基于测点时空特征的工程安全监测网络划分方法

Sensor Network Partition Based on the Spatial Temporal Features for Structure Safety Monitoring

大量传感器测点部署在大型土木结构体中观测多种类型物理量,直接根据海量监测数据评判工程运行状态存在困难.在实际应用中,采用分而治之的思想,根据结构体工程规范、设计资料和工程经验划分成若干区域,融合局部评价结果实现整体评判.但是,其没有考虑测点空间分布信息和监测数据时序变化规律,无法满足结构体安全监测实时评判要求.为了使安全监测网络划分结果能够及时反映结构体物理量动态变化规律,本文考虑监测网络中测点监测数据之间的相关性,提出时序降噪自动编码器(TSDA)压缩高维监测数据,以表征测点时空特征.在此基础上,提出基于测点时空特征的监测网络划分算法(NPA),该算法引入辅助目标变量优化网络划分目标函数,使网络划分结果反映大坝运行物理规律.利用公开数据集和某大坝实测数据进行实验验证,实验结果表明提出的基于测点时空特征的监测区域划分算法NPA在轮廓系数上较TSDA+K-Means和TSDA+GMM分别提高45.1%和58.4%;在CH指标上,较TSDA+K-Means和TSDA+GMM分别提高30.8%和61.6%,说明其可以得到很好的工程安全监测网络的划分结果.

With the rapid development of Internet of Things(IoT)technologies,many various types of sensor node have been deployed in the huge civil engineering to measure the different physical quantities and monitor their changes in various regions of the structure,such as formation,stress,strain,etc.In the safety monitoring systems of a huge civil engineering,the massive monitoring data are generated from the deployed sensor nodes.However,it is difficult to process a large amount of data with the traditional mechanical models,so it is impossible to directly evaluate the safe operation states of the engineering.In real applications,divide and conquer strategy is adopted.The sensor network is divided into multiple regions according to the design specifications,simulation data,and engineering experience.The local results from sub-regions are integrated to achieve overall evaluation.Due to management specifications,instrument failures,environmental changes,and actual monitoring requirements,the spatial distribution of sensor nodes will change.Meanwhile,during the operation period,the measured data and their temporal feature dynamically changes over time.The existing network partition based on the mechanical models ignored the spatial distribution of sensor nodes and their variations of time series,which can not reflect the spatial-temporal features of the measured data change in a real-time manner,resulting in the low accurate evaluation performance.The monitoring data in the large civil engineering is high-dimensional and dynamic spatial-temporal data.The network partitions can timely reflect the dynamic changes of engineering structure,it should consider the similarity of structure and force in the local area of the engineering,and the correlation among the monitoring data.In this paper,a time series denoising autoencoder(TSDA)is proposed to represent the spatial and temporal features of sensor nodes.Considering the correlation among the spatial positions of the different sensor nodes and the change law of the time series,the auxiliary distribution variables are introduced to optimize the objective function of deep clustering.Finally,the Network Partition Algorithm based on TSDA(NPA)is presented to cluster all sensor nodes and partition the sensor network into the different regions based on the spatial-temporal features of nodes.The clustering results can better reflect the physical laws of the large civil engineering.We compare our NPA partition algorithm,with several baseline and state-of-the-art algorithms,including K-Means,AE+K-Means,GMM,AE+GMM,on the public datasets(MNIST,Fashion-MNIST,STL-10,and Reuters)and a real data set from an arch dam.The evaluation metrics are Clustering Accuracy(ACC),Normalized Mutual Information(NMI)and Adjusted Rand Index(ARI).Experimental results demonstrate that the proposed network partition algorithm NPA can achieve better partition performance.On the public data set(CIFAR-10),NPA can enhanceACC 49.4%and 59.1%higher,NMI 46.5%and 54.1%higher,and ARI 47.0%and 41.3%higher than the AE+K-Means and AE+GMM,respectively.On therealdata set,the NPA can improve Silhouette Coefficient 45.1%and 58.4%higher than the TSDA+K-Means and TSDA+GMM,respectively.In the Calinski-Harabaz Index,the NPA can increase by 30.8%and 61.6%,respectively.