Short Term Load Forecasting In Power Systems Based On Time Series And Artificial Neural Network

Electric power system load forecasting is one of important operations in dispatching, demanding, scheduling and planning of power system management sectors. Accurate load forecasts will lead to economical and reasonable scheduling of a unit commitment, maintaining security and stability of power network operations, decreasing unnecessary spinning reserves, scheduling reasonable maintenance plan, reducing generating costs efficiently and improving efficiency of economic and society.With the development of electric power industry, more and more attention is paid to load forecasting, especially the short-term load forecasting (STLF).However, the load complexity and randomicity caused by plenty of influence(affection) make it very difficult to be solved well. There are many methods for STLF in the word nowadays, such as, time series technique, regression method, expert system and artificial neural network(ANN) etc.. They have their advantages and disadvantages. But there isn't a method can obtain the satisfied forecasting results on any occasion, so people have been researching it to improve forecasting accuracy.At first the significance and principle of STLF is elaborated. After discussing the regularity generally existing on short-term load, this paper introduces several kinds existing methods for short-term load forecasting,and points out their advantages and disadvantages as well.Then this paper studies the classical stochastic-time-sequence analytical method and artificial neural network (ANN) technique, which is the most potential way of intelligence forecasting methods. In this paper ,the basic ideas and detailed steps are concluded for Box-Jenkins' linear ARIMA prediction model, the sticking points and difficulties of Back-Propagation(BP) artificial neural networks applied in STLF are discussed thoroughly, such as slow training speed and possibility of leading to a local minimum of optimized function, etc.In order to solve these problems, improved BP algorithm type is presented in this paper. The selection and optimization methods of the network training samples are proposed, adaptively updating strategy for the network parameters and the period of...