【作者】 朱瑜馨；

【导师】 赵军；

【作者基本信息】 西北师范大学 ， 自然地理学， 2003， 硕士

【摘要】 地理信息系统从20世纪60年代萌芽，经过30多年的发展、推广和应用，以数据采集、存储、管理、查询检索与可视化输出功能为主的传统GIS已经不能满足社会和区域可持续发展在空间分析、预测预报、辅助决策等方面的需求，GIS与专家系统、神经网络的结合成为GIS新的发展趋势之一。 论文在已完成的宁夏森林资源调查管理信息系统的基础上，采用人工神经网络(Artificial Neural Network，简称ANN)方法对森林资源调查管理信息系统的森林资源预测模型进行初步探讨。模型利用GIS处理基础数据、可视化输出预测结果，利用人工神经网络模型来扩展GIS空间辅助决策功能。 人工神经网络从20世纪40年代的萌芽到今天的开发现有模型以及在应用中根据实际运行情况对模型、算法的改造，已经广泛应用于各种行业。人工神经网络是人类在对其大脑神经网络认识理解的基础上人工构造的能够实现某种功能的非线性动力系统，是理论化的人脑神经网络的数学模型，是基于模仿大脑神经网络结构和功能而建立的一种信息处理系统。 人工神经网络方法是基于实例的方法，不需要考虑数学模型的内部结构，不需要假设前提条件，不需要人为地确定因子权重，作为一个黑箱综合地映射出研究对象的整体性，建模简单。 论文采用改进了的三层误差反向传播神经网络(Back-Propagation Network，简称BP网络)模型，以甘肃连城林场为例，对森林资源进行了预测。算法上采用含有动量因子的自适应调整学习率的变学习率算法对网络进行学习训练，以提高网络的学习速度，并且可以增加算法的可靠性。 论文在前人研究成果的基础上，总结了森林资源预测的基本思路，分析比较了多种预测方法，建立GIS空间数据库，运用GIS空间分析方法，对实验区现状进行评价，在现状评价的基础上提出了基于GIS的人工神经网络动态预测模型的理论与方法。论文中应用人工神经网络滚动预测和多步预测方法，分别采用5-25-5和4-10-1神经网络结构建立实验区红桦5个龄组的蓄积量预测模型与有林地面积预测模型，预测了实验区2000～2004年红桦各龄组蓄积量以及有林地面积。经过仿真结果与GM(1，1)模型预测结果的比较分析，蓄积量GM(1，1)预测模型相对误差的平均值为6.73犯14%，神经网络滚动预测模型相对误差的平均值为0.027571%;有林地面积GM (1，1)预测模型相对误差的平均值为1.41434%，神经网络多步预测模型相对误差的平均值为0.03863%，说明基于Gis的神经网络动态预测模型的精度高，建模简单，实用性强。 本项研究中，基于GIS的神经网络预测模型主要侧重的是地理实体数量时间结构序列，模型结合森林资源复杂的空间和属性特征，不仅使用了Gis关系数据库中的属性时间序列值，同时也使用了一定的空间模型，实现了空间模型与属性模型的有效结Z、口0 在程序的实现上采用M八TLAB开发环境，其中的神经网络工具箱以人工神经网络理论为基础，构造了网络分析和设计的许多工具函数。运用神经网络工具箱进行预测研究，程序代码书写简便，与VB开发语言集成方便，预测过程易于实现。 论文最后根据预测结果及实验区实际情况提出了相应的林业发展对策，总结了基于GIS的森林资源神经网络动态预测模型的特点与不足.更多还原

【Abstract】 GIS originated in 1960s, with the development and application of more than 30 years the traditional GIS which functions are gather, storage, management, query and visual output of data can’t meet with the demands of social and regional sustainable development. Integration between GIS with expert system and ANN becomes one of tendency.Thesis primarily studies prediction model of the forest resources inventory management information system in ANN based on finished information system of the forest resources inventory management of NingXia Hui-autonomous region. By primary data process, visual output of GIS and ANN model, thesis has expanded the spatial assistant decision function.ANN has already applied in many aspects since it originated in 1940s and applied in design of existing model and bettering its model and arithmetic today. ANN is an artificial nonlinear dynamic system based on the recognition of cerebra neural network theory. ANN is a theoretic cerebra neural network mathematic model and an information processing system based on imitating cerebra neural net structure.The method of ANN based on example does not need to deal with internal structure of mathematic method, suppose premise, decide factor weight. ANN model is simple because it maps integration of object.Thesis forecasts forest resources of Liancheng forest fields in Gansu province by improved three-layer BP network. ANN is trained through fast BP algorithm with variable learning rate that mixed with momentum factor. This algorithm can improve network’s reliability.Thesis summarizes the basic thought of forest resources forecast, compares many forecasting method based on pioneer achievement. Thesis evaluates actuality of experimental district in GIS spatial analysis method through GIS spatial database and then puts foreword the theory and method of dynamic prediction based on GIS in ANN. Thesis establishes forest volume prediction model of red-birch’s five-age group and the wooded area prediction model by rolling prediction and multi-step prediction of ANN which structures are 5-25-5 and 4-10-1. Thesis predicts red-birch’s five-age group volume and thewooded area of experimental district from 2000 to 2004. In order to evaluate the precision of the model, the author establishes the GM(1,1) model. The average of relative error is 6.733214 percent through GM(1,1) model of forest volume and the ANN model’s is 0.027571 percent. The average of relative error is 1.41434 percent through GM(1,1) model of wooded area and the ANN model’s is 0.03863 percent. We can conclude that the model of dynamic prediction Based on GIS in ANN is better than the model of GM(1,1) by the comparison ANN model to GM(1,1) model.In this study, the model emphasizes particularly on time series of geological entity and at the same time it realizes the integration of the spatial model and the attributive model by integrating complicated spatial and attributive character of forest resources.Program is realized by MATLAB. The ANN toolbox of MATLAB established many tool functions based on ANN theory. Prediction is easily realized and program codes are easily written and integration with visual basic is easy by ANN toolbox.At last, thesis puts forward measures of forestry development according to prediction and actual experimental district condition, analyzes model’s character and points out model’s limitation for improving in further study.更多还原