Prediction Of Rotary Kiln Sintering Temperature Based On Deep Bidirectional Weighted GRU

Rotary kiln is the core production equipment in the metallurgy,power generation,cement and other industries.The sintering temperature in the kiln is a key parameter that reflects the status of its working conditions,and plays an important role in ensuring product quality and reducing pollutant emissions.Rotary kiln also is a typical complex controlled object with multivariable,nonlinear,time-varying and large inertia.The directly physical detection of sintering temperature is difficult,and traditional control methods are ineffective.At present,the fire-fighting workers still estimate the sintering temperature based on the flame image in the kiln and perform pre-kiln control.There are problems such as large fluctuations in operating conditions,low production efficiency,and high energy consumption.Therefore,predicting the calcination temperature of the rotary kiln has an important practical significance.In this paper,the on-site operating data from a company’s alumina rotary kiln is used to establish a calcination temperature prediction model in the kiln based on deep neural network method.The main works of the thesis are as follows:(1)The difficulties encountered in the modeling process of rotary kiln and the existing research results are analyzed.The research status of deep neural network and recurrent neural network are summarized.(2)According to the needs of the kiln time-series analysis,the network structures and calculation models of multiple recurrent neural networks are studied.The advantages and disadvantages of long-term memory network,bi-directional recurrent neural network and Multi-layer bi-directional gated recurrent neural network are analyzed,which provides a theoretical basis for selecting a appropriate deep network model of subsequent flame temperature prediction.(3)A deep bi-directional weighted GRU(DBWGRU)model is proposed.Aiming at the problem that the traditional gated recurrent unit(GRU)only learns single-direction information,we propose bi-directional weighted GRU algorithm which combines the forward with reverse information in the bidirectional recurrent neural network,and join the full connection layers to enhance the depth of the network and enhance the learning ability of the model.We use the Theano deep learning platform and the root mean square propagation(RMSProp)optimization method to compare the robustness of the model on Lorenz time series data.(4)A sintering band temperature prediction model is proposed by combining offset compensation and deep bidirectional weighted GRU.Firstly,we de-noise the field data,remove redundant variables,and delay the input variables in turn;then train the deep bidirectional weighted GRU model.The offset compensation is used to correct the model prediction value,which ensures the adaptability of the method under various working conditions and improves the robustness of the prediction model.(5)The analysis and application of sintering temperature prediction experiment results of rotary kiln.The experimental results show that the method proposed in this paper can provide more accurate guidance for the sintering temperature detection and make it possible for the prediction of the temperature of the rotary kiln.This paper focuses on the modeling of complex industrial processes.Deep learning and recurrent neural network techniques are used to deeply study the soft-sensing methods for the sintering kiln’s sintering temperature.The experimental results show that the predictive model based on depth bidirectional weighted GRU has better prediction results,which lays the foundation for the intelligent control of rotary kiln sintering process,so as to improves the clinker production and quality effectively.