Linear Regression Model of Li-Ion Battery Capacity Losing Rate Based on Equivalent Circuit Model Parameters and Operation Modes

The battery is a high-price consumable unit. Its degradation depends on a method of battery usage. The knowledge of losing capacity in real-time can be employed to design a control battery unit or to test a complex system. Therefore, an idea arose to design a linear regression model that can predict the current capacity fading rate. For this purpose, a particular data convolution is proposed. The training process of this model is performed on time normalized and selected battery cell states between a couple of nearby capacity tests. An additional feature of the proposed approach is that the resulting vectors are divided into parts. These parts describe different types of aging that occur during charging, discharging, and idle modes. Information for the regression model is obtained from conventional battery sensors and from observed parameters of battery equivalent circuit model. To train and validate the model NASA randomized battery usage data set was applied. The model consists of a great number of elements, which demonstrates a possible large degree of freedom. However, at the same time, the trained model uses less than 1 % of the elements of the linear regression model, which makes it fast and reliable. It has sufficient accuracy and can be used in real-time or in prior capacity loss rate evaluation. Long-term tests show that the model outputs satisfactorily follow the actual data from capacity tests. The best practices of this work can be taken to design optimization problems for various battery energy storage systems as well.