A Multi-Agent And Extremal Optimization System For Integrated Scheduling Of Steel Production

From a system point of view, an integrated steel mill can be modeled as a complex network with multi-stages' mass and energy flows as well as the buffer areas. The production network starting from hot iron metal to steelmaking, continuous casting (CC) and finally to hot strip mill (HSM) can be a good example in both methodology and application research. There are critical criteria and constraints in the capacity and steel grade of each production stage. The real complexities in production management and manufacturing come from the "market-driven" or "make-to-order" business model, the significant differences or unbalances between the required job processing times at each production line or stage, the huge losses in energy and/or productivity due to the idles of either jobs (waiting for machine) or machines (waiting for incoming jobs), and the generation of work-in-processing (WIP) inventories. Consequently, generating reasonably well schedules for BOF, CC and HSM respectively will not provide satisfactory results in reductions in both energy losses and WIP, and job flow smoothness.This thesis focuses on the research and development on the integrated scheduling of "BOF-CC-HSM" with the major goal in increasing the hot charge rate of the slabs directly transferred from CC to the HSM. The main contents studied in this thesis are as follows:Firstly, this thesis builds up individual math models of steelmaking, continuous casting and hot rolling from real production line, mapping to a kind of prize collecting traveling salesman problem (PCTSP). Considering time and slabyard, the math formulation for the whole production line from steelmaking to casting and hot rolling is given. To deal with this typical hybrid-multistage-constrained optimization problem, a multi-agent system associated with extremal optimization method is proposed. The schedule involves horizontal production flows along timelines on each process and vertical flows of materials from upstream to downstream.For horizontal scheduling of each process, a self-organized criticality (SOC) oriented extremal optimization (EO) method is employed. Experimental results show that EO is effective in solving round scheduling of hot rolling process with high efficiency. Results from EO are considered as the initial solutions for the integrated scheduling system.To deal with vertical job matching of the whole production line, this thesis proposes a multi-agent system model. In this model, each agent is in charge of the scheduling jobs of their own process. Initial solutions for each process are generated by agents using rule based method or EO method respectively. Furthermore, preferable solution for the whole scheduling system would be given under the cooperation and coordination of agents.Finally, a software system prototype is developed and simulations with real world production data are studied. The experimental results show that the multi-agent with extremal optimization method is able to provide entire production scheduling solutions for all sequences of heat, cast and rolling subject to the relevant constraints. The significant benefits could be provided through increasing hot charge rate, better mass flow balance and lower inventory costs.