optimizing sensor type and location for rapid restoration
TRANSCRIPT
![Page 1: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/1.jpg)
Optimizing Sensor Type and Location for Rapid Restoration of Power Grids
Lina Al-Kanj, PhD
Warren B. Powell
Operations Research and Financial Engineering DepartmentPrinceton University
Princeton, NJ
Federal Energy Regulatory Commission Technical ConferenceWashington, DC
![Page 2: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/2.jpg)
Outline
Problem Description
Probability Model for Grid Faults
Sequential Stochastic Optimization Model for Utility Crew Routing Multistage Lookahead Policy (Monte Carlo Tree Search)
Stochastic Optimization Model for Sensor and Protective Device Placements
2
![Page 3: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/3.jpg)
Problem Description – Power System
3
![Page 4: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/4.jpg)
Problem Description – Power System
4
![Page 5: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/5.jpg)
Problem Description – Power System
XX
X
X
5
![Page 6: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/6.jpg)
Problem Description – Power System
XX
X
X
6
![Page 7: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/7.jpg)
Problem Description – Power System
X
X
XX
X
X
7
![Page 8: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/8.jpg)
Problem Description - Emergency Storm Response
XX
X
X
X
X
8
![Page 9: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/9.jpg)
Problem Description - Emergency Storm Response
XX
X
X
X
X
9
![Page 10: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/10.jpg)
Problem Description - Emergency Storm Response
XX
X
X
X
X
10
![Page 11: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/11.jpg)
Problem Description - Emergency Storm Response
XX
X
X
X
X
11
![Page 12: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/12.jpg)
Problem Description - Emergency Storm Response
XX
X
X
X
X
12
![Page 13: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/13.jpg)
Problem Description - Emergency Storm Response
XX
X
X
X
callcall
X
call
call
call
call
call
13
![Page 14: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/14.jpg)
XX
X
X
X
X
Problem Description - Emergency Storm Response
call
call
callcall
call
call
call
14
![Page 15: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/15.jpg)
Known customers in outage
Unknown outages
Outage calls(known)
Network outages(unknown)
Storm
15
![Page 16: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/16.jpg)
How to create a belief about outages?
How to reconstruct the grid to minimize customer outage-minutes?
What is the value of a sensor/re-closer in terms of reducing customer outage-minutes?
What is the best place to put a sensor or re-closer?
What is the value of upgrades to SCADA systems?
What is the value of investments in reconfigurable grids?
Problem Description – Research Goals
16
![Page 17: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/17.jpg)
Outline
Problem Description
Probability Model for Grid Faults
Sequential Stochastic Optimization Model for Utility Crew Routing Multistage Lookahead Policy (Monte Carlo Tree Search)
Stochastic Optimization Model for Sensor and Protective Device Placements
17
![Page 18: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/18.jpg)
Exploiting the information from phone calls We have to blend the following…. What we knew before the phone calls came in – this is called the prior
belief.
The outage calls – this is called information.
…to produce the updated estimates of outage – this is called the posterior.
To compute this we have to use Bayes theorem:
Prob[lights-out calls | segment is out]Prob[ is out]Prob[segment is out|lights-out calls]Prob[lights-out calls]
l ll
Probability Model for Outage Identification
[1] L. Al-Kanj, B. Bouzaeini-Ayari and W. Powell, “A Probability Model for Grid Faults Using Incomplete Information”, IEEE Transactions on Smart Grids, July 2015.
18
![Page 19: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/19.jpg)
Exploiting the information from phone calls We have to blend the following…. What we knew before the phone calls came in – this is called the prior
belief.
The outage calls – this is called information.
…to produce the updated estimates of outage – this is called the posterior.
To compute this we have to use Bayes theorem:
Prob[lights-out calls | segment is out]Prob[ is out]Prob[segment is out|lights-out calls]Prob[lights-out calls]
l ll
The prior
Probability Model for Outage Identification
19
![Page 20: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/20.jpg)
Exploiting the information from phone calls We have to blend the following…. What we knew before the phone calls came in – this is called the prior
belief.
The outage calls – this is called information.
…to produce the updated estimates of outage – this is called the posterior.
To compute this we have to use Bayes theorem:
Prob[lights-out calls | segment is out]Prob[ is out]Prob[segment is out|lights-out calls]Prob[lights-out calls]
l ll
The prior
The information
Probability Model for Outage Identification
20
![Page 21: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/21.jpg)
Exploiting the information from phone calls We have to blend the following…. What we knew before the phone calls came in – this is called the prior
belief.
The outage calls – this is called information.
…to produce the updated estimates of outage – this is called the posterior.
To compute this we have to use Bayes theorem:
Prob[lights-out calls | segment is out]Prob[ is out]Prob[segment is out|lights-out calls]Prob[lights-out calls]
l ll
The prior
The informationThe posterior
Probability Model for Outage Identification
21
![Page 22: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/22.jpg)
Exploiting the information from phone calls We have to blend the following…. What we knew before the phone calls came in – this is called the prior
belief.
The outage calls – this is called information.
…to produce the updated estimates of outage – this is called the posterior.
To compute this we have to use Bayes theorem:
Prob[lights-out calls | segment is out]Prob[ is out]Prob[segment is out|lights-out calls]Prob[lights-out calls]
l ll
The prior
The informationThe posterior
The conditional outage distribution
Probability Model for Outage Identification
22
![Page 23: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/23.jpg)
Exploiting the information from phone calls We have to blend the following…. What we knew before the phone calls came in – this is called the prior
belief.
The outage calls – this is called information.
…to produce the updated estimates of outage – this is called the posterior.
To compute this we have to use Bayes theorem:
Prob[lights-out calls | segment is out]Prob[ is out]Prob[segment is out|lights-out calls]Prob[lights-out calls]
l ll
The prior
The informationThe posterior
The phone call distribution
The conditional outage distribution
Probability Model for Outage Identification
23
![Page 24: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/24.jpg)
Probability Model – Prior Probabilities
0.2 0.3 0.6 0.6 0.6
0.2
0.2
0.2
0.2
0.3
0.4
0.4
0.5
0.5
0.50.5 0.5
0.5 0.5
0.5 0.5 0.5
0.6 0.6
0.50.50.5
0.5
0.50.5
0.7
0.50.5
0.50.5
0.5
0.50.5
24
![Page 25: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/25.jpg)
XX
X
X
X
X
Posterior Probabilities
callcall
call
call
call
call
call
0.26 0.38 0.78 0.6 0.6
0
0
0
0.02
0.03
0.04
0.04
0.05
0.064
0.0640.064 0.064
0.502 0.502
0.502 0.502 0.502
0.603 0.603
0.670.670.05
0.05
0.5030.5
0.76
0.5030.503
0.540.54
0.503
0.5030.503
30 Customers
30 Customers
30 Customers
25
![Page 26: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/26.jpg)
Outline
Problem Description
Probability Model for Grid Faults
Sequential Stochastic Optimization Model for Utility Crew Routing Multistage Lookahead Policy (Monte Carlo Tree Search)
Stochastic Optimization Model for Sensor and Protective Device Placements
26
![Page 27: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/27.jpg)
Grid Restoration Model
Objective: Develop an optimal policy that routes the utility truck in order to minimize the number of customers in outage at any point in time.
[4] L. Al-Kanj, W. Powell and B. Bouzaeini-Ayari, “The Information-Collecting Vehicle Routing Problem: Stochastic Optimization for Emergency Storm Response”, Submitted to Operations Research, http://arxiv.org/abs/1605.05711, 2016.
, … ,00 0
0 0
: Physical State
: Information State
: Knowledge State
?
?
?
??
? ?
Information-Collection utility truck routing
27
![Page 28: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/28.jpg)
Grid Restoration Model
Objective: Develop an optimal policy that routes the utility truck in order to minimize the number of customers in outage at any point in time.
[4] L. Al-Kanj, W. Powell and B. Bouzaeini-Ayari, “The Information-Collecting Vehicle Routing Problem: Stochastic Optimization for Emergency Storm Response”, In preparation.
Information-Collection utility truck routing
, … ,00 0
0 0
?
0
??
? ?: Physical State
: Information State
: Knowledge State
28
![Page 29: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/29.jpg)
Sequential Stochastic Optimization Model
Five fundamental elements of sequential stochastic optimization:
State - information capturing what we know at time t;, , where represents the physical state of the network.
Decision - captures the decision made at time ; Let be the policy that determines ∈ given .
Exogenous information - new information that arrives between 1and ;
Includes new arriving calls, travel time, repair time and outages.
Transition function - , , represents the evolution of the states
e.g., , , 1| , 1 0 in addition to .
Objective function - min ∑ , ;minimizes the number of customers in outage at any point in time.
29
![Page 30: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/30.jpg)
Sequential Stochastic Optimization Model
Cost function -
Figure 6: Objective function; Customer outage-minute is represented by the shaded areaunder the curve.
,
30
![Page 31: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/31.jpg)
Sequential Stochastic Optimization Model
min ,
Two fundamental strategies for designing policies [5]: Policy Search Lookahead Approximations Value function approximation
Direct lookahead
where
, ,
Optimization problem
[2] W. B Powell, A Unified Framework for Optimization under Uncertainty, Informs Tutorials, November 2016.31
![Page 32: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/32.jpg)
Lookahead approximations – Approximate the impact of a decision now on the future: An optimal policy (based on looking ahead):
Lookahead Approximations
*' ' ' 1
' 1( ) arg max ( , ) max ( , ( )) | | ,
t
T
t t x t t t t t t t tt t
X S C S x C S X S S S x
32
![Page 33: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/33.jpg)
Lookahead approximations – Approximate the impact of a decision now on the future: An optimal policy (based on looking ahead):
2a) Approximating the value of being in a downstream state using machine learning (“value function approximations”)
2b) Approximate lookahead models – Optimize over an approximate model of the future:
*1 1
1 1
( ) arg max ( , ) ( ) | ,
( ) arg max ( , ) ( ) | ,
arg max ( , ) ( )
t
t
t
t t x t t t t t t
VFAt t x t t t t t t
x xx t t t t
X S C S x V S S x
X S C S x V S S x
C S x V S
' ' , !' 1
( ) arg max ( , ) max ( , ( )) | | ,
T
LAt t t t tt tt t t t t
t t
X S C S x C S X S S S x
*' ' ' 1
' 1( ) arg max ( , ) max ( , ( )) | | ,
t
T
t t x t t t t t t t tt t
X S C S x C S X S S S x
Lookahead Approximations
33
![Page 34: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/34.jpg)
We can then simulate this lookahead policy over time:
. . . .
t 1t 2t 3t
The
look
ahea
dm
odel
The base model
Multistage Lookahead Approximations
34
![Page 35: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/35.jpg)
. . . .
t 1t 2t 3t
The
look
ahea
dm
odel
The base model
Multistage Lookahead Approximations
We can then simulate this lookahead policy over time:
35
![Page 36: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/36.jpg)
. . . .
t 1t 2t 3t
The
look
ahea
dm
odel
The base model
Multistage Lookahead Approximations
We can then simulate this lookahead policy over time:
36
![Page 37: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/37.jpg)
. . . .
t 1t 2t 3t
The
look
ahea
dm
odel
The base model
Multistage Lookahead Approximations
We can then simulate this lookahead policy over time:
37
![Page 38: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/38.jpg)
Multistage Lookahead Policy
∗ argmin ∈ , , ∈ , min, ∈ , , , , ,
, ∈ , … , ∈ , , | , … | , |
The optimal policy is computationally intractable, requiring approximations:
where , , , , , , … , 1
Discretizing the time, states and decision
Limiting the horizon from , to , Dimensionality reduction e.g., by limiting some variables (e.g., fixing the set of calls,
limiting the fault types and travel times)
Aggregating the outcome or sampling by using Monte Carlo sampling
38
![Page 39: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/39.jpg)
Monte Carlo Tree Search (MCTS)
MCTS is a recent research area; the first MCTS algorithm has been developed by Chang et al [2005].
MCTS growth [5]
[3] H. S. Chang, M. C. Fu, J. Q. Hu, and S. I. Marcus. An adaptive sampling algorithm for solving Markov decision processes. Operations research, 53(1):126139, 2005.[4] Kocsis Levente and Csaba Szepesvári. "Bandit based monte-carlo planning“, Machine Learning: ECML 2006. pp. 282-293.[5] R. Coquelin, Pierre-Arnaud and Munos, “Bandit Algorithms for Tree Search,” in Proc. Conf. Uncert. Artif. Intell. Vancouver,Canada: AUAI Press, 2007, pp. 67–74.[6] A. Coutoux, J. B. Hoock, N. Sokolovska, O. Teytaud, and N. Bonnard. Continuous upper confidence trees. In International Conference on Learning and Intelligent Optimization, pages 433445. Springer Bernlin Heidelberg, 2011.
MCTS biases the growth of the tree towards the most promising moves which decreases the search space [3].
MCTS mainly applied for deterministic problems but Coutoux et al [2011] extended it to stochastic problems based on double progressive widening.
39
![Page 40: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/40.jpg)
Monte Carlo Tree Search
Decision selection based on UCB (Upper Confidence Bounds for Trees) [7]: Visit count ofstate
Visit count of decision from state Exploitation Exploration
MCTS Construction Steps
[7] D.Auger, A. Couetoux, and O. Teytaud, “Continuous upper confidence trees with polynomial exploration consistency,” in Proc. ofthe European Conference on Machine Learning and Knowledge Discovery in Databases, (Prague, Czech Republic), pp. 194–209, 2013.
∗ argmax ∈ ,log
,
40
![Page 41: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/41.jpg)
Monte Carlo Tree Search – Convergence Theory
For deterministic MCTS, the UCB policy samples the actions infinitely often and Kocsis and Szepesvari [2006] exploit this to show that the probability of selecting a suboptimal action converges to zero at the root of the tree.
Auger et al. [2013] provides convergence results for stochastic MCTS with double progressive widening under an action sampling assumption. The asymptotic convergence of also MCTS relies on some form of “exploring every node infinitely often“.
In Jiang et al. [2017], we design a version of stochastic MCTS that asymptotically does not expand the entire tree, yet is still optimal!
[6] Kocsis Levente and Csaba Szepesvári. "Bandit based monte-carlo planning“, Machine Learning: ECML 2006. pp. 282-293.[7] D.Auger, A. Couetoux, and O. Teytaud, “Continuous upper confidence trees with polynomial exploration consistency,” in ProcThe European Conference on Machine Learning and Knowledge Discovery in Databases, (Prague, Czech Republic), pp. 194–209,[8] D. Jiang, L. Al-Kanj, W. B. Powell, “Monte Carlo Tree Search with Information Relaxation Dual Bounds”, In preparation for Operations Research, 2017.
41
![Page 42: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/42.jpg)
Simulation Policy
Generate sample path from Ω , that determines all random events (faults, travel time, repair time).
Utility truck should visit each fault once to repair it.
Objective: find the optimal route that minimizes the customer-outage minute.
In the worst case, computational complexity ! .
X
X
X
X
42
![Page 43: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/43.jpg)
Simulation Policy – Optimal Solution
Define the graph , with connection costs .
: subset of nodes visited by the truck.
, : customer-outage minute starting from node 1 and ending at node . .
: function returning the number of customers in outage after visiting the nodes of .
The cost of moving from node to node is, , ∗
Dynamic Program
For all ∈ do 1, , ∑For 3 to
For all Subsets of of size doFor all ∈ , 1 do
min ,
, min∈ ,
, ∗ Computational Complexity
243
![Page 44: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/44.jpg)
XX
X
X
X
X
Lookahead Policy
callcall
call
call
call
call
call
0.26 0.38 0.78 0.6 0.6
0
0
0
0.02
0.03
0.04
0.04
0.05
0.064
0.0640.064 0.064
0.502 0.502
0.502 0.502 0.502
0.603 0.603
0.670.670.05
0.05
0.5030.5
0.76
0.5030.503
0.540.54
0.503
0.5030.503
44
![Page 45: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/45.jpg)
XX
X
X
X
X
Lookahead Policy
callcall
call
call
call
call
call
0.26 0.38 0.78 0.6 0.6
0
0
0
0.02
0.03
0.04
0.04
0.05
0.064
0.0640.064 0.064
0.502 0.502
0.502 0.502 0.502
0.603 0.603
0.670.670.05
0.05
0.5030.5
0.76
0.5030.503
0.540.54
0.503
0.5030.503
45
![Page 46: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/46.jpg)
XX
X
X
X
X
Lookahead Policy
callcall
call
call
call
call
call
0.0 0.0 0.0 0.715 0.715
0
0
0
0.02
0.03
0.04
0.04
0.05
0.064
0.0640.064 0.064
0.502 0.502
0.502 0.502 0.502
0.603 0.603
0.670.670.05
0.05
0.5030.5
0.76
0.5030.503
0.540.54
0.503
0.5030.503
46
![Page 47: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/47.jpg)
XX
X
X
X
X
Lookahead Policy
callcall
call
call
call
call
call
0.0 0.0 0.0 0.715 0.715
0
0
0
0.02
0.03
0.04
0.04
0.05
0.064
0.0640.064 0.064
0.502 0.502
0.502 0.502 0.502
0.603 0.603
0.670.670.05
0.05
0.5030.5
0.76
0.5030.503
0.540.54
0.503
0.5030.503
47
![Page 48: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/48.jpg)
XX
X
X
X
X
Lookahead Policy
callcall
call
call
call
call
call
0.0 0.0 0.0 0.0 0.0
0
0
0
0.02
0.03
0.04
0.04
0.05
0.064
0.0640.064 0.064
0.502 0.502
0.502 0.502 0.502
0.603 0.603
0.670.670.05
0.05
0.5030.5
0.76
0.5030.503
0.540.54
0.503
0.5030.503
48
![Page 49: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/49.jpg)
XX
X
X
X
X
Lookahead Policy
callcall
call
call
call
call
call
0.0 0.0 0.0 0.0 0.0
0
0
0
0.02
0.03
0.04
0.04
0.05
0.064
0.0640.064 0.064
0.502 0.502
0.502 0.502 0.502
0.603 0.603
0.670.670.05
0.05
0.5030.5
0.76
0.5030.503
0.540.54
0.503
0.5030.503
49
![Page 50: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/50.jpg)
XX
X
X
X
X
Lookahead Policy
callcall
call
call
call
call
call
0.0 0.0 0.0 0.0 0.0
0
0
00.032
0.048
0.056
0.056
0.08
0.0604
0.06040.0604 0.0604
0.51 0.51
0.51 0.51 0
0.62 0.62
00.990.08
0.08
0.5030.5
0.76
0.5030.503
0.540.54
0.503
0.5030.503
50
![Page 51: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/51.jpg)
XX
X
X
X
X
Lookahead Policy
callcall
call
call
call
call
call
0.0 0.0 0.0 0.0 0.0
0
0
00.032
0.048
0.056
0.056
0.08
0.0604
0.06040.0604 0.0604
0.51 0.51
0.51 0.51 0
0.62 0.62
00.990.08
0.08
0.5030.5
0.76
0.5030.503
0.540.54
0.503
0.5030.503
51
![Page 52: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/52.jpg)
XX
X
X
X
X
Lookahead Policy
callcall
call
call
call
call
call
0.0 0.0 0.0 0.0 0.0
0
0
00.0
0.0
0.0
0.0
0.0
0.0603
0.06030.0603 0.0603
0.52 0.52
0.52 0 0
0.63 0.63
000.0
0.0
0.5030.5
0.76
0.5030.503
0.540.54
0.503
0.5030.503
52
![Page 53: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/53.jpg)
XX
X
X
X
X
Lookahead Policy
callcall
call
call
call
call
call
0.0 0.0 0.0 0.0 0.0
0
0
00.0
0.0
0.0
0.0
0.0
0.0603
0.06030.0603 0.0603
0.52 0.52
0.52 0 0
0.63 0.63
000.0
0.0
0.5030.5
0.76
0.5030.503
0.540.54
0.503
0.5030.503
53
![Page 54: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/54.jpg)
XX
X
X
X
X
Lookahead Policy
callcall
call
call
call
call
call
0.0 0.0 0.0 0.0 0.0
0
0
00.0
0.0
0.0
0.0
0.0
0.0603
0.06030.0603 0.0603
0.59 0
0.59 0
0 0
000.0
0.0
0.5030.5
0.76
0.5030.503
0.540.54
0.503
0.5030.503
0
54
![Page 55: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/55.jpg)
XX
X
X
X
X
Lookahead Policy
callcall
call
call
call
call
call
0.0 0.0 0.0 0.0 0.0
0
0
00.0
0.0
0.00.0
0.06
0.060.06 0.06
0 0
0 0
0 0
000.0
0.0
0
0
0
00
00
0
0.670.67
00.0
55
![Page 56: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/56.jpg)
Industrial Heuristics
Escalation Algorithm
Step 1. For each circuit do
Step 1a. Collect all calls and back trace to find the first node that is
common to all calls say node .
Step 1b. Send the truck to node and then back trace to the substation
to make sure that there is no upstream fault
Step 1c. From node , perform down tracing to reach the first segment
from which a call was initiated and place it in set .
Step 2. For each segment in do
Step 2a. Perform down tracing to cover all nodes that called.
56
![Page 57: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/57.jpg)
XX
X
callcall
call
call
call
Escalation Algorithm
call
call
X
X
X
57
![Page 58: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/58.jpg)
XX
X
callcall
call
call
call
Escalation Algorithm
call
call
X
X
X
58
![Page 59: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/59.jpg)
XX
X
callcall
call
call
call
Escalation Algorithm
call
call
X
X
X
59
![Page 60: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/60.jpg)
XX
X
callcall
call
call
call
Escalation Algorithm
call
call
X
X
X
60
![Page 61: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/61.jpg)
XX
X
X
X
callcall
call
call
call
Escalation Algorithm
call
call
X
61
![Page 62: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/62.jpg)
X
callcall
call
call
Escalation Algorithm
call
call
call
X
X
X
X
X
62
![Page 63: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/63.jpg)
X
callcall
call
call
Escalation Algorithm
call
call
call
X
X
X
X
X
63
![Page 64: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/64.jpg)
Numerical Results – Customer Outage-Minutes
36150
55110
Escalation algorithm
Lookahead policy
Performance metrics Total outage minutes SAIDI, CAIDI, SAIFI, …
64
![Page 65: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/65.jpg)
Studies
Possible questions:
What is the effect of higher call-in rates?
What is the best place to locate a sensor/protective device?
What is the value of a sensor vs. protective device?
65
![Page 66: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/66.jpg)
Numerical Results
Figure 10: Average customer outage-hours vs. MCTS budget for ten networks.
66
![Page 67: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/67.jpg)
Comparison to Industrial Heuristics
67
![Page 68: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/68.jpg)
Outline
Problem Description
Probability Model for Grid Faults
Sequential Stochastic Optimization Model for Utility Crew Routing Multistage Lookahead Policy (Monte Carlo Tree Search) Monte Carlo Tree Search with Information Relaxation Dual Bounds
Stochastic Optimization Model for Sensor and Protective Device Placements
68
![Page 69: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/69.jpg)
Optimal Protective Device and Sensor Placement
Protective Device vs Sensor Placement to minimize Customer-Outage Minute:
A protective device prevents power flow to the downstream segments when a fault is detected.
A sensor feeds back information to the utility center whether power is on or off.
: a binary vector; 1if a protective device is placed at node and 0 otherwise.
combinations of protective devices which results in high computation complexity.
Solved sequentially by placing one protective device at a time.
Optimization problem for optimal protective device and sensor placements:
69
![Page 70: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/70.jpg)
XX
X
X
X
X
callcall
call
call
call
call
call
0.26 0.38 0.78 0.6 0.6
0
0
0
0.02
0.03
0.04
0.04
0.05
0.064
0.0640.064 0.064
0.502 0.502
0.502 0.502 0.502
0.603 0.603
0.670.670.05
0.05
0.5030.5
0.76
0.5030.503
0.540.54
0.503
0.5030.503
?
?
?
?
Optimal Protective Device and Sensor Placement
?
70
![Page 71: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/71.jpg)
XX
X
X
X
X
Case study A: sensor vs. protective device
callcall
call
call
call
call
call
0.26 0.38 0.78 0.6 0.6
0
0
0
0.02
0.03
0.04
0.04
0.05
0.064
0.0640.064 0.064
0.502 0.502
0.502 0.502 0.502
0.603 0.603
0.670.670.05
0.05
0.5030.5
0.76
0.5030.503
0.540.54
0.503
0.5030.503
Customer Outage-Minute Reduction:Value of sensor = 0Value of protective device = 225
30 Customers
30 Customers
30 Customers
71
![Page 72: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/72.jpg)
XX
X
X
X
X
callcall
call
call
call
call
call
0.26 0.38 0.78 0.6 0.6
0.01
0.01
0.01
0.18
0.27
0.36
0.36
0.45
0.47
0.470.47 0.47
0.501 0.501
0.501 0.501 0.501
0.601 0.601
0.580.580.45
0.45
0.5010.5
0.71
0.5010.501
0.5070.507
0.501
0.5010.501
10 Customers
10 Customers
10 Customers
Customer Outage-Minute Reduction:Value of sensor = 3360Value of protective device = 0
Case study B: sensor vs. protective device
72
![Page 73: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/73.jpg)
Summary
We have developed: A detailed storm and grid simulator that simulates storms, outages and lights-out
calls
Models the behavior of protective devices and grid reconfigurations
An industry-standard escalation policy and a probabilistic lookahead model that uses prior knowledge and anticipates likely outages
Shown that the lookahead policy produces near-optimal performance when compared against optimal on a deterministic model
Simulations We have quantified the value of sensors and protective devices at different
locations
We can find the best locations for each type of device
Conclusion We believe that our simulator can be used to develop effective rate cases for the
BPU
73
![Page 74: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/74.jpg)
Summary
We have developed: A detailed storm and grid simulator that simulates storms, outages and lights-out
calls
Models the behavior of protective devices and grid reconfigurations
An industry-standard escalation policy and a probabilistic lookahead model that uses prior knowledge and anticipates likely outages
Shown that the lookahead policy produces near-optimal performance when compared against optimal on a deterministic model
Simulations We have quantified the value of sensors and protective devices at different
locations
We can find the best locations for each type of device
Conclusion We believe that our simulator can be used to develop effective rate cases for the
BPUThank you for your attention!
74
![Page 75: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/75.jpg)
[1] L. Al-Kanj, B. Bouzaeini-Ayari and W. Powell, “A Probability Model for Grid Faults Using Incomplete Information”, IEEE Transactions on Smart Grids, July 2015.
Problem Description – Grid Restoration
I set of poles, i.e., I , 1, … ,
U set of circuits, i.e., U , 1, … ,
I set of power lines on circuit u, i.e., I , 1, … ,
Nu Set of nodes on circuit u, i.e., N , ∀ ∈
Number of customers attached to node
, Number of customers that called on node
Ω sample space at time ; each scenariorepresents phone calls, faults, travel & repair times
random vector representing the realizations of received calls at time
random vector representing the realizations of power line faults at time
random variable representing the travel time from node to node at timerandom variable representing the repair time of power line on circuit
random vector representing the trajectory of the truck at time ;
, ∈
Bayes’ Theorem:Illustration of distribution grid
Prior
Likelihood of callsgiven faults
Probability of calls
|| |
∑ ∈ |
Posterior
?
?
??
?
?
call
call
call
? ??
?
75
![Page 76: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/76.jpg)
Features of the simulator: Complete model of the distribution grid: Substations Sensors, protective devices, reconfiguration rules, …
Able to simulate: Many storm trajectories with varying sizes Process of outages occurring as storm progresses Model of who loses power (from grid configuration) Process of lights-out calls (with variable call-in rates)
Probabilistic model of outages Captures prior knowledge and probabilistic model of lights-out calls
Dispatch policies: Industry-standard escalation policy – Uses lights-out calls and grid
configuration Stochastic lookahead policy – Exploits uncertainty model of outage
probabilities
Problem Description – Simulator
76
![Page 77: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/77.jpg)
Monte Carlo Tree Search
Primal-Dual MCTS Algorithm Propose a Primal-Dual MCTS, that takes advantage of the information relaxation bound idea that
asymptotically converges [8] to the optimal solution while ignoring suboptimal parts of the tree Explore a set of actions:
Smooth with previous estimates:
Expand it if the value is greater than
[8] D. Jiang, L. Al-Kanj, W. B. Powell, “Monte Carlo Tree Search with Information Relaxation Dual Bounds”, In preparation for Operations Research, 2017. 77
![Page 78: Optimizing Sensor Type and Location for Rapid Restoration](https://reader031.vdocuments.mx/reader031/viewer/2022012514/618d6a07b93b6500f8262ad9/html5/thumbnails/78.jpg)
Simulation Policy – Integer programming
Minimizes the customer-outage minute
Computes number of customers with restored power
Indicates whether power line is still in outage
Computes required travel and repair time by going from node to node
Guarantee that the required travel and repair times are met
Guarantees that each arc is visited at most once in one direction
Indicates whether the truck can move to the next node
The problem can be formulated as a non-linear integer program as follows:
Domains of the variables 78