This paper presents a neural network-based control strategy designed to achieve maximum power tracking (MPT) in smart public buildings. Unlike conventional rule-based energy systems, our approach integrates a recurrent neural network (RNNRecurrent Neural Network (RNN)) for real-time forecastingReal-time forecasting of power demand and environmental dynamics, coupled with an ensemble learning model to optimize control actions. The system aims to maximize the extraction and utilization of renewable energy sources (such as solar or piezoelectricPiezoelectric systems), while dynamically responding to changing load profiles and building occupancy patterns. We demonstrate that the proposed control architecture enhances power utilization efficiency, ensures system robustness under uncertainty, and supports autonomous operation on embedded edge devices. Experimental evaluation on realistic datasets confirms improved energy stability, tracking precision, and fault tolerance compared to traditional predictive models.