Fast Self-Learning of Turbulence Feedback Laws Using Gradient-Enriched Machine Learning Control

Created by W.Langdon from gp-bibliography.bib Revision:1.7877

@InProceedings{cornejo-maceda:2024:GECCOcomp,

• author = "Guy Y. {Cornejo Maceda} and Zhutao Jiang and Francois Lusseyran and Bernd R. Noack",
• title = "Fast {Self-Learning} of Turbulence Feedback Laws Using {Gradient-Enriched} Machine Learning Control",
• booktitle = "Proceedings of the 2024 Genetic and Evolutionary Computation Conference Companion",
• year = "2024",
• editor = "Ting Hu and Aniko Ekart",
• pages = "495--498",
• address = "Melbourne, Australia",
• series = "GECCO '24",
• month = "14-18 " # jul,
• organisation = "SIGEVO",
• publisher = "Association for Computing Machinery",
• publisher_address = "New York, NY, USA",
• keywords = "genetic algorithms, genetic programming, hybrid method, feedback control, downhill simplex: Poster",
• isbn13 = "979-8-4007-0495-6",
• DOI = "doi:10.1145/3638530.3654395",
• size = "4 pages",
• abstract = "We apply genetic programming (GP) to solve the most complex turbulence control problems. We simultaneously optimize up to dozens of parameters and dozens of control laws, listening to dozens of sensor signals with 100 to 1000 short test runs. Unlike reinforcement learning, our implementation of GP does not require any meta parameter tuning and is many orders of magnitudes faster than vanilla versions thanks to numerous enablers: 1. Parameter optimization over many dozens of experiments and simulations. 2. Smart formulation of control laws by preprocessing inputs and outputs. 3. Gradient-enriched simplex optimization of promising subspace. This work focuses on the resulting algorithm, referred to as gradient-enriched Machine Learning Control (gMLC). The applications comprise learning a multiple-input multiple-output control law to stabilize the flow past a cluster of three rotating cylinders in numerical simulations and the stabilization of the shear layer in an open cavity flow experiment. The learning acceleration achieved by gMLC opens the path to complex and time-limited experiments, including evaluation in varying operating conditions and optimization of distributed-input distributed-output control laws, i.e., functions with O(100) inputs and outputs.",
• notes = "GECCO-2024 GP A Recombination of the 33rd International Conference on Genetic Algorithms (ICGA) and the 29th Annual Genetic Programming Conference (GP)",

}

Genetic Programming entries for Guy Yoslan Cornejo Maceda Zhutao Jiang Francois Lusseyran Bernd R Noack

Citations