Synthetic Data Generation using Resilient Sample Convolution and Interactive Learning Approach

Abstract

There’s been an increase in attempts to control the surge in pollution levels due to extensive exploitation of conventional fossil fuels. These efforts have fueled research for alternative green solutions. Lithium-ion batteries are immensely beneficial for energy storage system. They are extremely advantageous in the automobile industry, particularly as a source to power Electric Vehicles (EVs). Lithium-ion batteries are also vital for powering consumer electronics. The State of Charge (SOC) measurement is used to calculate the remaining usage time of batteries, is one of the most pertinent metric. The goal of current research has been to develop accurate State of Charge (SOC) prediction algorithms. All existing methods require significant amount of superior-quality curated dataset. However, battery researchers have minimal access to commercial battery datasets and therefore must rely on open-access public datasets that lack the required heterogeneity to generate generalised SOC algorithms. To resolve this issue of lack of data, we introduce a Sample Convolution and Interaction Networks (SCINet) to produce resilient synthetic battery data. The code implementation can be found on: https://github.com/vinayakrajurs/Sample-Convolution-Interaction-Syntheic-Data