Meeting Agenda 02-11-14 (1) Overview (概观) (2) Genetic Algorithm and Levenberg-Marquardt Algorithm for SEAS scenario (遗传算法和莱文贝格 - 马夸特方法 结果) (3) Priority Rule (零件优先级) (4) Optimization (优化) (1) Overview • Previous Week and Current Week: – Implemented SA + LM for Neural Network weight training. (使用退 火法+莱文贝格 - 马夸特方法训练人工神经网路) – Compared the results obtained by using SA + LM to those obtained by just using LM. (成果比较) – Implemented GA + LM for SEAS case study. (遗传算法+莱文贝格 马夸特方法结果) – Included a priority rule in the feasibility check. (建立零件优先级) • Next Week: – Continue our initial research on different optimization algorithms (持续研究不同的优化算法) (2) Genetic Algorithm and Levenberg-Marquardt Algorithm (遗传算法+莱文贝格 - 马夸特方法) • Simulation results with 28 scenarios for training, validation and testing • Refer to PDF (3) Priority Rule (零件优先级) • Ensures that higher priority work orders are processed before lower priority work orders. • Please see figure on next page. 4 5 6 11 12 1 2 3 32 13 34 7 8 9 10 14 15 16 17 18 19 20 21 33 22 23 24 25 26 27 28 29 30 31 Six priority rules are established: - 7 and 8 need to be completed before 11 starts - 9 and 10 need to be completed before 12 starts - 4-6 and 11-12 need to be completed before 13 starts - 1-3 and 13 needs to be completed before 32 starts - 14-21 needs to be completed before 33 starts - 22- 33 need to be completed before 34 starts (4) Optimization (优化) Inputs/Outputs (given, decision) Trained weight values/neural network Neural Network Optimization Inputs (given) Genetic Algorithm/Particle Swarm Generation of suggested decision Trained Neural Network Repeat until stopping criteria satisfied KPI/Optimized Schedule