Multi-dimensional ANN Application for Active Power Flow State Classification on a Utility System

A multi-dimensional ANN has been applied to identify the states of a power network in real-time. Large power systems have multiple transmission lines and crucial parameters on each line to be monitored in real-time which is very complicated, considering the number of parameters to be monitored on each line. After any contingency the power system could transit from normal state to the emergency state within a matter of seconds. These transitions of states could be unprecedented or unexpected at times, and in such a scenario, deciding which lines should be highly ranked in the new steady state, for operator decisions, can be very complicated at times. In this paper, extensive load flow studies have been studied, considering N-1 contingencies. Then for each scenario the power flow on the lines remaining in service have been recorded. Using these recorded data, the power system operation has been classified in the 5 different states, namely, normal, alert, emergency, extreme emergency and restorative states. These data are then used to train a multi-dimensional feed-forward neural network to identify the states in real-time in Software-InLoop (SIL). This paper proposes a multi-dimensional neural network with as many numbers of input nodes as the number of lines and 5 output nodes for the 5 state classifications. This type of ANN could help the Power System Engineer (PSE) to take crucial decisions in the event of severe contingencies in the power network.