Lithium battery SOE prediction method based on deep learning time sequence algorithm

Abstract

The invention belongs to the technical field of lithium battery residual energy estimation, and particularly relates to a lithium battery SOE prediction method based on a deep learning time sequence algorithm, and the method comprises the steps: obtaining a real-time data set of a target battery under a target working condition, carrying out the data analysis of the real-time data set, carrying out the feature extraction, and obtaining a plurality of independent observation variables, dividing the observation variables into a training set and a test set; constructing a Gaussian mixture model based on the observation variables in the training set to obtain a plurality of Gaussian components, and optimizing hyper-parameters of the Gaussian components in the Gaussian mixture model by adopting an expectation maximization algorithm; and importing the test set into the optimized Gaussian mixture model to obtain an SOE prediction result in the lithium battery. According to the prediction method, the SOE prediction of the lithium battery is realized by collecting part of time sequence data characteristics such as current, voltage and capacity in the real-time charging stage, the static and precision bottlenecks of a traditional prediction method are broken through, the SOE prediction analysis of the lithium battery can be dynamically performed with high precision, and the time and energy consumption of a traditional process are greatly reduced.

Description

Technical Field

[0001] This invention belongs to the field of lithium battery remaining energy estimation technology, specifically involving a lithium battery SOE prediction method based on deep learning time series algorithm. Background Technology

[0002] Lithium-ion batteries are increasingly widely used in electric vehicles due to their high energy density, low self-discharge, and long cycle life. Because of the complex operating conditions and customer requirements, electric vehicles place higher demands on the functionality of their battery management systems (BMS). Estimating the battery's state of energy (SOE) has gradually become an important function of BMS, especially for lithium-ion battery operating equipment and systems that require continuous operation. Traditionally, the state of charge (SOC) is considered an indicator of the battery's usable energy, but SOC actually represents the state of the battery's usable capacity, not the state of its usable energy. In practical applications, within the same SOC time interval, the discharge energy that the battery can output decreases as the SOC discharge time decreases; in actual battery energy storage systems, the discharge current often changes drastically, and changes in the load current cause the battery's remaining energy to deviate from its true value. Therefore, it is essential to study the estimation of the remaining energy of lithium-ion batteries using SOE instead of the traditional SOC parameter.

[0003] In recent years, there has been an increasing amount of research on residual energy estimation. The most direct algorithm is the power integration method, which integrates the product of voltage and current per unit time. However, due to the drift caused by current or voltage sensors, which leads to cumulative errors, the estimation error of the integration method increases significantly after long-term use. This not only consumes a lot of manpower and resources, but also has poor real-time applicability and low accuracy. Summary of the Invention

[0004] The purpose of this invention is to provide a lithium battery SOE prediction method based on deep learning time series algorithms to solve the problems mentioned in the background art.

[0005] The present invention achieves the above objectives through the following technical solutions:

[0006] A method for predicting SOE (Sequence Execution) of lithium batteries based on deep learning time-series algorithms, the method comprising:

[0007] S1. Obtain the real-time dataset of the target battery under the target operating conditions. After completing the data analysis of the real-time dataset, extract features to obtain multiple independent observation variables. Divide the observation variables into a training set and a test set.

[0008] S2. Construct a Gaussian mixture model based on the observed variables in the training set to obtain multiple Gaussian components. Use the expectation-maximization algorithm to optimize the hyperparameters of the Gaussian components in the Gaussian mixture model.

[0009] S3. Import the test set into the optimized Gaussian mixture model to obtain the SOE prediction results in lithium batteries.

[0010] As a further optimization of the present invention, step S1 includes:

[0011] S101. Collect real-time current, voltage, and capacity data during the charging and discharging process of lithium batteries to form a real-time dataset;

[0012] S102. Perform data analysis on the real-time dataset, and extract features to obtain multiple independent observation variables by combining the real-time change patterns of current and voltage during the constant current charging or discharging phase.

[0013] The observed variables must include at least: the maximum slope of the data curve, the area enclosed by the curve, the standard deviation, kurtosis, skewness, extreme values, and the mean;

[0014] S103. Divide the observed variables into training set and test set.

[0015] As a further optimization of the present invention, step S2 includes:

[0016] S201. Represent the observed variables in the training set as a Gaussian distribution, and approximate any continuous distribution using a finite number of single Gaussian density functions to obtain a Gaussian mixture model Deep-GMM, where the probability density function is:

[0017]

[0018] In the formula, α k μ is the weight of the k-th Gaussian component. k Let be the mean of the k-th Gaussian component; ∑ k Let be the covariance of the k-th Gaussian component; k represents the number of individual Gaussian components in the Deep-GMM.

[0019] S202. The EM algorithm, which consists of alternating E-step and M-step, is used for iterative calculation to optimize the weight α, mean μ, and variance Σ of each Gaussian group in Deep-GMM.

[0020] As a further optimization of the present invention, in step S3, the observed variables in the test set are used as the input x of Deep-GMM, and the expected output data y is represented as SOE. Then the Gaussian mixture joint density is:

[0021]

[0022]

[0023] The mean μ of the conditional distribution j(x) Sum of variances ∑ j(x) They are respectively:

[0024] μ j(x) =μ jy +Σ jyx Σ jxx -1 (x-μ jx )

[0025] Σ j(x) =Σ jyy -Σ jyx Σ jxx -1 Σ jxy

[0026] The regression expression is:

[0027]

[0028]

[0029] As a further optimization of the present invention, step S3 also includes model evaluation of the prediction results, including: selecting the mean absolute percentage error (MAPE) and mean squared error (MSE) as evaluation indicators, based on the following formula:

[0030]

[0031]

[0032] In the formula, Represents the predicted value, y i This represents the observed value, and n is the sample size.

[0033] The beneficial effects of this invention are as follows:

[0034] (1) This prediction method achieves SOE prediction of lithium battery by collecting time-series data features such as current, voltage and capacity during the real-time charging stage. It breaks through the static and accuracy bottleneck of traditional prediction methods and can dynamically and accurately predict and analyze SOE of lithium battery, greatly reducing the time and energy consumption of traditional processes.

[0035] (2) In the lithium battery SOE prediction technology based on deep learning time series algorithm proposed in this invention, the model estimation error can reach within 2%, which is far lower than the industry standard of 5%, and realizes accurate real-time estimation of SOE of lithium-ion battery under dynamic current; moreover, SOE has a significant advantage over SOC in reflecting the energy behavior of lithium-ion battery under dynamic load current. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating the execution of the prediction method in this invention;

[0037] Figure 2 This is a structural block diagram of the prediction method in this invention;

[0038] Figure 3 This is a diagram of the Deep-GMM model architecture proposed in this invention. Detailed Implementation

[0039] The present application will now be described in further detail with reference to the accompanying drawings. It should be noted that the following specific embodiments are only used to further illustrate the present application and should not be construed as limiting the scope of protection of the present application. Those skilled in the art can make some non-essential improvements and adjustments to the present application based on the above application content.

[0040] Example 1

[0041] like Figure 1-3 As shown in the figure, this embodiment proposes a lithium battery SOE prediction method based on deep learning time series algorithm, which includes the following steps:

[0042] S1. Obtain the real-time dataset of the target battery under the target operating conditions. After completing the data analysis of the real-time dataset, extract features to obtain multiple independent observation variables. Divide the observation variables into training set and test set.

[0043] S2. Construct a Gaussian mixture model based on the observed variables in the training set to obtain multiple Gaussian components. Use the expectation-maximization algorithm to optimize the hyperparameters of the Gaussian components in the Gaussian mixture model.

[0044] S3. Import the test set into the optimized Gaussian mixture model to obtain the SOE prediction results in lithium batteries.

[0045] As a further preferred option, step S1 includes:

[0046] S101. Collect real-time current, voltage, and capacity data during the charging and discharging process of lithium batteries to form a real-time dataset;

[0047] S102. Perform data analysis on the real-time dataset, and extract features to obtain multiple independent observation variables by combining the real-time change patterns of current and voltage during the constant current charging or discharging phase.

[0048] The observed variables must include at least: the maximum slope of the data curve, the area enclosed by the curve, the standard deviation, kurtosis, skewness, extreme values, and the mean;

[0049] S103. Divide the observed variables into training set and test set.

[0050] Understandably, the feature construction involved in step S1 includes: collecting real-time data such as current, voltage, and capacity during the charging and discharging process of the battery cell for data analysis to obtain time-series data features; based on the real-time change law of charging current and voltage, determining the inflection point by judging the current value decrease by ΔI during constant current charging or the voltage value decrease by ΔV during constant voltage charging; and calculating relevant features such as the maximum slope of the real-time data curves for current, voltage, and capacity, the area enclosed by the curves, the standard deviation, kurtosis, skewness, extreme values, and mean values. The calculation formula is as follows:

[0051] Formula for curve slope:

[0052] Formula for the area under a curve:

[0053] Standard deviation formula:

[0054] Kurtosis calculation formula:

[0055] Skewness calculation formula:

[0056] Where μ is the mean and σ is the variance.

[0057] As a further preferred option, step S2 includes:

[0058] S201. Represent the observed variables in the training set as a Gaussian distribution, and approximate any continuous distribution using a finite number of single Gaussian density functions to obtain the Gaussian Mixture Model Deep-GMM, where the probability density function is:

[0059]

[0060] In the formula, α k μ is the weight of the k-th Gaussian component. k Let be the mean of the k-th Gaussian component; ∑ k Let be the covariance of the k-th Gaussian component; k represents the number of individual Gaussian components in the Deep-GMM.

[0061] The hyperparameters of the Deep-GMM model for the k Gaussian components are:

[0062] Q={{α1,μ1,Σ1},{α2,μ2,Σ2},…,{α n ,μ n ,Σ n}}.

[0063] S202. The EM algorithm, which consists of alternating E-step and M-step, is used for iterative calculation to optimize the weight α, mean μ, and variance Σ of each Gaussian group in Deep-GMM.

[0064] Understandably, the Expectation-Maximization (EM) algorithm is the most widely used method for solving for optimal hyperparameters in Gaussian mixture models. The EM algorithm consists of alternating E-step and M-step operations, and through iterative calculations, it ultimately obtains the hyperparameters of the Deep-GMM model.

[0065] E-step:

[0066]

[0067] M-step:

[0068]

[0069]

[0070]

[0071] In this embodiment, based on the Deep-GMM model established above, the EM algorithm is used to solve the hyperparameters of Gaussian mixtures to obtain the weight α, mean μ, and variance Σ of each Gaussian mixture. Gaussian mixture regression is performed based on the linear combination of Gaussian conditions and Gaussian distributions. The lithium battery data consists of two parts: input data and output data. The input training set data x includes a high-dimensional array after feature engineering of dimensions such as current A and voltage V. The expected output data y is...

[0072] As a further preferred embodiment, in step S3, the observed variables in the test set are used as the input x of the Deep-GMM, and the expected output data y is represented as SOE. Then the Gaussian mixture joint density is:

[0073]

[0074]

[0075] The mean μ of the conditional distribution j(x) Sum of variances ∑ j(x) They are respectively:

[0076] μ j(x) =μ jy +Σ jyx Σ jxx -1 (x-μ jx )

[0077] Σ j(x) =Σ jyy -Σ jyx Σ jxx -1 Σ jxy

[0078] The regression expression is:

[0079]

[0080]

[0081] As a further preferred embodiment, step S3 also includes model evaluation of the prediction results, including: selecting the mean absolute percentage error (MAPE) and mean squared error (MSE) as evaluation indicators, based on the following formula:

[0082]

[0083]

[0084] In the formula, Represents the predicted value, y i This represents the observed value, and n is the sample size.

[0085] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A lithium battery SOE prediction method based on deep learning time series algorithm, characterized in that, The methods include: S1. Obtain the real-time dataset of the target battery under the target operating conditions. After completing the data analysis of the real-time dataset, extract features to obtain multiple independent observation variables. Divide the observation variables into a training set and a test set. S2. Construct a Gaussian mixture model based on the observed variables in the training set to obtain multiple Gaussian components. Use the expectation-maximization algorithm to optimize the hyperparameters of the Gaussian components in the Gaussian mixture model. S3. Import the test set into the optimized Gaussian mixture model to obtain the SOE prediction results in lithium batteries.

2. The lithium battery SOE prediction method based on deep learning time series algorithm according to claim 1, characterized in that: Step S1 includes: S101. Collect real-time current, voltage, and capacity data during the charging and discharging process of lithium batteries to form a real-time dataset; S102. Perform data analysis on the real-time dataset, and extract features to obtain multiple independent observation variables by combining the real-time change patterns of current and voltage during the constant current charging or discharging phase. The observed variables must include at least: the maximum slope of the data curve, the area enclosed by the curve, the standard deviation, kurtosis, skewness, extreme values, and the mean; S103. Divide the observed variables into training set and test set.

3. The lithium battery SOE prediction method based on deep learning time series algorithm according to claim 2, characterized in that: Step S2 includes: S201. Represent the observed variables in the training set as a Gaussian distribution, and approximate any continuous distribution using a finite number of single Gaussian density functions to obtain a Gaussian mixture model, Deep-GMM, where the probability density function is: In the formula, α k μ is the weight of the k-th Gaussian component. k Let be the mean of the k-th Gaussian component; ∑ k Let be the covariance of the k-th Gaussian component; k represents the number of individual Gaussian components in the Deep-GMM. S202. The EM algorithm, which consists of alternating E-step and M-step, is used for iterative calculation to optimize the weight α, mean μ, and variance Σ of each Gaussian group in Deep-GMM.

4. The lithium battery SOE prediction method based on deep learning time series algorithm according to claim 3, characterized in that: In step S3, the observed variables in the test set are used as input x of the Deep-GMM, and the expected output data y is represented by SOE. Then the Gaussian mixture joint density is: The mean μ of the conditional distribution j(x) Sum of variances ∑ j(x) They are respectively: m j(x) =μ jy +S jyx S jxx -1 (x-μ jx ) S j(x) =S jyy -S jyx S jxx -1 S jxy The regression expression is:

5. The lithium battery SOE prediction method based on deep learning time series algorithm according to claim 1, characterized in that: Step S3 also includes model evaluation of the prediction results, including: selecting the Mean Absolute Percentage Error (MAPE) and Mean Squared Error (MSE) as evaluation indicators, based on the following formula: In the formula, y i Represents the predicted value, y i This represents the observed value, and n is the sample size.

Images