Genetic Algorithm Optimization for Accurate Hydraulic and Water Quality Analysis of Water Systems Z. Y. Wu, T. Walski, R. Mankowski, G. Herrin and R. Gurrieri Bentley Systems, Incorporated, USA E. F. Arniella, E. Gianellaand, Envirosoft Eng. & Sci., Inc., USA C. Clark, City of Sidney, Ohio, USA P. Sage, United Utilities PLC, UK
Outline Needs for innovative technology Water system analysis in a nutshell Long-standing and emerging challenges Competitive solution methods Practical applications
Sustainable Needs Water systems are vital role in mankind history Systems are deteriorated over years Must be upgraded systematically Need comprehensive and accurate analysis Hydraulics throughout system Water quality characteristics Water security Cannot be achieved without innovation
WDS Analysis Overview Hydraulic model since 1960’s Water quality model Flow conservation law Energy conservation law Water quality model Reynold’s Transport Theorem (RTT) Mass balance Chemical reactions Hundreds of millions invested in modeling
WDS Model Examples Model tens or hundreds of thousands of elements (pipes, pumps, valves, tanks and reservoirs) Model a few pipes
Challenges Long-standing build accurate and robust model Identify roughness coefficients for all pipes Identify demand amount of water out of system Identify pump and valve operating settings Emerging challenges Water security Tougher regulations for water quality Higher customer expectation Tighter financial budget
Call for the best innovations!! Challenges (cont.) Customers are expected to save water Water companies are losing more water than the saved In average, water loss > 15% Call for the best innovations!!
Innovation Since 1960’s Fight for model accuracy via calibration Adjust model parameters Minimize the model predicted and the observed Hundreds of papers published Lack of robustness for handling growing complexity Mixed continuous and discrete parameters Static and dynamic parameters Large model size Astronomically large solution space
Generalized Formulation Search for: Minimize: Subject to: Where: fi is the roughness coefficient for pipe i mj,t is the demand factor for node j at time t Sk,t is operating setting for element k at time t is defined in four distance functions
fmGA Optimizer Multi-era and two-loop evolution Initialization Building Block Filtering Generation = 1 Selection Cut Splice Mutation Generation++ Era++ Generation Loop Era Loop Multi-era and two-loop evolution Start with short strings Enable partial solutions for a large system optimization
Core Method Darwin Calibrator Integrated into WaterCAD and WaterGEMS standalone version and multiple platforms of MicroStation, AutoCAD and ArcGIS
Handle Parameter Dynamics Snapshot dataset system wide data at one time step Allow multiple snapshots Optimization for all snapshots
Application Guidance Make parameter sensitive grouping Decompose system into subsystems Progressive calibration/optimization in multiple inherited runs
Competitive Case I Water system model for city of Guayaquil, Ecuador Supply 2.3 million people Water loss > 50% Optimize model parameters Improve project productivity 40 man-hours with the innovative tool At least four times as long (160 man-hours) with conventional modeling method
Competitive Case I: Benefits Forge non-revenue water reduction plan Simulate water loss in low pressure zones Identify pipes to be replaced or rehabilitated Analyze effects of future system expansions Produce informed 30-year master plan for City
Competitive Case II Water system model for city of Sidney, Ohio More than 150 miles water mains Identify system demand and pipe roughness coefficient Enable informed system analysis
Competitive Case II: Benefits Model new City subdivisions Report annexation assessment Model new industrial users Provide fire flow data to developers, engineers, architects and fire fighters Develop a new Hydrant Tagging System Blue hydrant >1500 gpm Green hydrant 1000 – 1500 gpm Orange hydrant 500 – 1,000gpm Red hydrant < 500 gpm
Vasconcelos et al (1997)-Benchmark Competitive Case III Oberlin zone of Harrisburg, PA Water quality benchmark funded by AWWARF Items GA Solution one GA Solution two GA Solution three Vasconcelos et al (1997)-Benchmark Sum of absolute mean differences 1.3090 1.3106 1.3119 2.4670 Average absolute mean difference (mg/L) 0.0450 0.0860 Excel benchmark results More robust and effective at handling all types of chemical reactions
Competitive Case IV No technique for both water loss detection and model calibration Apply Darwin Calibrator to a District Meter Area (DMA) in UK Optimize nodal demand Locate actual demand differences Predict leakage hotspots Minimize leak detection uncertainty Facilitate a better detection rate
Conclusions Solve the indisputable difficult problem of model calibration (HUMIES criteria G) Better the methods for the long-standing difficult problem of hydraulic and water quality model calibration (HUMIES criteria E) Produce better results than the research project supported by America Water Works Research Foundation (HUMIES criteria F) Calibrated modeling results have been published and also used in practice (HUMIES criteria D)
Conclusions (cont.) Outperform the previously published methods in robustness, flexibility and effectiveness (HUMIES criteria B & C) Provide the new method for water loss/leakage detection (HUMIES criteria D) Excel human-competitive criteria Generalize the human-competitive results for practical applications Integrate as a off-shelf modeling tool in multiple CAD and GIS platforms Develop the application guidelines for industry applications Bring the benefit of the technology advancement to water industry The technology has been applied around the world
Full Citations [1] Wu, Z. Y. (2006) "Optimal Calibration Method for Water Distribution Water Quality Model.", Journal of Environmental Science and Health Part A, Vol. 41, No. 7, pp1363-1378. [2] Wu, Z. Y. and Sage P. (2006) “Water Loss Detection via Genetic Algorithm Optimization-based Model Calibration” ASCE 8th Annual International Symposium on Water Distribution Systems Analysis, Cincinnati, Ohio, August 27-30, 2006. [3] Clark, C. and Wu, Z. Y. (2006) "Integrated Hydraulic Model and Genetic Algorithm Optimization for Informed Analysis of a real system" ASCE 8th Annual International Symposium on Water Distribution Systems Analysis, Cincinnati, Ohio, August 27-30, 2006. [4] Wu Z. Y. and Walski T. (2005) “Diagnosing error prone application of optimal model calibration.” International Conference of Computing and Control in the Water Industry, Sept. 5-7 2005, Exeter, UK. [5] Wu, Z. Y., Elio F. A. and Ernesto G. (2004) "Darwin Calibrator--Productivity and Model Quality for Large Water System", Journal of America Water Works Association, Vol. 96, No.10, pp27-34. [6] Wu, Z. Y, Walski, T., Mankowski, R., Herrin G., Gurrieri R. and Tryby, M.(2002) “Calibrating Water Distribution Model Via Genetic Algorithms”, in Proceedings of the AWWA IMTech Conference, April 16-19, Kansas City, MI.
Thank You! Zheng Y. Wu, Ph.D Bentley Systems, Incorporated Haestad Methods Solution Center 27 Siemon Co Dr. Suite200W Watertown, CT06759, USA Email: zheng.wu@bentley.com Website: www.bentley.com