Study On Modeling And Optimization For Stochastic Flexible Manufacturing System

Flexible Manufacturing System has the characteristics of resources no uniqueness, the operation can be performed by any available machine in a set of machines. Due to reduce the constraints of the machine, it becomes higher flexible. But it has high complexity and the uncertain factors exist in the actual production system, making its complexity is higher. In today’s competition, how to use computer technology to realize the production scheduling problem of uncertainty environment optimization, become an important issue facing now.Firstly, we focus on three classical evolution algorithms and two modified evolution algorithms with grouping mechanism, and do the experiments under the certain environment on different size of data. We find that as the data growing, the evolution algorithms with grouping mechanism can get a better solution with larger probability. Due to we use average value of 30 times, we have reason to believe that grouping mechanism is the key factor to the improving optimal solutions.Motivated by the experiment results before, Firstly, we choose the particle swarm algorithm for basic algorithm, use the set-based grouping and parameter adaptive adjustment mechanism. It is given a set number of available grouping, choose a grouping number and calculate adaptive value, if adaptive value become better, we will use the grouping number continually; otherwise, when the first time that good adaptive value does not change even worse, we abandon the grouping number, randomly choose another number in the grouping set. Using the algorithm before, we get better solutions and improve its efficiency.Finally, through experiments, we concluded that the proposed hybrid evolutionary algorithm based on critical path grouping mechanism would get better solution than original algorithm and improve robustness of algorithm. Meanwhile, the paper also have objective perspective, that is that we can group the data different from each other, make the whole population into sub-populations, and then make the experiment separately on different and parallel machines in distributed environment, so that not only optimizes the optimal solution, but also enhance the efficiency and shortened the time.