Research On STLF Of Electrical Power System Based On NFN And Wavelet Theory

At present time, the electric load forecasting is of a great importance for the development of the power industry. Electric load forecasts represent the main source information for decision making in the planning of optimum performance of the EPS(Electrical Power Systems). The accuracy of energy consumption forecast is the main criterion of quality performance of the whole system. Accuracy increase contributes to energy saving and leads to profits increase of the energy enterprises.Complex unsteady behavior of the time series, describing consumption is lowering the precision of the forecast. A large number of factors, both systematic and random nature, are affecting the level of consumption, the continuous change of the energy market, as well as the development of EPC(Electrical Power Consumption) themselves make the task of improving the accuracy of forecasting more difficult.The aim of dissertation is to develop a system of electric load forecasting on the basis of intelligent computing technology, which will enable to solve the problem of improving the accuracy of short-term forecasts effectively.Aim is achieved by solving the following tasks:(1) Analysis of current approaches to forecasting consumption of electric load and identifying the most perspective directions for improving the quality and reliability of forecasts;(2) Development of a technique of building a system of forecasting electrical power consumption;(3) Building a system for electricity consumption forecasting on the basis of previous research;(4) Experimental investigation of the developed system Matlab program;(5) Comparative analysis on the quality and reliability of the system built.To achieve the stated objectives the following methods were applied: methods of mathematical modeling and forecasting time series, regression analysis, fuzzy set theory, foundations of wavelet theory, the theory of artificial neural and hybrid networks, the methods of evolutionary modeling, the mathematical package Mat Lab.Originality of scientific results:(1) The presentation model of retrospective load data is distinguished by the identification and description of the additive components of time series, that have different properties;(2) The procedure of short-term load forecasting is distinct in the preliminary step of separating time series into components with different dynamics, in performing the prediction for each component separately, and also in the combined usage of intelligent computational technologies for the prediction;(3) A method of constructing the electrical load forecasting system is characterized by using wavelet filtering and neuro-fuzzy network committee that reflect the properties of the time series in high frequency, low frequency and mid-frequency regions.The forecast results of the developed system of electrical load forecasting based on wavelet theory and neuro-fuzzy approach is superior to previously used classical methods. Thus, it is possible to assume that objectives of this work were successfully accomplished and the goal achieved.