Application Of LSSVM In Soft Sensor Modeling

The increased demand of industrial enterprises and the developments of control theory and computer technology have been promoting the development of soft sensor technique.In modern complicated industrial process,some variables are very hard to be measured or even cannot be measured on-line by existing instruments and sensors.Soft sensor is an effective means of implementing the on-line evaluation of these variables.At Present,soft sensor technique has become one of the most important research areas in Modern industrial control field.The paper studies LSSVM to establish a soft sensor model for the advantages of solving aspects of nonlinear,High dimension,time-varying and large sample data fitting etc.The main contents of this paper are as follows. According to the problem of parameters optimization in the modeling process of LSSVM, the paper analyzes the effects of the regularization parameter and kernel parameter carefully. It studies two step grid search algorithm for parameter optimization based on the traditional grid search algorithm,improving the efficiency greatly. According to the sensitive to noise, poor robustness of LSSVM, the paper proposes adaptive weighted method based on the weighted method proposed by Suykens. Make full use of the contribution of the outliers to the model by changing the the original linear weight function to adaptive ways. And achieve good results. For LSSVM missing sparsity problem, the genetic algorithm(GA) is used for sparse model.The paper use binary coding method to code the kernels of initial LSSVM model. setting the reciprocal of errors’ quadratic sum of samples as fitness function, it use an elitist strategy to screen the binary strings. Decode the best individual and model again by the new sample until the deviation of sample’s standard is more than 10%. The simulation experiment indicates that the sparse rate of support vectors of model can reach about 70 percents.The algorithm improves the efficiency of model greatly without lowering the prediction precision.