Solution Evaluation

Evaluating solutions (i.e., configurations of the base-level metaheuristic) is a critical component of the meta-optimization process in Evolver and it is performed by class MetaOptimizationProblem. This section explains how configurations are evaluated and how their quality is assessed.

As any other multi-objective optimization problem in jMetal, class MetaOptimizationProblem has a method evaluate(solution) that takes a solution as a parameter and returns the same solution with the objective values set.

The evaluation of a solution is as follows:

1. For each solution generated by the meta-optimizer, a number \(N\) of independent runs of the base-level metaheuristic are performed on the set of \(P\) base-level optimization problems. This ensures that the results are statistically robust.

2. The resulting fronts from each run are normalized with the reference fronts (if available) before computing the quality indicators.

3. For each quality indicator, the median of the quality indicators for each problem is calculated.

4. The mean value of the medians of each problem (i.e., the mean of the \(P\) values) is the resulting objective value.

The parameter \(N\) is a parameter of the meta-optimization problem and it must be set carefully to avoid an excessive running time. By default, \(N\) is set to 1.