滇池草海水质等级预测模型研究

综合考虑影响水质的不同类别因子之间的关系,为水质等级预测提供平均预测精度更高的模型。选取的p H、DO、CODMn、NH3-N、历史水质等级5个水质因子数据来源于中国环境保护部官方发布的水质数据,降雨量、光照时间2个气象因子数据来源于云南省气象局官方发布的气象数据。首先利用改进的灰色模型(Adaptive Grey Model,AGM)进行单因子预测,从而获取BP人工神经网络(BP Artificial Neural Network,BPANN)训练集和水质等级残差序列;然后使用经过训练集训练后的BPANN进行水质等级残差纠正;最后利用AGM模型得到未来水质等级,以滇池草海2006-2013年水质周报资料和气象资料为数据基础进行了仿真分析和验证实验。结果表明:(1)AGM模型对水体因子和气象因子的单项指标预测理想,保证了作用于BP人工神经网络数据的可靠性,同时降低了预测误差的传输;(2)来源于中国环境保护部与云南省气象局的数据保证了水质等级预测中数据的权威性,采用AGM-BPANN组合模型预测滇池草海水质等级精度达到90.2%,说明模型适用于同一地区短时间内的水质变化研究;(3)AGM-BPANN组合模型借助BP网络的高维非线性克服了数据突变对预测的影响,在AGM预测基础上,通过纠正预测残差获得最终的水质等级值,实现了对滇池草海短时间内水质的预测。

Water Quality Grade Prediction Model for the Caohai Basin of Dianchi Lake

Pollution is becoming ever more serious in the Caohai section of Dianchi Lake, leading to socio-economic losses in the drainage area and arousing wide social concern. A forcasting model for water quality would support water pollution control efforts and reduce the losses. In this study, a robust model (AGM-BPANN) was used to achieve higher accuracy of water quality predictions, combining two models: an adaptive model (AGM) and a back propagation artificial neural network (BPANN). The relationships among various water quality factors were fully developed and seven parameters (pH, DO, CODMn, NH3-N, historical water quality grade, rainfall and duration of sunshine) were selected as model parameters. Data on the five water quality parameters were obtained from the Chinese Ministry of Environmental Protection, and data on the two meteorological factors were obtained from the Yunnan Meteorological Bureau. First, the adaptive grey model was used to predict parameter values for use as a training set for back propagation of water quality by the artificial neural network (BPANN) and a sequence of water quality grades were obtained. The model was then calibrated with BPANN, using historical water quality data, and future water quality levels were predicted using AGM-BPANN. The simulation analysis and verification experiments were carried out based on weekly water quality data and meteorological data for Dianchi Lake from 2006 to 2013. Results indicate that: (1) The predicted result of a single water quality parameter with a single meteorological factor by AGM is satisfactory, improving predictions, increasing data reliability for BPANN and reducing the error propagated through the model. (2) The data derived from government authorities ensures reliable prediction of water quality. The accuracy of water quality prediction in the Caohai basin of Dianchi Lake was up to 90.2% using the combined model, indicating that the proposed model is applicable to studies of water quality changes in a short timeframe. (3) The AGM-BPANN model combines the high-dimensional nonlinearity of the BP network to overcome the influence of data mutation on the prediction result and achieved the goal of predicting water quality in the Caohai section of Dianchi Lake in the short term.