In these days, there are more and more requirement to make sure to keep schedule in any types of product in Japan.Partly because of supply-chain management becomes popular, and partly because keeping product out time becomes one of effective service for the customer.However, it is not always easy to make scheduling that we can really keep in various situations.It is said that pull type scheduling works well to keep schedule.As it is easy to handle and also it is more robust than optimum scheduling.But to make it more reliable, we need to set certain amount of buffer for each production.Here, buffer means certain amount of time for each miss-estimation of production time at scheduling.Thus, if we make too much buffer on it, schedule becomes more and more robust, but at the same time it reduces product capability.Therefore, we need to set it properly.In this study, we make virtual factory that has 1/8 size of Maruzen Industry, and gave 1/8 orders that this virtual factory must proceed.Then, we make scheduling with (Push or Pull) type with certain amount of buffer for each product.We take type of scheduling and amount of buffer as input (Design variable) and we basically take average of leading time, total amount of extra-working time as basic objective function with deadline breaking as its constraints.As one simulation cost 20 second to 4hours, we use multi-objective approximate optimization using Radial Basis function Network as its solver.From this optimization, we made sure that pull type scheduling is better than push type scheduling that Maruzen Industry usually took, and also we can set appropriate buffer to meet seasonal situation for Maruzen Industry.In this paper, we also would like to report the results of real situation of factory, before and after we had followed the results of the optimization.