随机森林回归模型及其在污水排放量预测中的应用

Application of random forest regression model for wastewater discharge forecasting

以万元GDP用水量及万元工业增加值用水量为影响因子,提出了基于随机内插构造样本的随机森林回归（RFR）年污水排放量预测模型.结果表明,采用RFR模型对全国2000—2011年污水排放量进行预测,随机连续运行100次的平均相对误差绝对值和最大相对误差绝对值分别为1.50％ ～2.58％和3.36％～6.11％,均值分别为1.92％和4.39％;模型对全国2012-2015年、2020年及2030年污水排放量的预测平均值分别为679.1×108,679.5×108,700.6×108,710.6×108,735.0×108和765.4×108 m3.RFR模型具有预测精度高、泛化能力强、稳健性能好以及调节参数少等特点.

Taking the ten thousand yuan GDP water consumption and ten thousand yuan of industrial added value of water consumption as the impact factor, the random forest regression （RICR） annual wastewater discharge forecasting model which based on random interpolated structure sample was put forward. The results indicated that with random and continuous operation of RFR model for 100 times, the average relative error absolute value and maximum relative error absolute valve of annual wastewater discharge forecasting were from 1.50% to 2.58% and from 3.36% to 6.11% respectively from 2000 to 2011 in the whole country, the average was 1.92% and 4.39%. The annual wastewater discharge forecasting results were 679.1 × 108, 679.5 × 108, 700.6 ×108, 710.6 × 108, 735.0 × 108 and 765.4×108 m3 respectively in 2012, 2013, 2014, 2015, 2020 and 2030 by this model. The RFR had the characteristics of high precision prediction, strong generalization ability, good robustness and less adjustable parameters etc.