<div>This paper presents a suite of algorithms for detecting and localizing attacks in cyber-physical systems, and performing improved resilient state estimation through a pruning algorithm. High performance rates for the underlying detection and localization algorithms are achieved by generating training data that cover large region of the attack space. An unsupervised generative model trained by physics-based discriminators is designed to generate successful false data injection attacks. Then the generated adversarial examples are used to train a multi-class deep neural network which detects and localizes the attacks on measurements. Next, a pruning algorithm is included to improve the precision of localization result and provide performance guarantees for the resulting resilient observer. The performance of the proposed method is validated using the numerical simulation of a water distribution cyber-physical system.</div>

<div>This paper presents a suite of algorithms for detecting and localizing attacks in cyber-physical systems, and performing improved resilient state estimation through a pruning algorithm. High performance rates for the underlying detection and localization algorithms are achieved by generating training data that cover large region of the attack space. An unsupervised generative model trained by physics-based discriminators is designed to generate successful false data injection attacks. Then the generated adversarial examples are used to train a multi-class deep neural network which detects and localizes the attacks on measurements. Next, a pruning algorithm is included to improve the precision of localization result and provide performance guarantees for the resulting resilient observer. The performance of the proposed method is validated using the numerical simulation of a water distribution cyber-physical system.</div>