Approach on random weighted deep neural learning model for electricity customer classification

Abstract

With the increase of electrical sensors and smart meters installed in distribution networks, the consumption load data of local electricity customers gradually shows its following properties: large scale, big variety, fast generation and low value density. These properties have brought new challenges to load pattern analysis and pattern classification, since traditional methods and technologies cannot be able to meet the current performance requirements of pattern classification on both of the classification accuracy and time consuming. In this paper, facing electricity consumption load data, a novel electricity customer classification method is proposed based on random weighted deep neural learning. In this proposed method, the effective feature information of electricity consumption load data is extracted by training multi-layer auto-associative random weight neural networks with a small size central layer. Then, by combining the well-trained feature information and basic load feature indexes, single-layer neural network is employed to fulfill the electricity customer classification tasks of test samples. Comparative experimental results verified the outstanding performances of our proposed method.