A novel deep learning approach for estimating and classifying short-term voltage stability events in modern power systems with composite load and distributed energy resources

Short-term voltage stability in modern power systems is greatly impacted by the existence of composite loads with dispersed distributed energy resources (DERs) in the low- and medium-voltage power networks, as well as by their interactions. With composite loads and dispersed distributer energy resources (DERs) in low- and medium-voltage power networks, this study presents a novel way to estimate voltage breakdown in current power systems. In order to avoid voltage collapse, it also attempts to categorize the severity of such incidents well in advance. Combining long short-term memory (LSTM) and Convolutional Neural Network (CNN) based on deep learning, one can forecast and categorize the severity of short-term voltage stability (STVS) incidents. The CNN-LSTM model uses the post-failure voltage time series data as input. Using time domain modeling, these voltage time series data were produced for a range of composite load and distributed energy resource combinations during sporadic three-phase fault scenarios in the IEEE Bus14. The random permutations of these generated data are used during training for the introduced model in order to maximize the diversity of the three-phase short circuit fault, DERs, and composite load. The Kerala Grid at 220 kV is used to assess the created model. The outcomes produced for the actual system, 220 kV Kerala Grid, confirm the applicability of the proposed method for real-time STVS analytics and show a greater accuracy of 97.93% together with superior efficiency. Additionally, this aids in identifying and averting the likely voltage breakdown in contemporary power networks.