基于图神经网络聚类的土壤监测点位优化  

作　　者：陈志奎[1,2] 杨志朋 陈轩 

机构地区：[1]大连理工大学软件学院,辽宁大连116620 [2]大连理工大学辽宁省泛在网络与服务软件重点实验室,辽宁大连116620

出　　处：《环境保护与循环经济》2024年第1期72-75,88,共5页environmental protection and circular economy

基　　金：大连市创新基金(2021JJ12SN44)。

摘　　要：在土壤污染的监测过程中,由于需要大面积的监测点位布设,所以需要大量的人力和物力,同时原有的土壤监测点位存在冗余和代表率低的问题,对监测结果的准确性和全面性产生了影响。使用一种基于自编码器降维和图卷积网络(GCN)推理的土壤点位优化方法,通过将原有监测点位聚类成不同簇,来实现用尽量少的监测点位全面地表现区域污染情况的目的。使用GCN推理结合自编码器降维的土壤点位优化方法,该方法同时利用土壤监测点位数据的结构信息和高维特征表示,将监测点位聚类成不同簇。通过对相关性进行分析,发现聚类后的簇中心点位可以作为原始簇的代表,从而降低土壤监测点位的冗余性,实现了监测点位的优化。实验证明GCN方法可以有效地减少监测点位的数量,同时保持监测结果的准确性和全面性。During the monitoring process,due to the need for a large area of monitoring points,a large amount of manpower and material resources are required.At the same time,the existing soil monitoring points have redundancy and low representativeness,which has an impact on the accuracy and comprehensiveness of the monitoring results.Therefore,this article uses a soil point optimization method based on autoencoder dimensionality reduction and graph convolutional network(GCN)inference to achieve the goal of comprehensively representing regional pollution with as few monitoring points as possible by dividing monitoring points into different clusters.Specifically,this article uses a soil point optimization method that combines graph convolutional network inference and autoencoder dimensionality reduction.This method utilizes the structural information and higher-order feature representations of soil monitoring point data to cluster monitoring points into different clusters.By analyzing the correlation,it was found that the clustered center points can serve as representatives of the original cluster,reducing the redundancy of soil monitoring points and achieving optimization of monitoring points.Finally,experiments have shown that the GCN method can effectively reduce the number of monitoring points while maintaining the accuracy and comprehensiveness of monitoring results.

关 键 词：土壤监测 点位优化 图神经网络 聚类 

分 类 号：X833[环境科学与工程—环境工程]