A lithium battery capacity prediction method based on chaos theory and nonlinear system

Abstract

The present invention discloses a lithium battery capacity prediction method based on chaos theory and nonlinear system. The method deeply explores the intrinsic dynamic characteristics of lithium battery formation and capacity division data based on nonlinear analysis method, constructs a chaotic phase space containing rich information, and provides an excellent feature space for discharge capacity prediction. Therefore, feature dimensionality reduction extraction can be performed on the feature space, and the extracted dimensionality reduction features are filled into the training set, and the machine learning model is trained using the filled training set, so as to establish a lithium battery capacity prediction model. The present invention breaks through the static and precision bottlenecks of traditional methods, can dynamically and highly accurately predict the discharge capacity of lithium batteries, and greatly reduce the time and energy consumption of the capacity division process.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery production, in particular to a method for predicting battery discharge capacity based on a lithium battery formation process, and specifically to a method for predicting lithium battery capacity based on chaos theory and nonlinear system. Background Art

[0002] Lithium batteries have been widely used in consumer electronics, electric vehicles, aerospace and other fields due to their unique advantages such as high energy density, high discharge power and low cost. However, due to the complex electrochemical system, various failure mechanisms and differences in the manufacturing process, lithium batteries in the same batch often have large discreteness, which inevitably increases the difficulty of judging the process and time of lithium battery discharge capacity. The performance of battery modules composed of lithium batteries with large differences is poor and cannot meet the market's usage needs. Therefore, accurate capacity prediction of lithium batteries is one of the issues of concern in the lithium battery manufacturing process.

[0003] In actual production, lithium batteries need to undergo a 100% discharge test, which takes 6 to 8 hours to measure a more accurate discharge capacity. This part of the production process seriously increases the space, equipment and time required for production.

[0004] At present, most of the existing methods for predicting the discharge capacity of lithium batteries are linear analysis models. By studying the relationship between the voltage, current, temperature and other parameters during the formation and capacity division of lithium batteries and the discharge capacity of lithium batteries, the discharge capacity of lithium batteries can be calculated in advance during the production process. However, this method also has certain defects. For example, the characteristics of this method are based on traditional linear analysis methods, which cannot reflect the dynamic characteristics and complex nonlinear factors of lithium batteries. The generalization effect of the fitted model is also poor and cannot be applied to other lithium battery data. The prediction accuracy is not ideal and is far from the accuracy required by lithium battery manufacturers. In addition, the stability is insufficient, and it is difficult to apply in practice. Summary of the invention

[0005] The purpose of the present invention is to solve the above-mentioned defects in the prior art and to provide a lithium battery capacity prediction method based on chaos theory and nonlinear systems. This method breaks through the static and precision bottlenecks of traditional methods, obtains the internal kinetic energy characteristics of the lithium battery by performing dynamic and nonlinear analysis on the lithium battery production process data, and predicts the discharge capacity of the lithium battery with high precision, thereby eliminating the time and energy consumption of the capacity division process.

[0006] The purpose of the present invention can be achieved by adopting the following technical solutions:

[0007] A lithium battery capacity prediction method based on chaos theory and nonlinear system, the lithium battery capacity prediction method comprising the following steps:

[0008] S1. Original voltage data of lithium battery Perform second-order difference calculation to obtain the second-order difference sequence of each lithium battery Its function expression is shown as follows:

[0009]

[0010] in, Respectively represent the voltage values ​​of the lithium battery at time n, n+1, and n+2, Represents the second-order difference data sequence of lithium batteries;

[0011] S2, second-order difference sequence of lithium battery Perform Euclidean distance calculation under the sliding window and set the size of the sliding window The calculated Euclidean distance is compared with the distance threshold ε. The sliding window of the sequence greater than ε is shifted forward by one data point and then calculated and compared again until it is less than or equal to ε and stops. The stop position is the synchronization point of the voltage data of each lithium battery. The function expression for calculating the distance is shown in the following formula:

[0012]

[0013] in, Indicates The first lithium battery differential data sequence data points, Indicates that except for The jth data point in the differential data sequence of other lithium batteries that are considered as lithium batteries, Indicates A sliding window of lithium battery differential data, Indicates that except for A sliding window of the differential data of other lithium batteries of a lithium battery, Represents the deviation between differential data, Indicates the difference between The second-order difference data of lithium batteries;

[0014] S3. The original voltage data is aligned to obtain a new sequence De-noising is performed to remove the noise that masks weak features. A Gaussian filter is used to perform one-dimensional linear smoothing filtering on the raw voltage data of the lithium battery. The function expression is as follows:

[0015]

[0016] in, represents the voltage data after noise reduction at time n, represents the voltage data before noise reduction at time n, σ represents the standard deviation of the voltage data;

[0017] S4. Use the noise-reduced data Perform nonlinear feature extraction and map the battery voltage data to phase space. Since the formation process of lithium batteries is a dynamic process of charging, traditional linear analysis such as mean, kurtosis, and first-order difference cannot represent this dynamic process. It is necessary to map the formation data to a multidimensional vector space to represent the dynamic process of lithium batteries. Set the embedding dimension of the mapping operation. Time delay τ, find the vector

[0018]

[0019] in, Representation vector The dimension of , τ represents the time delay;

[0020] S5. Create a two-layer neural network feature extraction model to extract the dynamic data of lithium batteries As input data, data As output data, the feature extraction model is fitted with training data. The loss function of the feature extraction model uses the root mean square error (RMSE). When the error e is less than or equal to the set threshold ε, the model stops and extracts the hidden layer parameters as the dynamic characteristics of the lithium battery. The error e is calculated as follows:

[0021]

[0022] Where u represents the number of output data;

[0023] S6. Divide the lithium battery feature data set into a training set, a test set, and a validation set according to the distribution of discharge capacity;

[0024] S7. Take the extracted dynamic features of the lithium battery as input data and the lithium battery discharge capacity data as target data, establish a gradient boosting tree (GBDT) model according to the grouping of lithium batteries in high-dimensional space, calculate the functional relationship between the dynamic features and the discharge capacity, and obtain the lithium battery capacity.

[0025] Furthermore, the actual lithium battery voltage sampling is a finite sequence with noise, and the embedding dimension The time delay τ must be selected according to different battery types or production processes. In step S4, the Mutual Information Method is used to calculate the appropriate time delay τ. From the principle of information theory, it calculates the maximum possible uncorrelation between the original data and the data after the delay τ, which enables us to extract the most possible uncorrelated features from the similar sampling data of each lithium battery. The process is as follows:

[0026]

[0027] To standardize:

[0028]

[0029] in, Indicates the voltage sampling of the lithium battery at time l, Indicates the voltage sampling of the lithium battery at time l+τ, express The number of voltage sampling data, express The number of voltage sampling data, and Respectively represent and The information entropy of express and The information entropy of the joint distribution, is a function of the delay time τ, and is calculated The τ corresponding to the first minimum value obtained is the optimal delay time τ.

[0030] Furthermore, the embedding dimension The complexity and calculation amount of the subsequent model are closely related to whether the dynamic characteristics of lithium batteries can be extracted. Therefore, reasonable dimension selection is crucial. In step S4, the minimum pseudo neighbor method is used to calculate the embedding dimension. When the embedding dimension is continuously increased and the difference before and after the increase is calculated, and the difference is found to have no significant change by comparing the difference, it is proved that the current embedding dimension is the most suitable dimension to represent the lithium battery data. The process is as follows:

[0031]

[0032] in, Indicated in Different from The other vectors of express The vector formed by dimensional collection, express The vector formed by dimensional collection, Indicated in Different from The other vectors of Indicates the difference between The denoised data of lithium batteries, r represents the threshold of the distance between dimensions, σ represents the standard deviation of the data, represents the Heaviside step function, when x≤0, When x>00, express and The gap between dimensions, e represents and The threshold of the gap between dimensions. If the gap is less than the threshold, it means that the current Dimension is the optimal dimension.

[0033] Furthermore, there is a certain functional relationship between the dynamic characteristics of lithium batteries and their discharge capacity. Deep neural networks have powerful expressiveness, and finding functional relationships between variables is the first choice. However, the prerequisite for using deep neural networks is that a large amount of training data is required, and the actual lithium battery production capacity cannot meet the data volume requirements of deep neural networks. Therefore, a gradient boosting tree based on statistical methods that does not require huge training data and has very mature technology is selected. The gradient boosting tree (Gradient Boosting Decision Tree, GBDT) is an additive model based on the idea of ​​boosting integration. The forward distribution algorithm is used for greedy learning during training. Each iteration learns a CART (Classification and Regression Tree) tree to fit the previous The residual between the prediction result of the tree and the true value of the training sample, where Represents the gradient boosting tree The CART tree is a binary tree that generates two child nodes at each split with a strategy that minimizes the sample variance of the two nodes. The gradient boosting tree can be found in "Journal of Computer and System Sciences", Yoav Freund, Robert E Schapire, August 1997, Pages 119-139.

[0034] Compared with the prior art, the present invention has the following advantages and effects:

[0035] Conventional lithium battery capacity prediction is a linear analysis method for lithium battery formation and capacity data, but the actual lithium battery is a complex collection of many nonlinear factors, and the linear model cannot effectively represent the nonlinear factors of lithium batteries. The present invention provides a lithium battery capacity prediction method based on nonlinear analysis and chaotic phase space characteristics. The method is based on a nonlinear analysis algorithm, further mining the inherent dynamic characteristics of lithium batteries from lithium battery formation and capacity data, reconstructing the charging and discharging trajectories of lithium batteries on the constructed chaotic phase space, and using the rich information of the chaotic phase space and neural network to extract dimensionality reduction features, and filling the dimensionality reduction features into the training set. The capacity prediction model trained in this way can accurately and reliably predict the discharge capacity of lithium batteries, providing an effective method with reference significance for the actual production needs of lithium batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0037] Figure 1 This is a schematic diagram of the lithium battery voltage before synchronization in the present invention;

[0038] Figure 2 It is a schematic diagram of the lithium battery voltage after synchronization in the present invention;

[0039] Figure 3 It is a schematic diagram of the lithium battery voltage feature extraction model in the present invention;

[0040] Figure 4 This is a schematic diagram of the characteristic distribution of lithium batteries No. 1 to No. 100 in the present invention;

[0041] Figure 5 It is a schematic diagram of the characteristics and discharge capacity scatter points of lithium batteries No. 500 to No. 599 in the present invention;

[0042] Figure 6 The present invention discloses a flow chart of a lithium battery capacity prediction method based on chaos theory and nonlinear system. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] Example 1

[0045] The following will further explain in detail the lithium battery capacity prediction method based on chaos theory and nonlinear system of the present invention in combination with the actual formation data of a batch of 3000 lithium batteries. The present embodiment discloses a lithium battery capacity prediction method based on chaos theory and nonlinear system, and the specific implementation steps are as follows:

[0046] Step T1: The original formation data records the entire charging process of each battery from the beginning to the end of activation. For example, the formation data of a lithium battery is expressed as All lithium battery data sets are Since the activation state of lithium batteries is discrete during the production process, the original data of lithium batteries are aligned to make them consistent. The alignment data is as follows: Figure 1 shown.

[0047] Step T2: Use a Gaussian filter to remove noise from the aligned data, making the data smoother. Because there is noise in the voltage sampling of lithium batteries, the amplitude of the noise will mask the weak characteristics of the voltage. Effectively removing part of the noise is more conducive to presenting the true characteristics of lithium batteries.

[0048] Step T3: Choose the appropriate embedding dimension and time delay τ, mapping the lithium battery formation data into phase space, and obtaining 3000 sets of vector groups In this experiment, the appropriate time delay τ = 3 and embedding dimension τ are calculated according to the mutual information method and the minimum pseudo neighbor method. The phase space of each lithium battery formation data consists of 100 vectors and is represented as

[0049] Step T4: Input the 3000 vector groups into the feature extraction model for feature extraction. In this embodiment, the number of neurons in the input layer and the output layer of the feature extraction model is The number of neurons in the hidden layer is 1000, and the activation function of the neuron is the hyperbolic tangent function (tanh), and its function expression is shown in the following formula: According to the embedding dimension Establish a lithium battery data feature extraction model, such as Figure 3 shown.

[0050] In this embodiment, the vector group of the lithium battery phase space is As input data, the output vector group The training is carried out, and after the training is completed, the hidden layer is extracted as the lithium battery feature.

[0051] from Figure 5It can be seen that the extracted lithium battery features show a normal distribution superposition, and at the same time have a positive or negative proportional relationship with the lithium battery discharge capacity, and distinguish abnormal points, indicating that the extracted lithium battery features can reflect the discharge capacity.

[0052] Step T5: Divide the 3000 battery features into a training set, a test set and a validation set according to the distribution of discharge capacity. In this embodiment, the training set consists of 2800 batteries, and the test set and validation set each consist of 100 batteries.

[0053] Step T6: Use the training set data to train the capacity prediction model, the test set data to verify the overfitting point of the model, and the verification set is used to select the most suitable model. In this embodiment, the gradient boosting tree (GBDT) is used to establish the model, the input feature number is 1000, the output is the discharge capacity, and the root mean square error (RMSE) is used as the loss function. The prediction results are shown in Table 1.

[0054] Table 1. Lithium battery capacity prediction error table based on chaos theory and nonlinear system

[0055] Validation set Feature 1 Feature 2 Feature 3 Feature 4 Forecast capacity Actual capacity error No. 1 battery 0.099311 -0.02577 -0.46747 0.018658 49.91695 49.960 0.09％ AA battery 0.101807 0.452593 -0.15426 0.491103 49.5366 49.512 0.05％ AA battery -0.44202 0.115126 0.478594 -0.57585 49.78867 49.750 0.08％ AA battery 0.199478 0.323783 -0.13992 -0.89576 50.50497 50.464 0.08％ AA battery -0.04682 -0.44916 -0.15221 -0.59996 50.22633 50.212 0.03％

[0056] As can be seen from Table 1, the method has good prediction results for the battery discharge capacity of the validation set, and the predicted value is not much different from the actual value. From the quantitative index point of view, the maximum error is within 0.1%, the error is small, and the prediction result is relatively accurate, which verifies that the present application can accurately predict the discharge capacity of lithium batteries.

[0057] Example 2

[0058] Based on the lithium battery capacity prediction method based on chaos theory and nonlinear system disclosed in the above embodiment 1, this embodiment further discloses a lithium battery capacity prediction method based on chaos theory and nonlinear system. The specific implementation steps are based on the steps in embodiment 1, and the different steps are as follows:

[0059] Step T1: Replace 2000 sets of lithium battery capacity data. The capacity division station in the lithium battery production process will give the lithium battery a full charge and discharge process. The capacity at the end of the final discharge is marked as the discharge capacity of the lithium battery. We select 100% charging data and 30% discharge data in the capacity division data and input them into the model.

[0060] Step T3: In this experiment, the appropriate time delay τ = 2 and embedding dimension are calculated according to the mutual information method and the minimum pseudo neighbor method. The phase space of each lithium battery capacity data consists of 83 vectors.

[0061] Step T5: Divide the 2000 battery features into a training set, a test set and a validation set according to the distribution of discharge capacity. In this embodiment, the training set consists of 1800 batteries, and the test set and validation set each consist of 100 batteries.

[0062] Step T6: Use the training set data to train the capacity prediction model, the test set data to verify the overfitting point of the model, and the verification set is used to select the most suitable model. In this embodiment, the gradient boosting tree (GBDT) is used to establish the model, the input feature number is 1000, the output is the discharge capacity, and the root mean square error (RMSE) is used as the loss function. The prediction results are shown in Table 2.

[0063] Table 2. Lithium battery capacity prediction error table based on chaos theory and nonlinear system

[0064] Validation set Feature 1 Feature 2 Feature 3 Feature 4 Forecast capacity Actual capacity error No. 1 battery 0.687521 -0.44695 -0.35818 -0.22531 2.5012 2.4956 0.2244％ AA battery -0.23098 -0.10302 -0.40036 0.227802 2.5025 2.5001 0.0960％ AA battery -0.22456 -0.14869 -0.04745 -0.52779 2.5114 2.5081 0.1316％ AA battery -0.33269 -0.11021 0.082826 -0.13842 2.5128 2.5153 0.0994％

[0065] From the comparison between Table 2 and Table 1, it can be seen that although the performance of this method is better on lithium batteries with larger capacity, there are two main reasons. On the one hand, the production process of large-capacity lithium batteries takes longer, and more information can be extracted from the data. On the other hand, from the gap between battery characteristics, it can be seen that the details of large-capacity lithium batteries change more obviously. However, the results of this method in calculating the discharge capacity of lithium batteries with smaller capacity are also within the acceptable range of the lithium battery production industry.

[0066] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A lithium battery capacity prediction method based on chaos theory and nonlinear system, characterized in that: The lithium battery capacity prediction method comprises the following steps: S1. Original voltage data of lithium battery Perform second-order difference calculation to obtain the second-order difference sequence of each lithium battery Its function expression is shown as follows: in, Respectively represent the voltage values ​​of the lithium battery at time n, n+1, and n+2, Represents the second-order difference data sequence of lithium batteries; S2, second-order difference sequence of lithium battery Perform Euclidean distance calculation under the sliding window and set the size of the sliding window The calculated Euclidean distance is compared with the distance threshold ε. The sliding window of the sequence greater than ε is shifted forward by one data point and then calculated and compared again until it is less than or equal to ε and stops. The stop position is the synchronization point of the voltage data of each lithium battery. The function expression for calculating the distance is shown in the following formula: in, Indicates The first lithium battery differential data sequence data points, Indicates that except for The jth data point in the differential data sequence of other lithium batteries that are considered as lithium batteries, Indicates A sliding window of lithium battery differential data, Indicates that except for A sliding window of the differential data of other lithium batteries of a lithium battery, Represents the deviation between differential data, Indicates the difference between The second-order difference data of lithium batteries; S3. The original voltage data is aligned to obtain a new sequence To perform denoising, use a Gaussian filter to perform one-dimensional linear smoothing on the raw voltage data of the lithium battery. The function expression is as follows: in, represents the voltage data after noise reduction at time n, represents the voltage data before noise reduction at time n, σ represents the standard deviation of the voltage data; S4. Use the noise-reduced data Perform nonlinear feature extraction, map the battery voltage data to phase space, and set the embedding dimension of the mapping operation. Time delay τ, find the vector in, Representation vector The dimension of , τ represents the time delay; S5. Create a two-layer neural network feature extraction model to extract the dynamic data of lithium batteries As input data, data As output data, the feature extraction model is fitted with training data. The loss function of the feature extraction model uses the root mean square error. When the error e is less than or equal to the set threshold ε, the model stops and the hidden layer parameters are extracted as the dynamic characteristics of the lithium battery. The error e is calculated as follows: Where u represents the number of output data, where u represents the number of output data, Indicates Vector data, Indicates Vector Calculated values; S6. Divide the lithium battery feature data set into a training set, a test set, and a validation set according to the distribution of discharge capacity; S7, taking the extracted dynamic characteristics of lithium batteries as input data and the lithium battery discharge capacity data as target data, according to the grouping of lithium batteries in high-dimensional space, a gradient boosting tree model is established, and the functional relationship between dynamic characteristics and discharge capacity is calculated to obtain the lithium battery capacity, wherein the gradient boosting tree is an additive model based on the boosting integration idea, and the forward distribution algorithm is used for greedy learning during training, and a CART tree is learned in each iteration to fit the previous The residual between the prediction result of the tree and the true value of the training sample, where Represents the gradient boosting tree The CART tree is a binary tree that generates two child nodes at each split, using a strategy that minimizes the sample variance of the two nodes.

2. The method for predicting lithium battery capacity based on chaos theory and nonlinear system according to claim 1, characterized in that: In step S4, the appropriate time delay τ is calculated using the mutual information method, and the process is as follows: To standardize: in, Indicates that lithium batteries Voltage sampling at all times, Indicates that lithium batteries Voltage sampling at all times, express The number of voltage sampling data, express The number of voltage sampling data, Express the probability function, and Respectively represent and The information entropy of express and The information entropy of the joint distribution, is a function of the delay time τ, and is calculated The τ corresponding to the first minimum value obtained is the optimal delay time τ.

3. The method for predicting lithium battery capacity based on chaos theory and nonlinear system according to claim 1, characterized in that: In step S4, the embedding dimension is calculated using the minimum pseudo neighbor method. The process is as follows: in, Indicated in Different from The other vectors of express The vector formed by dimensional collection, express The vector formed by dimensional collection, Indicated in Different from The other vectors of Indicates the difference between The denoised data of lithium batteries, r represents the threshold of the distance between dimensions, σ represents the standard deviation of the data, represents the Heaviside step function, when x≤0, When x>0, express and The gap between dimensions, e represents and The threshold of the gap between dimensions. If the gap is less than the threshold, it means that the current Dimension is the optimal dimension.

Images