Intelligent Optimization Methods For Steelmaking-continuous Casting Batch Planning

Steelmaking & continuous casting production process in modern large steel plant is constituted by multiple converters, refining furnaces, casters. First, Molten steel in converters was poured into ladle (we call molten steel in a converter with a charge). Then ladle filled with molten steel was transported to refine process or directly to the continuous casting process. Last the molten steel was poured into tundish and cast into slabs with a certain width, length, thickness by continuous caster. We call the total molten steel cast by continuous caster from its start to end as a cast.For satisfying the capacity of downstream and continuously casting the steel molten, the plant need to make steelmaking & continuous casting batch planning and scheduling. On the basis of the known information about the contract properties, the size properties and processing properties, etc and the production capasity of steelmaking & continuous casting and the requirement of downstream process, the batch planning’s targets is determined the relation between slab and charge, the sequence of charges in cast, and the casting width of slab, subjecting to the constraints of volumn of converter, the service time of the tundish, the ingredient in charge, tundish, cast. On the basis of the cast plans, the steelmaking-continuous casting production scheduling aims at determining the heats in the cast plan production machines including converter, refining furnace, and their starting and end process time corresponding converter, refining furnace and caster. The target that the cast plan is carried out punctually, the heats in the cast is cast on caster continuously, the heats production time cannot be conflicted mutually in the same machine, and in front of the caster, waiting time of the heats do not exceed given threshold, subjecting to the constraints of fixed production process, fixed production time in each equipment and transporting time between different machines, eventually obtaining the production timetable (is called the schedule).Because of the following difficulties, steelmaking & continuous casting batch planning cannot be solved by existing optimal methods:1) the constraint about casting width in charge planning and tundish cannot satisfy the conformance requirements in casting planning; 2) the constrains and objectives about the production capasity of steelmaking & continuous casting conflict to the constrains and objectives about the requirement of downstreams. In existing literature, some shortcomings can be found:1)The relationship between the charge planning, tundish planning and cast planning is not thought.2) the relation between supply and demand between steelmaking & continuous casting and downstream is not considered.3) slab width is considered the fixed value. So planner adopts three steps:charge planning, tundish planning, casting planning. But the method adopt by the planner cannot improve the efficiency of equipment.Under the support of the National 863 high-tech projects "metallurgical industry MES key technology research and demonstration (EMS-EAM-IPS) application (2004AA412010)", we consider a steelmaking-continuous casting production line in a large iron and steel enterprises as the background, and make a study of steelmaking continuous casting batch planning method, and the main result are as follows:1.The overall optimization strategy consisting of charge planning,tundish planning,cast planning was presented.The different requirement of width in charge planning, tundish planning,cast planning were considered as the performance indicators and constraints equations respectively. The processing capacity of targets and its range of the converter and refining furnace, the target weight and its range of requirement for hot rolling process and its downstream, were considered as performance indicators and constraint equations in tundish planning. Based on the iterated local search (ILS), variable neighborhood search (VNS), and ant colony optimization, hybrid optimal strategy was adopted.2 We build a charge planning optimal model by considering the following objective and constraints:minimize the number of charges, minimize the number of high priority of slabs not production, minimize the difference between the slabs in charge in. width and ingredient, the total weight of slabs in a charge not exceeding the volume, and satisfying the jump range and the time of width. Based on the iterated neighborhood search algorithm and variable neighborhood search algorithm, we present a hybrid algorithm as ILS VNS.3 We consider the lifetime of each tundish, the jump range and the times of width, the specifics of the melton steel using tundish as constraints, minimize the difference between the total number of charges and the its goal, minimize the difference between the total weight of hot material slabs and its goal, minimize the total weight of the slabs each downstream required and their goals, minimize the number of tundishes, minimize the leftover of each tundish, minimize the difference between charges in tundishes as objectives, build a multi-objective tundish planning model. Designing a two layer ILSVNS with adjustable weight and the Utilization coefficient of tundish solves the model.4 Considering the life time of each tundish, the specific of using tundish, the jump range and the times of width in each tundish, the jump range of width in each cast as constraints, considering minimizing the number of cast, minimizing the difference of width and ingredient between tundish in each cast as objectives, considering the sequence of charges in each tundish, the sequence of tundishes in each cast, and the width as decision variable, we build cast planning model with multi-objective.Base no ACO, we design a two layers ant colony algorithm5 For using the methods mentioned above, we develop a simulation system, including the modules:charge planning making and querying the result, tundish planning making and result querying, cast planning making and result querying, system management, data maintance. The system has the characteristics of modular, friendly interface, good scalablity. Using multiple sets of real data, we do the experiments research for the method. Compared with the artificial planning, applying our proposed methods can at least reduce a tundish, start-stop frequency of the continuous caster, the number of adjusting the width, and the number of adjusting ingredient.