Optimising Complex Systems

Dec03

Posted by SE-Training
Posted in Courses

Complex systems, by their very nature, are cross-functional, high-dimensional, have inherently complex interdependences between the individual components and generally exhibits, amongst other characteristics, non-linearity and dependency to changes of external conditions. Optimising complex systems therefore generally requires a well orchestrated multi-disciplinary modelling effort, besides setting clear goals and defining the right optimisation strategy and trade-offs. Complex systems are found not only in engineering, but also across various fields of science and social sciences.

Inevitably, the quality of the optimisation strongly relies on the quality of the modelling, and therefore designing the correct model architecture and defining the right modelling strategy should be treated as a cross-functional project and carefully planned from the optimisation goals. Once the requirements of the optimisations are set, the modelling phase requires decomposition and simplification, breaking down the complete system into subsystems, and then it requires consistent integration of all the subsystems, reassembling them into a cohesive whole.

Engineering systems can normally rely on the physical modelling but also non-physical model like ‘behavioural’ or empirical models, common also in other fields, can be usefully employed at subsystem level when knowledge of the component is limited, uncertainty is high or speed of execution is a critical factor. Finally, models should be designed to allow quick and efficient correlation with experimental data to assess predictive accuracy. They also should be designed to be easily ‘accessible’ to carry out quick updates, this to ensure the adaptability of the model and fast response to project or requirement changes.

What defines ultimately a good model for optimisation? It lies in the ability of the model to capture the essence of the system with just enough details to serve the optimisation goal and accuracy, but no more. In this sense, the famous quote often attributed to Albert Einstein ‘Everything should be made as simple as possible but not simpler’ is particularly true. Striving for simplicity in modelling is not only a matter of efficiency, it is also importantly connected to smoother cost functions, crucial for an effective and reliable optimisation.

Models are not only the essential tools for optimising complex systems but also for enabling and driving a deep understanding of their behaviour. The very act of modelling a complex system requires prior knowledge and insight into its underling principles. Yet, once developed, models – both of the system and its subsystems – can be interrogated and characterised, to analyse and understand a wide range of operating conditions and sensitivities. While not directly tied to optimization, this understanding is essential for detecting anomalies, correlate to real world data, managing complexity, and orchestrating effective solutions across disciplines. All of these contribute indirectly, but significantly, to a more robust and informed optimisation process.

Optimising a complex system often requires balancing trade offs between competing requirements and different strategies can be employed to manage them and guide the decision-making process. Whether through a Pareto front, a weighted-sum method, or another multi-objective approach, transparency about the chosen approach is essential when interpreting the results. Moreover the outcomes of the optimisation should demonstrate robustness to change in external conditions, acknowledging the inherent uncertainty present in real-world application.

Importantly, optimisation in complex, cross-functional environment also requires harmonising the contributions of experts across disciplines. Success depends not only on the technical excellence but also on fostering trust, promoting collaboration, and aligning diverse perspectives towards a common goal. In this sense, optimising the systems also become a matter of orchestrating human expertise, today also increasingly augmented by the fast-evolving capabilities of a connected world and artificial intelligence.

– Lucia Conconi, SE-Training Guest Trainer