Communication-Cognizant Hybrid Voltage Control in Power Distribution Networks

Transcription

1 February 8, Switzerland Communication-Cognizant Hybrid Voltage Control in Power Distribution Networks Hao Zhu Assistant Professor Dept. of Electrical & Computer Engineering University of Illinois, Urbana-Champaign Acknowledgements: Max Liu (UIUC), Wei Shi (Boston Univ), NSF ECCS , PSERC, DoE Sandia Lab

2 Contemporary Modern Distribution Systems Modern design refers to a period of time, it is a design style that was created in the 1920 s 1950 s. It doesn t change, it is a defined style, and will remain such for ever. Contemporary design is ever changing. It is of the moment. Lack of and the reliability issues of communication links, or cyber infrastructures in general, pose as major challenges to the coordination of new devices/resources in distribution systems. 2

3 A contemporary look at voltage regulation LTC Xformer Voltage Regulators High PV Output 126V Voltage 114V Real power could also affect voltage due to high R/X ratio Reactive power (VAR) support ( imaginary power! ) 3

4 Our focus Towards communication-cognizant distribution system management (NSF collaborative project with Harvard) Local Distributed Hybrid 4

5 Prior art Local control: voltage measurements only Droop control (IEEE ) [Farivar et al 13] Integral control for stability [Zhang et al 12,Li-Dahleh 14] Static and dynamic performance under VAR limits [Zhu-Liu 15][Liu et al 16] HZ and H. Liu, Fast Local Voltage Control Under Limited Reactive Power: Optimality and Stability Analysis, IEEE TPWRS, H. Liu, W. Shi, and HZ, Decentralized Dynamic Optimization for Power Network Voltage Control, IEEE Trans. Signal & Info. Proc. over Networks, Distributed control: an optimization framework Consensus averaging for VAR balancing [Robbins et al 13] Distributed optimization using dual decomposition or splitting methods [Dall anese et al 13] [Sulc et al 14][Bolognani et al 15], [Robbins et al 16] [Liu et al 17] E. Dall anese, HZ, G.B. Giannakis, Distributed Optimal Power Flow for Smart Microgrids, IEEE TSG, H. Liu, W. Shi, and HZ, Distributed Voltage Control in Distribution Networks: Online and Robust Implementations, IEEE TSG, 2017 (revised). 5

6 Distribution flow model Distribution network Tree topology system Line impedance (r + jx) Complex power load/line flow Linearize distribution flow [Baran-Wu 89][Bolognani et al 16] (LDF-P) (LDF-Q) (LDF-V)

7 Matrix flow model Bv q + w Lemma: Matrices X and B are positive definite (PD) is the Bbus matrix a full-rank weighted graph Laplacian Linear approximation works under realistic distribution networks with line losses and unbalanced three-phase couplings [Gan et al 14] 7

8 Static problem formulation minimize V mismatch + weighted V mismatch 2 + γ 2 v μ B Convex VAR limits - Vector μ specifies a reference voltage profile (e.g.,μ = 1) - Input w depends on the system operating condition, which can be fast and dynamically varying in practice Weighted norm motivated by local control design 8

9 A partial primal-dual (PPD) approach Network flow constraints only couple neighboring nodes PPD works by alternatively updating v, q, λ (S1) unconstrained quadratic program for v (S2) (projected) gradient-descent for q (hence the name partial) (S3) linear update for λ 9

10 Online implementations Instantaneous feasible as VAR input signal A closer look at (S2) Similarly, (S3) uses 10

11 Algorithmic architecture k λ j 1 j k λ j+1 (S1) v j k+1 v j k (S2) q j k+1 λ j k+1 (S3) k+1 v j 1 j k+1 v j+1 w j k+1 11

12 Stability conditions Prop: For given operating condition, by the PPD updates (S1)-(S3) converges to the optimum if the step-size with / is the min/max singular values of, respectively. Saddle-point flow analysis [Feijer et al 10, Li et al 16] Strong convexity of the objective function Projection operation slightly complicates the contraction proof 12

13 Communication link failure Freeze all other variables (unchanged) and update Remark: Under a total link failure, the proposed control design boils down to a surrogate local integral droop update using the instantaneous voltage measurement (hence hybrid) 13

14 Numerical tests Lossy power flow solutions (not the linear approximation!) Each bus randomly activates with a given probability 14

15 Features of hybrid voltage control (HVC) Minimal sensing and communication overhead Adaptively adjusting step-size to ensure stability Implemented online by incorporating dynamic voltage measurement Robustness to (even total) link failures Thanks to the weighted norm term in the objective Adaptive to local voltage variation even at no communications Almost all existing approaches totally freeze the updates 15

16 Dynamic tests Daily profile of residential load and solar PV output every minute Heavy loading during the evening (18:00-22:00) High solar variability in the afternoon (12:00-17:00) Source: 16

17 Online HVC performance Total communication failure enforced from 16:00-24:00 HVC scheme can effectively reduce the voltage mismatch even in the worst-case communication scenario 17

18 What s next? Quantify the value of communication links (where to place?) Joint voltage regulation and energy management Modeling of the fast power-electronics dynamics Interactions with traditional voltage regulation devices 18

19 Cyber-physical situational awareness (NSF CAREER) 19

20 DOE ARPA-E GRIDDATA project Led by Tom Overbye (ACTIV: Asu-Cornell-Tamu-Illnois-Vcu) Synthetically generating power system models of ,000 buses Illinois200 SouthCarolina

21 Thank you! Hao Zhu 21