Power System State Estimation and Forecasting using CNN based Hybrid Deep Learning Models

Abstract

Power System State Estimation (PSSE) uses statistical criteria, which is essential for different power system studies. As the power system involves non-linearity and complexity, the conventional state estimation techniques become computationally inefficient and give a suboptimal performance. To overcome these difficulties, this paper utilizes hybrid data-driven Deep Learning (DL) techniques for state estimation. Also, to make the system independent on measurement data, the DL models are developed by driving them with historical information of system states, thereby it enables state forecasting. To deal with the complexity of data, the models Convolutional Neural Networks (CNNs) and Multi-Layer Perceptron (MLP) network are hybridized for estimating the states from measurement data. Similarly, the hybrid CNN and Long Short-Term Memory (LSTM) model is developed for state forecasting. The performance of the hybrid models is compared with individual models such as CNN, MLP for state estimation and CNN, RNN, LSTM for state forecasting. In this work, the IEEE 118 transmission system measurement data is considered for comparing the hybrid models with individual deep learning models. Numerical results indicate that the hybrid Convolutional MLP (CMLP) model performs better for mapping measurement data with states and in the case of state forecasting, the hybrid Convolutional LSTM (CLSTM) model provides better performance compared to individual models.