Control scheme of green building renewable energy system based on reinforcement learning

Abstract

The traditional construction industry relies heavily on non-renewable energy sources, resulting in high energy consumption. However, traditional renewable energy system regulation models have the problem of weak adaptability. This study proposes a reinforcement learning based regulation model for green building renewable energy systems. The results validated that the reward value of the research model remained stable between −1.3 and −1.4, demonstrating good convergence. When the learning rate was 0.1, the annual cumulative expenditure cost of the proposed model remained stable at 8200–8300 yen, which was better than the comparative model. The similarity between the model and the basic rules reached 94.62%, which was higher than other models. The model had a minimum net load demand of 1.75 kWh in winter and 0.42 kWh in summer. In terms of battery utilization, the model achieved the highest cycle numbers of 2.81 and 3.42 in winter and summer, respectively. In terms of renewable energy utilization, the winter photovoltaic self-consumption rate of the model reached 81.2%, while the summer grid connection rate and self-sufficiency rate were 9.2% and 67.2%, which were better than other models. Overall, this study explores new methods in the field of green building energy regulation, enhances the flexibility and adaptability of energy regulation, and has certain research significance for achieving efficient utilization and sustainable development of building energy.