提升小波多尺度分解法与RBF神经网络在混凝土重力坝应力监测中的应用

Application of Lifting Wavelet Multi-scale Decomposition Method and RBF Neural Network in Stress Monitoring of Concrete Gravity Dam

将混凝土坝的应力监测数据视为一组含有不同频率信号的时间序列,针对信号中的低频信号和高频信号,采用提升小波的多尺度分解法提取混凝土坝应力的时效分量、水压和温度分量以及噪声分量,采用计算信噪比极大值的方法提取周期性分量中的水压、温度分量和噪声分量,得到最优的去噪效果。使用RBF神经网络对去噪信号进行建模,预测结果表明,该模型能够很好地反映混凝土坝应力变化的趋势和规律,可应用于混凝土坝安全监测中。

The observation data of concrete dam stress is regarded as a set of time series which contains different frequency signals. Aiming at the signal with high frequency or low frequency, a new method of lifting wavelet multi-scale decomposition is proposed to extract the time dependent component, the water pressure and temperature component as well as the noise component from the observation date of dam stress, and then the method of calculating maximum SNR is used to extract the components of water pressure and temperature and noise component from cyclical component, which can get the most optimal denoising effect. The RBF neural network is finally used to model the denoising signals for stress prediction. The prediction results show that the model can well reflect the trends and patterns of stress changes of concrete dam.