Multiobjective optimization apparatus, multiobjective optimization method and multiobjective optimization program

A multiobjective optimization method in which parameters of an individual are fed to an optimization target, and said parameters of the individual are optimized on the basis of sampled values of fitnesses for a plurality of fitness functions that correspond to a plurality of objectives which are outputted from said optimization target, said multiobjective optimization method comprising the steps of:

• storing in a storage parameters of individuals and sampled values of fitnesses which are outputted from said optimization target;
• determining estimated values of true fitnesses that correspond to an individual of interest on the basis of the sampled values which are stored in said storage;
• and
• generating a new individual on the basis of the estimated values obtained, and feeding parameters of the generated new individual to said optimization target and said storage, while evaluating, in accordance with a multiobjective evolutionary algorithm, a population to be evaluated on the basis of a plurality of estimated values obtained, thereby determining a Pareto-optimal population;
• wherein
• said step of determining the estimated values includes the step of assigning a weight to the sampled values that corresponds to each individual which is stored in said storage, and determining a linear sum of the weighted plurality of sampled values, thereby determining estimated values of fitnesses that correspond to the individual of interest, said weight for each individual being a function that contains a distance between the individual of interest and the others of said individuals on a parameter space;
• and
• said step of determining the Pareto-optimal population includes the steps of:
• applying, for each of said plurality of fitness functions, dominance comparison on estimated values that correspond to a plurality of individuals of said population to be evaluated, and assigning weights to a plurality of results obtained for each of said plurality of fitness functions, so as to rank the plurality of individuals of said population to be evaluated on the basis of a linear sum of the plurality of obtained results which are weighted for each of said plurality of fitness functions;
• and
• generating a new individual on the basis of an index of distribution that indicates a degree of sparseness on a distribution of individuals at the highest rank of said population to be evaluated on a fitness function space;
• wherein
• said optimization target includes an evaluation system for evaluating a plurality of items of performance of a piece of equipment.

A computer readable storage medium encoded with a computer-executable multiobjective optimization program in which parameters of an individual are fed to an optimization target, and said parameters of the individual are optimized on the basis of sampled values of fitnesses for a plurality of fitness functions that correspond to a plurality of objectives which are outputted from said optimization target, said multiobjective optimization program allowing a computer to execute the processes of:

• storing in a storage parameters of individuals and sampled values of fitnesses which are outputted from said optimization target;
• determining estimated values of true fitnesses that correspond to an individual of interest on the basis of the sampled values which are stored in said storage;
• and
• generating a new individual on the basis of the estimated values obtained, and feeding parameters of the generated individual to said optimization target and said storage, while evaluating, in accordance with a multiobjective evolutionary algorithm, a population to be evaluated on the basis of a plurality of estimated values obtained thereby determining a Pareto-optimal population;
• wherein
• said process of determining the estimated values includes the process of assigning a weight to the sampled values that correspond to each individual which is stored in said storage, and determines a linear sum of the weighted plurality of sampled values, thereby determining estimated values of fitnesses that correspond to the individual of interest, said weight for each individual being a function that contains a distance between the individual of interest and the others of said individuals on a parameter space;
• and
• said process of determining the Pareto-optimal population includes the processes of:
• applying, for each of said plurality of fitness functions, dominance comparison on estimated values that correspond to a plurality of individuals of said population to be evaluated, and assigning weights to a plurality of results obtained for each of said plurality of fitness functions, so as to rank the plurality of individuals of said population to be evaluated on the basis of a linear sum of the plurality of obtained results which are weighted for each of said plurality of fitness functions;
• and
• generating a new individual on the basis of an index of distribution that indicates a degree of sparseness on a distribution of individuals at the highest rank of said population to be evaluated on a fitness function space;
• wherein
• said optimization target includes an evaluation system for evaluating a plurality of items of performance of a piece of equipment.