Evolutionary algorithms (EAs) have been a popular optimization tool for two decades in academia. However, the industry applications of EAs to real-world optimization problems are infrequent, due to the strong assumption that objective function evaluations are straightforward and easily accessed. In fact, such objective functions may not exist, instead computationally expensive numerical simulations or costly physical experiments must be performed for evaluations. In more extreme cases, only historical data are available for performing optimization and no new data can be generated during optimization. Such problems driven by only historical data are formulated as offline data-driven optimization problems, which poses challenges to conventional EAs. Although surrogate models that approximate objective functions can be used as function evaluations in EAs to guide the search, the search accuracy is limited by the offline data and cannot be validated or enhanced by sampling new data. The research on offline data-driven evolutionary optimization has not received sufficient attention, although techniques for solving such problems are very practical. One main reason is the lack of benchmark problems that can closely reflect real-world challenges, which leads to a big gap between academia and industries.
In this competition, there are two tracks: constrained and integer single-objective optimization. We carefully select 4 benchmark problems for each track from two real-world applications (aerodynamic optimization and software configuration tuning). The objective functions of those problems cannot be calculated analytically, but can be calculated by calling a simulation to provide true black-box evaluations for offline data sampling and the final comparison. Participants are required to find the optimal solutions of each benchmark problem based on the provided offline data. To assess the performance of compared algorithms, we will evaluate the found optimal solutions using the same simulation which generates the offline data.
Track 1: constrained single-objective optimization (aerodynamic optimization):
DDCSOP1: This problem is a drag minimization of an RAE 2822 transonic airfoil subject to lift, pitching moment and sectional area constraints, which has 18 decision variables and 3 constraints (not smaller than 0).
DDCSOP2: This problem is a drag minimization of the wing of CRM wing-body configuration subject to pitching-moment coefficient and internal volume of the wing, which has 39 decision variables and 2 constraints (not smaller than 0).
DDCSOP3: This problem is a drag minimization of the wing of DLR-F4 wing configuration subject to lift coefficient and thickness of different positionis of the wing, which has 48 decision variables and 5 constraints (not smaller than 0).
DDCSOP4: This problem is a drag minimization of the wing of M6 wing configuration subject to lift coefficient and thickness of different positionis of the wing, which has 60 decision variables and 4 constraints (not smaller than 0).
Track 2: integer single-objective optimization (software configuration tuning):
DDISOP1: This problem is a maximization of throughput, which has 10 integer decision variables.
DDISOP2: This problem is a maximization of requests per second, which has 20 integer decision variables.
DDISOP3: This problem is a minimization of latency, which has 30 integer decision variables.
DDISOP4: This problem is a minimization of latency, which has 40 integer decision variables.
For both tracks, the boundary of each decision variable is the range between the maximum and minimum of the corresponding dimension in the offline data.
Download links: https://github.com/HandingWang/DDEO-WCCI2020
Important Dates:
For participants planning to submit a paper to the 2020 IEEE World Congress on Computational Intelligence:
Paper submission: 15 Jan 2020
Notification to authors: 15 Mar 2020
Final submission: 15 Apr 2020
Note: You are encouraged to submit your paper to the Special Session on Data-Driven Optimization of Computationally Expensive Problems
For other participants (only result entry but without a paper):
Results submission deadline: 1 May 2020
Note: Please send your results directly to Dr Handing Wang (hdwang@xidian.edu.cn)
Organizers::
Handing Wang, School of Artificial Intelligence, Xidian University, China
Liang Bao, School of Software, Xidian University, China
Fei Liu, School of Aeronautics, Northwestern Polytechnical University, China
Zhonghua Han, School of Aeronautics, Northwestern Polytechnical University, China
Yaochu Jin, Department of Computer Science, University of Surrey, UK
Competition entries::
Track 1: constrained single-objective optimization (aerodynamic optimization):
CDDEA-DSGPR: Constrained offline data-driven evolutionary optimization based on dynamic selection of Gaussian process classification, Pengfei Huang, Handing Wang, Xidian University
SA-IBEA: A Surrogate-Assisted Indicator-Based Evolutionary Algorithm Using Model Selection for Constrained Offline Data-Driven Optimization, Zhening Liu, Handing Wang, Xidian University
SA-2Archive: A Surrogate-Assisted Indicator-Based Evolutionary Algorithm Using Model Selection with Two Archives for Constrained Offline Data-Driven Optimization, Zhening Liu, Handing Wang, Xidian University
EGP-DEPSO: Ensembled Genetic Programming with Differential Evolutionary Particle Swarm Optimization, Diego Rodriguez, David Alvarez, Ameena Al-sumati, Sergio Rivera, Universidad Nacional de Colombia, Khalifa University
HPPSO: Surrogate-Assisted Teaching Learning Based Optimization Algorithm, Weixi Chen, Jie Liu, Huachao Dong, Peng Wang, Northwestern Polytechnical University
CGP-ODCO: A Cotraining Method for Gaussian Process Assisted Offline Data-driven Constrained Optimization, Xilu Wang, Yaochu Jin, University of Surrey
HE-DDEA: Offline Data-Driven Evolutionary Optimization with Heterogeneous Ensembles , Fan Li, Liang Gao, Weiming Shen, Xiwen Cai, Huazhong University of Science and Technology
DE&ACO-K: DE&ACO hybrid optimization algorithm based on Kriging, Jia Xuyi, Yang Lianbo, Li Chunna, Northwestern Polytechnical University
ParetoHyperNet: Pareto HyperNetwork-based Optimization, Xi Lin, Qingfu Zhang, City University of Hong Kong
SurroOpt: Surrogate-base Optimization, Yang Zhang, Jiangbo Chi, Haodong Dang, Guanyu Meng, Chenzhou Xu, Northwestern Polytechnical University
Track 2: integer single-objective optimization (software configuration tuning):
EGP-DEPSO: Ensembled Genetic Programming with Differential Evolutionary Particle Swarm Optimization, Diego Rodriguez, David Alvarez, Ameena Al-sumati, Sergio Rivera, Universidad Nacional de Colombia, Khalifa University
STLBO: Hybrid population-assisted PSO integer optimization, Weixi Chen, Jie Liu, Huachao Dong, Peng Wang, Northwestern Polytechnical University
HE-DDEA: Offline Data-Driven Evolutionary Optimization with Heterogeneous Ensembles , Fan Li, Liang Gao, Weiming Shen, Xiwen Cai, Huazhong University of Science and Technology
SE-DPSO: Surrogate Ensembles Assisted Discrete Particle Swarm Optimization, Xijun Huang, Shenzhen University
ParetoHyperNet: Pareto HyperNetwork-based Optimization, Xi Lin, Qingfu Zhang, City University of Hong Kong
BDDEA-LDG: Boosting data-driven evolutionary algorithm with localized data generation, ?Jian-Yu Li and Zhi-Hui Zhan, South China University of Technology
All the results can be found at: https://github.com/HandingWang/DDEO-WCCI2020
Winer algorithms for Track 1:
Winner: SurroOpt
First runner-up: ParetoHyperNet
Second runner-up: CDDEA-DSGPR
Winer algorithms for Track 2:
Winner: SE-DPSO
First runner-up: HE-DDEA