Soft Sensor Multiple Model Based On LS-SVM Papermaking Black Liquor Concentration

Papermaking alkali recovery of evaporation section black liquor concentration is a very important control target, the black liquor composition is relatively complex, accurate hardware online measurement is difficult to solve this problem, this paper uses the soft measuring technique. Soft measurement technology is mainly composed three parts of an auxiliary variable, the main variable and the model, the idea is that combined the automatic control theory and the actual production process effectively, using auxiliary variables are easily measured to estimate the dominant variables, which are hard to be measured, through computer technology construction the mathematical model of auxiliary variables and the main variables. The software instead of hardware function, not only economic and reliable, but also quickly have dynamic response.Firstly, this paper deals with the modeling data, because of the lag in distinct extent of different variables with the dominant variables in the actual production process, auxiliary variables at the same time value and the dominant variable value can not accurately reflect the intrinsic relation between them, this paper use the method of grey correlation analysis, to determine the time sequence delay according to the degree of similarity slope of auxiliary variables and the dominant variables for a certain time period, improve the authenticity of internal relation between the variables. Secondly, use all auxiliary variable modeling may lead to redundant information, influence the result of forecast, calculation of correlation between the auxiliary variables by the correlation coefficient method, the smaller correlation between variables as a input group, to build the submodel, and use of the count model prediction error as the submodel weight factor, effectively improve the measuring accuracy. Thirdly, there are a variety of soft measurement modeling method, the least squares support vector machine is a new machine learning method, based on the structural risk minimization principle, with small training samples and computation speed, so choice the least squares support vector machine to modeling. The regularization parameters and kernel parameters of least squares support vector machine model need to be optimized by optimization algorithm, this paper propose improved inertia weight particle swarm optimization algorithm, based on the tangent and arctangent monotonicity, by adding weighting parameter set inertia weight iteration method.In order to testify the effectiveness of the proposed method, we use the data from a papermaking mill in Nanning of Guangxi, then make a series of simulation experiments on MATLAB platform. Analyze the relation of various auxiliary variables and the dominant variables by grey correlation analysis, the auxiliary variable with small association be removed, and the auxiliary variables remaining were grouped used correlation coefficient method, let the redundancy information of auxiliary variables grouped are smallest. Through a plurality of test function test the improved inertia weight particle swarm optimization algorithm, use the improved particle swarm optimization algorithm to optimizing least squares support vector machine model parameters. By comparing the gray correlation analysis of single and multiple least squares support vector machine modeling method, proved that multi model modeling method has better prediction results.