Research On Daily Load Forecasting Theory And Method Of Power System

The daily load forecast of electric power system is an important routine for power dispatch departments. Its precision will influence the economic and secure operation of power systems and quality of power supply. The features of daily load forecast can be generalized as followings: many data need to be forecasted, the physical factors which influence forecast are complicated and random, and high precision of forecast is demanded. This thesis covers three aspects: the process method of daily load data, the research of forecast model and the software development of daily load forecast. Since many data need to be forecasted for daily load forecast, the methods of principal component analysis and wavelet transform are applied to process daily data to make the information centralized effectively, Than we can build forecast model more efficiently and improve the forecast precision. The linear sparse number AR model is suggested as a forecast model for short梩erm daily load forecast. The feasibility is discussed theoretically that the linear sparse number AR model is used as a general linear forecast model in stead of linear ARIMA model and the model searching algorithm of the linear sparse number AR model is offered. Therefore a program will enable the process of building the linear forecast model to go on automatically and the complicated and trivial process of building linear ARIMA model is avoided. Also inthis thesis, the problems about the identification, the analysis and building model of nonlinear forecast model are discussed. A statistic principle is offered to identify of nonlinear forecast model of time series. A searching algorithm of optimal embedding dimension and optimal delay step is proposed for building phase space of nonlinear time series with chaotic property. A local forecast model based on optimal embedding phase space is offered. A amended newton fast learning algorithm of forward neural networks is proposed. The evolving learning algorithm of neural networks that have global searching capability is discussed. The general nonlinear ARMA(p, q) forecast model based on the Eleman recurrent neural networks is offered. The climate factors that influence the changing of daily load are analysed. The building model and forecast method of daily load forecast based on climate influence are discussed. A discussion about the building model and forecasting of daily load forecast based on influence of electric price is offered. An approach of building forecast based on influence of climate and price is proposed. The problem of code reuse, which occurs in the software development of daily load forecast under Windows Operating System, is discussed. Two method of Active control and Active component based on COM technique, which are used to solve the problem of code reuse in the software development of short-term daily load forecast, are reconinended.