Lithium battery residual life prediction method and system based on machine learning algorithm
The lithium battery data is preprocessed and feature decomposed through machine learning algorithms, and the prediction is performed using the TCN-Attention model, which solves the problem of inaccurate prediction of lithium battery life and achieves higher prediction accuracy.
Patent Information
- Application Number
- CN202510637419.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-19
- Publication Date
- 2025-07-04
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art is difficult to accurately predict the remaining life of lithium batteries, which affects the service life and performance of the equipment.
Using machine learning algorithms, the battery data is acquired for preprocessing, and the health data is decomposed using the VMD method to obtain feature components, and input it into the TCN-Attention model for prediction.
It improves the accuracy of lithium battery life prediction, can accurately obtain global minimum values and ensure complete information.
Smart Images
Figure CN120254641A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of lithium battery life prediction, and particularly to a method and system for predicting the remaining life of a lithium battery based on a machine learning algorithm. Background Art
[0002] As an efficient, environmentally friendly, and portable power storage device, lithium batteries are widely used in fields such as mobile communication, electric transportation, and energy storage systems. However, as the usage time of lithium batteries increases, their performance gradually deteriorates, resulting in a decline in energy storage capacity and affecting the service life and performance of devices. Therefore, predicting the remaining life of lithium batteries has become crucial.
[0003] Therefore, how to provide a method and system for predicting the remaining life of a lithium battery based on a machine learning algorithm to solve the difficulties existing in the prior art is an urgent problem to be solved by those skilled in the art. Summary of the Invention
[0004] In view of this, the present invention provides a method and system for predicting the remaining life of a lithium battery based on a machine learning algorithm, which can accurately obtain the global minimum value and ensure complete information, effectively improving the accuracy of lithium battery life prediction.
[0005] To achieve the above object, the present invention adopts the following technical solutions:
[0006] A method for predicting the remaining life of a lithium battery based on a machine learning algorithm includes the following steps:
[0007] Obtain battery data, and preprocess the battery data to obtain preprocessed battery data;
[0008] Based on the preprocessed battery data analysis, obtain the battery capacity, and further obtain the battery health data;
[0009] Use the VMD method to decompose the battery health data to obtain characteristic components;
[0010] Input the characteristic components into the TCN-Attention model to obtain battery life prediction data.
[0011] Optionally, obtaining battery data includes: obtaining voltage information, current information, charge and discharge time information, cycle number information, and temperature information.
[0012] Optionally, the preprocessing includes: cleaning the battery data to remove noise and abnormal data, and using data filtering to reduce the influence of high-frequency noise to obtain preprocessed battery data.
[0013] Optionally, the battery capacity includes: drawing a current-voltage curve based on the charge and discharge time information in combination with the voltage information and current information, and calculating the battery capacity in combination with the ampere-hour integration method.
[0014] Optionally, the VMD method includes:
[0015] Taking the battery health data as the original signal and converting it into multiple modal components;
[0016] Setting constraint conditions to make the sum of the modal components consistent with the original signal;
[0017] Initializing the training parameters in the constraint conditions and determining the maximum number of iterations;
[0018] Training the constraint conditions and stopping after the constraint conditions converge or reach the maximum number of iterations.
[0019] Optionally, the TCN-Attention model includes:
[0020] Obtaining a training data set;
[0021] Constructing a TCN network including an input layer, a hidden layer, and an output layer, and setting network parameters, where the network parameters are the filter coefficients and dilation coefficients in the hidden layer;
[0022] Constructing an Attention rule and adding it to the TCN network to form a TCN-Attention model;
[0023] Training the TCN-Attention model using the training data set to optimize the network parameters;
[0024] Judging whether the training times are reached. If so, generating a trained TCN-Attention model.
[0025] Optionally, the Attention rule includes: weighting different features based on the feature sequence to increase the weight of key features in the hidden layer.
[0026] A lithium battery remaining life prediction system based on a machine learning algorithm, which is used to execute a lithium battery remaining life prediction method based on a machine learning algorithm described in any one of the above, including a data acquisition module, a battery capacity acquisition module, a feature component construction module, and a life prediction module connected in sequence;
[0027] Data acquisition module: acquiring battery data and preprocessing the battery data to obtain preprocessed battery data;
[0028] Battery capacity acquisition module: analyzing the preprocessed battery data to obtain the battery capacity, and further obtaining the battery health data;
[0029] Feature component construction module: using the VMD method to decompose the battery health data to obtain feature components;
[0030] Lifetime prediction module: Input the feature components into the TCN-Attention model to obtain battery lifetime prediction data.
[0031] As can be seen from the above technical solutions, compared with the prior art, the present invention provides a method and system for predicting the remaining lifetime of a lithium battery based on a machine learning algorithm, having the following beneficial effects: 1) The present invention collects battery data, providing a sufficient basis for subsequent data analysis and improving the accuracy of comprehensively reflecting the state of the lithium battery pack; 2) The TCN-Attention model proposed by the present invention can accurately obtain the global minimum value and ensure the integrity of information, effectively improving the accuracy of lithium battery lifetime prediction. Description of the Drawings
[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the provided drawings.
[0033] Figure 1 It is a flowchart of a method for predicting the remaining lifetime of a lithium battery based on a machine learning algorithm disclosed by the present invention;
[0034] Figure 2 It is a block diagram of a system for predicting the remaining lifetime of a lithium battery based on a machine learning algorithm disclosed by the present invention. Detailed Embodiments
[0035] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.
[0036] Refer to Figure 1 As shown, the present invention discloses a method for predicting the remaining lifetime of a lithium battery based on a machine learning algorithm, including the following steps:
[0037] Obtain battery data and preprocess the battery data to obtain preprocessed battery data;
[0038] Based on the analysis of the preprocessed battery data, obtain the battery capacity, and then obtain the battery health data;
[0039] Use the VMD method to decompose the battery health data to obtain feature components;
[0040] Input the feature components into the TCN-Attention model to obtain battery life prediction data.
[0041] Furthermore, obtaining battery data includes: obtaining voltage information, current information, charge and discharge time information, number of charge and discharge cycles information, and temperature information.
[0042] Furthermore, the preprocessing includes: cleaning the battery data to remove noise and abnormal data, using data filtering to reduce the impact of high-frequency noise, and obtaining preprocessed battery data.
[0043] Specifically, removing noise and abnormal data includes deleting records containing noise and abnormal data or treating abnormal data as missing data: when there is a large amount of noise or abnormal data in the collected data set, the data set is deleted; when there is a small amount of noise or abnormal data in the collected data set, calculate the average value of adjacent data to replace the original noise or abnormal data.
[0044] To reduce high-frequency noise, the binning method is used for data smoothing to eliminate high-frequency noise, including: dividing the data into several data segments with the same number of data, and then replacing each value in the bin with the bin median to obtain smoothed data.
[0045] Furthermore, the battery capacity includes: drawing a current-voltage curve based on the charge and discharge time information in combination with the voltage information and current information, and calculating the battery capacity in combination with the ampere-hour integration method.
[0046] Specifically, the ampere-hour integration method is one of the most commonly used methods for accumulating electric energy. By calculating the integral of current and charge and discharge time over a period of time, calculating the percentage of the changed electric energy, and then finding the difference between the initial battery capacity and the changed battery capacity, which is the remaining battery capacity, and performing multiple charge and discharge cycles on the battery.
[0047] Furthermore, obtaining the battery health data SOH is based on the battery rated capacity Q N and the actual battery capacity Q real obtained, and the expression is:
[0048]
[0049] Furthermore, the VMD method includes:
[0050] Taking the battery health data as the original signal and converting it into multiple modal components;
[0051] Setting constraint conditions to make the sum of the modal components consistent with the original signal;
[0052] Initializing the training parameters in the constraint conditions and determining the maximum number of iterations;
[0053] Train the constraint conditions and stop when the constraint conditions converge or reach the maximum number of iterations.
[0054] Specifically, VMD is an adaptive and completely non - recursive method for modal variational and signal processing. Based on the three concepts of classical Wiener filtering, frequency mixing, and Hilbert transform, it decomposes complex non - linear sequences into multiple stable linear sequences, ultimately minimizing the sum of the estimated bandwidths of each mode. Based on this, the social constraint variation of battery data is as follows:
[0055]
[0056] where k is the number of decomposed modes, m t is the time derivative of the function, δ(t) is the Dirac function, u k (·) is the k - th modal component after decomposition, ω t (t) is the k - th center frequency after decomposition, t is time. Solve it and introduce the multiplication operator Z to convert the constrained variational problem into an unconstrained variational problem, and use the alternating direction multiplier method to solve the variational problem, finally obtaining the stationary intrinsic mode components as output.
[0057] Furthermore, the TCN - Attention model includes:
[0058] Obtain the training data set;
[0059] Construct a TCN network including an input layer, hidden layers, and an output layer, and set the network parameters. The network parameters are the filter coefficients and dilation coefficients in the hidden layers;
[0060] Construct the Attention rule and add it to the TCN network to form the TCN - Attention model;
[0061] Use the training data set to train the TCN - Attention model and optimize the network parameters;
[0062] Judge whether the training times are reached. If so, generate the trained TCN - Attention model.
[0063] Furthermore, TCN is a network that combines one - dimensional fully convolutional network and causal convolution, which can effectively extract the correlation between data and is more suitable for solving time - series problems. Its main structure is dilated causal convolution and residual units. TCN continuously increases or decreases the number of convolutional layers, thus forming a larger dilation coefficient and a larger convolutional kernel layer by layer. For the input of one - dimensional sequences, the convolutional kernel can expand the receptive field through the filter coefficient j and the dilation coefficient d. Then the dilated convolution expression is:
[0064]
[0065] Among them, f(j) is the filter function, and s - dj is the historical data in the input sequence. As the network depth increases, the training process becomes increasingly difficult, and the multi - layer backpropagation of the error signal will cause gradient dispersion or gradient explosion. Therefore, residual units are introduced, including 2 sequentially connected residual blocks. The residual block includes: a one - dimensional dilated causal convolution module, a weight normalization module, an activation function module, and a dropout operation module.
[0066] Furthermore, the Attention rule includes: weighting different features based on the feature sequence to increase the weight of key features in the hidden layer.
[0067] Specifically, the definition of the attention mechanism is to determine the hidden state y generated at each specific time point i , and take the vector x n as the weighted average of the state sequence y, and the expression is:
[0068]
[0069] where N is the total number of time steps of the input sequence, and α ni is the weight calculated for each state y i at time step n.
[0070] Furthermore, the TCN - Attention model includes an input layer, a TCN layer, an Attention layer, and an output layer. First, the stationary intrinsic mode components are input into the TCN layer; secondly, the Attention layer strengthens the feature weights of the hidden layer in the TCN layer, so as to process and analyze the data input into the TCN layer; finally, the prediction result is output through the output layer.
[0071] Furthermore, it also includes using an error evaluation function to analyze the effectiveness and accuracy of the prediction model.
[0072] A lithium - battery remaining life prediction system based on a machine - learning algorithm is used to execute a lithium - battery remaining life prediction method based on a machine - learning algorithm described in any one of the above, as shown in Figure 2 , and includes a data acquisition module, a battery capacity acquisition module, a feature component construction module, and a life prediction module connected in sequence;
[0073] The data acquisition module: acquires battery data and pre - processes the battery data to obtain pre - processed battery data;
[0074] The battery capacity acquisition module: analyzes the pre - processed battery data to obtain the battery capacity, and further obtains the battery health data;
[0075] Feature component construction module: The variational mode decomposition (VMD) method is used to decompose the battery health data to obtain feature components;
[0076] Battery life prediction module: The feature components are input into the TCN-Attention model to obtain battery life prediction data.
[0077] In a specific embodiment, the original data is analyzed, and the equal-depth binning method is used to process the noise values and outliers in the data. Then, the VMD method is used to decompose the preprocessed data to obtain multiple stationary intrinsic mode functions (IMFs). The multiple stationary mode components IMF are respectively input into the TCN-Attention model. The TCN-Attention model is used to calculate, analyze, and predict the components, and the mean squared error (MSE) function is selected to calculate the error of the obtained data.
[0078] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for predicting the remaining life of a lithium battery based on a machine learning algorithm, comprising the following steps: Obtain battery data, and preprocess the battery data to obtain preprocessed battery data; Analyze the preprocessed battery data to obtain the battery capacity, and further obtain the battery health data; Use the VMD method to decompose the battery health data to obtain characteristic components; Input the characteristic components into the TCN-Attention model to obtain battery life prediction data.
2. The method for predicting the remaining life of a lithium battery based on a machine learning algorithm according to claim 1, wherein Obtaining battery data includes: obtaining voltage information, current information, charge and discharge time information, number of cycles information, and temperature information.
3. The method for predicting the remaining life of a lithium battery based on a machine learning algorithm according to claim 1, wherein Preprocessing includes: cleaning the battery data to remove noise and abnormal data, and using data filtering to reduce the influence of high-frequency noise to obtain preprocessed battery data.
4. The method for predicting the remaining life of a lithium battery based on a machine learning algorithm according to claim 2, wherein The battery capacity includes: drawing a current-voltage curve based on the charge and discharge time information in combination with the voltage information and current information, and calculating the battery capacity in combination with the ampere-hour integration method.
5. The method for predicting the remaining life of a lithium battery based on a machine learning algorithm according to claim 1, wherein The VMD method includes: Taking the battery health data as the original signal and converting it into multiple modal components; Setting constraint conditions to make the sum of the modal components consistent with the original signal; Initializing the training parameters in the constraint conditions and determining the maximum number of iterations; Training the constraint conditions, and stopping when the constraint conditions converge or reach the maximum number of iterations.
6. The method for predicting the remaining life of a lithium battery based on a machine learning algorithm according to claim 1, wherein The TCN-Attention model includes: Obtaining a training data set; Constructing a TCN network including an input layer, a hidden layer, and an output layer, and setting network parameters, where the network parameters are the filter coefficients and dilation coefficients in the hidden layer; Constructing an Attention rule and adding it to the TCN network to form a TCN-Attention model; Training the TCN-Attention model using the training data set to optimize the network parameters; Judging whether the training times are reached. If so, generate a trained TCN-Attention model.
7. The method for predicting the remaining life of a lithium battery based on a machine learning algorithm according to claim 6, wherein The Attention rule includes: weighting different features based on the feature sequence to increase the weight of key features in the hidden layer.
8. A lithium battery remaining life prediction system based on a machine learning algorithm, which is used to execute the method for predicting the remaining life of a lithium battery based on a machine learning algorithm according to any one of claims 1-7, characterized in that, It includes a data acquisition module, a battery capacity acquisition module, a characteristic component construction module, and a life prediction module connected in sequence; Data acquisition module: Obtain battery data, and preprocess the battery data to obtain preprocessed battery data; Battery capacity acquisition module: Analyze the preprocessed battery data to obtain the battery capacity, and further obtain the battery health data; Feature component construction module: Decompose the battery health data using the VMD method to obtain feature components; Life prediction module: Input the feature components into the TCN-Attention model to obtain battery life prediction data.