Random Forest based Fault Detection and Localization in Microgrid using Simplified Measurements

Electrical microgrids are quite vulnerable to fault conditions because of being in close proximity to the load centres. Due to differences in operating dynamics of microgrids as compared to conventional distribution system, designing reliable protection is a major concern for protection engineers. Therefore, detection and localization of fault in microgrids has become difficult and less trustworthy using traditional overcurrent relay based protection schemes. This paper presents a new protection scheme where faulted line number and its exact location from substation is determined using Random forest algorithm. A microgrid model has been created in MATLAB/SIMULINK platform having various distributed generation sources along with the substation grid. Common types of shunt faults have been created at different locations and the root mean square value of post fault one cycle faulted voltage and current signal is collected from each of the source buses. Further various practical cases such as islanded mode, change in DG(distributed generation) penetration level, increment and decrement of loads by 50% has also been taken into consideration while creating the training data. The collected data is then used for designing a machine learning (ML) model. Three types of prediction are being made by this model i.e. (i) type of fault, (ii) line in which fault has occurred and (iii) distance of fault from substation. Two different approaches are used for model, first simply training the model on all the data set that has been collected and the second one is by filtering out the data according to type of fault and then on the filtered data the faulted data set training is done. Random forest method which is implemented using Python coding stands out to be the best algorithm for all the three problem in terms of accuracy and time of execution of model even with simplified measurement.