MEAN FIELD ANNEALING BASED COMMITTEE MACHINES FOR OUTAGE ESTIMATION IN POWER DISTRIBUTION S Y STEMS

摘要

Electric reliability is an important concern for the utilityrncompanies. Weather related outages have a significantrnimpact on it. There are many regression based models tornestimate outages from weather factors in overheadrndistribution system. This paper proposes the use ofrncommittee machines composed of multiple neuralrnnetworks to estimate outages. A major challenge for usingrna committee machine is to properly combine predictionsrnfrom multiple networks, since the performance ofrnindividual networks is input dependent due to mappingrnmisrepresentation. This paper presents a new method inrnwhich the individual network predictions are combinedrndynamically. The error minimization is performed usingrnthe mean field annealing theory. Results obtained for thernstudy area in Kansas are compared with observed outagesrnto evaluate the performance of the model for estimatingrnthese outages. The results are also compared withrnpreviously studied regression and neural network modelsrnto determine an appropriate model to represent effects ofrnwind and lightning on outages.