Convolutional neural network and long-short term memory based for identification and classification of power system events

In this present era, power system delivery has to be reliable and sustainable. The growth of demands increasing the complexity of the power system operations. An interrupted power supply must not occur for any reason. Hence, the improvement of the controller and protection devices is mandatory. One of the unnecessary interruptions in the power system is a false trip due to the incorrect setting of the protection devices. Therefore, a method to classify the symptom of the power system based on the voltage, current, and frequency measurements is required. However, since there are a ton of maneuver options and fault types, the number of data becomes complex, enormous, and irregular. This is where deep learning takes place. This paper proposed the use of Convolutional Neural Networks (CNN) combined with Long-Short Term Memory (LSTM) to recognize the categorize the type of events in a medium voltage power distribution network. As CNN’s models are great at decreasing frequency variation, LSTM is great for temporal modeling, we take benefit of CNN’s and LSTM’s complementarity in this study by integrating it into a unified architecture. The simulation results indicate that CNN and LSTM can recognize the symptoms in power system operation with accuracy up to 79 % with a total epoch 350.