Adaptive Battery Charger Using Elman Backpropagation Neural Network-PI Algorithm for Varying Voltage on ERSOW Robot

This study presents an adaptive battery charging controller that integrates an Elman Backpropagation Neural Network (EBNN) with a Proportional-Integral (PI) algorithm to manage varying battery voltages on the ERSOW robot. The system identifies the connected battery type- $12 ~\mathrm{V}, 16 ~\mathrm{V}$, or 24 V lithium-ion-and dynamically adjusts the charging voltage and current to optimize performance and extend battery life. Operating under the Constant Current-Constant Voltage (CCCV) mode, the EBNN processes real-time voltage and current inputs to generate appropriate setpoints, while the PI controller ensures stable regulation during charging. A SEPIC converter is used to handle the voltage adjustments. Simulation results demonstrate that the EBNN-PI method delivers higher precision than the Feedforward Backpropagation Neural Network-PI (FFBNN-PI) approach. In current regulation, EBNN-PI achieves an average error of 0.5 %, compared to 1.2 % with FFBNN-PI. For voltage regulation, the EBNN-PI controller attains an average error of only 0.072 %, outperforming the 0.267 % of the FFBNN-PI system. These improvements underline the advantages of the EBNN's temporal learning capability. Moreover, the proposed control framework is adaptable to various battery chemistries, including LiFePO 4, VRLA, and NiMH. By extending the training dataset, the system can scale to support a wide range of energy storage applications in robotics and renewable energy systems with enhanced reliability and safety.