【作者】 陈广胜；

【导师】 王金满； 郭明辉；

【作者基本信息】 东北林业大学 ， 木材科学与技术， 2006， 博士

【摘要】 本文以人工林落叶松(Larix ssp.)木材为研究对象,以建立木材材质早期预测模型为目标,通过分析人工林落叶松木材的解剖性质(包括:管胞长度、管胞弦向直径、管胞长宽比、管胞壁厚、璧腔比、胞壁率和微纤丝角)和物理性质(包括:生长轮宽度、晚材率、生长速率和木材密度),研究人工林落叶松木材各项材质指标的变异规律,建立人工林落叶松木材材质变异规律模型和木材材质预测模型。 首先,采用计算机视觉分析系统测量人工林落叶松木材解剖性质,采用x-射线微密度测试系统测量木材物理性质,研究人工林落叶松木材材质的变异规律,界定人工林落叶松木材的幼龄期与成熟期的界限,评定人工林落叶松木材幼龄材与成熟材材质的差异。 其次,针对目前广泛采用的预测方法,选取人工林落叶松木材管胞长宽比和生长轮密度两项指标作为基础数据,采用回归分析方法、时间序列方法和神经网络方法分析建立木材材质预测模型的可行性,比较预测模型的预测相对误差和预测精度,并进行模型检验。初步得出如下结论:神经网络预测方法建立的木材材质神经网络预测模型,预测误差小,预测精度高,为相对最优模型。 第三,利用神经网络良好的非线性映射能力,自学习适应能力和并行信息处理能力,及其用于未知不确定非线性系统建模的优势,针对不同材性指标时间序列数据的特点,确定各模型网络结构、传递函数、网络训练函数,建立人工林落叶松木材解剖性质和物力性质的神经网络预测模型,并利用测试集数据,检验预测模型精度,各材性指标的预测模型相对误差最大值为4.55%,最小为-4.73%,在全部的25项指标中有22项相对误差在-2%～2%之间,说明预测精度较高。通过比较分析预测值与实测值的差异,相对误差较小,预测效果良好,预测结果可以满足实际预测要求。 总之,采用神经网络方法建立人工林落叶松木材材质预测模型可行,预测误差小,预测精度高,预测模型科学,为人工林的集约经营和定向培育提供了基础理论依据。更多还原

【Abstract】 To establish a forecast model of wood properties, the variation regularity of larch plantation was studied by analyzing the properties of larch wood, including the anatomical (tracheid length, tracheid diameter, ratio of tracheid length and width, thickness of tracheid wall, ratio of thecal opening, and microfibril), and physical (width of growth ring, growth rate, late wood rate, and wood density) properties of wood, based on the study on the larch plantation. Moreover, the models of variation regularity and forecast for wood qualities in larch plantation is established.Firstly, the anatomical characters of timber for larch plantation were tested by computer visual analysis system, and physical properties of timber were tested by X-ray density testing system, respectively. The variation regularity of larch plantation was studied, juvenile plantation and matured forest were ploted, and the difference of timber quality for larch juvenile plantation and matured forest were evaluated.Secondly, according to the forecasting methods applied widely nowadays, the most typical characteristic index of wood properties, namely, ratio of length/width of trachied and growth ring density, were selected as analytical data. The feasibility of wood properties forecasting model were established by the methods of linear regression, time series, and neural network and its forecasting precision were compared and proof-tested.. Results indicated that the forecasting model established by neural network method was the best, with smallest error and best precision, and the model was the optimization.Thirdly, the neural network has the advantages of powerful non-linear mapping, self-learning adaptability, simultaneous information processing, modeling for unknown non-linear system. Aimed at the data characters of time series wood properties, the different model network structures, transfer functions and network training functions were determined by the neural network, and the neural network models of different wood properties indexes for larch plantation were established. The precision of the model was verified by data in test group, which indicated that maximal relative error was 4.559%, and minimal was -4.73%. Of all 25 items, the relative errors of 22 items ranged from -2% to 2%, forecasting precision was very high. The forecasting results may entirely meet the demands in practice.In conclusion, the forecast model of timber quality for larch plantation established by neural network method were optimization, with the advantages of small forecasting error, high forecasting precision, and scientific forecasting model. As a result, the model can offer a good theoretical foundation for targeted cultivation of larch plantation.更多还原