An Early Remaining Useful Life Prediction Method for Lithium-ion Batteries using Optimization Algorithm-Assisted Gaussian Process Regression

Lithium-ion batteries are widely used as power devices for electric vehicles due to their high energy density and fast charging capability. However, battery life degradation adversely affects safety management and utilization maintenance. Therefore, research on battery life prediction can help early health diagnosis and residual value assessment to prevent system downtime. Aiming at the problems of poor early prediction ability of current methods, this paper proposes an early life prediction method for lithium-ion batteries using optimization algorithm-assisted Gaussian Process Regression. Firstly, six health features strongly correlated with end-of-life are extracted from the battery's first 100 cycles of charge/discharge data. Then, a battery life prediction model is established based on Gaussian Process Regression with composite kernel functions. The particle swarm optimization algorithm is employed to optimize the hyperparameters of the composite kernel functions, enhancing the adaptiveness of the proposed prediction method. The validation result shows that the prediction is highly accurate, and the mean absolute percentage error of early life prediction can reach 10.66% overall. The Gaussian Process Regression model based on hyperparameter optimization proposed in this paper has better prediction accuracy compared with the prediction results without hyperparameter tuning and with a single kernel function. The method proposed in this paper also outperforms other common machine learning algorithms in terms of prediction accuracy.

I. Introduction

New energy vehicles are favored due to their alignment with the sustainable concept of modern society. Lithium-ion batteries are widely used in new energy vehicles because of their high energy density, fast charging capability, and low self-discharge rate. However, it is a dynamic, nonlinear fading system with complicated internal mechanisms. With the increase of charge/discharge cycles, physicochemical reactions will lead to loss of lithium-ion inventory (LLI) and loss of anode/cathode active materials (LAM) [1], and that results in the degradation of battery performance. When the battery’s capacity decays to 80% of initial capacity or internal resistance increases to 200% of initial resistance, the battery may experience strong failures such as accelerated attenuation and unexpected downtime. At that time the battery should be replaced to ensure expected performance and safety. Besides, retired batteries may also have a high residual value and can be used in applications where high performance is not critical, such as energy storage. Therefore, battery cycle life prediction before severe degradation is crucial for early health diagnosis, timely safety maintenance, residual value assessment, and regulation of secondary utilization.