In the UK supply market, customers can purchase electricity from any supplier regardless of size and location. Accordingly there is special interest in understanding the nature of variations in load shape, to devise better competitive tariff structures and facilitate aggressive niche marketing. Utilities have databases of half-hourly loads too large to be interpreted by hand and eye; potentially valuable information is hidden therein which is not revealed by coarse statistics. The heterogeneity of response, the large number of predictors, and the sheer size of these databases impose severe theoretical and computational difficulties on load shape modeling. Data mining refers (in part) to the use of adaptive nonparametric models (which vary their strategy according to the local nature of the data) for efficiently discovering knowledge in just such databases. A method centering on adaptive decision tree clustering of load profiles is presented, and results utilising an actual database are discussed.
展开▼
