【作者】 刘德胜；

【导师】 曹军； 张佳薇；

【作者基本信息】 东北林业大学 ， 控制理论与控制工程， 2007， 硕士

【摘要】 木材是全球应用最广泛的工程材料之一。面临世界森林资源日益减少所带来的环保和生态问题，如何有效地利用木材资源、降低能耗、提高木制品质量已引起各国政府的广泛关注。我国是一个森林资源不足的国家，木材的需求量已大于供给量，改善木材使用性能，提高其利用率，已成为木材科学工作者研究的前沿课题。木材干燥是木材加工中的必备环节，是改善木材物理力学性能、保证木制品质量、减少木材降等损耗、提高木材利用率的重要技术措施。木材干燥是一个复杂的强耦合非线性动力学系统，在干燥过程中存在外界的干扰和模型的不确定性，如何建立有效的干燥模型是木材干燥的重要基础研究内容之一，也是实现干燥全自动控制，提高干燥质量，减少能量消耗，缩短干燥时间的先决条件。本文针对木材干燥过程存在非线性的特点，分别建立了木材BP神经网络模型和动态递归神经网络模型，从静态和动态两个侧面对木材干燥过程进行建模；分别利用自适应加权数据融合算法和算术平均值与递推估计算法对两模型输出进行融合，建立了能够根据木材含水率变化的不同阶段在线调整的融合模型，提高了模型的泛化能力。本文针对木材干燥过程存在强耦合的特点，利用偏最小二乘的多重共线性分析和非线性回归能力，分别建立了基于偏最小二乘的木材干燥过程回归方程和偏最小二乘与神经网络的混合模型，建立的模型能更准确的反应影响木材含水率变化的环境因子，通过实际测试干燥数据进行仿真，验证了数据融合算法解决木材干燥动态建模问题的可行性。更多还原

【Abstract】 Wood is one of the most popular engineering materials. Facing the problems of environmental protection and ecology, which are caused by the decreasing of the world’s forest resources, how to use the wood resources effectively、reduce energy consumption and improve the quality of wood products has aroused widespread concern by the governments of many countries. China is a country which is short of forest resources, and timber demand has exceeded supplementation, so in order to make the performance of the timber better, improving timber utilization ratio has become a leading edge research subject in the face of wood scientific workers.Wood drying is an essential link in the wood processing and it is the important technical measure to improve the physical mechanical properties of wood, ensure wood products qualities, and to reduce the loss of wood dropping and to raise wood utilization ratio. Wood drying is a strong coupling、nonlinear and complex dynamic system, and external interference and model uncertainty exists during the drying process. So how to create an effective model of drying is an important foundation study of wood drying, and it is also the pre-condition of realizing full automatic control of drying, improving drying quality, reducing energy consumption and reducing the drying time.In this paper, according the nonlinear characteristics of wood drying process, we set up a wood BP neural network model and a dynamic recursive neural network model respectively, modeling from the static and dynamic aspects of the wood drying process; use self adapting weighted data fusion algorithm and the arithmetic average based on recursive estimation algorithm fused the outputs of the two models, established a fusing model which can be adjusted online according to the different stages of the changes of wood moisture content and improved generation ability of the models.In this paper, in allusion to the strong coupling characteristic of the wood drying process, using multicollinearity analysis of partial least squares and nonlinear regression ability, the regression equation of wood drying process based on the and partial least squares and the mixed model of neural network based on partial least squares are established respectively. The model can response to the more environmental factors which affect the moisture content changes of wood more accurately, and the feasibility and practicality of the methods has been validated through data simulation of a real dry.更多还原