Research on the Optimization of Hospital Contract Energy Management Strategy Based on Reinforcement Learning

Abstract This paper proposes an optimization method for hospital contract energy management strategies based on reinforcement learning, aimed at improving the intelligence of strategies and resource allocation efficiency in long-term energy-saving cooperation for healthcare buildings. Addressing the high volatility of hospital energy consumption behavior, the information incompleteness during contract execution, and the uncertainty of service responses, a master-slave game framework is constructed to model the interaction between hospitals and Energy Service Companies (ESCOs). The improved Deep Deterministic Policy Gradient (Hospital-DDPG) algorithm is introduced for strategy optimization. The proposed method adopts an Actor-Critic structure to implement strategy generation and value evaluation, with a Recurrent Neural Network (RNN) embedded in the critic network to capture the temporal dependencies of energy-saving behavior. Additionally, the Burn-In initialization mechanism is incorporated to enhance the model's ability to fit historical states. Moreover, a Value Decomposition Network (VDN) is introduced to achieve collaborative learning among multiple agents, optimizing global energy-saving performance while ensuring data privacy and decision autonomy. Simulation experiments are conducted based on real or constructed hospital energy consumption data. The results show that, compared to traditional DDPG and MADDPG methods, Hospital-DDPG exhibits significant advantages in strategy convergence speed, energy-saving contract benefits, and market adaptability. It effectively reduces reliance on the main platform, minimizes penalty for non-performance, and enhances local energy-saving response efficiency. This research provides a feasible strategy optimization approach and engineering practice reference for smart hospital contract energy management systems.