马尾松人工林单木生长神经网络模型研究

The study on neural network models of individual tree growth in Pinus massoniana plantation

以马尾松人工林间伐试验林为研究对象 ,用林分内单木相对直径、林分年龄、地位指数和单株地积作为输入变量 ,以单木直径生长量作为输出变量 ,构建了 4∶S∶1的BP神经网络模型。用间伐试验林的单木生长数据对网络模型进行训练 ,得适宜的网络结构为 2∶3∶1,网络对象名取为DMnet。网络模型理论值与实际值比较的结果 ,各密度级的平均训练精度为 90 10 %～ 97 73%。与常规模型的拟合精度 83%～ 93%相比 ,神经网络模型取得了较好的拟合效果。网络模型的仿真结果显示 ,随相对直径和单株地积的增大 ,单木胸径定期生长量逐浙增大 ,其变化细节与林木在林分中所处的空间位置相符合 ,说明所建的单木生长神经网络模型能有效地描述林木生长随相对直径和林分密度指数的变化规律。模型可供同类条件的林分在进行经营设计时进行有关的分析、计算、模拟。

By diameter of individual tree, age of stand, site index and individual tree’s area as input variables, and by diameter increase of individual tree as output variable, a BP neural network model was created, in which the model structure is 4∶S∶1. The neural network model was trained by growth data of individual trees in thinning stands so that the best network structure is 2∶3∶1 that was named as DMnet. The diameter increase through the model was compared with actual diameter increase. The results show that the average training precision is 90.10% to 97.73% that is higher than usual technology’s 83% to 93%. In the stand management, the model can be used for analysis, computing, simulation, prediction and so on.