Applications Of Support Vector Machine In Lake Eutrophication Assessment And Water Quality Prediction

Lake eutrophication has been becoming one of the key matters in terms of water pollution. Lake Eutrophication assessment and water quality prediction are two important approaches to understand and research lake water quality. The results can be used to represent lake water quality and pollution status as well as prediction of future water quality. They are essential for protecting and restoring lake water environment. There have been many mathematical models that are used to assesse and predict water quality, however, there is few generalized acceptable model. Different department or individual always chooses different model(s). Those models are widely used with some defects.Support vector machine(SVM), a new method developed in resent years, is a advanced research field in complex nonlinear science and artificial intelligence. The method has been widely used in many research field for the excellent capability on classification and regression. This thesis try to introduce it into water environment quality assessment, especially into lake eutrophication assessment and water quality prediction based on existing research results. Meanwhile, the effect of parameters to the model's precision was evaluated detailedly.The results show that the SVM method may be used in lake eutrophication assessment and water quality prediction. The precision of simulation results by SVM method is obviously equivalent with or better than that by other normal methods. The main research contents of the thesis are:(1) Summarize the current models which used in lake eutrophication assessment and water quality prediction systematically. Review the development research status briefly. The strongpoints of SVM method are summed up.(2) Introduce the targets, models and the three main problems of machine learning. Introduce the complexity, generalization and principle of experience Risk Minimize. Expounde the basal concepts, theory and research contents of Statistical Learning Theory.(3) Systematically introduce the principle of SVM, SVR and their standard methods and transformative methods. Analyze the strongpoints and disadvantages of those methods. Discussed the prediction trust degree and summarized the characteristic of SVR. (4) Took Wuliangsuhai Lake for example, Introduced SVM method into lake eutrophication assessment and water quality prediction using Weka software. Compare the SVM methods with current methods and summe up the strongpoints.(5) Discusse the effects of parameters to the model's precision. Compare the fit precisions betweenε-SVR and v-SVR. Predict the pH of Wuliansuhai Lake and compare with linear regression, BP neural network and RBF network. The results support the conclusion that SVM methods better than the other methods.(6) Analyze the reasons what caused the eutrophication of Wuliangsuhai Lake in Frozen period.The innovation points of this thesis concluding:(1) Introduced SVM into the resrarch field of lake eutrophication assessment. Extended the application ranges of SVM method.(2) Introduced SVR into the resrarch field of water quality prediction based on Time Series Analysis. The prediction precision has farther improvement.(3) Assessed the nutrition status of Wuliangsuhai Lake in the Frozen period and Analyzed the reasons what caused the eutrophication.(4) Many researchers concentrate on searching the optimal parameters for models however the effects of parameters'changes on models'precision is normally ignored. This thesis analyzed the model precision with respect to the changes of parameters and provided some valuable results.