基于典型相关系数和随机森林的水质预警方法

针对突发水污染事件提出一种高精度的预警方法.首先，通过模拟实验建立包含22种常见污染物的突发水污染事件数据库，然后采用典型相关系数准确揭示污染事件发生后多元水质参数之间的协同反馈规律.最后，基于多参数协同反馈规律构建“典型相关系数-随机森林”水质预警模型.结果表明预警模型对已知和未知污染物的平均准报率分别为96.78%和98.33%，对水质监测基线的平均误报率为0.16%.本研究成果可为降低突发水污染事件损失和保障供水安全提供有效的技术支撑.

Water quality warning method based on canonical correlation coefficient and random forest

This study proposed a high-precision early-warning method for detecting sudden water pollution incidents. Firstly, a database of sudden water pollution incidents containing 22common pollutants was established through simulation experiments. Secondly, the canonical correlation coefficients were used to accurately reveal the synergetic feedback law among various water quality parameters after pollution incidents. Finally, a water quality early-warning model, called "canonical correlation coefficients-random forest", was developed based on the multi-parameter synergetic feedback law identified above. Results show that the early-warning model's average true positive rates for known and unknown pollutants are 96.78% and 98.33%, respectively, while the average false positive rate under baseline status of water quality monitoring is 0.16%. The proposed early-warning model can provide practical technical support for reducing the loss of sudden water pollution incidents and ensuring the drinking water supply's safety.