Slot Decomposition Rate At The End Of The Process Of Alumina Carbon Prediction Model And Its Correction

In the process of sintering alumina production, the alumina continuous carbonation decomposition process (ACCDP) is a critical procedure bridging the preceding and subsequent procedure. Combination reaction of the sodium aluminate solution produced by upstream desiliconization process, and aluminium hydroxide with certain quality index is produced, at the same time qualified mother solution is supplied in the ACCDP. An important process index i.e. the last resolution ratio can be obtained only by analysis, and nonlinearly and long time-delay are formed in the ACCDP. Optimal operation was influenced greatly, and optimal running can not realize. Therefore, it is very important that the prediction model of the last resolution ratio was constructed and the on-line correction was studied.On the basis of production technology and influence factors analyzed, the main influence factors of the last resolution ratio were determined quantitatively through grey correlative analysis. The architecture of the prediction model of the last resolution ratio was proposed. By the characteristics of the ACCDP considered comprehensively, a back-propagation neural network (BP) prediction model was presented to forecast the last resolution ratio where the main influence factors were reduced by Principal Component Analysis (PCA), and parameters of BP neural network was optimized by Particle Swarm Optimization(PSO), so preferable precision and generalization capability of the model were improved. At the same time, the correction method of the prediction model of the last resolution ratio was studied. Online correction policy based on the filter method and short-period correction policy based on stochastic learning algorithm to the prediction model of the last resolution ratio were studied, so generalization capability of the model were improved.The simulation research was done to prediction model of the last resolution ratio and the correction method of the model by large quantity of actual data. The result shows that the prediction model of the last resolution ratio has a high precision and better generalization ability. And the prediction precision would not decrease along with the changes of working condition by introducing the correction method of the model. Then guidance to operation optimization of the ACCDP was afforded.