Advances in Water Resources Systems Engineering: Applications of Machine Learning

Abstract

AbstractThere has long existed a dichotomy in the field of water resources systems engineering between simulation and optimization modeling, with each approach having its own advantages and disadvantages. Simulation models provide a means of accurately representing the complex physiochemical, socioeconomic, and legal-administrative behavior of complex water resources systems, but lack the capability of systematically determining optimal water planning and management decisions. Optimization models, on the other hand, excel at automatic determination of optima, while often sacrificing the accurate representation of the underlying water system behavior. Various means of effectively establishing a synergy between simulation and optimization models that accentuates their advantages while minimizing their shortcomings have evolved from the field of artificial intelligence within the province of computer science. Artificial intelligence was defined by John McCarthy in 1955 as “the science and engineering of making intelligent decisions.” Machine learning, as a branch of artificial intelligence, focuses on the development of specific algorithms that allow computerized agents to learn optimal behaviors through interaction with a real or simulated environment. Although there are many aspects of machine learning, the focus here is on agent-based modeling tools for learning optimal decisions and management rule structures for water resources systems under conflicting goals and complex stochastic environments. A wide variety of machine learning tools such as reinforcement learning, artificial neural networks, fuzzy rule-based systems, and evolutionary algorithms are applied herein to complex decision problems in integrated management of multipurpose river-reservoir systems, real-time control of combined sewer systems for pollution reduction, and integrated design and operation of stormwater control systems for sustaining and remediating coastal aquatic ecosystems damaged by intensified urbanization and development.Key WordsArtificial intelligenceCoastal environmentDetention reservoirsEcologyEstuariesFuzzy setsGenetic algorithmsHydraulic sewer modelsMarkov decision processesMultireservoir systemsNeural networksOptimal real-time controlReinforcement learningStochastic dynamic programmingUrban stormwater management.