基于贝叶斯累加回归树模型的溶解氧影响因素分析及预测研究——以长江中游为例
Analysis and Prediction of Factors Affecting Dissolved Oxygen Based on a Bayesian Additive Regression Tree Model—A Case Study of the Middle Reaches of the Yangtze River

本文以长江中游三个河流断面每隔4小时的水质数据和气象指标为研究对象进行溶解氧的影响因素分析和预测研究。首先，通过模拟数据对BART、SVR、RF和XG模型进行实验对比，分析了BART模型的稳健性优势。其次，运用BART模型变量重要性(变量包含比例)来筛选重要变量，筛选出影响长江中游水质溶解氧浓度的关键变量为pH、水温、电导率、总磷、高锰酸盐指数、露点温度、气温。最后，选择筛选出的最优输入变量组合来构建BART、SVR、RF和XGB模型来对溶解氧进行预测研究。结果表明，BART模型的预测性能最佳，在100次预测统计中R²、RMSE、MAE、MAPE的中位数(M_e)分别为0.936、0.480、0.333、4.36%。
This study focuses on the analysis and prediction of dissolved oxygen (DO) in the middle reaches of the Yangtze River, using water quality data and meteorological indicators measured at three river cross-sections every 4 hours. First, experiments were conducted to compare the performance of four models: BART, SVR, RF, and XGB, using simulated data. The robustness advantage of the BART model was highlighted. Next, the variable importance of the BART model was utilized to select key influencing factors, identifying the critical variables affecting DO concentration in the middle reaches of the Yangtze River. These key variables include pH, water temperature, electrical conductivity, total phosphorus, permanganate index, dew point temperature, and air temperature. Finally, the optimal combination of input variables was selected to build the BART, SVR, RF, and XGB models for DO prediction. The results show that the BART model performs best, with median values of R², RMSE, MAE, and MAPE of 0.936, 0.480, 0.333, and 4.36%, respectively, from 100 prediction trials.