iTransformer Based Voltage Estimation of Lithium-Ion Batteries

Accurate battery voltage modeling is fundamental to optimizing performance, ensuring safety, and extending battery life in Electric Vehicles (EVs). Precise voltage modeling also enables unswerving estimates of range, reducing range anxiety for drivers. Conventional battery voltage models have been equivalent circuit models or other empirical models. However, these methods frequently fall short in accurately capturing the complex characteristics of batteries. To overcome these limitations, this paper introduces the iTransformer machine learning model as an alternative for voltage prediction in EV batteries. iTransformers enhance time-series predictions by applying attention across variates rather than time steps, improving accuracy for multivariate tasks like battery voltage modeling, while maintaining the parallel processing efficiency of transformers. Its inverted architecture enables it to better capture cross-feature relationships and temporal dependencies, as seen in battery data, while maintaining computational efficiency. Our experimental results demonstrate the model's effectiveness, achieving an RMSE of 30 mV and an MAE of 21 mV on the test dataset from an LG 18650HG2 Li-ion cell, which are within the range required in most practical applications, indicating high accuracy for battery voltage modeling.