EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
2356. Simulator-based surrogate optimisation: an enhanced framework for surrogate modelling and optimisation
Invited abstract in session WC-28: Advancements of OR-analytics in statistics, machine learning and data science 10, stream Advancements of OR-analytics in statistics, machine learning and data science.
Wednesday, 12:30-14:00Room: 065 (building: 208)
Authors (first author is the speaker)
| 1. | Yu Liu
|
| Department of Mathematics and Systems Analysis, Aalto University | |
| 2. | Fabricio Oliveira
|
| Mathematics and Systems Analysis, Aalto University |
Abstract
This work introduces a methodology for optimising complex functions common in high-fidelity simulations across engineering fields, emphasising the strategic bridging of simulators and optimisation processes. Initially, we employ low-discrepancy sequence sampling to select simulation points, followed by training a surrogate model using a piecewise linear neural network with Rectified Linear Unit (ReLU) activation. Using Mixed Integer Programming (MIP), we reformulate the ReLU neural network as a MIP optimisation problem and solve it to find an optimal parameterisation of the simulator’s input. Based on the best solutions obtained, we then enter an iterative domain-refined phase of resampling the simulator, retraining the surrogate model, and rebuilding and resolving the MIP problem. For resampling, we use an infill strategy that incorporates error assessment to balance exploration and exploitation. This process continues until the accuracy of the objective function reaches the desired tolerance, ensuring high surrogate accuracy near the optimum, with techniques like memory structure reuse and warm starts for efficiency. Validation against standard test functions indicates that refining the surrogate model via an error-informed resampling strategy significantly enhances optimisation efficiency. This framework advanced in surrogate-based optimisation by synergistically combining adaptive sampling, neural networks, and MIP for enhanced performance in complex simulations.
Keywords
- Machine Learning
- Programming, Mixed-Integer
- Simulation
Status: accepted
Back to the list of papers