Masoud Asadzadeh, Bryan A. Tolson, A. J. MacLean. Dept. of Civil & Environmental Engineering, University of Waterloo Hydrologic model calibration aims.

Slides:



Advertisements
Similar presentations
Topic Outline ? Black-Box Optimization Optimization Algorithm: only allowed to evaluate f (direct search) decision vector x objective vector f(x) objective.
Advertisements

MOEAs University of Missouri - Rolla Dr. T’s Course in Evolutionary Computation Matt D. Johnson November 6, 2006.
Angers, 10 June 2010 Multi-Objective Optimisation (II) Matthieu Basseur.
1 An Adaptive GA for Multi Objective Flexible Manufacturing Systems A. Younes, H. Ghenniwa, S. Areibi uoguelph.ca.
Biased Random Key Genetic Algorithm with Hybrid Decoding for Multi-objective Optimization Panwadee Tangpattanakul, Nicolas Jozefowiez, Pierre Lopez LAAS-CNRS.
Multi-Objective Optimization NP-Hard Conflicting objectives – Flow shop with both minimum makespan and tardiness objective – TSP problem with minimum distance,
Multi-objective optimization multi-criteria decision-making.
Elitist Non-dominated Sorting Genetic Algorithm: NSGA-II
Developing Multi-Lake Regulation Plans for the Great Lakes through Multi-Scenario Optimization Saman Razavi, Bryan A. Tolson, and Masoud Asadzadeh Dept.
Multiobjective Optimization Chapter 7 Luke, Essentials of Metaheuristics, 2011 Byung-Hyun Ha R1.
Kuang-Hao Liu et al Presented by Xin Che 11/18/09.
Spring, 2013C.-S. Shieh, EC, KUAS, Taiwan1 Heuristic Optimization Methods Pareto Multiobjective Optimization Patrick N. Ngatchou, Anahita Zarei, Warren.
Adaptive Multi-objective Differential Evolution with Stochastic Coding Strategy Wei-Ming Chen
1. Elements of the Genetic Algorithm  Genome: A finite dynamical system model as a set of d polynomials over  2 (finite field of 2 elements)  Fitness.
A New Evolutionary Algorithm for Multi-objective Optimization Problems Multi-objective Optimization Problems (MOP) –Definition –NP hard By Zhi Wei.
Multi-Objective Evolutionary Algorithms Matt D. Johnson April 19, 2007.
Diversity Maintenance Behavior on Evolutionary Multi-Objective Optimization Presenter : Tsung Yu Ho at TEILAB.
Genetic Algorithms in Materials Processing N. Chakraborti Department of Metallurgical & Materials Engineering Indian Institute of Technology Kharagpur.
Advisor: Yeong-Sung Lin Presented by Chi-Hsiang Chan 2011/5/23 1.
Torcs Simulator Presented by Galina Volkinshtein and Evgenia Dubrovsky.
The Pareto fitness genetic algorithm: Test function study Wei-Ming Chen
The Calibration Process
Resource Allocation Problem Reporter: Wang Ching Yu Date: 2005/04/07.
A New Algorithm for Solving Many-objective Optimization Problem Md. Shihabul Islam ( ) and Bashiul Alam Sabab ( ) Department of Computer Science.
Masoud Asadzadeh 1, Masoud Asadzadeh 1, Saman Razavi 1, Bryan Tolson 1 David Fay 2, William Werick 3, Yin Fan 2 2- Great Lakes - St. Lawrence Regulation.
Optimal Arrangement of Ceiling Cameras for Home Service Robots Using Genetic Algorithms Stefanos Nikolaidis*, ** and Tamio Arai** *R&D Division, Square.
Quality Indicators (Binary ε-Indicator) Santosh Tiwari.
Evolutionary Multi-objective Optimization – A Big Picture Karthik Sindhya, PhD Postdoctoral Researcher Industrial Optimization Group Department of Mathematical.
On comparison of different approaches to the stability radius calculation Olga Karelkina Department of Mathematics University of Turku MCDM 2011.
Hydrologic Modeling: Verification, Validation, Calibration, and Sensitivity Analysis Fritz R. Fiedler, P.E., Ph.D.
Example II: Linear truss structure
MOGADES: Multi-Objective Genetic Algorithm with Distributed Environment Scheme Intelligent Systems Design Laboratory , Doshisha University , Kyoto Japan.
Constrained Evolutionary Optimization Yong Wang Associate Professor, PhD School of Information Science and Engineering, Central South University
A two-stage approach for multi- objective decision making with applications to system reliability optimization Zhaojun Li, Haitao Liao, David W. Coit Reliability.
THE BATTLE OF THE WATER NETWORKS (BWN-II): PADDS BASED SOLUTION APPROACH Bryan A. Tolson, Ayman Khedr, Masoud Asadzadeh.
Optimization Problems - Optimization: In the real world, there are many problems (e.g. Traveling Salesman Problem, Playing Chess ) that have numerous possible.
Doshisha Univ., Japan Parallel Evolutionary Multi-Criterion Optimization for Block Layout Problems ○ Shinya Watanabe Tomoyuki Hiroyasu Mitsunori Miki Intelligent.
Omni-Optimizer A Procedure for Single and Multi-objective Optimization Prof. Kalyanmoy Deb and Santosh Tiwari.
Optimization of the ESRF upgrade lattice lifetime and dynamic aperture using genetic algorithms Nicola Carmignani 11/03/2015.
Locating Multiple Optimal Solutions Based on Multiobjective Optimization Yong Wang
A Comparison of a SWAT model for the Cannonsville Watershed with and without Variable Source Area Hydrology Josh Woodbury Christine A. Shoemaker Dillon.
Assessing the Impact of Alternative Pipe Groupings on Multi-Objective Water Distribution Network Masoud Asadzadeh Bryan Tolson.
Kanpur Genetic Algorithms Laboratory IIT Kanpur 25, July 2006 (11:00 AM) Multi-Objective Dynamic Optimization using Evolutionary Algorithms by Udaya Bhaskara.
DIVERSITY PRESERVING EVOLUTIONARY MULTI-OBJECTIVE SEARCH Brian Piper1, Hana Chmielewski2, Ranji Ranjithan1,2 1Operations Research 2Civil Engineering.
2/29/20121 Optimizing LCLS2 taper profile with genetic algorithms: preliminary results X. Huang, J. Wu, T. Raubenhaimer, Y. Jiao, S. Spampinati, A. Mandlekar,
Multiobjective Optimization for Locating Multiple Optimal Solutions of Nonlinear Equation Systems and Multimodal Optimization Problems Yong Wang School.
Target Releas e Component 1 Component 3 Baseline Flow Component 2 Design a regulation plan for Lake Superior that:  is easily interpretable (piecewise.
Multi-objective Evolutionary Algorithms (for NACST/Seq) summarized by Shin, Soo-Yong.
Tamaki Okuda ● Tomoyuki Hiroyasu   Mitsunori Miki   Shinya Watanabe  
Neural and Evolutionary Computing - Lecture 9 1 Evolutionary Multiobjective Optimization  Particularities of multiobjective optimization  Multiobjective.
Effect of Potential Future Climate Change on Cost-Effective Nonpoint Source Pollution Reduction Strategies in the UMRB Manoj Jha, Philip Gassman, Gene.
Evolutionary multi-objective algorithm design issues Karthik Sindhya, PhD Postdoctoral Researcher Industrial Optimization Group Department of Mathematical.
Evolutionary Computing Chapter 12. / 26 Chapter 12: Multiobjective Evolutionary Algorithms Multiobjective optimisation problems (MOP) -Pareto optimality.
1 ParadisEO-MOEO for a Bi-objective Flow-Shop Scheduling Problem May 2007 E.-G. Talbi and the ParadisEO team
- Divided Range Multi-Objective Genetic Algorithms -
Constrained Optimization by the  Constrained Differential Evolution with an Archive and Gradient-Based Mutation Tetsuyuki TAKAHAMA ( Hiroshima City University.
ZEIT4700 – S1, 2016 Mathematical Modeling and Optimization School of Engineering and Information Technology.
L-Dominance: An Approximate-Domination Mechanism
The 2st Chinese Workshop on Evolutionary Computation and Learning
Department of Computer Science
Evolutionary Technique for Combinatorial Reverse Auctions
Presented by: Dr Beatriz de la Iglesia
Masoud Asadzadeh Bryan Tolson Robert McKillop
Masoud Asadzadeh Bryan A. Tolson University of Waterloo
Heuristic Optimization Methods Pareto Multiobjective Optimization
A New multi-objective algorithm: Pareto archived dds
Multi-Objective Optimization
Doshisha Univ., Kyoto Japan
Chen-Yu Lee, Jia-Fong Yeh, and Tsung-Che Chiang
MOEA Testing and Analysis
Presentation transcript:

Masoud Asadzadeh, Bryan A. Tolson, A. J. MacLean. Dept. of Civil & Environmental Engineering, University of Waterloo Hydrologic model calibration aims to find a set of parameters that simulates observations. Many performance metrics have been proposed to evaluate the effectiveness of calibration; however a single metric cannot do it comprehensively. Therefore, several metrics must be used to assess calibration quality. Ideally, the best set of parameters in a perfect model with perfect data must be optimal for all metrics; however as no perfect hydrologic model exists, the metrics are usually conflicting, especially when objectives are based on different fluxes and/or state variables. In this study, the performance of variety of multi-objective (MO) optimization algorithms are compared for solving a bi-objective hydrologic model calibration problem. An improvement to the popular hypervolume MO performance metric is also introduced. The performance of the following Multi-Objective algorithms are assessed in solving the above bi-objective calibration problem : SPEA2 SPEA2 - Strength Pareto Evolutionary Algorithm [3] is a GA-based multi- objective optimization algorithm that selects the parents based on the strength (number of solutions that each solution is dominated with) and considering the distance to the k th neighbor. NSGAII NSGAII - Non-dominated Sorted Genetic Algorithm [4] is another GA- based multi-objective optimization algorithm that selects the parents from non-dominated sorted fronts of a generation with the priority to the solutions in first front and considering crowding distance. AMALGAM AMALGAM - A Multi-Algorithm Genetically Adaptive Multi-objective Method [5] utilizes several algorithms simultaneously (e.g. GA, Particle Swarm Optimization, Adaptive Metropolis Search, and Differential Evolution in this study) to search for non-dominated solutions. PADDS PADDS - Pareto Archived Dynamically Dimensioned Search [6] uses DDS as a search engine and archives non-dominated solutions during the search. To maintain the diversity of solutions, PADDS samples from less crowded parts of the set of non-dominated solutions in each iteration. PADDS inherits the parsimonious nature of DDS, so it has only 1 algorithm parameter. Tolson and Shoemaker [1] used the Soil and Water Assessment Tool version 2000 (SWAT2000) to model the Town Brook sub-watershed. SWAT2000 is a spatially distributed continuous simulation model for predicting flow, sediment and nutrient fluxes. Town Brook is a sub- watershed in Cannonsville watershed with a 37 km 2 drainage area. MO optimizers try to find a set of solutions that approximates the true set of non-dominated or tradeoff solutions. Many MO performance metrics have been proposed to assess the quality of an approximate tradeoff. However, only the Hypervolume (HV) metric [7] always prefers approximate tradeoffs that weakly dominate other approximate tradeoffs. normalized HV measures the area in normalized objective space (area < 1) that is weakly dominated by an approximate tradeoff; i.e. the area between the approximate tradeoff and a reference point (yellow area in Figure 1). The reference point in HV leads to very similar HV values (i.e. differences only in 2 rd or 3 rd decimal place) between quite different tradeoffs. Here, a revised hypervolume metric [8] is used to evaluate approximate tradeoff quality. This revised HV metric measures, in normalized objective space, the fraction of the hypervolume between the best and worst attainable tradeoffs that is weakly dominated by the approximate tradeoff (yellow area divided by area between red and green lines in Figure 2). The revised HV metric is specific to a set of comparative MO algorithm results and better highlights MO algorithm performance differences. We focus on the calibration of SWAT2000 to measured flow and total phosphorus loads in Town Brook and as such, selected 25 model parameters to be calibrated. As in [2], the reduced Nash-Sutcliffe (NS) metric is used to assess the effectiveness of calibration for simulated flow and total phosphorus against measured daily data. As presented in the following equation, the reduced NS penalizes the NS of solutions that have %Bias beyond a specified threshold. A threshold of 10% for flow and 30% for phosphorus was used. Therefore, the bi-objective calibration problem aims to maximize the reduced NS for flow and total phosphorus simultaneously.  All algorithm parameters set to recommended values from other studies. e.g. Pop. Size=100, simulated binary crossover for SPEA2, polynomial mutation, and uniform crossover for NSGAII.  All 4 algorithms are stochastic optimizers; therefore, a fair comparison should consider multiple trials of them. In this study, results are based on 5 independent trials of each algorithm with a budget of 10,000 model simulations per optimization trial.  Figure 3 shows the best attainable tradeoffs for each algorithm based on the combined result of the 5 independent optimization trials. Results of NSGAII and SPEA2 are nearly weakly dominated by AMALGAM and PADDS. Therefore, results of AMALGAM and PADDS are more closely compared. Figure 3. Non-dominated solutions for each MO algorithm based on combining results of 5 optimization trials PADDS AMALGAM Figure 4. Attainable objective function region based on PADDS and AMALGAM optimization results over 5 trials. Figure 4 represents the region between the best and the worst attainable tradeoffs for AMALGAM and PADDS.  The best attainable tradeoff consists of all non-dominated solutions after combining all AMALGAM and PADDS results.  The worst attainable tradeoff contains all tradeoff solutions that are weakly dominated by the combined AMALGAM and PADDS results. The following table computes the revised hypervolume metric based on the spread of results in Figure 4 as well as the normalized hypervolume metric for comparison. Metrics are computed for all 5 optimization trials. MetricAlgorithm Trials 12345Avg. Normalized HV AMALGAM PADDS Revised HV AMALGAM PADDS AMALGAM and PADDS perform better than NSGAII and SPEA2 here. PADDS, which is a very simple extension of the single objective DDS algorithm, has comparable but slightly worse avg. results than AMALGAM. The difference between PADDS and AMALGAM is more clearly detected by revised HV metric than Normalized HV (see Avg. metrics in above Table). The revised HV value is more directly interpretable, since it measures algorithm performance relative to the observed performance variation across all algorithms in the comparison. Acknowledgement: We would like to thank the Systems Optimization Group at ETH Zurich, Jasper Vrugt and Aravind Seshadri for sharing the source codes of SPEA2, AMALGAM and NSGAII, respectively. [1]. Tolson, B. A., and C. A. Shoemaker (2004), Watershed modeling of the Cannonsville Basin using SWAT2000: Model development, calibration and validation for the prediction of flow, sediment and phosphorus transport to the Cannonsville reservoir, version 1.0, technical report, Sch. of Civ. and Environ. Eng. Cornell Univ., Ithaca, N. Y. [2]. Tolson, B. A., and Shoemaker, C. A. (2007). “Dynamically dimensioned search algorithm for computationally efficient watershed model calibration.” Water Resour. Res., 43(1), [3]. Zitzler, E., Laumanns, M., and Thiele, L. (2001). “SPEA2: Improving the strength pareto evolutionary algorithm for multiobjective optimization.” Proc., Evolutionary Methods for Design, Optimization, and Control, Barcelona, Spain, 95–100. [4]. Deb, K., Pratap, A., and Agarwal, S. (2002). “A fast and elitist multiobjective genetic algorithm: NSGAII.” IEEE Trans. Evol. Comput., 6(2), [5]. Vrugt, J. A., and Robinson B. A. (2007). “Improved evolutionary optimization from genetically adaptive multi-method search.” Proc. Natl. Acad. Sci. U.S.A., 104(3), [6]. Asadzadeh, M., and Tolson, B. A. (2009). “A new multi-objective algorithm, Pareto archived DDS”. Proc. 11th Genetic and Evolutionary Computation Conference GECCO., Montreal, Canada, [7]. Zitzler, E., and Thiele, L. (1998), “Multiobjective optimization using evolutionary algorithms-A comparative case study,” in Parallel Problem Solving from Nature (PPSN V), Germany: Springer, 1998, pp. 292–301. [8]. Asadzadeh, M., and Tolson, B. A. “Hybrid Pareto Archived Discrete Dynamically Dimensioned Search, a New Multi- Objective Optimization Algorithm, for Solving Water Distribution Network Design Problems”. To be submitted. Upper bound is best attainable tradeoff in comparison Lower bound is worst attainable tradeoff in comparison AGU Fall Meeting, Dec 17, Paper Number: H41A-0869 f2f2 1 f1f1 1 Best attainable tradeoff Approximate tradeoff Worst attainable tradeoff Figure2. Revised Hypervolume Metric for two maximization objectives f2f2 1 f1f1 1 Approximate tradeoff Figure1. Normalized Hypervolume Metric for two maximization objectives Best attainable tradeoff Reference point