| Summary: | 碩士 === 國立屏東科技大學 === 環境工程與科學系所 === 98 === Kao-Ping and Tung-Kang rivers are two major rivers that provide water resources for livelihood, industrial and agricultural uses in the Kaohsiung-Pingtung area. Kao-Ping River is the major source of drinking water for 2.7 million people in the Kaohsiung metropolitan area. While the Tung-Kang River is the major source of industrial water because it is seriously polluted by nitrogen and phosphorus. This study utilizes factor analysis to search for principal factors among the water quality data collected from Kao-Ping and Tung-Kang rivers, and then applies cluster analysis to classify the sampling stations according to their pollution levels. Furthermore, the analysis techniques of the auto regressive integrated moving average (ARIMA) model of time series and back-propagation neural network were applied to predict the variation tendency of water quality.
The results from correlation analysis showed parameters including biochemical oxygen demand (BOD5), chemical oxygen demand (COD), total organic carbon (TOC), ammonia nitrogen (NH3-N), total phosphorus (TP), total nitrogen (TN), and total Kjeldahl nitrogen (TKN) had significant correlation for Kao-Ping River. While the parameters including suspended solid (SS), turbidity (TB), chloride (Cl-), electrical conductivity (EC), NH3-N and TKN showed significant correlation for Tung-Kang River.
Congruence index (Kaiser-Meyer-Oklin, KMO) was utilized to evaluate the water quality data collected from Kao-Ping and Tung-Kang rivers whether the data were feasible for factor analysis. For Kao-Ping River, three principal factors were obtained including eutrophication factor, nitrite factor and environmental factor with principal component analysis (PCA). All three factors can interpret 74.78 % of total variances. While for Tung-Kang River, three principal factors were also obtained including eutrophication factor, turbidity factor and inorganic factor. All three factors can interpret 67.61 % of total variances. In addition, three clusters were classified according to the three principal factors obtained for Kao-Ping River. Three clusters were also classified based on the three principal factors determined for Tung-Kang River. The results showed the severe pollution area of Kao-Ping River located on the upstream and downstream sampling stations in Pingtung city. The principal pollution factor was eutrophication factor. It showed the pollution source may be due to the improper discharges from the livestock wastewater, municipal wastewater and industrial wastewater. While for Tung-Kang River, the severe pollution area situated in the downstream sampling stations. The principal pollution factor was eutrophication factor, too. The pollution source may be from the improper discharge of livestock wastewater.
The results from analysis of ARIMA model showed moderate correlations were observed between predicted data and measured data for DO, BOD and NH3-N. But negative correlation for SS was observed. It might be due to the effects of intense precipitation in the spring of 2009 that caused the variation tendency of SS differing from that of the previous years. Single parameter model was not able to readily answer the effects of environmental changes. The results from back-propagation neural network analysis showed significant correlations were observed for training and learning stages. Therefore multiple parameters model can obtain better learning results. The results from ARIMA model analysis showed significant correlations were observed for training and verification, but moderate even negative correlations for testing. The results from back-propagation neural network model analysis showed significant correlations for training, verification, and testing stages.
Application of multivariate statistical analysis is able to extract and obtain fewer factors from a group of variables. The major parameters can be determined from those factors. The water quality in the future can be simulated and predicted by the data collected over the years to provide a reference to authority for the evaluation of water quality. The modeling results are evaluated to draft the corresponding strategies to reduce the monitoring cost and to enhance the cost-effective benefits. In the future, application of ARIMA model and back propagation neural network are proposed to drop the outliners first to enhance the correlations of model training and verification.
|
|---|
