CNN-LSTM optimized with SWATS for accurate state-of-charge estimation in lithium-ion batteries considering internal resistance

Abstract

Accurately estimating the state-of-charge (SOC) of lithium-ion batteries is of great significance for the energy management and range calculation of electric vehicles. With the development of graphics processing units, SOC estimation based on data-driven methods, especially using recurrent neural networks, has received considerable attention in recent years. However, existing data-driven methods often neglect internal resistance, which is highly detrimental to the accuracy of SOC estimation. In addition, commonly used network optimization algorithms do not always maximize the convergence speed and performance simultaneously. To solve these problems, this paper describes a battery test bench for producing an effective lithium-ion battery dataset containing current, voltage, temperature, and more importantly, internal resistance measurements. To improve the estimated SOC performance, the internal resistance is considered in the construction of a data-driven model. Using a convolutional neural network (CNN) and long short-term memory (LSTM), we propose an optimization model that switches from Adam to stochastic gradient descent (SWATS). A well-known public battery dataset and an experimentally measured dataset are used to verify the feasibility of the SWATS scheme. The results show that, compared with existing data-driven methods, the proposed method is effective, especially in terms of robustness and generalization.