CONCLUSION and future work
This work proposed the use of WMS for future optimization by similarity.
The method intends to find the experimental region where a model with
desirable characteristics is a good descriptor of the data at hand. An
evaluation case using function AOG_1 was presented. According to
these results, the method demonstrates the potential to find regions of
similarity between two responses where optimality can be a pattern of
interest.
The evaluations of the method in seven unconstrained global optimization
test functions served to show the use of window of maximum similarity in
examples of functions with different shapes. Also, it was observed in
the evaluations that the WMS method potentially detected zones of
maximum similarity between the different test functions and a quadratic
function.
According to these results, the method demonstrates the potential to
find regions of similarity between two responses where optimality can be
a pattern of interest and can be a useful tool for exploration of
simulated data to find, at least a local optimum.
In many cases, the WMS obtained by the method were limited to take the
minimum size or epsilon value assigned, which is why future work
includes:
- Substitute single composite objective by multiple criterion
optimization, as presented in [5].
- Include more variables to evaluate test functions on.
- Experiment using alternative metamodels with different forms.
- Use design of experiment to sample from the experimental region.