A CNN-ABC model for estimation and optimization of heat generation rate and voltage distributions of lithium-ion batteries for electric vehicles

Abstract

Accurate estimation of battery voltage and heat generation rate (HGR) at different conditions is critical to manage a battery system on behalf of making safe and efficient operations for electric vehicles (EVs). However, due to malfunction structures within the battery cell, it is extremely challenging task to estimate the HGR and voltage by measuring the external parameters. Although there are many parameter estimation algorithms, it has become essential to make better estimations with deep networks. In this study, we propose a novel scheme, namely convolutional neural network (CNN)-artificial bee colony (ABC) leveraged from CNN and ABC algorithm for HGR and voltage estimation. Unlike other CNN methods, in this proposed CNN model, hidden features in the data are dynamically revealed and thanks to artificial feature extraction. The Voltage and HGR data are used as inputs of the proposed model after the preprocessing operation is terminated. To optimize the HGR and voltage estimates and minimize the error functions, new objective and fitness functions are introduced using the ABC algorithm. The proposed model is tested with several experiments that are carried on 20Ah Lithium phosphate (LFP) battery at different discharge rates of 1C, 2C, 3C, and 4C, and various temperature ranges of 5°C, 15°C, 25°C, and 35°C, and voltage distributions, as making long-term HGR and voltage predictions. The proposed methods are compared with artificial intelligence methods such as Linear Regression (LR), Multi Linear Regression (MLR), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), Artificial Neural network (ANN), Long Term Short Memory (LSTM), and basic CNN. The performance results strikingly show that the proposed CNN-ABC scheme is better than others. The proposed scheme produces 1.38% and 99.72% root mean square error (RMSE) and R2 in HGR estimation, while 3.55% and 99.82% in voltage data estimation, respectively, when applying the ABC to the proposed CNN architecture.