Event Fusion and Decision Approach Based on Node Reliability in Dam Safety Monitoring

The sensor networks are widely deployed in the large-scale dams to monitor the structural deformation. The deformation of multi-type dam structure results in the transition of the corresponding events states, so it can evaluate the dam safety through the analysis of multi-type events. The dam safety evaluation includes two phases: the events classification in each single region and the features fusion for all regions. In the real applications, individual node's measurement is unreliable due to the network instability. The current approaches do not distinguish the impacts on the results of the event classification among the different nodes, it may result in the classification deviation. With regard to the accurate and real-time decision for operating conditions of the structure, we propose a Sensor Network Decision Framework based on regional division. In the event classification phase, it can improve the accuracy of the event classification via computing the node decision vectors based on the node reliabilities. In the event fusion phase, the attention mechanism is adopted to extract the key features of single regions and the DNN is established to perform global safety evaluation. Experimental results illustrate that SNDF can improve the accuracy by 17% and reduce the decision time by 8.15 seconds compared with the Expert Weighting Method.