【摘要】 为探究目标断面低溶解氧(DO)变化规律，构建反演预测模型预测其浓度水平。首先基于随机森林(RF)特征重要性分析得出“硝化耗氧、有机物耗氧是导致目标断面季节性低氧两个重要因素”的结论，然后利用优选出的NRBO-LSSVR反演模型开展溶解氧敏感度因子分析;基于随机森林特征重要性分析筛选出的8项重要参数开展溶解氧短时预测，结果表明RF性能显著优于其他3种模型，且RF预测未来3天决定系数分别为0.83、0.62和0.57，预测未来1天时精度最高，可将DO平均误差控制在±0.18 mg/L以内; 24、48 h预测准确率分别为79.6%、71.5%，其中20 h决定系数(R~2)最高，为0.912，平均误差控制在±0.15 mg/L以内。本研究优选出的模型能够在夏秋季节缺氧时段保持良好的预测性能，可实现目标断面的快速精准预测。更多还原

【Abstract】 In order to explore the change pattern of low dissolved oxygen in the target section and accurately predict the dissolved oxygen concentration level,multiple dissolved oxygen inversion and prediction models were constructed. First,based on the importance analysis of random forest characteristics,it was concluded that “oxygen consumption by nitrification and oxygenconsumption by organic matter are two important factors causing seasonal hypoxia at the target section”. Then sensitivity factor analysis on dissolved oxygen was conducted by using the optimized NRBO-LSSVR inversion model. Finally,based on the random forest feature importance analysis,8 key parameters were selected for short-term dissolved oxygen prediction. The results showed that the RF model performed significantly better than the other three models,with coefficients of determination(R~2) of 0. 83,0. 62,and 0. 57 for predicting the next 1,2,and 3 days,respectively. The highest accuracy was achieved for 1-day predictions,with the mean error controlled within ± 0. 18 mg/L. The prediction accuracy rates for 24 h and 48 h were 79. 6% and 71. 5%,respectively. Among these,the 20 h prediction achieved the highest R~2 value of 0. 912,with the mean error controlled within ± 0. 15 mg/L. Therefore,it is considered that the preferred model selected in this study can maintain good prediction performance in summer and autumn,which can achieve fast and accurate prediction of target section.更多还原