The fitting method of wind power forecast error under power-time dimension

Randomness and volatility are the inherent attributes of wind power, and the present method for forecasting wind power is not accurate enough. However, the exact description of forecast error has obvious effect on optimized scheme of power flow calculation, unit commitment and operation in the power system with wind power. The wind forecast error is different under the power dimension and correlative under time dimension, so it can describe the actual wind power attributes more accurately on the basis of fitting the forecast error divided into different groups according to the power-time attributes. The parameters distribution like normal distribution can not describe the error in different bins accurately, but this paper proposed non-parameter kernel density estimation (NKDE) to fit the error data in various bins, which is also suitable for various forecast methods. If ignoring the characteristic of NKDE, it has poor fit ting effect because of too many bins and too few data in some bins. This paper proposed a method to reduce the number of bins to improve the fitting effect and reduce calculation time. Results for this case show the proposed probability density function (pdf) is more close to the actual wind power error distribution.