Deep Learning-Based Stability Prediction for Electrical Grids: A Data-Driven Approach

Stability assessment and prediction is crucial in today’s electrical grids due to the integration of many renewable energy sources, in particular the inverters (DC to AC converters), that could extremely affect grid stability. With the current dependency on electricity, it is not expected at all to have any problem in electricity supply even for a short period of time. Accordingly, the stability of electrical grids is paramount. Traditional techniques to assess and predict grid stability are based on modelling and simulations, besides human experience. This is obviously not suitable with the current complexity and scale of electrical grids. Data-driven approaches, including machine learning and artificial intelligence, offer a real-time and accurate solutions to assess, predict and control grid stability. In this paper, machine learning and deep learning algorithms have been used to assess and predict grid stability. Classical machine learning methods such as logistic regression (LR), random forest (RF), and support vector machines (SVM) have been used, besides more advanced deep learning methods, including multilayer perceptron (MLP) and convolutional neural networks (CNN), have been employed and compared to the classical methods. The results of this work have shown that classical machine learning methods, such as SVM, have shown the highest prediction accuracy ($\mathbf{96.1 \%}$), with deep learning methods such as MLP scoring $95.9 \%$ accuracy. On the precision side, SVM scored 97.2%, while MLP scored 97.4%. Complex deep learning methods, such as the CNN, performed well, but other deep learning and machine learning methods were more competitive. In conclusion, classical machine learning methods are particularly useful to balance model complexity with accuracy.