多目的遗伝的プログラミングを用いた自动回路设计 - Doors.PDF

HE CIENCE AND NGINEERING EVIEW OF OSHISHA NIVERSITY, OL , O T S E R D U V . 49 N . 2 July 2008 Automatic Circuit Design by Multi-Objective Genetic Programming ** * * Yasuyuki YUKI , Toshiji KATO , and Kaoru INOUE (Received March 21, 2008) An efficient genetic programming (GP) method which synthesizes circuits automatically for multi-objective problem is proposed. Conventional methods like the weighting method of multiple objective functions find only one solution which is often not enough to balance trade-offs between the functions. The proposed method finds a set of optimal solutions which can be selected for user’s choice. The SPEA2 (Strength Pareto Evolutionary Algorithm 2) is used to find possible optimal solutions called Pareto solutions which give no further improvement without violating at least one objective function. Furthermore a parallel computing process using a PC cluster system with MPI is used to reduce its processing time. The DRMOGA (Divided Range Multi-Objective Genetic Algorithm) is used to divide the process to find Pareto solutions systematically by allocating one sub-group for each CPU. The proposed method is applied to one test function and four circuit design examples. The results of the test function example validated basic behaviors of the proposed method which finds a set of Pareto solutions. The results of the circuit design examples validated applicability of the proposed method to practical problems and showed how users can select their best solutions out of the Pareto sets. Key words Multi-Objective Genetic Programming, Circuit Design, SPEA2, DRMOGA, MPI , , SPEA2, DRMOGA, MPI