Yield model of Cunninghamia lanceolata plantation based on back propagation neural network and support vector machine.
-
摘要: 以闽西北杉木人工林为研究对象,选取涵盖中龄林、近熟林、成熟林3个龄组的700个小班作为样地进行调查,以林龄、地位指数、林分密度、平均胸径作为输入变量,单位蓄积量为输出变量,运用BP神经网络和支持向量机2种机器学习方法建立林分收获模型,并采用遗传算法对模型参数进行优化。随机将样本数据分成350个训练样本和350个验证样本,对不同模型的拟合精度、预测精度进行对比分析,其中参数优化后的BP神经网络和支持向量机模型训练样本精度分别达到0.935 37和0.936 33,预测结果精度分别为0.921 30和0.926 97,训练样本和验证样本的总体拟合平均相对误差值均低于7%。分析结果表明,2种模型拟合精度高、预测性能好,为杉木人工林林分收获模拟和预测奠定了基础。为比较2种方法预测结果的差异性,将350个验证样本样地平均分为7组,分别用优化后的2种模型计算各组的预测精度,对预测精度与训练精度的差值进行t检验,结果表明,2种建模方法的预测结果不存在显著性差异,但模型精度的提高对森林资源的精确监测和森林生长动态预测具有重要的理论价值。同时,研究发现支持向量机模型的拟合精度和泛化能力均优于BP神经网络,该方法为收获模型研究提供了新思路。Abstract: Based on data from 700 sample plots of Cunninghamia lanceolata plantations in the northwestern area of Fujian Province, consisting of middle-aged, near-mature and mature trees, we established a yield model with stand age, site index, stand density and average diameter at breast height (DBH) as input variables and stand volume as the output variable. We used two machine learning methods, i.e., a back propagation (BP) neural network and a support vector machine (SVM). The parameters used in both modeling methods were optimized by a genetic algorithm. We randomly divided the plots into two halfs, i.e., a 350 plot training set and a 350 plot test set and compared the fit and prediction accuracies of both models. After parameter optimization, the accuracies in fitting both models were 0.935 37 for BP and 0.936 33 for the SVM, with prediction accuracies of 0.921 30 for BP and 0.926 97 for the SVM. The average relative errors of both the training and test sets were less than 7% for both models. We conclude that both models established a basis for simulating and predicting stand yield of C. lanceolata plantations, given that the accuracy of both models was quite high with good performance of prediction. In order to analyze the differences in the results of both models, the 350 plot test set was evenly divided into seven groups. We calculated the prediction accuracy in each group with two optimization models and used a t-test to compare the absolute difference between the prediction and fitting accuracy. The results showed no significant difference between the two methods. All the same, the slight improvement in precision is important and valuable for monitoring forest resources as well as for predicting the dynamics of stand growth. We found that the accuracy of the fit and generalization ability of SVM were better than those of BP. Therefore, we recommend the SVM for providing new insights for research in forest yield models.
-
Key words:
- Cunninghamia lanceolata /
- yield model /
- machine learning /
- predicting /
- fitting
-
[1] HONG W, WU C Z, HE D J. A study on the model of forest resources management based on the artificial neural network[J].Journal of Natural Resources, 1998,13(1): 69- 72. [2] 洪伟, 吴承祯, 何东进. 基于人工神经网络的森林资源管理模型研究[J]. 自然资源学报, 1998, 13(1): 69- 72. [3] HUANG J R, GAO G Q, MENG X Y. Forecasting stand diameter distribution based on artificial neural network.[J].Journal of Beijing Forestry University,2010,32(3):21- 26. [4] DIAMANTOPOULOU M J, MILIOS E. Modelling total volume of dominant pine trees in reforestations via multivariate analysis and artificial neural network models[J]. Biosystems Engineering, 2010, 105(3): 306- 315. [5] 黄家荣, 高光芹, 孟宪宇, 等. 基于人工神经网络的林分直径分布预测[J]. 北京林业大学学报, 2010,32(3): 21- 26. [6] LI L S. Research on growth and profit model of Taxus chinensis var. mairei plantation [D]. Beijing: Beijing Forestry University, 2011. [7] XU B Q, ZHANG Q L, MI H Z, et al. Growth model of Pinus tabulaeformis plantation based on BP neural network [J]. Journal of Northeast Forestry University, 2012, 39(12):33- 35. [8] 李良松. 南方红豆杉人工林生长与收益模型研究 [D]. 北京:北京林业大学, 2011. [9] 徐步强, 张秋良, 弥宏卓, 等. 基于BP神经网络的油松人工林生长模型[J]. 东北林业大学学报, 2012, 39(12): 33- 35. [10] YANG X, ZHANG Q L. Based on artificial neural network modeling of the Larix principis rupprechtii Mayr plantation at daqing mountain in Inner Mongolia[J]. Journal of Inner Mongolia Agricultural University: Natural Science Edition, 2012,33(5- 6):76- 79. [11] CHE S H. Growth modeling for Chinese fir plantation based on artificial neural network [D]. Beijing: Chinese Academy of Forestry, 2012. [12] 杨潇, 张秋良. 基于BP人工神经网络的大青山自然保护区华北落叶松人工林全林分生长模型研究[J]. 内蒙古农业大学学报:自然科学版, 2012, 33(5- 6):76- 79. [13] CHEN W P. Analysis of the influence of young Chinese fir's different plantation tending modes on the amount of growth [J]. Straits Science, 2008 (8): 26- 27. [14] 车少辉. 基于神经网络方法的杉木人工林林分生长模拟研究[D]. 北京:中国林业科学研究院, 2012. [15] 陈文平. 杉木人工林不同幼林抚育模式对生长量影响的分析[J]. 海峡科学, 2008(8): 26- 27. [16] HU X X. Analysis of landscape pattern of Longqishan national nature reserve and its ecology evaluation [D]. Fuzhou: Fujian Agriculture and Forestry University, 2009. [17] AVERY T E, BURKHAR H E. Forest Measurements[M]. 4th ed. New York: McGraw-Hill Book Company, 1994. [18] YUE C R. Forest biomass estimation in Shangri-La County based on remote sensing [D]. Beijing: Beijing Forestry University, 2011. [19] LIN H, HONG T, CHEN H, et al.Multi-species design in planting industrial forests by genetic algorithm[J]. Scientia Silvae Sinicae, 2010,46(5):92- 101. [20] DAVIS L S, JOHNSON K N. Forest Management[M]. New York: McGraw-Hill Book Company, 1987. [21] REINEKE L H. Perfecting a stand-density index for even-aged forests[J]. Journal of Agriculture Research, 1933,46: 627- 638. [22] DANIEL T W, HELMS J A, BAKER F S. Principles of silviculture[M]. New York: McGraw-Hill Book Company, 1979. [23] GULLU M, YILMAZE I·, YILMAZE M, et al. An alternative method for estimating densification point velocity based on back propagation artificial neural networks[J]. Studia Geophysica et Geodaetica, 2011, 55(1): 73- 86. [24] KOTB M T, HADDARA M, KOTB Y T. Back-propagation artificial neural network for ERP adoption cost estimation[J]. Enterprise Information Systems, 2011,220: 180- 187. [25] CORTES C, VAPNIK V. Support-vector networks[J]. Machine Learning, 1995, 20(3): 273- 297. [26] BURGES C J C. A tutorial on support vector machines for pattern recognition[J]. Data Mining and Knowledge Discovery, 1998, 2(2): 121- 167. [27] 胡欣欣. 龙栖山国家级自然保护区森林景观格局分析及其生态评价[D]. 福州:福建农林大学, 2009. [28] 岳彩荣. 香格里拉县森林生物量遥感估测研究[D]. 北京:北京林业大学, 2011. [29] 林晗, 洪滔, 陈辉, 等. 应用遗传算法的工业原料林多树种造林设计[J]. 林业科学, 2010, 46(5): 92- 101. [30] SEDIGHHI M, AFSHARI D. Creep feed grinding optimization by an integrated GA-NN system[J]. Journal of Intelligent Manufacturing, 2010, 21(6): 657- 663. [31] FUNAHASHI K I. On the approximate realization of continuous mappings by neural networks[J]. Neural Networks, 1989, 2(3): 183- 192. -

计量
- 文章访问数: 617
- HTML全文浏览量: 60
- PDF下载量: 10
- 被引次数: 0