基于神经网络的GPS坐标转换方法研究

全球定位系统(GPS)与地理信息系统(GIS)结合是湿地研究的关键技术之一.将神经网络方法用于GPS全球坐标系统和地方坐标系统之间数据坐标转换研究中,建立了三层前馈神经网络模型.训练方法采用Levenberg-Marquardt算法,并改进其性能函数以提高网络的泛化能力.对神经网络模型的建立、网络的训练方法、仿真曲线、神经网络的输出结果进行了详细的讨论.仿真结果证明改进的Levenberg-Marquardt算法适用于GPS数据坐标转换问题.选择松辽流域为研究区域,以流域内扎龙湿地为中心进行了考查.结果表明,神经网络方法能够建立坐标转换的复杂模型,是一种效率高、准确性高的方法,在湿地保护的动态数据分析方面具有一定的意义.

Research on application of neural network to GPS coordinates transformation

The integration technology of Global Positioning System (GPS) and Geographic Information System (GIS) is one of the key technologies emerging in the wetland study. The neural network method is applied to the research on coordinates transformation method between local map coordinates system and global coordinates system of GPS data. A three-layer feed-forward neural network model is built to transform data. Levenberg-Marquardt algorithm is used in the transformation process, and a modified performance function is adopted in neural network training to improve the network generalization. Neural network model, principle of learning algorithm, simulation curves and output results are discussed in detail. The simulation results prove that Levenberg-Marquardt algorithm is suitable to the GPS data coordinates transformation. The study is undertaken on Songliao valley and Zhalong wetland in Heilongjiang Province which is taken as the central study area. The result of practice proves that neural network method of transformation succeeds in building the complex model of transformation and is an effective method with high precision in dynamic data analysis of wetland protection.