[关键词]
[摘要]
基于高频水质在线监测数据, 结合遗传算法和神经网络模型, 建立基于遗传-神经网络( Improved Genetic Algorithm-Back Propagation Neural Network, IGA2BPNN) 的河流水质预测模型, 实现对河流水质的实时预测预警。 将该方法应用于美国波托马克河流中, 对其水质参数浊度( TURB) 和电导率( SC) 进行实时预测, 并对预测结果进行 性能分析, 以验证基于 IGA2BPNN 的河流水质预测模型的准确性与可靠性。与 BPNN 模型的水质预测结果进行对 比分析, 结果表明: IGA2BPNN 模型对水质参数 TURB 和 SC 有更准确的预测效果。同时, IGA2BPNN 模型对正常 平稳条件下的水质参数 TURB 和 SC 预测结果的区间覆盖率 PICP 分别为 99.81% 和 100% , 预测结果具有一定的 可靠性。IGA2BPNN 水质预测模型可以有效地识别长时间的水质异常或瞬时显著的水质变化情况, 可实现对河流 水质的风险预警, 最终可为河流突发水污染的应急处置措施的制定提供科学依据。
[Key word]
[Abstract]
A surface w ater quality predictio n model is developed to analyze the inherent water quality variation tendencies and provides r eal-time early warnings according to the high-frequency water quality historical observation data. The developed water quality prediction model is integrated by an improved genetic algorithm ( IGA) and a Back Propagation Neural Network ( BPNN) . To verify the accuracy and reliability of the river water quality prediction model based on IGA-BPNN, the method is applied to the Potomac River in the United States to predict its water quality parameters, turbidity ( TURB) and conductivity ( SC) , and to analyze the performance of the prediction results. The prediction results demonstrate that the developed IGA-BPNN model can provide a more accurate prediction result than the BPNN model. Since the PIPC values of TURB and EC predictio n can reach 991.81% and 100% under normal stable conditions, GA2BPNN model also can provide a reliability prediction result. Meanwhile, the developed IGA-BPNN models can reflect the long period and iso lated sharp peaks of the water quality variations, and effectively provide real-time early warning for emergency response.
[中图分类号]
[基金项目]
国家自然科学基金( 51779066) ; 国家重点研发计划课题( 2018YFC0408001) ; 中国博士后科学基金面上项目( 2018M631935) ; 哈尔滨学院大学生科技创新项目( HXS20171506)