Enhancing State of Charge Estimation in Ni-MH Batteries Through a Hybrid Approach Incorporating RRBF and K-Means

The accurate prediction of the State of Charge (SOC) in batteries is crucial for optimizing their usage and extending their end of life. In this paper, we focus on the utilization of two neural network architectures, namely Radial Basis Function (RBF) networks and Recurrent Radial Basis Function (RRBF) networks, for predicting the SOC of Ni-MH (Nickel-Metal Hydride) batteries. RBF networks are commonly employed in static modeling tasks but struggle to capture the temporal dependencies present in many dynamic modeling problems. RRBF networks have been specifically designed to overcome this limitation through their recurrent architecture, which allows them to incorporate data history and model temporal dependencies. Additionally, in this conference, we introduce the k-means clustering algorithm to optimize the parameters of Gaussian functions used in RRBF and RBF networks. The k-means clustering groups the training data into clusters, enhancing the prediction quality of SOC by utilizing specialized sub-networks for each cluster.