Sustainable Energy Dispatch Optimization with Threshold-Gated Reinforcement Learning

Ensuring demands are satisfied while minimizing operating costs remains a challenging problem for reinforcement learning (RL), particularly in long-horizon energy management settings. In this work, we propose a unified training framework that combines curriculum learning with a threshold-gated reward mechanism for sequential energy dispatch with battery storage under uncertain operating conditions. Unlike conventional weighted reward formulations that optimize competing objectives simultaneously, the proposed reward design encourages the agent to first minimize demand–supply imbalance. Once a prescribed imbalance threshold is achieved, the learning process focuses on improving secondary objectives, allowing the policy to select lower-cost operating strategies among the solutions that satisfy the primary reliability requirements. Curriculum learning progressively increases task complexity, enabling the agent to gradually learn the desired reliability and economic objectives. Extensive experiments demonstrate that the proposed curriculum-guided threshold-gated framework maintains the absolute value of mean imbalance gaps below 3% while achieving operating costs close to those of the linear programming benchmark. Comparative studies against conventional reward formulations and ablation analyses further show that the framework effectively enforces objective priorities during learning, resulting in improved constraint satisfaction, enhanced renewable energy utilization, and competitive operating costs. These results highlight the framework’s ability to support reliable, cost-effective, and sustainable power system operation.