基于微粒群算法的海堤渗压RBF神经网络监测模型

Seawall Seepage Pressure RBF Neural Network Monitoring Model Based on Particle Swarm Optimization

为提高海堤安全监控能力,从渗压实测数据以及潮位因子和降雨因子入手,使用影响因子的合理形式,利用径向基函数(RBF)神经网络建立渗压监测模型,推导微分进化微粒群优化算法(DPSO)速度和位移进化的数值计算方程,以此确定渗压RBF神经网络模型的聚类中心,并由此对渗压进行拟合和预测。以120组实测样本对模型进行训练拟合,并对后期60组渗压进行预测,得到拟合段平均相对误差为0.83%,相应预测段为1.71%。实际应用表明,经微分进化微粒群算法优化后,渗压RBF神经网络模型可以有效反映及预测渗压变化。

In order to improve the monitoring capacity of seawall, a seepage pressure monitoring model based on radial basis function（ RBF） neural network with reasonable forms of influencing factors is established based on observed seepage pressure data as well as tidal level data and rainfall factor data that influence seepage pressure. The numerical calculation equations of differential equations for speed and displacement evolution are derived, and the RBF centers are conformed by differential evolutionary particle swarm optimization（ DPSO） to fit and forecast seepage pressure. 120 measured samples are used for fitting and training the monitoring model with average relative error of 0. 83%, and the later 60 seepage pressure values are predicted with average relative error of 1. 71%. The actual application shows that, after optimized by differential evolutionary PSO, the seepage pressure monitoring model can effectively reflect and predict the general law of seepage pressure.