Description
This model helps a farmer to decide how to allocate his or her land. The yields are uncertain. Here we use Lindo NBD algorithm and need to modify the model since Lindo's NDB algorithm requires that a RV occurs only once in the model.
Small Model of Type : SP
Category : GAMS EMP library
Main file : farmnbd.gms
$title The Farmer's Problem - Stochastic with NBD (FARMNDB,SEQ=100)
$onText
This model helps a farmer to decide how to allocate
his or her land. The yields are uncertain. Here we use
Lindo NBD algorithm and need to modify the model since
Lindo's NDB algorithm requires that a RV occurs only once
in the model.
Birge, R, and Louveaux, F V, Introduction to Stochastic Programming.
Springer, 1997.
Contributor: Michael Ferris
$offText
* Only run for Lindo
$ifI not "%gams.emp%"=="lindo" $exit
Set crop / wheat, corn, sugarbeets /
ch header for data table /
yield yield in tons per acre
cost plantcost on dollars per acre
pprice crop seed purchase price in dollars per ton
minreq minimum requirements of crop in ton to feed cattle /
alias (c,crop);
Table cd(crop,ch) crop data
yield cost pprice minreq
wheat 2.5 150 238 200
corn 3 230 210 240
sugarbeets 20 260
;
Parameter
yf(crop) yield factor / #crop 1 /
land available land in acres /500/;
Set seq price curve segments / s1*s2 /;
Table pricecurve(crop,seq,*) dollars per ton
price ub
wheat.s1 170 inf
corn.s1 150 inf
sugarbeets.s1 36 6000
sugarbeets.s2 10 inf
;
set pcs(crop,seq) relevant segments; option pcs<pricecurve;
set errorPC(crop) price curve is not concave;
errorPC(c) = smin(pcs(c,seq), pricecurve(c,seq,'price')-pricecurve(c,seq+1,'price'))<0;
abort$card(errorPC) errorPC;
Variables
x(c) crop planted in acres of land
w(c,seq) crops sold in segment of cost curve in tons
y(c) crops purchased in tons
profit objective variable in dollars;
Positive variables x,w,y;
Equations
profitdef objective function
landuse capacity
bal(c) crop balance;
profitdef.. profit =e= sum(pcs, w(pcs)*pricecurve(pcs,'price'))
- sum(c, cd(c,'cost')*x(c) + cd(c,'pprice')*y(c));
landuse.. sum(c, x(c)) =l= land;
bal(c).. yf(c)*cd(c,'yield')*x(c) + y(c) - sum(pcs(c,seq), w(pcs)) =g= cd(c,'minreq');
* No purchase of crops that don't have a purchase price
y.fx(c)$(cd(c,'pprice')=0) = 0;
w.up(pcs) = pricecurve(pcs,'ub');
model farm_emp /all/;
Set s scenarios / s1*s3 /;
parameter probab(s) / s1 0.25, s2 0.50, s3 0.25 /;
parameter yfac(s) / s1 0.8, s2 1.0, s3 1.2 /;
file emp / '%emp.info%' /; put emp '* problem %gams.i%'/;
put 'jrandvar ';
loop(c, put yf.tn(c); );
put /;
loop(s, put probab(s);
loop(c, put yfac(s));
put /;
);
putclose 'stage 2 yf y w bal profit';
Parameter
srep(s,*) scenario attributes / #s.prob 0 /
s_yf(s,c) yield factor realization by scenario and crop
s_profit(s) profit by scenario
s_w(s,c,seq) crops sold in segment of cost curve in tons by scenario
s_y(s,c) crops purchased in tons by scenario;
Set dict / s .scenario.''
'' .opt. srep
yf .randvar. s_yf
profit.level. s_profit
w .level. s_w
y .level. s_y /;
farm_emp.optcr = 1e-6;
farm_emp.optca = 1e-2;
* The first two options will select the NBD algorithm within Lindo
* The NBD algorithm has the requirement that a random variable can only
* occur once. Otherwise Lindo issues the following message and terminates:
* *** Lindo does not allow multiple occurrences of one RV with NBD method
* Therefore, we have replaced the RV yf (yield factor) that occured in the
* bal equation for every crop c by a a set of joined RV yf(c). The original
* model with a single RV is captured in farmsp.
$onEcho > lindo.opt
STOC_MAP_MPI2LP 1
STOC_METHOD 1
STOC_CALC_EVPI 0
$offEcho
option emp = lindo;
farm_emp.optfile = 1;
solve farm_emp using emp maximizing profit scenario dict;
display srep, s_yf;