optimizing economic operations for residential …...to appear at ieee smartgridcomm 2016 in nov....
TRANSCRIPT
Wei-Peng Chen and Daisuke Mashima
Fujitsu Laboratories of America, Inc.October 11th, 2016
Optimizing Economic Operations for Residential Demand Response Programs
Copyright 2016 Fujitsu Laboratories of America, Inc.
Residential DR has great potential but not yet popular
Objective 1: to expand residential DR adoption
Copyright 2016 Fujitsu Laboratories of America, Inc.1
$-
$100
$200
$300
$400
$500
$600
$700
$800
$900
$1,000
2015 2016 2017 2018 2019 2020 2021 2022 2023 2024
C&I
Residential
Unit($ Million)
Market forecast for Demand Response (DR) services in C&I (Commercial & Industrial) and Residential sectors
Source: Navigant Research
Complexity of managing multiple DR programs
Multiple types of DR programs exist e.g. Critical Peak Pricing (CPP), Peak Time Rebate (PTR), Direct Load
Control (DLC) program
Utilities would like to employ different DR programs to attract residential customers with different preferences
BUT, the complexity of handling multiple programs is non-trivial
Copyright 2016 Fujitsu Laboratories of America, Inc.2
CPP PTR DLC
Category Time-based Time-based Incentive-basedRewards Lower off-peak rates Performance-basis Compliance-basis
Electricity rate Dynamic Flat Flat
Limited calls Yes No Yes
Direct control No No Yes
Objective 2: jointly managing multiple DR programs and optimally controlling DR resources in a large scale deployment
Solution: Comprehensive residential DR framework
Mobile app serves as an end-customer facing DR client Support multiple DR programs, and enables fine-granularity
control.
Copyright 2016 Fujitsu Laboratories of America, Inc.3
VTN: Virtual Top NodeVEN: Virtual End Node
•"Residential Demand Response System Framework Leveraging IoT Devices." To appear at IEEE SmartGridComm 2016 in Nov. 2016. •"SPOT: A Smartphone-Based Platform to Tackle Heterogeneity in Smart-Home IoT Systems." To appear at the IEEE 3rd World Forum on Internet of Things in Dec. 2016.
Advantages of mobile app DR solution
A bi-directional, real-time channel between utilities and consumers Better reliability
Cost-effective platformNo hardware hub/gateway neededLow cost, easy installationSupport multiple DR programsLower customer acquisition cost
Automated, informative features to enhance customer engagement Automated control of multiple IoT devices Informative energy data analytics
Copyright 2015 FUJITSU LABORATORIES OF AMERICA4
Summary of contributions Present mathematical modeling of optimal DR resource control
for three representative DR programs, (CPP, PTR, and DLC programs), with simple solutions that minimize utilities’ operation cost for DR. The model is based on the two-settlement electricity market adopted by
most of the wholesale electricity markets in the U.S. such as ERCOT, PJM, MISO, and CAISO.
That is, we consider both day-ahead and real-time market.
Compare performances of multiple DR programs through simulations based on real-world data.Real statistical data from PJMReal DR program settings from utilities
Copyright 2016 Fujitsu Laboratories of America, Inc.5
Basics of two-settlement electricity market
𝑪𝑪𝑬𝑬𝒅𝒅𝒅𝒅 = cost of energy procurement at the hour ℎ of the day 𝑑𝑑from both day-ahead (DA) and real-time (RT) markets
Copyright 2016 Fujitsu Laboratories of America, Inc.6
Real-Time Market
Day-Ahead Market
DemandSupply
𝒑𝒑𝑫𝑫𝑫𝑫𝒅𝒅𝒅𝒅𝑫𝑫𝑫𝑫𝑫𝑫𝒅𝒅𝒅𝒅
𝒑𝒑𝑹𝑹𝑹𝑹𝒅𝒅𝒅𝒅𝑫𝑫𝑹𝑹𝑹𝑹𝒅𝒅𝒅𝒅
𝑪𝑪𝑬𝑬𝒅𝒅𝒅𝒅 = 𝒑𝒑𝑫𝑫𝑫𝑫𝒅𝒅𝒅𝒅 � 𝑫𝑫𝑫𝑫𝑫𝑫𝒅𝒅𝒅𝒅 + 𝒑𝒑𝑹𝑹𝑹𝑹𝒅𝒅𝒅𝒅 � 𝑫𝑫𝑹𝑹𝑹𝑹𝒅𝒅𝒅𝒅 − 𝑫𝑫𝑫𝑫𝑫𝑫𝒅𝒅𝒅𝒅
DA price RT energy demandRT priceDA energy
procurement
Optimized DR operations Objective function = to maximize the utility’s profit of DR operations Control variables: utilities decide when and which participants to
dispatch DR events in each DR programWhen 𝑠𝑠𝑖𝑖𝑖𝑖 = 1, the utility send a DR event to a DR participant 𝑖𝑖 at the day 𝑑𝑑 Fine granularity of DR resource selection and control is key to efficiency
Model of DR participants’ behaviors Regular load of a DR participant 𝑖𝑖 when 𝑠𝑠𝑖𝑖𝑖𝑖 = 0 ⇒ 𝐷𝐷𝑖𝑖𝑖𝑖𝑖 Curtailment at hour ℎ of day 𝑑𝑑 of participant 𝑖𝑖 when 𝑠𝑠𝑖𝑖𝑖𝑖 = 1 ⇒ ∆𝐷𝐷𝑖𝑖𝑖𝑖𝑖
Optimization problem:
𝐸𝐸𝑖𝑖𝑖 = profit of a utility at hour h of day 𝑑𝑑 = 𝑅𝑅𝑖𝑖𝑖 − 𝐶𝐶𝐷𝐷𝐷𝐷𝑑𝑑𝑑 − 𝐶𝐶𝐸𝐸𝑑𝑑𝑑= Revenue – DR cost – Energy procurement cost
𝐶𝐶𝐸𝐸𝑑𝑑𝑑 = 𝑝𝑝𝐷𝐷𝐷𝐷𝑑𝑑𝑑𝐷𝐷𝐷𝐷𝐷𝐷𝑑𝑑𝑑 + 𝑝𝑝𝐷𝐷𝑅𝑅𝑑𝑑𝑑 ∑𝑖𝑖 𝐷𝐷𝑖𝑖𝑖𝑖𝑖 − 𝑠𝑠𝑖𝑖𝑖𝑖∆𝐷𝐷𝑖𝑖𝑖𝑖𝑖 + 𝐷𝐷𝑂𝑂𝑑𝑑𝑑 − 𝐷𝐷𝐷𝐷𝐷𝐷𝑑𝑑𝑑Copyright 2016 Fujitsu Laboratories of America, Inc.7
Constraints depend on the contract of each program
max[𝑠𝑠𝑖𝑖𝑑𝑑]
∑𝑖𝑖=1𝑖𝑖=365∑𝑖=𝑖𝑠𝑠𝑖=𝑖𝑒𝑒 𝐸𝐸𝑖𝑖𝑖
s.t. the constraints on 𝑠𝑠𝑖𝑖𝑖𝑖 ∀𝑖𝑖 ∈ 𝑁𝑁
Comparison of formulations of 3 DR programs
DR program Critical Peak Pricing (CPP)
Peak-Time Rebate (PTR)
Direct Load Control (DLC)
Characteristic During critical peak day, the high rate 𝑝𝑝𝑖 is applied compared to the low rate 𝑝𝑝𝑙𝑙is used in normal days
The pre-determined rate of the PTR credit is 𝑝𝑝𝐷𝐷𝐷𝐷
A lump sum DLC credit is paid in a year, so no need to consider the cost of individual DR event
revenue and DR cost
𝑅𝑅𝑖𝑖𝑖 = 𝑝𝑝𝑓𝑓𝐷𝐷𝑂𝑂𝑑𝑑𝑑 + 𝑝𝑝𝑙𝑙 ∑𝑖𝑖 1 − 𝑠𝑠𝑖𝑖𝑖𝑖 � 𝐷𝐷𝑖𝑖𝑖𝑖𝑖 +𝑝𝑝𝑖 ∑𝑖𝑖 𝑠𝑠𝑖𝑖𝑖𝑖 𝐷𝐷𝑖𝑖𝑖𝑖𝑖 − ∆𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝐶𝐶𝐷𝐷𝐷𝐷𝑑𝑑𝑑 = 0
𝑅𝑅𝑖𝑖𝑖 = 𝑝𝑝𝑓𝑓�𝐷𝐷𝑂𝑂𝑑𝑑𝑑 + ∑𝑖𝑖 𝐷𝐷𝑖𝑖𝑖𝑖𝑖 − 𝑅𝑅𝑖𝑖𝑖 = 𝑝𝑝𝑓𝑓�𝐷𝐷𝑂𝑂𝑑𝑑𝑑 + ∑𝑖𝑖 𝐷𝐷𝑖𝑖𝑖𝑖𝑖 −
Constraint Maximal CPP events in a year = 𝑀𝑀
No limitation on the number of PTR events
Maximal DLC events in a year = 𝑀𝑀
Optimization problem
𝑚𝑚𝑚𝑚𝑚𝑚[𝑠𝑠𝑖𝑖𝑑𝑑]
∑𝑖𝑖=1𝑖𝑖=365∑𝑖𝑖 𝑠𝑠𝑖𝑖𝑖𝑖 ∑𝑖=𝑖𝑠𝑠𝑖=𝑖𝑒𝑒� 𝑝𝑝𝑖 − 𝑝𝑝𝑙𝑙 𝐷𝐷𝑖𝑖𝑖𝑖𝑖 − �𝑝𝑝𝑖 −
𝑚𝑚𝑚𝑚𝑚𝑚[𝑠𝑠𝑖𝑖𝑑𝑑]
�𝑖𝑖
𝑠𝑠𝑖𝑖𝑖𝑖 � �𝑖=𝑖𝑠𝑠
𝑖=𝑖𝑒𝑒
𝑝𝑝𝐷𝐷𝑅𝑅𝑑𝑑𝑑 − 𝑝𝑝𝐷𝐷𝐷𝐷 − 𝑝𝑝𝑓𝑓 � ∆𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑚𝑚𝑚𝑚𝑚𝑚[𝑠𝑠𝑖𝑖𝑑𝑑]
∑𝑖𝑖=1𝑖𝑖=365∑𝑖𝑖 𝑠𝑠𝑖𝑖𝑖𝑖 � ∑𝑖=𝑖𝑠𝑠𝑖=𝑖𝑒𝑒 𝑝𝑝𝐷𝐷𝑅𝑅𝑑𝑑𝑑 − 𝑝𝑝𝑓𝑓 � ∆𝐷𝐷𝑖𝑖𝑖𝑖𝑖
s.t. ∑𝑖𝑖=1𝑖𝑖=365 𝑠𝑠𝑖𝑖𝑖𝑖 ≤ 𝑀𝑀 ∀𝑖𝑖 ∈ 𝑁𝑁
Solution Greedy algorithm: compare today’s potential profit of each CPP participant against the highest X values of the rest of the year (X is the remaining quota)
Simple comparison with a threshold (a function of real-time price)
𝑠𝑠𝑖𝑖𝑖𝑖 = �1, 𝑖𝑖𝑖𝑖 𝑝𝑝𝐷𝐷𝑅𝑅𝑑𝑑𝑑 − 𝑝𝑝𝐷𝐷𝐷𝐷 − 𝑝𝑝𝑓𝑓 > 00, 𝑖𝑖𝑖𝑖 𝑝𝑝𝐷𝐷𝑅𝑅𝑑𝑑𝑑 − 𝑝𝑝𝐷𝐷𝐷𝐷 − 𝑝𝑝𝑓𝑓 ≤ 0
∀𝑖𝑖 ∈ 𝑁𝑁, 𝑑𝑑 = 1, . . , 365
Similar to the solution of CPP problem
Copyright 2016 Fujitsu Laboratories of America, Inc.8
Simulation settings
Simulation parameters are based on real-world data. Real statistical data from PJM: DA and RT price and energy procurement
data from Dominion Virginia Power (corresponding to DOM zone in PJM) Real DR program settings from utilities
• CPP: Dominion Virginia Power Smart Pricing [28], • PTR: Duke Energy Peak Time Rebate Pilot [29], and • DLC: Dominion Virginia Smart Cooling program
Model participant’s load, 𝐷𝐷𝑖𝑖𝑖𝑖𝑖, and curtailment behavior, ∆𝐷𝐷𝑖𝑖𝑖𝑖𝑖, based on simple mathematical models.
Copyright 2016 Fujitsu Laboratories of America, Inc.9
𝐷𝐷𝑖𝑖𝑖𝑖𝑖 = �𝑚𝑚 � 𝒩𝒩 𝐿𝐿𝑖𝑖𝑖, 𝑏𝑏 � 𝐿𝐿𝑖𝑖𝑖 , if 𝑖𝑖 is not on vacation~ 𝑈𝑈 0.8 � 𝐷𝐷𝑣𝑣𝑣𝑣𝑣𝑣 ,𝐷𝐷𝑣𝑣𝑣𝑣𝑣𝑣 , if 𝑖𝑖 is on vacation
Program CPP PTR DLC∆𝐷𝐷𝑖𝑖𝑖𝑖𝑖
(𝑛𝑛𝑛𝑛−𝑣𝑣𝑣𝑣𝑣𝑣) 𝑈𝑈(𝑘𝑘ℓ , 𝑘𝑘𝓊𝓊 ) � 𝐷𝐷𝑖𝑖𝑖𝑖𝑖 𝑈𝑈(𝑘𝑘ℓ , 𝑘𝑘𝓊𝓊 ) � 𝐷𝐷𝑖𝑖𝑖𝑖𝑖 𝑈𝑈(∆𝐷𝐷ℓ ,∆𝐷𝐷𝓊𝓊 )
Parameters 𝑘𝑘ℓ(𝐶𝐶𝐶𝐶𝐶𝐶) 𝑘𝑘𝓊𝓊
(𝐶𝐶𝐶𝐶𝐶𝐶) 𝑘𝑘ℓ(𝐶𝐶𝑅𝑅𝐷𝐷) 𝑘𝑘𝓊𝓊
(𝐶𝐶𝑅𝑅𝐷𝐷) ∆𝐷𝐷ℓ ∆𝐷𝐷𝓊𝓊Value 0.3 0.8 0.1 0.4 2 2.5
Performance comparison of 3 programs PTR
Compare with baseline (no PTR)
Show increased profit
Copyright 2016 Fujitsu Laboratories of America, Inc.10
CPP
Compare with: DOM: utility’s real
events Price: broadcast basis
Proposal reduces the loss significantly
DLC Compare with: DOM: utility’s real
events Price: broadcast basis
Proposal reduces the loss
DLC program is usually applied for ancillary service
Video: Residential Demand Response
Copyright 2016 Fujitsu Laboratories of America, Inc.11
https://drive.google.com/file/d/0B3okJ_ljkYZUU1E4ai1VaGhCWG8/view?usp=sharing
Conclusions and future works Propose optimal DR resource control for utilities to deal with
multiple, heterogeneous DR programs Presented mathematical models for popular DR programs and the
computationally-lightweight solutions for optimization problems
Simulation results show the proposed solution either increase the profit or reduce the loss significantly
As the next step, we plan to find an opportunity to evaluate the effectiveness in a real-world setting.
Future works: Explore the bidding strategies when considering DR operationsCombine coupons with DR incentive: partner with merchants to jointly
offer coupons
Copyright 2016 Fujitsu Laboratories of America, Inc.12
Copyright 2016 Fujitsu Laboratories of America, Inc.13
Support Multiple DR ProgramsDirect Load Control (Smart Thermostat) Peak Time RebateDynamic Pricing
/ Critical Peak Pricing
• Initialization of a DR event can be constraint driven (e.g. emergency) or price driven (e.g. high wholesale electricity price)
Copyright 2016 Fujitsu Laboratories of America, Inc.14