基于PSO-BP神经网络的北江大堤渗流预测模型

Seepage Prediction Model of Beijiang Dike Based on BP Neural Network

以北江大堤石角灵洲段作为研究对象,以降雨量、外江水位等作为输入,测压管水头值作为输出,建立了BP、GA-BP、PSO-BP 3种渗流预测模型。在前期对模型参数设置优化的基础上,考虑通过增加训练集、优化迭代步数,进一步提高渗流模型的精度。最后从北江大堤安全监测系统导出11个月监测数据,筛选使用其中292组数据进行训练,取30组数据作为输出,分析发现PSO-BP模型为最佳模型,此时的RMSE、MAPE、R2分别为0.31、5%、0.983。相对于大型有限元仿真软件、多元线性回归模型等传统方法,预测模型可以减少计算成本,处理非线性问题时也能够获得更好的效果。

This paper takes the Shijiao Lingzhou section of the Beijiang Dike as the research object,takes rainfall and the water level of the outer river as input and the piezometer head value as the output,and establishes BP,GA-BP,PSO-BP three kinds of seepage predictions Model.The 11-month research data was derived from the Beijiang Dike Safety Inspection System,and 292 sets of data were selected for training,and 30 sets of data were taken as output.After analysis,it was found that the PSO-BP model was the best model.At this time,RMSE,MAPE,and R2 were respectively It is 0.31,5%,0.983.Finally,the influence of iteration steps and training set on the model is studied,and the most accurate prediction model is selected.Compared with traditional methods such as large-scale finite element simulation software and multiple linear regression models,the prediction model proposed in the article can reduce the calculation cost and can also obtain better processing effects in the face of nonlinear problems.