王梓璇,王圃,王颖,彭翰,华佩,张晋.基于小波分解及遗传BPNN耦合模型的地表水As浓度预测研究[J].环境科学学报,2021,41(7):2942-2950
基于小波分解及遗传BPNN耦合模型的地表水As浓度预测研究
- Wavelet decomposition and genetic BPNN hybrid model based modelling approach for As concentration prediction in surface water
- 基金项目：国家重点研发计划（No.2017YFC04047064-1）；中央高校基本科研业务费资助项目（No.2019CDCGHJ326）
- 王梓璇
- 重庆大学环境与生态学院, 重庆 400044
- 王颖
- 重庆大学建筑规划设计研究总院有限公司, 重庆 400045
- 华佩
- 华南师范大学环境学院, 华南师范大学环境理论化学教育部重点实验室, 广州 510006
- 张晋
- 暨南大学地下水与地球科学研究院, 广州 510632
- 摘要：随着工业的快速发展，水体中污染物超标事件时有发生，造成了较严重的水环境污染问题.水环境监测与预报是环境科学研究的重要内容.为了实现地表水砷（As）污染的准确预报，本研究提出小波分解、遗传算法与BP人工神经网络的耦合建模方法，并结合某河流监测站1998—2016年共19年的地表水质监测数据，通过皮尔逊相关系数和信息指标评价法对模型输入变量进行筛选，最后对比分析了在不同水质参数输入情况下BP人工神经网络（BPNN）、遗传算法改进的BPNN（GABP）、小波-遗传BPNN耦合模型（W-GABP）对后6年（2011—2016年） As浓度预测结果的均方根误差（RMSE）、决定系数（R2）、平均绝对百分比误差（MAPE），以确立最优模型.结果表明：①多水质参数BPNN、GABP与W-GABP耦合模型预测结果的MAPE分别为17.51%、15.98%、14.46%，单水质参数BPNN、GABP与W-GABP耦合模型预测结果的MAPE分别为18.78%、16.74%、7.83%；②小波分解数据前处理及遗传算法均能较大程度地提高预测模型的精度；③对于地表水水质预报，需对比不同模型在不同输入变量下的预测结果，以获得最佳的预测精度.单水质参数输入的W-GABP耦合模型能较准确地预报地表水As浓度的变化情况，对数据缺乏地区水质监控和地表水As污染防治具有重要意义.
- Abstract：With the rapid urbanization, anthropogenic pollutants in surface waters have resulted in serious water pollution problems. Nowadays, water status monitoring and prediction is an important part of environmental management. In order to provide an accurate arsenic (As) prediction in surface water, this study proposed a hybrid modelling method consisted wavelet decomposition, genetic algorithm and BP artificial neural network. based on a 19-year of surface water quality monitoring data, the input variables of the models were screened through the Pearson correlation coefficient and information index evaluation method. A comparative analysis of BP artificial neural network (BPNN), genetic algorithm-improved BPNN (GABP), wavelet-genetic BPNN hybrid mode l (W-GABP) under different water quality parameters input for the next six years of As prediction. The mean square root error (RMSE), coefficient of determination (R2), average absolute percentage error (MAPE) were used to filter out the optimal model. The results show that: ①MAPE values of the BPNN, GABP and W-GABP hybrid models with multi-water quality parameter were 17.51%, 15.98%, 14.46%, respectively. MAPE values of BPNN, GABP and W-GABP with single water quality parameter were 18.78%, 16.74%, 7.83% respectively; ②Wavelet decomposition data pre-processing and genetic algorithm could significantly improve the accuracy of the prediction results; ③the prediction results of different models under different input variables should be compared to obtain the best prediction accuracy. Therefore, the single water quality parameter W-GABP hybrid model could provide a more accurate prediction in the changes of surface water As concentration, which is of great significance for water quality monitoring and pollution prevention in areas with insufficient data.
