Estimating State Of Health Of Li-ion Batteries Based On QGA And GRNN

As the safety of lithium-ion batteries has attracted more and more attention,the estimation of State of Health(SOH)as an important indicator of safety has become a research hotspot.However,SOH is difficult to measure directly,it is mainly estimated by directly measurable parameters.Due to the complex chemical reaction of the battery,there is a highly complex nonlinear relationship between the SOH and the measurement parameters.Currently,there is no estimation method that satisfies the practical application,and there are problems of unstable algorithm and low estimation accuracy.The existing methods are mainly based on laboratory measurement data,because the laboratory environment is relatively stable and the data changes relatively smoothly.However,the actual operating data is affected by factors such as driving behavior,road,temperature,humidity and communication lines,so there are a lot of abnormalities.Therefore,it is necessary to study the anomaly detection algorithm based on the characteristics of actual driving data and the SOH estimation algorithm with high stability and high accuracy.Aiming at the time sequence characteristics of battery data,this thesis proposes an anomaly detection algorithm based on Gated Recurrent Unit(GRU)and Support Vector Domain Description(SVDD)to identify anomalies.This thesis also proposes an SOH estimation algorithm based on Optimized General Regression Neural Network(OGRNN)to improve the stability and accuracy of estimation.The main research work of the thesis is as follows:(1)An anomaly detection algorithm based on GRU-SVDD is proposed to extract parameters with strong correlation with capacity,which lays the foundation for accurate SOH estimation.GRU is used to predict the change trend of voltage,the difference between voltage measurement value and predicted value is taken as the dataset,and then the abnormal is identified by SVDD.Aiming at the problem that SVDD mistakenly judges the point with the minimum difference value as the abnormal point,the judgment condition is modified to correctly identify the abnormal point.According to Pearson correlation coefficient,the parameters with strong correlation with capacity are extracted from the parameters after exception processing to obtain stable estimated capacity.Experiments based on operating dataset show that the proposed anomaly detection algorithm can effectively identify the anomalies,and the correlation between parameters and capacity after exception processing is improved to more than 0.9,with an average increase of 0.37,which meets the needs of the model for strong correlation of input parameters.(2)In order to improve the stability and accuracy of the SOH estimation of lithiumion batteries,this thesis proposes a novel SOH estimation algorithm based on particle filter(PF),quantum genetic algorithm(QGA)and generalized regression neural network(GRNN).A denoising method based on grouping and improved PF is proposed to make the network input parameters more stable.In order to improve the estimation stability,a new method is proposed to optimize the GRNN smoothing factor by using QGA to eliminate the subjectivity of setting the smoothing factor manually.The correlation coefficients between parameters and capacity are used to optimize the transfer function of the pattern layer to improve the estimation accuracy.Comparative experimental results show that the minimum error of the relevant SOH estimation algorithms on the NASA dataset is 0.4%,and the proposed algorithm is 0.26%.On the operating dataset,the error of the proposed algorithm is 0.64%,and its smoothing factor is automatically estimated by the algorithm compared with the traditional GRNN,thus reducing the human dependence.Therefore,the proposed algorithm is more accurate and stable,and realizes the automatic optimization of smoothing factor.(3)This thesis designs and implements a prototype system for SOH estimation of lithium-ion batteries.The system has the functions of data import,parameter preprocessing,state of health estimation,data query and visualization.