Rafiqul Islam Roky, Shahin Shaikh, Md Nazmulhuda
School of Electronic and Electrical Engineering, Ningxia University, Electrical Engineering, Chongqing University of Technology, Electrical and it’s Automation Engineering, North China Electric Power University
rokyrafiqul@gmail.com; shahinshaikhkpc@gmail.com; nazmul.huda2450@gmail.com;
Abstract
This paper introduces an enhanced Unscented Kalman Filter (UKF) algorithm integrated with
Radial Basis Function (RBF) neural networks to advance the accuracy of nonlinear state
estimation in dynamic systems. Our approach specifically addresses the estimation of the State
of Charge (SOC) of a battery, leveraging a second-order equivalent circuit model to capture
the battery’s complex behavior. The innovation of our method lies in the integration of RBF
neural networks into the UKF framework, which enhances the algorithm’s capability to model
nonlinearities and improve prediction accuracy.The standard UKF algorithm, while robust in
handling nonlinear systems, often struggles with certain nonlinearities inherent in battery SOC
estimation. By incorporating an RBF neural network, which excels at approximating complex,
nonlinear relationships, our proposed UKF-RBF algorithm achieves superior performance. The
RBF network is trained to capture the nonlinear Open Circuit Voltage (OCV) vs. SOC
relationship, which is crucial for accurate SOC estimation.Experimental results demonstrate
that the UKF-RBF algorithm significantly outperforms the traditional UKF in terms of Mean
Squared Error (MSE), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE).
The UKF-RBF algorithm shows marked improvements in SOC estimation accuracy across
varying operating conditions and temperatures, making it a robust solution for practical
applications in battery management systems. The integration of RBF neural networks into the
UKF framework represents a novel approach that bridges the gap between traditional Kalman
filtering and modern neural network techniques, providing a substantial enhancement in the
estimation of nonlinear states.
Keywords:
Neural Networks, CNN, UKF-RBF, OCV
Citation
Rafiqul Islam Roky, Shahin Shaikh, & Md Nazmulhuda. (2025). Enhanced Nonlinear Estimation with Unscented Kalman Filter and RBF Neural Networks. In Journal of Global Knowledge and Innovation (Version 1, Vol. 2, Number 2, pp. 65–81). Journal of Global Knowledge and Innovation (JGKI). https://doi.org/10.5281/zenodo.14623346