Uncertainty is ubiquitous in decision-making problems across the engineering & life sciences disciplines. Whether decisions are made by governments or industrial companies, failing to incorporate uncertainty within the decision-making models that are used, can lead to undesirable outcomes regarding economic performance, safety and overall resilience. In 2021 Novartis shut down a US plant due to overestimation in gene therapy demand. Meanwhile, global efforts to tackle the climate change crisis necessitate governments and national energy regulators to make decisions now, about the long-term net zero landscape despite the deep uncertainty surrounding technological performance, demand and deployment costs among others.
In that context, the question of how to make uncertainty-aware optimal decisions becomes urgent. Leveraging advances in computational decision-making and data science constitutes a systematic approach for this problem. To this end, instead of using average or “best estimates” of parameters that are needed for the models that systematically inform decision-making, one can exploit past observations of these parameters and derive data-driven scenarios[3] or uncertainty sets[4] as illustrated in Fig.1. The benefit of such approach is that it provides the decision-maker with: (i) the freedom to tune their outlook on risk-averseness & (ii) the ability to quantify systematically the resilience of their options and identify best corrective actions. A recent research work from the SCPSE[4] highlights the crucial role of uncertainty in net zero energy systems planning. In this work, a novel computational framework for hedging against demand uncertainty in net zero models was introduced. The graph in Fig. 2 illustrates the resulting operating and total system cost when the optimal decisions are exposed to a stochastic environment for the deterministic (blue) and the uncertainty-aware approach (green). As shown, although neglecting demand uncertainty results initially in an optimistic capital cost, when the system is exposed to demand uncertainty the resulting operating and total system costs are 42% and 21% higher compared to the uncertainty-aware approach.
