In each iteration Evaluates the effects of each alternative schedule on the constraints Evaluates the expected loss over time Selects the fuel treatment schedule that provides the minimum overall expected loss over time while satisfying the constraints ID: 928096
Download Presentation The PPT/PDF document "Heuristic Solver Builds and tests altern..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Heuristic Solver
Builds and tests alternative fuel treatment schedules (solutions) at each iterationIn each iteration:Evaluates the effects of each alternative schedule on the constraintsEvaluates the expected loss over timeSelects the fuel treatment schedule that provides the minimum overall expected loss over time while satisfying the constraints
Slide2Heuristic Solver
Uses Simulated Annealing (SA) Algorithm to select treatments The SA algorithm is based on simulating cooling of materials in a bath (annealing)Heuristic optimization technique widely used to solve large combinatorial problems in various fieldsAssignment/scheduling problems
Transportation network problems
Manufacturing problems
Monte Carlo approach that uses a local search
A subset of all possible solutions is explored by moving to through neighbor solutions
Some lower quality solutions are accepted to avoid solutions stagnation at local optimum
Slide3Building Clusters of Polygons for Treatment
Assigns fuel treatments to stand polygons (GIS layer)Treating individual stands is generally not effective at changing fire behavior at the landscape level (Finney 2006)Solver builds clusters of adjacent polygons to form larger treatment units
Slide4Clustering adjacent stand polygons
No action
Select random polygon
Building Clusters of Polygons for Treatment
Slide5Clustering adjacent stand polygons
Minimum cluster 100.0 ac
Selected polygon
5.0 ac
Clustered area
5.0 ac
No action
Selected
Adjacent
Select random polygon
Update cluster area
Is current cluster area > minimum cluster area ?
Identify adjacent polygons
No
Select random adjacent polygon
Building Clusters of Polygons for Treatment
Slide6Clustering adjacent stand polygons
Minimum cluster 100.0 ac
Selected polygon
4.5 ac
Clustered area
9.5 ac
No action
Selected
Adjacent
Clustered
Select random polygon
Identify adjacent polygons
Select random adjacent polygon
Update cluster area
Is current cluster area > minimum cluster area ?
No
Building Clusters of Polygons for Treatment
Slide7Clustering adjacent stand polygons
Minimum cluster 100.0 ac
Selected polygon
5.8 ac
Clustered area
15.3 ac
No action
Selected
Adjacent
Clustered
Select random polygon
Identify adjacent polygons
Select random adjacent polygon
Update cluster area
Is current cluster area > minimum cluster area ?
No
Building Clusters of Polygons for Treatment
Slide8Clustering adjacent stand polygons
Minimum cluster 100.0 ac
Selected polygon
6.1 ac
Clustered area
21.4 ac
No action
Selected
Adjacent
Clustered
Select random polygon
Identify adjacent polygons
Select random adjacent polygon
Update cluster area
Is current cluster area > minimum cluster area ?
No
Building Clusters of Polygons for Treatment
Slide9Clustering adjacent stand polygons
Minimum cluster 100.0 ac
Selected polygon
5.2 ac
Clustered area
26.6 ac
No action
Selected
Adjacent
Clustered
Select random polygon
Identify adjacent polygons
Select random adjacent polygon
Update cluster area
Is current cluster area > minimum cluster area ?
No
Building Clusters of Polygons for Treatment
Slide10Clustering adjacent stand polygons
Minimum cluster 100.0 ac
Selected polygon
7.2 ac
Clustered area
33.8 ac
No action
Selected
Adjacent
Clustered
Select random polygon
Identify adjacent polygons
Select random adjacent polygon
Update cluster area
Is current cluster area > minimum cluster area ?
No
Building Clusters of Polygons for Treatment
Slide11Clustering adjacent stand polygons
Minimum cluster 100.0 ac
Selected polygon
8.1 ac
Clustered area
41.9 ac
No action
Selected
Adjacent
Clustered
Select random polygon
Identify adjacent polygons
Select random adjacent polygon
Update cluster area
Is current cluster area > minimum cluster area ?
No
Building Clusters of Polygons for Treatment
Slide12Clustering adjacent stand polygons
Minimum cluster 100.0 ac
Selected polygon
8.8 ac
Clustered area
50.7 ac
No action
Selected
Adjacent
Clustered
Select random polygon
Identify adjacent polygons
Select random adjacent polygon
Update cluster area
Is current cluster area > minimum cluster area ?
No
Building Clusters of Polygons for Treatment
Slide13Clustering adjacent stand polygons
Minimum cluster 100.0 ac
Selected polygon
7.7 ac
Clustered area
58.4 ac
No action
Selected
Adjacent
Clustered
Select random polygon
Identify adjacent polygons
Select random adjacent polygon
Update cluster area
Is current cluster area > minimum cluster area ?
No
Building Clusters of Polygons for Treatment
Slide14Clustering adjacent stand polygons
Minimum cluster 100.0 ac
Selected polygon
7.8 ac
Clustered area
74.0 ac
No action
Selected
Adjacent
Clustered
Select random polygon
Identify adjacent polygons
Select random adjacent polygon
Update cluster area
Is current cluster area > minimum cluster area ?
No
Building Clusters of Polygons for Treatment
Slide15Clustering adjacent stand polygons
Minimum cluster 100.0 ac
Selected polygon
8.9 ac
Clustered area
82.9 ac
No action
Selected
Adjacent
Clustered
Select random polygon
Identify adjacent polygons
Select random adjacent polygon
Update cluster area
Is current cluster area > minimum cluster area ?
No
Building Clusters of Polygons for Treatment
Slide16Clustering adjacent stand polygons
Minimum cluster 100.0 ac
Selected polygon
8.2 ac
Clustered area
91.1 ac
No action
Selected
Adjacent
Clustered
Select random polygon
Identify adjacent polygons
Select random adjacent polygon
Update cluster area
Is current cluster area > minimum cluster area ?
No
Building Clusters of Polygons for Treatment
Slide17Clustering adjacent stand polygons
Minimum cluster 100.0 ac
Selected polygon
7.8 ac
Clustered area
98.9 ac
No action
Selected
Adjacent
Clustered
Select random polygon
Identify adjacent polygons
Select random adjacent polygon
Update cluster area
Is current cluster area > minimum cluster area ?
No
Building Clusters of Polygons for Treatment
Slide18Clustering adjacent stand polygons
Minimum cluster 100.0 ac
Selected polygon
7.8 ac
Clustered area
106.8 ac
No action
Selected
Clustered
Select random polygon
Identify adjacent polygons
Select random adjacent polygon
Update cluster area
Is current cluster area > minimum cluster area ?
No
Stop clustering polygons
Yes
Building Clusters of Polygons for Treatment
Slide19Clustering adjacent stand polygons
Minimum cluster 100.0 ac
Selected polygon
7.8 ac
Clustered area
106.8 ac
Select random polygon
Identify adjacent polygons
Select random adjacent polygon
Update cluster area
Is current cluster area > minimum cluster area ?
No
Stop clustering polygons
Yes
Building Clusters of Polygons for Treatment
Slide20Heuristic Solver
Objective is to minimize expected loss
where :
c
: Index of grid cells (pixels)
t
: Index of time period
Loss
c,t
: Expected loss value for grid cell
c
for period
t
, based
on the flame length predicted by MTT
P
c,t
: Probability of cell
c
being burned in period t, based
on the fire arrival time predicted by MTT
Slide21Steps in
each IterationBuild or modify a solution
(timing and placement of treatments)
Is SA stopping criteria met ?
No
Data passed to Solver
:
Landscape fuel
parameters
Fire scenario
Objective Function
Constraints
Adjacent Polygons
topography
Run MTT for each planning period and retrieve results
by pixel
( flame length , arrival time)
Calculate
objective function value
(total expected loss value)
Report Best Found Solution
Yes
Update the landscape fuel parameters for each period
Slide22Build or modify a solution
(timing and placement of treatments)
Is SA stopping criteria met ?
No
Data passed to Solver
:
Landscape fuel
parameters
Fire scenario
Objective Function
Constraints
Adjacent Polygons
topography
Run MTT for each planning period and retrieve results
by pixel
( flame length , arrival time)
Calculate
objective function value
(total expected loss value)
Report Best Found Solution
Yes
Update the landscape fuel parameters for each period
Steps in
each Iteration
Slide23Build or modify a solution
(timing and placement of treatments)
Is SA stopping criteria met ?
No
Data passed to Solver
:
Landscape fuel
parameters
Fire scenario
Objective Function
Constraints
Adjacent Polygons
topography
Run MTT for each planning period and retrieve results
by pixel
( flame length , arrival time)
Calculate
objective function value
(total expected loss value)
Report Best Found Solution
Yes
Update the landscape fuel parameters for each period
Steps in
each Iteration
Slide24Build or modify a solution
(timing and placement of treatments)
Is SA stopping criteria met ?
No
Data passed to Solver
:
Landscape fuel
parameters
Fire scenario
Objective Function
Constraints
Adjacent Polygons
topography
Run MTT for each planning period and retrieve results
by pixel
( flame length , arrival time)
Calculate
objective function value
(total expected loss value)
Report Best Found Solution
Yes
Update the landscape fuel parameters for each period
Steps in
each Iteration
Slide25Build or modify a solution
(timing and placement of treatments)
Is SA stopping criteria met ?
No
Data passed to Solver
:
Landscape fuel
parameters
Fire scenario
Objective Function
Constraints
Adjacent Polygons
topography
Run MTT for each planning period and retrieve results
by pixel
( flame length , arrival time)
Calculate
objective function value
(total expected loss value)
Report Best Found Solution
Yes
Update the landscape fuel parameters for each period
Steps in
each Iteration
Slide26Build or modify a solution
(timing and placement of treatments)
Is SA stopping criteria met ?
No
Data passed to Solver
:
Landscape fuel
parameters
Fire scenario
Objective Function
Constraints
Adjacent Polygons
topography
Run MTT for each planning period and retrieve results
by pixel
( flame length , arrival time)
Calculate
objective function value
(total expected loss value)
Report Best Found Solution
Yes
Update the landscape fuel parameters for each period
Steps in
each Iteration
Slide27Build or modify a solution
(timing and placement of treatments)
Is SA stopping criteria met ?
No
Data passed to Solver
:
Landscape fuel
parameters
Fire scenario
Objective Function
Constraints
Adjacent Polygons
topography
Run MTT for each planning period and retrieve results
by pixel
( flame length , arrival time)
Calculate
objective function value
(total expected loss value)
Report Best Found Solution
Yes
Update the landscape fuel parameters for each period
Steps in
each Iteration
Slide28Build or modify a solution
(timing and placement of treatments)
Is SA stopping criteria met ?
No
Data passed to Solver
:
Landscape fuel
parameters
Fire scenario
Objective Function
Constraints
Adjacent Polygons
topography
Run MTT for each planning period and retrieve results
by pixel
( flame length , arrival time)
Calculate
objective function value
(total expected loss value)
Report Best Found Solution
Yes
Update the landscape fuel parameters for each period
Steps in
each Iteration
Slide29Build or modify a solution
(timing and placement of treatments)
Is SA stopping criteria met ?
No
Data passed to Solver
:
Landscape fuel
parameters
Fire scenario
Objective Function
Constraints
Adjacent Polygons
topography
Run MTT for each planning period and retrieve results
by pixel
( flame length , arrival time)
Calculate
objective function value
(total expected loss value)
Report Best Found Solution
Yes
Update the landscape fuel parameters for each period
Steps in
each Iteration
Slide30Treatment Selection
Selection of treatments and location of clusters to develop alternative schedules (solutions) is conducted in two phasesPhase I: Add clusters until the maximum area feasible to treat is reached in each periodIteration zero – “no action”Following iterations – randomly locate and add feasible clusters over the landscape
Phase II: Replace clusters with new clusters to find the best allocation and timing of fuel treatments
At each iteration – randomly select a new cluster and remove a previously selected cluster
Continues until stopping rule is reached
Slide31Lower bound Upper bound
400 650P1P2
Phase I – add a cluster
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide32Lower bound Upper bound
400 650
P1
P2
Phase I – add a cluster
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide33Lower bound Upper bound
400 650
P1
P2
Phase I – add a cluster
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide34Lower bound Upper bound
400 650
P1
P2
Phase I – add a cluster
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide35Lower bound Upper bound
400 650
P1
P2
Phase I – add a cluster
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide36Lower bound Upper bound
400 650
P1
P2
Phase I – add a cluste
r
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide37Lower bound Upper bound
400 650
P1
P2
Phase I – add a cluster
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide38Lower bound Upper bound
400 650
P1
P2
Phase I – add a cluster
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide39Lower bound Upper bound
400 650
P1
P2
Phase I – add a cluster
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide40Lower bound Upper bound
400 650
P1
P2
Phase I – add a cluster
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide41Lower bound Upper bound
400 650
P1
P2
Phase I – add a cluster
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide42Lower bound Upper bound
400 650
P1
P2
Phase I – add a cluster
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Numerous tries found no addition clusters could be added without exceeding the 650–ac Upper Bound
===> Go to Phase II
Treatment Selection
Slide43Lower bound Upper bound
400 650
P1
P2
Phase II – change cluster locations and timing
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide44Lower bound Upper bound
400 650
P1
P2
Phase II – change cluster locations and timing
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide45Lower bound Upper bound
400 650
P1
P2
Phase II – change cluster locations and timing
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide46Lower bound Upper bound
400 650
P1
P2
Phase II – change cluster locations and timing
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide47Lower bound Upper bound
400 650
P1
P2
Phase II – change cluster locations and timing
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide48Lower bound Upper bound
400 650
P1
P2
Phase II – change cluster locations and timing
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide49Lower bound Upper bound
400 650
P1
P2
Phase II – change cluster locations and timing
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide50Lower bound Upper bound
400 650
P1
P2
Phase II – change cluster locations and timing
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide51Lower bound Upper bound
400 650
P1
P2
Phase II – change cluster locations and timing
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide52Lower bound Upper bound
400 650
P1
P2
Phase II – change cluster locations and timing
Evaluate feasibility
Run MTT algorithm
arrival time
flame length
Calculate expected loss
Period 1 Period 2
Area constraints
Period 1
400 – 650 acres
Period 2
400 – 650 acres
Treatment Selection
Slide53Solver Performance
Algorithm performance
Phase I
Phase II
“no action”
Expected Loss
Iteration Number
Slide54Hardware Requirements and Solution Times
OptFuels has parallel computing capabilities that use all available processors in a machine
For an area of 34,600 acres with 2 treatment period the solver takes:
Computer Type
(Processor )
Number of processor/Threads
Processor Speed
Iterations
per minute
Solution
times
Risk Assessment
Run
Low Intensity Run
(~400 iterations)
Regular
desktop
(Intel Core 2 4300)
2
1.8 GHz
0.4
Several
minutes
18 hrs
Multi-processor Laptop
(Intel i7-2720QM)
8
2.33 GHz
2.5
Seconds
2.5 hrsMulti-processor workstation(Intel Xeon X5570)
8
2.93 GHz3.5Seconds2 hrs
Slide55Heuristic Solver Enhancements
Adding selection of individual stands for treatments in the latter stages of Phase II.This selects individual stands for treatment that lie between treatment clusters where the area is too small to generate another clusterAdding the ability to schedule treatments to minimize expected loss from two or more fire scenarios.
Different ignition points
Different wind direction and speed
Integrating OptFuels into IFT-DSS
Slide56Results Available from OptFuels
Spatial and temporal treatment schedule that can be mapped in GIS Expected loss Effects functions evaluated for the treatment schedule could include:Treatment costs, treatment acres, volumes of treatment products (if any), treatment revenues (if any), sediment yields, resource effects (
eg
. acres of aspen)
Outputs produced by
FlamMap
MTT for both the treated and untreated landscapes:
GIS display of flame length and arrival time.
LCP files that could be used for additional
FlamMap
analyses.
FVS stand parameters for the projected stands for both the treated and untreated landscapes:
Tree species, stand size, stand volume, etc.
Slide57Thank You!
Questions?