측정 잡음을 고려한 1기 무한모선 시스템의 고장 판별을 위한 DQN 기반 외란 관측기 설계

Abstract

Increasing demand on electric power supply results in a need of an intelligent method for fast detection of various failures in the power system. This paper presents a reinforcement learning-based disturbance observer (DOB) design for the determination and protection against a line fault occurred in the single-machine infinite bus (SMIB) power system. Whilst a high gain disturbance observer could estimate the system states and the external disturbance successfully, the high gain of the observer can cause problems in the presence of the measurement noise. When measurement noise exists in the output, fault detection methods based on the estimated states may often result in false alarms. To solve the problem, this paper designs an adaptive DOB using Deep Q-Network (DQN) which is one of reinforcement learning algorithms. For the proposed observer design, this paper explains the definitions of the state, the action, and the reward for the reinforcement learning. Matlab simulations have been conducted based on the observer gains trained using the power angle data from the swing equation. The results show that the estimation performance of the proposed DQN-based observer can be satisfactory against both an external disturbance and the measurement noise.