Parameter identification of lithium-ion battery pseudo-2-dimensional models using genetic algorithm and neural network cooperative optimization

The electrochemical model parameters of a lithium-ion battery are important indicators of its state-of-health, and many previous studies have proposed methods for identifying them. These identification methods must solve highly nonlinear optimization problems with many local optima. Hence, metaheuristic approaches are often employed. Most metaheuristics take a way to abandon worse solutions and make the most use of better solutions only. Such inefficient use of data leads to local optima problem in metaheuristics. To overcome these limitations, this paper proposes a novel parameter identification method in which a neural network cooperates with a genetic algorithm. The proposed method adopts an 1-dimensional convolutional neural network to learn the dynamics between the known input current and the corresponding simulated voltage. Although estimated parameters cause large output voltage errors, they are useful for building an electrochemical model and can be used to recommend highly probable parameter candidates. We clearly show through simulation and experiment that the electrochemical model parameters are identified more accurately and reliably compared with various existing results, owing to the high data efficiency of the proposed method.

Automated summary

This paper proposes a novel parameter identification method for lithium-ion batteries using a neural network and genetic algorithm, resulting in more accurate and reliable identification of electrochemical model parameters with high data efficiency.