ON METHODS OF DETECTING THE CYBERATTACKS ON SMART ENERGY GRIDS WITH HYBRID DEEP LEARNING MODELS

Abstract

The article presents a new method of detecting cyberattacks in intelligent energy networks based on methods of detecting anomalies with partial training. It is based on the neural network of an autoencoder with a flexible encoder model used in the method to extract representations of the input data of the power system. One of the methods of teaching with a teacher was used as an autoencoder decoder in the cyberattack detection model. The peculiarity of the proposed method is that in the learning process it uses only the signs of regular events. A comparative analysis of the performance of the proposed method with popular machine learning algorithms with a teacher, such as KNN, SVM, Random Forest, showed that it is more effective in detecting cyberattacks, and the accuracy of their detection was 91.67%.