Fast Task Scheduling With Model Predictive Control Integrating a Priority-Based Heuristic

Abstract

This paper presents a scalable Model Predictive Control (MPC) algorithm for task scheduling and real time re-scheduling. The use case motivating the work is given by the problem of managing the integration activities involved in the final assembly of the Vega rocket at the European space center in Kourou, French Guiana. There are two main objectives. The algorithm shall suggest to the planning operators an optimized scheduling of the activities, i.e., one which minimizes the total completion time (the makespan), while satisfying all the applicable constraints. In addition, the algorithm shall provide in real time an update of the planning, in case some unforeseen events require a re-scheduling of the activities. While a standard application of mixed-integer optimization would not be feasible in practice due to the combinatorial complexity of the problem, the scalable MPC algorithm proposed in this paper retains all the flexibility and modelling power of optimization-based techniques, and is almost as fast as the state of the art scheduling heuristics, which in real scenarios can provide a sub-optimal solution in few seconds, or less. Extensive simulations on randomly generated realistic scenarios are carried out to validate the proposed approach. On average, the proposed MPC algorithm decreased by nearly 2% the makespan, compared to a state of the art scheduling heuristic, while having a comparable solving time, in the order of milliseconds, and while retaining (contrary to heuristics) all the flexibility and modelling power of the optimization based approaches (which took several hours to run on the test scenarios).

Keywords

• Task scheduling
• assembly lines
• makespan reduction
• model predictive control
• global optimization
• mixed-integer programming