PV Farm Power Generation Forecast using PV-Battery Model with Machine Learning Capabilities

This study presents a photovoltaic (PV) battery model with machine learning capabilities to predict PV farm power generation, offering a valuable tool for real-world energy management and planning. Existing models often suffer from limitations in predictive accuracy, computational efficiency, and adaptability to complex temperature and irradiance data variations, reducing their practical effectiveness. To overcome these gaps, the proposed model integrates an Adaptive Neuro-Fuzzy Inference System (ANFIS) and a multi-input multi-output (MIMO) prediction algorithm, using historical temperature and irradiance data for accurate and efficient power forecasting. Simulation results demonstrate the model’s robustness, achieving high prediction accuracies of 95.10% for temperature and 98.06% for irradiance, while reducing computational demands and outperforming conventional curve-fitting and Artificial Neural Network (ANN) techniques. Similarly, the electrical model utilizes ANFIS outputs to estimate PV farm power generation while efficiently managing the battery’s state of charge (SOC), exhibiting a minimal SOC reduction ofonly 0.88% (from 80% to 79.1212%) over a seven-day charge-discharge cycle and providing up to 11 hours of battery-bank autonomy under specified load conditions. Validation with three distinct datasets further demonstrated the ANFIS network’s ability to handle diverse and complex data variations with consistent accuracy.