A multi-model-integration-based prediction methodology for the spatiotemporal distribution of vulnerabilities in integrated energy systems under the multi-type, imbalanced, and dependent input data scenarios. (15th August 2022)

• Record Type: Journal Article
• Authors: Sun, Chenhao; Zhou, Zhuoyu; Zeng, Xiangjun; Li, Zewen; Wang, Yuanyuan; Deng, Feng
• Is Part Of: Applied energy. Volume 320(2022)
• Publication Date: 2022-08-15
• Subjects: IES vulnerability -- Spatiotemporal distribution prediction -- Pattern recognition -- Two-expert-systems-integrated methodology -- Structure-based impact weight evaluation; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2022.119239
• Languages: English
• ISSNs: 0306-2619

Abstract

Abstract: The reliability of an integrated energy system (IES) is most likely menaced by its weakest spots, and hence the prediction-based maintenance (PBM) is deployed to forecast these risky periods and locations where inspection and maintenance (I&M) actions are requisite. The key to accomplishing this is the pinpoint vulnerabilities prediction with sufficient lead time to prepare the PBM. With such motivations, one short-term prediction methodology for the spatiotemporal distribution of vulnerabilities, the fuzzy association pattern recognition covering rare and dependent factors (FAPRrdf), is developed in this paper. In the first preprocessing stage, a parallel learning process is formed. The discrete and continuous features are assessed through two expert models, respectively, to differentiate the common and rare components in each feature; Secondly, in the qualitative analysis, a two-expert-systems-integrated single entity approach is established to distinguish the "risky" components. The separated common and rare components are further evaluated to extract the high-impact (HI) and high-impact-low-probability (HILP) components, respectively. Ergo, the imbalanced data distribution can be solved, and the optimization step of weights between two independent expert models in some ensemble methods is no longer required; The third step is the quantitative analysis, and a structure importance measure (SIM)-based impact weight evaluation framework is proposed to rate the specific risk level of the "risky" components. The impact of each component on the variation of total system risks, as well as all the potential paths that will lead to a fault event in this system, are incorporated. Thereof, both the self-impacts of each component and the dependence on other components can be taken into account. Finally, this methodology is validated via an empirical case study, and its flexibility and feasibility during real applications can therefore be demonstrated. Graphical abstract: Highlights: The spatiotemporal distribution prediction of the IES vulnerabilities is achieved. A two-expert-system-integrated method is built to tackle multi-type input features. Fuzzy set and fuzzy significance measurement are redesigned for imbalanced data. Components' dependence is considered in a SIM-based impact weight evaluation model. An empirical case study validates the capability of the method during applications.