Real-time Prediction of Highway Equipment Faults Based on GCN and GRU Algorithms

Abstract

The maintenance and renewal of the highway equipment requires a large amount of financial investment, and it is extremely expensive to directly inspect all the equipment in the highway equipment. Therefore, a cost-effective equipment failure prediction technology is needed, such as a failure prediction model, which can detect highway equipment failures in time and reduce maintenance costs. In this paper, knowledge map technology is introduced to construct a highway equipment map, and a highway equipment fault prediction model based on the combination of semi-supervised Graph Convolution Network (GCN) and Gated Recurrent Unit (GRU) is proposed. This model is based on the GCN’s device network feature perception process, with the fault prediction of highway equipment, which uses GRU to extract the real-time features of the device. The initial values of parameters in GCN and GRU adopt Gaussian distribution and uniform distribution respectively. We apply our model to the problem of active maintenance of highway equipment, and have an experiment with the model using a real data set from a Chinese highway company. The results show that our model can effectively use equipment map information and real-time information, and effectively predict the fault of highway equipment only by partial samples, thereby saving fault response time and maintenance costs.