Machine Learning based identification of topology and event classification in distribution system

The integration of renewable energy sources for decentralized power generation in the existing power system network has resulted in continuous changing topology of the network. The existing adaptive protection schemes to protect such a dynamic network have certain limitations which need to be addressed in time to make the power system protection more reliant and accurate. Supervised machine learning can be used to train the relay to estimate the topology, classify the fault, estimate the status of the circuit breaker and isolate the faulty section from the healthy section. All these functions can be performed even in the case of limited communication infrastructure. Scientists and utility engineers are looking for adopting machine learning techniques to protect a distribution system network in case of communication failure. This paper deals with the application of machine learning algorithms to identify distinct topologies resulting from the connection and disconnection of renewable energy sources. The second objective is to train the algorithm to find the status of the network. A comparison between SVM and CNN is carried out and their efficiencies are compared. The effectiveness of this approach is tested on a modified IEEE 15 bus system using MATLAB.