Research On Data Mining Based Production Scheduling

Production scheduling plays an important role in the planning and operation of manufacturing systems. Effective scheduling could dramatically optimize the allocation of resources in workshops, reduce the production loss, and improve the production efficiency of enterprises, so as to enhance the enterprises’competitiveness. In the past decades, numerous modeling and optimization methods have been developed for different types of scheduling problems. There is, however, some discontinuity between scheduling practice and the academic research of traditional scheduling models and algorithms. On the other hand, with the development of information technologies, the production styles of manufacturing factories have been changing dramatically. The growth of data in information systems of manufacturing factories, including scheduling related data, has proceeded at an explosive rate. Naturally, data based scheduling has been an emerging area of research and applications. This paper mainly discusses how to use data mining techniques to realize data based scheduling. The innovative points are shown as follows:(1) In discrete manufacturing industries, job shop scheduling problem (JSP) has been considered as one of the most typical scheduling problems. For classic JSP, timed Petri net was used to formulate the scheduling process and an improved Petri net based branch and bound algorithm was developed to solve JSP. On this basis, a novel dispatching rule extraction method was proposed to mine the scheduling knowledge implicit in the Petri net modeling based schedules. The extracted knowledge, formulated by a decision tree, could be used as a new dispatching rule. Besides, a novel method was developed to combine the extracted knowledge with traditional heuristics to gain better performance. Simulation experiments were used to verify the feasibility and validity of the proposed methods.(2) The practical scenarios are filled with dynamic events. For example, the jobs always arrive over time and the information about the jobs arriving in the future is unknown in advance. Therefore, the dynamic job shop scheduling problem is considered. A novel approach was proposed to automatically generate customized rules for dynamic job shop scheduling problems with different operation conditions. Genetic algorithm was used to generate optimal or near-optimal scheduling data. Due to the genetic inheritance, the characteristics of these schedules should be similar. Among them, similar relationships may exist between characteristics of operations and their sequential orders. Based on this, extreme learning machine (ELM) was used to learn new dispatching rules from these schedules. The target concept to learn and the input features for the learning algorithm were also defined based on the characteristics of dynamic job shop scheduling problem. Simulation experiments were used to verify the validity of the proposed method.(3) Up to now, research on data based production scheduling is mainly focused on the scheduling problems in discrete manufacturing industries rather than process industries. This paper explored how to extend data mining techniques to scheduling problems in process industries by using a typical batch scheduling problem-single-stage multi-product scheduling problem (SMSP) as an example. An ELM based scheduling knowledge extraction method was proposed. The extracted knowledge could be used for the prediction of order’s position in a sequence. By combining with existing heuristic rules, it could generate complete schedules for SMSP. Simulation experiments were used to verify the validity of the proposed method.