关键词: 太阳辐照度, 预测, AdaBoost算法, 双层协同架构, LightGBM, 多隐层极限学习机

Abstract: In order to enhance the accuracy and universality of daily solar radiation intensity prediction, a bi-layer collaborative prediction model based on multi-feature analysis is proposed. Firstly, a bi-layer collaborative architecture is established, which divides the entire model into two parts, the base layer and the promotion layer. It tracks the multi-feature and changing trends of the target object using a layered prediction method. Secondly, a benchmark prediction model is also established base on the feature learning prediction method, by using LightGBM and numerical weather prediction （NWP） as input. Then, on the basis of the former, the correlation between the solar irradiance at the target moment and the historical time series data is mined. The improved AdaBoost algorithm and the multiple hidden layer extreme learning machine （MH-ELM） are introduced as the main body of the lifting layer to improve the stability of time series prediction. Finally, the actual measured solar irradiance data of a photovoltaic power station in the central region of China is selected to analyze the calculation example. The rationality and validity of the model are verified.

Key words: solar irradiance, prediction, AdaBoost algorithm, bi-layer collaborative architecture, LightGBM, multiple hidden layer ELM