Soft Measurement Of F-CaO Content In Cement Clinker Based On Dualdistribution GAN Data Augmentation

The amount of free calcium oxide(f-CaO)in cement clinker is a crucial indicator of cement quality in the production process.Currently,cement companies mainly rely on manual sampling and detection techniques for their measurements.The sampling duration is usually one hour,which leads to a meager amount of f-CaO content data for cement clinker and a considerable delay.This paper proposes a combination of data enhancement and soft sensing,and a soft sensing model of f-CaO content of cement clinker is constructed to enhance cement production data and detect f-CaO content.The research content is as follows:(1)In view of the multiple time scales caused by different timing,high dimension and sampling cycles of cement production data,a data enhancement model for cement calcining process based on Dual Distribution Metric-Generator Adversarial Network(DDM-GAN)is proposed.First,the sliding window technique is used to extract the data in time order,and the input layer of the model is constructed to ensure that the model can learn the timing properties of the data.The effects of non-equilibrium data volume and different dimensionality between the data are then removed through the operation of effective period expansion and data normalization.DDM-GAN’s dual distribution metric is utilized to direct the generator in comprehending the distributional characteristics of the data,thus producing a considerable quantity of dependable information.Finally,a variety of evaluation mechanisms are used to evaluate the reliability of the generated data to ensure the accuracy and reliability of the modeling.(2)A soft sensing model of f-CaO content of cement clinker(C-DDM-GAN)was proposed to address the issue of low accuracy in the soft sensing model,which is caused by the nonlinear and time-delay characteristics of production data during cement calcining.The trained DDM-GAN model is used to augment the dataset,and then the real data is mixed with the generated data.Sliding window technique was employed to input the mixed data set into the convolution prediction model in time series,and multi-layer convolution was used to extract the data features to measure the f-CaO content in cement clinker.Experiments demonstrate that the soft sensor model with data enhancement is more accurate.(3)In order to further improve the soft-sensing accuracy of f-CaO content of cement clinker,considering the weak long-term dependence of convolutional neural network,a softsensing model of f-CaO content data enhancement for cement clinker was proposed based on convolutional cycle network(CR-DDM-GAN).Firstly,DDM-GAN is used to generate reliable data to augment the dataset.The sliding window technique is then used to feed the data into the soft sensor model in a timely order.Two 1D convolutional layers are used to first extract the data,yielding feature sequences.These sequences are then fed into two bidirectional LSTM layers,further extracting the data features and sequence information.Finally,the loop optimization of the model yields the measured f-CaO content values.Experiments show that the CR-DDM-GAN model has better ability to extract data features and higher soft sensing accuracy.