Application Of Recognition Model For Power Load In Cluster Ensemble And Deep Learning

In order to help power company have a understanding of power consumer’s behavior,Studying the characteristics of the power consumer’s load mode is necessary. Better customersegmentation creates opportunities to provide personalized service and better service.To improve the reliability and universality of consumer’s load mode recognition in peakshifting management, this paper presents a new consumer’s load mode recognition methodbased on Bootstrap cluster ensemble. First, we ocuses on selecting suitable clusteringalgorithm(CA) for load curve data from30different kinds of CAs. And the results ofeffectiveness evaluation proves that selected are useful. Then, Bootstrap method is used togenerated cluster members with diversification. I also define the distance of each sample inconsensus function. In order to get the final clustering results, CSPA method is use to mergeconsensus function. The results is typical load curves into four groups. There are Bimodal,Trimodal, Stationary and Averting. And the overall accuracy of results of cluster ensemble are74.61%, which proves that cluster ensemble is better than single CA in consumer’s loadmode.Establishing Deep Learning model in consumer’s load mode recognition with usingresults of cluster ensemble as input. And model parameters are given by Cross-Validation. Theresults of model shows that the accuracy of results in prediction set is76.91%which approvesits advantages in training layer by using self-learning method. After Comparing with BP nets,SVM and Random Forest, the outcome proves the good performance of Deeping Learning inconsumer’s load mode.