Research Of Wood Drying Process Modeling Based On Parameter Optimization Of Support Vector Machines

Wood drying is an important technical measure to better the physical and mechanical properties, use it reasonably, reduce the degradation loss, and improve the utilization ratio of wood. It is also one of the key technologies to ensure the quality of wood. Using conventional control based on linear model can't achieve the target and effect of optimal control as wood drying process shows strong nonlinearity, so to build the model which describes wood drying regular accurately and integrally is the key to realize full automatic control and improve control level.Support Vector Machines(SVM) is suitable for the modeling of wood drying process with the characteristics of strong nonlinear, strong coupling, and large delay, as a new machines learning method based on statistical learning theory with outstanding advantages, such as generalization ability, overall optimization, and high computing speed and so on. Fitting accuracy and generalization ability of predictive model directly depends on choose of related parameters, however, with the lack of guidance on general theory and method, the application effect of SVM is affected to some extent. This paper researched on the modeling of wood drying process based on parameters optimization of SVM to aim at the problem of its parameters selection. By analyzing the effect of SVM parameters on the performance of wood drying schedule model,this paper considered parameters selection as combinatorial optimization problem, built objective function, applied Particle Swarm Optimization(PSO) algorithm and Genetic Algorithm(GA) to SVM predictive model respectively, so as to realize automatic selection and optimization of the parameters. Simulation results showed that two kinds of swarm intelligence optimization algorithm above could select the SVM wood drying model parameters effectively through establishing wood drying model based on PSO-SVM and GA-SVM. In comparison with GA, PSO showed better learning ability and generalization in wood drying process modeling, produced predictive model with better accessibility.In order to realize online modeling of wood drying process, this paper introduced Kernel Principal Component Analysis(KPCA) to pretreat wood drying data because test sample data had more noise in drying process and SVM could not distinguish samples redundant, useful or not in the training process. Using particle swarm algorithm with better optimization effect optimized SVM parameters to research offline modeling. Simulation results showed that the wood drying schedule model built by the pretreated training samples data had strong practicability, and could get better predictive accuracy, less computations and higher computing speed. On the basis of KPCA data pretreatment, this paper researched drying schedule online modeling with online optimizing SVM parameters based on PSO algorithm, and it could predict the moisture content in drying process accurately, which would provide powerful guarantee to realize online predictive control of wood drying process, also had significance for improving wood drying control level.