Forecasting of Power Generation in a Single-Axis Solar Tracking PV System Using an Enhanced Artificial Neural Network-Based Method

In order to anticipate photovoltaic (PV) power output in both fixed and tracking solar systems, this study proposes a strong neural network-based framework that models nonlinear dependencies by utilising meteorological factors such as temperature, wind speed, and sun radiation. Strong correlation coefficients (𝑅²) and low mean squared errors (MSE) throughout the training, validation, and testing phases demonstrate the model's high predictive accuracy, which was attained by combining a 10-layer artificial neural network (ANN) architecture optimised with the Adam algorithm and a dynamic learning rate scheduler. To guarantee generalisability, the dataset—which included 8,761 hourly samples over a full year—was carefully divided into three categories: 70% training, 15% validation, and 15% testing. The impact of system design on productivity was highlighted by a comparative analysis that showed a 21% improvement in annual energy yield for tracking systems (231 kWh) versus fixed systems (184 kWh). Regression plots, error histograms, and monthly power generation profiles were among the visual and statistical assessments that showed how well the model captured seasonal and diurnal variations while reducing bias. With error rates lowered to less than 10% and prediction accuracies over 90% in both contexts, the combination of the MATLAB and Python frameworks further confirmed the method's consistency and scalability.