基于GIS和BP神经网络模型的PM_(10)浓度预测与空间分布研究

Prediction and Spatial Distribution of PM_(10) Concentration in Different Algorithms and Neurons by Using GIS and BP Neural Network Model

耦合GIS和BP神经网络模型,探讨不同算法和隐藏神经元数对PM_(10)浓度预测和空间分布的影响,结果显示:不同算法的PM_(10)浓度预测值与监测值的平均相关系数和平均相对误差分别为0.85和17.58%,Levenberg-Marquardt优化算法在隐藏神经元数为20时预测精度最高。相同算法,不同隐藏层神经元数对PM_(10)浓度的预测结果影响较大,不同算法,相同神经元数对PM_(10)浓度的预测结果影响较小。不同算法的PM_(10)浓度空间分布模拟在中北部的高风险区和东南部的低风险区与监测数据结果基本一致。

This study analyzed the impact of different algorithms and the hidden neurons on forecast and spatial distribution of PM10 by coupling of geographic information system GIS and BP neural network model. The results show that the average correlation coefficient and relative error of simulated and observed PM10 concentration by different algorithms are 0. 85 and 17. 58%,the Levenberg-Marquardt optimization algorithm（ Trainlm） had highest prediction accuracy when the number of hidden neurons was 20.The same algorithm,different number of hidden layer neurons had a greater influence on the prediction effect of PM10 concentration,while different algorithms,the same number of neurons had a less influence on it. Spatial distribution of PM10 concentration by different algorithms indicated that the simulation of high risk distributed in north-central areas and low risk located in southeast zone,which are basically identical for monitoring value.