CAJ | 학술논문

批量评估具有效率高、费用低且满足大量评估等优点。论文以中龄林为例，将BP神经网络应用于林木资源资产批量评估。通过比较学习算法、隐含层节点数，运用敏感性分析法确定影响因子对评估值的贡献程度，筛选输入层因子，从而优化了林木资源资产批量评估BP神经网络模型结构。结果表明：贝叶斯正则化法优于L-M算法；年龄、利率、蓄积、树种为强影响因子，这4个因子对评估值的贡献度超过60％；最优模型结构为BR 9-10-1，该模型平均绝对误差为32．46元/hm2，平均相对误差为1．28％，决定系数达0．9997，模型拟合精度高，泛化能力强，能够满足中龄林林木资源资产批量评估的要求。
비량평고구유효솔고、비용저차만족대량평고등우점。논문이중령림위례，장BP신경망락응용우림목자원자산비량평고。통과비교학습산법、은함층절점수，운용민감성분석법학정영향인자대평고치적공헌정도，사선수입층인자，종이우화료림목자원자산비량평고BP신경망락모형결구。결과표명：패협사정칙화법우우L-M산법；년령、리솔、축적、수충위강영향인자，저4개인자대평고치적공헌도초과60％；최우모형결구위BR 9-10-1，해모형평균절대오차위32．46원/hm2，평균상대오차위1．28％，결정계수체0．9997，모형의합정도고，범화능력강，능구만족중령림림목자원자산비량평고적요구。
Mass appraisal is of high efficiency,high precision,low cost,satisfies the needs of vast-amount evaluation.In this study,BP neural network was applied to mass appraisal of mid-age forest assets evaluation. By comparing different learning algorithms and the numbers of hidden layer nodes,selecting layer factors,using sensitivity analysis method which revealed the factors’ influence degree to the assessed value,the model struc-ture of BP neural network was optimized.The results showed that Bayesian regularization method was better than L-M algorithm;the contribution to the assessed values of the four factors including age,rate,accumula-tion,tree species was more than 60%;the best model structure was BR9-10-1.Its mean absolute error was 32.46 yuan/hm2 ,mean absolute percentage error was 1.28%,and decision coefficient was 0.999 7.The model has high fitting accuracy and generalization ability thus meets the requirement of mass appraisal of mid-age forest resource assets.