Optimise transient control against DoS attacks on ESS by input convex neural networks in a game

Abstract

For improving the transient control of power grids, an alternative fast acting feature is offered by the vast application of energy storage systems (ESS). However, Denial of Service (DoS) attacks can interfere with ESS, significantly hindering the performance of the transient control. In the present study, a framework of a non-zero sum difference game between a transient controller and a DoS attacker was established to determine an optimal control strategy of ESS in DoS attacks. In order to solve the aforementioned game, an input convex neural networks (ICNNs) based adaptive dynamic programming (ADP) scheme was proposed. Regarding the DoS attacker and the transient controller, two long term strategy utility functions are approximated by ICNNs, thereby ensuring the existence of the Nash equilibrium (NE) of the game without requiring a small sampling period or linearity of power system models. Through neural training to reach the NE from the framework of the game, an approximated optimal controller is derived. The simulations of the 9-bus and 30-bus system experiments validate the effectiveness of the proposed transient control scheme against DoS attacks. Further, increased performance of the ICNNs based ADP in multiple sampling periods and the nonlinearity of the power grids are illustrated.