Sequence-dependent setup time estimation in tactical production planning using machine learning

Tactical production planning with sequence-dependent setup times is a frequently studied topic in literature. Production schedule needs to be known to determine the total sequence-dependent setup time required in a period, which accounts for the capacity loss due to setup operations. Determining the schedule with minimum capacity loss and lot sizes for a multi-period tactical plan simultaneously is the main challenge in literature. This study aims to introduce a tactical production planning model to determine production amount and inventory levels with more accurate capacity loss estimation due to sequence-dependent setup times. Production amounts in tactical model are discretized into equally sized buckets to generate production mixes. Production mixes are used to estimate the sequence-dependent setup times by taking which products are produced and the amounts they are produced in to account. A scheduling model is used to determine the optimal setup times for a set of randomly generated production mixes and machine learning methods are used to estimate for the rest of the production mixes due to computational complexity. The estimation methodology introduced in this study is compared with a baseline and a basic estimation model based on the accuracy of estimating the total capacity loss due to sequence-dependent setup times.