A spatiotemporal characteristics based multi-nodes state prediction method for byproduct gas system and its application on safety assessment

The state prediction of by-product gas system in steel industry plays a pivotal role in its safety assessment, so as to maintain stable operation and production. The fluctuation caused by some units in a time window will then affect others by spatially distributed pipeline network, which may lead to potential safety threats, such as shortage supply, unstable transmission, etc. As such, a multi-node state prediction model considering spatial and temporal characteristics for by-product gas system is proposed in this paper. Considering that the distribution of the key nodes including generation, transmission, storage and consumption presents a non-Euclidean spatial structure, the byproduct gas network is intuitively addressed as a graph model according to the state features in this study, which not only innovatively defines both nodes and edges with regard to their practical consideration, but also establishes physics-related constraints to efficiently and accurately capture the correlation. Then, an interactive extraction mechanism of the node–edge features is designed to achieve dynamic updating of the graph neural network, so that the transient characteristic of gas transportation process can be fully reflected. Finally, the Gated Recurrent Unit (GRU) is introduced to capture the temporal-dependent relationship. Based on the actual data of an iron and steel enterprise in China, the experimental results verified that the proposed method exhibits an advanced accuracy for multi-node prediction. In addition, the prediction interval is constructed to quantify reliability based on the numeric prediction results, which is then verified to be effective for supporting the safety assessment.