Tri-objective optimization of noisy dataset in blast furnace iron-making process using evolutionary algorithms

Data-driven models are used to solve many objective blast furnace problems by using evolutionary approach. The process variables and the objectives of the blast furnace are highly complex and hence, solving of these multi-objective optimization problems is not an easy task. Here, a deep evolutionary neural network (EvoDN2), bi-objective genetic programming (BioGP), evolutionary neural net (EvoNN) and constrained version reference vector evolutionary algorithm (cRVEA) are used to solve these optimization problems. Pareto tradeoffs between various conflicting objectives such as gas velocity in tuyere, total heat loss, productivity, heat loss owing to tuyere cooling, coke rate, heat loss during plate cooling, and the amount of gas flow in the blast furnace and the carbon requirement per ton of hot metal, etc. are determined and subsequently used to solve tri-objective optimization problems pertinent to the process. The optimized results are compared against a commercial software KIMEME with multiple modules. The results obtained and the performances of these models are analyzed in three-dimensional space and their influence on the iron-making process is discussed.