基于时间序列单维卷积神经网络的水泥熟料游离钙软测量方法

Soft Measurement Method for Cement Clinker fCaO Based on Time Series Single-dimensional Convolutional Neural Network

水泥熟料游离钙(fCaO)含量对水泥质量和生产能耗有着重要影响,现阶段主要通过化学分析的方法离线测得水泥熟料fCaO含量,但是该方法对于烧成系统操作指导具有明显的滞后性。针对熟料fCaO无法在线实时监测的问题,提出基于多变量时间序列单维卷积神经网络(TS-CNN)熟料fCaO软测量建模方法。该方法利用影响熟料fCaO的多个过程变量历史时间段的时间序列作为输入,结合水泥数据特性,采用单维卷积池化的方式提取各过程变量特征,同时降低网络的复杂度,最后经全连接层整合提取的局部信息。通过实验对比,结果表明基于TS-CNN的软测量方法预测精度更高、泛化能力更强。

The content of free calcium oxide(fCaO)in cement clinker has an important impact on cement quality and production energy consumption.At this stage,the fCaO content of cement clinker is measured offline by chemical analysis method,but this method has obvious hysteresis for the operation guidance of the production process.Aiming at the problem that clinker fCaO could not be monitored online,a soft measurement modeling method of clinker fCaO based on multi-variable time series single-dimensional convolutional neural network(TS-CNN)was proposed.The time series of a certain historical time period that affects multiple variables of clinker fCaO was used as the model s input,and the cement data was combined to extract the characteristics of each variable by using single-dimensional convolution pooling method to reduce the complexity of the network,and finally the extracted local information was integrated by the fully connected layer.Through the experimental comparison,the results show that the soft measurement method based on TS-CNN has higher prediction accuracy and more generalization ability.