基于人工神经网络的街道峡谷NO_x浓度的数值模型研究

通过对反向传播人工神经网络的算法和网络结构的研究,发现拟牛顿算法训练速度较快,能够较好地接近误差目标值,同时建立了包括输入层、隐含层、输出层的人工神经网络三层拓扑结构。通过对街道峡谷人工神经网络的训练,模拟计算了街道峡谷NOx浓度分布值。结果显示,训练误差和测试误差比为1.11,训练样本的模拟值与实测值的相关系数为0.93,测试样本的模拟值与实测值的相关系数为0.87,模拟值与实测值的相关系数均高于显著水平为α=0.05与α=0.01所对应检验性表的相关系数临界值。该模型能够用于街道峡谷污染物浓度的模拟计算,具有较好的泛化能力。

Study on NOx concentration prediction in a street canyon based on artificial neural networks

The studies on the algorithm and architecture of a back-propagation artificial neural network showed that the quasi-Newton algorithm could exhibit good training speed as well as fine close to set error. Meanwhile, the artificial neural network, which included an input layer, an output layer and a hidden layer, was constructed. The simulation of the NO_x concentration in a street canyon was carried out based on the training of its artificial neural network. The results showed that the ratio of testing error to training error was 1.11. The correlation coefficient between a simulated value and measured value of training sample was 0. 93, while one of the testing samples was 0.87. The correlation coefficient between simulated value and measured value was higher than the R-values of the significant levels at 0. 01 and 0.05. The model showed remarkable generalization capacity and well accessibility to be employed to simulate the pollutant concentration in a street canyon.