A Study Of Water Quality Prediction Based On Probability And Combination Method

The safety of drinking water which has been threaten seriously by human over-exploitation of the water resources and environment is closely related to people's livelihood. Water quality early-warning system can effectively reduce and control the harm caused by the deterioration of water quality through the real time water quality monitor, assessment and early-warning. Prompt and effective water quality prediction can provide a reliable basis for water quality assessment and early-warning.In the current study of water quality prediction, researchers prefer single prediction methods for water quality prediction, and the existing combination methods for water quality prediction lack of a framework approach. The probability prediction is always based on an assumption that water quality index follows certain probability distribution. Base on probability and combination, a method for water quality prediction is proposed in this thesis. A water quality early-warning system for typical sustained water pollution is built with the demands of the practice of a certain water quality early-warning project.The main contents and innovative points are summarized as follows:(1)A water quality prediction method based on probability and combination is proposed. The method combines the prediction results of different single methods through Odds-Matrix method and it can improve the performances of prediction effectively. It is worth noting that the combination-forecast approach can be extended to new methods. The probability of prediction is established through statistical analysis of historical prediction data and hence the validation of the method is achieved along with interval estimation under certain confidence level.(2)A water quality prediction experiment aimed to verify the effectiveness of the method based on probability and combination is carried out. Two groups of water quality prediction methods are developed for a monitoring section of Qiantang River to conduct this work. One of the groups is the methods based on Grey Model theory. Support Vector Regression and the combination of them and the other one is the methods based on Grey Model theory, BP neural network and the combination of them. The analysis of the effects of these prediction methods and the experiment of probability prediction is also developed. The experimental results indicate that the combination-forecast approach performs better than single prediction methods. The validity of probability establishment can be checked effectively. According to the results, the interval prediction under certain confidence level can be given.(3) A water quality early-warning system for typical sustained water pollution is built. The system integration is carried out with the combination of the modules of the water quality prediction method based on probability and combination, water quality safety assessment and early-warning information issue. And the early-warning system for typical sustained water pollution which can achieve a complete water quality early-warning process from the monitoring data to water quality prediction, safety assessment and information issue is established. With the practice of the water quality early-warning project, the system has been well applied in three demonstration areas.