Soft Sensor Modeling For Temperature In Rotary Kiln Based On Improved Elman NN And Its Application

The real-time online monitoring and control of pellet production process in chain grate rotary kiln is significant for improving pellet mineral quality and ensuring production safety,which greatly depends on the real-time online measurement for temperature of the rotary kiln in the pellet production process of chain grate rotary kiln.However,the temperature in the rotary kiln may be very difficult to measure online in time due to the complex measuring environment,expensive measuring instruments and time delay in the production process of chain grate rotary kiln.Therefore,soft sensing technology is used to establish a mathematical model between the temperature in rotary kiln and those easy-to-measure process variables,which is to predict the temperature in rotary kiln online in time.Due to the complicated structures of chain grate rotary kiln and a series of complex physical and chemical reactions in the pellet production process,the pellet sintering production process is intrinsically characterized with non-linearity,dynamics,time variability and high dimension of variable data,etc.Therefore,this paper mainly focuses on the soft sensor modeling for temperature of the rotary kiln based on improved Elman NN,data preprocessing technology and intelligent optimization algorithm to solve the real-time online measurement for temperature of the rotary kiln,which provides theoretical guidance to control the grate rotary kiln pellets production process effectively and in time.The main work of this paper is as follows:A new manifold learning data preprocessing method based on isometric feature mapping(ISOMAP)and locally linear embedding(LLE)under kernel framework is applied to solve the problems of the nonlinear variable data with high dimension,redundancy information and random noises.By constructing positive definite kernel function matrix and introducing the balance adjustment factor between ISOMAP and LLE,the ISOMAP and LLE is integrated into each other,which can reduce data dimension and remove redundant information while keeping the neighborhood structure relation and global mutual distance among data set to the greatest extent.Experimental results show that the new method has better dimensionality reduction effect.A soft sensing model based on improved Elman neural network with feed-forward strategy is proposed to deal with the production process nonlinearity,time delay and time variability simultaneously.To solve the nonlinearity and delay effectively,the output layer is feedforward adjusted by the context layer with dynamic memory ability,which makes the network advance to some extent.At the same time,the corrected and updated adaptively soft sensor model,which is based on the moving window strategy under the time difference framework,is used for soft sensor model of the nonlinear and time-varying processes.Finally,the nonlinear objective function is used for fitting test.The experimental results show that the improved Elman NN model has better fit and generalization ability.A novel grey Wolf optimization algorithm(LGGWO)with Levy flight disturbance and gravity search strategy is proposed.To shift the ability between exploration and exploitation during the search process,the Levy fight disturbance and gravity search are added in grey wolf optimizer algorithm by proposed excitation factors,which makes the grey wolf population have a stronger global exploration ability through Levy flight disturbance in the early search stage and more careful and thorough local search by gravity search in the later search stage.The LGGWO has been evaluated on a set of well-known benchmark functions.Experimental results show that LGGWO has better population diversity,search accuracy and convergence speed.Aiming at the real-time online measurement of temperature in rotary kiln during the pellet oxidation sintering production process of grate rotary kiln,the proposed soft sensing model is applied to predict the temperature in rotary kiln.Firstly,the data preprocessing method based on ISOMAP and LLE under the kernel framework is used to reduce the data dimension and remove the redundancy information or random noises in the pellet production process.Then the soft sensor model of temperature in rotary kiln based on improved Elman NN is established by using the LGGWO to optimize the structure parameters.And 10-fold cross validation is used to estimate the performance of soft sensing model.Finally,the external independent sample test data set is used to test the established soft-sensing model of temperature in rotary kiln.Experimental results show that the proposed soft-sensing model of temperature in rotary kiln has good robustness and prediction ability.