The increasing penetration of renewable energy sources requires energy system planning models to explicitly account for uncertainty. Stochastic programming, applied in this thesis to generation expansion planning, is a powerful tool for this purpose, but its computational burden often makes realistic applications intractable. This thesis investigates temporal reduction techniques based on representative periods and aims to improve understanding of the mechanisms and data characteristics that most strongly influence investment decisions.
In particular, the thesis compares selecting representative periods independently for each scenario and jointly across scenarios in stochastic energy models, as well as traditional clustering methods and the more recent hull clustering approach. The results show that per-scenario selection can have advantages when scenarios differ significantly, while cross-scenario selection generally performs better as the number of scenarios increases. Furthermore, hull-based methods are shown to better capture extreme conditions that strongly influence investment decisions.
The thesis also builds on recent work that improves feasibility under extreme conditions through the inclusion of artificial worst-case periods, highlighting the resulting trade-off between system reliability and over-conservative investment decisions. To mitigate this issue, new and less conservative definitions of artificial worst-case periods are introduced, together with the Artificial Period Guided Selection (APGS) algorithm, which combines artificial and real worst-case periods and adapts blended weighting strategies to balance feasibility and cost-efficiency. Computational experiments on stylized European case studies show that APGS can significantly reduce over-conservatism, leading to cost-efficient and reliable solutions. ... Read more

The global transition to renewable energy forces electricity systems to integrate large shares of variable generation, making investment planning increasingly complex. Energy System Optimisation Models (ESOMs) jointly optimise capacity investment and operational dispatch to support these decisions, but their computational cost grows rapidly with increasing temporal resolution. Temporal aggregation reduces this cost by replacing the full time series with a smaller set of representative partitions. A known weakness of standard aggregation methods is the smoothing effect: extreme demand peaks and near-zero renewable generation periods are averaged with their neighbours, causing the optimisation model to underestimate capacity requirements and incur avoidable costs when its investment plan is evaluated against the full time series.

This thesis proposes two extensions to chronology-preserving hierarchical clustering that actively counteract this smoothing effect. Post-hoc Extreme Correction (PEC) runs standard Ward hierarchical clustering and then replaces the centroid of any partition containing an extreme time step with the corresponding maximum or minimum value. Extreme-Aware Clustering (EAC) integrates extreme-event information directly into the merge priority, delaying the mixing of extreme and non-extreme time steps and thereby forming more coherent partitions before any representative correction is applied. Both strategies are embedded within the Fully Flexible Temporal Resolution (FFTR) framework, enabling automated, data-driven, variable-specific temporal partitions.

We evaluate all three strategies --- baseline Hierarchical Clustering (HC), PEC, and EAC, together with Uniform Temporal Resolution (UTR) as an additional baseline --- using load duration curves, Sum of Squared Error analysis, and comparison against the full-resolution benchmark. We find that EAC's advantage over HC is conditional rather than universal: EAC outperforms HC when temporal variability is high and cluster counts are low, but HC achieves comparable or lower regret when variability is moderate and the resolution assignment for non-profiled assets is consistent. We additionally find that the resolution assigned to non-profiled assets and inter-location flows has a first-order effect on investment quality: aligning these to the common highest resolution of their connected profiled endpoints dramatically reduces HC regret, substantially reordering the method rankings at low to moderate resolutions. These results suggest that the resolution assignment strategy matters as much as the choice of clustering algorithm, and that EAC is a robust improvement over HC specifically when extreme events dominate the cost structure. ... Read more

The rapid integration of renewable energy is making long-term power system planning more important and more computationally demanding. Generation Expansion Planning (GEP) is a key problem in energy planning because it determines where and how much to invest in generation capacity. As renewables grow, GEP must be solved across many scenarios, demand forecasts, and technology options, making the speed of solving it a key bottleneck for timely policy analysis. GEP is commonly formulated as a large-scale mixed-integer linear program and is often solved using Benders Decomposition, where investment decisions are optimized in a master problem and the Economic Dispatch (ED) problems are solved as subproblems, but this remains computationally expensive at scale. Recent work proposed learning the dual solution of the ED subproblem with a machine-learning surrogate to generate approximate Benders cuts during an inexact phase, followed by exact refinement to recover optimality guarantees. The runtime benefit of this framework, however, depends on whether the learned surrogate produces cuts that are useful within the Benders algorithm, which has not been studied empirically. This thesis investigates how machine-learning dual surrogates can be more effectively designed and integrated into Benders Decomposition to accelerate GEP. Three aspects are studied. First, a constraint-aware data generation strategy is proposed to construct surrogate training instances that better reflect the investment decisions encountered during Benders iterations. Second, limitations of self-supervised dual learning are diagnosed. To address these limitations, this thesis proposes Heuristic Soft Label Augmentation (SLA), which adds cheap, structure-aware auxiliary guidance for learning ED shadow-price classes without requiring exact dual labels. Third, the thesis studies how each subproblem’s dual solution should be converted into Benders cuts, comparing single-cut, full multi-cut, and a proposed K-means clustered cut-aggregation strategy and analysing their effect on overall runtime. The results show that constraint-aware data generation improves the usefulness of surrogate-generated cuts during the inexact phase, and that SLA reduces the dual surrogate’s optimality gap and can lead to faster GEP solve time. Cut management has the largest effect: the proposed K-means clustered multi-cut strategy achieves the largest runtime improvement in the tested setting by preserving timestep-level dual information while limiting master-problem growth. Combining the three improvements, ML-enhanced Benders solves GEP 3 times faster than both exact Benders and a direct solver on the tested instances. Overall, the thesis shows that accelerating MLenhanced Benders requires improving the surrogate’s integration across the pipeline: training-data alignment, dual learning, and cut management, with cut management the dominant contributor at the tested scale.
... Read more

Planning large-scale energy system investments requires solving optimization models over long time horizons, which quickly becomes computationally intractable, due to the problem size. The standard solution is time series aggregation: replacing the full time series with a smaller set of representative periods through clustering. Standard clustering methods select for typical behaviour, so extreme conditions are under-represented and the resulting investment decisions are insufficient to meet demand. Appending an artificial worst-case period corrects this underestimation but pushes the solution to an upper bound where investments become too expensive. This paper investigates whether adding real best-case periods selected from historical data can reduce this gap. Four selection strategies are evaluated — maximum availability, minimum demand, maximum availability-to-demand ratio, and maximum demand-adjusted availability — combined with the upper bound model using two aggregation methods. Results show that all four strategies plateau at similar regret regardless of strategy, number of representative periods, or integration method, indistinguishable from the artificial worst-case baseline. When k-medoids coverage is removed, best-case selection does improve on the worst-case baseline, but performs no better than randomly selected days, indicating that the improvement is not driven by the best-case nature of the selected periods. These results suggest that best-case period selection is most likely not an effective complement to the worst-case approach. ... Read more

The transition toward sustainable and reliable energy systems requires long-term investment and operational planning. However, accurately capturing the variability of renewable energy sources and the growing complexity of power systems requires large-scale energy system optimization models (ESOMs) with high temporal resolution, which are computationally demanding and often intractable to solve. Clustering-based representative periods (RPs) are widely used alongside ESOMs to improve tractability of the problem. Yet, as clustering optimizes for typical conditions, extreme periods are often underrepresented, giving reason to include extreme representatives so that feasibility is achieved. In this paper, we investigate the combination of typical and worst-case representative periods, with a particular focus on the effect of dynamically assigning weights to worstcase representatives. We consider two worst-case inclusion strategies: a Global approach, which adds a single worst-case representative constructed from the full dataset, and a proposed Local approach, which constructs a worst-case representative within each cluster. These strategies are combined with standard Dirac weighting and blended weight variants that allow original periods to be represented as weighted combinations of representatives. The Local approach consistently outperforms the Global approach by producing less conservative worst-case RPs. However, blended weights provide only some improvements over standard Dirac weighting, which do not hold consistently across configurations in the studied case, suggesting they should not be applied blindly. Overall, the findings support the use of cluster-level worst-case representatives while showing the limited practical benefit of more complex blended weight approaches.
... Read more

Large-scale energy system planning requires solving capacity expansion problems
over complete hourly time series, which is computationally intractable. Selecting a
small set of representative periods compresses the input, but standard clustering meth-
ods miss rare extreme events and smooth over many local peaks, producing investment
plans that underestimate the true optimal cost. A construction from Elgersma [5]
addresses this by building an artificial worst-case period, but the result can be more
extreme than any historically observed day. This work investigates whether selecting a
real observed period to represent the worst case can match this quality while preserv-
ing the temporal coherence of each selected day, meaning its demand and availability
values reflect conditions that genuinely occurred together in the same 24-hour window.
Three real-period selection strategies are proposed and compared against two artifi-
cial worst-case variants and the standard k-medoids baseline using the Tulipa Energy
Model [4]. All non-fractional methods approach near-zero regret by approximately
k = 400 periods; no single method is consistently fastest, and the differences between
methods become negligible beyond that point. No real worst-case method consistently
improves reliability over plain k-medoids, but this result is explained by a dataset cal-
ibration issue rather than a fundamental failure of the approach. Two lessons emerge
regardless of the dataset: global weight scaling in the fractional-weight variant intro-
duces a persistent non-zero regret plateau that cannot be corrected by adding more
periods, and model-guided period selection doubles computation cost without benefit
when the planning model is not sensitive to reliability shortfalls. The weight-scaling
issue is structural, and the computational cost of the model-guided method follows
directly from its two-solve design. A realistically calibrated dataset is still needed to
fully evaluate the reliability benefit of real worst-case selection. ... Read more

Capacity-expansion energy models are vital because they help policy-makers make better decisions when investing into the energy grid, thus leading to more up-time, greener energy and possibly more funds being injected in other sectors. These models require temporal data that makes full-resolution optimisation computationally
intractable. Time series aggregation reduces this burden by replacing the full year with a small set of representative periods, but standard clustering methods produce investment decisions that under-estimate true system costs. Recent work tightens this gap by injecting a global artificial worst-case representative day, but its fixed weight leads to unnecessary over-investment.

This paper proposes per-cluster worst-case augmentation: instead of a single global worst-case, each cluster receives its own artificial worst-case day injected alongside the cluster’s medoid. Three methods are explored: a distance-driven reassign- ment, a fixed-fraction weight split, as well as a se- lective cluster split. Per-cluster split reduces mean regret 2-4x compared to the global benchmark, at the cost of higher sensitivity to k-medoids initialisation as well as slower solves. A selective cluster split only adds artificial periods to a limited amount of clusters, covering for the second method’s weaknesses while maintaining good performance. ... Read more

Planning a long-term energy system relies on models that simulate system operation over many years at an hourly level, which is computationally expensive. A common remedy is temporal aggregation: grouping similar time periods and representing each group by one typical period to shrink the dataset the model must process. This speeds up the computation but tends to average away rare yet demanding conditions, such as days with high energy demand and little energy availability. These extreme periods, however, often determine how much capacity the system requires. This paper introduces three adaptations of widely used clustering algorithms that deliberately embed synthetic worst-case periods into the clustering process, ensuring the representative periods do not ignore the most demanding conditions. We evaluate them against four standard baselines (K-Means, K-Medoids, K-Medoids WC (worst-case), and Hull clustering) by measuring how closely each method's investment decisions match those of a benchmark model that uses the full, unaggregated data: a gap we call relative regret. The standard methods often require a large number of representative periods to approach the benchmark, whereas the proposed worst-case method WCA-K-Means reaches near-benchmark decisions with far fewer periods. By capturing the conditions that drive capacity needs without partitioning the data into excessive detail, it represents a full year with a much smaller dataset, giving planners results that closely match a full-resolution model while substantially reducing the computational cost of solving the energy model.  ... Read more

Optimization models are widely used in energy system planning to identify cost-effective investment strategies. However, relying solely on a single optimal solution can be misleading, as it fails to account for model uncertainty, competing objectives, and stakeholder preferences. To address this, near-optimal alternatives, solutions that are close in cost to the optimum but structurally different, are increasingly used to support robust and flexible decision-making.

This thesis explores the generation and evaluation of near-optimal alternatives within energy systems, with a focus on improving the decision relevance of the generated alternatives. This thesis introduces a unified analytical framework, formalizing existing Modeling to Generate Alternatives (MGA) methods using weight vector formulations. This formulation enables a clearer comparison of different techniques that generate these alternatives. This analysis highlights the limitations of current evaluation metrics, particularly their inability to distinguish decision-relevant alternatives from decision-irrelevant ones.

To overcome this gap, the thesis proposes a novel evaluation metric based on dominance relations from multi-objective optimization. This metric identifies non-dominated alternatives, those not strictly worse than any other across all decision variables, as decision-relevant. The thesis introduces a new method that uses Directionally Weighted Variables to generate alternatives aligned with this dominance criterion.

The proposed approach is evaluated using a stylized energy investment model and benchmarked against existing MGA techniques. Results show that traditional methods tend to generate fewer non-dominated alternatives, while the new method generates more non-dominated alternatives within the near-optimal space. This work contributes a new perspective on alternative generation, bridging the gap between mathematical optimality and practical decision support. ... Read more

This paper extends the Tulipa energy system optimisation model by incorporating start-up and shut-down capability constraints formulated for Tulipa's fully flexible temporal resolution. The impact of adding these constraints for thermal generators is assessed using a greenfield case study with 7 European countries. Results show that including these constraints increases computation time, but they more realistically represent generator behaviour, which also results in a higher objective function value. Cases where the resolution of assets is not a multiple of the resolution of flows result in uniquely long solving times. The investments, as well as the unit operation trends remain similar on a high level. Batteries are utilised to improve the reduced flexibility, and units with the most flexible start-up/shut-down capabilities become used slightly more often, while the opposite holds for those with the least flexible capabilities. Units also tend to be turned on and off less often. This research contributes to understanding the trade-offs between model complexity and runtime in long-term energy planning. ... Read more

The transition of the energy grid into a system with higher shares of renewable energy production requires careful investment planning while considering operational characteristics of generators. Generation Expansion Planning (GEP) is used for finding optimal investments, while Unit Commitment (UC) can be used to limit generator operational capabilities, such as via ramping limits, for more accurate modelling. The system may be extended to model custom thermal generator capabilities at the time of their start-up or shut-down, where they may differ from traditional ramping, promoting more precise modelling. However, the inclusion of such detail requires careful managing of model complexity to keep solving time feasible, which can be done with techniques such as Clustered Unit Commitment (CUC), and clustering time blocks while allowing flexible resolution combinations. The latter is known as Fully Flexible Temporal Resolution, and promises managing of model complexity via temporal resolution reductions, while maintaining modelling versatility. The paper targets the unexplored area of including Start-Up and Shut-Down (SU/SD) capability ramping limits alongside a fully flexible temporal resolution in a GEP & CUC model, and contributes by examining the effect of the new capabilities on the run times, investment and operational solutions, and the total system cost of a model with enabled Battery Energy Storage System (BESS) investments. The resulting findings show that the inclusion of the SU/SD capabilities has little effect on the investments and total cost of the model, significantly increases computation time, yet has a noticeable effect on the operational schedule of generators. ... Read more

In recent literature fully flexible temporal resolutions have been proposed as a new form of temporal clustering in generation expansion planning models, showing promising benefits in terms of the tradeoff between solution accuracy and computation time. However, unit commitment constraints such as minimum up and down times have not yet been considered in combination with these resolutions. This paper introduces minimum up and down time (MU/MD) constraints to fully flexible time resolutions and shows the effects of including them when doing generation expansion planning. This is done by constructing a case study based on the European energy grid and comparing the effects of adding MU/MD constraints to a model with a fully flexible time resolution in the form of the geographically decreasing resolution. The paper shows that the addition of minimum up and down time constraints maintains the benefits of fully flexible time resolutions, but comes with additional computation time and also has little effect on the optimal solution cost for the case study examined. ... Read more

This research explores the integration of start-up and shut-down trajectory constraints into the Tulipa energy system optimisation model, which uses fully-flexible temporal resolution. These constraints aim to more realistically represent the behaviour of large thermal generators during operation. Case studies with varying time resolutions and generator configurations were evaluated to measure impacts on computation time and solution accuracy. Results show a significant increase in computation time with small gains in accuracy, compared to introduction of minimal down-time constraints. ... Read more

Fully flexible temporal resolutions have shown to be a useful tool for improving the tradeoff between the runtime and accuracy of generation expansion planning models. However, no research has been done into the effects of considering short-term operational dynamics of thermal generators in models with such resolutions. Therefore, this paper complements the existing literature by adding start-up and shut-down costs to a large-scale energy system optimisation model with a fully flexible temporal resolution. The results suggest that the addition of start-up and shut-down costs significantly increases the runtime of the model, while providing a small increase in accuracy. Additionally, they show that a compact set of start-up and shut-down constraints outperforms a full set of constraints in terms of model runtime, while having the same accuracy. ... Read more

The energy transition is one of the major challenges of the 21st century, impacting the way energy is generated, conserved and consumed. Energy generation becomes more and more decentralised, intermittency and fluctuations suddenly are becoming topics of interest within day-to-day life and energy system operators are facing many new obstacles never encountered before. In this context, the anticipation for hydrogen as a resource for energy conservation and -management is big. This research focuses on the optimisation of the hydrogen pathway for investment and operational models.

The hydrogen pathway as aforementioned is divided over three technologies: hydrogen generation with means of water electrolysis, also known as 'green hydrogen', storage in compression vessels and reconversion of hydrogen into electricity in the form of a fuel cell technology (also known as Power-to-Gas).

The research focuses on identifying technical parameters and operational policies of the water electrolysis systems that can be translated into optimisation constraints, assessing the level of detail required to create an accurate optimisation model. A generic model is developed that can be scaled for further research, making different case studies and sizing possible. The research compares the performance of the models in terms of accuracy to the computational burden. The comparison is done for the level of detail and complexity added to the model.

After a literature review of technical parameters and operational policies regarding the technologies, two models were created in a mathematical framework. The two models proposed were Linear Programming (LP) and a Mixed-Integer Programming (MIP) Model. On the LP model 6 different sensitivity analysis has been performed, to be precise on Capital Expenditures (CAPEX), efficiency, lifetime, ramping rates, interest rates and finally different time horizons. The outcome of these analyses is that the technology mix can best be used in a combined manner, whereby each component of the mix contributes towards minimising the objective value: the Total Annualised Cost.

Lastly the two models are compared with different types of configurations, each with a different set of constraints. The constraints to be modelled were: minimum uptime and downtime, start-up costs, degradation due to cycling and finally the part-load operation. ... Read more

With the decision made to act upon climate change, the remaining question is: "How?". Economic theory suggests that the most efficient method is by means of market-based policies. These policies are often designed based on Integrated Assessment Models like DICE, which is the subject of this thesis. Integrating uncertainty into this model has been the aim of the thesis. The stochastic version of DICE, EICE, showed to be sensitive to the distribution of climate uncertainty, but results were less explicit then expected. Overall, the conclusion is that both modest and more strict mitigation policies can be justified based on EICE. ... Read more

Increased use of the distribution grid due to the uptake of distributed energy resources and the expected penetrations of Electrical Vehicles (EVs) could lead to congestion problems in the distribution grid. Congestion refers to issues related to the overheating of components or voltage issues in the distribution network. Avoiding these issues is crucial to maintain a stable, economical and reliable electricity grid. By using the flexibility of aggregated EVs large investments in grid reinforcement can be avoided. However, a holistic approach is necessary to manage the procurement of flexibility services for all stakeholders involved. One approach is the Universal Smart Energy Framework (USEF), a framework that integrates the existing electricity market with a market for flexibility services from the aggregator to the Distribution System Operator (DSO). This master thesis presents a study on the flexibility market as described by USEF from the perspective of a commercial aggregator of EVs. USEF presents a framework in which flexibility potentially provides financial opportunities for aggregators of EVs. However, it is not clear what the financial impact on the charging costs of an aggregator of EVs is and in what way an aggregator has to adapt its charging logic when trading with DSOs. Therefore, the aim of this thesis is to answer the following question: how can an aggregator of EVs offer flexibility services to a USEF compliant market at distribution level? This report presents an in-depth analysis on USEF, determines the impact of network constraints from USEF on the charging costs of an aggregator of EVs and improves the charging strategy under USEF constraints. ... Read more