Estimation of State of Charge for Lithium Ion Batteries using Data Based Deep Learning Approach

In recent decades, urbanization and population growth have led to increased carbon emissions, exacerbating climate change and global warming. This trend has compelled the power industry to transition to alternative energy sources due to heightened fossil fuel consumption and decreasing availability. The utilization of renewable sources for power supply is recognized as a pivotal step toward achieving sustainable energy goals. These sources, of the likes of solar and wind, are variable in nature and do not provide constant dispensable power supply, hence the need for battery energy storage systems (BESS). BESS store surplus energy from these intermittent sources and provide this energy as backup supply in the time of need. In BESS, the state of charge (SoC) holds significant importance for optimizing charge and discharge schedules which directly affect the rate of dispatch of power. The SoC of the BESS depends directly or indirectly on factors such as voltage supply, current source, operational temperatures, and time of use. Therefore, it varies throughout the life cycle of the BESS, highlighting the need for accurate prediction. This study proposes a deep feed forward neural network (DNN) based SoC prediction model for a Li-ion battery. The input layer of the DNN consists of five nodes: voltage, current, temperature along with average current and voltage values. The battery is studied under a variety of temperature conditions in a controlled environment, and data is recorded for constant and variable ambient temperatures. The trained DNN model had an RMSE of 0.025% and showed an accuracy of 2.6% for testing phase.