Robust and Decentralized Operations for Managing Renewable ...€¦ · problems in power systems,...

Robust and Decentralized Operations for

Managing Renewable Generation and

Demand Response in Large-Scale

Distribution Systems

PSERC Industry-University Meeting

December 2-4, 2015

Research Team

Andy Sun, Georgia Tech

Duncan Callaway, UC Berkeley

Industry Advisors

• Tongxin Zheng, ISO-NE

• Hong Chen, PJM

• Jim Price, CAISO

• Masoud Abbaszadeh, GE

Research

• Bahman Darynian, GE

Research

• Santosh S. Veda, GE

Research

• Lei Fan, GE Energy

Management

• Eduard Muljadi, NREL

• Mirrasoul J. Mousavi, ABB

• Harish Suryanarayana,

ABB

• Evangelos Farantatos,

EPRI

• Erik Ela, EPRI

• Jens Boemer, EPRI

• Xing Wang, Alstom Grid

• Ying Xiao, Alstom Grid

• Curtis Roe, ATC

Outline

• Motivation

• Project Description

• Technical Approaches

• Robust optimization for scheduling of demand

response and distributed generators

• Fully decentralized optimal dispatch in distribution

systems

• Potential Applications and Benefits

• Summary

Increasing Renewable Penetration

• Peak demand: 69,621MW (Aug 10, 2015)

• Wind Capacity: ~ 16,000MW

• Wind Generation record: 12,971MW (Nov 25, 2015) ~32%

of load at that time

http://www.ercot.com/content/news/presentations/2015/ERCOT_Quick_Facts_12715.pdf

ERCOT 2015

• ERCOT: 6.6 million advance meters

• 97% ERCOT load in competitive area settled with 15-min interval

• More than 2100 MW in demand response, including

• Load resources (mostly large industrial) ~ 1,390 MW

• Emergency response service (commercial & industrial) 850MW

• Utility load management programs ~220MW

• Demand response provider manages large portfolios of DR

• E.g. Enernoc 24-27GW peak load under management over

14000 sites

• DR resources can be highly uncertain

Demand Response Management

Project Description

• Develop Robust Scheduling Tools for managing

uncertainty in demand response portfolios

• Robust operation of DG and DR in distribution systems

with interface to transmission systems

• Fully decentralized optimal dispatch for active distribution

systems

• Study flexibility and reliability performance of proposed

models in distribution systems

• Develop Simulation platform for large-scale systems

TA 1: Robust Scheduling of DR

• A quick motivation on using robust optimization for

power system management:

• Systematic and practical approach to model uncertainty

in variable resources --- “uncertainty set”

• Dispatch decision is uncertainty-aware and adaptive

• Able to save cost and increase reliability comparing to

deterministic approaches

[BLSZZ, 2013][LS, 2015][TSX, 2014][LSLZ, 2015]

TA 1: Uncertainty Set – A Primer

• Uncertainty set for renewable variation

t hour • This classic uncert set is Static

• We want to develop Dynamic

uncertainty set

TA 1: The Need to Model Correlation

• Modeling temporal and spatial correlation of

renewable resources is crucial for operations

Kennewick

Vansycle

Goodnoe Hill

39 km

146 km

wind

• Spatial and temporal correlation

of 3 wind farms

• Wind and solar production are

also correlated

Source [Xie et. al. 2011]

TA 1: Dynamic Uncertainty Set – A New Proposal

• A dynamic uncertainty set for wind speed:

• Dominates performance of static uncertainty set

Seasonal pattern

Residual

Linear dynamics:

Temporal & Spatial

correlation

Uncertainty in

Estimation with

Budget Constraints

[LS 2015]

TA 1: Demand Response Uncertainty

• A demand response event:

• DR resource ramps up, holds reduction, ramps down

• DR aggregator/scheduler plans for DR events

• Final realization of DR performance can be quite different

• Uncertainty in realization depends on planning decision

• How to model such a correlation?

TA 1: A New Dynamic Uncertainty Set

• We propose to develop a new type of dynamic uncertainty

sets that model this decision dependence:

• Scheduled DR reduction decision:

• Deviation in realized DR reduction:

• Final realized DR reduction:

Total variations controlled

Uncertainty depends

on decision

TA 1: Robust Scheduling of DR Portfolio

• Now imagine a DR portfolio of hundreds of C&I DR

resources

• Managing such a DR portfolio with uncertainty in DR

performance is a challenging task

• No commercial software is available

• We propose the following robust scheduling model

•

• is a set of operational constraints on DR reduction decision

TA 1: Robust Scheduling of DR Portfolio

• Preliminary results

• Type A: Highest profit and highest uncertainty

• Type B: Medium profit and uncertainty

• Type C: Lowest profit and lowest uncertainty

• Type A: most favored in deterministic model

• Types B, C: favored in robust model, balance

btw profitability and operation uncertainty

TA 2: Fully Decentralized Dispatch

• With thousands of distributed generators in the

grid, centralized controlling is challenged

• Can we do decentralized control down to the

device level?

• Yes?...!

TA 2: Fully Decentralized Dispatch

• Literature review:• Parallelization of certain computation steps (matrix factorization) in centralized optimal

power flow (OPF) algorithms [Huang,Ongsakul 94] [Lin, Ness 94] [Oyama et. al. 90]

• Market coordination: dividing high-voltage control area into a few sub-regions, each sub-

region solves a OPF [Kim, Baldick 97] [Baldick et. al. 99][Conejo et. al. 02][Ji, Tong, 15]

• Linearized approximation of OPF and decentralization to sub-systems [Biskas, Bakirtzis 06]

• Convex relaxation formulation of OPF and decentralization to cliques [Jabr 06] [Lavaei, Low

12] [Zhang, Tse 11,12] [Boyd et al. 12][Zhu et. al. 14]

• Mostly on linearized DC OPF, regional coordination,

convexification etc.

• We want to do down to the nodal level decentralized

control

• Full decomposition & AC OPF

TA 2: Fully Decentralized Dispatch

• Centralized AC OPF:

Power Flow

Equations

Nodal power/voltage bounds

Minimize cost or loss

TA 2: Fully Decentralized Dispatch

• Nodal decomposition:

• At each node 𝑖, a set of artificial variables are

introduced:

• 𝑒𝑗𝑖, 𝑓𝑗

𝑖,𝜃𝑗𝑖 are node 𝑖’s estimate of node 𝑗’s variables

i

TA 2: Fully Decentralized Dispatch

• Solve the following problem:

• Augmented Lagrangian:

• ADMM consists of three steps:

TA 2: Fully Decentralized Dispatch

One generator at node 0,

N loads at nodes 1…N

• Convergence to a stationary point

• Linear scaling of computation time vs N

Potential Benefits

• This research will provide the industry partners with a set of new

tools for managing large-scale distribution systems with intrinsic

uncertain and active resources, such as distributed renewable

generation and complicated demand response portfolios.

• The new operational models and solution algorithms are anticipated

to substantially increase the utilization of the DG and DR resources,

therefore, encouraging their further adoption in the distribution

system.

• The proposed models and algorithms will also provide new

computational tools for the solution of fundamental operational

problems in power systems, such as the multi-time scale optimal

power flow problem in distribution systems.

• The proposed methodology is not limited to distribution system, but

can be applicable to other power system analysis functions.

Expected Outcomes

• Robust scheduling tools for managing large DR

portfolios under uncertainty of DR resources.

• Uncertainty modeling techniques for distributed DGs,

DRs, and customer owned resources in the distribution

system.

• A hierarchical and decentralized control scheme for

solving multi-time scale optimal power flow problems in

distribution systems.

• Software platform that implements the proposed

modeling and operation tools with data management

and processing functions.

• A comprehensive evaluation of the proposed methods

and models in a real-world power system.

Potential Applications

• The proposed work can be used by utility companies and demand

response aggregators for managing their operational portfolios and

hedge against significant variations in DR resources and renewable

generations.

• The decentralized control scheme provides a scalable approach for

the distribution system operator to operate a large-scale distribution

network with a significant number of active devices.

• If successful, the proposed models and algorithms can help

distribution system operators to upgrade their operational scheme

to allow much more accurate and robust control of the

heterogeneous devices in the system and to improve the flexibility

and reliability of the entire distribution system.

• The models and algorithms from this project can also be developed

into commercial software packages.

Summary• The distribution system is becoming more complex and active. Distribution

system operators may face a portfolio of an extremely large number of devices

including distributed generators (DG), demand response (DR) resources, storage

devices, and emerging proactive customers with various resources (electric

vehicles, smart appliances, rooftop PVs, TCLs).

• Many of these devices may exhibit stochastic supply or consumption patterns.

• The goal of this project is to develop new operational models and algorithms to

efficiently operate such a large portfolio of controllable but uncertain resources in

an active distribution system with the aim to increase flexibility and reliability of

both distribution and transmission systems.

• The proposed models will provide the industry with computational tools to

manage various types of uncertainties through robust optimization techniques

and a mixture of centralized and decentralized control schemes in order to improve

scalability of the operational algorithms.

• The project will also explore efficient solution methods for incorporating

unbalanced multi-phase power flow models in the proposed scheduling

algorithms in order to accurately model the distribution system.

Work Plan

Number Task Year 1 Year 2

1 Develop robust optimization models and algorithms for the DR portfolio management problem

2 Develop robust operational models including renewable DGs, DRs, and storage

3 Develop decentralized operation model and algorithms for the multiphase AC OPF problem

4 Development of simulation platform with data processing functions

5 Comprehensive evaluation on real-world power systems

6 Project documentation

THANK YOU!

References:

• D. Bertsimas, E. Litvinov, X. A. Sun, J. Zhao, T. Zheng, “Adaptive

robust optimization for the security constrained unit commitment

problem,” IEEE Trans. Power Syst., vol. 28, no. 1, pp. 52-63, 2013.

• A. Lorca, X. A. Sun, “Adaptive robust economic dispatch with

dynamic uncertainty sets for significant wind,” IEEE Trans. Power

Syst., vol. 30, no. 4, pp. 1702-1713, 2015.

• A. Lorca, X. A. Sun, E. Litvinov, T. Zheng, “Multistage robust

optimization for the unit commitment problem,” accepted for

publication at Operations Research, 2015.

• A. Thatte, X. A. Sun, L. Xie, “Robust Optimization Based Economic

Dispatch for Managing System Ramp Requirement”, HICSS 2014.

Multi-period, Multi-phase AC OPF

• min 𝑡=1

𝑇 𝑓𝑡 𝒑𝑡 ∶ 𝒑𝑡 ∈ 𝑃𝑡 , −𝑹𝑡 ≤ 𝒑𝑡 − 𝒑𝑡+1 ≤ 𝑹𝑡 ,∀𝑡 = 1,… , 𝑇

• Time decoupling:

• Lagrangian relaxation of ramping constraints

• Resource decoupling:

• Ramping is resource specific

• Multiphase AC OPF

• 𝑃𝑖𝑠 = 𝑗=1𝑁 𝑡=𝑎

𝑐 𝐺𝑖𝑗𝑠𝑡 𝑒𝑖𝑠𝑒𝑗𝑡 + 𝑓𝑖𝑠𝑓𝑗𝑡 − 𝐵𝑖𝑗𝑠𝑡 𝑒𝑖𝑠𝑓𝑗𝑡 −