DREOA: A Deep Reinforcement Learning Approach for Energy Optimization in Power Collection Systems

Energy efficiency optimization in power collection systems is critical for sustainable economic development and environmental protection. Traditional methods face challenges in handling the vast, heterogeneous, and non-linear data generated by modern power systems. In this paper, we propose the Deep Reinforcement Energy Optimization Algorithm (DREOA), which combines deep learning and reinforcement learning to address these challenges. DREOA employs a Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) network for effective feature extraction and temporal dynamics modeling, respectively. An adaptive learning rate mechanism and communication pruning are incorporated to enhance real-time performance and reduce computational overhead. We evaluated DREOA on three open-source datasets: the UCI Energy Efficiency Dataset, Pecan Street Dataset, and Electricity Load Diagrams Dataset. The results show that DREOA outperforms traditional machine learning models and existing deep reinforcement learning algorithms, achieving higher accuracy and significant energy efficiency improvements of up to 15.3%. An ablation study confirms the effectiveness of each component of the proposed algorithm. DREOA demonstrates robustness and scalability, making it a promising solution for real-world power systems. The proposed method contributes to the advancement of intelligent power management and offers a pathway towards more sustainable energy utilization.