Simulation of a Dual-axis Solar Tracking System Based on Artificial Neural Networks: A Case Study of Misrata City

Abstract

This research addresses the problem of low efficiency of fixed solar panels and the shortcomings of traditional tracking systems based on photonic sensors, which are highly susceptible to environmental factors. In this research, a control unit was developed using a multi-layer perception model (MLPRegressor) to receive time variables and predict optimal orientation angles. To ensure the highest levels of reliability, astronomical laws were combined with real-world climate data recorded via NASA POWER satellites. Furthermore, a transfer learning strategy was applied to train Artificial Neural Network (ANN). The results showed a 96.97% accuracy in matching real-world data, with orientation deviations remaining within safe geometric tolerance limits. Cosine loss did not exceed 0.4%, and the simulated radiant gain demonstrated a tremendous summer system advantage of 68.13%, achieving an average total annual increase of 25.60% in incident radiation compared to the stationary system.