Abstract: In order to solve the ubiquitous problems on cable-stayed bridge’s loading test which are expensive cost and long testing time, this paper builds a practical cable-stayed bridge’s finite element model, analyzes the mechanics characteristics of cable-stayed bridge, and puts forward a method of loading test conditions’ merging. Finally, Dezhou bridge’s loading tests are finished with the merging method. The research results show that for cable-stayed bridges, beam maximum moment condition and beam maximum deflection condition can combine into one condition; and tower maximum moment condition and top tower horizontal displacement can combine into one condition. The merging method can be applied to practical bridge’s loading test and can popularize to other long-span bridge type’s loading test. In summery, the research result has a wide application prospect.

Key words: pit deformation, engineering safe, predict research, wavelet neural network, parameter optimization

摘要： 针对一般小波神经网络存在的学习时间长，网络预测精度低的问题，提出了对网络输入层权值初始值进行归一化处理的优化方法，改进了原有小波神经网络。将改进后的模型应用于某市轨道交通1号线珠江路站深基坑水平变形预测中。监测结果表明，网络输出值与实测值吻合很好，优化后的小波神经网络收敛速度也更快；同时随着大量最新的监测数据输入到网络中学习，将使深基坑水平变形预测更加精确。

关键词: 基坑变形, 工程安全, 预测研究, 小波神经网络, 参数优化