Applying a Machine Learning Model to Estimate the Current State of Charge of Energy Storage Devices

This research examines the role of applying machine learning to estimate the state of charge (SoC) of energy storage devices. SoC is a vital parameter that can be used to reflect the performance of the energy storage device and is key in the management of these energy storage systems, such as batteries or a hot water buffer. It is used to optimize the performance and to extend the lifetime of storage systems. A simple artificial neural network (ANN) and a time-series long short term memory (LSTM) neural network is investigated to create a model that can be used on multiple forms of energy storage devices. The simple ANN achieved results of over 80% accuracy for the Yuasa and Multipower batteries but did poorly at 25% accuracy for the Conrad battery. The LSTM achieved an accuracy of over 90% for 4 batteries and a hot water buffer. The LSTM model is then altered so that it is able to estimate the SoC in real time for the Conrad battery, this achieved a lower accuracy of 70.1%.