A new Machine Learning based Approach for Power System State Forecasting

State estimation in power systems is utilized to provide an ongoing data set for controlling and observing power systems' performance. There are a few traditional methods like Weighted Least Square (WLS), Weighted Least Average Value (WLAV), and settled utilizing iterative techniques, for example, Gauss-Newton methods. However, these techniques are sensitive to the operating conditions of power systems and uncertainties associated with Renewable Energy Systems (RESs). These techniques yield poor results in the presence of missing data arising out of unreliable communication networks and false data injection, as in the case of a cyber-attack. This paper proposes Bi-directional LSTM (BiLSTM) based Machine Learning method to forecast the state values of power systems. The BiLSTM model utilizes forward and backward layers, having bidirectional memory, making it capable of estimating both past and future hidden layers data. This paper investigates the performance of the proposed method studied on different test system datasets. The efficacy of the proposed method is investigated by comparing the obtained results with other Machine Learning techniques results in the literature. Further, the BiLSTM model's robustness for state estimation is analyzed in the presence of Gaussian noise and missing data points.