Deeppipe: An intelligent monitoring framework for operating condition of multi-product pipelines

• Published in: Energy (Oxford) 2022-12, Vol.261, p.125325, Article 125325
• Main Authors: Wang, Chang, Zheng, Jianqin, Liang, Yongtu, Wang, Bohong, Klemeš, Jiří Jaromír, Zhu, Zhu, Liao, Qi
• Format: Article
• Language: English
• Subjects: Condition recognition; Condition status change; Intelligent monitoring framework; Pipeline system safety
• ISSN: 0360-5442
• DOI: 10.1016/j.energy.2022.125325

The operation monitoring of multi-product pipelines helps to grasp the operation dynamics, detect abnormal situations in time, and assist on-site operation management. However, due to the complexity of the scheduling plan, the operating conditions of pipelines change frequently, which makes it difficult to accurately recognise condition types. To solve the above problem, an intelligent monitoring framework for operating conditions is proposed to simultaneously achieve the system recognition of steady, unsteady, and abnormal conditions. (i) The proposed monitoring framework extracts temporal and spatial characteristics of condition samples through four modules: Modules 1 and 2 form an unsupervised model for monitoring state changes and capturing temporal characteristics of condition samples; Module 3 is utilised to capture the spatial characteristics; the fusion layer based on Module 4 is applied to nonlinearly fit the spatiotemporal characteristics, and while monitoring the status changes of condition, it can also accurately recognise whether the condition is normal operation adjustment or abnormal condition. (ii) Taking a simulated pipeline and a real pipeline as examples, the effectiveness of the proposed monitoring framework is verified, and the accuracy, precision, recall, and F1 score of the recognition results reach 98.56%, 98.56%, 97.68%, and 98.12%. (iii) Through the sensitivity analysis of each module, accuracy, precision, recall, and F1 score are reduced to 96.10%, 96.10%, 95.83%, and 96.83% (i.e., only 2.46%, 2.46%, 1.85%, 1.29% differences) without Module I, which proves that the framework has strong robustness and generalisation. (iv) Finally, an intelligent analysis and control system of multi-product pipelines is designed for future applications. Consequently, the proposed intelligent monitoring framework can guide the safe operation and management of multi-product pipelines on-site. •An intelligent framework is proposed for operating conditions monitoring of multi-product pipelines.•An unsupervised method is proposed to monitor status changes.•A fusion layer is conducted to recognise whether operation adjustment or abnormal conditions.•Two pipelines are used to verify the effectiveness and reliability of the proposed framework.