Stability Prediction Model for Smart Grid Network That Handles Missing Input Variables

This paper focuses on developing a smart grid stability prediction model to handle the missing input variables. This paper implements two models. The first model is a feedforward neural network designed to predict stability with complete input data. The second model includes the novel work carried out, wherein two missing input variables are considered, which are predicted using a sub-neural network. This sub-neural network's obtained testing output variables are taken to run the primary neural network. The Levenberg-Marquardt algorithm is used to train both the models. The transfer functions used are tansig for the hidden layer and purelin for the output layer. The performance of the proposed models is measured in terms of mean square error and $\mathrm{R}^{2}$ values. The performance of the final leading stability prediction model that handles missing input variables depicted the system's superior performance in both the training and testing phases.