基于神经网络技术的天气雷达超折射回波识别

Identification of Superrefraction Echoes of Weather Radar Based on Neural Network

基于SA多普勒天气雷达资料,从其基本反射率、径向速度和谱宽3个基本产品中提取出6个能反映它们之间差异的特征参量,并对它们进行概率统计分析,作为BP神经网络的识别因子。通过建立适当的训练集对神经网络进行训练,从而得到最优的网络结构,再利用测试集对经过训练的网络做进一步测试,对其识别效果进行评判。结果表明:当神经网络的输入层、隐含层和输出层的神经元个数分别为6、6、2时,能够对超折射回波达到最优的识别效果。最终通过实际个例对识别效果做再次验证。

Based on the SA Doppler weather radar data,six characteristic parameters are extracted from the basic reflectivity,radial velocity,and spectrum width of three basic products,which can reflect the differences between them,and their probability analysis are used for the recognition factors of BP neural network.Through the establishment of the proper training set,the neural network is trained,and the optimal network structure is obtained.Then the test set is used to further test the trained network,and the identification effect is evaluated.The results are as follows：when the number of neurons in the input layer,the hidden layer,and the output layer of neural networks is 6,6 and 2,and the best recognition effect can be obtained for the super-refraction echoes.Finally,the validity of the recognition is verified again by actual cases.