Protocols and Techniques to Manage
Distribution System Resources
James Uber, CU
This project will develop real-time predictive analytics tools that leverage existing utility information resources to help manage distribution residence times, and disinfectant residual and DBP concentrations. In collaboration with the USEPA National Homeland Security Research Center, the project team has developed an object-oriented software library called Epanet-RTX (Epanet “Real-Time eXtension”), which comprises the core data access, cleanup, and modeling components of a real-time analytics system (Rossman, 2000; Hatchett et al., 2011). After several years of development, RTX was released as an open-source software project in September, 2012, to support further research and commercialization opportunities (see http://openwateranalytics.github.com/epanet-rtx/). RTX is currently being applied at full-scale with the Northern Kentucky Water District (NKWD) and the Greater Cincinnati Water Works (GCWW), to develop and test real-time hydraulic and water quality network models. The RTX software will be serve as the ideal framework for building real-time analytics applications for this project.
Description of EPANET-RTX. RTX is an object library for building real-time analytics and hydraulic modeling environments. It is a set of building blocks, which can be used and extended to create real-time data fusion applications. In essence, RTX provides interoperable access to several different technologies which are foundational to real-time modeling. These technologies involve accessing a supervisory control and data acquisition (SCADA) historian database, using time series transformation, filtering, and forecasting methods, and running hydraulic and water quality simulations.
RTX interfaces with existing technologies to form a scaffolding to enable the smooth migration of data from the measurement domain into the analysis and modeling domain. These form the three pillars of real-time modeling -- data acquisition, cleaning, and simulation. As part of the pilot studies, the research team has assembled several prototype applications, including a geographical user interface, and an interactive analytics configurator. The former provides a platform with which to display the results of real-time analytics in a mapping environment, while the latter provides convenient access to SCADA data and the basic RTX analytics objects. In particular, new analytics applications can be created by dragging and arranging RTX time series transformations on a virtual canvas.
Real-Time Analytics of Storage Tank Operations. Real-time analytics will be created and deployed to estimate and forecast tank residence time, chlorine residual, and degree of mixing. These analytics require only real-time water level data, available from SCADA, tank geometry, and a simple chlorine decay model (e.g., Boccelli et al, 2003). By bringing real-time information about tank residence time, mixing, and chlorine residual into the control room, operators will be naturally made aware of and challenged by important water quality information. The research approach essentially consists of three sequential steps: 1) build the analytics software applications; 2) deploy the application at 1-2 small utilities and provide operator training; and 3) measure the impacts of water quality awareness through analysis of operational changes and operator interviews.
Hydraulic mixing in tanks has been studied by the chemical process industries. Rossman and Grayman (1999) built on these results to derive and test operating rules that maintain a complete-mixed condition in finished water storage tanks. The upward movement of a turbulent inlet jet creates side circulation patterns that, given sufficient time, can mix the entire tank contents. Indeed, sufficient mixing time is the key to achieving a completely mixed condition, depending on the inlet diameter, flow rate, and water volume in storage. Rossman and Grayman showed that the following relationship is a good predictor of measured mixing times (R2 = 0.9):
where τm is the required fill time for mixing, V0 is the stored water volume at the beginning of the fill cycle, Q is the inlet flow rate, and u is the inlet water velocity. This expression is dimensionally correct and any consistent set of units may be used. The requirement for complete mixing becomes τf > τm where τf is the time of the filling period. This expression is convenient for practical use, since it requires only as-built drawings and a SCADA system that records water levels.
The turnover time, τt, is the amount of time required to replace the average volume of water stored. For example, if 1 million gallons is the average stored volume, and 250,000 gallons are drained and replaced each day, then τt = 1,000,000/250,000 = 4 days. For well-mixed tanks, the turnover time is similar to the residence time in the tank if the inlet water age is relatively small. We propose to use tank turnover time as a pragmatic, but approximate, real-time analysis that relates operating decisions to water quality evolution. The turnover time is easily calculated from SCADA tank water levels and tank geometry. To supplement this water quality metric, we propose to add a simple single-specie model of chlorine residual based on first or second order kinetics. The parameters required would include the rate constant and the tank inlet concentration. The tank inlet concentration is proposed to be estimated from point-of-entry (POE) concentration - if available from SCADA - and an estimate of the residence time from the plant to the tank. If the POE concentration is not available, then the tank inlet concentration will be a parameter that would be calibrated periodically, in order to match field monitoring data.
Real-Time Network Modeling for Operational Decision Support. This project will develop and deploy real-time network models to simulate distribution system hydraulics and chlorine residuals, in the control room of a small utility partner. These models link distribution system infrastructure models with real-time SCADA data on pump operations, tank levels, pump station flows, etc, to provide a real-time predictive capability that continually adapts to operational conditions. The thought is that operators will use these tools to routinely engage in situational response training, and to conduct operational analyses to achieve optimization goals related to pressure, leakage, energy, and water quality management – just as a pilot uses a flight simulator. In addition to the technical research issues, it is of broader interest to understand better the barriers to using such technology within a small utility environment.
This project builds on the experience gained through the NKWD and GCWW full-scale deployments, as well as the software tools developed for speeding the deployment of real-time models. Full-scale real-time network modeling presents technical challenges, and to date is unproven. We anticipate that the configuration and development of the real-time model at a small utility will present special challenges as well as opportunities for efficiency, in comparison with an advanced utility like GCWW. Certainly, the distribution system at the small utility will be far less complex both in terms of number of elements, pressure zones, and complexity of infrastructure. On the other hand, we anticipate information access challenges, and possibly also greater resistance to integration within their operational protocols.
The research steps involve the development of the real-time model for one small utility, deployment and training of operational personnel, and measurement of performance through analysis of operational data and personnel interviews. We would measure improvements to consumer water quality through SCADA data records and regulatory compliance monitoring, but also through the implementation of a limited field campaign to collect residual disinfectant and DBP distribution system samples, both before and after deployment. Of the above steps, the development of the real-time model presents the greatest technical hurdles, and will thus require the greatest effort. The main tasks involved in real-time model development include: 1) data gathering, 2) data quality analysis and correction, 3) development of appropriate SCADA data transformations, 4) association of (transformed) SCADA data with network elements, and 5) assessment and calibration of real-time model results, through comparison of simulation results and SCADA data for an extended historical period.