Study On Data-driven Based Early Warning And Emergency Management Of Sudden Water Pollution

With the development of China’s economy and the acceleration of the industrialization,China is confronted with increasingly serious water security problem.Water pollution incidents caused by chemical spill or deliberate behavior are one of the most important threats to China’s water security.Water pollution incidents happened in recent years not only endanger people’s life and property safety,but also have serious social impacts and cost considerable money during the subsequent ecological restoration procedure.Therefore,the analysis of automatic monitoring time series of water parameters for detecting anomalous water quality and early warning based on risk assessment after a sudden water pollution incident are of great significance.With this background,the paper mainly studies river pollution early warning based on anomaly detection and risk assessment.For early warning based on anomaly detection,an abnormal detection method for surface water quality based on autoregressive model(AR)and isolation forest algorithm(i Forest)is proposed.The time series of water quality parameters can be dynamically predicted with high accuracy.Then the isolation forest algorithm is used to detect the abnormal water quality by comparing to the threshold.The time series of turbidity,specific conductance and dissolved oxygen from a monitoring station in the Potomac River basin of the United States are selected as a case analysis.The receiver operating characteristic curve(ROC)is used to evaluate the performance of proposed method and good performance for abnormal surface water quality detection is achieved.For risk-based early warning,a risk warning approach based on risk assessment index system and fuzzy comprehensive evaluation is proposed.The risk index system after the occurrence of sudden water pollution was built with three aspects considered: the characteristics of water pollution incidents,the impact on the recipients,and the emergency response.The quantification of indicators involves Monte Carlo simulations of one-dimensional water quality model and experts grading method.The weights of indicators are obtained through the constrained fuzzy analytic hierarchy process(FAHP)and experts’ judgment of importance matrixes between the indicators.The fuzzy comprehensive evaluation method is used to assess the risk level in the last step,with which risk-based early warning of sudden water pollution incidents is realized.This approach is used to analyze the Songhua River water pollution incident in 2005 as case analysis,the results show that the approach can provide risk-based early warning after sudden water pollution.Based on the two chapters above and the three tier architecture for software development,the overall design of the platform for sudden water pollution early warning was carried out.Two functions are integrated in the platform,namely water quality time series anomaly detection and risk-based early warning for sudden water pollution.At the same time,the key stakeholders in the water pollution emergency management were identified.In the design of the early warning system,the people-centered design concept advocated by the UNISDR was emphasized.Finally,suggestions were made for the key stakeholders from the perspective of 4 key elements in early warning system.