基于BP神经网络的海河干流叶绿素浓度短时预测研究

BP neural network-based short-term prediction of chlorophyll concentration inmainstreamof Haihe River

为实时预测海河干流水体藻华的暴发时段及影响程度,提高环境管理部门决策能力,以海河干流段典型断面的水质在线监测及气象站高频、实时数据为基础,基于BP神经网络,以实时叶绿素浓度、气温、光照强度和气压四项指标为输入变量,建立了叶绿素浓度日变化量的预测模型,对海河干流大光明桥处水域叶绿素浓度随时间的变化进行预测。结果表明:对海河干流叶绿素浓度短时预测影响较大的因素依次为溶解氧(叶绿素)、气温、光照强度、气压、降雨、电导率、相对湿度;预测时长越短,预测精度越高。当预测时长分别为24 h、12 h、6 h时,Nash效率系数分别为0.77、0.85、0.93,预报误差的标准误差分别为5.7μg/L、4.6μg/L、3.1μg/L;12 h内的预测精度可满足海河河道藻华预警的实际需求,为其短期预警提供了数据支撑。

In order to timely predict the outbreak time and effect of algae bloom in the mainstream of Haihe River and enhance the decision-making capacity of the relevant environment management department,a model for predicting the daily variation of the chlorophyll concentration is established based on BP neural network by taking the indexes of the real-time chlorophyll concentration,temperature,light intensity and atmospheric pressure as the input variables on the basis of the on-line water quality monitoring data for the typical cross-section of the mainstream of Haihe River and the high frequent and real-time data from the relevant meteorological stations,and then the variation of the chlorophyll concentration depending on time at Daguangming Bridge is predicted. The result shows that the factors having the larger impacts on the short-term prediction of the chlorophyll concentration are dissolved oxygen（ chlorophyll）,temperature,light intensity,atmospheric pressure,rainfall,conductivity and relative humidity.The shorter the prediction time interval is,the higher the prediction accuracy is to be. When the prediction time intervals are 24 h and 12 h,6 h respectively,the Nashefficiencycoefficients are 0. 77,0. 85 and 0. 93 with the prediction errors of 5. 7 μg/L,4. 6μg/L and 3. 1 μg/L respecively,thus the prediction accuracy within the time interval of 12 h can meet the actual demand from the pre-warning of the algae bloom in the channel of Haihe River and then provide the relevant data support for the short-term prewarning of algae bloom therein as well.