Research On Soft Measurement Model For Finished Product Quality And Production Energy Consumption Of Cement Clinker Grinding System

Cement clinker grinding is one of the important links in the whole process of cement production.The running state of this link is closely related to the qualified rate of cement final product.Specific surface area is a key index to reflect the grinding quality of cement clinker.However,there is still a lack of real-time online measurement model to guide workers to adjust the production process in time.In addition,the cement industry has been faced with the problem of excessive energy consumption for a long time,so the realization of the power consumption prediction of cement clinker grinding system under different working conditions can provide a scientific basis for cement production control,and can improve the core competitiveness of enterprises.This paper aims to use the neural network based soft sensing method,according to the characteristics of different input data of cement clinker grinding system,establish the concrete specific surface soft sensing model and the power consumption soft sensing model,to achieve the effective prediction of concrete specific surface area and power consumption,and provide scientific guidance for the subsequent production decision.Specific research work is as follows:(1)The process flow of cement product preparation and the production process of cement clinker grinding system were analyzed,the required variables were extracted from the cement industry database,and the process variables affecting the specific surface area and power consumption of cement were screened out.According to the characteristics of cement grinding process and data,the modeling difficulties of cement specific surface area related data with feature redundancy and nonlinearity,and the modeling difficulties of power consumption related data with strong coupling and uncertainty were analyzed,and the corresponding design scheme was put forward.(2)According to the nonlinear and dynamic characteristics of specific surface area data redundancy,module based on the Inception and improved Quasi-Recurrent Neural Networks cement specific surface area of the soft measurement model.The model uses the characteristics of multi-branch structure of Inception module to extract the diversity features of data and reduce the feature redundancy.At the same time,in order to improve the generalization ability of the model and better fit the nonlinear characteristics,the model fused the residual module to adjust the weight inside the QRNN on the basis of QuasiRecurrent Neural Networks.The Residual-Quasi Recurrent Neural Networks(R-QRNN)model is proposed for the measurement of specific surface area of cement.(3)Due to the strong coupling of power consumption related data structure and the difficulty of in-depth analysis of time sequence characteristics,the power consumption soft measurement model of cement clinker grinding system is established on the basis of Involution operator and Nbeats X model,combined with Gated Recurrent Unit.In this model,the Involution operator is used to reduce the coupling between the involution data by its "deconvolution" characteristic.At the same time,the improved Nbeats X model,namely the Last GRU model,is used for time series decomposition to improve the accuracy of the model prediction.(4)The final product quality and production energy consumption models proposed in this paper were compared with the current mainstream machine learning and neural network models to verify the prediction accuracy and generalization ability of the model.