Research Of The Combined Short-term Load Forecasting Model In Power System

The short-term load forecasting of electric power system is to predict electric load for a period of hours, days, or weeks, and especially twenty-four hours, which is the primary gist for making the plan of power generation and the scheme of power transmission. It is important for economic arrangement of generating capacity, scheduling of fuel purchases and planning for energy transaction. Its precision will greatly influence the economy and secure operation of power system. Furthermore, with the establishment of power market, load forecasting will play a more important role in the future.The principle, current status and development of the electric power system short-term load forecasting are generalized in this thesis. Varieties of traditional and modern prediction techniques for load forecasting are summarized, and the differences and features of these methods are also emphasized. Based on the research of back propagation neural network (BPNN), autoregressive integrated moving average(ARIMA), and Ant colony clustering (ACOC) algorithm, the analogous theories with ARIMA and BPNN, ACOC and BPNN models in load forecasting are proposed. The ARIMA-BPNN model forecasts the load of the predicted day at first by use of the strong approaching capacity to linear time series of ARIMA, then the load forecasted by ARIMA is modified by BPNN; The ACOC-BPNN model differentiates the load series having the different characteristics using ACOC at first, then the load seires forecasted by BPNN using the load seires in the same cluster. The corresponding program of this load forecasting model is developed and applied to the power systems of the province of North China. The results show they are credible and practicable in short-term load forecasting. Compare with the forecasting result of ANN, these models can get a more precise result. At the end of the paper, the main problems in the research of short-term price forecasting are analyzed and the futher work is pointed out as well.