|
Vars.
The variable being optimised is Opt.
|
Curr, and then we apply
reduced cost pruning to each variable in the list. This is achieved
as follows:
|
cons5, which prevents
it from changing so far that the optimum Opt exceeds its upper bound.
For maximisation problems a different constraint would be imposed.post goal.
This is a goal that is executed immediately after each waking of the
linear solver.
|
reduced_cost_pruning/2
available in the eplex library.)overlap
constraints used in the first example above. All the constraints can
therefore be handled by eplex alone.
However the probing algorithm does not send the resource constraints
to eplex.
Instead it takes the start times returned from the optimal eplex
solution, and computes the associated resource profile.
The resource bottleneck is the set of tasks running at the time the
profile is at its highest.before constraint
(defined in the example above) between one task and the start time of
another.before constraints, and so the
algorithm is complete and terminating.post goal as follows:
|
Three kinds of information can be used Reduced costs allow values to be pruned from variable domains. The solution can be checked for feasibility against the remaining constraints, and even if infeasible can be used to support search heuristics. Dual values are used in other hybridisation forms, devised by the mathematical programming community.
- Reduced Costs
- The solution (the value for each variable at the linear optimum)
- Dual values
Figure 17.4: Using information returned from the linear optimum