Study On Ultra-short-term Probability Interval Prediction Of Wind Power Based On Spectral Analysis

With the decrease of fossil fuels and the increasingly serious environmental pollution,wind energy,as a renewable energy with the value of large-scale development and commercial utilization,has been widely used in the field of power generation,but its volatility,intermittence and randomness have brought great difficulties to large-scale wind power Grid-connected operation.Wind power prediction is conducive to the formulation of scheduling arrangements,providing reference for the operation and maintenance of wind turbines,and is an important way to solve the hidden dangers of wind power grid safe and stable operation.Wind power has its own fluctuation characteristics,and the scale of wind power Grid-connected will have a negative impact on power quality.With the development of wind power,power system will inevitably face the challenge of wind power fluctuation characteristics.In this paper,the distribution characteristics of wind power in frequency domain are analyzed based on the measured data of wind power.The convergence effect of wind power is verified in frequency domain.The fluctuation of wind power can be suppressed by the installed capacity of wind power.The measured data of wind power is the basis of wind power forecasting.Through the data acquisition and monitoring control system of wind farm,the related data of actual operation of wind farm can be obtained.Because of equipment failure,human factors,natural factors and other reasons,there are abnormal data points.Starting with the missing data of the measured data,this paper introduces the evaluation index of the missing data completion results by using various data completion methods,and predicts the data after completion by using the same prediction method.The relationship between the effect of data completion and prediction accuracy is inconsistent.The essence of wind power forecasting is to accurately grasp the changing law of wind power.A real-time combined forecasting method of wind power based on singular spectrum analysis and dynamic weighting correction is proposed in this paper.Singular spectrum analysis is used to decompose the time series of wind power and filter out the noise sequence.The combination model of auto-regressive moving average,persistence method and weighted coefficient of least squares support vector machine is used to predict the noise sequence.This method can effectively improve the prediction accuracy and show good universality.Based on the error analysis of real-time wind power prediction results,a power interval prediction method based on non-parametric and semi-parametric kernel density estimation is proposed in this paper.Statistical probability density distribution of original wind power data,dividing wind power into high and low output levels,combining non-parametric with semi-parametric,making ultra-short-term interval prediction of wind power at different output levels.By comparing the three interval prediction evaluation indexes under different confidence levels,the proposed algorithm has the best effect and embodies its superiority.