Combination of load curves and tangent curve for customer’s load curve clustering

Abstract

The load curve clustering for electrical customers traditionally is based on the 24-dimension input space. It means that every load curve is considered as an element with 24 attributes corresponding to 24 load values per 0day. But in some cases, the load curve itself can not lead to the right cluster when the two curves have different forms but have the same distance to the third one. To overcome this limitation, the present paper pays attention to the selection of the input space. From each load curve, the tangent curve will be received. Now the clustering will be based not only on the load curve but on the tangent curve. The clustering techniques used the Pulsar algorithm with some modifications to fit the input space. Examining for the electrical load curve of Ho Chi Minh city constumers, the propose approach has achieved the best results.