Research Of Very-short-term Prediction For Wind Power

As an important factor, Energy affects development of human productivity and energy efficiency is related directly to the development of the entire human-society. Because of the non-renewability of the energy and the free use of wind as well as no pollution during the process, its status is becoming more and more vital. However, the non-controllability of the wind brings a lot of resistance to wind power grid, which means serious challenges to the stability of the grid. Therefore, the accurate prediction of wind power becomes indispensable.In the forecast modeling of wind power, the historical data as the basis for modeling, which needs the assessment and treatment on its integrity and authenticity. Therefore, this paper will be validate and revise the date that collected from SCADA. In this paper, through the reasonable range of reference values of all measured parameters to test data for error data detected by interpolation correction. Through the correlation of wind speed and revision operation data for missing the fan. The historical operation and measurement data of the wind farm are studied and some conclusions are obtained.Artificial neural network has powerful nonlinear fitting ability,and it is easy to operate, which means it plays an essential in the wind power prediction method. However, the performance of neural network has strong dependence on the input variables and training samples, so the research on the optimal input variables has certain engineering practical value. The paper presents a method based on mutual information redundancy analysis, to filter Neural network input variable wind power prediction model that make the correlation between the input variables and objective variables reach maximum, and the redundancy among the variable reach minimum. A cross validation method is introduced to obtain the optimal redundancy analysis parameters. The actual data of the wind field of Hunan Chenzhou verified the analytical method based on mutual information redundancy can effectively remove the redundant input variables and reduce the input variables to improve the prediction performance of the neural network.In addition, because of a nonlinear relationship between the transient degeneration of the wind and instability, and the wind speed and wind power, the wind power error is inevitable, so there is not only uncertainty of wind power forecasting analysis but also great practical significance. In this regard, this paper presents an interval neural network prediction model which predicts the uncertainty of wind power, by introducing interval prediction coverage probability(prediction interval coverage probability, PICP) and the normalized average width(PI normalized average width, PINAW) to construct a fitness function based on fuzzy membership function on the effect of the neural network training function and adaptation assessment, and combined with the theory of molecular motion optimization algorithm to construct the neural network training function. The feasibility of this method is verified by the measured data of a wind field in Chenzhou, Hunan, and the uncertainty of wind power prediction can be analyzed accurately.