Research On Water Quality Prediction And Early Warning Based On Rough Bayesian Network

At present,the water quality in China is poor.Moreover,the ecosystem of river basin is seriously damaged.The protection of water resources is closely related to the interests of residents and the development of economy and society.The “Action Plan for Prevention and Control of Water Pollution” and “Strictest Management System for Water Resources” issued by the State Council both indicate the firm attitude of government towards water environmental protection and management.Water quality is one of key factors restricting the development,utilization and protection of water resources.However,due to the lack of reasonable and effective assessment,prediction,and early warning methods,water resources management is still in the state of experience and semi-empirical pattern.Time series data has characteristics of uncertainty,high dimension,complexity,time-varying,and dynamic,so the traditional methods limit their application in the real system.To address this issue,a model based on rough Bayesian network(RS-BN)is proposed in this dissertation.The presented model is applied for water quality evaluation,prediction and early warning in the Jialing River Basin(three monitoring stations of Chongqing’s main urban area).Furthermore,the specific management policies were proposed based on the results.The main works of the dissertation are as follows:(1)Extract and analyze significant influence factors based on rough set theory.Rough set(RS)not only can effectively deal with uncertain,missing and dynamic data,but also can eliminate information redundancy and reduce modeling complexity.According to the features of time series and requirements in the water quality prediction and evaluation,the RS theory is used to attribute reduction of multiple water quality information(external and internal environment factors),thus achieving the main influence factors.(2)Establish rough Bayesian network(RS-BN)for water quality prediction.Bayesian network(BN)with the better abilities of probability expression and uncertainty analysis is one of the most effective models for uncertain knowledge representation and reasoning.On the basis of the attribute reduction using the RS approach,an RS-BN model is established via the structure modeling and parameter learning.The ability of forward / reverse reasoning of the RS-BN model was used to obtain important pollutants in the river basin(knowledge expression),and at the same time obtain prediction results superior to other comparative models.(3)Research on early warning strategy of water environment capacity based on the overall reach standard method.The overall reach standard method(ORS)with a simple computing framework without anthropogenic impact is a typical model for estimation of environment capacity.Hence an ORS method based on the prediction results using the RS-BN model is addressed and applied for the water environment capacity estimation and early warning.Combine the calculation results with the analysis of the social and economic status of the area where the corresponding section is located,some water resources management suggestions are recommended from the aspects of total emission reduction(industrial,living,and agricultural)and ecological maintenance.This dissertation constructs rough Bayesian network water quality prediction model and water environment capacity early warning model through analysis and fusion of attribute reduction,probability expression,and prediction modeling methods.To evaluate its performance,a case study of the Jialing River Basin(section of Chongqing’s main urban area)is investigated.This study has important theoretical and practical significance for effective intelligent management of assessment,classification,prediction and decision-making in water environment system.