Experience in using and improving the usability of mathematical modeling system of production at a metallurgical enterprise

Due to the world trade development, metallurgical plants have a great variability in preparation of the production plan. At EVRAZ West Siberian Metallurgical Combine JSC (JSC “EVRAZ ZSMK”), the charge is optimized from 110 components only in terms of agglomeration and blast furnace production. The product range consists of more than 2000 units and varies from month to month. Usually, the production plan is optimized only within individual redevelopments. The purpose of optimization is to minimize the cost of conversion and to maximize production. The article describes development and implementation of a mathematical modeling system of production of the entire metallurgical plant on the example of JSC “EVRAZ ZSMK”. In contrast to the existing conversion modeling systems, the goal of the system is a one-time end-to-end optimization of all the plant’s conversions. The ultimate goal is to maximize the profit of the entire plant. During the operation of the new system in 2019 – 2020, high labor costs were found during the work of the users. For example, more than 10,000 test iterations of calculations were performed for the release of 60 plans for the year and calculation of 30 economic cases. A form of work statistics was developed and analyzed, which showed the following main problems: unsolvability of the model due to the input of mathematically incorrect data; repeated calculations of economic cases to isolate and interpret individual changed factors of the plan from the budget; errors at the data verification stage due to incorrect chemical composition of elements or incorrect model settings for a specific period of work. To eliminate the shortcomings of the system, the authors developed the validation systems for input data at the stage before and after calculations, sensitivity analysis and factor calculation modules and automatic chemical composition filling system. Also the possibility of retraining the model on historical data was added.