Research On The Methods Of Soft-sensing For The Kernel Function Of Alumina Powder Flow

Modern large prebaked aluminum production, aluminum production and alumina conveying system as an important component of the efficient production of aluminum an increasingly important role. Alumina conveying process, the need for aluminum in the alumina powder flow as a precise measurement of the main parameters on-line, and applied to the alumina production line analytical instruments due to the accuracy is not high, expensive, unable to adapt to the harsh environment of the scene and other factors, it is difficult to meet the alumina powder flow line measurement. Soft-sensing technology for the alumina powder flow online provides a good way to predict. But now the technology based on the alumina powder flow forecasting even less, and more with a single model for modeling, especially for the auxiliary variables nonlinear factors to consider very little. This article was based on the idea to the kernel function, combined with the commonly used three kinds of soft sensor modeling method for the alumina powder flow forecasting techniques are studied.1) Based on KPCA-PLS alumina powder flow soft sensor modeling method. Engineering practice, the first in-depth analysis of alumina powder ultra dense phase conveying technology, based on the process identified two measurable variables as auxiliary variables required for modeling, considering nonlinear contact auxiliary variable field has a certain noise using KPCA treatment was carried out, nonlinear feature extraction and removal of random noise; thus using partial least squares (PLS) its robustness and good prediction accuracy, the establishment of alumina powder traffic prediction model.2) Based on KPCA-RBF alumina powder flow soft sensor modeling method. Because PLS is essentially a linear operation, process data can not be a good solution to nonlinear problems, it is difficult to establish predictive models of high precision, considering RBF neural network for nonlinear approximation of continuous functions with consistent, learning speed, with better generalization ability, the results will be as KPCA input RBF established the appropriate predictive models.3) Based on KPCA-LSSVM alumina powder flow soft sensor modeling method. As RBF neural network modeling for large sample data with higher prediction accuracy and generalization ability, but in the actual data obtained from the industrial process is limited, taking into account the small sample LSSVM has, preferably non-linear processing capability and speed fast and generalization ability, etc., the results will be as KPCA input LSSVM established alumina powder flow forecasting model.Industrial field data using these three soft sensor modeling method is simulated in MATLAB software, and verification of three kernel-based alumina powder flow soft sensor model validity and applicability; same batch of data on the simulation results, based on KPCA treated alumina powder flow soft sensor model prediction accuracy and generalization ability are superior to non-nuclear approach.