A Condition Knowledge Representation and Feedback Learning Framework for Dynamic Optimization of Integrated Energy...

Area of Science:

• Energy Systems Engineering
• Artificial Intelligence
• Optimization Theory

Background:

• Integrated energy systems (IESs) require efficient energy scheduling to enhance energy utilization and minimize carbon emissions.
• The large state space of IESs, influenced by uncertainties, poses challenges for model training and requires effective state-space representation.
• Existing methods may struggle with the dynamic and uncertain nature of IES operations, impacting cost-effectiveness.

Purpose of the Study:

• To design a condition knowledge representation and feedback learning framework for optimal energy scheduling in IESs.
• To develop a dynamic optimization model that partitions condition samples based on pre-optimized daily costs.
• To improve the state-space representation and policy learning performance in uncertain IES environments.

Main Methods:

• A contrastive reinforcement learning framework is proposed, incorporating a contrastive network for time-dependent state-space representation.
• A dynamic optimization model based on deterministic deep policy gradient is established for condition sample partitioning.
• A Monte-Carlo policy gradient-based learning architecture is utilized to optimize condition partitioning and enhance policy learning.

Main Results:

• The proposed method demonstrates superior cost-effectiveness compared to human experience strategies and state-of-the-art approaches.
• The framework effectively constrains uncertain states within the IES environment through improved state-space representation.
• Simulations on typical IES load operation scenarios validate the approach's adaptability in uncertain conditions.

Conclusions:

• The developed framework offers an effective solution for optimal energy scheduling in integrated energy systems.
• The approach significantly improves cost efficiency and adaptability in dynamic and uncertain operational environments.
• This research contributes to advancing intelligent control strategies for sustainable energy management.