Basic Explanations Of The Forecast Combination And Studies Of White Noise Denoising For Wind Forecasting

With the rapid development of the wind power, it has been widely utilized as akind of new green energy and paid more attention. Separating the white Gaussiannoise from the close sampling of the wind series as the first step of the "mechanism+identification" forecasting strategy can improve the accuracy of the forecasting of thewind power. And it is very significant for the safe running of the power system andthe valid utilization of the wind energy.The "forecast combination puzzle" has been paid attention as a famous topic inthe forecasting field since it is put forward. It describes the strange phenomena thatthe simple average is usually better than the complex combination in practice, and thestrange phenomena will be studied by mathematical statistics.The main contents and results are:(1)In2009, Smith and Wallis put forward the empirical discovery that "the finitesample size" is a factor which affects the "forecast combination puzzle". Besides that,we propose the other two factors "the population distribution deviations", and explainthe three factors by mathematical statistics. We also find that the robust statistics canresist the outlier worsen the performance of the combining weights.(2) The wavelet denoising is not robust and is caused by the wavelet base, therules of wavelet threshold, the wavelet decomposition level and so on. As a result ofthat, this paper puts forward two kinds of Fourier analysis denoising. The first kindFourier analysis denoising partitions the white Gaussian noise by a given frequencythreshold. The second kind Fourier analysis denoising partitions the white Gaussiannoise by a given power spectral density threshold, and it is better than the first in thecase that there are superposition of the frequency bands of the signal and white noise.Finally, verify the second kind Fourier analysis denoising is feasible and can be betterwithout the interference of singularities to a certain via the numerical experiments.(3) Using the variance and the chi-square test of the white noise defines thewavelet decomposition level so that it improves the performance of the waveletdenoising.(4) Because that Tianjin is in the monsoon area, the paper puts forward the newmethod that can improve the accuracy of the forecasting of the wind power combining the close sampling wind speed series and the spatial correlation.