Surrogate optimization algorithms generally consist of four components:
- Optimization problem: All of the available information about the optimization problem, e.g., dimensionality, variable types, objective function, etc.
- Surrogate model: Approximates the underlying objective function. Common choices are RBFs, Kriging, MARS, etc.
- Experimental design: Generates an initial set of points for building the initial surrogate model
- Adaptive sampling: Method for choosing evaluations after the experimental design has been evaluated.
The surrogate model (or response surfaces) is used to approximate an underlying function that has been evaluated for a set of points. During the optimization phase information from the surrogate model is used in order to guide the search for improved solutions, which has the advantage of not needing as many function evaluations to find a good solution. Most surrogate model algorithms consist of the same steps as shown in the algorithm below.
The general framework for a Surrogate Optimization algorithm is the following:
Inputs: Optimization problem, Experimental design, Adaptive sampling method, Surrogate model, Stopping criterion, Restart criterion
1 2 3 4 5 6 7 8 9 10
Generate an initial experimental design Evaluate the points in the experimental design Build a Surrogate model from the data Repeat until stopping criterion met If restart criterion met Reset the Surrogate model and the adaptive sampling method go to 1 Use the adaptive sampling method to generate new point(s) to evaluate Evaluate the point(s) generated using all computational resources Update the Surrogate model
Outputs: Best solution and its corresponding function value
Typically used stopping criteria are a maximum number of allowed function evaluations (used in this toolbox), a maximum allowed CPU time, or a maximum number of failed iterative improvement trials.