The Study Of Wind Speed Fusion Forecasting Based On Model Optimization And Combined Strategy

As a kind of clean and renewable energy, wind energy is favored in many countries because of its characteristic of extensive distribution, rich storage capacity and pollution-free. The coming mature of the wind energy technology ensure the rapid growth of wind power generation industry. However, the characteristic of uncontrollable and intermittent will cause a series of problems about power system operation and power dispatch, which limits the futher development of wind power generation industry. Accurate wind speed forecasting can effectively reduce the bad influences casued by wind energy. Thus, it is of great significance to study how to improve the wind speed forecasting accuracy.At present, there are many forecasting models for wind speed forecasting. According to different classification standards, these models can be divided into different categories. On the basis of the structure of forecasting models, the wind speed forecasting models can be divided into individual models and combination models. In combination models, data information from a variety of single forecasting methods are used to improve the forecasting accuracy. How to excavate the information from individual models, namely, how to assign weight for them, is studied widely in combination model. However, how to select individual models is rarely studied, which is of great importance to forecasting accuracy. Therefore, a fusion forecsting method is introduced into wind speed forecasting to improve forecasting accuracy. Firstly, individual forecasting models are optimized to improve their forecasting ability. Then multi-criteria decision making is proposed to select individual models in established model library. How to eatablish the best fusion models is studied through the combination between a variety of optimization methods and combination methods. The major research content is as listed below:(1) The current development situation and existing problems of wind power are summarized in the paper. Morever, the significance of accurate wind speed forecasting is elaborated. Aslo, the research status and the common classification methods of wind speed forecasting models are summarized.(2) Based on the object of study, the paper analyzed the change characteristic, distribution characteristic and shear characteristic of wind speed; analyzed the transformational relation between wind speed and output power of fan; The national standard GB/T18709-2002 is used to the test of wind tower data, and least squares support vector machine is used for data mending. Also, evaluation indexes are discussed here.(3) With regarding to the exiting problems in least squares support vector machine, such as the parameters are difficult to deal with, the least squares support vector machine wind speed forecasting model based on the coupled simulated annealing- Nelder Mead simplex method is proposed in the paper. The simulation results show that the method can improve the forecasting ability of LSSVM wind speed forecasting model. And, the forecasting model is added to individual model library, laying foundation for wind speed fusion forecsting.(4) The selection and combination of individual forecasting models are two key problems in wind speed fusion forecasting. Aiming at the optimal selection problem of single forecasting methods in fusion forecasting, this paper firstly presented a multi-criteria decision method to select individual forecasting models from individual forecasting model library. Meanwhile, co-integration analysis method also is introduced to select individual forecasting models. In order to further improve the fusion accuracy, the optimal selection methods(multi criterion decision, cointegration analysis method) and the combination methods(reciprocal method, simple weighted average method, induced ordered weighted operator) are combined to explore the best fusion forecasting model. The simulation results show that when multi-criteria decision-induced ordered weighted operator, cointegration analysis-reciprocal method, cointegration analysis-simple weighted average are combined, these fusion models show excellent forecasting performance.And, different methods of optimal selection methods are suit to different combination methods, and the best fusion models require the combination between the appropriate optimal selection method and the combination method.