由于不同气象条件会影响太阳辐照度的有效利用,这制约了太阳能的应用和发展.为了基于不同站点不同采样时刻的气象属性预测中尺度站的太阳能辐照度,依据传统卷积神经网络的框架,建立了一种新型的卷积神经网络结构并用于太阳能辐照度预测...由于不同气象条件会影响太阳辐照度的有效利用,这制约了太阳能的应用和发展.为了基于不同站点不同采样时刻的气象属性预测中尺度站的太阳能辐照度,依据传统卷积神经网络的框架,建立了一种新型的卷积神经网络结构并用于太阳能辐照度预测.为了缓解新型网络由超参数选取不当导致预测性能差的问题,利用融合算法对新型网络的超参数进行优化.为了提高融合优化算法的全局搜索能力,引入帐篷映射对粒子的初始位置和初始速度进行混沌初始化.首先,导入训练集更新新型卷积神经网络框架,训练结束后导入验证集检验当前模型参数下新型卷积框架的性能.其次,混沌融合算法依据新型卷积神经框架在验证集上的预测性能更新模型的超参数.对更新模型的超参数多次检验,直至最优的预测模型在验证集上的性能趋于收敛.最后,输出模型的最优超参数,建立太阳能辐照度预测模型.基于气象实测数据建立太阳能辐照度预测实验,引入其他两种预测方法进行对比仿真研究,并尽可能复现了Eustaquio and Titericz团队的预测方法(GBRT)作为太阳能辐照度预测性能的评估基准.实验数据表明:混沌融合算法可以有效地提高新型卷积神经网络的预测性能,所提出预测方法的全年太阳能辐照度的均方误差较GBRT降低25.9%,绝对平均误差较GBRT降低了10.7%;全年太阳能辐照度平均误差率降低了18.4%,误差率小于0.1的样本量增加了21.1%.展开更多
The effective temperature of the solar photosphere is usually obtained according to the solar constant, based on the Stefan-Boltzmann law. However its temperature distribution is not homogeneous. A hopeful way to obta...The effective temperature of the solar photosphere is usually obtained according to the solar constant, based on the Stefan-Boltzmann law. However its temperature distribution is not homogeneous. A hopeful way to obtain the area-temperature distribution of the solar photosphere is to solve the Black-body Radiation Inversion (BRI) problem. In this paper, a new practical solution method for BRI is developed. The theoretical analysis and numerical calculations show the low-temperature distribution difficulty of BRI is solved by this new method. Then the area-temperature distribution of the solar photosphere is obtained, according to the measured absolute solar spectral irradiance. It is the first realization of BRI for a real system after almost three decades of efforts. The results are comparable to that from the Stefan-Boltzmann law.展开更多
文摘由于不同气象条件会影响太阳辐照度的有效利用,这制约了太阳能的应用和发展.为了基于不同站点不同采样时刻的气象属性预测中尺度站的太阳能辐照度,依据传统卷积神经网络的框架,建立了一种新型的卷积神经网络结构并用于太阳能辐照度预测.为了缓解新型网络由超参数选取不当导致预测性能差的问题,利用融合算法对新型网络的超参数进行优化.为了提高融合优化算法的全局搜索能力,引入帐篷映射对粒子的初始位置和初始速度进行混沌初始化.首先,导入训练集更新新型卷积神经网络框架,训练结束后导入验证集检验当前模型参数下新型卷积框架的性能.其次,混沌融合算法依据新型卷积神经框架在验证集上的预测性能更新模型的超参数.对更新模型的超参数多次检验,直至最优的预测模型在验证集上的性能趋于收敛.最后,输出模型的最优超参数,建立太阳能辐照度预测模型.基于气象实测数据建立太阳能辐照度预测实验,引入其他两种预测方法进行对比仿真研究,并尽可能复现了Eustaquio and Titericz团队的预测方法(GBRT)作为太阳能辐照度预测性能的评估基准.实验数据表明:混沌融合算法可以有效地提高新型卷积神经网络的预测性能,所提出预测方法的全年太阳能辐照度的均方误差较GBRT降低25.9%,绝对平均误差较GBRT降低了10.7%;全年太阳能辐照度平均误差率降低了18.4%,误差率小于0.1的样本量增加了21.1%.
基金supported by the National Natural Science Foundation of China (Grand Nos. 10675031, 10375012 and 19975009)supported in part by the Department of Education of Zhejiang Province (Grant No. Y200906911)
文摘The effective temperature of the solar photosphere is usually obtained according to the solar constant, based on the Stefan-Boltzmann law. However its temperature distribution is not homogeneous. A hopeful way to obtain the area-temperature distribution of the solar photosphere is to solve the Black-body Radiation Inversion (BRI) problem. In this paper, a new practical solution method for BRI is developed. The theoretical analysis and numerical calculations show the low-temperature distribution difficulty of BRI is solved by this new method. Then the area-temperature distribution of the solar photosphere is obtained, according to the measured absolute solar spectral irradiance. It is the first realization of BRI for a real system after almost three decades of efforts. The results are comparable to that from the Stefan-Boltzmann law.