Genetic algorithms (GA) can work with both discreteDecisions in which the alternatives are individually distinct. Examples are decisions which are the number of individual items (machines in a factory), an quantity that can be enumerated (machine type) and particularly yes/no decisions. and continuousDecisions that can be expressed as a real numbers representing a quantity that selected in smoothly rather than in steps. decision variables on a wide range of problems. GA differ from other methods by considering populations of solutions rather than individual 'best' solutions. GA are broadly based on evolution. The success of the members of the current population is given by the optimisation cost functionA function that returns a value from a well ordered set for each of the outcomes from a set of decisions. The well ordered set is usually the set of all real number and the cost can often be interpreted in monetary terms.. A new generation of solutions is formed by combining features (decision variable values) from the successful members of the current generation. The process is repeated until the best member of the populations reaches a reasonably steady value (and is similar in value to the best member found in any generation).
Apart from the modelA simplified description providing a basis for an empirical understanding and representation of a system that can be used to forecast its behaviour according to a set of decisions. and cost functions, which are required for any optimisation method, the key aspect required for Genetic Algorithms is how the decision variables are encoded as 'genes'. For the GA to work the decision variable encoding and the representation of this coding must be able to capture features of 'good' sets of decisions. The gene representation must support the process of combining features of parent individuals to form new individuals for the next generation. This combination will include gene crossover (all genes from either parent) and gene mutation (the spontaneous creation of new genes). Premature convergence of the population can be a problem so that while selection of parents is based primarily on success as measured by the cost function there must be a random component in the selection and the retention of dissimilar individuals.
Once the decision variable representation has been formulated, the genetic algorithm optimisation can be developed using one of the publicly available libraries of genetic algorithm objects. These libraries support most of the genetic representations and operators that have been developed so far. The main work required in developing a genetic algorithm optimisation is in the identification of a suitable representation, however programming is also required to enable the one of the GA libraries to be used.
Optimal Solutions 4U has the experience to implement Genetic Algorithm optimisations and can develop applications and modules to your requirements. We would be pleased to offer consultancy on this, or any other aspect of Optimisation. You can get in touch by sending a message from our Contact Us page, or by calling us on the number below.