Data-driven real-time power dispatch for maximizing variable renewable generation. (15th May 2016)

• Record Type: Journal Article
• Authors: Li, Zhigang; Qiu, Feng; Wang, Jianhui
• Is Part Of: Applied energy. Volume 170(2016)
• Page Start: 304
• Page End: 313
• Publication Date: 2016-05-15
• Subjects: Data-driven -- Real-time dispatch -- Renewable energy generation -- Uncertainty; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2016.02.125
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: A novel real-time power dispatch framework for variable renewable generation (VRG) is developed. A data-driven approximation model is formulated to determine the do-not-exceed limits of VRG without parameter preselection. A data selection strategy is proposed to obtain the most relevant samples for DNE limit calculation. The proposed method has greater potential for improving VRG integration than the original do-not-exceed limit method. Abstract: Traditional power dispatch methods have difficulties in accommodating large-scale variable renewable generation (VRG) and have resulted in unnecessary VRG spillage in the practical industry. The recent dispatchable-interval-based methods have the potential to reduce VRG curtailment, but the dispatchable intervals are not allocated effectively due to the lack of exploiting historical dispatch records of VRG units. To bridge this gap, this paper proposes a novel data-driven real-time dispatch approach to maximize VRG utilization by using do-not-exceed (DNE) limits. This approach defines the maximum generation output ranges that the system can accommodate without compromising reliability. The DNE limits of VRG units and operating base points of conventional units are co-optimized by hybrid stochastic and robust optimization, and the decision models are formulated as mixed-integer linear programs by the sample average approximation technique exploiting historical VRG data. A strategy for selecting historical data samples is also proposed to capture the VRG uncertainty more accurately under variant prediction output levels. Computational experiments show the effectiveness of the proposed methods.