基于RS、GIS和ANN的珠江流域水质分析技术的研究

【作者】 王孝武；

【导师】 孙水裕；

【作者基本信息】 广东工业大学 ， 环境工程， 2004， 硕士

【摘要】 本论文以珠江水系广州河段为研究对象，提出了联合使用遥感(Remote Sensing，简称RS)、地理信息系统(Geographical Information System，简称GIS)和人工神经网络(Artificial Neural Networks，简称ANN)技术对珠江水质进行分析监测的新技术。该新技术能大大减轻传统的监测所耗人力和物力，并能对大范围水域进行快速监测。 基于RS、GIS和ANN的监测技术主要是利用GIS对RS图像进行地理定位，然后利用RS专用软件提取定位点的RS数据，最后通过建立前馈误差反传人工神经网络(BP网络)，确定TM(Thematic Mapper，专题成像仪，简称TM)数据前5个波段在定位点的反射率与水质三个主要参数的映射关系，即建立RS数据与水质的映射模型。本研究对模型建立过程、模型检验进行了探讨，得出了该监测技术完全能满足当前实际需要。 本研究是建立在当前水色RS技术理论之上的，即RS数据与河水组成是呈复杂函数关系的，但当前水色RS卫星主要是针对海洋设计的，并不适用于内陆江河的水质分析，所以采用了陆地卫星Landsat系列的TM数据。又由于陆地卫星并不是针对水色RS理论建立起来的，所以需重新建立TM数据与水质参数的映射关系，这里采用人工神经网络技术建立两者之间的复杂关系。 数据采集整理和模型建立是本研究中的两个重点。数据采集主要利用GIS和RS技术，同时采集相同时段的历史监测数据，数据整理主要是针对模型的训练数据，要采取均匀性原则。人工神经网络模型是最常用的含有一个隐层的两层BP(误差反向传播，Error Black Propagation，简称BP)网络，充分利用训练集数据和校验数据，确定了最佳隐节点数目是7个，最佳训练次数是1000次。 利用这种新技术，可使预测误差小于20％，已经能够满足现实需要。更多还原

【Abstract】 In this paper, remote sensing (RS) , geographical information system (GIS) and artificial neural networks (ANN) are combined to monitor River Pearl in Guangzhou city. This combination is regarded as a new technology for water quality. Compared to conventional measurement methods, it has advantages in decreasing manpower and material resource. In addition, it makes the rapid monitoring of big river possible.In the application of the new measurement technology based on RS, GIS and ANN , firstly, geography location is received by GIS and RS pictures. Secondly, RS data of location is gained by exclusive RS software. Finally, BP network is built up , so the relation between echo rate of five frontal bands of TM data is found.The new technology is built on current theory of water color RS, which shows that the relation between RS data and constitutes of river water. But the technology uses TM data of Landsat series because the satellites of water color RS are not designed for rivers inland but only for ocean. In the same way, the relation between TM data and water parameters is built by ANN in the technology, because Landsats are not made on theories of water RS.Two important contents in the technology are collection and pretreatment of data, and model’s building process. GIS and RS are used during the process of collection of data, at the same time, synchronous history monitoring data is gained. Average principle is adopted during data’s pretreatment. The model of ANN is one BP network of two layers including one hidden layer, which has seven nodes in hidden layer and whose perfect train times is 1000.The error of forecast by the new technology is less than 20 percentage, and so the new technology can reach practical need.更多还原