Application of Reinforcement Learning-Based Adaptive PID Controller for Automatic Generation Control of Multi-Area Power System

This paper presents an application of an actor-critic reinforcement learning (RL) algorithm-based adaptive proportional-integral-derivative (PID) controller for automatic generation control of multi-area power systems. The proposed approach has several advantages over other deep RL algorithm-driven PID controllers, such as simplicity in structure, elimination of pre-learning requirements, and prior tuning of PID parameters. Online adaption of PID parameters is achieved through actor-critic policy. The proposed method implements a single radial basis function (RBF) based neural network for actor and critic networks. Three different case studies are demonstrated with proper illustration and analysis of the result to present the effectiveness and robustness of the proposed control strategy against various uncertainties. The outcomes of the proposed controller are compared with the conventional PID controller tuned by the Particle Swarm Optimization (PSO) algorithm. The results seem competent enough to maintain the frequency within an acceptable limit under various uncertainties.
Note to Practitioners
—This paper describes the application of an online adaptive PID controller for automatic generation control of power systems. The controller is designed using a model-free reinforcement learning algorithm, which enables it to control the system without requiring prior knowledge of the system dynamics. Additionally, the controller does not need any global optimization algorithm for tuning the parameters (
$K_P$
,
$K_I$
,
$K_D$
) beforehand. This controller can be implemented for both linear and non-linear systems.