iWIN

Intelligent Water Infrastructure Network

EXPERT PANEL DISCUSSION-DEMYSTIFYING ARTIFICIAL INTELLIGENCE CONCEPTS IN WATER

RESOURCES

RECORDING

_Expert_Panel_Discussion_-_Demystifying_Artificial_Intelligence_Concepts_in_Water_(Basic_Small_-_WEB_MBL_(H264_400)).mp4

COMMENTS

  • It is important to understand the difference between ordinary computer modeling versus AI and machine learning.

  • Gravity flow wastewater metering systems can induce problems at high flows, especially with systems that have high I&I challenges. Utilities spend a lot of staff time and effort reviewing / cleaning data.

  • Communicating benefits to employees on how the application of AI or other technologies can enhance their work and allow them to do higher value activities. Also, it is important to make sure everyone is included in the work and understand they are partners and have a say.

  • It is important to identify "low-hanging" AI technologies that utilities can start with low investment.

  • The input data quality for AI or machine learning models are very important. One of the function of AI or machine learning is to distinguish good data from bad data. How to achieve that is a question.

  • Flow monitors are notoriously inaccurate, especially under Average Dry Weather Flow (ADWF) conditions. They do provide a good relative reference for the diurnal and Wet Weather Flow (WWF) response patterns. How do we train an AI to understand the difference between relative and absolute information?

  • How much data is required to implement AI. Is it 10years or 20 years worth of data?

  • Sensor health can be a major concern of contaminant warning system development, and needs to be part of data collection efforts for other purposes as well.

  • Non-hydraulic data could include manufacturer, material of construction, site-specific conditions (soils, traffic), date of installation, etc.

  • A Radial Basis Function ANN will classify/categorize determining variables. I've had some success building algorithms using outputs from those runs, and developing inputs for predictive networks.

  • Often we fall into thinking about only data that we (as a utility or organization) create or measure. There is so much available data that exists in public domain that (when understood and used appropriately) can reap great value and benefit when interacting with internal data streams.

  • Water Quality ( pH, Turbidity, Chlorine, Temperature, Hardness, Alkalinity, THMs, HAAs) data is very important in day to day operations for water utilities industry.