Research On Soft Sensing Of Cement Specific Surface Area Based On Multi-step Time Extraction Network

Cement specific surface area is one of the important quality indicators of cement products.It is very important to realize accurate soft sensing for making reasonable cement production schedule to realize low-carbon,energy-saving and improve the quality of cement.Due to the particularity of cement processing industry,including the time lag and strong coupling of the acquired variables and the uncertainty of production conditions,it is a challenging task to establish an accurate soft measurement model for cement quality detection.In this paper,we propose a soft measurement method based on Multi-step Time Extraction Network(MTENet)to address the problem of time lag and strong coupling of data features in cement clinker grinding process.For the uncertainty of production conditions,this paper proposes a variable condition adaptive hyperparameter optimization soft measurement model IDOA-MTENet based on Improved Dingo Optimization Algorithm(IDOA).This paper achieves accurate measurement of cement specific surface area through the above soft measurement model.The specific research work is as follows.Firstly,this paper introduces the whole process of cement production,and focuses on the clinker grinding system process.This paper also analyzes the important variables affecting cement specific surface area,and provides a method for selecting these variables.In addition,the paper also describes some difficulties in soft sensing modeling of cement specific surface area,and proposes a soft sensing framework to solve these problems.At the same time,the establishment process and function of the framework are explained in detail in order to better understand its application value.Secondly,MTENet soft measurement model is constructed for the problem of time lag and strong coupling.MTENet is divided into an encoder and a decoder.The encoder contains a new temporal feature extraction unit proposed in this paper,which can solve the time lag problem of the feature data very effectively by a clever combination with Conv LSTM.Also,this paper introduces a parallel unit to make the network perceive the changing trend of surface area values.In the decoder part,this paper adopts the method of concatenating parallel features to ensure the accuracy of the final decoding results.Thirdly,for the problem that the generalization ability of the model of cement clinker grinding system is not high in the case of variable production conditions,and manual tuning of the parameters is required.In this paper,we propose IDOA based on DOA by adopting chaotic strategy to initialize the population and introducing Levy flight strategy into the survival behavior.IDOA adopts a global search strategy,which can avoid falling into the local optimal solution,and can be well used to automatically find the reasonable hyperparameter of MTENet soft sensor model when the production condition changes.Finally,to evaluate the accuracy of the soft measurement model MTENet,comparison experiments are conducted in this paper,and the models compared include the classical models CNN-LSTM,LSTM,CNN,XGBoost and SVR.The effectiveness of MTENet is verified through comparison experiments and analysis of evaluation metrics.At the same time,this paper also conducted a series of comparative experiments to determine the effectiveness of the IDOA optimization algorithm on the MTENet model.These experiments compare some common hyperparameter optimization models as well as the original DOA model.The experimental results show that the IDOA-MTENet soft-sensing model has excellent accuracy and can meet the measurement requirements under variable operating conditions.