摘要
针对传统的BP神经网络存在的缺点,提出了用附加动量法、自适应学习速率和L-M优化算法等几种算法进行优化。通过对比分析,证明了采用L-M优化和附加动量因子算法相结合取得了最优的预测效果。该方法克服了BP神经网络模型存在的收敛速度慢、易陷入局部极小点的缺点。结合现场实测数据,将该优化模型与传统的BP神经网络预测模型对比,预测结果表明改进的BP神经网络在路基沉降预测中精度最高,适宜广泛采用。
Aimed at the shortcomings of traditional BP neural network,the optimization methods such as additional momentum method,adaptive learning rate and L-M optimization algorithm are proposed.Through comparison and analysis,the combination of L-M optimization and additional momentum factors algorithm is proved to be the optimal prediction effect.The shortcomings of BP neural network model are overcome,such as the slow convergence speed and falling into local minimum value easily.Combined the measured data in the field,the new model and traditional BP neural network prediction model are compared,the predicted results show that the improved BP neural network has the highest accuracy in settlement prediction of subgrade,and it has wide application.
出处
《勘察科学技术》
2010年第5期28-31,共4页
Site Investigation Science and Technology
关键词
神经网络
附加动量法
自适应学习速率
L-M优化算法
沉降预测
neural network
additional momentum method
adaptive learning rate
LM optimization algorithm
settlement prediction
