Protection of Power System during Cyber-Attack using Artificial Neural Network

Abstract

Impacts of frequency and voltage disturbance on an isolated power system caused by cyber-attack have been discussed, and a neural network-based protective approach has been proposed in this research work. Adaptive PID controllers for both load frequency control and automatic voltage regulator have been implemented using an artificial neural network-oriented by genetic algorithm. The parameters of the PID controller have been tuned offline by using a genetic algorithm over a wide range of system parameter variations. These data have been used to train the neural network. Three input switch has been implemented to control governor speed regulation and amplifier gain. For load frequency control neural network tuned PID controller mitigate frequency disturbance 48% faster than manually tuned PID and for the automatic voltage regulator, neural network tuned PID controller mitigate voltage disturbance 70% faster than manually tuned PID during cyber-attack.

A new linear quadratic regulator model to mitigate frequency disturbances in the power system during cyber-attack

“…3. Comparisons among the results of self-implemented models of LQRMFD, LFC, AGC and AGC-PID using new power system as well as existing models of LFC [21], AGC [21], AGC-PID [21], CSU [21], and AGC-PID-ANN [24]. Here, "--" refers to not available…”

“…Some works have been done in the literature to protect CA on PS (e.g., [18]- [24]). Particularly, load frequency control (LFC) and automatic generation control (AGC) units are introduced in [18]- [20] that are bound to retain stable frequency with load variation during CA.…”

“…Furthermore, a three-input switch for LFC of an isolated PS has been proposed in [23] to reduce FD during CA but failed to achieve PS stability because of demanding a considerable time. Islam et al [24] have tried to solve the frequency disturbances problem during CA using ANN in the AGC-PID model but failed to significantly improve because of demanding larger time. Recently, Li et al [25] proposed an LFC-based method for GAN networks to mitigate the frequency disturbances during CA, but no significant performance analysis has been reported.…”

“…As we know, LQR can remarkably improve the static and dynamic response of a closed-loop PS [8], [9]. Since the conventional controllers (e.g., LFC [23], integral controller, or AGC [23], AGC-PID [21], CSU [21], ANN [24]) produce a very slow dynamic response. One optimal controller (e.g., LQR) that is much more effective than conventional controllers has been invented.…”

“…Second, the proposed model uses a new PS network to efficiently mitigate the frequency disruption in PS during CA. It is noted that most findings in this arena were done based on similar PSs of the same ratings (e.g., [21], [23], [24]). In such cases, the potentiality of finding better problem solutions is low as the chance of diversity of PS platforms is low.…”