JavaCplex结果不同于IBM优化工作室

Question

我在IBM优化工作室中实现了一个LP问题，使用OPL来创建模型。验证模型后，我想把它放在Java，为模拟目的脚本参数。我找到了在Java使用我的OLP模型的最快方法，代码如下：

IloOplFactory.setDebugMode(false);
IloOplFactory oplF = new IloOplFactory();
IloOplErrorHandler errHandler = oplF.createOplErrorHandler(System.out);
IloOplModelSource modelSource = oplF.createOplModelSource("myModel.mod");
IloCplex cplex = null;
cplex = oplF.createCplex();
IloOplSettings settings = oplF.createOplSettings(errHandler);
IloOplModelDefinition def = oplF.createOplModelDefinition(modelSource, settings);
IloOplModel opl = oplF.createOplModel(def, cplex);

String inDataFile = "myData.dat";
IloOplDataSource dataSource = oplF.createOplDataSource(inDataFile);
opl.addDataSource(dataSource);

opl.generate();
opl.convertAllIntVars(); // converts integer bounds into LP compatible format
if (cplex.solve()) {
        double obj = opl.getCplex().getObjValue();
        System.out.println("OBJECTIVE: " + obj);
}

现在的问题是，如果我在IBM优化工作室和Java运行“myModel.mod”和“myData.dat”，我会得到非常不同的客观结果。

在IBM Optimization Studio中：

solution (optimal) with objective 125

在Java中：

Parallel mode: deterministic, using up to 4 threads for concurrent optimization.
Tried aggregator 1 time.
LP Presolve eliminated 0 rows and 1 columns.
Reduced LP has 5280 rows, 5325 columns, and 25525 nonzeros.
Presolve time = 0.01 sec. (3.77 ticks)

Iteration log . . .
Iteration:     1   Dual objective     =             0.000000
Iteration:   345   Dual objective     =            90.297455
Iteration:   568   Dual objective     =           117.206047
Perturbation started.
Iteration:   707   Dual objective     =           117.206047
Removing perturbation.
Reinitializing dual norms . . .

Dual simplex solved model.

OBJECTIVE: 117.20608137232513

我看了数据集和我的模型，低于125的目标甚至不可能（我用极值来确保我的目标变量之一是125，所以低于这一点的任何东西都不可能）。

有人知道为什么这些结果不同吗？与IBM相比，Java的设置中是否存在这样的问题？我是否可以将我的IBM optimization studio设置导入Java来测试这一点？

非常感谢。

编辑：这是IBM Optimization studio日志，我忘了包括这些日志。它确实显示了更多关于整数被截断的信息，但这是我第一次使用这两个程序，所以有人能帮我找到问题所在或这一切意味着什么吗？

Found incumbent of value 125.000000 after 0.00 sec. (0.53 ticks)
Tried aggregator 1 time.
MIP Presolve eliminated 5027 rows and 4764 columns.
MIP Presolve modified 1191 coefficients.
Reduced MIP has 253 rows, 562 columns, and 1958 nonzeros.
Reduced MIP has 562 binaries, 0 generals, 0 SOSs, and 0 indicators.
Presolve time = 0.02 sec. (8.46 ticks)
Probing fixed 8 vars, tightened 0 bounds.
Probing time = 0.00 sec. (2.86 ticks)
Tried aggregator 1 time.
MIP Presolve eliminated 158 rows and 306 columns.
MIP Presolve modified 293 coefficients.
Reduced MIP has 95 rows, 256 columns, and 632 nonzeros.
Reduced MIP has 256 binaries, 0 generals, 0 SOSs, and 0 indicators.
Presolve time = 0.02 sec. (1.73 ticks)
Probing fixed 6 vars, tightened 0 bounds.
Probing time = 0.00 sec. (0.12 ticks)
Tried aggregator 1 time.
MIP Presolve eliminated 29 rows and 30 columns.
MIP Presolve modified 2 coefficients.
Reduced MIP has 66 rows, 226 columns, and 522 nonzeros.
Reduced MIP has 226 binaries, 0 generals, 0 SOSs, and 0 indicators.
Presolve time = 0.00 sec. (0.42 ticks)
Probing time = 0.00 sec. (0.07 ticks)
Clique table members: 34.
MIP emphasis: balance optimality and feasibility.
MIP search method: dynamic search.
Parallel mode: deterministic, using up to 4 threads.
Root relaxation solution time = 0.00 sec. (0.36 ticks)

    Nodes                                         Cuts/
   Node  Left     Objective  IInf  Best Integer    Best Bound    ItCnt     Gap

*     0+    0                          125.0000      123.0000             1.60%
      0     0      123.8469     4      125.0000      123.8469       62    0.92%
      0     0        cutoff            125.0000                     81    0.00%
Elapsed time = 0.09 sec. (17.09 ticks, tree = 0.00 MB, solutions = 1)

Zero-half cuts applied:  1
Gomory fractional cuts applied:  1

Root node processing (before b&c):
  Real time             =    0.09 sec. (17.22 ticks)
Parallel b&c, 4 threads:
  Real time             =    0.00 sec. (0.00 ticks)
  Sync time (average)   =    0.00 sec.
  Wait time (average)   =    0.00 sec.
                      ------------
Total (root+branch&cut) =    0.09 sec. (17.22 ticks)

编辑2：我发现矩阵中的整数值为0。。1不是四舍五入为0或1，而是计为0.932。。。如何强制Java cplex对整数进行四舍五入？

已解决：它是“opl.convertalintvars（）；”这把一切都变成了双打。我从示例代码中删除了它，现在一切正常。

Answer 1

你的模型在某些方面会有所不同。我怀疑整数变量可能被处理的方式——值125是一个整数，而117.2显然不是整数，日志显示它是作为一个直LP求解的。

查看两个模型解决方案中变量的值，并找到值不同的地方。希望两个解决方案中的许多变量值相似，这将有助于缩小搜索范围。

尝试将模型导出为LP文件或两者中的类似文件，并比较生成的模型。可以在CPLEX命令行解算器中解算两个导出的模型，并比较日志和结果。

尝试从另一个模型中的一个解决方案中修复一些值，然后重新求解。

另外，给我们提供更多来自OPL模型的日志输出。