Short-term Forecasing Of ElecticityPower System Based On FNN

The level of load forecasting is one of the measures of modernization of power system management. So load forecasting, especially accurate short-term load forecasting is of great importance to power system. There are many factors that afect system load, such as history data of load, many non-load factors.After analyzed the meaning and method of power system load forecasting, based on the analysis of fuzzy inference and neural network biology characteristic, the paper presents a new method of constructing Fuzzy Neural Networks (FNN), which merges the fuzzy inference into BP network. In the paper, two kinds rule-extracted method of fuzzy inference is presented: one is based on combinatorial pattern; the other is based on tree formation in which the definition of fuzzy rule tree is presented. The problem of dimension explosion accompanied by combinatorial pattern can be avoided by the tree formation rule-extracted in some degree. At the same time, the well integration of fuzzy inference and trams-calculating is obtained through the controlling of information feed-forward controller in the FNN model. The detailed algorithms of controller, two kinds rule-extracted and FNN model training are presented. Moreover some proofs and corresponding mathematical models about FNN model are presented.In the last, the FNN is applied to forecast the short-term load of power system. Based on the characteristics of load and the basis of load forecasting, the pretreatment of load data is given firstly, the load data of 4 hours are selected and generated 2 rates of change; the rates of similar and trend change are used as the input of FNN model, the neuront is used as controller in the input layer;the 24 neurons are set in the hidden layer to undertake every hour' s fuzzy inference and trans-calculating respectively. Finally the FN'N model is trained and simulated by using actual load, the object of short-term load forecasting is achieved. Seen the forecasting result, the FNN model in the paper not only expands the function of BP network and accelerates the speed of training, but also conforms to demand of actual application.