When applying deep reinforcement learning to the scheduling tasks of integrated energy systems (IES)

the multi-energy coupling structure of IES often leads to reward functions composed of multiple coupled components. The design of reward functions fundamentally determines how agents perceive task objectives and action values; however

excessive reward terms can easily lead to policy confusion and suboptimal optimization. To address this

this paper proposes a reward decoupling method based on hierarchical reinforcement learning. Firstly

a hierarchical decision-making approach for electric-thermal units is developed according to the coupling mechanism of reward functions in IES economic scheduling. Subsequently

the optimal scheduling problem is formulated as a Markov decision process

with specifically designed algorithms for the rapid response characteristics of electrical systems and the thermal inertia features

respectively. Ultimately

a hierarchical reinforcement learning model is constructed to separate decision-making processes for electrical and thermal units

thereby achieving decoupling of electric-thermal components in the reward function. Case simulation results demonstrate that the proposed model effectively avoids mutual interference between cross-energy action reward signals

enhances operational economic benefits while aligning with different scheduling timescales for electricity and heat

verifying the effectiveness and superiority of the reward decoupling method.