Scalable imputation technique for effective computation of missing battery parameters

The collection of high quality battery datasets is critical for applications such as State-of-Charge or State-of-Health estimations. Missing features within the dataset is a challenging problem. The issue is further exacerbated in on-field electric vehicle applications. One method to resolve this challenge is to impute missing battery parameter values with best possible estimates. In this paper, we explore a self-attention technique for imputing missing battery parameters. A masked self-attention block enables accurate estimates of missing battery data. The introduced technique helps provide potential solutions to resolve the issue of missingness in battery parameters. We provide two primary contributions, (a) we illustrate the effectiveness of self-attention mechanism to provide best estimates of missing battery parameters, (b) we highlight the efficacy of the approach over available state-of-the-art imputation models. We release the code used in this scenario in the results section of our paper. This will allow researchers to effectively reproduce the results presented here.