Based On A Combination Of Kernel Function Support Vector Machine Soft Measurement Technology And Its Applications

The soft measurement of the basic idea is,will the advanced automatic control theory and practical production cut combining process, using computer technology to existing technologies cannot be detected, the more difficult to measure or target variables, choose some other related, easy to measure, secondary variables by constructing a mapping relationship to forecast, estimate. Usually this kind of method is fast response, which gives the target variables, and continuous information of low cost, simple maintenance etc.Modern statistical learning theory is special research based on sample information for limited circumstances machine learning theory, for limited sample machine learning problems finding a solution. In the industrial production control system mostly acquisition limited sample information, to build pattern recognition or predict estimation model, thus theoretically, statistical learning theory has certain advantages. Support vector machine is in the basis of statistical learning theory. The development of a relatively new general machine learning method, because its use of structural risk minimization principle and has high accuracy and good study the promotion of generalization ability.Based on support vector regression machine algorithm and application as the main content, in the thorough analysis soft measurement technology principle, based soft sensor modeling process, target variables and auxiliary variable exists between the larger nonlinear, fuzzy relations, was proposed based on combination of kernel function of support vector regression machine soft sensor modeling method. Full text main job is as follows:1. Put forward a kind of method of linear kernel function, will have different characteristics in the kernel function is the form of linear combination of a new and satisfy the combination of Mercer theorem kernel functions concurrently, combination kernel function the global nuclear function and local kernels, and the advantages of by the weight coefficient factor adjust them to combination of kernel function, the function size in support vector machine (SVM) model have been achieved in good comprehensive prediction effect. Experimental results show that: based on the combination of kernel function of support vector regression machine model can obtain higher learning precision.2. Will be based on the combination of kernel function of support vector regression machine soft sensor modeling algorithm was applied in the process, for limited sample, high dimension, nonlinear soft sensor modeling provides an effective ideas. Not only for realizing complex nonlinear control system modeling provides control theory new methods, and expand the research contents of modern control field.Through YinRanYe wastewater of textile "acidic magenta determined.the decolored rate" modeling application shows that based on support vector machine regression machine, a soft sensor model in the generalization ability and measurement precision has certain advantages, practicability, achieved based on traditional mechanism analysis of the soft measurement technology better estimate prediction effect; For textile printing and dyeing wastewater treatment process, data acquisition delay big, noise is strong, parameters such as nonlinear, acidic magenta organic solvent difficult to real-time on-line detection problem, will be based on the combination of kernel function of support vector regression machine algorithm is applied to target parameters in the soft measurement modeling, simulation results show: this method can obtain than using a single base nuclear method of fitting accuracy is higher in promotion in generalization capability also made the ideal effect, can satisfy the printing and dyeing wastewater treatment process important parameter prediction requirements of online variables.