Extraction Of Health Factors And Prediction Of SOH In Lithium Ion Batteries

Lithium ion battery has the characteristics of high energy density,high power density and long cycle life,which is one of the research focuses of energy storage equipment.In order to curb environmental pollution,land desertification,energy shortage and other problems,the state vigorously develops new energy,vigorously excavates the development of green energy,vigorously supports the production of energy vehicles,so as to achieve the goal of "carbon peak,carbon neutral" as soon as possible,promote economic transformation and promote sustainable social development.Due to the composition and structure of lithium-ion batteries,it is currently impossible to directly obtain the real-time SOH of batteries using sensing technology.Therefore,the development of models for estimating SOH and predicting RUL of batteries will help to promote the development of new energy industry.This paper first introduces the working principle of lithium ion battery,analyzes and defines the SOH,RUL and other indicators of lithium ion battery,and uses the instrument to measure the battery decay experimental data,combined with NASA data to analyze the data decline trend and experimental conditions.Secondly,the improved Douglas Peucker feature extraction algorithm is used to analyze the feature extraction of various feature projects in combination with Pearson correlation coefficient,and the prediction accuracy of each algorithm is compared to obtain the best feature extraction algorithm.Filter and Wrapper are used to realize the secondary extraction of features,eliminate redundant and irrelevant features,and avoid multicollinearity and dimensional disasters.Finally,the machine learning algorithm and optimization algorithm are used to establish the SOH and RUL prediction models of lithium ion battery,and the models are implemented on dataset A and dataset B of this paper respectively to verify the generalization ability of the models established.The prediction effect of the feature engineering and the prediction model proposed in this paper is summarized as follows:(1)In this paper,the Douglas Peucker algorithm is improved by thresholding.The improved Douglas Peucker algorithm can realize the fixed dimension feature extraction,and features with high Pearson correlation coefficient can be obtained in the feature extraction of battery data.The features extracted by the improved Douglas Peucker algorithm cannot determine the dimensions of feature extraction.Less dimensions will discard many feature information,and extracting more dimensions will cause feature confusion.The Min distance is combined with the improved Douglas Peucker algorithm to solve the problem of feature confusion.(2)The improved Douglas Peucker algorithm is used to establish feature engineering,which shows a strong generalization ability on data sets A and B.The secondary feature extraction is carried out by combining Filter and Wrapper methods,and the 9-dimensional optimal feature subset is obtained on data set A.On the GPR model,R2 of all batteries exceeds 0.99,showing a strong prediction effect.In terms of RUL prediction,the AE index of battery B0018 is 0,and the AE index of other batteries has errors,RUL prediction error shall be controlled within 10%.(3)Combined with Cat Boost model and LSTM model,the prediction effect of battery is better than GPR model.On the basis of improved Douglas Peucker algorithm feature extraction and various optimization algorithms,CGTSSA-LSTM model has the best prediction effect among the six prediction models established.In the SOH prediction aspect,the R2 of CGTSSA-LSTM model exceeds 0.999.In terms of RUL prediction,the AE index of CGTSSA-LSTM model is 0,and the RUL prediction error is 0.The R2 of the six prediction models on B0006,B0007 and B0018 batteries exceeded 0.99,and the established feature engineering could extract better health factors.In combination with dataset B,LSTM model and Cat Boost model,R2 of all batteries in dataset B exceeds 0.99,and the feature engineering established has strong generalization ability.To sum up,the feature engineering established by using the improved Douglas Peucker algorithm is more accurate than other feature extraction algorithms.The model is predicted on multiple batteries of two different materials.The experimental results show that LSTM model and Cat Boost model have strong generalization ability under the feature engineering established by using the improved Douglas Peucker algorithm.Figure [44] Table [11] Reference [85]...