基于小波网络的次日太阳逐时总辐射预测技术研究  

Study of Forecasting the Hourly Solar Irradiance of Next Day Using Wavelet BP Networks

作　　者：林星春[1] 曹家枞[1] 刘春雁[1] 

机构地区：[1]东华大学环境科学与工程学院,上海201620

出　　处：《能源技术》2007年第2期70-75,共6页Energy Technology

摘　　要：现有的地面太阳逐时总辐射预测模型的预测精度及泛化能力尚不能令人满意。利用小波神经网络在提升非线性函数影射能力方面的优势,以及递归网络的优良的动态性能,建立了对角递归小波BP网络(DRWBPN)模型,用以对次日地面太阳逐时总辐射进行精确预测。进一步提高预测精度的措施还包括将ASHRAE太阳辐射确定性模型的计算结果和经模糊化处理的气象预报中的云量信息加入到网络输入向量中,以充分利用已知可靠信息。采用分阶段训练网络的方法,提高了有限次数下的训练质量。太阳逐时总辐射预测实例及与其它典型模型预测结果的比较表明,提出的地面太阳逐时总辐射预测模型具有更高精度和实际可行性。The existing models for forecasting hourly total solar irradiance are unsatisfactory in forecast errors and the generalization capability. The wavelet neural networks （ WNN） in enhancing the ability of mapping nonlinear functions, and of the prominent dynamic property of the recurrent neural networks （ RNN）, a new diagonal recurrent wavelet BP network （DRWBPN） is established so as to carry out fine forecast of the next day hourly total solar irradiance. Furthermore, some new measures are adopted to help the enhancement of forecast precision. They include applying dependable information to the input vector of networks, i.e. , the calculation of ASHRAE model is considered as an input to the DRWBPN, and the fuzzilized cloudiness from weather forecast is input to the network as well. As the DRWBPN has been trained according to a specially scheduled 2 - phase- training algorithm, the training time is shortened. As an example an hourly irradiation forecast is completed using the sample dataset in Shangha, and comparisons between irradiation models show that the RWBPN model is definitely higher precise and feasible.

关 键 词：太阳逐时总辐射 预测 对角递归小波BP网络 模糊技术 

分 类 号：P422.1[天文地球—大气科学及气象学] TP183[自动化与计算机技术—控制理论与控制工程]