太阳逐日曝辐量预测建模方法研究  

Modeling method for daily solar radiation energy prediction

作　　者：王建平[1] 郑董文 朱程辉[1] 徐晓冰[1] 

机构地区：[1]合肥工业大学电气与自动化工程学院,合肥230009

出　　处：《电子测量与仪器学报》2014年第12期1332-1339,共8页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金(51177034);安徽省自然科学基金(2011AKZR1598);安徽省国际科技合作计划(11030603014)资助项目

摘　　要：建立一种太阳逐日辐射能量预测的小波过程神经网络等效模型。将神经网络与傅里叶变换以及小波多尺度分解相结合,构建了太阳逐日曝辐量的短期预测和中长期预测问题的两种实时在线预测模型,以提高预测精度。将太阳逐日辐射能量混沌时间序列进行相空间重构,其维数作为网络输入数据的傅里叶变换长度,以得到的傅里叶变换系数作为神经网络的输入,避免输入随意性导致预测结果不一定是其解空间的表达问题;通过小波多尺度分解使太阳逐日曝辐量在一定尺度上表现出准平稳性,以此确定神经网络隐层节点数,以减小试凑性造成的误差。将日照百分比、云量与太阳逐日曝辐量的时间序列同时作为输入量,对两种模型进行学习训练和预测分析。仿真结果表明,2种预测模型可在不同的时间尺度上有效地预测太阳逐日曝辐量,且能有效地实时在线递推预测太阳逐日辐射能量。A kind of wavelet process neural network model to predict daily solar radiation energy is established.Neural network combined with Fourier transform and wavelet multi-scale decomposition is used to build two realtime online prediction models for short-term and medium-and-long-term of daily solar radiation energy to increase the forecast precision. Embedding dimension of the reconstructed phase space is adopted as the network input length of the Fourier transform points,and takes this Fourier transform coefficient as the node point number of input layer to avoid the forecasting result is not in the solution space by arbitrariness of selected inputs. Wavelet multiscale decomposition is employed to make the solar irradiance become quasi-steady performance at a certain scale and determine the number of hidden layer of network to reduce the deviation caused by cut-and-try. Those two models are trained by the data of daily solar irradiance,and time series of insolation and cloud-age. The simulation results show that both of the models can effectively forecast the daily solar radiation energy in different time scales and can effectively estimate daily solar radiation energy in real-time and online.

关 键 词：太阳逐日曝辐量预测模型 相空间重构 小波神经网络 傅里叶变换 

分 类 号：TK51[动力工程及工程热物理—热能工程]