DiscoverSim Optimization Methods

Optimization Methods

Back to Table of Contents

DiscoverSim optimization features include:

• Dividing Rectangles (DIRECT) Global Optimization
• Genetic Algorithm (GA) Global Optimization
• Sequential Quadratic Programming (SQP) for Fast Local Optimization
• Powerful hybrid of the above 3 methods
• Stochastic Optimization — minimize dpm or maximize Ppk
• Multiple Response Optimization using Desirability, Weighted Sum or Deviation from Target
• Optimization with Input Controls
• Specify linear and nonlinear Constraints

Components of Optimization

Output: The output will be a function of the model equation, intermediate cell calculations, as well as input distributions and input controls.

Input Control: The permissible range for the control is specified, and the control, which unlike an input distribution, has no statistical variation. Think of this as a control knob like temperature. This is also known as a "Decision Variable". An input control can be referenced by an input distribution parameter, constraint and/or an output function. It is possible to have a model that consists solely of controls with no input distributions. (In this case, the optimization is deterministic, so the number of replications, n, should be set to 1.) An input control can be continuous or discrete integer.

Constraint: A constraint can only be applied to an Input Control and cannot reference an Input Distribution or Output Response. A constraint cannot be a part of the model equation (i.e., an output cannot reference a constraint). Constraints can be simple linear or complex nonlinear.

Each constraint will contain a function of Input Controls on the Left Hand Side (LHS), and a constant on the Right Hand Side (RHS). Optimization will usually occur more rapidly if the initial values of the parameters and/or controls satisfy the constraint. In addition, optimization is much harder with equality constraints than with inequality constraints. Constraints are managed automatically as part of the SQP process; for DIRECT and GA, a penalty function is specified for each constraint - if the constraint is violated, the penalty should be increased. The default penalty function is 100.

Specifying the Optimization Objective Function

The following table gives the options available to specify an optimization objective function in the Run Optimization dialog:

Notes on Optimization:

1. Minimize/Maximize Weighted Sum: minimize or maximize the weighted sum of statistic. Weights are specified for each output in the Output Response dialog. For example, if the selected statistic was the Mean, and there were two outputs, the objective function would be:

Weight₁Mean₁ + Weight₂Mean₂

If there is a single output in the model, this simplifies to minimize or maximize the statistic value. Note that the Output Goal specified in the Output Response dialog is not used here. It is only used in Maximize Desirability.



2. Minimize Deviation from Target: minimize the square root of weighted sum of deviations squared. A target must be specified for each output. The only statistic available is the mean.

SQRT( Weight₁(Mean₁ − Target1)² + Weight₂(Mean₂ − Target₂)² )


3. Maximize Desirability: maximize the weighted linear sum of the Derringer and Suich desirability measure.
• Weight (default = 1).
• Output Goal (Target, Maximize or Minimize).
• If the output goal is Target, then LSL, Target, and USL are required.
  
  
• If the output goal is Minimize, then Target and USL are required.
  
  

The only statistic available for the desirability option is the mean.

4. If the number of replications is n = 1, the statistic used is the mean.


5. Stochastic optimization requires a fixed seed to avoid inconsistent comparisons.

Back to Table of Contents