Eutrophication Prediction Based On ANN And SVM For Xiangxi Bay Of The Three Gorges Reservoir

Since the Three Gorges Reservoir impoundment, a large area of eutrophication has occurred in the reach of back water of some tributaries because the hydrological conditions have changed obviously. The prominent feature was the increaseing chlorophyll a concentration caused by algae bloom. Therefore, simulation and prediction of chlorophyll a concentration can directly reflect the trend of eutrophication. Base on analyzing the main factors of eutrophication for Xiangxi Bay, the prediction model was established to study the short-term forecasting and warning of the eutrophication status with artificial neural network and support vector machine. The main works and results of this dissertation are as follows:(1) Relying on the ecological and environmental field station at Xiangxi Bay of the Three Gorges University, water environmental data from 2008 to 2010 was collected and collated. The time series variation of water environment-related factors and the eutrophication status was analyzed and discussed for Xiangxi Bay of the Three Gorges Reservoir.(2) Based on the principal component analysis, cluster analysis and gray relational analysis, screened out six categories eutrophication parameters, Zeu/Zmix, TP, WT, TN/TP, D-Si and SD, which have a close connection with Chla from 18 environmental parameters in field monitoring.(3) Based on the water environment monitoring data from 2008 to 2010 in Xiangxi Bay, GA-BP model was established to forecast eutrophication. The Zeu/Zmix, TP, WT, TN/TP, D-Si, SD and Chla in previous week are used to predict the chlorophyll a concentration for next week. The results showed that the model can basically reflect the trend of chlorophyll a concentration, but can not provide a relatively accurate predictions of the values.(4) Based on the same data, GA-SVM model was established to forecast eutrophication and can predict the trend of chlorophyll a concentration. The predicted and measured values showed a significant linear correlation with the linear slope is 0.86, correlation coefficient is 0.97974. When eutrophication occurred(the chlorophyll a concentration is greater than 10 mg/m3), the relative error of model predicts is little, the minimum relative error is 6.49%, the maximum relative error is 27.94%, the average relative error is 14.75%. When eutrophication didn't occur(the chlorophyll a concentration is less than 10 mg/m3), despite the large relative error predicted by the model in the condition of not eutrophication outbreak, but the prediction results can still reflect the fact with no occurrence of eutrophication in Xiangxi bay.(5) The fitting and prediction of the two prediction methods were compared. The fitting ability and predictive ability of GA-SVM model is better than GA-BP model, the same time, GA-SVM model has better robustness. The GA-SVM model can be used to the short-term forecasts of chlorophyll a concentration in Xiangxi bay based on its good prediction effect.(6) This paper puts forward a grade division scheme of eutrophication warning on the basis of the eutrophication standard in the Three Gorges Reservoir which was proposed by Zheng Binghui and others. Then, early eutrophication warning was carried out and obtained good prediction effect by using the predicted values of chlorophyll a concentration. In the 10 early warning, the results of 8 times is true and accurate without only 2 times lower than the actual level.