Research On Short-term Power Interval Prediction Method Of Wind Farm Based On B-ELM Model

Wind energy is one of the more important renewable energy sources in the current energy structure.It has been widely used all over the world.New energy power generation has been rapidly developed on a global scale,and wind power generation is particularly important.Although large-scale wind power has alleviated China's energy pressure to a certain extent,it has brought huge economic and environmental benefits.However,due to the randomness and intermittent nature of wind,wind energy has uncertainty,and this uncertainty gives electricity.The scheduling and regulation of the system have brought enormous challenges.While increasing the difficulty of safe and stable operation of the power system,it has also brought huge difficulties to wind power consumption.Therefore,it is especially important to predict the output power of wind farms.At present,most of the predictions of output power are point predictions,which have greater uncertainty and are easy to introduce risks into decision-making.First,this thesis analyzes the sources of wind power uncertainty factors,and states that the existence of prediction errors can be divided into: lateral error and longitudinal error,and introduces the error evaluation index and the distribution characteristics and time characteristics of prediction error,and several main sources of wind power prediction errors are deeply studied,including:accuracy of input data;dispersion of fan output power;uncertainty of wind turbine faults in wind farms;prediction model errors.Aiming at the two main uncertainties of model structure and parameter setting and data noise,the wind power prediction model is proposed.A wind power interval prediction model B-ELM based on extreme learning machine and Bootstrap sampling method is proposed.Multiple training samples are obtained by using Bootstrap sampling method.The extreme learning machine network obtains the systematic error variance and the data noise error variance.By calculating the upper and lower limits of the interval between the wind power prediction value and the prediction error value,get a prediction interval that satisfies the confidence level prediction interval.According to the criteria of interval coverage PICP,prediction interval average bandwidth NMPIW,and prediction interval comprehensive index CWC,the B-ELM method is compared with several common neural network-based interval prediction methods,including Delta method and Bayesian Method,lower upper bound estimation(LUBE)method.Taking the historical operation data of a wind farm in Xinjiang as an example,the wind farm example simulation is carried out.The simulation results show that the power uncertainty can be better predicted in the case of severe power fluctuations compared with the other three methods,the B-ELM method is suitable for engineering applications.It can be an important basis for the planning and operation decision-making of electricity,and greatly improve the anti-risk ability of the power system.The interval prediction performance and computational efficiency of the proposed method are verified.