遗传-BP神经网络法预测叶绿素a浓度变化

Prediction of Chlorophyll-a by Genetic BP Neural Network

基于2009年—2010年对临江河回水区水质指标的监测数据,采用遗传算法结合BP神经网络的方法对回水区的叶绿素a(Chl-a)浓度变化进行动态模拟预测。通过灰色关联法确定了对Chl-a浓度有显著影响的指标与网络输入变量,即水温、DO、流速、透明度(SD)、TP、CODMn及Chl-a。模拟结果表明,遗传-BP神经网络的预测值和实测值吻合较好,其相对误差约为9.8%,模型可良好地用于次级河流回水区叶绿素a浓度的短期预测。预测结果表明,在春末夏初季节,当水库蓄水位为150～160 m时,临江河回水区富营养化潜势较高,尤其应注重临江河该时段富营养化的防控工作。

Based on the monitoring data of water quality in the backwater area of Linjiang River from 2009 to 2010, the dynamic variation of chlorophyll-a concentration in the backwater area was simulated and predicted by genetic algorithm with BP neural network. The input variables of the network which have significant effects on chlorophyll-a concentrations were determined by grey relational analysis, and they were water temperature, dissolved oxygen （DO）, flow rate, transparency （SD）, total phosphorus （TP）, CODMn and chlorophyll-a. The simulation results indicate that the predictive values of genetic BP neural network and measured values fit well with the relative error of 9.8%. The prediction model can be well used for short-term prediction of chlorophyll-a concentration in backwater area of tribu- tary. The prediction results show that the backwater area of Linjiang River has high eutrophication potential when the reservoir water level is in the range of 150 to 160 m during the late spring and early summer. Particular emphasis on prevention and control of eutrophication of Linjiang River should be given during this period.