Deep Reinforcement Learning Microgrid Optimization Strategy Considering Priority Flexible Demand Side.

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Energy Systems

Background:

• Microgrids face challenges with volatile distributed energy resources (DERs) and demand-side uncertainty.
• Traditional microgrids lack flexible energy dispatch capabilities for complex loads.
• Existing control methods struggle to balance microgrid power constraints with individual flexible load needs.

Purpose of the Study:

• To propose an integrated microgrid environment including wind power, thermostatically controlled loads (TCLs), energy storage systems (ESSs), and price-responsive loads.
• To develop a control strategy that prioritizes TCLs and ESSs for reliable flexible load power supply and cost savings.
• To enhance microgrid operational efficiency and reduce energy input costs through intelligent resource allocation.

Main Methods:

• Formulating the microgrid optimization as a Markov decision process (MDP).
• Implementing a Memory A3C (M-A3C) algorithm, incorporating an experience replay pool, to address data correlation and non-static distribution issues in training.
• Utilizing a multi-threaded approach within M-A3C for efficient learning of demand-side resource priority allocation.

Main Results:

• The M-A3C algorithm demonstrates efficient learning of resource priority allocation for demand-side management in microgrids.
• Flexible scheduling of microgrid demand-side resources is significantly improved, leading to substantial reductions in input costs.
• Comparative analysis shows M-A3C outperforms the Proximal Policy Optimization (PPO) algorithm in convergence and economic optimization.

Conclusions:

• The proposed M-A3C approach effectively manages microgrid resources, ensuring power supply and minimizing costs.
• This method enhances the flexibility and economic viability of microgrids with diverse energy sources and loads.
• The study highlights the potential of advanced reinforcement learning techniques for optimizing microgrid operations.