Efficient Water Distribution Presentation

Modelling Toois Dinamis dan studi kasus di water gems. Misalnya scheduling pompa, hidran pembila san, darwin calibrator untuk deteksi kebocoran dan pengaturan katup dengan menghubungkan permintaan data secara real time

Pedro Pina, Water Industry Solution Architect, Bentley Systems

© 2008 Bentley Systems, Incorporated

Rangkuman

• Pandangan atas masalah • Alat seadanya • Features Update• Examples

2 © 2008 Bentley Systems, Incorporated

Masalah

Data

Report

Persiapan Input

Simulation/ Analysis

Verification


Features

• Input/Integrasi Data- Pengembangan Model - SCADA Connect - Water Objects

• Simulation/Analsys - Model Calibration - Leakage Detection - Close Valve Detection - Pump Optimization - Data Logger Placement


Input - Model Builder

CAD, GIS, TableData ModelBuilder WaterGems/CAD


Connectivity Issues Pipes without end nodes

Pipes that do not connect but should

Pipes that appear to connect but are not

Pipes that cross without junctions


Complex Demand Paterns

SCADA


Input - SCADA Connect


Input - Water Objects

• .NET Development Environment • Means of extending capability of model • Can do

- Pre-processing - Post-processing - Add engines


Real Time modelling

Current Time


Water Objects Examples


Available Tools

• Commercially Available - Model Calibration - Leakage Detection - Closed Valve Detection - Pump Optimization - Pipe Renewal Planning and Optimization

• Prototypes (also available under conditions) - Fire Hydrant/Flushing optimization - Data logger optimization - Improved Pump Optimization


Darwin Calibrator - Calibration Module Including in WaterGEMS - Addition in WaterCAD

• Theory of natural selection developed in the 70’s

• Applied to water systems in the 90’s

• Optimization through genetic algorithms

• Uses multiple field data sets to calibrate: Roughness, Demands and States

• It generates tests of successive populations

•Comparison of Field data: • The strong will survive-Pressures or gradients at nodes-Flows in pipes, pumps, and valves


Leak Detection Model Overview

• Data-driven models - Apply data-driven modeling methods, e.g. statistics/

regression methods, artificial neural network, support vector machine etc.

- Purely based on data analysis (pressures and flows) to uncover new abnormalities as possible new pipe bursts

- Unable to detect the leaks that have already existed in the system

• Physics-based models - Simulate leakage through network hydraulic model - Model leakage as pressure dependent demand for known

leakages - Key is to predict the whereabouts and size of leaks

throughout a distribution system

14 | WWW.BENTLEY.COM © 2008 Bentley Systems, Incorporated

Tota

l Hea

d (m

)

Tota

l Hea

d (m

)To

tal H

ead

(m)

Tota

l Hea

d (m

)

Perfect Hydraulic Modeling 120.00

120.00 100.00

100.00 80.00

80.00 60.00

00:00 08:00

Model16:00

Field Test00:00

60.00 00:00 08:00 16:00 00:00

KALL2 Source

120.00 120.00

100.00 100.00

80.00 80.00

60.00 00:00 08:00 16:00 00:00 60.00

Model Field Test 00:00 08:00 16:00 00:00

Model Field Test


Model Calibration Demand Leakage Actual HGL

100 gpm

Calibration Time

Pre Calibration

Post Calibration

80 gpm

Modelled HGL

HGL

80 gpm

A B C


Leakage Hydraulics

HGL

Leakage is allocated to the node in a model Q1 =k1 (p1)^n

• Leakage Qi is pressure dependent, given as emitter flow as above • Ki is the emitter coefficient to be optimized as leakage indicator • Ki > 0 indicates a leakage at node i, while Ki = 0 indicate no leakage

at node i.


Parameter Identification • Genetic Algorithm based network calibration/optimization

tool • Made up of a GA (Darwin Calibrator) and hydraulic

engine • Has three functionalities

- Demand based calibration - “Pressure Dependent” based calibration – Leakage Hotspot Detection

• The GA optimizes any combination of: - Nodal outflow (Consumption and/or Leakage) - Links roughness - Links operational status

• Attempts to generate nodal heads and flow rates that best matches recorded field data


19

Enable Active Water Loss Control

• Predict the location and size of water losses (both real and apparent)

• Guide field engineer to quickly locate leaky pipes and/or apparent water losses

Field personnel can focus on area(s) detected by Darwin Calibrator


Integrated Framework Darwin Calibrator for Leakage Detection & Model Calibration


Real DMA Leakage Detection • Darwin Calibrator for leakage detection• 28 observed pressure points and total

inflow into the system • Apply it to watersystems in UK

• Optimize emitter flow

• Predict leakage hotspots

• Minimize leak detection uncertainty

• Facilitate a better detection rate

• Narrow down leakage spots


Case I Leak repaired, 10 l/s saving

Posi-Tect & field survey

Forest Farm

Leakage spots identified with Darwin Calibrator


Case I (cont) Historical leak

Predicted leak



Predicted leak


Case I (cont)

Historical leak

Predicted leak


Closed Valve Detection


Pump Optimization

Tank Supply

PumpStation


CSP Case Study (Wu, Woodward & Allen 2009) • DMZ system • 57 Ml/day• 11 pump

stations and 9 tanks

• Energy cost: £330K/year

• Recorded daily energy cost: £912

• Modeled daily energy cost: £923


Energy Cost comparison Pump Existing controls Optimized controls

Pump utilizationID (%) Daily cost (£) Pump utilization (%) Daily cost (£)X2420052_ 100 181.99 100 181.73X2420014_ 40 142.11 41 120.51X2420075_ 42 201.95 37 141.19X2410361_ 50 31.99 42 22.65X2419963_ 50 31.99 42 22.65X241998C_ 26 7.92 31 5.18X2450024_ 40 37.35 21 13.87PILWTH 82 236.19 40 98.33NEWMRKT 23 111.63 22 88.98Total cost(£) 983.12 695.10

• Overall saving is 29% of original energy cost• By shifting pumping hours and increasing supply of 3.5 Ml/d from

gravity source


Pipe Renewal Planning and Design Optimization

• Condition assessment tool

• Tool to rank pipe links based on several “aspects”

• Calculate a score for each aspect

• Combine scores for overall ranking

• Part of WaterGEMS, or WaterCAD add-on • Results display tools


WorkflowPipe Score

Pipe Break History

BreakAnalysis

Normalized Break Score

System Inventory

Model

Fire Flow CriticalityAnalysis Analysis

Normalized NormalizedFire Score Criticality Score

Other Property Of Interest

Analysis

Normalized Score

Weighting

OverallScore


Color Coding by Score


Rehabilitation Optimization


Available Tools

• Comercially Available - Model Calibration - Leakage Detection - Closed Valve Detection - Pump Optimization - Pipe Renewal Planning and Optimization

• Prototypes (also available under conditions) - Fire Hydrant/Flushing optimization - Data logguer optimization - Improved Pump Optimization


GPU

Ad

vanc

emen

t

A New Paradigm… Heterogeneous

CPU Advancement Computing

Single-core Many-core

HomogeneousComputing

CP CPU U1 N

U Coh

Generalpurpose

Graphicsdriver

GPU

Memor


Accelerated Modeling14001200

1000

800

600

400

200

0

2.84

1.51

48 hr 96 hr

CPU GPU

3.943.52

144 hr 192 hr

SpeedUp

1500

43.532.521.510.50

3.94 4

1000 2.46

500

014K Pipes

CPU

3

2

1

081K Pipes

GPU

SpeedUp Linear (SpeedUp)

Spee

dUp

Seco

nds

Spee

dUp

Seco

nds


Com

puta

tion

Tim

e (m

in)

Data-Driven Model• Big data, big opportunity• Data ≠ information• Capture data relationships• Fast ANN modeltraining/calibration

765

Host CPU Neural Computing on GPU

Uploaddata to

GPU

800700600500400300200100

0On CPU

ReadfromGPU

14

On GPU 38

Pump Scheduling • Optimize pump operation • Minimize energy cost

Fitness

Solutio n

Respon se


Accelerated Pump Scheduling

Fitness

Solutio n

Trained A Response


Flushing Problem

• Opening hydrant changes head loss and flow velocity of pipes, which is useful - Greater the change, more helpful for the model

calibration - Changing velocity helps remove bad accumulations

in the pipe 41 © 2008 Bentley Systems, Incorporated• Very common operation in practice

Hydrant Selection - Find Best Hydrants To Open

• We don’t want to open all hydrants <- Limited number of hydrants should be opened

• Which one to open? -> Affect as much as possible pipes [Efficiency]

• How many to open? -> Require as few as possible [Cost]

• How much hydrant flow should be used? -> Smaller the better


Case study • Use Hydrant Selection

Tool to find optimal combination of hydrants - A water system with 429

pipes - Currently 8 hydrants are

selected for flow testing, selected by experience

- Head loss change threshold: 0.1 m H2O (0.14psi)

- Hydrant Flow Range: 32 - 126, Interval: 4


Sum of Pipe Lengths Comparison Current 8 Hydrants Optimum 8 Hydrants


Affec

ted

Perc

enta

ge

Sum of Pipe Lengths / Total Pipe Lengths Comparison

Optimal VS Existing 8 90.00%

80.00%

70.00%

60.00%

50.00%

40.00% OptimalExisting 8

30.00%

20.00%

10.00%

0.00% 1 2 3 4 5 6 7 8 12 20

Number of open hydrants

• Optimal solution outperforms existing 8hydrant setting with even 2 hydrants


Logguer Placement


Aknowlegments

• http://inside/bsw/AppliedResearch/Watertown/S itePages/Home.aspx

• Zheng Wu et al. 2011,2012,2114


Efficient Water Distribution Presentation

Documents

Transcript of Efficient Water Distribution Presentation