| Lithium-ion batteries are widely used in household appliances,smart phones,energy storage systems,electric vehicles and other fields because of their high energy density,high output voltage,low self-discharge rate,low voltage drop and easy management.Accurate modeling of battery working characteristics and battery state evaluation are important guidance for the control,management and optimization of battery system.Due to the complexity of the battery itself,its working characteristics and state are affected by working conditions,ambient temperature,historical use mode and other factors.These complex characteristics of mutual coupling make the model-based method difficult to apply.In order to solve the above problems,this thesis studies the application of data-driven method in simulation modeling,capacity estimation,life prediction and anomaly detection of lithium-ion batteries.Firstly,based on the DC impedance test data of the battery,the equivalent circuit model and long short-term memory(LSTM)model describing the characteristics of the battery are established.The performance of the model is verified by the experimental data of different discharge rates.The result shows that the LSTM model has higher simulation accuracy than the equivalent circuit model.Secondly,this thesis analyzes the cycle aging data of the battery,extracts the features from the voltage-capacity curve of the battery,and establishes the battery capacity estimation model based on the convolutional neural networks(CNN).The combined CNN-LSTM model is proposed to predict the remaining useful life of batteries.The experimental results show that the voltage-capacity curve is a rich data source for battery state prediction and diagnosis.In order to evaluate the uncertainty of prediction,the selected features are used as the input of Gaussian process regression(GPR),and the capacity estimation model is established.Experiments are carried out on two different battery datasets to verify the effectiveness of the method.Finally,in order to solve the problem that the ground truth of training data is difficult to obtain or the cost of collecting is high,an unsupervised learning model based on the autoencoder is proposed to evaluate the aging state of batteries.Based on the degradation mechanism of batteries,the data set is divided into health and abnormal,and the health data are used to train the autoencoder to learn the internal representation of health data.The reconstruction error of the model is used as the criterion to judge the battery state.The experimental results show that the method can evaluate the battery state according to the input features. |
