A Physics-Based Artificial Intelligence General Framework for Optimal Control of Sewer Systems to Minimize Sewer Overflows

联合下水道系统将雨水径流、生活污水和工业废水收集在同一管道中。在正常情况下，联合下水道系统将收集的废水输送到工厂，在工厂进行处理并排放到包括河流、湖泊、河口和海洋在内的地表水系统。在暴雨或融雪期间，联合下水道系统输送的废水有时会超过处理厂的处理能力，导致溢流进入附近的溪流和地表水体。在美国，每年大约有8500亿加仑未经处理的下水道溢出物被排放。这些污水泛滥对环境和社区造成不利影响，包括污染饮用水水源和关闭娱乐海滩。这个项目的首要目标是开发和验证机器学习工具，以预测潜在的下水道溢出的位置和数量。为了推进这一目标，首席调查员建议整合数据科学(大数据算法)、实验室规模的实验和基于物理的数值模型，以加快机器学习模型的可用性，以指导和优化污水系统的运行，以管理和减少污水溢出对环境的影响。该项目的成功完成将通过开发基础知识和新的建模工具来支持管理和减少下水道溢流，从而造福社会。还将通过外展和教育活动为社会带来更多好处，包括指导佛罗里达国际大学的两名研究生和六名本科生。随着气候变化导致的暴雨和洪水等极端天气事件的频率和强度增加，下水道溢出将变得更加频繁和严重。为了管理下水道溢流，需要新一代超高速模型来预测它们可能发生的时间和地点，以及在暴雨发生之前将溢流降至最低所需的决策序列。该项目的总体目标是集成人工智能(AI)、大数据和基于物理的数值模型，以加快可用于预测、管理和减少下水道溢出的机器学习(ML)模型的可用性。这项研究的具体目标是：1)为目前不能模拟下水道溢出的现有开源下水道流动动力学模型实施和验证下水道溢出模型；2)开发一个基于物理学的通用人工智能开源框架，用于在给定的一组固定运行场景(例如，闸门位置固定)下预测组合下水道溢出的位置和流量；以及3)开发一个人工智能开源框架，用于确定控制门处决策变量/场景的最佳顺序，以最大限度地减少合并下水道溢出。除了上述具体目标外，首席调查员(PI)建议将他们建议的新ML模型与各种基于开放源代码的物理模型结合和集成，以建立一个新的建模框架，名为Impacto(用于预测污水溢流的集成建模和用于优化封闭管道和隧道中的闸门控制的分析，以将溢流降至最低)。最后，PIS建议通过训练Impacto来验证Impacto：(1)预测现有两个下水道系统中组合下水道溢流的位置和流量，以及(2)确定控制门处的最佳决策变量序列(例如，流量排放)，以最大限度地减少下水道溢流。该项目的成功完成具有潜在的变革性影响，通过开发和验证可用于预测、管理和减少混合下水道溢流发生的集成和开放源码模型。为了实施该项目的教育和外展活动，PIS计划与佛罗里达国际大学(FIU)工程师在车轮上计划合作，为代表不足的群体的中学开发一个关于下水道溢出的实践活动的教育模块。此外，PIs建议利用FIU Louis Stokes少数群体参与联盟计划从代表不足的群体中招募六名本科生参与该项目。这一奖励反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

A combined sewer system collects rainwater runoff, domestic sewage, and industrial wastewater in the same pipe. Under normal conditions, a combined sewer system transports the collected wastewater to a plant where it is treated and discharged to surface water systems including rivers, lakes, estuaries, and oceans. During heavy rainfall events or snowmelts, the volume of wastewater transported by a combined sewer system can sometimes exceed the treatment plant's capacity resulting in overflows into nearby streams and surface water bodies. Approximately 850 billion gallons of untreated combined sewer overflows are discharged every year in the United States. These sewer overflows have an adverse impact on the environment and communities including the contamination of drinking water sources and the closures of recreational beaches. The overarching goal of this project is to develop and validate machine learning tools to forecast the location and volume of potential sewer overflows. To advance this goal, the Principal Investigators (PIs) propose to integrate data science (big data algorithms), lab-scale experiments, and physics-based numerical models to accelerate the availability of machine learning models to guide and optimize the operation of sewer systems with the aim of managing and reducing the environmental impact of sewer overflows. The successful completion of this project will benefit society through the development of fundamental knowledge and new modeling tools to support the management and reduction of sewer overflows. Additional benefits to society will be achieved through outreach and educational activities including the mentoring of two graduate students and six undergraduate students at Florida International University.As the frequency and intensity of extreme weather events such as heavy rainfalls and flooding increase due to climate change, sewer overflows will become more frequent and severe. To manage sewer overflows, a new generation of ultrafast models are needed to predict when and where they are likely to occur, and the sequence of decisions needed to minimize overflows before heavy rainfall occurs. The overarching goal of this project is to integrate artificial intelligence (AI), big data, and physics-based numerical models to accelerate the availability of machine learning (ML) models that could be used to predict, manage, and reduce sewer overflows. The specific objectives of the research are to: 1) Implement and validate a sewer overflow model for an existing open-source sewer flow dynamics model that currently cannot simulate sewer overflows; 2) Develop a general physics-based AI open-source framework for predicting the location and volume of combined sewer overflows for a given fixed set of operational scenarios (e.g., gate positions are fixed); and 3) Develop an AI open-source framework for determining an optimal sequence of decision variables/scenarios at control gates for minimizing combined sewer overflows. In addition to the specific objectives listed above, the Principal Investigators (PI) propose to combine and integrate their proposed new ML model with various open-source physical-based models to build a new modeling framework named IMPACTO (Integrated Modeling for Prediction of sewer overflows and Analytics for optimal control of gates in Closed-conduits and Tunnels to minimize Overflows). Finally, the PIs propose to validate IMPACTO by training it to (1) predict the location and volume of combined sewer overflows in two existing sewer systems and (2) determine the optimal sequence of decision variables at control gates (e.g., flow discharges) to minimize sewer overflows. The successful completion of this project has the potential for transformative impact through the development and validation of an integrated, and open-source model that could be used to predict, manage, and reduce the occurrence of combined sewer overflows. To implement the education and outreach activities of the project, the PIs plan to develop an educational module with hands-on activities on sewer overflows for middle schools from underrepresented groups in collaboration with the Florida International University (FIU) Engineers on Wheel program. In addition, the PIs propose to leverage the FIU Louis Stokes Alliances for Minority Participation program to recruit six undergraduate students from underrepresented groups to work on the project.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.