Black-box Stealthy Frequency Spectrum Attack on LSTM-based Power Load Forecasting In An Energy Management System with Islanded Microgrid

This research paper introduces a novel approach called Frequency Spectrum Attack (FSA) and assesses its impact on load forecasting and energy management within a microgrid. FSA leverages Fast Fourier Transform to convert load data into the frequency domain, thereby uncovering significant patterns during the learning phase. It strategically adjusts the amplitudes of dominant frequencies within the healthy amplitude range to maintain stealthiness against statistical analysis, contributing to its effectiveness. We evaluate the performance of FSA using LSTM network as a state-of-the-art technology for load forecasting. Our findings reveal that FSA results in a threefold increase in Mean Absolute Error (MAE) compared to normal conditions and a 70% increase compared to noise injection attacks. Additionally, FSA indirectly boosts battery utilization in the Energy Management System (EMS) by 45%. Furthermore, FSA effectively evades detection by frequency monitoring and control units in the microgrid, making frequency deviations inconspicuous.