Water Journal : Water Journal December 2012
intelligent water networks refereed paper technical features 104 DECEMBER 2012 water proof-of-concept TaKaDu Trial Set-up In late 2010, Yarra Valley Water was approached by TaKaDu to conduct a trial of their system. Yarra Valley Water commenced an initial proof-of-concept trial for six months on a third of its network in March 2011. The purpose of this trial was to assess the feasibility and benefits of an IWN type solution in water operations with particular reference to NRW. For the limited trial, 40 distribution zones out of the total 138 were selected. An emphasis was given to zones with the following characteristics: • The more leak-prone pipe materials, such as 1960s cast iron mains; • Varying topography with strong pressure variation; • Clusters of zones in areas close to reservoirs (zones around Greensborough reservoir, zones around Mitcham reservoir); • A mixture of both metropolitan and regional zones. This diversity in distribution zone selection was to ensure that the results from the trial would encompass a representative sample of the Yarra Valley Water’s network. As a result of the Zone Metering program (commenced in 2001), Yarra Valley Water had implemented metering in these 138 distribution zones across its network. Prior to the TaKaDu trial, the primary method of identifying non-revenue water consisted of Infrastructure Leakage Index (ILI) and night-flow monitoring and analysis of the distribution zones using Access and Excel reports based on SCADA. A key test was to investigate the benefit performance of the TaKaDu leakage detection in comparison to prior methods. The following information exchanges and transfers were set up with TaKaDu for the trial (as shown in Table 1). Due to the data extraction limitations of the previous SCADA system at the time of the trial, six-hourly intervals of data for approximately 400 points were extracted and transferred to TaKaDu’s cloud servers. The turnaround time between starting FTP transfer and having the information and alerts visible in the interface was approximately 10 minutes, as shown in Figure 5. This enables almost real-time feedback on the performance of the network. The inbox style of the software enabled the reviewing user to load and view any changes in the system where it was behaving not as expected. This deviation from ‘expected’ values helped operators make the judgement on what appropriate action should be taken. The user’s information and follow-up response also tunes the system and provides feedback to the models. After tuning, the system learns other expected normal patterns and understands what information is useful to alert on. In the day-to-day operation, event alerts were issued for unexpected flow changes classified as bursts, leaks, abnormal pressure changes and metering errors. These were followed up and actioned by the NRW team, as would have occurred during previous processes. Key Trial Results and findings The proof-of-concept trial between March and August 2011 aimed to provide detailed information about the feasibility of IWN systems. As discussed earlier, TaKaDu claims a wide variety of benefits and Yarra Valley Water sought to confirm these, specifically in five major areas: • Ability of models to accurately interpret and classify system behaviour; • Accuracy of geolocation classifications; • Ability to detect meter issues and failures; • Burst prevention by detecting leakage earlier; • Efficiency savings. Over the six-month period of the trial, there were two distinct stages: the initial three months of the trial with its setup, training and tuning of the system, and the second three months where the system was running in close to an optimal state. Ability of TaKaDu to accurately interpret and classify system behaviour The key concern of Yarra Valley Water in this trial was the reliability of the classifications performed by TaKaDu. The concern from operators and the business was that it would be another source of annoying ‘false alerts’ that actually decrease efficiency by requiring additional work to respond to. The numbers and classifications of the events were recorded, and these were then correlated to the known works, issues and problems in the network. During the initial three months of the trial, 417 events were created with the breakdown of the classifications shown in Figure 6. In the early stages of the trial, flow increase was used as a generic notifier for burst, leakage and small increases. As the system was tuned and calibrated, further classifications as Burst, Leak and Flow increase were used with greater accuracy. The key message from Figure 7 is that only 10% of the events were deemed as non-useful or were not directly attributable to confirmed events in the system. The majority of these were in the initial tuning and set-up stage, and were reduced over time as the integrated learning algorithms improved. In particular, if we observe the second stage, the period from June 2011 to September 2011, we can see a substantial improvement in classification ability, as shown in Figures 8 and 9. The notification and classification turnaround time for events took, on average, two hours. This was in line Table 1. Information sources for TaKaDu proof-of-concept trial. Information Purpose GIS of Yarra Valley Water’s network For building model of connections and system for analysis 18 months of historical SCADA data To build a basis for the systems models 18 months of works history To assist in analysis of the network and tune the models Flow and Pressure (~ 400 SCADA data points) For monitoring and alerting. Six-hourly extract of continuous information Figure 5. Flow of information from field sensors to the software interface.
Water Journal February 2013
Water Journal November 2012-1