| 摘要 |
PROBLEM TO BE SOLVED: To efficiently provide an excellent optimum parameter by suppressing influence by drift, and variation of a characteristic, or increase of a learning time by use of an actual machine while excluding an approximation error of a simulation model by use of the actual machine. SOLUTION: This automatic adaptation system automatically operates the actual machine to calculate an evaluation value, and a learning calculation part 50 optimizes a control parameter by use of multipurpose genetic algorithm. The learning calculation part 50 has: an adaptability calculation part 52 calculating individual adaptability; a parent selection part 53 deciding relative merits of a solution by use of an index according to a degree of density with another individual when the adaptability is similar when selecting a parent individual; a generation change part 54 generating a child individual from the parent individual; a child individual evaluation part 55 selectively evaluating the child individual a plurality of times; and a life selection part 56 selecting the individual stored in next generation. The learning calculation part 50 suppresses the influence of the drift, the variation of the characteristic or the increase of the learning time by the use of the actual machine to provide efficiently the excellent optimum parameter at high speed. COPYRIGHT: (C)2008,JPO&INPIT
|
