城市空气质量的BP和RBF人工神经网络建模及分类评价

BP and RBF Artificial Neural Network Modeling and Classified Evaluation for Urban Air Quality

根据MATLAB提供的人工神经网络模型,将其应用到城市空气质量评价,研究并对比分析BP和RBF两种人工神经网络的建模方法及评价结果。首先构建BP神经网络模型,确定输入层、隐含层和输出层的神经元数,选择Sigmoid型函数作为激励函数,应用内插扩展出的训练样本对BP网络进行学习,再用训练成熟的BP网络对待评价样本进行仿真;其次构建RBF神经网络模型,确定其输入层和输出层的神经元数,选择Gauss函数作为隐含层激励函数,再用同样的训练样本进行学习和仿真;最终进行归一化论证,验证归一化预处理在空气质量评价中的必要性。结果表明:应用BP和RBF人工神经网络可以得出较好的城市空气质量分类评价结果,其中RBF神经网络模型与改进的灰色聚类法评价结果一致,具有较高的准确率,是一种快捷、有效的综合评价方法。

To study the differences in the modeling and evaluate results betwee Back Propagation（BP）and Radial Basis Function（RBF）neural network,this paper applies the artificial neural network models in MATLAB to evaluating urban air quality.Firstly,the paper determines the neuron number of input layer,hidden layer and output layer and chooses Sigmoid function as the excitation function to build the BP neural network model.Then the paper uses the extended samples obtained by interpolation to train BP neural network,and applies the trained BP network to simulating the samples.Secondly,to determine the neuron number of input layer and output layer,the paper builds a RBF neural network model and chooses Gauss function as the excitation function of hidden layer and uses the same samples to train and simulate.Finally,the paper applies the pretreatment method of normalization and demonstrates its necessity in atmospheric environment quality assessment.The results show BP and RBF neural networks have good effects on the classified evaluation for atmospheric environment quality.In addition,the evaluation results obtained by RBF neural network are in line with improved gray clustering method,so it has been proved to be a quick,effective,precise and comprehensive evaluation method.