王丽婧,雷刚,韩梅,吴光应,李虹.数据缺失条件下基于MLP神经网络的水华风险预警方法研究[J].环境科学学报,2015,35(6):1922-1929
数据缺失条件下基于MLP神经网络的水华风险预警方法研究
- Early warning method for algae bloom risk based on MLP neural network under the conditions of missing data
- 基金项目：十二五国家水体污染控制与治理重大科技专项(No.2012ZX07503-002)
- 王丽婧
- 中国环境科学研究院, 国家环境保护饮用水水源地保护重点实验室, 北京 100012
- 雷刚
- 1. 中国环境科学研究院, 国家环境保护饮用水水源地保护重点实验室, 北京 100012;2. 华南农业大学工程学院, 广州 510642
- 韩梅
- 中国环境科学研究院, 国家环境保护饮用水水源地保护重点实验室, 北京 100012
- 李虹
- 中国环境科学研究院, 国家环境保护饮用水水源地保护重点实验室, 北京 100012
- 摘要：针对水华风险预警过程中相关监测指标数据缺失的问题,借鉴多元统计和随机分析构建了一种缺失数据插补方法,用于弥补现场调查数据的不足.基于主成分分析,对水华相关影响指标进行降维,确定水体水华风险预警模型的输入层变量.同时,采用多层感知器(MLP)人工神经网络模型对水华表征指标叶绿素a的浓度进行预测,并引入风险概率的概念,提出了水华风险概率计算公式,完善了水华预警的风险表达.最后以三峡库区典型支流大宁河为案例的研究证明了上述方法的可操作性.研究结果显示,插补数据条件下和未插补数据条件下的大宁河水华风险预警模型决定系数分别为0.9711和0.7769,前者的模型准确性更高,叶绿素a浓度预测效果更好;预测时段内大宁河11 d为水华蓝色预警(无警)级别,水华发生的风险概率为1.99%~18.61%;1 d达到水华橙色预警(中警)级别,水华发生概率为90.48%.
- Abstract：For solving the problem of missing or insufficient data of the key indicators in the early warning of algae bloom risk, we proposed a method for data supplement with missing data with the reference of multivariate statistical theory and stochastic analysis to compensate inadequate in-situ data collection. Based on principal component analysis, dimensionality reduction of the main causal indicators for algae bloom was carried out to finalize the input variables in early warning model. Using Multi-Layer Perceptron(MLP)neural network, the concentration of chlorophyll a, the indicator for algae bloom occurrence, were predicted. According to the concept of risk probability, calculation formula of algae bloom risk probability was put forward aiming to improve the expression of algae bloom risk warning. The case study in Daning River of Three Gorges Reservoir has proved the feasibility of the proposed method. The results showed that decision coefficient of the prediction model were 0.9711 and 0.7769 with and without data supplement respectively. The former one ensured more accurate modeling and better prediction of chlorophyll a concentration. In the prediction period, blue warning (no warning) level was shown at 11 days in Daning River, as well as the algae bloom risk probability was 1.99%~18.61%. The orange warning (moderate warning) level appeared at 1 day when the algae bloom risk probability was 90.48%.
摘要点击次数： 1496 全文下载次数： 3965
