我正在尝试使用AMPL来模拟问题，我希望能够看到替代方案，或者多个＆＃34;最佳或接近最优的＆＃34;解决方案。
我在这个网站上看到：http://orinanobworld.blogspot.com/2011/02/finding-multiple-solutions-in-binary.html
我试图将这种类型的东西放到我自己的模型中
并尝试实施以下内容：
set RECORDS; # items available to choose from
var Picked {RECORDS} binary; # the variables that were set to 1 for "picked"
#other conditions and model constraints....
# we now create some structure to record multiple solutions
param want := 5; # number of solutions we want
param nfound default 0; # number of solutions found
set ALTS {1..100} in {RECORDS}; # records which items were packed (and where) in each solution
# we add constraints to the previous model to exclude solutions we've already seen
s.t. Exclude {s in 1..nfound}:
sum {i in ALTS[s]} (1 - Picked[i]) + sum {j in {RECORDS} diff ALTS[s]} Picked[j] >= 1;
# solve until either the problem becomes infeasible or the right number of solutions have been found
for {s in 1..want} {
solve; # solve the model
if solve_result = 'infeasible' then break; # if the model has become infeasible, give up
#packed is getting the value of players that are in a lineup that have a "1" (>.5)
let ALTS[s] := {i in RECORDS : Picked[i] > 0.5};
let nfound := s; # bump the counter for solutions found
display s;
display ALTS[s];
}
当我使用此代码运行我的模型时，它给出了5个完全相同的解决方案。我究竟做错了什么？作为一个单独的问题，似乎Picked也会在其中设置一些非二进制值（例如0.7235），即使我认为将其设置为二进制会将其限制为1或0。
答案 0 :(得分：1)
您获得分数解决方案的事实表明您正在解决放松而不是MIP问题。例如，如果您使用不支持MIP的解算器（例如MINOS），通常会发出如下警告：
MINOS 5.51: ignoring integrality of 5 variables
并且您获得相同的解决方案，因为约束Exclude仅适用于二进制解决方案。如果使用CPIP或Gurobi等MIP求解器，则应获得不同的二进制解法。