摘要
针对传统BP神经网络容易发散、泛化能力差等问题，采用自适应Kalman滤波方法训练神经网络的连接权。与Kalman滤波训练连接权和传统的BP算法相比，该方法提高了BP神经网络计算精度，增强了泛化能力。实测数据的计算结果证明了该方法的可行性与有效性。
In this paper, adaptive Kalman filtering is proposed to train the weight of BP neural network for solving the problems of divergence and the low generalization ability. By comparisons, this method can improve the computing accuracy of BP newral network and generalization ability. It has been shown that BP based on adaptive Kalman filtering is feasible and efficient.
出处
《测绘科学与工程》
2007年第3期27-30,共4页
Geomatics Science and Engineering
基金
交通部科技项目（200531881203）和武汉大学地球空间环境与大地测量教育部重点实验室测绘基础研究基金（1469990324233-04-02）.
