Design Engineering

A Machine Learning Based Framework for the Design and Evaluation of Flexible, Sustainable and Resilient Energy Systems Under Uncertainty

Cesare Caputo
Dr Michel-Alexandre Cardin
Embracing the White Space

The impact of climate change has become more frequent, visible and extreme in recent years with increasingly adverse and unpredictable future effects predicted by the scientific community. As a result, there are increased challenges associated with the engineering design of energy systems related to economic, environmental and social considerations.

Designing for uncertainty from the early stages of the design process can help significantly increase the sustainability, resilience, and flexibility of these systems with important positive externalities related to the United Nations Sustainable Development Goals (UN SDGs). My research thus focuses on the development of an integrated framework based on Machine Learning which can be used for the design and optimization process of energy systems in the early decision making stage.

Due to the nature and extent of both the variables and the uncertainty associated with these systems, this kind of optimization problem is likely to become quickly intractable or overly biased toward a specific project in practice. The novelty of this research stems from the use of a deep reinforcement learning approach for exploration of design space combined with a unified framework including resilience, sustainability and flexibility which can be more easily applied and understood by decision makers than current available alternatives.

The system design space for different kinds of projects is proposed to be explored using a combination of physical design variables, subject to a number of feasibility constraints, and the decision rules for flexibility they allow via a deep reinforcement learning approach. An economic value-based formulation is implemented for the reward function to allow mapping of actions (such as including a real option “in” a system or exercising one “on” the system) and states(alternative resulting system design configurations variables) for the formulation of a learning agent policy based on a Markov decision process. New alternative metrics for resilience and sustainability are to be developed which allow better integration into the economic reward function and more efficient exploration of the trade-offs in the design

Incorporating these factors into a single model will provide a credible quantitative analysis tool for designers, allowing better exploration of the trade-offs among those complexities and improving decision making under uncertainty. Adopting an ML approach will help reduce computational costs and modelling requirements for the design and decisionmaking process. Ultimately, the development of such a framework will help enhance the transition towards the UN SDGs to yield more resilient and adaptable energy systems with reduced downside exposure in the face of climate change.

 — A Machine Learning Based Framework for the Design and Evaluation of Flexible, Sustainable and Resilient Energy Systems Under Uncertainty
Graphical representation of proposed design framework.


No comments have been posted on this project yet.

Outdated Browser

This website has been built using some of the latest web technologies. Unfortunately, your browser doesn't support these technologies. To update your browser, please visit Outdated Browser.