![Page 1: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/1.jpg)
Optimal control techniques for managementstrategies in biological models
Suzanne Lenhart
University of Tennessee, Knoxville
July 20, 2018
Lenhart (UTK) Management Strategies July 20, 2018 1 / 61
![Page 2: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/2.jpg)
Outline
1. Modeling of Management of Forest Products in Benin
2. An Epidemic Model of Rabies in Raccoons
3. Formulation of PDE model for Zika virus in Brazil
4. Management of Fire Model
Lenhart (UTK) Management Strategies July 20, 2018 2 / 61
![Page 3: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/3.jpg)
Objective Functional
System of ODEs (or PDEs) modeling situationDecide on format and bounds on the controlsDesign an appropriate objective functional–balancing opposing factors in functional–include (or not) terms at the final time
Examples:1. minimize infecteds and cost of vaccinationincluding number of persons vaccinated over time
2. Maximize profit due to fishery and minimize cost of fishing effort
Lenhart (UTK) Management Strategies July 20, 2018 3 / 61
![Page 4: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/4.jpg)
Big Idea
In optimal control applications, after formulating a problem appropriateto the scenario, my approach is :(a) to prove the existence of an optimal control,(b) to characterize the optimal control through states and adjoints,(c) to prove the uniqueness of the control,(d) to compute the optimal control numerically,(e) to investigate how the optimal control depends on various
parameters in the model.
Lenhart (UTK) Management Strategies July 20, 2018 4 / 61
![Page 5: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/5.jpg)
First Example: Harvest
Optimal control in a model of management of forest resources
GOAL Investigate management strategies in a model of harvestingforest products, motivated by forests in Benin
collaborators:O. Gaoue (at UT, from Benin), Wandi Ding (MTSU), Jiang Jiang,Folashade Agusto (U of Kansas)publication: J. Theoretical Ecology 2016
Lenhart (UTK) Management Strategies July 20, 2018 5 / 61
![Page 6: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/6.jpg)
Where is Benin?
Lenhart (UTK) Management Strategies July 20, 2018 6 / 61
![Page 7: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/7.jpg)
Khaya seneglansis: African Mahogany
Lenhart (UTK) Management Strategies July 20, 2018 7 / 61
![Page 8: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/8.jpg)
Lethal harvesting
Lenhart (UTK) Management Strategies July 20, 2018 8 / 61
![Page 9: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/9.jpg)
State System
x, the density of a species and r, its intrinsic growth rate
dx(t)
dt= r(t)x(t)
(1− x(t)
K
)− hL(t)x(t) (1)
τdr(t)
dt= re − r(t)− (αhN (t) + βhL(t)) (2)
Our species x has a Logistic-type growth function with a carryingcapacity K but a time dependent intrinsic growth rate r(t).Control, hN (t) , non-lethal harvesting rate, only affects r(t) directly.Control hL(t), lethal harvesting rate, affects r(t) and directly pulls downthe population in x(t) DE.re is the equilibrium growth rate without any harvesting.τ average lifespan of the plant in years
Lenhart (UTK) Management Strategies July 20, 2018 9 / 61
![Page 10: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/10.jpg)
Our goal is to find an optimal control pair, hL and hN , in order tomaximize the objective functional
J(hL, hN ) =∫ T
0e−δt(Ax(t) +B1hL(t)x(t) +B2hN (t)x(t)− C1h
2L(t)− C2h
2N (t))dt
The coefficients B1 and B2 represent prices from the two types ofharvestingTerms with B1hL(t)x(t) +B2hN (t)x(t) give the corresponding revenue.The weight coefficient A balances the relative importance ofconservation of species x.The quadratic terms with the controls give the costs.
Lenhart (UTK) Management Strategies July 20, 2018 10 / 61
![Page 11: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/11.jpg)
Solve numerically using forward-backward iterative method for statesand adjoints and optimal controls.
Using prices and growth rates from a variety of forest products inBenin.
Lenhart (UTK) Management Strategies July 20, 2018 11 / 61
![Page 12: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/12.jpg)
Optimal control results for slow growth plant
Lenhart (UTK) Management Strategies July 20, 2018 12 / 61
![Page 13: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/13.jpg)
Second Example: Model for Rabies in Raccoons
Collaborator: T. Clayton, S. Duke-Sylvester, L. Gross, L. A. Real, J. ofBiological Dynamics, 2010.
GOAL: Model outbreaks of rabies in raccoons considering variousfeaturesMODEL with Birth Pulse and unusual feature of dynamic equation forvaccine, with systems of ordinary differential equations
S susceptiblesE exposedsI infecteds (able to transmit the disease)R immune (from vaccine or recovery from disease)V vaccine
Lenhart (UTK) Management Strategies July 20, 2018 13 / 61
![Page 14: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/14.jpg)
State System
Note birth pulse and natural death rates are included.control u(t)- input of vaccine baits.
S′ = −(βI + b+
c0V
K + V
)S + a(S + E +R)χΩ(t) (3)
E′ = βIS − (σ + b)E
R′ = σ(1− ρ)E − bR+c0V S
K + V
I ′ = σρE − αIV ′ = −V [c(S + E +R) + c1] + u
The birth pulse acts only during the spring time of the year (March 20to June 21) and χΩ is a characteristic function of the set Ω.
Lenhart (UTK) Management Strategies July 20, 2018 14 / 61
![Page 15: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/15.jpg)
Goal with linear cost
Minimize infected population as well as cost of vaccine, the objectivefunctional is
minu
∫ T
0[I(t) +Bu(t)]dt,
where the set of all admissible controls is
U = u : [0, T ]→ [0,M1]|u is Lebesgue measurable
where coefficient B is a weight factor balancing the two terms. WhenB is large, then the cost of implementing the control is high.
This problem is linear in the control, which implies that the optimalcontrol is bang-bang, singular or combination. In this case, the optimalcontrol is bang-bang..
Lenhart (UTK) Management Strategies July 20, 2018 15 / 61
![Page 16: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/16.jpg)
Infection with no birth pulse
0 5 10 15 20 25 3010
−6
10−4
10−2
100
102
104
TIME
S
0 5 10 15 20 25 300
100
200
300
400
500
600
700
800
TIME
E
0 5 10 15 20 25 3035
40
45
50
55
60
65
70
75
TIME
I
0 5 10 15 20 25 300
50
100
150
TIME
R
0 5 10 15 20 25 300
0.02
0.04
0.06
0.08
0.1
0.12
0.8
V
0 5 10 15 20 25 300
0.2
0.4
0.6
0.8
1
1.2
1.4
TIME
CO
NT
RO
LFigure: birth pulse not encountered
Lenhart (UTK) Management Strategies July 20, 2018 16 / 61
![Page 17: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/17.jpg)
Infection, 3 weeks before birth pulse
0 10 20 3010
−4
10−2
100
102
104
TIME
S
0 10 20 300
100
200
300
400
500
600
700
800
TIME
E
0 10 20 3035
40
45
50
55
60
65
70
75
TIME
I
0 10 20 300
50
100
150
TIME
R
0 10 20 300
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.8
V
0 10 20 300
0.2
0.4
0.6
0.8
1
1.2
1.4
TIME
CO
NT
RO
L
Figure: Notice two control pulses
Lenhart (UTK) Management Strategies July 20, 2018 17 / 61
![Page 18: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/18.jpg)
Thrid Example: Zika and Vaccination
Optimal control of vaccination in a vector-bornereaction-diffusion model applied to Zika virus -Preliminary Report
collaborators: T. Miyaoka, J. MeyerU of Campinas - Brazil
Lenhart (UTK) Management Strategies July 20, 2018 18 / 61
![Page 19: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/19.jpg)
Modeling Zika Virus in Brazil
Zika Virus is a Flavivirus and is primarily transmitted to humansmainly by Aedes aegypti mosquitoesZika can also be transmitted by vertical transmisson, sexualrelations and blood transfusions.Zika virus: concern about children born with neurologicalconditions (microcephaly)Vaccines still in development (clinical trial)How to balance the cost benefit in vaccination?
Goal: Apply optimal control of vaccination in a partial differentialequation model using data in a state in Brazil .
Lenhart (UTK) Management Strategies July 20, 2018 19 / 61
![Page 20: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/20.jpg)
Spatial region for simulations
Figure: Rio Grande do Norte State.
Lenhart (UTK) Management Strategies July 20, 2018 20 / 61
![Page 21: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/21.jpg)
Mathematical modeling
Reaction–Diffusion PDE model.
SIR dynamics for humans and SI for mosquitoes.S, I,R susceptible, infected, recovered (vaccinated) for humansSv, Iv susceptible, infected for mosquitoes
Vaccination rate u gives immunity to susceptible humans.
Control using the vaccination rate u(x, t).
Lenhart (UTK) Management Strategies July 20, 2018 21 / 61
![Page 22: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/22.jpg)
Compartmental dynamics
S I
R
Sv Iv
uS
βSIv
βdSI
βvSvI µvIv
δI
Figure: Flow chart for model (4).
Lenhart (UTK) Management Strategies July 20, 2018 22 / 61
![Page 23: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/23.jpg)
Model in Weak solution sense
∂S
∂t−∇ · (α∇S) = −βSIv − βdSI − uS,
∂I
∂t−∇ · (αI∇I) = βSIv + βdSI − δI,
∂R
∂t−∇ · (α∇R) = uS + δI,
∂Sv∂t−∇ · (αv∇Sv) = −βvSvI + rv (Sv + Iv)
(1− (Sv + Iv)
κv
),
∂Iv∂t−∇ · (αv∇Iv) = βvSvI − µvI, in Q = Ω× (0, T ).
(4)
Plus initial conditions and no flux boundary conditions.Logistic growth for mosquitoesIn the simulations, sexual transmission coefficient was estimatedto be 0.
Lenhart (UTK) Management Strategies July 20, 2018 23 / 61
![Page 24: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/24.jpg)
Optimal control
Goal: minimize cost of infecteds and administering the vaccinecontrol:
J(u) =
∫Q
(c1I(x, t) + c2u(x, t)S(x, t) + c3u(x, t)2)dxdt.
The control setu ∈ L2(Q), 0 ≤ u ≤ umax
Lenhart (UTK) Management Strategies July 20, 2018 24 / 61
![Page 25: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/25.jpg)
Necesary Condtions for Optimal control
To derive necessary conditions for an optimal control, we need todifferentiate the map u→ J(u)
Since J(u) depends on the states, we must differentiate the map,u→ S, I,R, Sv, IvUsing sensitivities (derivatives of states with respect to control),we derive an adjoint PDE system:
L∗(λ) +MTλ = (c2u, c1, 0, 0, 0)T ,
MT =
βIv + βdI + u −βIv − βdI −u 0 0
βdS δ − βdS −δ βvSv −βvSv0 0 0 0 0
0 0 0 −rv +2rvκv
(Sv + Iv) + βvI −βvI
βS −βS 0 −rv +2rvκv
(Sv + Iv) µv
,
RHS of adjoint system is derivative of integrand of J w.r.t. statesLenhart (UTK) Management Strategies July 20, 2018 25 / 61
![Page 26: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/26.jpg)
The derivative operators in L∗ are −(λ1)t −∇(α∇λ1)... −(λ5)t −∇(αv∇λ5)
Plus no flux boundary conditions and transversality conditionsλi(x, T ) = 0 for i = 1, ..., 5
Differentiating the map u→ J(u) and using the sensitivity andadjoint systems, we can characterize our optimal control
u∗(x, t) = min(max((λ1 − λ2 − c2)S∗(x, t)
2c3, 0))
Lenhart (UTK) Management Strategies July 20, 2018 26 / 61
![Page 27: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/27.jpg)
Numerical Simulations
Forward-Backward sweep for optimality system with PDEs solvedby finite elements.Data from Rio Grande do Norte state in BrazilSome parameters from literature and other parameters wereestimatedFor estimation, used incidence from simulated system and data
Yij =
∫ tj+τ
tj
∫ωi
(βSIv + βdSI) dxdt.
Least Squares Approach with normalized residual:
R =
√√√√√√√∑i,j
(Yij − Yij
)2∑i,j
(Yij)2 .
Lenhart (UTK) Management Strategies July 20, 2018 27 / 61
![Page 28: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/28.jpg)
Data set
2015
epi
dem
iolo
gica
l wee
k
25
20
15
10
1234567Cities
8910111213
0
1000
1500
500
Incid
en
ce
1: Santa Cruz
2: São Gonçalo do Amarante
3: Rio do Fogo
4: João Câmara
5: Apodi
6: Santo Antônio
7: Areia Branca
8: Parnamirim
9: Canguaretama
10: Caicó
11: Ceará Mirim
12: Mossoró
13: Natal
Figure: Incidence in selected cities by 2015 weeks, accounted forunder-reporting of cases.
Lenhart (UTK) Management Strategies July 20, 2018 28 / 61
![Page 29: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/29.jpg)
Spatial region for simulations
x (km)
0 50 100 150 200 250 300 350
y (
km
)
0
50
100
150
200
Apodi
Areia Branca
CaicóCanguaretama
Ceará MirimJoão Câmara
Mossoró
Natal
Parnamirim
Rio do fogo
Santa CruzSanto Antônio
São Gonçalo do Amarante
Figure: Finite elements mesh and city locations.
Source term due to immigration added at 7 and 21 days in thehuman infected PDE
Lenhart (UTK) Management Strategies July 20, 2018 29 / 61
![Page 30: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/30.jpg)
Initial conditions for simulations
Initial conditions :
S : 3.4 million distributed over space.
I : small amount in one city.
R : none.
Sv : 17 million distributed over space.
Iv : none.
Two sources of infecteds added in locations indicated by the data
Lenhart (UTK) Management Strategies July 20, 2018 30 / 61
![Page 31: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/31.jpg)
No control
x (km)
0 100 200 300
y (
km
)
0
100
200
I(x,y,35)
0
0.2
0.4
x (km)
0 100 200 300
y (
km
)
0
100
200
I(x,y,52)
0
0.5
1
x (km)
0 100 200 300
y (
km
)
0
100
200
I(x,y,70)
0
1
2
x (km)
0 100 200 300
y (
km
)
0
100
200
I(x,y,91)
0
2
4
x (km)
0 100 200 300
y (
km
)
0
100
200
I(x,y,112)
0
2
4
x (km)
0 100 200 300
y (
km
)
0
100
200
I(x,y,140)
0
2
4
Figure: Solutions at different times, no control. The three infection sourcesspread over space. Difference in scales.
Lenhart (UTK) Management Strategies July 20, 2018 31 / 61
![Page 32: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/32.jpg)
Control starting at t = 35 days
x (km)
0 100 200 300
y (
km
)
0
100
200
I(x,y,35)
0
0.2
0.4
x (km)
0 100 200 300
y (
km
)
0
100
200
I(x,y,52)
0
0.5
1
x (km)
0 100 200 300
y (
km
)
0
100
200
I(x,y,70)
0
1
2
x (km)
0 100 200 300
y (
km
)
0
100
200
I(x,y,91)
0
1
2
3
x (km)
0 100 200 300
y (
km
)
0
100
200
I(x,y,112)
0
1
2
3
x (km)
0 100 200 300
y (
km
)
0
100
200
I(x,y,140)
0
1
2
x (km)
0 100 200 300
y (
km
)
0
100
200
u(x,y,35) ×10-3
0
2
4
x (km)
0 100 200 300
y (
km
)
0
100
200
u(x,y,70) ×10-3
0
2
4
x (km)
0 100 200 300
y (
km
)
0
100
200
u(x,y,112) ×10-5
0
2
4
Figure: Solutions at different times, control starting at t = 35 days. Differencein scales.
Lenhart (UTK) Management Strategies July 20, 2018 32 / 61
![Page 33: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/33.jpg)
Time plots
t (days)
0 35 70 105 140
S(t
)
×10 6
2.6
2.8
3
3.2
3.4
S(t)
t (days)
0 35 70 105 140
I(t)
×10 4
0
5
10
15I(t)
t (days)
0 35 70 105 140
R(t
)
×10 5
0
2
4
6
R(t)
t (days)
0 35 70 105 140
Sv(t
)
×10 7
1
1.2
1.4
1.6
Sv(t)
t (days)
0 35 70 105 140
I v(t
)×10 6
0
1
2
3
Iv(t)
without control control starting at t=35 days control starting at t=52 days control starting at t=70 days
t (days)
0 35 70 105 140
u(t
)
×10 -4
0
5
10u(t)
Figure: Solutions integrated over space.
Lenhart (UTK) Management Strategies July 20, 2018 33 / 61
![Page 34: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/34.jpg)
Quantities of interest
Table: Optimal control results compared to the scenario without control
without control starting control starting control startingcontrol at t = 35 days at t = 52 days at t = 70 days
Total incidence 4.719× 105 2.746× 105 3.271× 105 3.811× 105
% of no control 58.18% 69.32% 80.75%
Total vaccinated 4.951× 105 4.072× 105 2.939× 105
J(u∗) 3.341× 108 2.454× 108 2.759× 108 3.040× 108
% of no control 73.44% 82.58% 90.98%
J(umax) 3.093× 108 3.326× 108 3.536× 108
% of no control 92.56% 99.53% 105.83%
J(umax)/J(u∗) 126.03% 120.53% 116.33%
Lenhart (UTK) Management Strategies July 20, 2018 34 / 61
![Page 35: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/35.jpg)
Conclusions
Successful application of vaccination using optimal control.
Model has been applied to real data from Brazil.
Global sensitivity analysis performed with optimal control J(u∗)value as output.
Other kinds of control could be applied and we are currentlyworking on that.
Same model can be adapted to other similar diseases.
In future work, we are including better spatial varying ICs andefficacy of vaccine.
Lenhart (UTK) Management Strategies July 20, 2018 35 / 61
![Page 36: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/36.jpg)
Fourth Example : Fire Management
Assessing the Economic Tradeoffs Between Prevention andSuppression of Forest Fires
collaborators: Betsy Heines,Charles Sims, paper in NaturalResource Modelng, 2018
GOALCombine optimal control, ecology, and economics in order todetermine the optimal fire prevention and suppression spendingby maximizing the value of a forest under threat of fire usingPontryagin’s Maximum Principle.
Lenhart (UTK) Management Strategies July 20, 2018 36 / 61
![Page 37: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/37.jpg)
Introduction
The total number of forested acres burning in the US is increasing,despite fewer fires total.Federal suppression spending is increasing.
Strict fire exclusion policies have produced overgrown forests,leading to larger and more severe fire events.Active prevention management of forests has the potential tomitigate these effects.
Lenhart (UTK) Management Strategies July 20, 2018 37 / 61
![Page 38: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/38.jpg)
Introduction
Fire PreventionActions before a fire.Prescribed burning,mechanical thinning,...Decreases severity andprobability of ignition.
Fire SuppressionActions to control fire.Aerial spraying, boots onthe ground,...
Lenhart (UTK) Management Strategies July 20, 2018 38 / 61
![Page 39: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/39.jpg)
Motivation
Reed, William J., and Hector Echavarria Heras. ”The conservationand exploitation of vulnerable resources.” Bulletin of MathematicalBiology 54.2-3 (1992): 185-207.
Resource management models include risk of catastrophiccollapse at an unknown time.Reed’s Method allows us to convert a stochastic problem into adeterministic optimal control problem.
Lenhart (UTK) Management Strategies July 20, 2018 39 / 61
![Page 40: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/40.jpg)
Formulating the OC Problem
Some Assumptions:At most one fire occurs in [0, T ]. We determine the optimalprevention schedule up to the time of fire.The fire event and all associated costs are taken to beinstantaneous.The spread of fire is not modeled. However, the uncertainty in thetiming of a fire is captured through our application of Reed’sMethod.
Lenhart (UTK) Management Strategies July 20, 2018 40 / 61
![Page 41: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/41.jpg)
Formulating the OC ProblemValue of Forest Before Fire
Let A(t) represent the number of unburned acres in an A acre forest.Suppose a fire occurs at time τ ∈ [0, T ].
Net Value of Forest Before Fire∫ τ
0
[B(A(t)
)− h(t)
]e−rtdt (5)
Flow of benefits BPrevention management spending rate h
where A(t) is given by the solution to
A′(t) = δ(A−A(t)
)with A(0) = A0. (6)
Lenhart (UTK) Management Strategies July 20, 2018 41 / 61
![Page 42: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/42.jpg)
Formulating the OC ProblemValue of Forest After Fire
Suppose K(h(τ), x(τ)
)acres are burned in the fire at time τ and A(t)
represents the number of unburned acres after the fire.Net Value of Forest After a Fire
∫ T
τB(A(t)
)e−rtdt−
[D(K(h(τ), x(τ)
))+ x(τ)
]e−rτ (7)
Flow of benefits BNontimber damages (instantaneous) DSuppression costs (instantaneous) h
where A(t) is given by the solution to
A′(t) = δ(A− A(t)
)with A(τ) = A(τ)−K
(h(τ), x(τ)
). (8)
Lenhart (UTK) Management Strategies July 20, 2018 42 / 61
![Page 43: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/43.jpg)
Formulating the OC ProblemMaximize Value of Forest After Fire
Define the optimal value of the forest after a fire by
JW ∗(τ,A(τ), h(τ)
)=
maxx(τ)
∫ T
τB(A(t)
)e−r(t−τ)dt−
[D(K(h(τ), x(τ)
))+ x(τ)
]subject to x(τ) ≥ 0
where A(t) = A−(A−
(A(τ)−K
(h(τ), x(τ)
)))e−δ(t−τ). (9)
We solve the problem above using scalar optimization with respect to xfor a given τ , A(τ), and h(τ).
Note: JW ∗(τ,A(τ), h(τ)
)will be a function with an explicit closed form
due to our functional choices.Lenhart (UTK) Management Strategies July 20, 2018 43 / 61
![Page 44: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/44.jpg)
Formulating the OC ProblemValue of a Forest Over [0, T ] - Fire at τ ∈ [0, T ]
Suppose that a fire occurs at time τ ∈ [0, T ] and that suppressionspending is optimal, then the value of the forest over [0, T ] is
∫ τ
0
[B(A(t)
)− h(t)
]e−rtdt + JW ∗
(τ,A(τ), h(τ)
)e−rτ (10)
Net value before fireNet value after fire w/ optimal suppression expenditures x∗
where
A(t) = A− (A−A0)e−δt. (11)
Lenhart (UTK) Management Strategies July 20, 2018 44 / 61
![Page 45: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/45.jpg)
Formulating the OC ProblemValue of a Forest Over [0, T ] - No Fire
Suppose a fire does not occur in [0, T ]. Then the value of the forestover [0, T ] is
∫ T
0
[B(A(t)
)− h(t)
]e−rtdt. (12)
Benefits minus prevention over full time horizon
where
A(t) = A− (A−A0)e−δt. (13)
Lenhart (UTK) Management Strategies July 20, 2018 45 / 61
![Page 46: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/46.jpg)
Formulating the OC ProblemTime of Fire as a Random Variable
To capture the uncertainty of the time of fire τ ∈ [0,∞), we treat it as arandom variable T .
Hazard Function:
ψ(h(t)
)= lim
∆t→0
Pr(fire in [t, t+ ∆t)| no fire up to t
)/∆t
(14)
Survivor Function:S(t) = e−
∫ t0 ψ(h(z))dz (15)
Cumulative Distribution Function:
F (t) = 1− S(t) (16)
Lenhart (UTK) Management Strategies July 20, 2018 46 / 61
![Page 47: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/47.jpg)
Formulating the OC ProblemTime of Fire as a Mixed Type Random Variable
We are considering a finite time interval [0, T ] and thus consider atruncated random variable TM :
TM =
T if τ ≤ TT if τ > T.
(17)
Cumulative Distribution Function:
FTM (τM ) =
1− S(τM ) if τM < T
1 if τM = T(18)
Lenhart (UTK) Management Strategies July 20, 2018 47 / 61
![Page 48: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/48.jpg)
Formulating the OC ProblemFrom Stochastic to Deterministic
The expected value of the forest J(h) is given by the expectation, withrespect to the RV TM , of the piecewise function for the value of theforest:
J(h) = ETM
∫ τM
0
[B(A(t)
)− h(t)
]e−rtdt
+e−rτMJW ∗(τM , A(τM ), h(τM )
)if τM < T∫ T
0
[B(A(t)
)− h(t)
]e−rtdt if τM = T.
(19)
Lenhart (UTK) Management Strategies July 20, 2018 48 / 61
![Page 49: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/49.jpg)
Formulating the OC ProblemFrom Stochastic to Deterministic
After a little calculus we arrive at
J(h) =
∫ T
0
[B(A(t)
)− h(t) + ψ
(h(t)
)JW ∗
(t, A(t), h(t)
)]e−rt−y(t)dt
(20)where we have introduced a new state variable y defined by
y′(t) = ψ(h(t)
)with y(0) = 0. (21)
This allows us to write S(t) = e−y(t).
The stochastic problem has been converted to deterministic.
Lenhart (UTK) Management Strategies July 20, 2018 49 / 61
![Page 50: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/50.jpg)
The Optimal Control Problem:
maxh
∫ T
0
[B(A(t)
)− h(t) + ψ
(h(t)
)JW ∗
(t, A(t), h(t)
)]e−rt−y(t)dt (22)
subject to y′(t) = ψ(h(t)
)with y(0) = 0 (23)
h(t) ≥ 0 (24)
where A(t) = A− (A−A0)e−δt. (25)
Next, we present the conditional current-value Hamiltonian andoptimality system and introduce our chosen functional forms.
Lenhart (UTK) Management Strategies July 20, 2018 50 / 61
![Page 51: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/51.jpg)
Conditional Current-Value Hamiltonian & OptimalitySystem
Let H be the Hamiltonian with adjoint λ. Then the conditional current-valueHamiltonian is H = ert+y(t)H with corresponding adjoint equation ρ(t) = ert+y(t)λ(t).
Conditional Current-Value Hamiltonian
H = B(A(t)
)− h(t) + ψ
(h(t)
)JW ∗(t, A(t), h(t)
)+ ρ(t)ψ
(h(t)
)(26)
Optimality Condition, in interior of control set
∂H∂h
= −1+JW ∗(t, A(t), h(t))∂ψ∂h
+∂JW ∗
∂hψ(h(t)
)+ρ(t)
∂ψ
∂h= 0 (27)
Adjoint Equation
ρ(t) = rρ(t) +B(A(t)
)− h(t) + ψ
(h(t)
)(ρ(t) + JW ∗(t, A(t), h(t)
))(28)
Transversality Conditionρ(T ) = 0 (29)
Lenhart (UTK) Management Strategies July 20, 2018 51 / 61
![Page 52: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/52.jpg)
Functional Forms
Benefits Function : B1 - benefits parameter
B(A(t)
)= B1A(t)
Hazard Function : b - background hazard of firev - hazard management effectiveness parameter
ψ(h(t)
)= be−vh(t)
Kill Function : k - fire severity parameterk1 - severity management effectiveness parameterk2 - severity suppression effectiveness parameter
K(h, x) =k
(k1 + h)(k2 + x)
Nontimber Damage Function : c - cost parameter
D(K(h, x)
)= cK(h, x)
Lenhart (UTK) Management Strategies July 20, 2018 52 / 61
![Page 53: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/53.jpg)
Numerical Methods
Due to the complexity of ψ and JW ∗ an explicit closed formcannot be determined for h∗ from the optimality condition.
We numerically determine h∗ by maximizing the conditionalcurrent-value Hamiltonian with respect to h at each time step.
The MATLAB function fminbnd is used to optimize h. Since thecontrol h is not bounded above, a large upper bound was used onfminbnd.
Lenhart (UTK) Management Strategies July 20, 2018 53 / 61
![Page 54: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/54.jpg)
Results - 2011 Las Conchas Fire, NM
When: June - August 2011Location: Santa Fe NationalForest, Bandelier NationalMonument, Valles CalderaNational PreserveAcres Burned: > 150,000Suppression Costs: ≈ $40.9MillionStructures Destroyed: 63homes, 49 outbuildingsParameters chosen for thisspecific fire
Lenhart (UTK) Management Strategies July 20, 2018 54 / 61
![Page 55: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/55.jpg)
Results - Las Conchas
No Prevention
Value of Forest: J(0) =$773MExpected Time to Fire: 10 yrs
Optimal Prevention
Value of Forest: J(h∗) =$801MExpected Time to Fire: 45 yrs
Lenhart (UTK) Management Strategies July 20, 2018 55 / 61
![Page 56: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/56.jpg)
Results - Las Conchas
Preliminary Results
When the optimal prevention management spending rate is applied,we see:
An increase in the expected value of the forest.An increase in the mean time of fire.
Lenhart (UTK) Management Strategies July 20, 2018 56 / 61
![Page 57: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/57.jpg)
Fire Sequences
Using a different initial condition for A(t), it is possible to considersequences of fires by successively applying our optimal controlproblem.
We can determine the time of fire by sampling from the cumulativedistribution function for the time of fire random variable.
Lenhart (UTK) Management Strategies July 20, 2018 57 / 61
![Page 58: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/58.jpg)
Fire SequencesResults for 500 Simulations - Number of Fires
With optimal prevention, on averagethere are fewer severe fire events in atime period of 50 years.
Number of FiresPrev. Optimal NoneMean 1.4 5.0
Median 1 5Std. 1.1 2.4
Lenhart (UTK) Management Strategies July 20, 2018 58 / 61
![Page 59: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/59.jpg)
Fire SequencesResults for 500 Simulations - Value of the Forest
Larger mean/median for optimalprevention case.Furthermore, std. for no preventioncase is over triple the std. of optimalprevention case.
Value of Forest $MPrev. Optimal NoneMean 671 536
Median 677 556Std. 34.0 111.7
Lenhart (UTK) Management Strategies July 20, 2018 59 / 61
![Page 60: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/60.jpg)
Conclusions
From our work with fire sequences, we see that on average withprevention management:
The overall value of a forest is increased by 25% and has lessvariation than when no prevention management efforts are made.
There is a 72% reduction in the number of forest fires.Furthermore, the forest is at less risk for fire.
There is an 82% reduction in suppression spending and a 55%reduction in management and suppression spending in total.
This work showcases a valuable tool which could guide forestmanagers and policymakers in their development of forest firemanagement plans.
Lenhart (UTK) Management Strategies July 20, 2018 60 / 61
![Page 61: Optimal control techniques for management strategies in ...juliu/cbms18/img/Lenhart.pdfModeling of Management of Forest Products in Benin 2. An Epidemic Model of Rabies in Raccoons](https://reader035.vdocuments.site/reader035/viewer/2022071416/6111db5b2ae4521e99631642/html5/thumbnails/61.jpg)
Acknowledgments and Thanks!
National Institute for Mathematical and Biological Synthesiswww.nimbios.org
THANK YOU
Lenhart (UTK) Management Strategies July 20, 2018 61 / 61