【作者】 蒋朝辉； 黄良； 蒋珂； 方怡静；

【Author】 Zhaohui Jiang;Liang Huang;Ke Jiang;Yijing Fang;School of Automation,Central South University;Peng cheng laboratory;

【机构】 中南大学自动化学院； 鹏城实验室；

【摘要】 FeO含量的实时检测对于烧结过程精细准确调控、稳定烧结矿质量和保证高炉平稳顺行具有重要意义。然而现场对FeO含量的检测采用人工化验的方式,造成FeO含量数据获取的严重滞后,导致烧结矿检测结果无法实时指导烧结配矿操作。因此,为了解决烧结过程FeO含量信息无法实时获取的问题,本文提出了一种基于模糊推理和R-C-LSTM的烧结过程FeO含量预测方法。首先,针对料层内部反应温度无法检测的问题,通过对烧结过程的传热机理进行分析,建立了料层最高温度模型,并结合专家经验,提出了基于模糊推理定性分析烧结过程FeO含量等级的方法;其次,针对烧结过程参数检测频率不一致导致的大量未标签数据的问题,考虑到过程参数检测的连续性和波动性,提出了基于核函数高维映射的核心数据选择的方法;最后,针对烧结过程的时序性和动态性的问题,选用长短期记忆网络(LSTM)建模分析,通过将LSTM的遗忘门和输入门进行耦合以及将烧结过程FeO等级融入LSTM,建立R-C-LSTM预测模型,理论分析和实验对比均表明了本文所建的模型可以实时准确预测烧结过程FeO含量,能给烧结现场操作人员实时提供有效信息,确保烧结过程平稳运行。更多还原

【Abstract】 Real-time detection of FeO content is of great significance for fine and accurate control of the sintering process,stabilizing the quality of sintering ore,and ensuring the smooth and direct blast furnace.However,the on-site detection of FeO content uses manual testing,which causes a serious lag in the acquisition of FeO content data,resulting in sinter detection results that cannot guide sintering and ore blending operations in real time.Therefore,the purpose is to solve the problem that the FeO content information in the sintering process cannot be obtained in real time,this paper proposes a method for predicting the FeO content in the sintering process based on fuzzy reasoning and R-C-LSTM.Frist,Aiming at the problem that the internal reaction temperature of the material layer cannot be detected,the maximum temperature model of the material layer is established by analyzing the heat transfer mechanism of the sintering process,and combined with expert experience,a method for qualitatively analyzing the FeO content level in the sintering process based on fuzzy inference is proposed;Then,Aiming at the problem of a large amount of unlabeled data caused by the inconsistent detection frequency of the sintering process parameters,considering the continuity and volatility of the process parameter detection,a core data selection method based on the high-dimensional mapping of the kernel function is proposed;Finally Aiming at the problem of the timing and dynamics of the sintering process,a long short-term memory network(LSTM) was selected for modeling and analysis.By coupling the forget gate and the input gate of the LSTM and integrating the FeO level of the sintering process into the LSTM,the R-C-LSTM was established.The theoretical analysis and experimental comparison show that the model built in this paper can accurately predict the FeO content in the sintering process in real time,it can give sintered site operators to provide effective real-time information to ensure the smooth operation of the sintering process and sintered ore quality and high yield.更多还原