Gao, W.; Wan, L.; Qi, S., and Wang, D., 2019. The tracing of wastewater in enterprises based on hybrid neural network. In: Hoang, A.T. and Aqeel Ashraf, M. (eds.), Research, Monitoring, and Engineering of Coastal, Port, and Marine Systems. Journal of Coastal Research, Special Issue No. 97, pp. 1–9.
Over-standard discharge of enterprise sewage had severely weakened the processing wastewater ability of the wastewater and shortened the service life of the facility in the sewage treatment plant, more importantly, posed a serious threat to the water environment safety in the watershed. Based on the historical daily monitoring data of key monitoring enterprises and sewage treatment plant, temperature and precipitation data, the boosted regression tree (BRT) model was established. The feature variables (discharge flow, ammonia nitrogen (NH3-N) daily emission and COD daily monitoring concentration) were obtained by BRT model. The total contribution of feature variables accounted for 80.92%. The BRT model prediction accuracy and kappa coefficient was 0.88 and 0.78, respectively. Then, three single hidden-layer feedforward neural network models were constructed to analyze the importance ranking of enterprises according to each feature variable. The results showed that heavy industry and pharmaceutical enterprises ranked ahead. Last, using generalized extreme studentized deviate test (ESD) algorithm, the anomaly detection analysis was conducted to obtain the proportion of abnormal data and distribution regularities on the suspicious enterprises. Our study may provide scientific evidence and auxiliary decision support in the identification of illegal emission enterprises.
