Early warning model for industrial internet platform based on graph neural network and time convolution network

The control system is an important part of the industrial infrastructure. Once the abnormal is likely to cause equipment failure, production interruption and other serious consequences, a timely and effective anomaly detection method is of great significance to the industrial control system. This paper presents a predictive maintenance strategy–early warning model that includes two modules, prediction and alarm, which can also realize online status monitoring. In the prediction module, a two-channel input model using a graphical neural network and a temporal convolutional network, and the graph structure learning module in the graph neural network is improved to optimize the graph structure. For the alarm module, combining expert experience and mathematical statistics knowledge, a hierarchical alarm mechanism based on abnormal scores and abnormal interval accumulation is proposed. We use the real data to verify the performance of the prediction algorithm and the effectiveness of the early warning index. Experimental results with the real data and the open data show that the prediction performance of the proposed early-warning model is better than the state-of-the-art methods. On the real TRT dataset, the evaluation indicator MAPE is 6.5693%, and on the open dataset, MAPE is 1.3368%. At the same time, its early-warning performance is better than the existing automatic interpretation and manual interpretation methods on the industrial internet. © 2022, The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature.