Study And Apply On Wind Speed Forecasting

With the intensification of the global energy crisis, a consensus has been reached in the world on developing renewable energy. Currently, wind power, one of renewable energy, with mature technology, has been used most widely and the most promising in the large-scale commercial production. As wind characterized by strong random, intermittent and uncontrollable, consequently the output of wind turbines becomes highly volatile as well. However, this is not conducive to the safe and stable operation of power system. In order to reduce the impact of wind power on the grid, operating costs and spinning reserve, additionally to increase penetration limit of wind power, it is necessary to predict wind speed and power accurately.Artificial neural network with self-learning, self-organizing and adaptive capacity in dealing with strong non-linear problems, is widely used in the field of wind speed forecasting. But it requires samples with a high similarity, as the effect of wind speed prediction is bound to affect by wind characteristics. Therefore, in this paper features of wind speed and changes and the relationship between wind and meteorological factors are studied deeply. According to a certain periodic and continuous changes of wind speed in the characteristics, samples are selected with weighted pattern recognition to improve the similarity of the samples, thereby the wind speed prediction accuracy.Since traditional neural networks have shortcomings such as low convergence speed and large amount of dealing data, Radical Basis Function (RBF) neural network model with fast learning and convergence speed and universal approximation is adopted as a predictor in this paper.A new model is proposed to forecast the wind speed in wind farm with dual-RBF neural network, particularly taking the meteorological factors speed into consideration. Historical similar days choosing in periodic and continuous aspects respectively to predict the wind speed of specified day with RBF neural network, and the results are then used to give the terminal value with RBF neural network. In addition, a method of pattern recognition with weighted Euclidean distance are used to sieve proper samples to improve the accuracy of the RBF neural network.Finally, the above method are simulated with Matlab software. The wind speed forecasting results based on data of a wind farm in Neimenggu show that the model holds more high accuracy and smooth. The model in this paper can be feasible and effective in practical applications.