Lithium battery life prediction method based on unmanned equipment side GRU + BLR algorithm

Abstract

The invention discloses a lithium battery life prediction method based on an unmanned equipment side GRU + BLR algorithm, and the method comprises the steps: collecting the working data of a target lithium battery, and carrying out the preprocessing of the working data; performing multi-modal feature extraction on the working data to obtain feature data, performing classification fusion on the feature data to obtain fusion data, and capturing a nonlinear trend of lithium battery degradation according to the fusion data by adopting a gating cycle unit to obtain a first lithium battery life; and optimizing the first lithium battery life by adopting a Bayesian linear regression model to obtain a second lithium battery life, performing uncertainty estimation on the second lithium battery life, and outputting the predicted lithium battery life. The method not only can improve the precision of lithium battery life prediction, but also has good interpretability, and can be directly applied to a lithium battery life prediction system.

Description

Technical Field

[0001] The present invention relates to the field of life prediction, and in particular to a method for predicting the life of a lithium battery based on a GRU+BLR algorithm on an unmanned equipment side. Background Art

[0002] With the widespread application of unmanned equipment in military, industrial and civilian fields, the requirements for the reliability and endurance of its energy system are becoming higher and higher. As the main power source of unmanned equipment, the prediction and management of the performance and life of lithium batteries are of great significance to the service life, operation safety and maintenance plan of the equipment. Accurately predicting the remaining service life of lithium batteries can help extend the operation time of equipment, optimize maintenance cycles, and reduce the risk of unexpected downtime.

[0003] At present, traditional statistical life prediction methods based on linear regression and ARIMA models are difficult to capture the complex nonlinear changes and random fluctuations in the battery degradation process; methods that simulate the chemical processes inside the battery for life prediction require complex parameter calibration and high computing resources in practical applications. Data-driven machine learning methods such as support vector machines, random forests, and deep neural networks are widely used for battery life prediction, but they are insufficient in providing uncertainty estimates, especially when dealing with computing resource constraints on edge devices.

[0004] In view of the shortcomings of existing technologies in lithium battery life prediction, a hybrid algorithm combining GRU and BLR is proposed for lithium battery life prediction on the unmanned equipment side. It effectively captures the nonlinearity and long-term trend of battery degradation; adds uncertainty estimation to the prediction results, improves the reliability of battery life prediction and the lack of risk assessment; reduces computational complexity and is easy to apply to resource-limited edge devices. Summary of the invention

[0005] The purpose of the present invention is to provide a lithium battery life prediction method based on the GRU+BLR algorithm on the unmanned equipment side.

[0006] To achieve the above object, the present invention is implemented according to the following technical solutions: The present invention comprises the following steps: S1 collects the working data of the target lithium battery and pre-processes the working data; S2 performs multimodal feature extraction on the working data to obtain feature data, classifies and fuses the feature data to obtain fused data, and uses a gated cyclic unit to capture the nonlinear trend of lithium battery degradation according to the fused data to obtain the life of the first lithium battery; including: Input the feature data into the feature fusion model and calculate the total mean square error of the feature data: ; The total mean square error is , the weighting factor of the kth feature data is , the kth feature data is , the number of feature data is , the expected function is ; Construct an auxiliary function, the expression is: ; The Lagrange multiplier is , the variance of the kth data is , the auxiliary function is K; The weighted coefficient calculation formula is obtained according to the total mean square error, extreme value and Lagrange multiplier, and the expression is: ; The weight coefficient of the kth feature data is , the average value of the characteristic data is , perform mean standardization on feature data; Extract the feature vector of feature data, the expression is: ; The kth feature data at the i-th attribute is , feature data The eigenvector of , the average value of the feature data at the i-th attribute is , feature data The state at time s is , the upper limit of the moment is M; Constructing a feature vector set according to the feature vector, and classifying the feature vector using a support vector machine according to the weighted coefficient feature vector set to obtain classification features; The classification features are fused and the expression is: ; The fused data is , the generalization parameter is , the optimal solution for the kth classification feature is , the optimal solution for the jth classification feature is , the optimal solution mean of the a-th classification feature is , the optimal solution mean of the jth feature data is , the fusion data type is , the optimal solution mean of the j-th classification feature is , the kernel function is , the ath feature vector set is , the eigenvector set is Z; S3 optimizes the life of the first lithium battery using a Bayesian linear regression model to obtain the life of the second lithium battery, estimates the uncertainty of the life of the second lithium battery, and outputs the predicted life of the lithium battery; including: Input the second lithium battery life into the uncertainty estimation model and select the parameter domain as , taking the uniform distribution of the parameter domain as the prior distribution, calculate the prior probability density distribution function: ; The prior probability density distribution function of parameter u is , the number of a is M, and the ath weight coefficient is , the ath parameter ; Calculate the optimized parameter posterior probability density function: ; The total error caused by the ath experiment and modeling is , the total error caused by experiment and modeling is , the predicted life of lithium battery with parameter u is , the predicted life of the lithium battery is y, and the posterior probability density function of the parameter u when the predicted life of the lithium battery is y is ; When the confidence interval is 96%, given the likelihood function, the expression is: ; The predicted life of lithium batteries The likelihood function of the parameter u is , the ath standard deviation is , the predicted life of the a-th lithium battery with parameter u is , the predicted life of the a-th lithium battery with parameter u squared is ; Calculate the adjusted predicted life of lithium batteries: ; The adjusted predicted life of lithium battery is The adjusted standard deviation is , the kernel function of parameter u is , transposed to T, the variance of the posterior distribution is E, and the ath parameter is , the predicted battery life is y, and the kernel function is H.

[0007] Furthermore, the method of performing multimodal feature extraction on the working data to obtain feature data includes: The image features of the working image data are extracted by using a convolutional neural network, and the sound features of the working sound data are extracted by using Fourier transform; The original signal for extracting working data is specifically: Determine all extreme points of the target battery capacity according to the original signal of the working data, the extreme points including the maximum value and the minimum value; The upper envelope is generated by connecting all the maximum extreme value points, and the lower envelope is generated by connecting all the minimum extreme value points; Calculate the average value of the upper envelope and the lower envelope to obtain the mean envelope; The mean envelope is subtracted from the original signal in the working data to obtain an intermediate signal; Two conditions are given for the IMF condition: 1. At any point, the number of local maxima and local minima must be the same; 2. At any point, the average of the local maxima and local minima must be zero; If the intermediate signal satisfies both conditions of the IMF condition, the first IMF of the intermediate signal is extracted to obtain the first feature; Eliminate the original signal of the first feature and update the original signal, and calculate the complexity of the updated original signal: ; The updated signal is w, and the complexity of the updated original signal is , the updated signal power is , the updated signal-to-noise power is ; When the signal complexity is greater than 0.2176, the updated original signal is subjected to empirical mode decomposition to obtain the second characteristic data; otherwise, the decomposition is stopped; When the updated original signal cannot be further decomposed, the updated original signal is used as a residual term; The image feature, the sound feature, the first feature, the second feature data and the residual item are output as feature data.

[0008] Furthermore, a method for capturing the nonlinear trend of lithium battery degradation according to the fused data using a gated recurrent unit includes: The gated recurrent unit includes an input layer, a gating mechanism for a hidden layer, and an output layer; The gated recurrent unit contains two gate structures, namely the update gate and the reset gate; The input gate and forget gate are combined into one as the update gate. The update gate determines how much historical information needs to be retained to the current state. The update gate controls the amount of memory information from the previous moment that can continue to be retained to the current moment. The reset gate determines how much past information needs to be forgotten and controls the degree of influence of historical information on the current state; The linear self-update memory unit of the gated recurrent unit is cancelled, and the gating is used to directly perform linear self-update in the hidden unit.

[0009] Furthermore, the method for optimizing the life of the first lithium battery using a Bayesian linear regression model includes: The Bayesian linear regression model updates the posterior distribution of model parameters by combining prior knowledge and the likelihood of the first lithium battery life, thereby providing predicted values ​​and their confidence intervals, optimizing parameter estimation and uncertainty quantification, and obtaining the confidence intervals of the parameters and the second lithium battery life.

[0010] The beneficial effects of the present invention are: The present invention is a lithium battery life prediction method based on the GRU+BLR algorithm on the unmanned equipment side. Compared with the prior art, the present invention has the following technical effects: The present invention can improve the accuracy of lithium battery life prediction through preprocessing, multimodal feature extraction, obtaining the first lithium battery life, obtaining the second lithium battery life, uncertainty estimation and outputting the predicted lithium battery life steps, thereby improving the accuracy of lithium battery life prediction, optimizing lithium battery life prediction, greatly saving resources, and improving work efficiency. It can realize intelligent prediction of lithium battery life and optimize uncertainty estimation of lithium battery life prediction in real time. Intelligent prediction of lithium battery life is of great significance, and can adapt to the needs of intelligent prediction of lithium battery life of different standards and intelligent prediction of different multi-lithium battery life, and has a certain universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is a flowchart of the steps of the lithium battery life prediction method based on the GRU+BLR algorithm on the unmanned equipment side of the present invention; Figure 2 This is a block diagram of the lithium battery life prediction system based on the GRU+BLR algorithm on the unmanned equipment side. DETAILED DESCRIPTION

[0012] The present invention is further described below by means of specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0013] The present invention provides a method for predicting the life of a lithium battery based on the GRU+BLR algorithm on the unmanned equipment side, comprising the following steps: like Figure 1 As shown, in this embodiment, the following steps are included: S1 collects the working data of the target lithium battery and pre-processes the working data; In the actual evaluation, the working data include voltage, temperature, current, lithium battery status data, battery capacity, energy density, charge and discharge rate and voltage, ambient temperature, discharge current, discharge depth, average charging current and actual health status, nominal voltage, specific energy, specific power, charging efficiency, battery health status, ambient temperature, discharge depth, working image data, working sound data; The lithium battery of model XXX is used as the research object to study the life prediction of lithium batteries on the unmanned equipment side; S2 performs multimodal feature extraction on the working data to obtain feature data, classifies and fuses the feature data to obtain fused data, and uses a gated cyclic unit to capture the nonlinear trend of lithium battery degradation according to the fused data to obtain the life of the first lithium battery; including: Input the feature data into the feature fusion model and calculate the total mean square error of the feature data: ; The total mean square error is , the weighting factor of the kth feature data is , the kth feature data is , the number of feature data is , the expected function is ; Construct an auxiliary function, the expression is: ; The Lagrange multiplier is , the variance of the kth data is , the auxiliary function is K; The weighted coefficient calculation formula is obtained according to the total mean square error, extreme value and Lagrange multiplier, and the expression is: ; The weight coefficient of the kth feature data is , the average value of the characteristic data is , perform mean standardization on feature data; Extract the feature vector of feature data, the expression is: ; The kth feature data at the i-th attribute is , feature data The eigenvector of , the average value of the feature data at the i-th attribute is , feature data The state at time s is , the upper limit of the moment is M; In the actual evaluation, the voltage fluctuation range, current peak-to-valley ratio, temperature average value, internal resistance stability index, and capacity retention rate; the life of the first lithium battery is 5 years and 6 months; Constructing a feature vector set according to the feature vector, and classifying the feature vector using a support vector machine according to the weighted coefficient feature vector set to obtain classification features; The classification features are fused and the expression is: ; The fused data is , the generalization parameter is , the optimal solution for the kth classification feature is , the optimal solution for the jth classification feature is , the optimal solution mean of the a-th classification feature is , the optimal solution mean of the jth feature data is , the fusion data type is , the optimal solution mean of the j-th classification feature is , the kernel function is , the ath feature vector set is , the eigenvector set is Z; S3 optimizes the life of the first lithium battery using a Bayesian linear regression model to obtain the life of the second lithium battery, estimates the uncertainty of the life of the second lithium battery, and outputs the predicted life of the lithium battery; including: Input the second lithium battery life into the uncertainty estimation model and select the parameter domain as , taking the uniform distribution of the parameter domain as the prior distribution, calculate the prior probability density distribution function: ; The prior probability density distribution function of parameter u is , the number of a is M, and the ath weight coefficient is , the ath parameter ; Calculate the optimized parameter posterior probability density function: ; The total error caused by the ath experiment and modeling is , the total error caused by experiment and modeling is , the predicted life of lithium battery with parameter u is , the predicted life of the lithium battery is y, and the posterior probability density function of the parameter u when the predicted life of the lithium battery is y is ; When the confidence interval is 96%, given the likelihood function, the expression is: ; The predicted life of lithium battery The likelihood function of the parameter u is , the ath standard deviation is , the predicted life of the a-th lithium battery with parameter u is , the predicted life of the a-th lithium battery with parameter u squared is ; Calculate the adjusted predicted life of lithium batteries: ; The adjusted predicted life of lithium battery is The adjusted standard deviation is , the kernel function of parameter u is , transposed to T, the variance of the posterior distribution is E, and the ath parameter is , the predicted battery life is y, and the kernel function is H; In the actual evaluation, the life of the second lithium battery is 5 years and 5 months, and the predicted life of the lithium battery is 5 years, 4 months and 25 days; The EMD module is the empirical mode decomposition; the GRU module is the gated recurrent unit; the BLR model is the Bayesian linear regression model; GRU module: includes update gate and reset gate. The update gate determines how much past information to retain and how much new information to incorporate, while the reset gate is used to control the degree to which the previous state is ignored. GRU has fewer parameters than LSTM. Due to its simpler structure, fewer parameters need to be trained. GRU is unique in information propagation and update mechanism. The update gate can directly merge and update historical information with current input, allowing information to be transmitted more smoothly and effectively in the sequence. GRU, with its unique gating mechanism, effectively alleviates the common gradient vanishing and gradient exploding problems in recurrent neural networks.

[0014] In this embodiment, the method of performing multimodal feature extraction on the working data to obtain feature data includes: The image features of the working image data are extracted by using a convolutional neural network, and the sound features of the working sound data are extracted by using Fourier transform; The original signal for extracting working data is specifically: Determine all extreme points of the target battery capacity according to the original signal of the working data, the extreme points including the maximum value and the minimum value; The upper envelope is generated by connecting all the maximum extreme value points, and the lower envelope is generated by connecting all the minimum extreme value points; Calculate the average value of the upper envelope and the lower envelope to obtain the mean envelope; The mean envelope is subtracted from the original signal in the working data to obtain an intermediate signal; Two conditions are given for the IMF condition: 1. At any point, the number of local maxima and local minima must be the same; 2. At any point, the average of the local maxima and local minima must be zero; If the intermediate signal satisfies both conditions of the IMF condition, the first IMF of the intermediate signal is extracted to obtain the first feature; Eliminate the original signal of the first feature and update the original signal, and calculate the complexity of the updated original signal:

[0015] The updated signal is w, and the complexity of the updated original signal is , the updated signal power is , the updated signal-to-noise power is ; When the signal complexity is greater than 0.2176, the updated original signal is subjected to empirical mode decomposition to obtain the second characteristic data; otherwise, the decomposition is stopped; When the updated original signal cannot be further decomposed, the updated original signal is used as a residual term; The image feature, the sound feature, the first feature, the second feature data and the residual item are output as feature data.

[0016] In this embodiment, the method of using a gated recurrent unit to capture the nonlinear trend of lithium battery degradation according to the fused data includes: The gated recurrent unit includes an input layer, a gating mechanism for a hidden layer, and an output layer; The gated recurrent unit contains two gate structures, namely the update gate and the reset gate; The input gate and forget gate are combined into one as the update gate. The update gate determines how much historical information needs to be retained to the current state. The update gate controls the amount of memory information from the previous moment that can continue to be retained to the current moment. The reset gate determines how much past information needs to be forgotten and controls the degree of influence of historical information on the current state; The memory unit with linear self-update of gated recurrent unit is cancelled, and linear self-update is directly performed in hidden unit by using gating; The first lithium battery life is obtained based on the long-term dependence of the nonlinear trend.

[0017] In this embodiment, the method for optimizing the life of the first lithium battery using a Bayesian linear regression model includes: The Bayesian linear regression model updates the posterior distribution of model parameters by combining prior knowledge and the likelihood of the first lithium battery life, thereby providing predicted values ​​and their confidence intervals, optimizing parameter estimation and uncertainty quantification, and obtaining the confidence intervals of the parameters and the second lithium battery life.

[0018] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A lithium battery life prediction method based on the GRU+BLR algorithm on the unmanned equipment side is characterized by: The following steps are involved: S1 collects the working data of the target lithium battery and pre-processes the working data; S2 performs multimodal feature extraction on the working data to obtain feature data, classifies and fuses the feature data to obtain fused data, and uses a gated cyclic unit to capture the nonlinear trend of lithium battery degradation according to the fused data to obtain the life of the first lithium battery; including: Input the feature data into the feature fusion model and calculate the total mean square error of the feature data: ； The total mean square error is , the weighting factor of the kth feature data is , the kth feature data is , the number of feature data is , the expected function is ; Construct an auxiliary function, the expression is: ； The Lagrange multiplier is , the variance of the kth data is , the auxiliary function is K; The weighted coefficient calculation formula is obtained according to the total mean square error, extreme value and Lagrange multiplier, and the expression is: ； The weight coefficient of the kth feature data is , the average value of the characteristic data is , perform mean standardization on feature data; Extract the feature vector of feature data, the expression is: ； The kth feature data at the i-th attribute is , feature data The eigenvector of , the average value of the feature data at the i-th attribute is , feature data The state at time s is , the upper limit of the moment is M; Constructing a feature vector set according to the feature vector, and classifying the feature vector using a support vector machine according to the weighted coefficient feature vector set to obtain classification features; The classification features are fused and the expression is: ； The fused data is , the generalization parameter is , the optimal solution for the kth classification feature is , the optimal solution for the jth classification feature is , the optimal solution mean of the a-th classification feature is , the optimal solution mean of the jth feature data is , the fusion data type is , the optimal solution mean of the j-th classification feature is , the kernel function is , the ath feature vector set is , the eigenvector set is Z; S3 optimizes the life of the first lithium battery using a Bayesian linear regression model to obtain the life of the second lithium battery, estimates the uncertainty of the life of the second lithium battery, and outputs the predicted life of the lithium battery; including: Input the second lithium battery life into the uncertainty estimation model and select the parameter domain as , taking the uniform distribution of the parameter domain as the prior distribution, calculate the prior probability density distribution function: ； The prior probability density distribution function of parameter u is , the number of a is M, and the ath weight coefficient is , the ath parameter ; Calculate the optimized parameter posterior probability density function: ； The total error caused by the ath experiment and modeling is , the total error caused by experiment and modeling is , the predicted life of lithium battery with parameter u is , the predicted life of the lithium battery is y, and the posterior probability density function of parameter u when the predicted life of the lithium battery is y is ; When the confidence interval is 96%, given the likelihood function, the expression is: ； The predicted life of lithium batteries The likelihood function of the parameter u is , the ath standard deviation is , the predicted life of the a-th lithium battery with parameter u is , the predicted life of the a-th lithium battery with parameter u squared is ; Calculate the adjusted predicted life of lithium batteries: ； The adjusted predicted life of lithium battery is The adjusted standard deviation is , the kernel function of parameter u is , transposed to T, the variance of the posterior distribution is E, and the ath parameter is , the predicted battery life is y, and the kernel function is H.

2. According to claim 1, the method for predicting the life of a lithium battery based on the GRU+BLR algorithm on the unmanned equipment side is characterized in that: The method for extracting multimodal features from the working data to obtain feature data comprises: The image features of the working image data are extracted by using a convolutional neural network, and the sound features of the working sound data are extracted by using Fourier transform; The original signal for extracting working data is specifically: Determine all extreme points of the target battery capacity according to the original signal of the working data, the extreme points including the maximum value and the minimum value; The upper envelope is generated by connecting all the maximum extreme value points, and the lower envelope is generated by connecting all the minimum extreme value points; Calculate the average value of the upper envelope and the lower envelope to obtain the mean envelope; The mean envelope is subtracted from the original signal in the working data to obtain an intermediate signal; Two conditions are given for the IMF condition:

1. At any point, the number of local maxima and local minima must be the same; 2. At any point, the average of the local maxima and local minima must be zero; If the intermediate signal satisfies both conditions of the IMF condition, the first IMF of the intermediate signal is extracted to obtain the first feature; Eliminate the original signal of the first feature and update the original signal, and calculate the complexity of the updated original signal: ； The updated signal is w, and the complexity of the updated original signal is , the updated signal power is , the updated signal-to-noise power is ; When the signal complexity is greater than 0.2176, the updated original signal is subjected to empirical mode decomposition to obtain the second characteristic data; otherwise, the decomposition is stopped; When the updated original signal cannot be further decomposed, the updated original signal is used as a residual term; The image feature, the sound feature, the first feature, the second feature data and the residual item are output as feature data.

3. According to claim 1, the method for predicting the life of a lithium battery based on the GRU+BLR algorithm on the unmanned equipment side is characterized in that: The method of using a gated recurrent unit to capture the nonlinear trend of lithium battery degradation according to the fused data includes: The gated recurrent unit includes an input layer, a gating mechanism for a hidden layer, and an output layer; The gated recurrent unit contains two gate structures, namely the update gate and the reset gate; The input gate and forget gate are combined into one as the update gate. The update gate determines how much historical information needs to be retained to the current state. The update gate controls the amount of memory information from the previous moment that can continue to be retained to the current moment. The reset gate determines how much past information needs to be forgotten and controls the degree of influence of historical information on the current state; The linear self-update memory unit of the gated recurrent unit is cancelled, and the gating is used to directly perform linear self-update in the hidden unit.

4. According to claim 1, the method for predicting the life of a lithium battery based on the GRU+BLR algorithm on the unmanned equipment side is characterized in that: The method for optimizing the life of the first lithium battery using a Bayesian linear regression model comprises: The Bayesian linear regression model updates the posterior distribution of model parameters by combining prior knowledge and the likelihood of the first lithium battery life, thereby providing predicted values ​​and their confidence intervals, optimizing parameter estimation and uncertainty quantification, and obtaining the confidence intervals of the parameters and the second lithium battery life.

Images