Load Probability Density Forecasting Under FDI Attacks Based on Double-Layer LSTM Quantile Regression

Accurate load prediction is critical for boosting high-quality electricity use, as well as safety in energy and power systems. However, the power system is fraught with uncertainty, and cyber-attacks on electrical loads result in inaccurate estimates. In this study, a probability density prediction method is proposed to provide reliable predictions in the face of false data injection (FDI) attacks. The method effectively integrates data-driven and statistical algorithms such as double-layer long short-term memory (DL-LSTM) networks, quantile regression (QR), and kernel density estimation (KDE). To acquire predicted values under diverse conditional quartiles, the FDI-attacked data of different types were first simulated and then utilized as the training set for the QR-DL-LSTM model. A probability density curve was drawn using the Gaussian kernel function, and interval estimates were used to more thoroughly analyze and assess predictive capability. Power load data from a wind farm in northeast China were used to confirm the availability and effectiveness of the QR-DL-LSTM model. The final results show that the proposed model has a 1.13 and 0.26 reduction in MAPE and MSE compared to the original LSTM. According to our research, the suggested model can successfully describe future power systems full of possible risks and uncertainties with great accuracy.