耦合自相关理论与小波神经网络的地下水模拟预测模型  

Groundwater Simulation and Prediction Model Based on Coupling Self-Correlation Theory and Wavelet Neural Network

作　　者：彭峰[1] 王林 PENG Feng;WANG Lin(Zunyi Survey and Design Institute of Water Conservancy and Hydropower,Zunyi,Guizhou,563002)

机构地区：[1]遵义水利水电勘测设计研究院,贵州遵义563002

出　　处：《红水河》2021年第2期96-100,共5页Hongshui River

摘　　要：地下水位演化趋势具有一定的振荡特性,使得其预测工作难以开展。针对该特性,引入自回归理论及小波神经网络(WNN)模型,提出了一种修正的地下水位预测方法。首先,该方法基于自相关理论对地下水位监测数据进行时间序列分析,选择合理先验区间作为预测模型的输入;然后,建立以合理先验区间为输入,当前地下水位为目标的WNN预测模型;最后,建立了基于传统回归预测算法的地下水位预测模型,预测结果与WNN进行验证分析。通过案例分析,选择了t-1至t-6作为合理的监测数据先验区间;进化131次的WNN能返回最优的预测结果,其精度明显优于KNN、AdaBoost、SVR和ANN等传统方法。笔者认为WNN是一种精确可靠的地下水位预测模型,能在该领域发挥一定的积极作用。The evolution trend of groundwater level has certain oscillation characteristics,which makes the prediction work difficult to carry out.Aiming at this characteristic,a modified groundwater level prediction method is proposed by introducing autoregressive theory and Wavelet Neural Network(WNN) model.Firstly,this method analyzes the time series of groundwater level monitoring data based on the self-correlation theory,and selects the reasonable prior interval as the input of the prediction model.Subsequently,a WNN prediction model with reasonable prior interval as input and current groundwater level as target is established.Finally,the prediction model of groundwater level based on traditional regression prediction algorithm is established,and the prediction results are verified and analyzed with WNN.Through case analysis,t-1 to t-6 are selected as the reasonable prior interval of monitoring data;The WNN with 131 evolutions can return the optimal prediction results,and its accuracy is significantly better than that of traditional methods such as KNN,AdaBoost,SVR and ANN.WNN is an accurate and reliable groundwater level prediction model,which can play a positive role in this field.

关 键 词：地下水位 模拟预测模型 自相关理论 小波神经网络 

分 类 号：P641[天文地球—地质矿产勘探]