A multi-objective optimization model based on long short-term memory and non-dominated sorting genetic algorithm II

Abstract

Addressing the conflicts of continuous ironmaking and intermittent steelmaking, the steel production process is always hard to be regarded as a whole. Most of the researches focus on the ironmaking process which is a complicated nonlinear system characterized by large scale, big hysteresis, strong distribution and random. In order to solve these difficulties, many researchers tried to propose models to describe the status of the blast furnace(BF). However, seldom have they succeeded due to the lack of direct measurement of key indexes and enough data of abnormal conditions. Because of the development of detection technology and computer science, researchers can easily construct data-driven models without knowing exactly what happens in the BF. While the goal is to keep the economical and environmental cost at the lowest level, a multi-objective optimization model based on long short-term memory(LSTM) neural network and non-dominated sorting genetic algorithm II(NSGA-II) is proposed to optimize the operating parameters in the BF. The model is mainly composed of two parts, one is the LSTM that serves as the mapping rule between the input and output variables, playing the role as a fitness function. The other is the NSGA-II, whose function is to optimize the model and search for the Pareto optimality and its corresponding solution set. Furthermore, the input variables are consist of four aspects, which are the quality of the coke, the composition of ore, the utilization of BF gas and the balance of basicity. Besides, the output variables have three parts, including the iron yield, fuel consumption and iron quality. Even though the safety is the first priority in the ironmaking plant, strange change of operating parameters is not allowed. With the help of this model, the field engineers can keep the BF profitable and safe simultaneously.