A Novel Automatic Generation Control Method Based on the Large-Scale Electric Vehicles and Wind Power Integration...

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Control Systems

Background:

• Power systems face frequency instability due to large-scale integration of electric vehicles (EVs) and wind power.
• Traditional reinforcement learning methods can fall into local optima or suffer from action value overestimation.

Purpose of the Study:

• To propose an improved reinforcement learning algorithm for automatic generation control (AGC) in power systems.
• To address frequency instability caused by random disturbances from EVs and wind power.

Main Methods:

• An optimistic initialized double Q-learning algorithm is developed.
• The algorithm incorporates optimistic initialization for broader action exploration and double Q-learning to mitigate overestimation.
• Hyperparameter ατ and exploration-based reward bτ are introduced to enhance learning efficiency and drive optimal solution discovery.

Main Results:

• Simulations on a two-area load frequency control model with EVs and a four-area grid with wind power were conducted.
• The proposed algorithm successfully obtained global optimal solutions.
• The algorithm demonstrated superior control performance compared to existing reinforcement learning methods.

Conclusions:

• The developed optimistic initialized double Q-learning algorithm effectively solves power system frequency instability issues.
• This approach provides a robust solution for grid-connected large-scale wind power generation and EV integration.
• The algorithm offers enhanced control performance and stability in power systems facing significant random disturbances.