降噪自编码神经网络下高炉冶炼质量在线预测

Online Prediction of Blast Furnace Smelting Quality under Noise Reduction Autoencoding Neural Network

高炉冶炼过程中,其内部发生的复杂反应会对产出的铁水质量产生重要影响,为了调整冶炼参数,并实时掌握产品质量变化趋势,有效提高冶炼产品的质量,提出降噪自编码神经网络下高炉冶炼质量在线预测方法。根据烧结工艺和烧结矿的同化性、液相流动性和粘结相强度等基础特征,确定烧结矿质量评价指标及包含9大类参量的主要工艺参数。基于神经网络并行处理、分布式存储和自适应性强等优势条件,改善神经网络模型对输入数据的泛化性,获取降噪自编码器代价函数,结合激活函数提取隐含层中任意神经元残差,建立液相神经元模型,得到降噪自编码网络神经元,以误差反向传播算法作为神经网络学习方法,通过输出误差实现层级之间的逆传播,确定学习步骤和学习模式,构建烧结矿质量在线预测模型,为了提高预测精度,定义学习速率修正预测模型,实现烧结矿质量预测。实验结果表明,采用所提方法对高炉冶炼质量进行在线预测后,烧结矿碱度预测误差较小,预测结果可信程度较高,预测时间较短,具有良好的预测能力,能够实现实时反馈。

During the smelting process of a blast furnace,the complex reactions that occur internally can have a significant impact on the quali-ty of the produced molten iron.In order to adjust smelting parameters,real-time grasp the trend of product quality changes,and effectively im-prove the quality of the smelting products,a noise reduction self coding neural network based online prediction method for blast furnace smel-ting quality is proposed.Based on the basic characteristics of sintering process and the assimilation,liquid phase fluidity,and bonding phase strength of sintering ore,the quality evaluation indicators of sintering ore and the main process parameters including 9 categories of parameters are determined.Based on the advantages of neural network parallel processing,distributed storage,and strong adaptability,the generalization of input data in the neural network model is improved,and the cost function of denoising autoencoder is obtained.The activation function is combined to extract any residual neurons in the hidden layer,and a liquid phase neural element model is established to obtain denoising au-toencoder network neurons.The error backpropagation algorithm is used as the neural network learning method,by outputting errors to achieve backpropagation between levels,learning steps and modes are determined,and an online prediction model for sinter quality is constructed.In order to improve prediction accuracy,a learning rate correction prediction model is defined to achieve sinter quality prediction.The experi-mental results show that after using the proposed method for online prediction of blast furnace smelting quality,the prediction error of sinter al-kalinity is small,the reliability of the prediction results is high,the prediction time is short,and it has good prediction ability and can achieve real-time feedback.