摘要
介绍了神经网络BP(Back Propagation)算法存在的几个缺点,针对这些缺点提出了2个改进措施:"学习速率自动调整法"和"误差分级迭代法"。结合一个工程实例,详细分析了BP改进算法的应用效果。工程实例为某地区"高程异常"模型的建立。就该工程实例,与常规BP算法相比较,"BP改进算法"的收敛速度提升了约28%,其模拟精度更高,模拟结果更稳定。"BP改进算法"的工程应用效果良好,值得借鉴与推广。
The shortcomings of neural network BP algorithm are introduced. Two improvement algorithms are proposed for overcoming the shortcomings. One is ＂Automatic Adjustment of Learning Rate Method＂, the other is ＂Error Grade Iterative Method＂. Combined with an engineering example, the application effect of the improved BP algorithm was analyzed in detail. The engineering example is to build the height anomaly model of a certain area. For the engineering example, the improved BP algorithm has higher simulation accuracy and more stable simulation results than the conventional BP algorithm, and its convergence rate is improved about 28%. The engineering application effect of the improved BP algorithm is good. The improved BP algorithm is worth learning and promotion.
出处
《现代测绘》
2016年第6期1-4,共4页
Modern Surveying and Mapping
基金
国家自然科学基金项目(41274028
41574022)
江苏省科技支撑计划项目(BE2014026)
关键词
神经网络
BP算法
高程异常
neural network
BP algorithm
height anomaly
