Study On Load Profile Analysis And Short-Term Load Forecasting Method Considering Demand Response

Demand response is the development of demand side management in competitive electricity market and an important part in smart grid, which draws more and more attentions. With wider and wider practice and application of demand response in the world and with massive amouts of information in smart grid, demand response and massive amouts of information bring up new demands for load characteristics and short-term load forecasting, which are the premise and foundation of analysis on electricity market and load management, and points out new research focus.Considering the new demands for load characteristics and short-term load forecasting in smart grid, this paper researches method of load analysis, modeling demand response under time-of-use pricing and short-term load forecasting on the basis of summarizing and analyzing demand response systematically.As means of dealing with large amouts of load information, some menthods of load analysis are rough and the cluster number of customers’ load curves is set by experience at present, this paper reseaches clustering algorithm and cluster validity indices to extract typical load cuves of industries. This paper uses fuzzy C-mean algorithm to cluster load curves, and uses validity indices to determine the cluster number and evaluate clustering results.Some kinds of customers’ load chracteristics are greatly influenced by peak and valley time-of-use prices, and it’s necessary to study the models of demand response. Considering differences among customers ’responses during peak (flat or valley) load period, a load shifting adjustment coefficient is proposed in this paper to improve the precision of the models. This paper also sets up mathematical relationship between the price-elasticity matrix of demand and the psychology model. Based on the relationship, the steps are presented to establish the price-elasticity matrix of demand based on customer psychology model, which uses relatively less data and avoids the negative impact of data in insensitive response zone on the matrix, and the matrix can be revised in real time by the real-time data. This method also shows how to forecast load curves under different TOU pricings using the price-elasticity matrix derived.When investigating method of short-term load forecasting, this paper adopts algorithm idea of clustering load cuves to select similar days from historical days, and forecasts subsequences after load wavelet decomposition by ARIMA and BP nepal network. To improve forecasting accuracy, this paper proposes the idea of multilevel coordinating of short-term load forecasting. This method determines reliability weighs of forecasting results of total load sequence and sub-sequences by historical predictions, then coordinates errors among total load forecast results and sub-sequence forecast results by state estimation ideas.