Surrogate modeling

🌎 This article is a stub, and is currently under construction. You can help by adding to it!

This page is about the applications of machine learning (ML) in the context of surrogate modelling. For an overview of surrogate models more generally, please see the Wikipedia page on this topic.

Machine learning can approximate simulations by learning the relation between inputs and outputs. These surrogate models can run considerably faster as they have much lower computational demands. This technique can be used in many different application areas in climate change mitigation and adaptation. For example, building energy simulation (BES) programs are software tools that simulate the complex physics of buildings and are key enabling tools for R&D in the building's domain. However, detailed BES models are notoriously difficult to design, tune and typically have high computational demands. ML, in conjunction with physics, can help to build accurate yet computationally efficient surrogate models for faster simulations.

Background Readings

Learning a surrogate model is a straightforward supervised machine learning task, given the inputs and outputs of the simulation. However, often the surrogate model is used to find inputs that give the desired output, such as a design for a building or machine with low energy costs. If the surrogate model is used in such an optimisation framework (finding optimal outputs), the technique is called surrogate-based optimisation. An introduction to these techniques can be found in chapter 10 of Engineering Design Optimization.

While these techniques have been used in many applications to reduce the number of computationally expensive simulations, thereby saving energy, their potential for climate change mitigation and adaptation are investigated in A Survey on Sustainable Surrogate-Based Optimisation.