| 引用本文: | 丁朋朋,王苏桐,郭 政,等.基于Stacking集成机器学习的银川平原地下水位动态预测[J].灌溉排水学报,2026,45(8):113-120. |
| Ding Pengpeng,Wang Sutong,Guo Zheng,et al.基于Stacking集成机器学习的银川平原地下水位动态预测[J].灌溉排水学报,2026,45(8):113-120. |
|
| 摘要: |
| 【目的】精准模拟与预测银川平原地下水位的动态变化。【方法】基于银川平原33 a逐月地下水位、气温、降水量和蒸散发量的监测数据，采用偏自相关函数与时序互信息确定历史地下水位的最优滞后阶数，构建支持向量机回归（SVR）、随机森林（RF）、人工神经网络（ANN）、梯度提升决策树（GBDT）和极限梯度提升（XGBoost）地下水位动态预测的机器学习模型；在此基础上，以岭回归（Ridge）为元学习器，融合单一模型的预测结果，建立Stacking集成预测模型，开展银川平原地下水位动态预测。【结果】各监测井的历史地下水位最优滞后阶数介于6~8阶；集成模型预测性能显著优于5种单一模型，平均决定系数（R2）、绝对误差（MAE）、均方根误差（RMSE）分别为0.933、0.357 m与0.500 m，平均残差分布和平均残差均值分别为1.393 m和0.006 m。【结论】基于地下水位多阶滞后特征与Stacking集成策略构建的动态预测模型，可有效提升银川平原地下水位的预测精度与稳定性。 |
| 关键词: 地下水位预测；机器学习；Stacking算法；最优滞后；银川平原 |
| DOI：10.13522/j.cnki.ggps.2026018 |
| 分类号: |
| 基金项目: |
|
| Predicting groundwater level in the Yinchuan Plain using stacking ensemble machine learning methods |
|
Ding Pengpeng, Wang Sutong, Guo Zheng, Yang Xi, Tang Lijun, Guan Hong
|
|
1. School of Civil and Hydraulic Engineering, Ningxia University, Yinchuan 750021, China; 2. Key Laboratory of Digital Water Management for Yellow River Water Networking, Yinchuan 750021, China; 3. Ningxia Survey and Monitoring Institute of Land and Resources, Yinchuan 750002, China; 4. Ningxia Technical College of Wine and Desertification Prevention, Yinchuan 750199, China
|
| Abstract: |
| 【Objective】Groundwater is a main water resource in the Yinchuan Plain, and its dynamics is controlled by complex interactions of various natural and anthropogenic activities. This paper proposes a method to predict groundwater level changes in the plain.【Method】The model was developed based on 33 years of field-monitored data, including groundwater level, air temperature, precipitation, and evapotranspiration across the plain. The optimal lag order of groundwater level was determined using the partial autocorrelation function and time-series mutual information method. Five machine learning models, including support vector machine regression (SVR), random forest (RF), artificial neural network (ANN), gradient boosting decision tree (GBDT), and extreme gradient boosting (XGBoost), were constructed to predict groundwater level dynamics. Based on the prediction results of each of the five models, a stacking ensemble model was established using the Ridge regression as the meta-learner to integrate outputs of the five single models for simulating groundwater level variation in the plain.【Result】The optimal groundwater level lag order for the monitored datasets ranged from 6 to 8. The proposed ensemble model was significantly superior to the five individual standalone models, achieving an average coefficient of determination of 0.933, mean absolute error of 0.357 m, and root mean square error of 0.500 m; the average residual distribution range and average residual mean were 1.393 m and 0.006 m, respectively.【Conclusion】The model we developed based on the multi-time-lag characteristics of groundwater levels and the Stacking ensemble strategy effectively improved the accuracy and stability of groundwater level prediction in the Yinchuan Plain. |
| Key words: groundwater level prediction; machine learning; stacking algorithm; optimal lag; Yinchuan Plain |
