Deeppipe: A hybrid model for multi-product pipeline condition recognition based on process and data coupling

Abstract

It is significant for field monitoring and management to identify the operating state of the pipeline and prevent an abnormal accident from occurring. Aiming at the few unsteady and abnormal conditions data samples, different condition switching duration, and strong time-space characteristics of multi-product pipeline condition recognition, a hybrid model for condition recognition based on process and data coupling is proposed. First, collecting the relevant parameters of the real multi-product pipeline system and using Stoner Pipeline Simulator (SPS) to establish a simulation model. Based on the simulation model, performing the simulation of unsteady conditions and abnormal conditions to obtain the corresponding simulated data. Second, the reliability of the simulation model is verified by calculating the distribution similarity between the simulated data and the real pipeline data. Third, the simulated data and real data are processed by variable-length series and input into the proposed hybrid model for training and verification. The results show that the proposed hybrid model has high accuracy, precision, recall, and F1 score of 93.1%, 93.1%, 87.2%, and 84.7%, respectively, which has essential guiding significance for monitoring and ensuring the safe and stable operation of multi-product pipelines.