基于水循环神经网络模型的海堤渗压预测研究

Seawall Seepage Pressure Prediction Based on the Water Cycle Neural Network Model

为提高海堤渗压的预测效果,在充分分析渗压影响因素的基础上,提出一种新型预测模型.将具有强大寻优能力的新型算法——水循环算法与神经网络相结合,以相关系数法筛选出的渗压主要影响因子作为模型输入,渗压作为模型输出,利用水循环算法搜索神经网络的最佳权值,建立海堤渗压水循环神经网络模型.通过上海浦东海堤实测信息对该模型进行验证,结果表明,水循环神经网络模型较BP模型具有更快的收敛速度和更高的预测精度.同时,基于等维新息思想,实现模型信息不断更新、渗压实时预测的效果.

In order to improve the predictive effect of the seawall seepage pressure,a new kind of prediction model was put forward based on the sufficient analysis of the influencing factors of seepage pressure. The Water Cycle Algorithm with strong searching ability was combined with the Neural Network. The primary influencing factors of seepage pressure selected by correlation coefficient method were taken as the input units of model. The seepage pressure was taken as the output unit of model. The Water Cycle Algorithm was used to search the optimum weights of the Neural Network to establish the Water Cycle Neural Network model of the seawall seepage pressure. This model’s effect was verified by the monitoring information from Pudong seawall of Shanghai. The results indicate that the Water Cycle Neural Network model has faster convergence rate and higher prediction accuracy than the BP model. Meanwhile,based on the concept of equal-dimension and new-information,the model ’s information can be updated constantly and seepage pressure can be predicted in real time.