Dynamic Optimization To Production Scheduling Problem In Acid-Rolling Based On Approximate Dynamic Programming

Acid rolling is located in the medium of iron&steel production process,whose material is supplied from hot rolling by truck.Due to the uncertain arrival time incurred by unqualified quality and so on,the material inventory is not volatile.Consequently,it is difficult to keep the production schedule obtained by classical deterministic scheduling method executable.Research on dynamic optimization method for coil scheduling in acid rolling is beneficial to production costs reduction and improvement of production efficiency.This thesis studies the coil scheduling problem considering arrival time in acid rolling,a mathematical programming model is formulated,and a learning based dynamic scheduling method is proposed.With consideration of arrival time,quality problem for coils,production ability volatility,dynamic coil scheduling in acid rolling is studied.And the approximation dynamic programming method based on myopic policy and the value function approximations learning is designed.The main contents of this thesis are summarized as follows:(1)For the coil scheduling problem considering arrival time in acid rolling,a mixed integer linear model is formulated with the objective of minimizing the production changeover costs and total tardiness.And a learning based scheduling algorithm is developed.In the algorithm,a weight parameter is developed,which is updated during iteration,in order to improve the searching efficiency.Finally,the efficiency of the proposed algorithm is evaluated by computational experiments based on randomly generated instances.(2)For the dynamic coil scheduling in acid rolling,considering the dynamic arrival of acid-rolling coils,the coil cancellation caused by quality defect,and the fluctuation of the production capacity,an approximation dynamic programming algorithm is developed which takes the rolling state as state,the adjacent relationship between coils as decision variables.Moreover,both myopic policy and the value function approximations learning are utilized.Finally,the efficiency of the proposed algorithm is illustrated by comparing with the schedule obtained by solving the case with given arrival time.(3)According to the demands of steel enterprises and the needs of production planners,the acid-rolling dynamic coil optimization decision support system is developed based on the above algorithms,which can display the real-time acid rolling production plan dynamically.By the design of easy-to-use graphical user interface and data management system,the application of this system can significantly reduce the workload and improves work efficiency of production planners.