Kiln Coal Feeding Trend Prediction Based On KPCA-BHMM

Rotary kiln clinker sintering plays a key role in the process of aluminaproduction. As a typical complex industrial process, it is of great difficulty to developa model with the method of mechanism analysis or system identification for thesintering process. In the case that the blowers, exhaust throttle opening remainsunchanged, the amount of coal supplying is the main operating variable forcontrolling the sintering temperature within the kiln.With the full consideration of the impact that each thermal parameters wouldhave on the sintering process during the rotary kiln sintering process, the Kernelprincipal component analysis (KPCA) and hidden Markov model (HMM) arecombined. The rotary kiln feed coal volume trend forecasting results are obtainedaccording minimum risk Bayesian decision required before the kiln operation, Thismodel can be expressed as KPCA-BHMM. The research work of this thesis is asfollows:The compression of the time sequence is analyzed, and according to the timesequence of thermodynamic data of rotary kiln, time sequence compression of the keypoints and first-order differential mean sequence description are presented.Since thedata collected during the rotary kiln sintering process has non-linear structures, thenoise in the data is eliminated utilizing the kernel principal component analysis. Theextracted data contains the main component information. Then the dimensionality ofdata is reduced. Due to the advantage in time sequence prediction, the HMM isintroduced to forecast the trends of the rotary kiln feed. An equal scaling vectorquantization method is proposed to solve the problem of vector quantization thatoccurs in the discrete HMM.Taking the loss caused by the trend misjudgment duringthe clinker sintering process into account, the minimum risk Bayesian decision isintroduced to estimate the coal feeding trends in rotary kiln,based on the posteriorprobability and decision-making table calculated by using the HMM, the minimumrisk Bayesian decision is conducted to obtain the feed coal quantity trends.The rotarykiln site data obtained is used to verify and simulate the method of the KernelPrincipal Component Analysis-hidden Markov-Bayesian decision. Compared withthe simulation results of Naive Bayesian classifier (NBC), principal componentanalysis-hidden Markov model(PCA-HMM), Kernel principal component analysis-hidden Markov model(KPCA-HMM), The results showed that KPCA-BHMM has a high prediction accuracy in kiln coal feeding.The research of the thesis not only helpful for effective implementation of thekiln sintering temperature controlling and clinker sintering process operationoptimization, but also create the conditions for improving the kiln conditions andproducts quality as well as saving the producing energy.