Research Of Wood Drying Modeling Based On Support Vector Machine

Presently, wood was more and more beloved by humans. But the contradiction is that the global forestry resource was reducing continuance. Wood drying is an important step for keep wood physics capability and wood production quality. Most domestic wood drying equipment have the weakness of low automation level and unreliability, which cause the serious problem of high energy consume and lowing wood grade. Building a wood drying model is the first step for design wood drying automation equipment. But the wood drying process is a nonlinear system, which has a characteristic of strong coupling and lagging. So the wood drying mechanism model is hard to use because of overabundance restriction and complex model frame. Machine learning based on sample data is an experiment modeling method. The model can reflect exterior system characteristic, only need to learn from easy measured system input and output sample data. So it's an effective method for nonlinear system modeling.Support vector machine was brought forward by Vapnik at ninety decade last century. Its theory basement is statistic learning theory. SVM is a learning machine for small sample data with strong extensive use and no over learning problem because of the structure risk minimization. A SVM wood drying model was built for its nonlinear characteristic in this article. Sample data was acquired from wood drying experiment, the result of simulation shows that SVM wood drying model has a perfect performance. A LSSVM wood drying model was also built for acquire more precise prediction and faster computation model. This article also analyzed the influence of SVM wood drying model between different kernel function and optimal method.Research of online LSSVM wood drying model was also launched. Leave line model has no ability to reflect the dynamic characteristic of wood drying process. Contradiction, online modeling can online updating training sample and model frame, so it has a perfect dynamic characteristic which can reflect currently system state. Time window training sample frame was used by classic online modeling method, which is deleting the earliest sample data when the new sample data was acquired. Based on sparse LSSVM theory, a new online method was used for wood drying. This method updating training sample data by delete the sample data corresponding smallest support vector modulus when the new sample data was added. Simulation experiment shows that both online modeling method all have perfect capability, which can precisely predict wood moisture content on process of wood drying.