Research On Dimensionality Reduction Strategy And Comprehensive Evaluation Ways Of Pulpwood Indexes

Pulp and paper industry’s development is inseparable from the rich supply of timberresources, but the imbalanced structure of China’s paper industry raw material, has beenhampered by the pulp and paper industry on the road to modernization. Therefore, we mustvigorously develop the fast-growing forest pulp, must keep taking the route offorestry-paperintegration, to solve the shortage of raw materials for pulp and paper industry bottlenecksgradually. The past development of fast-growing plantations has certain blindness. Usually weplant trees and then find their purpose, reducing the economic benefits of developingfast-growing plantations. We should know whether the tree having fast-growing, high yieldcharacteristic and good pulping performance before planting tree, having important researchvalue. So it is absolutely necessary for comprehensive evaluation of paper-making raw materials.Usually the indexes of comprehensive evaluation are large and we can get comprehensiveevaluation indexes by many ways. We can get massive amounts of data form pulpwood biomass,growth, chemical composition and fiber morphology and other property index, or near-infraredspectroscopy in this article. The massive data is multi-dimensionalvector in mathematics, havingthe different dimension and the specific values. Multiple attributes indexes often have multiplecorrelation problems. When studying the relationship between the independent and multipledependent variables, variables and dependent variables may have nonlinear problems exceptmultiple correlation problems. What’s more，if we deal with massive data in near-infraredspectrum, it will cost so much time. The massive data contain the information which is notcompletely useful in near-infrared spectrum. The massive data may have noise in near-infraredspectrum. In conclusion, we should make comprehensive evaluation of paper materialsaccordingto local conditions, choosing ways of comprehensive evaluation by the purpose of evaluation andthe feature of multiple indexes data.In order to effectively address the high-dimensional problem in the evaluation of pulpwoodspecies, a projection pursuit model based on particle swarm optimization (PSO-PP) is applied toevaluate14poplar varieties by pulpwood biomass, growth, chemical composition and fibermorphology and other property index. Using this method to study paper-making suitability ofpulpwood species to provide a scientific basis for pulpwood species oriented cultivation. In orderto establish pulpwood pulping performance evaluation and prediction model, partial least squaresregression (Partial least-squares regression, PLSR) is used to build a more stable and a morereasonable interpretation of pulpwood pulping performance evaluation and prediction model bydecomposing and recombining the information in variable system to overcome the multiplecorrelation problem of the performance indicators. Forty examples of five kinds of papermaking raw material’s near-infrared spectrum are collected to set up identification model of pulps bynear-infrared spectroscopy. Liner Fisher discriminate method and nonlinear mapping capabilityof BP-ANN are introduced to found discrimination model by near-infrared spectral analysistechnology and chemometrics method.