一种基于邻域保持的数值天气预报数据降维可信度评估准则

Since the Numerical weather prediction (NWP) data in the power forecasting of renewable power generation is always multidimensional and redundant, dimension reduction of the NWP data is very necessary for data preprocessing before power forecasting. After the feature selection and feature conversion algorithm being applied for NWP dimension reduction, a general reliability assessment of NWP dimension reduction based on neighborhood preservation was proposed. This evaluation criterion is based on the neighborhood changes of sample points in the process of data dimensionality reduction projection, without considering the dimensionality reduction principle and objective function of the algorithm itself, which can effectively evaluate the reliability of different dimensionality reduction algorithms in NWP data dimensionality reduction, thus providing theoretical basis for other researchers to select NWP algorithm.