LSTM-Based Smart Grid Stability Prediction Supported by Higher-Order Interaction Features

The stability of smart grids is critical to ensure reliable and efficient power distribution in modern energy systems. In this study, stability refers to the ability of the system to maintain power balance equilibrium under varying configurations of load behavior and price elasticity. Traditional approaches to stability prediction often overlook the complex interactions between input variables, limiting their predictive performance. In this study, a novel framework for the prediction of smart grid stability is presented, where a long-short-term memory (LSTM) network is supported by statistical indices of higher-order features-interaction (FI) to enrich the feature space. An LSTM model is developed to predict smart grid stability, achieving state-of-the-art performance. The required dataset was prepared by simulating a 4-node star network with one producer and three consumers, where each data sample reflects a specific configuration of power values, reaction times, and price sensitivities. The simulation results show that the proposed approach significantly outperforms the baseline models, reaching an accuracy of 99.3% along with superior precision. The inclusion of statistical indices of higher-order interaction features not only improves predictive performance, but also provides deeper insights into the underlying dynamics of smart grid stability. This work highlights the potential of feature engineering to improve the reliability and efficiency of machine learning (ML) in smart grid systems.