Research On Forecasting Method For Wind Power

Along with the development of wind power technology. wind power forecasting gradually gets the extensive attention from people. Wind power generation is influenced by weather conditions and the output power is uncertain. But it is required to keep a balance between generation and demand in a power system. An accurate wind power forecasting will decrease the reserve capacity, reduce the cost of the system and provide reference to the operation and dispatch of the power grid. With the deepening of the research for the methods of wind power forecasting, numerical weather forecasting is also gradually becomes the mainstream of the wind power forecasting system.In this context, some theories are researched with the time series of wind power generating capacity from Yilan. such as Chaos characteristic, phase space reconstruction. artificial neural network (ANN) method and time series analysis method. Some forecasting models, for example Back-propagation (BP). General Regression Neural Network (GRNN). ARMA. ARMA model on noise occasions, are structured. And the way how to use NWP data in the forecasting is discussed through GRNN neural network. Then, an integrated wind power forecasting System is established by LabVIEW and SQL language.After the study on the wake effect and seasonal factors, wind wake effect and the seasonal characteristic are verified by the historical data of Yilan wind farm.Research is done on the inherent characteristic of wind power time series. The largest Lyapunov exponent is bigger than zero, which proves that the wind power generation system has chaos characteristics. The accuracy from different training data which are colleted from different seasons, also proves that the wind power generation system has seasonal characteristic.Different forecasting models are structured on the basis of the research of BP neural network based on phase space reconstruction and GRNN method. Seasonal characteristic of the wind power generation system is tested and verified by the different forecast accuracy from the models, which are structured from training data.Three different methods of how to determine the order of ARMA are taken into account and different ARMA models are built. The measurement noise is considered and an ARMA model on noise occasions based on HOYW method is built.Considering the instability characteristic, the wind power time series is divided into determine item and random item two parts, and ARMA and BP neural network method integrated forecasting model is structured. Uncertainty factors from wind farms usually lower the accuracy of NWP. so when the NWP data is used in wind power forecasting, analysis of the relationship between NWP data and Actual meteorological data is very necessary. Here, artificial neural network method is well deserved.According to the wind farm wind power forecasting standard from National Energy Bureau, Synthesized so multiple forecasting model modeling methods, LabVIEW and SQL language are combined to establish integrated wind power forecasting System which includes identification function, data acquisition function, analysis and forecasting model function, data processing function and son on. And this forecasting System has been successfully used in large grid wind farms.