Expandable deep width learning for voltage control of three-state energy model based smart grids containing flexible energy sources. (1st July 2021)

• Record Type: Journal Article
• Authors: Yin, Linfei; Lu, Yuejiang
• Is Part Of: Energy. Volume 226(2021)
• Publication Date: 2021-07-01
• Subjects: Expandable deep width learning -- Three-state energy model -- Unified time-scale -- Coordinated primary voltage control framework -- Coordinated primary voltage controller; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2021.120437
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: The article establishes a three-state energy (TSE) model for flexible energy sources (FESs) connected to smart grids. The article proposes a unified time-scale (UTS) coordinated primary voltage control framework and a UTS coordinated primary voltage controller for voltage control of smart grids containing a high proportion of FESs. To mitigate uncoordinated voltage, the proposed control framework integrates traditional secondary and primary voltage control into a UTS. The article proposes an expandable deep width learning (EDWL) for the proposed controller. The proposed controller applies the EDWL for predictive control; the proposed controller outputs the reactive power reference value of each TSE unit in smart grids with the real-time voltages of smart grids pilot buses. The proposed algorithm is numerically simulated with the proportional-integral-derivative (PID) algorithm and deep neural networks (DNNs) in IEEE 118-bus, 300-bus, 1354-bus, and 2383-bus systems. The simulation results show that the proposed framework and controller can quickly and accurately control the grid voltage, and verify the feasibility and effectiveness of the proposed approach. The integral of squared error control performance index is 0.47% smaller than the PID and 0.06% smaller than DNNs. Highlights: The designed three-state energy model can accommodate flexible energy sources. A coordinated primary voltage control framework with unified time scale is proposed. An expandable deep width neural networks structure is proposed for the model. An expandable deep width learning algorithm is proposed for the framework. The algorithm can accurately provide commands for generator units of smart grids.