The OPTLSO Procedure

The OPTLSO Algorithm

The OPTLSO algorithm is based on a genetic algorithm (GA). GAs are a family of local search algorithms that seek optimal solutions to problems by applying the principles of natural selection and evolution. Genetic algorithms can be applied to almost any optimization problem and are especially useful for problems for which other calculus-based techniques do not work, such as when the objective function has many local optima, when the objective function is not differentiable or continuous, or when solution elements are constrained to be integers or sequences. In most cases, genetic algorithms require more computation than specialized techniques that take advantage of specific problem structures or characteristics. However, for optimization problems for which no such techniques are available, genetic algorithms provide a robust general method of solution.

In general, genetic algorithms use some variation of the following process to search for an optimal solution:

initialization:  

An initial population of solutions is randomly generated, and the objective function is evaluated for each member of this initial iteration.

selection:

Individual members of the current iteration are chosen stochastically either to parent the next iteration or to be passed on to it such that the members that are the fittest are more likely to be selected. A solution’s fitness is based on its objective value, with better objective values reflecting greater fitness.

crossover:

Some of the selected solutions are passed to a crossover operator. The crossover operator combines two or more parents to produce new offspring solutions for the next iteration. The crossover operator tends to produce new offspring that retain the common characteristics of the parent solutions while combining the other traits in new ways. In this way, new areas of the search space are explored while attempting to retain optimal solution characteristics.

mutation:

Some of the next-iteration solutions are passed to a mutation operator, which introduces random variations in the solutions. The purpose of the mutation operator is to ensure that the solution space is adequately searched to prevent premature convergence to a local optimum.

repeat:

The current iteration of solutions is replaced by the new iteration. If the stopping criterion is not satisfied, the process returns to the selection phase.

The crossover and mutation operators are commonly called genetic operators. Selection and crossover distinguish genetic algorithms from a purely random search and direct the algorithm toward finding an optimum.

There are many ways to implement the general strategy just outlined, and it is also possible to combine the genetic algorithm approach with other heuristic solution improvement techniques. In the traditional genetic algorithm, the solution space is composed of bit strings, which are mapped to an objective function, and the genetic operators are modeled after biological processes. Although there is a theoretical foundation for the convergence of genetic algorithms that are formulated in this way, in practice most problems do not fit naturally into this paradigm. Modern research has shown that optimizations can be set up by using the natural solution domain (for example, a real vector or integer sequence) and by applying crossover and mutation operators that are analogous to the traditional genetic operators but are more appropriate to the natural formulation of the problem.

Although genetic algorithms have been demonstrated to work well for a variety of problems, there is no guarantee of convergence to a global optimum. Also, the convergence of genetic algorithms can be sensitive to the choice of genetic operators, mutation probability, and selection criteria, so that some initial experimentation and fine-tuning of these parameters are often required.

The OPTLSO procedure seeks to automate this process by using hybrid optimization in which several genetic algorithms can run in parallel and each algorithm is initialized with different default option settings. PROC OPTLSO also adds a step to the GA; this step permits generating set search (GSS) algorithms to be used on a selected subset of the current population. GSS algorithms are designed for problems that have continuous variables and have the advantage that, in practice, they often require significantly fewer evaluations than GAs to converge. Furthermore, GSS can provide a measure of local optimality that is very useful in performing multimodal optimization.

The following additional "growth steps" are used whenever continuous variables are present:

local search selection:

A small subset of points are selected based on their fitness score and distance to (1) other points and (2) pre-existing locally optimal points.

local search optimization:

Local search optimization begins concurrently for each selected point. The only modification made to the original optimization problem is that the variables’ lower and upper bounds are modified to temporarily fix integer variables to their current setting.

These additional growth steps are performed for each iteration; they permit selected members of the population (based on diversity and fitness) to benefit from local optimization over the continuous variables. If only integer variables are present, this step is not used.