Adaptive resilient control and detection mechanisms for secure nanogrid operations facing FDI attacks

Abstract

Paralleled DC hybrid Nanogrid (DCHNG) with a communication network is configured for the profit of consumers. In this structure, parallel power converters with distinct generation units are established to supply constant power loads (CPLs). In particular, the impact of false data injection (FDI) attack on injecting false data into measurement signals is investigated from a systematical point of view. The case study is constructed from two energy storage units which supply the connected loads to DCHNG. An adaptive resilient scheme has been developed in the current paper to address the destructive effects of FDI attacks on DCHNG systems. In the first step, a detection mechanism based on the non-integer extended state observed is developed to estimate the system output while the occurrence of cyber threats is identified by the residual scheme. Then, the mitigation mechanism is applied to eliminate the effect of the FDI attack and stabilize the output of DC hybrid Nanogrid. The twine-delayed actor-critic (TDAC) scheme, as an advanced reinforcement learning (RL), is developed for the adaptive design of the mitigation mechanism. The twin critic networks assess the quality of selected variables of mitigation mechanism, and the policy is updated by the actor network according to the evaluation. The training of deep neural networks (DNNs) of TDAC is realized in such a way that the effect of FDI threats disappears and stabilizes the DCHNG under CPLs simultaneously. Compared with the conventional cyber defense schemes which are developed based on the system model, the proposed technique doesn't need the dynamic model of the system. The real-time tests based on the Arduino Mega2560 setup are carried out and real-time tests of DCHNG reveal the effectiveness of the suggested scheme to address cybersecurity issues.