1. Josef Bicik, Dragan A. Savić & Zoran Kapelan
Operation of Water Distribution Systems Using Risk-based Decision Making
Josef Bicik, Dragan A. Savić & Zoran Kapelan
Centre for Water Systems, University of Exeter, Exeter, UK

2. Outline Motivation WDS Failures Risk-based decision making Case study
Future work
Summary 2

3. Motivation Support operator’s decision making
WDS operation under abnormal conditions
Help prioritise actions of the operators
Reduce impact on customers
Meet regulatory requirements
EPSRC Neptune project
Many decision support systems / expert systems for operation under normal conditions (e.g., pump schedulling)

4. WDS Failures Exhibit abnormal flow/pressure patterns
Focus on pipe bursts
Exact cause typically unknown
Operational risk assessment
Failure risk is dynamic
True cause unknown until confirmed by customers/field technician
Proactive approach – try to deal with failures before they are detected by customers – i.e., rely on pressure and flow sensors in the network
Subjective impact – the impact is perceived differently by different people in different situations
Evaluation of intervention strategies – the impact assessment is not only used to evaluate the adverse effects of a burst, however as well to compare the possible interventions used to mitigate the impact
The time of the intervention can be decided depending on the severity of the burst and time of detection. E.g. if a burst is detected at 10PM and the burst does not affect any customers, it can be repaired in the morning thus reducing the cost of the repair however, this has more to do with the interventions and the impact model can only support such decision.

5. Risk Assessment Internal Alarm List Alarm 1 Alarm 2 … Alarm M … …
Potential
Incident 1
Internal
Alarm List
Impact 1
Likelihood
Alarm 1
Potential
Incident 2
The risk of each incident is defined as a triplet comprising (a) a risk scenario (describing our understanding of the current and future state of the system – to be defined below), (b) the incident’s probability of occurrence and (c) its impact over a specified period of time (typically 24 hours), under a particular risk scenario across a series of impact indicators.
In this work, the “risk scenario” associated with a potential incident is defined as the ensemble of: (1) the known initial, i.e. current network conditions (pressures/flows, tank levels, statuses of automatically regulating devices, etc.) and (2) the assumed future network conditions (e.g. forecasted nodal demands and assumed statuses of manually controlled devices) over some risk analysis horizon (e.g. next 24hr hours).
Impact 2
Alarm 2
Type
Size
Location
Timing
Impact
Network
State
Risk
Horizon
Forecasted
Demands …
Alarm M …
Impact X …
i.e. (water & energy losses, low pressure, supply interruption, discolouration, damage..)
Potential
Incident N

6. Pipe Burst Occurrence Likelihood
Combination of several bodies of evidence
Dempster-Shafer Theory

7. Pipe Burst Impacts ECONOMIC SOCIAL ENVIRONMENTAL Pipe Burst Lost Water
Water Utility
Customers
Low Pressure
ECONOMIC
SOCIAL
ENVIRONMENTAL
Pipe burst might cause: Lost Water, Low Pressure, Supply Interruption (direct because pressure < 7m – in the paper we mention that it was 0, this has changed after the meeting with YWS), Discolouration, Third Party damage, Energy losses
The above impact factors are affecting the two principal stakeholders (i.e., the water utility and the customers)
And the effect on the stakeholders has been categorized as Economic, Social and Environmental
Suitable surrogate models have been developed to reflect the above impacts
Supply Interruption
Pipe
Burst
Discolouration
Third Party Damage
Energy Losses

8. Risk-based Decision Making
Risk maps
Non-aggregated risk
Pipe burst investigation
Likelihood of burst occurrence
Impact of the burst over a given horizon
Low
High
Likelihood
Impact

9. Performance Considerations
Risk assessment computationally demanding
Database-centric distributed architecture
Negligible communication delay

10. Case Study 16 DMAs 25,000 properties 95% residential
>300 km of mains
Demand: 35 MLD
>8,700 Nodes
>9,000 Pipes
69% not metered
Urban DMA
1,600 properties
95% residential
19 km of mains
Demand: 1 MLD
447 Nodes
468 Pipes
Aim of the case study: Impact assessment of a hypothetical “what-if” failure scenario in an urban DMA
Number of relevant impact factors presented, no supply interruption has occurred
The results presented were obtained in E023 – Urban DMA in the city of Harrogate
However, impact evaluation has been done on the whole network so that e.g., cascading DMAs were considered. In the urban DMAs the pressures are generally very high and the impact was localised to E023 only. This would not be the case of rural DMAs (e.g. E054 and E093 in the southern part of the case study area)

11. Alarm 1 – Risk map (Low Impact)
WMS Order No
The purpose of these 3 slides is to show that by applying the risk-based approach we are able to prioritize alarms having the same abnormal flow (in this case 5l/s) which would presumably be considered as equally important (for the sake of simplicity the example is illustrated on a single DMA where the issue is more apparent but could be obviously used to prioritize alarms coming from different DMAs).
The risk map shows the likelihood of occurrence of a burst (estimated burst flow 5l/s) and its impact based on low pressure due to undelivered water (impact aggregation based on Lydia’s methodology not done yet). The studied DMA is E023. In this case the burst seems to be most likely located at the bottom part of the DMA where it would not cause much damage.

12. Alarm 1 – Risk plot
A scatter plot allowing the operator to look into detail where the most likely burst locations are by selecting a point in the scatter plot which will be immediately displayed in GIS. By presenting the potential incidents using a scatter plot we avoid the need to have a table and the ranking since this clearly indicates what are the pipes with high burst likelihood and the impact. Such display will be the primary potential incidents view in the DSS (i.e., instead of the list)

13. Alarm 2 – Risk map (Medium Impact)
WMS Order No
The burst of the same magnitude has this time most likely occurred in a “medium impact part” of the DMA.

14. Alarms 1&2 Risk Comparison
At the moment only “visual comparison” of the risk of the alarms has been done (in reality the alarms would be coming from different DMAs). The “Low impact” alarm seems to be clearly inferior to the “Medium impact” alarm although this would not be apparent if only the estimated burst flow would be considered to rank alarms.
Exeter DSS will visualise the potential failures (of one alarm – see the slide before: “Alarm 1 – Risk plot”) using the above scatter plot as well as the potential failures table as part of the investigation. Thus risk of potential failures will be presented in a non-aggregated form.

15. Future work Automated prioritisation of alarms
Based on the risk of all potential incidents
Further performance improvements
Grouping of similar pipes using clustering
Implementation in a near real-time DSS

16. Summary Supporting control room personnel
Non-aggregated risk presentation
Risk-aware decision making
Better insight into WDS behaviour
Improved response to contingency situations
Reduced failure consequences
Risk-aware decisions – the operator dispatches field crew to a part of a DMA, knowing if it is a high/low impact area

17. Thank you! Questions? www.exeter.ac.uk/cws/neptune
The work on the NEPTUNE project was supported by the U.K. EPSRC grant EP/E003192/1 and Industrial Collaborators.