Applications Of Fuzzy Clustering & Chaotic Prediction In Short-term Electric Power Load Forecasting

Electric power load forecasting, especially short-term forecasting is an important routine of the electric power utility and of great economic significance. Therefore, to pursue higher forecasting precision becomes great research task for the electric power utility. In the past decade, researchers have advanced many forecasting methods which improve the forecasting precision, but it is still hard to meet the increasing demands of both utility and customers. More excellent load forecasting methods should be developed.In this dissertation, fuzzy clustering and pattern recognition are used to the preprocessing of historical load and weather data to provide more effective data sets for forecasting. Then new load forecasting methods are developed based on the chaotic phase space reconstruction, in which climate factors are considered. The primary work in the dissertation is as follows:In the phase of load and weather data preprocessing: Firstly, by analyzing the correlation between the daily peak load data and the corresponding 9 kinds of weather data of certain northern area in china, it is confirmed that in summer, "maximum temperature, minimum temperature, mean temperature, precipitation, mean dew point and mean sea level air pressure" are the factors influencing the load most, while in winter, "precipitation, mean sea level air pressure, mean visibility, mean wind speed and maximum constant wind speed" are the main factors. Secondly, the specific data of year 2000 are used as demonstration to the fuzzy clustering for the load and weather data. On the choosing of fuzzy membership degree function, correlative coefficient function and Euclidean distance function are selected first. Then a new membership degree function "correlative coefficient + Euclidean distance" is constructed. Compared with the previous two on the basis of original data, it is approved that the result of the constructed method is more logical.In the phase of load forecasting: multiple second order polynomial regression combining with the local approach of chaotic phase space reconstruction is used to forecast the future loads. Compared with the linear regression, it achieves better forecasting results and sharply restrains the 12h periodical error augmentation phenomenon. Besides, the "amount" of the neighboring vectors instead of Euclidean distance ε between them is used to gain the vectors near the reference vector in the embedded phase space, and this guarantees that in any case adequate neighboring vectors can be obtained as well as enhance the forecasting precision. Another improvement is directly forecasting multi-steps other than the existing one step to eliminate the accumulative errors. Further, weather records of corresponding load records and weather forecast of the day to be forecasted are added to the neighboring vectors which form the new type vectors "load records, weather records and weather forecasts" when considering the weather factors.