Study on auxiliary operation control of machine learning in multiobjective complex drainage system

Abstract

Abstract Recently, urban waterlogging prevention and treatment of black–odorous rivers have become a social concern and the upgradation of drainage system and the development of river runoff pollution control projects have accelerated. The use of deep tunnels to upgrade old drainage systems and achieve pollution control-related engineering designs has complicated the drainage system operation control. The traditional operation control mainly relies on human experience or model simulation. This study provides a perspective of machine learning for controlling the operation of the drainage system and exploring whether the operation suggestions regarding facilities in this system can be given in real time while relying only on real-time data and avoiding the complex model simulation process. Herein, five drainage systems were used as examples: the initial water level of a pipeline, key point water level flow, pump station front pool water level, and most unfavorable point water level were selected as relevant variables and four machine-learning discrimination methods were used for to analyze the weir-lowering operation of a deep tunnel. This study found that the average error rate of the linear discrimination method was <10%, thereby exhibiting satisfactory performance. This study provides insights for improving the operation of complex drainage systems.