Deep Learning Based Effective Technique for Smart Grid Contingency Analysis Using RNN with LSTM

Abstract

Smart grids play a crucial role in modernizing power systems, yet their susceptibility to disruptions necessitates effective contingency analysis for reliable operation. Traditional algorithms lack prediction accuracy for mitigation. To address this, a proposed method employs LSTM networks within a DL RNN framework, utilizing a logical collection of pertinent data—historical smart grid operation data, sensor measurements, weather conditions, and contingency events. This data undergoes preprocessing to train an LSTM-based DL RNN model capable of capturing temporal dependencies and complex dynamics within the smart grid system. The trained model yields a substantial improvement in prediction accuracy, increasing from 88% to 93%, while decreasing false positive and false negative rates from 3.2% to 1.25% with optimal time detection. Leveraging deep learning and recurrent neural networks, this approach offers accurate predictions, proactive decision support, and efficient restoration strategies, ultimately elevating the resilience and reliability of smart grid systems. The solution’s dataset encompasses a comprehensive array of relevant information. The neural network architecture integrates DL RNN with LSTM networks, showcasing its efficacy in enhancing prediction accuracy for smart grid contingency analysis. Evaluation metrics such as RMSE and MAPE providing a reliable foundation for proactive decision-making and fortifying the smart grid’s resilience against various contingencies.