A Neural Network Based Control Strategy for Constant Current Battery Chargers with PV Source

• Main Authors: Becchi, Lorenzo, Bindi, Marco, Garzon Alfonso, Cristian Camilo, Intravaia, Matteo, Lozito, Gabriele Maria, Grasso, Francesco
• Format: Conference Proceeding
• Language: English
• Subjects: Accuracy; Batteries; constant current charging; DC-DC power converters; Feedforward neural networks; Integrated circuit modeling; Load modeling; Monitoring; neural networks; photovoltaics; Temperature measurement; Temperature sensors; Training
• ISSN: 2687-6817
• DOI: 10.1109/RTSI61910.2024.10761440

In this work, a novel strategy to determine the optimal duty cycle of a boost-type converter for battery charging applications from photovoltaic source is proposed. The optimal duty cycle is determined to keep the battery charging current constant at a fixed value. The strategy is based on a preliminary analysis of the conversion chain from the PV source to the battery load, from which a steady-state circuit model is formulated. From this model, a dataset is created and used to train a low complexity artificial neural network to estimate the optimal duty cycle of the converter for a desired output current. The network inputs are electrical quantities normally monitored in power conversion applications, and the temperature of the PV source. The network achieves good results in terms of training, validation, and test performance. In particular, the test performance on the optimal duty cycle is 0.0036 in terms of Root Mean Square Error and 0.98% in terms of Mean Absolute Percentage Error. This corresponds to an accuracy on the charging current of 5.32%.