Forecasting policies for scheduling a stochastic due date job shop

This work studies the problem of scheduling a production plant subject to uncertain processing times that may arise, e.g. from the variability of human labour or the possibility of machine breakdowns. The problem is modelled as a job shop with random processing times, where the expected total weighted tardiness must be minimized. A heuristic is proposed that amplifies the expected processing times by a selected factor, which are used as input for a deterministic scheduling algorithm. The quality of a particular solution is measured using a risk averse penalty function combining the expected deviation and the worst case deviation from the optimal schedule. Computational tests show that the technique improves the performance of the deterministic algorithm by 25% when compared with using the unscaled expected processing times as inputs.