a host product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosts and VPS services ("Virtual Private Servers" or simply "VPS"). The server may also be a server in a distributed system or a server integrated with blockchain.

[0214] It's important to note that artificial intelligence (AI) is the study of how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). This encompasses both hardware and software technologies. AI hardware technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, and big data processing. AI software technologies primarily encompass computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graphs.

Claims

1. A method for predicting the health status of a lithium battery, characterized in that: include: Selecting a second preset number of health factor data from a first preset number of health factor data based on the Pearson correlation coefficient, wherein the first preset number is greater than the second preset number; preprocessing the second preset amount of health factor data to obtain a preprocessed health factor data set; Performing feature extraction on the preprocessed health factor dataset based on two convolutional layers to obtain a feature data sequence; Performing a long-term dependency analysis on the feature data sequence based on a bidirectional long short-term memory network to obtain a target hidden state; The target hidden state is weighted based on a preset attention mechanism to obtain an indicator vector related to the current health state of the target battery, specifically: Calculating an attention score for the target hidden state based on an attention score function; Calculate the weight value corresponding to the target hidden state according to the attention score based on a weight calculation function; Calculating the indicator vector according to the attention score and the weight value based on a weighting function, the preset attention mechanism includes the attention score function, the weight calculation function and the weighting function; The formula of the attention score function is expressed as: in, is the attention weight matrix, is the target hidden state at time t, is the hyperbolic tangent function, is the attention score; The formula of the weight calculation function is expressed as: in, is the weight value, is the attention score, is the fifth activation function; The formula of the weighting function is expressed as: Where V is the indicator vector, t is the time step, is the weight value, is the target hidden state at time t; Processing the indicator vector based on a preset linear function in a fully connected layer to obtain the maximum available capacity of the target battery; Performing denoising and optimization on the maximum available capacity based on a preset encoder-decoder to obtain an optimized maximum available capacity, and determining the current health state of the target battery based on the optimized maximum available capacity; performing denoising and optimization on the maximum available capacity based on the preset encoder-decoder to obtain the optimized maximum available capacity includes: Performing nonlinear transformation processing on the maximum available capacity based on an encoder to obtain a nonlinear maximum available capacity; performing a linear transformation process on the nonlinear maximum available capacity based on a decoder to obtain the optimized maximum available capacity, wherein the preset encoder-decoder comprises the encoder and the decoder; The encoder formula is expressed as: in, is the weight matrix of the encoder, b is the bias of the encoder, is the nonlinear maximum available capacity, The maximum available capacity, is a nonlinear transfer function; The decoder formula is expressed as: in, is the weight matrix of the decoder, is the decoder bias, is the nonlinear maximum available capacity, This is the maximum available capacity after optimization.

2. The method according to claim 1, characterized in that The selecting a second preset number of health factor data from the first preset number of health factor data based on the Pearson correlation coefficient includes: Calculating the Pearson correlation coefficients between the first preset number of health factor data and the historical health status of the target battery based on a correlation coefficient function to obtain the first preset number of Pearson correlation coefficients; The second preset number of Pearson correlation coefficients are selected from the first preset number of Pearson correlation coefficients according to the order of absolute value from large to small, and the second preset number of health factor data corresponding to the second preset number of Pearson correlation coefficients are determined.

3. The method according to claim 2, characterized in that The preprocessing of the second preset amount of health factor data to obtain a preprocessed health factor data set includes: performing normalization processing on the second preset amount of health factor data based on a normalization function to obtain a normalized health factor data set; Gaussian noise is added to the standardized health factor dataset based on a data enhancement function to obtain the preprocessed health factor dataset.

4. The method according to claim 3, characterized in that The formula of the correlation coefficient function is expressed as: in, a is the health factor data, b is the historical health status, is the average value of the health factor data, is the average value of the historical health status, A is the Pearson correlation coefficient, is the first subscript variable, and n is the sequence length of the health factor data; The formula of the normalization function is expressed as: in, is the target health factor data after normalization, X is the feature value selected in the target health factor data, is the minimum eigenvalue in the target health factor data, is the minimum eigenvalue in the target health factor data, and the target health factor data is one of the second preset number of health factor data; The formula of the data enhancement function is expressed as: in, is the standardized health factor dataset, is the standard deviation of Gaussian noise, N(0,1) represents a standard normal distribution with mean 0 and variance 1, represents the pre-processed health factor dataset.

5. The method according to claim 4, characterized in that The two convolutional layers include a feature extraction function, and the formula of the feature extraction function is expressed as: Wherein, H is the feature data sequence, K is the size of the convolution kernel, is the pre-processed health factor dataset, is the second subscript variable, t is the time step, k is the weight value, Indicates The weight value at .

6. The method according to claim 5, characterized in that The bidirectional long short-term memory network includes an input gate, a forget gate, an output gate, a memory unit update function and a hidden state update function; The formula of the input gate is expressed as: in, is the input gate weight matrix, is the input gate activation value, is the target hidden state at time t-1, is the input data at the current time t, (.) is the first activation function; The formula of the forget gate is expressed as: in, is the forget gate weight matrix, is the activation value of the forget gate, is the target hidden state at time t-1, is the input data at the current time t, (.) is the second activation function; The formula of the output gate is expressed as: in, is the forget gate weight matrix, is the activation value of the forget gate, is the target hidden state at time t-1, is the input data at the current time t, (.) is the third activation function; The formula of the memory unit update function is expressed as: in, is the state of the memory unit at time t-1, is the candidate memory cell state, It is used to control the contribution of the memory cell state at time t-1 to the memory cell state at time t. For control right Contribution, is the state of the memory unit at time t, is the activation value of the forget gate, is the input gate activation value. The formula of the hidden state update function is expressed as: in, is the target hidden state at time t, is the activation value of the forget gate, is the state of the memory unit at time t, (.) is the fourth activation function.

7. The method according to any one of claims 1 to 6, characterized in that Before selecting a second preset amount of health factor data from the first preset amount of health factor data based on the Pearson correlation coefficient, the method includes: Acquiring historical operating data of the target battery during the discharge process; The first preset amount of health factor data is extracted from the historical operating data.

8. A device for predicting the health status of a lithium battery, characterized in that: include: a selection unit, configured to select a second preset number of health factor data from a first preset number of health factor data based on a Pearson correlation coefficient, wherein the first preset number is greater than the second preset number; a processing unit, configured to preprocess the second preset amount of health factor data to obtain a preprocessed health factor data set; an extraction unit, configured to perform feature extraction on the preprocessed health factor dataset based on two convolutional layers to obtain a feature data sequence; An analysis unit, configured to perform a long-term dependency analysis on the feature data sequence based on a bidirectional long short-term memory network to obtain a target hidden state; a weighting unit, configured to perform weighted processing on the target hidden state based on a preset attention mechanism to obtain an indicator vector related to the current health state of the target battery; a calculation unit, configured to process the indicator vector based on a preset linear function in a fully connected layer to obtain a maximum available capacity of the target battery; An optimization unit is configured to perform denoising optimization on the maximum available capacity based on a preset encoder-decoder to obtain an optimized maximum available capacity, and determine a current health state of the target battery according to the optimized maximum available capacity.

Images