(C) PLOS One This story was originally published by PLOS One and is unaltered. . . . . . . . . . . Human-in-the-loop MGA to generate energy system design options matching stakeholder needs [1] ['Francesco Lombardi', 'Tu Delft', 'Faculty Of Technology', 'Policy', 'Management', 'Department Of Engineering Systems', 'Services', 'Delft', 'Stefan Pfenninger'] Date: 2025-04 In the following sub-sections we illustrate in greater detail the proposed HITL-MGA workflow and the experimental set up by which we test its effectiveness (sub-section 2.1). This includes a discussion of the specific version of the SPORES MGA algorithm we use (2.1.1), and the original automated method we devise for translating human preferences into a guided SPORES search (2.1.2). Second, we discuss the energy system model of Portugal we employ for testing the workflow (sub-section 2.2). 2.1 HITL-MGA workflow and synthetic experiment setup The idea of the proposed HITL-MGA workflow is to refine the system design space generated via MGA by leveraging human preferences and knowledge of the problem, which get translated into machine-readable parameters. The steps to implement the idea, summarised in Fig 1, are the following. PPT PowerPoint slide PNG larger image TIFF original image Download: Fig 1. Conceptual representation of the proposed HITL-MGA workflow and of the synthetic experiment we carry out in this study to test the workflow. The theoretical framework we propose envisions that we collect the high-level system design preferences arising from a first iteration of an MGA design space with stakeholders. Then, we decode which technical features underlying such preferences and translate them into parameters that may inform and re-tune the MGA search for design options. The resulting updated design space is richer in design options that align with stakeholders’ high-level preferences, and it facilitates consensus formation. In our synthetic experiment, we replace stakeholder interaction with synthetic high-level stakeholder preferences, from which we derive five top-rated designs to initiate the HITL-MGA workflow. This way, we can test the outputs of the HITL-MGA workflow against the initial high-level preferences in a way that would not be possible if interacting with real-world stakeholders whose high-level preferences are typically unknown. https://doi.org/10.1371/journal.pclm.0000560.g001 First, we generate an initial, best-guess MGA system design space. (We provide some discussion on what MGA formulations might work best for an initial exploration of the design space in previous work [10] .) Second, we let stakeholders explore this initial design space, for instance, via a web interface similar to what we proposed in previous work [9], and ask them to select their favourite system design options. For example, stakeholders may prefer systems with a lower import dependency or perceived as more decentralised. We call these ‘high-level’ preferences because they do not necessarily relate to concrete, particular features of the system but can be more abstract or concerned with the macro system. For our method to work, it is not necessary to understand the rationale behind the expressed preferences; the method only requires a list of the most-selected designs. From this, we use automated decoding, which we discuss further in sub-section 2.1.2, to determine the technical features inherent to a given ‘high-level’ preference. For example, the decoding may allow us to recognise that substantial amounts of offshore wind power combined with significant domestic power-to-hydrogen capacity and little expansion of cross-border electricity transmission lines are the key features of a system design highly favoured by stakeholders. This might represent the high-level preference for ‘low import dependency’. Unlike the high-level preference, the technical features it can be decomposed into can be translated into parameters for a customised MGA search that favours or penalises the detected features while simultaneously generating different ways of doing so. The resulting customised MGA search gives rise to an updated system design space, now richer in designs that fulfil stakeholders’ preferences to varying degrees and by means of a diverse range of spatial and technological system configurations. In this study’s synthetic experiment, we test the method based on synthetic high-level stakeholder preferences, which allows us to benchmark the outputs in a way that would not be possible if interacting with real-world stakeholders whose high-level preferences, as anticipated above, may be unknown. We assume five high-level preferences: A low dependency of the system on energy imports from abroad. A limited concentration of onshore wind farms in any given region. A limited overall deployment of new infrastructure to achieve the energy transition. A limited reliance on large-scale infrastructure that requires central planning, such as onshore wind farms, and thus, an increased possibility for bottom-up community initiatives to lead the transition. A low reliance on hydrogen. We consider these five high-level preferences to be plausible based on the insights from our SEEDS project (seeds-project.org), which included interviews with stakeholders in Portugal on this matter. We filter the initial design space generated through the combination of an energy system model of Portugal (2.2) and our MGA SPORES algorithm (2.1.1) to alternatively subset it to only those system designs that perform within the top 10% of all designs on each of these high-level features. Table A in S1 Text provides further details on how we quantitatively assess a given design’s performance with respect to each preference. Within each of these top-10% subsets of designs, we randomly select one option as the representative most preferred or top-rated design for a given high-level preference, ultimately leading to a set of five synthetic top-rated designs. In a real-world case, we would have such a selection of designs based on stakeholder ranking, without knowing what high-level preferences the designs represent. We feed the set of designs into the HITL-MGA workflow outlined above, which produces an updated design space for Portugal. Finally, we test this new design space against the initial synthetic high-level preferences, assessing whether there are more or fewer design options that match those preferences. In addition, as a proxy for the likelihood of consensus formation, we quantify how many design options exist in the neighbourhood of the design that ensures the best compromise between all preferences. To quantify the above capacity of a system design to compromise between the different high-level preferences, we use a simple Multi-Crtieria Decision Analysis (MCDA) approach. In particular, we rely on a typical weighted-sum model [17], in which we assess the performance of the given system design for each of the five assumed synthetic high-level stakeholder preferences, then we sum these performances together with equal weights, as per Eq. 1. (1) where is thus the total score obtained by a given alternative design option when considering all high-level stakeholder preferences simultaneously (n = 5); this corresponds to our proxy for the likelihood of consensus formation of the given design. is the performance of the design in terms of the high-level preference k, and is the weight of preference k compared to the others. In our case, we assume all high-level preferences have equal weight. 2.1.1 SPORES MGA algorithm. For this study, we use a specific MGA algorithm, SPORES, which we presented in earlier work [3] and then continuously developed [9,10]. The key feature of SPORES is that it targets both technological and spatial diversity of system designs, under the assumption that not only what technologies are deployed but also, and in particular, where they are deployed is critical for real-world consensus formation. What is more, compared to other MGA algorithms, SPORES is highly customisable [10] and thus lends itself to being adapted for the guided search informed by stakeholder preferences. Fig 2a illustrates the high-level implementation of the algorithm, which envisions one default run with a single MGA objective and several parallel runs with a multi-objective MGA formulation aimed at generating alternatives from different corners of the near-optimal solution space. The basic idea underpinning this formulation is the following. We identify the cost-optimal solution as a starting point. Then, the algorithm searches for new feasible system designs that are different in terms of both what technologies are deployed and where compared to what is found in the previous solution(s). Mathematically, we achieve this by incrementally assigning penalties, or weights, to spatially explicit capacity investment decision variables that featured prominently in previous solutions. In those runs that feature a multi-objective MGA formulation, we combine the search for spatially and technologically distinct feasible system designs with the intensification of specific design features, such as the high or low deployment of a particular technology. We define ‘intensification’ as pushing the given feature either towards its minimum or maximum feasible value, depending on whether it is a desired or undesired feature. As illustrated elsewhere [10], this ensures that extreme technological boundaries of the design space are captured and that otherwise hard-to-discover design options are generated. In all cases, the newly generated feasible solutions are constrained to be only marginally more costly than the least-cost feasible solution (in this study, we assume this means a 10% cost increase is acceptable, in line with previous work [9,18]). PPT PowerPoint slide PNG larger image TIFF original image Download: Fig 2. Stylised representation of the SPORES algorithm, in its standard (a) and HITL (b) formulations. Starting from the cost-optimal solution, the algorithm iteratively looks for additional feasible and only marginally costlier system designs in parallel batches. The standard formulation (a) foresees S parallel batches that generate alternative designs based on the sole objective of making them spatially and technologically distinct from previously found solutions. Each of these S batches targets decision variables pertaining to a specific energy sector, such as power, heating, mobility or synthetic fuels. At the same time, 2xM additional batches generate (in parallel) alternatives around the intensification of specific features of the design space, for instance, the very high or very low deployment of offshore wind power overall. In the HITL formulation (b), each parallel batch is informed by a high-level stakeholder preference M and intensifies one or more technical features that underlie such a preference while also targeting spatial and technological distinctiveness from previously found solutions. One additional batch (M+1) targets all the given high-level preferences simultaneously. https://doi.org/10.1371/journal.pclm.0000560.g002 In mathematical terms, we can formulate the generic SPORES-MGA problem as in Eq. 2. (2) where i and j indicate the i-th technology type and the j-th location within the model; is the capacity investment decision variable for the ij-th location-technology pair; and is the capacity decision variable associated with the technology under intensification. The a and b coefficients are the weights associated with the different components of the objective function: when b has a positive sign, the technology under intensification is minimised; for a negative sign, it is maximised. Finally, if b is null, the formulation collapses into the default case with a single objective (see Fig 2a). A, b, are a matrix and a vector of coefficients that give rise to the physical constraints alongside the decision variables’ vector x; cost n is the total annualised system cost, which is bound to remain within a marginal relaxation of the optimal cost ( ); and s is the accepted cost relaxation, also known as cost slack. In this study, we apply SPORES to a model of the Portuguese energy system (see sub-section 2.2). Following the logic outlined in Fig 2a, we generate a total of 80 SPORES in four parallel batches using the default single-objective mode applied to decision variables pertaining to different energy sectors (S = 4): power (N = 50 SPORES), heating, mobility and synthetic fuels (N = 10 per sector). Then, we generate a total of 180 SPORES across multi-objective batches that alternatively maximise (N = 10) or minimise (N = 10) one of the following (M = 9) key technology assets: wind onshore, wind offshore, wind overall, open-field solar PV, roof-mounted solar PV, biofuels, battery storage, electrolysis and transmission capacity. This leads to a total of 260 (80+180) different system design options, or SPORES, of which about 70% adopting a multi-objective MGA formulation. [END] --- [1] Url: https://journals.plos.org/climate/article?id=10.1371/journal.pclm.0000560 Published and (C) by PLOS One Content appears here under this condition or license: Creative Commons - Attribution BY 4.0. via Magical.Fish Gopher News Feeds: gopher://magical.fish/1/feeds/news/plosone/