Energy Dispatch for CCHP System in Summer Based on Deep Reinforcement Learning.

Area of Science:

• Energy Systems Engineering
• Artificial Intelligence in Energy
• Sustainable Energy Management

Background:

• Combined cooling, heating, and power (CCHP) systems offer energy and environmental benefits.
• Energy dispatch optimization in CCHP is challenging due to load uncertainty, prediction errors, and demand charges.
• Existing methods often rely on prediction information, limiting adaptability.

Purpose of the Study:

• To develop an adaptive energy dispatch strategy for CCHP systems using deep reinforcement learning.
• To address the challenges of load uncertainty and demand charges in CCHP operation.
• To optimize CCHP system performance during summer operation.

Main Methods:

• A dispatch method based on the Deep Reinforcement Learning (DRL) algorithm, specifically DoubleDQN, was proposed.
• The DRL approach was integrated to generate optimal dispatch strategies without requiring prediction information.
• The method was tested and validated through simulations and extended scenarios.

Main Results:

• The proposed DoubleDQN strategy significantly reduced total intra-month costs by 0.13% to 31.32% compared to benchmark policies.
• Demand charges were notably reduced, ranging from 2.19% to 46.57%.
• The strategy effectively preserved thermal comfort while adapting to load uncertainties.

Conclusions:

• Deep Reinforcement Learning (DRL), particularly DoubleDQN, provides an effective and adaptive solution for CCHP energy dispatch optimization.
• The proposed method demonstrates practical applicability and potential for real-world implementation in CCHP systems.
• This approach offers substantial cost savings and maintains operational comfort in dynamic energy environments.