Integrating Data Intrusion Defense Strategy in PV Power Forecasting Framework

Smart grid networks are gradually participating in highly efficient and advanced intelligent cyber-physical systems with high renewable energy penetration and information technologies. The reliable operation of the grid depends on forecasting methods widely used in the industrial, commercial, and residential sectors. The forecast's accuracy depends upon the data's quality, which is vulnerable to cyberattacks. False data attacks in training data pose a significant threat to forecasting accuracy. This study introduces a data-driven, explainable artificial intelligence (XAI) data poisoning technique that effectively compromises the forecasting model, even in the presence of outlier identification. This work investigated the impact of poisoning attacks on short-term solar forecasting with outlier identification. The XAI and t-SNE models are used for feature extraction and reduction, respectively. The various forecasting models are utilised, and performance results show that CNN-LSTM-Attentation is the most effective for forecasting compared to others. The obtained performance matrices MAE, MSE, RMSE, R2 score with the proposed model are 0.089, 0.026, 0.161, and 99.69%, respectively.