Study On The Remaining Useful Life Of Lithium Battery With Multi-indicator Fusion Under Charging Curve

Lithium-ion battery has been widely recognized because of its high working voltage,small volume,low weight,high energy density,no memory effect,low cost,little pollution,low self-discharge rate and long lifespan.At present,it has been used in communication equipment,laptop,camera and other portable products.The prediction of remaining useful life can provide users with effective information of battery replacement to avoid safety accidents.In addition,the remaining useful life prediction of the lithium battery also has certain reference value for online battery management,used car evaluation and battery ladder utilization.Therefore,it is particularly important to evaluate the remaining useful life of batteries in the actual use.The traditional research methods usually carry out the remaining useful life prediction based on battery capacity.How to fuse multiple indicators to estimate the state-of-health and remaining useful life of the battery is a very important research direction.The main work of this thesis is as follows:1.In this thesis,the variation rule of the battery data under the state of charge is firstly analyzed.Through the analysis of the incremental capacity curves,current curves,voltage curves and temperature curves,four indicators are extracted as the health indicators.Pearson correlation analysis is used to analyze the correlations between the extracted indicators and battery capacity.In addition,the grey correlation analysis method and entropy weight method are combined to fuse the four health indicators.It is found that the correlation between the characteristics of the battery in the first cycle and the characteristics of the battery in the remaining cycles can be effectively used to estimate the state-of-health of the battery.2.In this thesis,the accuracy of state-of-health estimated by a single indicator is compared with the accuracy of state-of-health estimated by multiple indicators.The comparison shows that the latter is higher.And the method is used to estimate the state-of-health of the four batteries published by NASA,which verified the effectiveness of the proposed method.3.In this thesis,the algorithm of predicting the remaining useful life is studied.The particle filter algorithm and gradient descent algorithm are combined to predict the remaining useful life of battery.The sigmoid function is used to improve the loss function in the gradient descent algorithm.According to this improvement,the motion direction of particle is no longer random.On the contrary,both the local decay characteristics of the battery and the overall decay characteristics of the battery are considered in the process of updating the particle.After the improvement,the tracking ability of the algorithm and the prediction accuracy of the remaining battery life are improved.