Adaptive Extended Kalman Filtering for Predictive Modelling and Attack Detection in Electrical Systems

In modern-day power systems, the data-driven models play a very impactful role in forecasting future energy demand by analyzing the previous consumption trends. To ensure the reliability of these forecasts, the input data must undergo thorough training, validation, and testing, and it must be free from any irregularities, abnormalities, and noise. However, in the real-world scenarios, the performance of these forecasting models is greatly influenced by the malicious intrusions or any unpredictable conditions. Among these, adversarial attacks have emerged as a serious concern due to their ability to harm the data integrity and reduce the forecasting accuracy. This study explores the forecasting of actual load data and analyses the prediction outcome when the actual load data are subjected to attack vectors. For this purpose, the Adaptive Extended Kalman Filter (AEKF) serves as the basic forecasting technique. The evaluation of the model performance is carried out by comparing the forecasted values with the attacked data using two standard error metrics: Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE). The proposed method shows system good resilience even if subjected to abnormalities while resulting in a MAPE value of 2.44% and a MAE value of 28.5337 MW demonstrating the high reliability of the proposed method in upholding a good accuracy. Furthermore, it ensures the detection of abnormalities even in uncertain environments.