Leveraging Big Data to develop an expert system for the optimal operation of smart water networks

全球老化的饮用水系统（DWS）面临着越来越大的压力，以减少无收益水（NRW，估计超过140亿美元/年），最大限度地降低能源成本，并减少温室气体排放（GHG）。然而，美国水务协会在2019年估计，到2039年，更换这种老化基础设施的资本成本将超过4720亿美元，这是成本高昂的-引发了对创新解决方案的需求来解决这些重大问题。由于无线传感器的最新发展，许多DWS现在正在收集关键变量的高频流（例如，在不同的空间和时间尺度上的地块级消费者用水需求），从而产生大数据并促使传统DWS向智能水网（SWAN）的转换，从而能够缓解次优DWS操作（例如，最大限度地减少能源消耗）。SWAN依赖于基于数据驱动模型的专家系统（例如，机器学习）来识别满足DWS管理者的目标的最佳操作决策，DWS管理者在面对许多不确定性（例如，改变水的需求）。然而，SWAN仍处于起步阶段，没有专家系统可以处理大数据并以计算高效的方式解释DWS中的不确定性。我的长期愿景是通过使用天鹅来优化DWS。在接下来的5年里，我的研究计划将集中在实现专家系统来解决这些重要挑战的最迫切需求上。该专家系统将通过三个短期目标依次构建：1）将探索最先进的数据驱动模型，用于预处理和从与SWAN相关的大数据（例如，地块级消费者水需求）; 2）这些模型将被合并到新的随机框架中以考虑几个重要的不确定性源（例如，模型结构）和时间相关性，以提高专家系统的可靠性;以及3）前两个阶段将与强化学习相结合，以根据关键运营目标（最大限度地减少NRW，能源成本和/或GHG）优化SWAN（重点是泵计划优化）。为了证明它们的优越性，每个目标的新发展将与水务公司采用的当前最先进的方法和基准方法进行严格比较。拟议的研究将推进关于优化SWAN的新知识，并为水务公司提供一个新的专家系统，以应对紧迫的挑战，并有可能为水务公司每年节省46亿美元的运营成本。通过我的计划，8名学生研究人员将获得必要的技能，使水公用事业的重大影响（例如，渥太华市）和私人公司投资于天鹅（例如，Innovyze），为水资源的可持续管理做出贡献，并将加拿大置于SWAN研究的前沿。

Aging drinking water systems (DWS) worldwide are under increasing pressure to reduce non-revenue water (NRW, estimated at over $14B USD/year), minimize energy costs, and cut greenhouse gas emissions (GHGs). However, the capital costs to replace this aging infrastructure, estimated by the American Water Works Association in 2019 to over $472B USD in the USA by 2039, is cost-prohibitive - triggering a need for innovative solutions to address these significant issues. Due to the latest developments in wireless sensors, many DWS are now collecting high-frequency streams of key variables (e.g., parcel level consumer water demand) at varying spatial and temporal scales, resulting in Big Data and prompting the conversion of traditional DWS to smart water networks (SWANs) that enable the mitigation of sub-optimal DWS operations (e.g., by minimizing energy usage). SWANs rely on expert systems based on data-driven models (e.g., machine learning) to identify optimal operational decisions that meet the goals of DWS managers, who struggle to sustainably operate DWS in the face of numerous uncertainties (e.g., changing water demand). However, SWANs are still in their infancy and there are no expert systems that can handle Big Data and account for uncertainty in DWS in a computationally efficient manner. My long-term vision is to optimize DWS through the use of SWANs. The next 5 years of my research program will focus on the most immediate needs to realize an expert system to address these important challenges. The expert system will be built sequentially through three short-term objectives: 1) novel state-of-the-art data-driven models will be explored for pre-processing and making accurate forecasts from Big Data associated with SWANs (e.g., parcel level consumer water demands); 2) these models will be incorporated in a novel stochastic framework to account for several important uncertainty sources (e.g., model structure) and temporal correlations to improve the reliability of the expert system; and 3) the previous two stages will be coupled with reinforcement learning for optimizing SWANs (with a focus on pump schedule optimization) according to key operational goals (minimizing NRW, energy costs, and/or GHGs). To demonstrate their superiority, the novel developments from each objective will be rigorously compared against current state-of-the-art methods and benchmark approaches adopted by water utilities. The proposed research will advance new knowledge on optimizing SWANs and provide a novel expert system for water utilities to address pressing challenges, with the potential to save utilities $4.6B USD/year in operation costs. Through my program, 8 student researchers will gain the skills necessary to make significant impacts at water utilities (e.g., City of Ottawa) and private firms invested in SWANs (e.g., Innovyze), contributing to the sustainable management of water resources and placing Canada at the forefront of research in SWANs.