LAUSR

Abstract The work of recognizing load patterns for power big data accurately and efficiently is an indispensable basic task for supporting safe, reliable, and economical operation of the power grid. At present, it meets so many difficulties to handle high dimensional eigenvalues of tremendous amount of original collected data. To fulfill the demand of efficient and accurate classification of load pattern recognition, this paper reaches out a solution which based on dimensionality reduction of characteristic index and improved entropy weight method for power load pattern recognition. Firstly, six characteristic indexes such as daily load rate, maximum utilization hour rate, daily peak-to-valley difference rate, peak period load rate, flat load rate and valley load rate are extracted and taken as input to replace the original load data. Secondly, the improved entropy weight method is introduced to configure the weight coefficient of each characteristic index adaptively. Then, the elbow method is used to determine the optimal number of clusters, and the clustering method of weighted Euclidean distance K-means is used to get classification labels for sample data. Finally, the K-nearest neighbor (KNN) algorithm is used to identify the labels and the six characteristic indexes. The results of the example show that the algorithm based on dimensionality reduction of characteristic index and improved entropy weight method is an effective algorithm for power load pattern recognition, and has certain advantages in operating efficiency and accuracy.