A LLE-based HMM On Kiln Coal Feeding Trend Prediction

The rotary kiln is the key equipment in the process of alumina production by sintering.The sintering process of kiln is a typical kind of complex industrial process with strong coupling,multi-interference,large time delay and nonlinear time-varying characteristics,which make it difficult to realize the automatic control of the production process.In the clinker sintering process,as coal feeding is the main control variable of the kiln operation,accurately adjust the coal feeding can stabilize the kiln conditions and improve the quality and yield of clinker.However,the kiln conditions,the quality of coal,slurry composition and skin structure have great influence on the amount of coal feeding,which make the experienced workers in the operation are rather difficult to grasp the chance of coal feeding amount to add and subtract accurately.In this paper,we analyze the correlation between the thermal data and the trend of feed coal,put forward a kind of hidden Markov model based on locally linear embedding(LLE-HMM)to predict the trend of feed coal changing.First of all,as to the characteristics of the thermal data,we take the measure of de-noising,filter and standardization.Next,we use nonlinear transformation of LLE to extract features.Then,the symbolic value can obtain by quantitative processing sequence.Finally,we combine the DS evidence theory to apply LLE-HMM make predictions.Through the simulation,the results show that the method can predict the trend of former kiln feed coal effectively.The follows are the main contents in this paper:(1)Analyze the deficiency of the linear transform methods such as PCA,ICA and LDA for feature transformation and introduce the nonlinear transform for the thermal data at kiln.Use the locally linear embedding(LLE)of manifold learning for transformation,and update the cost matrix for weight calculation to the incremental data.(2)Use the HMM to the coal feeding prediction.In the model,the scaling quantitative method is adopted for the symbolic processing.Then,the training algorithm is optimized in order to weaken the impact on initial value and to prevent the fall into local optimum.Finally,we combine the DS evidence theory to make predictions by the HMM.(3)Compare the performance of HMM with other predict model such as BP,SVM and ELM,we can find HMM have the advantages at the establishment of model inclassification at the dynamic time-sequence signal such as thermal data in kiln.Apply the HMM in the thermal data at rotary kiln,the experiments of simulation show that the LLE-HMM has rather high precision to other linear transformation model in the kiln coal feeding.In this paper,by analyzing the characteristics of thermal data at rotary kiln,LLE-HMM is used to predict the coal feeding.The simulation results show that this model can judge the change of coal feeding and provide reliable reference guide information for the artificial operation and expert control of rotary kiln effectively.