基于BP神经网络的太阳辐射预测——以兰州市为例

Simulation of solar radiation in Lanzhou based on BP neural network

采用常规气象观测资料建立了兰州市太阳总辐射BP神经网络预测模型,利用神经网络释义图和连接权法剔除了模型中的冗余变量,用优化的BP网络模型预测了兰州市1996-2000年的太阳辐射,并用实测数据验证了该模型。结果表明:该方法增加了模型的透明度,提高了模型的可靠性和鲁棒性,模拟结果与实测值非常吻合,模拟值的各项误差指标值均很小,模拟值与实测值的拟合优度R2达到0.987,通过与其他经验模型的模拟结果进行对比,优化的BP网络模型的模拟效果最好,精度明显高于其它经验模型。因此,对于无太阳辐射观测的地区,优化后的BP网络模型是预测当地太阳辐射的一种有效方法。

A BP artificial neural network model was developed to predict monthly mean daily global solar radiation of Lanzhou by using the meteorological observation data.The model used the interpretation diagram method and connection weights method to compute the relative contribution of each input variable for estimating the importance to the output variable.Then,the redundant input variables can be eliminated.The measured meteorology values from 1971 to 1995 for Lanzhou station were used for the training of the network.Then we can predict the solar radiation from 1996 to 2000 with the simplified network.The measured values of monthly mean daily global solar radiation for Lanzhou station were analyzed to test the network model.The experimental results indicate that the method can increase the transparency of model and improve the reliability and robustness of the model.The model has the smaller error and the higher precision.The coefficient of determination is 0.987.Results have shown better agreement between the estimated and measured values of global solar radiation than other empirical models.Therefore,using the BP neural network model with the meteorological observation data is a very effective method to predict the solar radiation of some region which has no radiation observation sites.