Deep learning-based cyber resilient dynamic line rating forecasting

Increased integration of renewable energy resources into the grid may create new difficulties for ensuring a sustainable power grid which drives electric utilities to use a number of cost-effective techniques such as Dynamic line rating (DLR) that enable them to run power networks more efficiently and reliably. DLR forecasting is a technique devised to accurately forecast the maximum current carrying capacity of overhead transmission lines. DLR offers many advantages, including increased renewable energy penetration without system reinforcement, improved grid dependability, and lower congestion costs. So far, many solutions have been proposed for DLR forecasting, which, despite estimating the exact capacity of the DLR, have some problems, such as installing multiple sensors and measurement devices and communication networks with precise calibration, and also neglect cyberattacks which may lead to operators making inappropriate operational choices. To address these issues, in this paper, a novel hybrid deep learning-based DLR forecasting approach called the autoencoder bidirectional long short-term memory (AE-BiLSTM) is efficiently and precisely developed. Several scenarios were developed to test the robustness and accuracy of the proposed methodology using real-world data with and without cyber-attacks. Detection of cyber-attack is done based on the increase in the least square errors of forecasting models. Then, the carefully designed hybrid AE-BiLSTM method reconstructs the falsified measurement data and provides reliable DLR forecasting. Also, a comparative study is carried out. The numerical results demonstrated that the proposed hybrid approach can significantly provide acceptable performance even under cyber-attacks and forecast DLR values ​​with the least possible error, outperforming the existing conventional and deep learning-based techniques.