Verification of Neural Network Behaviour: Formal Guarantees for Power System Applications

This paper presents for the first time, to our knowledge, a framework verifying neural network behavior in power system applications. Up this moment, networks have been applied systems as black-box; has presented major barrier their adoption practice. Developing rigorous based on mixed integer linear programming, methods can determine range of inputs that classify safe or unsafe, and are able systematically identify adversarial examples. Such potential build missing trust operators networks, unlock series new applications systems. framework, assess improve robustness systems, addresses concerns related scalability accuracy. We demonstrate IEEE 9-bus, 14-bus, 162-bus treating both N-1 security small-signal stability.