On the Creation of a Chess-AI-Inspired Problem-Specific Optimizer for the Pseudo Two-Dimensional Battery Model Using Neural Networks

Abstract

In this work, an artificial intelligence based optimization analysis is done using the porous electrode pseudo two-dimensional (P2D) lithium-ion battery model. Due to the nonlinearity and large parameter space of the physics-based model, parameter calibration is often an expensive and difficult task. Several classes of optimizers are tested under ideal conditions. Using artificial neural networks, a hybrid optimization scheme inspired by the neural network-based chess engine DeepChess is proposed that can significantly improve the converged optimization result, outperforming a genetic algorithm and polishing optimizer pair by 10-fold and outperforming a random initial guess by 30-fold. This initial guess creation technique demonstrates significant improvements on accurate identification of model parameters compared to conventional methods. Accurate parameter identification is of paramount importance when using sophisticated models in control applications.