基于知识与AW-ESN融合的烧结过程FeO含量预测

氧化亚铁(FeO)含量是衡量烧结矿强度和还原性的重要指标, 烧结过程FeO含量的实时准确预测对于提升烧结质量、优化烧结工艺具有重要意义. 然而烧结过程热状态参数缺失、过程参数波动频繁给FeO含量的高精度预测带来巨大的挑战, 为此, 提出一种基于知识与变权重回声状态网络融合(Fusion of data-knowledge and adaptive weight echo state network, DK-AWESN)的烧结过程FeO含量预测方法. 首先, 针对烧结过程热状态参数缺失的问题, 建立烧结料层最高温度分布模型, 实现基于料层温度分布特征的FeO含量等级划分; 其次, 针对烧结过程参数波动频繁的问题, 提出基于核函数高维映射的多尺度数据配准方法, 有效抑制离群点的影响, 提升建模数据的质量; 最后, 针对烧结过程数据驱动模型缺乏机理认知致使模型预测精度不高的问题, 将过程数据中提取得到的FeO含量等级知识与AW-ESN (Adaptive weight echo state network)结合, 建立DK-AWESN模型, 有效提升复杂工况下FeO含量的预测精度. 现场工业数据试验表明, 所提方法能实时准确地预测烧结过程FeO含量, 为烧结过程的智能化调控提供实时有效的FeO含量反馈信息.

FeO Content Prediction in Sintering Process Based on Fusion of Data-Knowledge and AW-ESN

FeO content is an important index to characterize the strength and reducibility of sinter. Real-time and accurate prediction of FeO content in sintering process is of great significance for improving sintering quality and optimizing sintering process. However, the lack of thermal state parameters in sintering process and the frequent fluctuation of process parameters bring great challenges to the high-precision prediction of FeO content. In order to alleviate these problems, a method of FeO content prediction in sintering process by fusing data-knowledge and adaptive weight echo state network (DK-AWESN) is proposed in this paper. Firstly, aiming at the problem of lacking thermal state parameters in sintering process, the temperature distribution model of sinter bed is established, and the state of FeO content can be obtained based on the temperature distribution characteristics of sinter bed; Secondly, aiming at the frequent fluctuation of sintering process parameters, a multi-scale data registration method based on kernel function high-dimensional mapping is proposed, which effectively suppresses the influence of outliers and improves the quality of modeling data; Finally, to alleviate the problem of low prediction accuracy of data-driven model due to the lack of mechanism knowledge, the expert knowledge extracted from the process data is fused with adaptive weight echo state network (AW-ESN) to establish DK-AWESN, which improves the FeO content prediction performance of the model under complex working conditions. Industrial verification shows that the proposed method can accurately predict the FeO content in real time and provide effective FeO content information for the intelligent control of the sintering process.