A Study On Cement Clinker Free Calcium Data Enhancement And Soft Measurement Model Based On Generative Adversarial Network

The content of free calcium oxide(f-CaO)in cement clinker can directly reflect the quality of cement products.Achieving its accurate prediction plays a significant role in process optimization of cement production,energy saving and consumption reduction,and improvement of concrete quality.However,the lag of manual sampling leads to the problems of small amount of labeled data and unbalanced samples on multiple time scales for cement clinker f-CaO samples,which poses difficulties in establishing an accurate and efficient soft measurement model.Therefore,this paper proposes a generative adversarial network-based data augmentation and soft measurement model for free calcium of cement clinker.The idea of data augmentation is used to expand the multi-timescale sample data to further achieve accurate prediction of cement clinker clinker f-CaO.The study is as follows:Firstly,a cement clinker f-CaO data enhancement model based on semi-supervised Wasserstein Generate Adversarial Networks(WGAN)is proposed for the problems of small labeled sample data volume in cement production process.By reconstructing the input and output layers of the generator in WGAN,the unlabeled production variables of the input generator are mapped to low-dimensional labeled values,and the filling of the missing labeled values is achieved.Then,the labeled data and the corresponding production variable data are data spliced according to the time scale and fed to the discriminator uniformly for discriminating.Through the game process of generator and discriminator,the generator can catch the correspondence between labeled samples and production variables and generate a large amount of reliable f-CaO data.Finally,the reliability of the generated f-CaO data was proved by the introduction of high-dimensional data evaluation metrics for visual evaluation of the generated labels.Secondly,a data augmentation-based soft measurement model for cement clinker fCaO(SSP-WGAN)is proposed to solve the problem of low accuracy of the soft measurement model due to the nonlinearity,coupling,and long time lag characteristics of the cement firing process.Based on the data augmentation model in the previous chapter,the generated data are used to expand the training set of the SSP-WGAN soft measurement model.The model uses the sliding window principle to construct a neural network input layer based on the time-series relationship between cement clinker f-CaO data and production variables.The spatio-temporal features embedded between the production variable data are extracted using a stacking of convolutional neural networks and gated recurrent units.The model can improve the soft measurement model accuracy effectively by eliminating feature redundancy while retaining dynamic time delay information.Finally,a data enhancement-based continuous time series prediction model(Seq2SeqWGAN)is proposed for cement calcination process with unbalanced time scales,which is based on data enhancement and combined with industrial continuous production characteristics..The label values filled in the data enhancement model are mixed with the real labels and mapped to the input layer of the encoder to eliminate the influence of multiple time scales.seq2Seq-WGAN can achieve accurate prediction of cement clinker f-CaO under multiple time nodes by establishing relationships between the input and output data given to different time periods.