【摘要】 冶炼高炉烧结矿化学原料配比准确预测是对烧结矿化学成分提前预报、优化烧结工艺的关键方法。目前烧结矿化学原料配比准确预测的浅层网络算法未能充分发掘烧结过程的本质规律,预测精度不够高,难以应用在实际生产过程中。针对以上问题,通过对烧结过程工艺机理及特征进行深入分析,提出了应用深度学习技术中的深度置信网络算法对烧结矿化学成分进行预测,建立一种以深度置信网络为核心的烧结矿化学原料配比准确预测模型。首先设计深度置信网络结构和参数。通过无监督贪婪算法对模型进行预训练,采用BP网络有监督的反向微调权值,优化整个模型。最后和浅层预测算法进行对比。仿真结果表明,该方法预测值与实际值之间的误差小,预测精度高,相对于其它方法具有明显的优势,表明了深度置信网络应用于烧结矿化学原料配比准确预测的有效性。更多还原

【Abstract】 At present,the accuracy prediction for chemical raw material ratio of smelt blast furnace is the key method to predict chemical component of sintering mineral and optimize agglomeration process. But this kind of shallow network algorithm has not fully explored the essential law of sintering process,which is difficult to be applied in actual production process. Based on deep analysis of the technological mechanism and characteristic of sintering process,this article used the deep confidence network algorithm in the deep learning technology to predict the chemical composition of sinter. With deep confidence network as the core,we built a model of accuracy prediction for chemical raw material ratio of sinter mineral. Firstly,we designed the structure and parameter of deep confidence network. Secondly,we used the unsupervised greedy algorithm to pre-train the model and used the supervised reverse fine tuning weight in BP network to optimize the whole model. Finally,we compared this method with the shallow prediction algorithm. Simulation results show that the error between the predicted value of proposed method and the actual value is small. The prediction accuracy is high. Meanwhile,this method has obvious advantage. Therefore,the effectiveness that the deep confidence network is applied in the prediction of agglomerate chemical raw material ratio is proved.更多还原