基于BP神经网络的城市边缘带土壤重金属污染预测——以成都平原土壤Cd为例

随着成都平原城市化快速发展,城市边缘带土壤重金属污染风险逐渐增大。而关于社会经济对土壤重金属污染定量影响的研究方法还较为欠缺。本文利用BP人工神经网络方法,建立了12输入、1输出、1个隐含层的三层BP神经网络,定量研究成都平原城市发展中社会经济影响因素与土壤重金属Cd含量间的内在联系。网络拟合精度达97.02%,模型拟合程度高。运用该BP网络模型对城市化影响下城市边缘带土壤重金属Cd含量进行预测,其预测精度为84.19%,明显高于传统回归模型71.55%的预测精度,体现出神经网络预测模型的优越性。利用2005年和2010年各影响因素的预测值,将这两组值分别作为网络的输入,并和以前的样本合并再重新训练更新网络权值,得到2005年和2010年各区/县土壤重金属Cd预测值。

PREDICTION OF SOIL HEAVY METAL POLLUTION OF PERI-URBAN ZONE BASED ON BP ARTIFICIAL NEURAL NETWORK——A CASE STUDY OF THE CHENGDU PLAIN

With the rapid urbanization of the Chengdu Plain, the risk of peri-urban soils exposing to heavy metal pollution is aggravating gradually, however, so far no ready-made research method is handy to study quantitatively impact of socio-economic development on soil heavy metal pollution. An attempt was made to study internal relationships between Cd content in soil and its affecting factors related to the socio-economy of the urbanization of the Chengdu Plain with the aid of the BP Artificial Neural Network, which was made up of one input layer of 12 inputs, one output layer and one hidden layer. The network fit extremely well with precision reaching 97.02 %. This BP network model was used to predict Cd content in peri-urban soils, with results 84.19% in precision, which is obviously higher than 71.55% of the traditional regression model, showing superiority of the former in predicting heavy metal pollution. Then the predicted data of each affecting factor in year 2005 and year 2010 were input into the network, and merged with the previous samples. The model was trained over again to renew the network weight values. Thus soil Cd content in each county in the Chengdu Plain in year 2005 and year 2010 was predicted.