3种机器学习模型在地下水位动态模拟中的效果比较

Comparison of three machine learning models in dynamic simulation of groundwater level

为及时准确模拟区域地下水位动态变化,采用机器学习算法,在参数训练与特征子集参数筛选基础上,构建了基于极限学习机(ELM模型)、非线性自回归神经网络(NARX模型)和随机森林(RF模型)3种机器学习模型,对密怀顺区域地下水位动态变化进行了模拟,结果表明:3种机器学习模型在密怀顺区域地下水位动态模拟精度大小顺序为RF模型>NARX模型>ELM模型,与ELM模型相比,NARX模型和RF模型更适用于密怀顺区域地下水位动态模拟。地下水位动态变化模式分为波动上升型和稳定上升型2种类型,NARX模型适用于模拟地下水位变化呈波动上升型的监测井,RF模型适用于模拟地下水位变化呈稳定上升型的监测井。研究成果可为机器学习模型在地下水位分析中的应用提供方法参考。

In order to simulate the dynamic change of regional groundwater level,three machine learning models including Extreme Learning Machine(ELM),Nonlinear Auto-Regressive models with exogenous inputs(NARX)and Random Forest(RF)based on parameter subset selecting and training were established to simulate the dynamic change of groundwater level in Mi-huai-shun region.The comparison of the accuracy of the three machine learning models showed that RF model>NARX model>ELM model.Compared with ELM model,NARX model and RF model were more suitable for the simulation of groundwater level in Mi-huai-shun region.Additionally,the dynamic change modes of groundwater table in different monitoring wells were classified into two types:wavelike and steady rising.The NARX model was more suitable for simulating the wavelike rising of the groundwater level,while RF model was more suitable for simulating the steady rising of the groundwater level.The results of this study provide methodological reference for the application of machine learning model in groundwater level analysis.