基于深度学习的铜转炉终点多任务动态预测模型及应用

Multi-Task Dynamic Prediction Model and Application of Copper Converter End Point Based on Deep Learning

转炉终点判断的命中率直接关系到铜的质量和熔炼效率.针对铜转炉造渣、造铜终点判断智能化程度和命中率低的问题,当前工作提出了一套集转炉工作条件、熔炼参数数据采集和造铜、造渣终点智能预报动态模型.基于深度学习算法,作者提出了时间终点、铜熔池火焰数字化特征、SO_(2)浓度和烟气温度耦合的多任务模型,通过对100炉转炉熔炼历史数据进行训练,结果表明该模型造渣期平均预测误差为2.31%;造铜期平均预测误差为1.97%,终点预测命中率达到97.33%.相对于时间终点法和火焰模型命中率分别提高了18.55%和6.38%,显著提高了预报命中率.本研究在铜转炉终点判断智能化进程中具有积极的指导意义.

The hit rate judged at the end of the converter is directly related to the quality of copper and the efficiency of smelting.Aiming at the problem of low intelligence and low hit rate in determining the slagging and copper end points of copper converters,the current work has proposed a set of dynamic models that integrate converter working conditions,data collection of melting parameters,and intelligent prediction of copper and slagging end points.Based on the deep learning algorithm,we proposed a multi-task model coupled with time endpoint,molten pool flame characteristics,SO_(2)concentration and flue gas temperature.Through training on 100 converter smelting historical data,the results show that the average prediction error of the model during the slag-making period is 2.31%;the average prediction error during the copper production period was 1.97%,and the end-point prediction hit rate reached 97.33%.Compared with the time-end method and the flame model,the hit rate is increased by 18.55%and 6.38%respectively,which significantly improves the forecast hit rate.This research has far-reaching guiding significance in the intelligent process of determining the end point of the converter.