Configuring the Approximation Loop Strategy

You can control options such as the maximum number of sup-optimization runs and the execution time.

  1. Double-click the Exploration component icon .
    The Exploration Component Editor appears.
  2. From the Exploration Component Editor Exploration Strategy list, select Approximation Loop.
  3. In the Approximation Loop Options area, enter or select the following:
    OptionDescription
    Maximum sub-optimization runs The number of sub-optimization runs that the strategy will perform.
    Note: If the strategy converges before this number is reached, the execution will terminate. Each sub-optimization run requires one exact simulation process flow execution for verifying the results of the approximate optimization and checking the approximation errors. Several more exact executions may be required if the approximation is re-initialized.
    Maximum execution time, hrs This value sets a time limit on the Approximation Loop strategy execution. If you do not set a time limit, the algorithm does not check execution time. Execution terminates based on other settings such as Maximum sub-optimizations runs or Maximum number of subflow runs.
    Use multi-objective optimization This option enables the multi-objective optimization mode of the Approximation Loop. The optimization technique used by the Approximation Loop will be changed to AMGA when this option is selected.
  4. In the Execution Options area, select Enable parallel subflow execution to enable Isight to execute exact simulation process flow runs in parallel when appropriate.
  5. If desired, click Advanced Options, and enter or select the following:
    OptionDescription
    Maximum exact subflow runs (optional) This value sets a limit on the allowed number of exact simulation process flow executions. If this number is reached, the strategy execution terminates. If you do not enter this value, the algorithm uses other stopping criteria (e.g., Maximum sub-optimization Runs), or it can converge to a local optimum point and terminate when the change in ObjectiveAndPenalty is below the convergence tolerance.
    Important: The strategy execution is not terminated until the approximation initializes and at least one sub-optimization executes.
    Initial relative design space size This value controls the size of the initial local design space in relation to the global design space defined by the bounds on all design variables.
    Minimum relative design space size The value below which the relative size of the local design space will not be reduced during the strategy execution.
    Design space reduction factor The multiplier used for reducing the local design space when an approximate optimum point found by a sub-optimization run is not an improvement over the starting design point or when approximation errors are above the acceptable level.
    Design space expansion factor The multiplier used for expanding the local design space when an approximate optimum point found by a sub-optimization run is an improvement over the starting design point and the size of the local design space is at its minimum limit.
    Convergence tolerance The value at which Isight terminates the strategy execution. When the relative change in the ObjectiveAndPenalty parameter is below this value for a certain number of sub-optimization runs, Isight terminates the strategy execution.
    Number of converged cycles for termination The number of converged sub-optimization runs required to declare the strategy converged and terminate execution. When the relative change in the ObjectiveAndPenalty parameter is below the convergence tolerance for the number of sub-optimization runs equal to the value of this option, Isight terminates the strategy execution.
    Acceptable relative approximation error The value that Isight uses to re-initialize the approximation and to reduce the local design space. When the relative approximation errors on the output parameters exceed this value, Isight re-initializes the approximation and reduces the local design space size.
    Use fixed random seed When this option is selected, the Exploration component uses a fixed seed value for the random number generator. If you do not enter a value, the clock time is used as the seed value. The Random number generator is used for generating sampling points for the approximation used by the Approximation Loop.
    Always accept new approximate optimum design This option forces the strategy to always accept a new optimum point after sub-optimization, even when exact analysis shows that it is not an improvement over the starting point. This can help make progress toward an improved design in difficult cases when small constraint violations prevent the strategy from moving in the correct direction.
    Always re-initialize approximation before sub-optimization This option forces the strategy to always re-initialize the approximation with a new set of design points before every sub-optimization. This can help to progress toward an improved design in difficult cases at the expense of a higher number of exact simulation process flow executions.
    Do not store approximate subflow runs in database This option allows Isight to skip all approximation executions and to not store them in the database.
    Note: If you select this option, skipped approximation runs are not included on the Runtime Gateway History tab. Only exact simulation process flow executions are visible.
  6. To close the Advanced Options dialog box and to return to the Exploration Component Editor, click OK.
  7. Click OK to save your changes and to close the Exploration Component Editor.