Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 47

PhD Thesis Review Seminar

School of Information Technology and Electrical Engineering (ITEE)

Photovoltaic Variability and Uncertainty


Characterization and Management Strategies

Saeid Veysi Raygani

Advisors: Dr Daniel Martin


Dr Michael Forbes
Contents
1. Introduction
 Problem Statement, and Thesis Objectives

2. Solar Generation Characterization


 Solar Predictive Models
 Solar Characterization and Aggregation Analysis

3. Robust Unit Commitment (RUC)


 RUC with Characterized Solar PV Systems

4. Robust Risk-Averse Unit Commitment


 Unit Commitment under Worst-case Conditional Value-at-risk

5. Conclusion and Recommendation for Future Research

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 2


Contents
1. Introduction
 Problem Statement, and Thesis Objectives

2. Solar Generation Characterization


 Solar Predictive Models
 Solar Characterization and Aggregation Analysis

3. Robust Unit Commitment (RUC)


 RUC with Characterized Solar PV Systems

4. Robust Risk-Averse Unit Commitment


 Unit Commitment under Worst-case Conditional Value-at-risk

5. Conclusion and Recommendation for Future Research

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 3


Introduction
Generation Mix: World 2017
30000000
Coal
25000000 Natural gas
Hydro
Electricity Generation 20000000 3% 2%
2% 0% 0% 0% 0%
Nuclear
4%
by Source, World Wind

TWh
15000000 Oil
1990-2017 10000000
10% 38% Biofuels
Solar PV
5000000 Waste
16% Geothermal
0 Other sources
1990 Other sources
Solar thermal 1995 2000
Geothermal 2005
Waste 2010
Solar PV 2015
Biofuels Oil 2017
Wind Solar thermal
23%
Nuclear Hydro Natural gas Coal

300000 Generation Mix: Australia 2018


250000 Coal
2% 1% 0% Natural gas
Electricity Generation by 200000
4% Hydro
6% Wind
Source, Australia 1990-
TWh

150000 6% Solar PV
2018 Oil
100000 Biofuels
Solar thermal
50000 21% 60% Geothermal

0 3 GW
1990 1995 2000 2005 2010 2015 2018 of Large-scale Solar PV
Geothermal Solar thermal Biofuels Oil Solar PV Wind Hydro
Natural gas Coal (26% of solar PV Mix: 1%
of the total generation
Mix)
Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 4
Problem Statement
Source of Uncertainty in Power Systems Uncertainty Managed in Different time-scales

 Intermittency of Loads  Planning Procedures (Long- or Short- Term Forecast)

 Power Lines Outages  Day-ahead Unit Commitment (UC) (Usually 1 hour)

 Unavailability of Generation Units  Economic Dispatch (ED) (5-10 Minutes to Hours)

 Intermittency of Variable Energy Resources  Regulation Reserves (Seconds to Minutes)

How to Manage Interconnection of VERs?

 Generation, Distribution and Transmission Expansion

 Balancing Generation Portfolio

 Utilising Energy Storages

 Load Shedding and Solar Curtailment

 Spatial Displacement of VERs

 Changes in Current Procedures: UC, ED, LF, AGC

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 5


Thesis Objective
Design, Analysis, and Validation of UC Strategies, with Solar PV Systems
1. Characterization of PV Uncertainty and Variability under Various Weather Conditions:

 Categorizing PV Uncertainty, Analyse Energy Level and Performance of PVs

 Characterizing Solar Ramp Rates, and Assessment of the Effects of Aggregation

2. Development and validation of UC methods with grid-connected PV:

 Inclusion of PV Characterization Model into a Two-stage Security-constrained Unit Commitment Using a


Robust Optimization Technique.

 Consolidating the Risk Management Concept and Robustness in the Unit Commitment with Intermittent
Solar PV Systems

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 6


Contents
1. Introduction
 Background, Problem Statement, and Thesis Objectives

2. Solar Generation Characterization


 Solar Predictive Models
 Solar Characterization and Aggregation Analysis

3. Robust Unit Commitment (RUC)


 RUC with Characterized Solar PV Systems

4. Robust Risk-Averse Unit Commitment


 Unit Commitment under Worst-case Conditional Value-at-risk

5. Conclusion and Recommendation for Future Research

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 7


Solar Generation Characterization
Solar Variability: Expected Changes Solar Uncertainty: Unexpected Changes

 Clear Sky Diurnal Shape  Generally due to Moving Clouds

 Prediction: Clear Sky GHI Models  Prediction: Sophisticated Forecasting



  GHI (w/m2) Solar Power on a Horizontal Surface  Short Term Forecasting of Environmental Factors Like
Cloud Type, Speed, Size, Direction, etc.
 GHI = DNI.cos(z)+Diffuse

 Ineichen and Perez Model

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 8


Solar Generation Characterization
Daily-Based Characterization of Solar Irradiance

Daily Uncertainty Index

  ∑ √ ( 𝐺𝐻𝐼 𝑡 − 𝐺𝐻𝐼 𝑡 − 1 )2+ ∆ 𝑡 2


𝑡 ∈𝑇
𝐷𝑈𝐼 =
∑ √ ( 𝐶𝑆𝐼 𝑡 − 𝐶𝑆𝐼 𝑡 −1 )2 +∆ 𝑡 2
𝑡 ∈𝑇

Daily Clearness Index


DAY DUI DCI
Clear 1≤DUI<2 0.5<DCI≤1
DCI 
 Measured solar irradiation Overcast 1<DUI<2 DCI≤0.5
 Calculated clear sky solar irradiation mild 2<DUI≤5 -
moderate 5<DUI≤10 -
high DUI>10 -

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 9


Solar Generation Characterization
Solar Generation Predictive Model
UQ site A: 17/05/2013
250
40
Ambient Temperature
measured
Measured Tm simulated Tm
35 Estimated Tm fixed Tm
200

30

Temperature (C)
150

Power(kW)
25

20
100

Pout  I SC VOC  FF 15

50
10

VOC
   
 N .VOC 0 . 1   Tm  T0  1   (Tm ) POAI / 1000 
  5
4 6 8 10 12 14 16 18 20 0
Hour 0 5 10 15 20
Hour

I SC  POAI .N S .I SC 0 . 1    Tm  T0   / 1000
 
UQ site A: 03/05/2013
300

FF  a  ( b  POAI  c ) / Ln( POAI )  Tm ( d  e  POAI  f / POAI )


measured
simulated Tm
250 Fixed Tm

 
Tm  POAI ea1 b1 .WS  Ta a1  3.45 & b1  0.057 200

Power(kW)
150

100

50

0
4 6 8 10 12 14 16 18 20
Hour

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 10


Solar Generation Characterization
Performance Factors Sunlight, Technology, and Temperature
Total  Energy  Produced( kWh ) Total   insolation (kWh ) P roduction   Factor  (PF )
 
CF=  
SF= E  PF =
System Rating  (kW )× Time Interval (24 hours) IrradianceSTC × Time Interval ( daylight   hours ) Sun   Factor  (SF )
Energy Performance Factor for UQ and Rockhampton
Capacity Factor for UQ and Rockhampton
1 28
15 Sun Factor for UQ and Rockhampton
0.6 15
UQ UQ Hours
UQ UQ Hours 26
Rockhampton Rockhampton Hours 14.5
0.25 Rockhampton Rockhampton Hours
14 0.5 0.8

Energy Performance Factor


14
24
13.5
0.2 0.4
Capacity Factor

Temprature (°C)
13 0.6
13 22

Sun Factor
Hours

Hours
0.15 0.3 12.5

0.4 UQ 20
12 12
0.2
Rockhampton
0.1 11.5
18
11 0.2
11 0.1
0.05 UQ Ambient Temp 16
10.5
Rockhampton AmbientTem
0 10
0 10 Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct 0 14
Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Month Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct
Month Month

Coal CC-NG
Gas
Nuclear Turbine Wind
Solar
Hydro Thermal
 
CF 70-85% 70-80% 90% 5%-85% 25%-35% 17-50% 19%-
24%

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 11


Solar Generation Characterization
Wavelet Decomposition of Power Output, UQ site

where, Pnorm is the normalised solar output, T is


the sampling time, t is time and

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 12


Solar Aggregation Characterization
PV Uncertainty In Australia
 Quantitative Analysis on PV Geography and Climatology
uncertainty
 Temporal Ramp-rate Analysis
 Existing studies are not applicable
to Australia
 Uncertainty of an Individual Plant
 Effects of Mitigation Via
Aggregation
 Sites locations are far apart to facilitate
capturing the climatic impacts on
mitigating PV variability through
aggregation

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 13


Solar Generation Characterization
  𝑁 𝑁
𝜎 =
𝑡𝑜𝑡𝑎𝑙
𝛥𝑡 √∑ ∑
𝑖=1 𝑗=1
𝑖
𝜎 𝜎 𝜌
𝛥𝑡
𝑗 𝑖, 𝑗
𝛥𝑡 𝛥 𝑡

Summer Season UQ plant


Reduction in Aggregate Uncertainty compared to UQ PV

Cumulative distribution functions of 1-min ramp rates


Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 14
Solar Generation Characterization
Analyzing the Effects of Aggregation on Site 1 and Site 2 at UQ, RR: 1 Min
PDF for Site 1 PDF for Site 2 PDF for two Sites
0.016 0.025 0.015
Winter,  = 0.0587 p.u Winter,  = 0.0448 p.u.
0.014 Winter,  = 0.0604 p.u Spring,  = 0.0838 p.u Spring,  = 0.0605 p.u
Spring,  = 0.0799 p.u 0.02 Summer,  = 0.103 p.u Summer,  = 0.0732 p.u
0.012 Fall,  = 0.0899 p.u Fall,  = 0.0616 p.u
Summer,  = 0.102 p.u
Fall,  = 0.0865 p.u 0.01
0.01 0.015

PDF
PDF

PDF
0.008

0.01
0.006
0.005
0.004
0.005
0.002

0 0 0
-0.3 -0.2 -0.1 0 0.1 0.2 0.3 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 -0.15 -0.1 -0.05 0 0.05 0.1 0.15
Ramp Rate (p.u) Ramp Rate (p.u) Ramp Rate (p.u)

Correlation
Normalised Three-Sigma Ramp Rate (%)
Coefficient (ρ)
Season Site 1 Site 2 Site 1 & Site 2 Site 1 & Site 2
Winter 18.1 17.6 13.5 -0.72
Spring 23.9 25.1 18.2 -0.73
Summer 30.5 30.8 22 -0.75
Fall 25.8 26.9 18.5 -0.76
Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 15
Solar Generation Characterization
Correlation Coefficient for 30-min RR in summer.

Correlation Coefficient for 30-min RR in winter.

Correlation coefficient of ramp rates versus distance  Substantial Reduction In Uncertainty Via Aggregation

   Correlation of Ramps Reduces when Distance Increases

: distance between site x and y  Different Climate Patterns Hugely Affect the Correlation
∆t : timescale
a=5.76, b=1.7 and c=0.5  1-min and 5-min Ramps Loose Correlation in Order of 1
Km And 20 Km
Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 16
Contents
1. Introduction
 Background, Problem Statement, and Thesis Objectives

2. Solar Generation Characterization


 Solar Predictive Models
 Solar Characterization and Aggregation Analysis

3. Robust Unit Commitment (RUC)


 RUC with Characterized Solar PV Systems

4. Robust Risk-Averse Unit Commitment


 Unit Commitment under Worst-case Conditional Value-at-risk

5. Conclusion and Recommendation for Future Research

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 17


Robust Unit Commitment
Traditional Unit commitment (UC)
 Inputs  Current Practice
 Set of Available Generations  AEMO Runs Half-hourly Pre-dispatch UC Considering 10%,
50%, 90% Forecast Error
 Day-ahead Forecast of Load and VERs (e.g. solar, wind et.c)
 In US, ISOs Conduct Reliability UC With Resolution Of One-
 Outputs hour by Allocating Additional Capacity

 The Commitment Decisions: When a unit starts and shuts down cost-  Issues
effectively while meeting the demand
 Deterministic: Lack of Robustness to Uncertainty: Security
 Purpose (Load Shed), Cost

 Inform GENCOs of sufficient spinning and operating reserves, and provide  Lack of Risk Measures
regional pricing for GENCOs
 PV Forecast Accuracy with 1-hour Resolution: 18%-60%
 Maintain Reliability and Security of Power System Operations

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 18


08/09/2020

Robust Unit Commitment


Traditional Unit commitment (UC)
Mixed Integer Programming (MIP)
Load Unit 3: Peaking (OC Gas Turbines, pumped hydro)
𝑁𝑇 𝑁𝐺 𝑁𝑇 𝑁𝐺 𝑁𝑇 𝑁𝐷
 
Min
{∑ ∑
𝑡=1 𝑖=1
ST𝑖, 𝑡 ⋅𝛼𝑖, 𝑡 +SD𝑖, 𝑡 ⋅ 𝛽 𝑖, 𝑡 +∑ ∑ 𝐶𝑖 (𝑃𝑖,𝑡 )+{∑ ∑ VoLL𝑑 ,𝑡 ⋅𝑟 𝑑 ,𝑡
𝑡=1 𝑖=1 𝑡=1 𝑑=1
} G
?
G
?
G
? Load Profile Unit 2: load following
(CC Gas Turbine, Steam Turbine)
 
 Transmission and Network Const.:
Unit 1: base load (oil or coal fired, nuclear, hydro
 Load Flow Balance (DC) Eq.
  Ci(Pi,t): Production cost,
CC Gas Turbine)
 Si,t: Start-up and Shutdown cost Time
 Min and Max Gens Const. 0 6 12 18 24
 : Startup and shutdown binary Var.
 Up-ramp and Down-tamp limits of Gens Load Unit 1 Unit 2 Unit 3
 VoLL: Value of Lost Load 500 On Off Off
 Spinning and Operating Reserves Const. 700 On On Off
 rd,t: Load shed 800 On On Off
 ON and Off states of Gens Const. 1100 On On On
1000 On On Off
 Line Flow limits 900 On On Off
800 On On Off
 Bus Angles Limits 600 On Off Off

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 19


Robust Unit Commitment
Recent Advances in UC Models Hierarchy of the Models
 Stochastic Unit Commitment (SUC)  Two-stage Optimization Models

 Probability Distribution of Uncertain Scenarios  Commitment Decision: “Here-and-now”

 Large Number of Appropriately Weighted Scenarios  The First Stage Al. commits cheaper gens

 Decomposition Techniques and Scenario Reduction Alg.  Second Stage Alg. (SSA), runs an Hourly Economic Dispatch

 High Computational Burden for Large-scale Power Systems  SSA Generates Infeasibility Cuts

 Robust Unit Commitment (RUC)  Multi-stage Optimization Models

 Minimizes Worst-case Scenario and Robust against uncertainty  Adjustable Commitment Decisions Over Stages

 Distribution-free uncertainty set  Captures Dynamics of the Uncertainty

 Conservative Choice of Uncertainty Set for Practical Applications  Complex and Intractable

 Not Optimal For Non-worst-case Scenarios

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 20


Robust Unit Commitment
Motivation to Improve the Classic Robust Unit Commitment (CRUC)
 Lack of an Accurate Model for Intermittent Solar Systems  
 un,t  uˆn,t 
 Issues with Classic Uncertainty Set:

ut  NU :
Ut (ut , ut , t ) : 
ˆ nNU un,t
  t , 



 
 Based on the Forecast of Uncertain Quantity, Not Historical Data u  uˆ  u , uˆ  u  n  NU 
 n,t  n,t n,t n,t n,t  

 An Instantaneous Model: Does Not Present a Well-defined Bound for 1


Uncertainty: 0.9
Upper Bound
Actual generation
0.8 Lower Bound

 The Lower and Upper Bounds Proportional to the Inaccurate Day- 0.7

Solar Generation (p.u)


0.6
ahead Forecast 0.5

0.4

 Very Conservative for Some Time Instances and Not Enough for 0.3

0.2
Others 0.1

0
1 3 5 7 9 11 13 15 17 19 21 23
 Cannot Cover the Entire Region of Possible Solar Generation or Time (h)

Incorrectly can Exceed the Region

 Does Not Include an Individual and Overall Ramps of Uncertain Nodes

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 21


Robust Unit Commitment
Uncertainty Set for RUC with Characterized Solar Gen.

 DEI: ratio of the realization of solar energy  The uncertainty set for solar generations’ ramps
over the solar energy forecast
 pvt  pv  NPV : 
 t 1 
𝑝𝑣
 𝐷𝐸𝐼 = ∑ 𝑡 ∈𝑇
 
𝑗 ,𝑡
 
𝑗
∑^ 𝑝𝑣
𝑡 ∈𝑇
𝑗 ,𝑡
PVt  PVt 1 :  
 jNPV
pv j ,t  pv j ,t 1  k NPV 

 
j  NPV ,k [3,7] 
 DUI: ratio of the length of realized solar
generation over the solar forecast where  NPV   
jNPV j NPV
 j j  j , j , j, j   NPV

  ∑ √( 𝑝𝑣 𝑗 ,𝑡 − 𝑝𝑣 𝑗 ,𝑡 − 1 )2+ ∆ 𝑡2
𝑡 ∈𝑇
𝐷𝑈𝐼 𝑗 = 2 2
 Upper- and lower bounds of solar generation forecast
∑ √( ^
𝑝𝑣 𝑗 ,𝑡 − ^
𝑝𝑣 𝑗 ,𝑡 − 1 ) + ∆ 𝑡
pv j ,t  (1  eup )PVup
𝑡 ∈𝑇 j ,t

pv j ,t  (1  elow )PVlow
j ,t
 Days categorised as clear, overcast or highly-uncertain

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 22


Robust Unit Commitment
Uncertainty Set for RUC with Characterized Solar Gen.
2
  ∑ √( 𝑝𝑣 𝑗 ,𝑡 − 𝑝𝑣 𝑗 ,𝑡 − 1 ) + ∆ 𝑡2 DEI: Ratio of the Realization Of Solar Energy over Solar Energy Forecast
𝑡 ∈𝑇
𝐷𝑈𝐼 𝑗 = 2 2
∑ √( ^
𝑝𝑣 𝑗 ,𝑡 − ^
𝑝𝑣 𝑗 ,𝑡 − 1 ) + ∆ 𝑡 DUI: Ratio of The Length of Realized Solar Generation over Solar Forecast
𝑡 ∈𝑇
Type of Day DUI DEI
Clear

  ∑ ¿ 𝑝𝑣 𝑗 , 𝑡 − 𝑝𝑣 𝑗 , 𝑡 −1∨ ¿ Clear
 
𝑡 ∈𝑇
𝐷𝑈𝐼 𝑗= ¿
𝑝𝑣 𝑗 ,𝑡 − ^
∑¿ ^ 𝑝𝑣 𝑗 ,𝑡 − 1∨¿ ¿ Highly Uncertain Day
𝑡 ∈𝑇  
Overcast, Rainy or Snowy
xpj,t , xngj,t-1 ≥0  
Highly Uncertain Day
xpj,t - xngj,t-1 = pvj,t- pvj,t-1

xpj,t ≤MjB, xngj,t ≤Mj(1-B), B {0,1}


Overcast, Rainy or Snowy

Mj is the installed capacity of the solar gen


Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 23
Robust Unit Commitment
Two-stage Robust Unit Commitment Model

Traditional Unit commitment (UC)  “cTx”:Unit’s Start-up And Shutdown Costs

 “dTy”: Represents Dispatch Cost


min
x ,y
c T
x  d T
y  (1)
 “x”: Vectors of Binary Variables:
s.t. Ax  b, x binary (2)  shutdown, start-up and commitment states of Gens

 “y”: Vectors of Dispatch-related Variables:


Ex  F y  h (3)
 Generations and Reserve Levels
Gy  Hpv  Iu (4)
 Transmission Network Flows
 Phase Angle of Busses
 Load Sheds

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 24


Robust Unit Commitment
Two-stage Robust Unit Commitment Model

 ω(u,pv,x): The Set of Feasible Dispatch Solutions


 T
min
x 
c x  pvmax min d T
y  Under Overall Uncertainties For a Fixed Commitment
PV, uU y (u , pv , x )  Decision X and Realization Of Load and Solar
Generations

 (u, pv, x)   y : Ex  Fy  h, Gy  Hpv  Iu   The Second-stage (Dispatch) Cost Represented by


“max pv∈PV, u∈U min y∈ω(u, pv, x) dTy” Defines the Minimum
Dispatch Cost For a Fixed Commitment Decision
Ax  b, x binary under Worst-case Uncertainty Sets of U and PV

 This Representation of Dispatch-cost (Max-min) is


Non-linear and Not Suitable For LP Modelling

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 25


Robust Unit Commitment
Two-stage Robust Unit Commitment Model
 max pv∈PV, u∈U min y∈ω(u, pv, x) dTy can be
represented by the following bilinear eq.
 First step in the linearization of max-min function is
to write the dual form of the dispatch cost, (min Z ( x )  max λT (h  Ex)  ζ T (Iu  Hpv)
λ, ζ , u, pv
y∈ω(u, pv, x) d y ) as follows:
T

s.t.
P( x,u, pv )  max λ (h  Ex)  ζ (Iu  Hpv)
T T
λ, ζ
s.t. λT F  ζ T G  dT

λT F  ζ T G  dT λ  0, ζ free, pv  PV , u  U

λ  0, ζ free  The optimal solution Z*(x) is the extreme points of the bounded
polyhedral set of constraints involving variables λ and ζ and the
uncertainty sets of U and PV.

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 26


nNU un,t 

n,t  u , ˆ
u
n,t n,t  u  n  NU 
n,t  
  The non-concave bilinear function can be linearized using
linear lower approximation technique

 
 

“ζJ-1”, “uJ-1” and “pvJ-1” are the solutions passed from


the “J-1th’ iteration

  Bender’s cuts (dual form of the dispatch cost). the


extreme points of the dual polytope and below the
epigraph of Z(x).
Robust Unit Commitment
Test Results
 IEEE 118-Bus Test system, 54 Gen., 12 large-scale PV (150MW), 91 loads, peak load: 5,400 MW

 CPLEX solver, Convergence parameters, ε=1e-3 and Δ=5e-5

CRUC worst-case cost for the load uncertainty of CRUC worst-case cost with PV connected
5% and without PV Connection
1.4 1.5
Worst-Case Cost
1.45 Day-Ahead Forecast Cost
1.35
1.4
1.3

Commitment Cost ($MM)


1.35

Worst-Case Cost ($MM)


PV
t =1
1.25 1.3
PV
t =0.5
PV 1.25
Worst-Case Cost Day-Ahead
1.2 t =0.0

ΔtLoad 1.15
1.2

($) Forecast Cost 1.1


1.15

1.1

0 1,223.3k 1,223.3k 1.05 1.05

0.455 1.297.2k 1,262.7k 1


0 0.5
Load
1 1.5
1
0 20 40 60
PV Forecast Error (%)
80 100

0.91 1,360,4k 1,296.2k


t

≥1 1,374,5k 1,305.8k

Worst-case cost for the CRUC cost region ud,t = ud,t=5%,

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 28


Robust Unit Commitment
Test Results
Lower and Upper Bounds DUI DEI
 Highly Uncertain Day ෞ
𝑝𝑣𝑗 ,𝑡 = 𝑝𝑣 ത
തത

𝑗 ,𝑡 & 𝑝𝑣 ෞ
𝑗 ,𝑡 = 0.1𝑝𝑣𝑗 ,𝑡 0.0≤DUI≤2.2 0.0≤DEI≤1.1
DRUC and CRUC Comparison ෞ
𝑝𝑣𝑗 ,𝑡 = 0.7𝑝𝑣 ത
തത

𝑗 ,𝑡 & 𝑝𝑣 ෞ
𝑗 ,𝑡 = 0.1𝑝𝑣𝑗 ,𝑡 0.3≤DUI≤2.2 0.3≤DEI≤1.1

𝑝𝑣𝑗 ,𝑡 = 0. 5𝑝𝑣 ത
തത

𝑗 ,𝑡 & 𝑝𝑣 ෞ
𝑗 ,𝑡 = 0.1𝑝𝑣𝑗 ,𝑡 0.5≤DUI≤2.2 0.5≤DEI≤1.1

1.4 1.4
DRUC: 0.5DUI2.2, 0.5DEI1.1
1.35 -- PV
CRUC: u j,t= u-- j,t =50%, t =1
1.35 1.35

Day-ahead Forecast Cost ($MM)


1.3

Worst-Case Cost ($MM)


Worst-Case Cost ($MM)

1.3 1.3
1.25

1.25 1.2 1.25


DRUC: 0.0 DUI 2.2, 0.0 DEI 1.1
DRUC: 0.3DUI2.2, 0.3DEI1.1 1.15 DRUC: 0.3 DUI 2.2, 0.3 DEI 1.1
1.2 DRUC: 0.5DUI2.2, 0.5DEI1.1 1.2 --
CRUC: u j,t = u-- j,t =100%, t
PV
=1
-- PV
CRUC: u j,t= u-- j,t =50%, t =1 1.1

1.15 1.15
1.05

1.1 1 1.1
0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5
Load Load
t t t
Load

Worst-case cost Day-ahead Cost adding solar ramps constraints

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 29


Robust Unit Commitment
Worst-Case Scenarios, Conclusion
 An accurate and reliable two-stage RUC developed relying
on the clear-sky- and overcast solar forecasts.

 The solar generation categorized into a clear-, overcast- or


highly uncertain-day.

 The numerical results on the IEEE 118-bus test systems


verified that the utilized approach is substantially efficient
in cost management for all types of days.
Highly Uncertain Day: CRUC Highly Uncertain Day: DRUC
 The RUC can take advantage of the proposed approach
subject to the availability of historical data for DEI and DUI

 Inaccurate level of DEI and DUI might have significant cost


implications.

Clear Day: DRUC Overcast Day


Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 30
Contents
1. Introduction
 Background, Problem Statement, and Thesis Objectives

2. Solar Generation Characterization


 Solar Predictive Models
 Solar Characterization and Aggregation Analysis

3. Robust Unit Commitment (RUC)


 RUC with Characterized Solar PV Systems

4. Robust Risk-Averse Unit Commitment


 Unit Commitment under Worst-case Conditional Value-at-risk

5. Conclusion and Recommendation for Future Research

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 31


Robust Risk-Averse Unit Commitment
Motivations
 Stochastic Unit Commitment (SUC)  Distributionally Robust Unit Commitment (DRUC)

 Does not include ambiguity set  Emerging Framework

 Chance-constraint UC (CCUC):  Ambiguity Set Utilizing the Moment Information

 Incorporate The Risk Paradigm In the UC  UC-WCVaR: The Developed Approach

 Risk-averse control measures  Incorporates risk management concept and robustness

 Requires Exact Information of Probability Distribution  Minimizes CVaR over an ambiguity set over partial inf.

 Robust Unit Commitment (RUC) UC Model Robustness Risk Paradigm


SUC - -
 Distribution-free Uncertainty Set CCUC - 
RUC  -
 Practical Implantation is Challenging: Very Conservative DRUC  -
UC-WCVaR  

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 32


Unit Commitment under Worst-case Conditional Value-at-risk
Conditional Value-at-risk (CVaR)

  The amount of tail risk of the cost beyond value-at-risk (), i.e., the average value of the highest
costs

  1 +¿ 𝑝 ( 𝑧) 𝑑𝑧
𝐹 ( 𝑧 , 𝑦 )𝛼 =𝛽 + ∫ ( 𝑓 ( 𝑧 , 𝑦 )− 𝛽 ) ¿
1 −𝛼 − ∞
probability distribution function (pdf)

𝐶𝑉𝑎𝑅
  𝛼 =𝑚𝑖𝑛 𝐹 ( 𝑧 , 𝑦)𝛼
𝛽
(1- )

Var 
CVar 

 , y is the decision variable, and p(z) represents the probability


Cost
density function of random variable z. is the cost threshold for
which the minimum cost is attained. In this case, is the left
endpoint of .

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 33


Unit Commitment under Worst-case Conditional Value-at-risk
Definition of UC-WCVaR

  WCVaR for fixed with respect to is defined as

 
min ¿ ¿   is the average value of the highest dispatch
𝑥 costs.
∞  z represents the intermittency in solar
  1 𝐓 +¿ 𝑝 ( 𝑧 ) 𝑑𝑧 generation forecast
𝐶𝑉𝑎𝑅 𝛼 =𝑚𝑖𝑛 𝛽+ ∫ ( 𝐝 𝑦−𝛽 ) ¿
𝛽 1− 𝛼 −∞
𝑠  .𝑡 .  max-min problem, integral equation over a probability
𝐄
  𝑥+𝐅 𝑦 ≥ 𝐡
distribution.
𝐆
  𝑦+𝐇 𝑧=𝐈 𝑢
 cannot be directly solved using linear programming solvers

 𝑨 𝑥≥ 𝐛 , 𝑥 binary

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 34


Unit Commitment under Worst-case Conditional Value-at-risk
Robust Counterpart of UC- WCVaR:A Mixed-integer Linear Programming
  Is Convex In ,Concave (Linear) In P(z) and Has a Strong Max-min Property

 
𝑚𝑖𝑛 ¿ ¿ Lifted Ambiguity Set
𝑥, 𝛽   ( 𝑧 , 𝑣 ) P                 
𝒦 𝒦
P [( 𝑧 , 𝑣 )∈ 𝑃𝑉 ´ ] =1
Ambiguity Set 𝒢= {P ∈ 𝑀
´ ( R × R ¿| 1 2

´ 𝑧 ] =𝜇 }
E P [𝐺

ℱ =¿
  EP [ 𝑣 ] ≤𝜎

  ´ = {( 𝑧 , 𝑣 ) ∈ R 𝒦 × R 𝒦 ∨ 𝑧 ∈ 𝑃𝑉 , 𝑔(𝑧 )≤ 𝑣 }
𝑃𝑉 1 2

  ´ = {( 𝑧 , 𝑣 ) ∈ R 𝒦 × R 𝒦 ∨ 𝐂
𝑃𝑉 1 ´ 𝑧+𝐃
2
´ 𝑣≤𝑞}

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 35


Unit Commitment under Worst-case Conditional Value-at-risk
Dual form of the UC-WCVaR
  min
𝑥 , 𝛽, 𝑟,𝑠,𝜏
( 𝒄𝑻 𝑥 + 𝛽 +𝑟 +𝑠 𝑇 𝜇+𝜏 𝑇 𝜎 ) 𝑧=𝑧 0 + ∑ 𝑧 𝑖
 
𝑖 ∈𝒦 1
𝑠  .𝑡 .
  : Look-ahead/day-ahead solar generation forecast
1
𝑟  +𝑠 𝑧 +𝜏 𝑣 ≥
𝑇 𝑇 𝐓
(𝐝 𝑦 − 𝛽 )
1− 𝛼
𝐝  𝐓𝑦 −𝛽≥0 Intractable 𝑦 ( 𝑧 ,𝑣 ) = 𝑦 0 + ∑ 𝑦 1𝑖 𝑧𝑖 + ∑ 𝑦2𝑗 𝑣 𝑗
 
𝑖∈ 𝒦 1 𝑗∈ 𝒦 2
𝐄
  𝑥+𝐅 𝑦 ≥ 𝐡
𝐆  𝑦+𝐇 𝑧− 𝐈 𝑢=0  : decision variables for look-ahead/day-ahead forecast

 ´ 𝑧 + 𝐃
𝐂 ´ 𝑣 ≤𝐪 “” recourse constants associated with variable
 𝑨 𝑥≥ 𝐛, 𝑥 binary  
recourse constants associated with auxiliary variable
  r,,

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 36


Unit Commitment under Worst-case Conditional Value-at-risk
Adaptive Robust Optimization Problem With Lifted Decision Rule (LDR)

 m ∈ ( 𝒄 𝑻 𝑥 + 𝛽 +𝑟 + 𝑠𝑇 𝜇 + 𝜏 𝑇 𝜎 )
𝑥 , 𝛽 ,𝑟 , 𝑠 , 𝜏
𝑠  .𝑡 .
 (1− 𝛼 ) 𝑟 +𝑠 𝑇 𝑧 𝑖+ 𝜏
𝑇
𝑣 𝑗 + 𝛽 −𝐝
𝐓 0 1
𝑦 + ∑ 𝑦 𝑖 𝑧 𝑖 + ∑ 𝑦 𝑗 𝑣 𝑗 ≥0
2
( ∑ ∑ ) ( )
𝑖 ∈𝒦 1 𝑗 ∈ 𝒦2 𝑖∈ 𝒦 1 𝑗∈ 𝒦 2

𝐓 0 1 2
𝐝  𝑦 + ∑ 𝑦 𝑖 𝑧 𝑖 + ∑ 𝑦 𝑗 𝑣 𝑗 − 𝛽 ≥ 0 𝐂 ´ ∑ 𝑧 𝑖+ 𝐃´ ∑ 𝑣 𝑗 ≤ 𝐪
( 𝑖 ∈ 𝒦1 𝑗 ∈ 𝒦2 ) 𝑖 ∈𝒦 1 𝑗 ∈ 𝒦2

 𝐄 𝑥+𝐅 0 1 2
𝑦 + ∑ 𝑦 𝑧 + ∑ 𝑦 𝑣 𝑗 ≥𝐡
( ) 𝑖∈ 𝒦 1
𝑖 𝑖
𝑗∈ 𝒦 2
𝑗   r,,

0 1 2 0
𝐆 𝑦 + ∑ 𝑦 𝑧 + ∑ 𝑦 𝑣 +𝐇 𝑧 + ∑ 𝑧 − 𝐈 𝑢=0
 
( ) (
𝑖∈ 𝒦 1 )
𝑖 𝑖
𝑗 ∈𝒦 2
𝑗 𝑗
𝑖∈ 𝒦 1
𝑖  𝑨 𝑥≥ 𝐛, 𝑥 binary

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 37


Unit Commitment under Worst-case Conditional Value-at-risk
Dual form of UC-WCVaR with LDR
 
min ( 𝒄 𝑻 𝑥+𝛽 +𝑟+𝑠𝑇 𝜇+𝜏 𝑇 𝜎 )
𝑥 , 𝛽 , 𝑟 , 𝑠 , 𝜏 ,𝜉 𝑙 ,𝜉 𝑚 ,𝜉 𝑜 , 𝜉 𝑝
𝑠  .𝑡 .
𝑟  (1 −𝛼 )+( 𝛽 − 𝐝 𝐓 𝑦 0) −𝜉 𝑇𝑙 𝐪 ≥𝟎 𝐝 𝐓 0 𝑇
  𝑦 − 𝛽 −𝜉 𝑚 𝐪 ≥ 0 𝐄
0 𝑇
  𝑦 0+ 𝐇 𝑧 0 − 𝐈 𝑢 − 𝜉𝑇𝑝 𝐪=0
  𝑥+ 𝐅 𝑦 − 𝐡− 𝜉 𝑜 𝐪 ≥ 0 𝐆
  𝑦 11 1

[ ] 𝑦1 1
 

[ ] 𝑦1 1
 
𝑇
´ =− 𝐝
−𝜉 𝐂 𝑙
𝐓

1
𝑦𝒦

𝑦
2
1
𝑇
⋮ + 𝑠  (1 − 𝛼 ) 𝑇
´ =𝐝
−𝜉 𝐂 𝑚
𝐓

1
𝑦𝒦
2
1
𝑇 ´
−𝜉 𝑜 𝐂 = 𝐅 ⋮

𝑦
1
𝑦𝒦[ ]
2
1
 

 
𝑇 ´
𝑦1

[]
− 𝜉 𝑝 𝐂 =𝐆 ⋮ + 𝐇  
1
𝑦𝒦
𝑦1
2
1

[ ]
  𝑦1  
𝑇
´ =− 𝐝
−𝜉 𝐃 𝑙
𝐓

[ ]1

2
𝑦𝒦
𝑇
⋮ + 𝜏  (1− 𝛼 )
2
 
−𝜉 𝑇𝑚 𝐃=
´ 𝐝𝐓
[ ]

2
𝑦𝒦 2
−𝜉 𝐃
𝑇
´ =𝐅 ⋮
𝑜
2
𝑦𝒦 [ ]
1

2
𝑇 ´
−𝜉 𝑝 𝐃 =𝐆 ⋮
2
𝑦𝒦 2

𝜉  𝑙 ,𝜉 𝑚 , 𝜉 𝑜 , 𝜉 𝑝 ≥ 0   ,,

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 38


Unit Commitment under Worst-case Conditional Value-at-risk
Test Results

ℱ =¿
1

0.9  
0.8

0.7
Solar Generation (pu)

0.6

0.5

0.4

0.3

0.2

0.1

0
0 2 4 6 8 10 12 14 16 18 20 22 24
Time (h)

  𝑧 ∈ R 𝒦 ,  𝑣 ∈ R 𝒦              
1 2

𝑧 𝑚𝑖𝑛 ≤ 𝑧 ≤ 𝑧 𝑚𝑎𝑥
Histogram of solar generation in
number of cloudy days: On each box,
the central mark “*” represents “μ” and
´ ( R𝒦
𝒢= { P ∈ 𝑀 1
𝒦
× R ¿|2
{
P 0 ≤ 𝑣 ≤ 𝑣 𝑚𝑎𝑥 =1
𝑧 − 𝑣 ≤0
E P [ 𝑧 ]= 0
} } 
the top and bottom edges represent “μ
+ σ” and “μ – σ”, respectively. The EP [ 𝑣 ] ≤𝜎
whiskers extend to the minimum and
maximum values

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 39


Unit Commitment under Worst-case Conditional Value-at-risk
Modelling a source of uncertainty in the network
  ,, , and
𝑃  𝑡 ( 𝑧 ,𝑣 )= ∑ 𝑃𝑔, 𝑡0 + ∑ 𝑃 𝑔, 𝑡1 𝑧+ ∑ 𝑃 𝑔 ,𝑡 2 𝑣
Time (h) Start-up (αi,t)
Shutdown ∀ 𝑔∈ 𝑁𝐺 , ∀ 𝑡 ∈ 𝑁𝑇 ∀ 𝑔∈ 𝑁𝐺 ,∀ 𝑡 ∈𝑁𝑇 ∀ 𝑔∈ 𝑁𝐺, ∀ 𝑡 ∈𝑁𝑇
(βi,t)
140
G12 G25 G26 300
0.035
UC-WCVaR =0

120
G49 G54 G65

Probability distribution function (pdf)


0.03
00:00 - 250
G66 G69 G87 100
0.025
G92 G99 G100

Frequency
200 80
P(z) (MW)
PG10(z) 0.02
06:00 G80 - PG25(z)
150
PG61(z) 60
0.015
07:00 G10 - PG69(z)
PG89(z)
08:00 G89 - PG92(z)
0.01 40
100 PG100(z)
PG59(z)
09:00 G59 G61 - pdf(z) 0.005 20
50
18:00 G18 G56 G77 -
0 0
-150 -100 -50 0 50 100 150 1.194 1.196 1.198 1.2 1.202 1.204 1.206
19:00 G70 - z (MW) Unit Commitment Cost ($MM)
20:00 G104 -
21:00 - G59
22:00 - G61
G18 G56
23:00 -
G70 G77 G89
Generation output and pdf of the solar
intermittency, for α=0 at hour 09:00. Histogram of UC cost over 1000 samples and
UC cost of UC-WCVaRα=0.

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 40


Unit Commitment under Worst-case Conditional Value-at-risk
1.22
1.22 UC-WCVaR

1.215
1.215
UC-WCVaR UC-WCVaRα=99% cost

Unit Commitment Cost ($MM)


 1.21
Unit Commitment Cost ($MM)

1.21

1.205
UC-WCVaR
  α cost at
1.205
is 0.27% more than
1.2
1.2 and increases by optimal
1.195
1.195
One PV 0.09% and 0.82% 1.19
Two PV commitment
1.19
compared with 1.185
decision without
1.185

1.18
1.18
90% 95% 99%
solar PV system
90% 95% 99% 

140
140
c Tx + VaR90%
120 UC-WCVaR90% c Tx + VaR95%
120
UC-WCVaR95%
100
UC-WCVaRα=90% cost 100 UC-WCVaRα=95% cost
One PV
Frequency

80
matches the highest matches the worst

Frequency
80
60
Two PV
(1- ) = 10%
10% UC costs 60
5% cost derived
40 (1- ) = 5%

20
derived from 1000 40
from 1000 samples
0
samples (0.002% 20
(0.07% accuracy)
1.194 1.196 1.198 1.2 1.202 1.204 1.206
Unit Commitment Cost ($MM) accuracy) 0
1.188 1.19 1.192 1.194 1.196 1.198 1.2
Unit Commitment Cost ($MM)
1.202 1.204

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 41


Unit Commitment under Worst-case Conditional Value-at-risk
Comparison of RUC and UC-WCVaR, Conclusion
1.24
UC-WCVaR

1.34
 The RUC commitment cost is very conservative
compared with the UC-WCVaRα
Unit Commitment Cost ($MM)

1.32

1.22 1.3 RUC Worst-Case Cost

RUC Cost ($MM)


RUC Forecast Cost
1.28

1.26  The UC-WCVaRα=99% cost is significantly lower


1.2
1.24

1.22
than RUC cost for PV forecast errors of 40%-100%
1.2 by 7.6%-8.2% ($100k-$110k)
1.18
90% 95% 99% 1.18
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

PV Forecast Error (%)

 Multi-stage Model Fits With Limitation of


Computational Resources:

No. Max Solution No.


UC-WCVaRα cost RUC cost stages Time variables

 “Here-and-now” Model Consumes Extensive 1 1 hour 947,459


Computation Time 4 18 min 460,559

 Utilising Four PC Cores (threads) Instead of one 12 5 min 244,159


Dropped the Solution Time from 2 Hours to 25 Min 24 2 min 40 sec 190,059

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 42


Contents
1. Introduction
 Background, Problem Statement, and Thesis Objectives

2. Solar Generation Characterization


 Solar Predictive Models
 Solar Characterization and Aggregation Analysis

3. Robust Unit Commitment (RUC)


 RUC with Characterized Solar PV Systems

4. Robust Risk-Averse Unit Commitment


 Unit Commitment under Worst-case Conditional Value-at-risk

5. Conclusion and Recommendation for Future Research

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 43


Main Contributions of the Thesis
 Characterization of Solar PV systems  Two-stage Robust UC with Solar PV systems

 Development of a solar PV predictive model that uses the GHI,  The Uncertainty Set For PV is based on the Uncertainty
DHI, DNI, temperature, and wind and Energy Levels, and Types of Days

 Solar Days Categorized Based on Level of Intermittency  The Model can be used by AEMO

 Extensive Analysis of Performance and Energy Metrics  The Solution Time of the Model is Less than 5 Min

 Temporal ramp-rate analysis in Australia (season-wise)  Robust Risk-averse UC (UC-WCVaR)

 Correlation Coefficients Between RRs Decreases with Increasing  This risk-averse Model Shapes the Cost Distribution
Distance
 Maintains the Secure Operation of Power Systems Under
 Solar Aggregation Reduces Uncertainty Uncertainty of Variable Resources

 In Australia, Solar Dispersion Reduces the Uncertainty More  The Risk-averse Model is Less Conservative than the
Effectively In Summer Classic RUC Model

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 44


Main Contributions of the Thesis
Saeid Veysi Raygani, Daniel Martin, Michael Forbes "Robust Risk-averse Unit Commitment
with Solar PV Systems” submitted on 27/12/2019 to IET Renewable Power Generation

“Robust unit commitment with characterised solar PV systems”. IET Renewable Power
Generation, 2018 Nov 2, DOI:10.1049/iet-rpg.2018.5176 Print ISSN 1752-1416, Online ISSN
1752-1424

"PV power output uncertainty in Australia” 2015 IEEE Power and Energy Society General
Meeting, 26-30 July, 2015, Denver, CO, USA

“Analysing the PV output variability and its mitigation through aggregation in Queensland,
Australia”, 2014 IEEE PESGM, 27-31 July, 2014, Washington, DC, USA

Variability and performance analysis of the PV plant at The University of Queensland”


AUPEC, 2013

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 45


Future Research
 Risk-averse robust unit commitment: “N-1” contingency criterion with wind and solar PV systems
 robust commitment under “N-1” contingency criterion in presence of the VERs
 SDP can be implemented to reduce the computation time

 Multi-stage robust unit commitment


 the commitment decisions are adjustable
 solar generations can also be considered as scheduled
 Developing tractable adaptive models that incorporate LDR and support sets

 Robust unit commitment with demand-side management


 Integrating demand response and unit commitment could effectively manage the operating cost and the
demand response strategies

 Adaptive data-driven robust unit commitment


 RUC suffers from a high level of conservatism and interpretability
 data-driven and scenario based models capturing the mean and covariance matrix of the RPGs and loads

Photovoltaic Variability and Uncertainty Characterization and Management Strategies 08/09/2020 46


Thank you
Saeid Veysi Raygani| RPEQ
School of ITEE
s.veysiraygani@uq.edu.au
0490 299 552

https://www.linkedin.com/in/saeid-veysi-raygani-65b52331/

CRICOS code 00025B

You might also like