Detection of Smart Grids Instability with Convolutional Neural Networks and Global Explainability

A smart grid is an advanced concept of power systems devoted to harmonize electricity and communication in system networks. It is able to provide real-time information to producers, operators and consumers. There is an urgent demand to efficiently route supplied energy to consumer domains such as, for instance, households, organisations, industries and also smart cities. In this context, a smart grid with a stable system is required to supply the dynamic energy demand. In this paper, we propose a method aimed to detect, in real-time, whether a smart grid is in an instable state. We exploit deep learning, by considering a neural network developed by authors and global explainability, with the aim to determine the features useful in the model prediction. We evaluate a dataset composed by 60000 smart grid observations we obtain interesting results, by demonstrating the effectiveness of the proposed method.