A Crisis-Proof Electrical Power System: Desirable Characteristics and Investment Decision Support Approaches

Abstract

Electricity expansion planning is inherently subject to uncertainty, shaped by climatic, regulatory, and economic risks. In Brazil, this challenge is compounded by recurrent crises that have repeatedly reduced electricity demand. This study proposes a complementary decision-support approach to make planning more resilient to such crises. Using Brazil’s official optimization models (NEWAVE), we introduce two analytical elements: (i) a regret-minimization screen for choosing between conservative and optimistic demand trajectories and (ii) a flexibility stress test that evaluates the cost impact of compulsory-dispatch shares in generation portfolios. Key findings show that conservative demand projections systematically minimize consumer-cost regret when crises occur, while portfolios with lower compulsory-dispatch shares reduce total system cost and improve adaptability across 2000 hydro inflow scenarios. These results highlight that crisis-robust planning requires combining cautious demand assumptions with flexible supply portfolios. Although grounded in the Brazilian context, the methodological contributions are generalizable and provide practical guidance for other electricity markets facing deep and recurrent uncertainty.

1. Introduction

Electricity expansion planning has always been carried out under uncertainty, typically accounting for risks such as climatic variability, environmental constraints, regulatory changes, and economic factors, including exchange rate fluctuations. Although the COVID-19 pandemic was unprecedented in nature, abrupt and unforeseen reductions in electricity demand resulting from crises of various origins have become increasingly frequent. In Brazil, notable examples include electricity rationing (2001–2002), the global financial crisis (2008–2009), government intervention in the electricity sector combined with water scarcity (2012), and the presidential impeachment (2016). These recurrent disruptions motivate two research questions: (1) Given recurrent demand-contraction crises, does adopting conservative demand projections minimize consumer-cost regret relative to optimistic projections? (2) How does the portfolio’s compulsory-dispatch share (a proxy for supply rigidity) change consumer total cost under demand and hydrological shocks? To address this, we propose a complementary expansion planning approach explicitly designed for contexts of repeated recessions.

The article first reviews the history of crises in the Brazilian electricity sector (BES), identifying their drivers and assessing the effectiveness of measures adopted in response. This retrospective analysis seeks to determine whether lessons learned can be systematically incorporated into planning processes. Building on this, we propose an analytical framework to support investment decisions in both generation expansion planning (GEP) and transmission expansion planning (TEP). The framework is tested through simulations using official optimization models employed by the Energy Research Office (EPE, Empresa de Pesquisa Energética), the federal agency responsible for long-term planning studies. The methodology and results of these applications are presented and discussed.

Long-term generation and transmission expansion planning is inherently a large-scale, complex optimization problem. Its purpose is to define the least-cost portfolio of projects capable of meeting demand with acceptable reliability, including the selection of technologies, capacities, locations, and commissioning schedules, all under uncertainty. A wide range of optimization methods has been developed for this purpose. Koltsaklis and Dagoumas [1], in a comprehensive review, highlighted recent methodological advances in GEP, including integration with TEP and natural gas systems. Sadeghi et al. [2] similarly reviewed GEP strategies under different industry structures. Integrated GEP–TEP modeling has also been examined in works by Alizadeh and Jadid [3], Eghbali et al. [4], and Guerra et al. [5], given the challenge of reconciling high mathematical complexity with economic efficiency in systems with significant renewable penetration. Luz et al. [6], for instance, proposed a multi-objective extension to EPE’s methodology by incorporating renewable maximization and system security criteria.

The literature frequently introduces methodological advances to represent new resources and technologies, including distributed generation, electric vehicles, demand response, storage, and offshore wind. Approaches also address uncertainty in demand, water and wind availability, and fuel prices (Seddighi and Ahmadi-Javid [7]). However, little attention has been devoted to modeling adverse scenarios—low-probability but high-impact events that trigger crises in the BES. This gap constitutes the core focus of this study.

Objectives. We developed and tested a complementary decision-support approach for long-term expansion planning. Specifically, we (i) evaluated whether conservative demand projections minimize ex-post-consumer cost (regret) when crises recur and (ii) quantified how supply-side rigidity (compulsory dispatch shares) affects total cost under demand and hydrological shocks. The framework interfaces with standard GEP/TEP studies and is intended to be used alongside the least-cost optimization now in place.

This paper (1) proposes a regret-minimization screen for choosing conservative vs. optimistic demand trajectories in expansion planning under recurrent crises; (2) introduces a flexibility stress test that varies compulsory-dispatch shares and measures consumer-cost impacts; and (3) demonstrates both with Brazil’s official planning model (NEWAVE), yielding implementable guidance for planners.

To address these challenges and enrich current planning practices, Section 2 reviews crisis events in the BES, while Section 3 examines the international literature to identify lessons from other countries. Section 4 presents the proposed analytical complementary approaches, followed by Section 5, which discusses the results of applying them to planning simulations. Section 6 concludes with the main findings and outlines opportunities for future research.

2. Contextualization

Birol [8] highlighted an important aspect when examining the COVID-19 pandemic and its consequences for electricity demand. He emphasized how shifts in consumer behavior have reinforced the indispensability of electricity, driven by the widespread adoption of telecommuting, the surge in e-commerce activity, and the growing reliance on streaming platforms for entertainment. This global trend may influence consumer perceptions of electricity systems, leading to increasingly stringent expectations for reliability.

However, changes in electricity consumption behavior are not exclusive to the pandemic context. Crises can emerge from multiple origins—economic, political, social, or environmental—and their impacts are not uniform across sectors. For instance, during COVID-19, electricity demand contracted significantly in the industrial and commercial segments, while residential demand increased. Natural disasters, such as droughts or floods, can disrupt supply chains and infrastructure. Political and geopolitical crises can trigger fuel price shocks, contractual disputes, or infrastructure delays. These differentiated impacts underscore the importance of modeling crises as multi-faceted phenomena that affect not only aggregate demand but also consumer classes, supply reliability, and contracting structures.

The Brazilian electricity sector (BES) has repeatedly faced crises triggered by both domestic and international events. Experiences abroad also suggest that sectoral shifts and policy responses are key to resilience. Over the past two decades, at least five major crises have affected the BES—an average of one every four years. This recurring pattern underscores the need for expansion planning approaches that explicitly incorporate the possibility of repeated crises and shifting consumer expectations, ensuring that investment decisions are robust and resilient under conditions of deep uncertainty.

• 2001–2002: Energy Rationing: 12% to 14% drop in electricity demand. (The percentage drop in consumption for each episode is calculated by comparing the consumption verified in the SIN with that projected in the year prior to the event. For the 2001/2022 episode, exceptionally, as there were no demand forecasts available for the period, consumption was estimated based on the average percentage change in consumption verified in the previous five years (1996 to 2000))

The structural imbalance between supply and demand in the system, boosted by an unfavorable hydrological situation, led to rationing. The consequence was a sudden and abrupt drop in demand, between 12% and 14% of the country’s total consumption (with the residential sector reaching the stipulated target by the government of 20% off), with an impact on all economy. This also stimulated an increase in consumption efficiency as part of the rationing plan [9] drawn up by the Energy Crisis Management Chamber (CGE in Brazilian acronym) set up by the federal government. For the regulated market, the plan had four points: (i) consumption target; (ii) demand response by extra price (made more flexible at a later time); (iii) risk of power outage; and (iv) wide mobilization of the population through intensive communication. For the free market, it was decided to seek market solutions, respecting contracts and creating incentives. As pointed out by Campos [10], by bringing up the history of evolution of the electricity sector, CGE also adopted initiatives aimed at improving the market model adopted in sector reform. This included creating a Committee for Revitalization of the Electric Sector Model to seek solutions for attracting private capital for expansion and designing a competitive power generation market.

• 2008–2009: International financial crisis: 2% to 8% drop in electricity demand.

The international financial crisis hit Brazil’s economy, and with a gross domestic product (GDP) reduction of 0.3%, the impact on BES was mainly due to an 8% reduction in industrial sector demand. As it was an international crisis with an impact on all sectors of the economy, mitigation measures adopted by the Brazilian government were aimed at the economy as a whole. This included reducing value-added tax (VAT) on industrialized products with the aim of stimulating demand, mainly final consumption, and thus restoring production and employment levels in Brazil’s economy [11]. As industrial segment demand is concentrated in the free market, its performance was fundamental in seeking solutions through negotiation.

• 2012–2014: Government price intervention and unfavorable hydrological situation: 1% to 4% drop in electricity demand.

During this period, the largest state-owned generators (both federal and state) sold almost all of their energy to distribution companies. These contracts, which were the main source of supply for the distribution companies, ended in 2012 and 2013. Additionally, the electricity was generated by hydroelectric power plants whose concessions were set to expire during that decade.

In view of the 2014 presidential elections, the federal government proposed renewing the concessions of those hydroelectric power plants that agreed to sell their electricity at regulated and fixed prices to distribution companies. The goal was to reduce the electricity bill of captive electricity consumers by 20%.

However, only federal state-owned companies accepted the proposal. As a result, the Power Purchase Agreements (PPAs) between distribution companies and state-owned ones ended, leaving the former short on the market and having to buy electricity from the short-term market.

Furthermore, the country began a period of water scarcity, causing short-term market prices to increase from BRL 183/MWh in September 2012 to BRL 414/MWh in January 2013, reaching BRL 822/MWh in February 2014. Since distribution companies were involuntarily short (they are not allowed to buy electricity in the free market but only from government auctions), the cost of electricity purchased in the short-term market was passed on to captive consumers. This led to a price explosion for captive consumers (contrary to the original objective), resulting in a 4% drop in demand.

• 2015–2016: Impeachment of President Dilma Rousseff and political-economic crisis: 4% to 8% drop in electricity demand.

The political-economic crisis that emerged in Brazil during the impeachment of President Dilma Rousseff further aggravated the situation in the electricity sector, which was still suffering from the previous crisis. The industrial sector was most impacted, with its electricity demand in 2016 being 7.7% lower than in 2014. Brandão et al. [12] quantified and evaluated the economic performance of distribution companies in Brazil and found an adverse effect on the market in the years 2009 and 2014 to 2016, likely due to these economic crises.

• 2020: COVID-19—11% reduction in electricity consumption in May 2020, representing a 5% drop compared to 2019.

In the first wave of the health crisis in 2020, electricity consumption in Brazil fell by approximately 6.6% (April 2020) in just 50 days after the World Health Organization (WHO) declared a coronavirus pandemic. This was compared to the same period in 2019 due to the restrictive measures imposed on the population to contain the spread of the virus. As shown in Table 1 [13], this drop was concentrated in the industrial and commercial consumer classes (Only in the residential class there was an increase in consumption during this period, possibly due to the confinement of people at home, which led to a significant increase in domestic activities as well as professional activities, which until then were carried out in the companies’ offices). In May 2020, the reduction was even more significant, reaching 11% (36 TWh) [14], which corresponded to a reduction of 6.8 average GW in the total generation of the National Interconnected System—NIS [15]. The result was an average surplus of 9.3 GW of energy in the system this month of May—distributors were over-contracted by 23.8%, and prices in the short-term market fell by approximately 50% compared to May 2019 [16] (in April, the drop was 78%, around BRL 140/MWh in the southeastern and southern submarkets). On the other hand, this reduction in prices in the short-term electricity market was not noticed by consumers in the regulated electricity market, which even experienced an average increase of 3.2% [17] throughout 2020 (even after governmental measures tariff relief [18]). One of the reasons is that part of the energy already contracted by distributors (long-term contracts, up to 30 years duration) is not needed in the short-medium term due to the drop in consumption but should be paid anyway.

Table 1. Electricity consumption (GWh) in April 2019 and April 2020, by consumption class.

Consumption (GWh)	April 2020	April 2019	Difference (%)
Brazil	37,116	39,718	−6.6
Residential	12,308	11,610	6.0
Industrial	12,173	13,899	−12.4
Commercial	6393	7782	−17.9
Others	6243	6428	−2.9

Self-elaboration. SOURCE: Monthly Review of the Electric Energy Market. EPE, May 2020 [14].

The impact of these crises on electricity consumption can be better illustrated by Figure 1 (The hourly consumption database for Brazil has been available since 2004. Prior to this period, consumption data were presented as monthly averages [19]). After rationing in 2001, residential consumption levels remained below pre-2001 levels for several months. One reason is a change in consumption behavior, as highlighted by Birol [9] in his assessment of how crises can have far-reaching consequences. More specifically, energy efficiency policies were disseminated that encouraged more rational use of energy by the population, as analyzed by Carreno et al. [20].

Figure 1. Monthly Electric Energy Consumption on the Grid (TWh) from 1998 to 2022 for the Residential, Industrial and Commercial classes. Self-elaboration. Source: Monthly Electricity Consumption by Class (regions and subsystems). EPE, 2024 [20].

Regarding the industrial sector, consumption has remained practically the same over a decade. This is because as soon as it starts to recover from one crisis, another one occurs. About commercial consumers behavior, the effects of rationing are similar to the other two sectors, but this sector suffered the most from the COVID-19 crisis. The various economic and political crises also led to a stagnation of consumption growth for almost a decade.

Given this, this article questions the assumption that the Brazilian economy and therefore electricity demand will always grow as considered by the government for electricity expansion planning. Starting from the hypothesis that crises affecting the electricity sector are frequent and impact the generation and energy transmission expansion planning process, this article aims to propose improvements in expansion planning to make the electricity sector less susceptible to frequent and severe downturns in consumption. This is achieved by reviewing projections of variables such as electricity demand and scheduling forward procurement electricity auctions.

The authors believe that this topic is relevant, as suggested by Yin [21]. He denied a certain public assumption and presented a motivation for research based on “distrust” that the methodologies adopted in expansion planning studies can adequately represent crisis scenarios with an impact on demand reduction. The significant impact of these crises on the sector requires different approaches to deal with them in the field of expansion planning.

It is worth emphasizing that the Brazilian electricity market is structurally divided into regulated and liberalized segments. Within the free market, transactions are restricted to energy products, leading market participants to favor the lowest-cost alternatives—principally wind and solar in the Brazilian context. Capacity contracting, formerly centralized within the regulated environment to guarantee uniform reliability, has since 2021 been undertaken through capacity reserve auctions. Under this mechanism, costs are socialized via a tariff charge allocated across all consumers. This article analyzes the economic implications of contracting capacity under this framework.

3. Literature Review

Considering the crises reported in the previous section and their impacts on the electricity sector, it is necessary to plan for the possibility of future crises and mitigate their impacts. These events, whether domestic or international, such as the 2008–2009 crisis caused by the bankruptcy of the giant American bank Lehman Brothers, can affect the electricity sector on both the supply and demand sides. However, of the five events that triggered crises in the BES this century, four had a negative impact on the growth of electricity demand.

Given that new events may occur in the coming decades, it is worth assessing the need to review processes involving expansion planning, regulation, commercialization, and operation of the BES. This would allow for mapping actions that could be adapted to sustain the market and enable the government to act preventively to mitigate the risk of unfavorable scenarios in the near future.

Because four of the five major crises in Brazil this century depressed electricity demand, the key planning gap concerns downside demand risk and its tariff pass-through. Internationally, crisis-era studies often treat uncertainty on both the demand and supply sides; however, few explicitly prioritize recurrent demand contractions and their interaction with rigid supply commitments in expansion studies. By reviewing how other systems model crisis uncertainty and decision criteria, we position our contribution as a regret-based screen on demand projections and a flexibility stress test for supply portfolios.

In Japan, after the earthquake, tsunami, and nuclear disaster of March 2011, there was a drastic reduction in electricity generation capacity due to the withdrawal of nuclear power from operation. Fujimi and Chang [22] found that adaptations by private companies to the electricity shortage persisted after the crisis. These adaptations were classified into three types: (i) behavioral, with reduced use of air conditioning and lighting; (ii) schedule, with manufacturing companies shifting production to off-peak hours; and (iii) hardware, with the installation of more efficient devices. Apart from schedule adaptation, they noted that others persisted into the second post-crisis year.

The electricity sector in Ireland has undergone many transformations over 100 years between 1916 and 2015 [23]. These aimed to improve security of supply through diversification of generation portfolio and promotion of renewable energy sources, particularly wind energy. The transition was marked by market reforms such as liberalization and regulation, fragmentation into regions, and creation of a single electricity market (SEM). The country recognizes the importance of crisis events and learning from them, which has driven development and implementation of energy and climate policies that played a fundamental role in these transformations.

In the United States, the COVID-19 pandemic led to a reduction in electricity demand on regional transmission organizations (RTOs). According to Eryilmaz et al. [24], rapid changes in demand differently affected fuel sources for marginal generators. In some RTOs, significant changes were observed in base load and peak load as well as changes in electricity generation mix. Note that [23,24,25] are adduced for crisis impacts and adaptation patterns, informing our scenario design and evaluation criteria not as GEP models per se.

Expansion planning activities play a fundamental role in anticipating future scenarios and addressing the uncertainties inherent in their processes. They provide key information, solutions, and elements to aid the government’s decision making in facing crises in the electricity sector. A series of questions arise associated with these frequent episodes of reduced demand that could be evaluated from a medium- and long-term planning perspective. Among them are the costs to the electricity sector of frequent “course corrections” when faced with atypical scenarios not considered or given due importance in planning. It is also questioned whether these costs justify changes or improvements in planning processes to increase their credibility. These questions demonstrate the practical usefulness of this article’s contributions to the Brazilian electricity sector, where electricity consumers bear a significant part of the costs that could be avoided by viewing the phenomenon differently.

This article presents two types of simulations: “crises scenarios” and “business as usual”:

• Crises’ scenarios: These are atypical situations motivated by various reasons (economic, political, environmental, etc.) that have a high potential for interference in the management of the electricity sector and may generate harmful consequences affecting various levels of the supply chain. The variables that make up this scenario are reasons, impacts, measures and mitigations, effectiveness, and lessons learned. Some of these variables can be measured quantitatively, while others are qualitative;
• Business as usual: This is a central and strategic activity in the management of the electricity sector that guides new investment decisions in generation and transmission assets to ensure supply adequacy while considering cost and quality dimensions. The variables associated with this scenario are operating cost, investment cost, and adequacy supply criteria. All these variables must be measured quantitatively through simulations based on mathematical optimization models. The evaluation of these variables will be based on measures of scales already validated in the planning activities literature but may need to be adapted due to different characteristics of electricity matrices and markets.

The relationship between these scenarios highlights the importance of identifying desirable characteristics in power systems so that they can face unexpected demand reductions with minimal impact on consumers. This contributes to signaling necessary planning actions for constructing a more robust electrical matrix and designing appropriate products for sale in the electricity market.

In contrast to the surveyed approaches, our contribution is twofold: (i) explicitly treating recurrent demand-contraction crises as a distinct uncertainty category and applying regret minimization to demand-trajectory selection and (ii) introducing a flexibility stress test via compulsory-dispatch shares. While prior studies address resource variability and supply adequacy, few highlight consumer-cost regret from persistent demand downturns. This distinction positions our work as a complementary addition to the existing literature.

Recent studies also highlight advanced risk-aware frameworks, such as holistic bilevel optimization models for expansion under deep uncertainty [25]. These approaches strengthen the representation of multi-actor decision layers under risk. Our contribution complements this stream by focusing on the specific challenge of recurrent demand-contraction crises and translating insights into simple decision-support tools usable with official planning models.

4. Proposed Analytical Complementary Approach

Despite Brazil being a developing country with high electricity demand growth forecasts, frequent crises make it difficult to achieve these projections and often result in downward revisions. Figure 2 illustrates the projection made in December 2011(blue line) and some revisions made until April 2019 (gray, green and orange lines) [26]. There is a difference of 165 TWh/year between the demand forecasts for the year 2023 and that carried out in December 2011 and May 2019 (red dashed line). Given this, it is worth assessing the most appropriate way to consider the uncertainty of these demand forecasts in the investment decision process to reduce decision regret.

Figure 2. Differences between electricity demand projections from 2012 to 2023—Realized demand × Expected demand (GWh). Self-elaboration. Source: ONS, 2019 [26].

Brazil is also known for its rich natural resources for electricity generation, but this beneficial characteristic brings greater complexity to the expansion planning process. There are many possibilities for the composition of the electricity matrix, which must be carefully analyzed considering safety, economic, and socio-environmental aspects. It is necessary to assess whether there are characteristics of the system that could be desirable in mitigating the impact on the electricity market in crisis scenarios with negative impact on demand.

In this study, the perspective of electricity consumers is analyzed to assess the impacts associated with investment decisions during unexpected demand-reduction events. Electricity tariffs are reviewed annually and consist of portions reflecting costs of fuel use, expansion contracting, transport (transmission and distribution), and taxes. As the supply contracted was based on planning that considered higher demand growth than realized, part of the investment made in system expansion would no longer be necessary. Consumers bear this frustrated investment, but savings are expected in terms of system operation costs. Since investment decisions are based on optimization of investment and operating costs, the system planner can assess tariff impact through quantitative analysis of these portions.

Our modeling represents crises through negative GDP shocks, which serve as a tractable proxy for recessions and demand contractions. However, crises can take many forms beyond GDP downturns. The points below summarize a typology of crises, mapping their likely impacts on electricity demand and supply:

• Economic and financial crises: broad-based demand decline, especially in industry;
• Health crises (e.g., COVID-19): demand contraction in industry and commerce but residential increases;
• Natural disasters (e.g., droughts and floods): infrastructure disruption and hydrological scarcity;
• Political/geopolitical crises: fuel price volatility, contract disputes, and delays in expansion.

While this paper operationalizes crises through GDP-demand shocks for tractability, the analytical framework is flexible enough to incorporate differentiated sectoral or supply-side shocks, making it adaptable to a wide range of adverse conditions.

The first set of analyses focuses on the supply side, examining the degree of compulsory generation of the system, primarily from natural gas-fired power plants (For this exercise, any compulsory generation energy source with a non-zero-unit variable cost could have been considered, but gas-fired power plants were chosen for simplicity of modeling.). This study concentrates on gas-fired thermoelectric generation, which has emerged as the principal technology prioritized by the government in contracting mechanisms aimed at securing electricity supply. Although natural gas is used here as a proxy for inflexible generation, the same logic applies to other rigid technologies such as nuclear or long-term “take-or-pay” contracts. The second set of analyses evaluates demand projections under uncertainty, applying decision-making methods that emphasize minimizing regret. Together, these approaches allow planners to assess how system rigidity and forecasting assumptions interact to shape consumer costs in crisis contexts

The simulation considers as a reference the long-term plan called Ten-Year Energy Expansion Plan 2029—PDE 2029 [27] by EPE (Share of sources in the installed capacity of centralized generation estimated for 2029: hydroelectric—49%; thermoelectric—18%; other sources (wind, solar photovoltaic, biomass, and small hydroelectric plants)—33%.). It adopts as a premise that the year in progress is 2026, and all generation expansion between 2026 and 2029 would have already been contracted, as illustrated in Figure 3.

Figure 3. Premises of the study: horizon of analysis. Self-elaboration.

Regarding the first analysis, the impact of this obligation on operating and investment costs (reflected in the charges paid by the consumers) will be assessed when a sharp drop in demand is observed. Different levels of participation of natural gas thermoelectric power plants were considered, with 50% compulsory annual generation—as proposed by the ministry, at an operating cost of USD 5.00/MMBTu. This is equivalent to a variable cost (VC) of BRL 193/MWh and an investment cost of BRL 5000/kW [28].

For this simulation, in 2026, the additional expansion of gas thermoelectric plants was set at 0 GW, 3 GW, 8 GW, and 12 GW, totaling four supply sensitivity scenarios. The increments of 3, 8, and 12 GW of gas-fired capacity correspond to policy-relevant magnitudes: 3 GW reflects a single large auction round, 8 GW is the is the amount provided for by law 14182 [29], and 12 GW represents the upper range of recent governmental discussions on capacity reserve auctions, enabling a clean identification of rigidity effects on consumer cost. These benchmarks ensure that results are interpretable in terms of actual policy choices These scenarios are, respectively, called “Sc. 0 GW Gas”; “Sc. 3 GW Gas”; “Sc. 8 GW Gas”; and “Sc. 12 GW Gas”. Figure 4 illustrates the variation in the level of thermal inflexibility of the system (compulsory generation) between these four scenarios in relation to the available thermoelectric power of the NIS (National Integrated Systems). This considers the PDE 2029 reference demand and the reduced demand due to COVID-19, with a negative variation of GDP in 2020 of −5%.

Figure 4. Inflexibility of thermal power plants, in each scenario (% of available thermoelectric power). Self-elaboration. (GDP = Gross domestic product).

The model used for this simulation was Newave [30], officially used by both the ESO (Electricity System Operator) and EPE. This model performs hydrothermal coordination for a medium-term horizon and aims to meet demand at the minimum operating cost. It considers several restrictions, risk aversion [31], and the uncertainty of water affluence to hydropower reservoirs represented through scenarios generated by periodic autoregressive models [32,33,34]. The monthly seasonality of sources not dispatched by the system operator, such as wind power and photovoltaics, can be represented in the model by signaling the amount of generation each month of the study period, which makes it possible to assess the effect of the complementarity of these resources with hydrology in the various regions of the country. To find the operation policy that indicates the dispatch of each power plant each month, stochastic dual dynamic programming [35] is used. Hydroelectric plants are represented in a grouped manner, in equivalent energy reservoirs per electrical-geographic region, given the large dimension of the NIS. The individualized representation of hydroelectric plants significantly increases the complexity of the optimization problem.

Table 2 summarizes the simulations conducted in this study to test the proposed analytical complementary approaches, totaling 12 simulations scenarios, combining the variables that are being sensitized, which are (i) additional amount of expansion of inflexible thermoelectric plants and (ii) electricity demand projection.

Table 2. Summary of simulations scenarios, combining supply, and demand variables.

	0 GW Gas	3 GW Gas	8 GW Gas	12 GW Gas
GDP + 2.7%
(Reference Demand)	Sc. (1)	Sc. (2)	Sc. (3)	Sc. (4)
GDP − 5.0%
(Low Demand)	Sc. (5)	Sc. (6)	Sc. (7)	Sc. (8)
GDP − 10.0%

Self-elaboration.

The following Section 5.1 and Section 5.2 present the results obtained with the application of the proposed analytical complementary approach considering uncertainty in the investment decision process (i) on the demand side and (ii) on the supply side when demand is lower than projected.

5. Results and Discussions

Standard studies already include hydrological uncertainties in NEWAVE. Our contribution is orthogonal: we (i) explicitly test demand-path choices with a regret criterion under recurrent contractions and (ii) isolate how compulsory-dispatch rigidity shifts consumer cost across hydrology and demand shocks

5.1. Demand-Side Analytical Approach

Decision-making methods under uncertainty can provide interesting insights and aid in decision making by simulating different combinations of scenarios in the investment decision stage with realized scenarios, known as “states of nature”, whose probabilities of occurrence are unknown.

For this simulation, two scenarios were considered for both dimensions—reference demand (demand projection by PDE 2029) and low demand—seeking to find the combination that would lead to lower regret. This is calculated as the difference between the best return and that obtained considering a given decision for each state of nature. The total system cost, represented by the sum of the operation cost and the investment cost, is used as an indicator. This analysis was carried out for each of the same previous four electricity matrix scenarios, each with different levels of participation of natural gas thermopower plants. The objective was to evaluate whether there would be persistence in the solution between them.

Figure 5 presents decision matrices for extreme scenarios—Scenario 1 and Scenario 4—also called result or payoff matrices commonly used by decision-making methods under uncertainty. Minimax regret is a decision criterion that compares the loss from not having chosen the best action once the state of the world is revealed. It identifies the strategy that minimizes the maximum possible regret, thus favoring robust decisions under uncertainty.

Figure 5. Decision matrix: Scenario 1—0 GW gas; Scenario 4—12 GW gas. Self-elaboration.

This matrix serves to organize information by combining investment alternatives and states of nature to help choose the best decision. Thus, for each scenario, combinations of investment decisions (planned supply) are tested resulting from simulations that considered high and low demand projections as input data with demand realizations called “states of nature”, high and low. For example, the third quadrant of Scenario 1 (total cost = BRL 87.34 billion) presents the result of a simulation that considers planned supply based on a low demand projection (red box) with high realized demand. Moreover, the value of BRL 6.85 is the difference between BRL 94.19 and BRL 87.34, as BRL 27.46 is the difference between BRL 82.90 and BRL 55.43.

For all supply composition profiles tested, two possible solutions of lower regret were identified when applying the “Minimax” theory—Loulou and Kanudia [36], Conde and Leal [37], and Yue et al. [38]. These solutions are similar in that they both consider the low demand scenario for the investment decision. It may seem intuitive to use more optimistic forecasts of growth in electricity demand during the planning process to reduce the risk of supply. In the worst case, this would only result in an anticipation of investments for generation and power transmission expansion. However, this anticipation can lead to significant frustration of investments, resulting in financial damages. To achieve a more efficient power system, it is essential to adopt criteria that balance both the risk of adequate supply and economic risk from planning to system operation.

Thus, both scenarios with no new natural gas power plants and with 12 GW of new plants with compulsory generation indicate that when the system is planned considering a low demand projection, even when a higher demand occurs (the demand is met), the lower regret is planning the system with the low demand projection.

Of course, this conclusion was reached for the case study presented in this manuscript, which considers a specific configuration and a certain growth rate variation between the simulated demand projection scenarios. Growth rates are the average CAGR of each trajectory over 2026–2029 relative to its own 2025 base level; similar CAGRs can coexist with different absolute levels due to the 2020 shock (base effect) and include the 2026 starting values in the Figure 6 caption. Therefore, extrapolating the conclusion to configurations with very different characteristics from the Brazilian electricity system should be done sparingly. Thus, the main message of this manuscript is that it rejects the hypothesis that considering more optimistic demand projections is desirable when carrying out studies to plan the expansion of electricity systems. It is important to test the best strategy, applying decision-making methods under uncertainty, such as the one presented in this section, which make it possible to weigh not only the risk of not meeting electricity demand but also the economic risk associated with the impact on electricity tariffs.

Figure 6. Demand projections in average MW for the demand scenarios: Reference, Low, and +Low. Self-elaboration.

5.2. Supply-Side Analytical Approach

This section seeks to understand which level of compulsory thermoelectric generation is better in crisis scenarios (both supply and demand), i.e., which results in lower investment and system operating costs. The first part of the evaluation simulated the current scenario caused by the COVID-19 pandemic in 2026, considering matrices with different levels of inflexibility of thermoelectric generation represented in the four previous scenarios. For each of these scenarios, simulations were carried out where effective demand was lower than projected—the one used as a reference for the investment decision where GDP expectation for 2020 was 2.7%, representing 3.6% growth in demand. Simulations were carried out considering GDP = −5% in 2020, representing a 3.8% reduction in demand that year, and GDP = −10%, representing a 7.4% reduction in demand in 2020. These are called “Low Demand” and “+Low Demand” scenarios, respectively. Figure 6 presents the three demand projections between 2026 and 2029, in average MW (MWh per year divided by 8760 h per year), used in this simulation.

On an annual average between 2026 and 2029, the demand growth rate for the three scenarios has little variation (“Reference Demand”—3.2% per year, “Low Demand”—3.1% per year, and “+Low Demand”—2.7% per year). However, significant variation in demand in 2020 caused a base effect during this period, leading to a significant difference in absolute values between projections during these years, as observed in Figure 6.

A comparative analysis of these scenarios was conducted using the average total cost (operation + investment) to compare them. The greater the level of inflexibility of the electricity matrix, the higher the operating cost in all scenarios, from greater availability of resources (water/rain) for hydro generation to the worst 200 as well as low to high demand. This occurs because when cheaper resources may be dispatched, there is an obligation to dispatch power plants with high variable costs. Therefore, it is expected that the economic benefit in the operation of the system will be more pronounced in scenarios with less participation of inflexible thermopowers.

A reduction in investment cost was observed with an increase in the matrix’s level of inflexibility. This reduction is associated with the reduction in the marginal operating cost (MOC) when there is compulsory generation by thermoelectric plants, which makes the option of operating the system with the available resources cheaper than the option of investing in new plants. However, consumers pay the sum of operating and investment costs, and the operating cost prevails. This means that the total cost increases with the level of compulsory generation of the system.

Figure 7 shows that the result is the same when effective demand is the same as that used in the planning process for investment decisions. There is a gradual reduction in total cost difference between the scenario with the most inflexible matrix profile and the least inflexible one, but the more flexible remains better for consumers.

Figure 7. Total Cost (Billion BRL) = Operating Cost (orange column) + Investment Cost (grey column): comparative analysis of the scenarios presented in Table 2. (a) 2000 most unfavorable hydrological series; (b) 300 most unfavorable hydrological series; (c) 200 most unfavorable hydrological series; (d) 100 most unfavorable hydrological series. Self-elaboration.

On the demand side, results were already intuitive. Now, supply shocks must be examined. The same comparative analysis was carried out, accounting for only total system cost for simulated series with more adverse hydrological situations. For this purpose, from 2000 (the uncertainty of water affluence to hydropower reservoirs is represented through 2000 scenarios in the Ten-Year Expansion Plans for the Brazilian Electricity System) simulated series, three groups were selected with the 300, 200, and 100 worst hydrological series. It was expected that with worse hydrological simulations, there would be no differences between four supply scenarios since there would be a high probability of dispatching all available resources in the system, including those with high VC. However, this did not happen. Even considering the worst 100 hydrological series, an increase in average total system cost was observed as matrix becomes more inflexible. For this extreme set of hydrological series, in the event of demand much lower than reference demand—“+Low Dem.”—the matrix with the most inflexible profile—Sc. 12 GW Gas—is more expensive than the one with the most flexible profile—Sc. 0 GW Gas—by approximately BRL 475 billion, equivalent to a cost increase of 22.5%.

Figure 7 summarizes the results obtained, showing operating (orange column) and investment (grey column) cost portions for four options of the hydrological series set analyzed—the 2000 (a), 300 (b), 200 (c), and 100 (d) most unfavorable hydrological series—comparing the impact of carrying out lower demand than that used for system expansion planning—Low Dem. and +Low Dem.— for four supply scenarios tested with different profiles in terms of dispatch flexibility—Sc. 0 GW Gas, Sc. 3 GW Gas, Sc. 8 GW Gas, and Sc. 12 GW Gas The four supply scenarios differ in the amount of expansion of inflexible natural gas thermoelectric plants. By setting this thermal expansion, there is naturally an impact on the investment decision signaled by the mathematical optimization model for other sources, especially flexible gas thermoelectric plants.

In simulations with 2000 hydrological scenarios and three demand forecasts, it is observed that the system with compulsory generation (12 GW of natural gas thermopower plants with 50% of compulsory annual generation) has a total cost (operation + investment) between 18% and 24% higher than without compulsory generation. The benefit to the consumer is reduced as the hydrological scenario becomes more stressed, but even in the worst condition, considering the 100 worst scenarios, when there is no dispatch obligation, the total cost of the system is 7.4% to 21.5% lower than the condition with the highest amount of compulsory generation. In all scenarios, the total cost for the consumer is lower without compulsory generation.

Given that the variation in investment costs is not significant between the scenarios with different levels of thermoelectric inflexibility, the impact on operating costs is decisive for the outcome of this analysis, which takes total costs as the decision variable. That said, considering the low risks of energy and power deficits shown in Figure 8 and Figure 9, it is expected that only in very severe unfavorable hydrological scenarios (less than the 100 worst simulated) would the dispatch of compulsory thermoelectric plants be entirely due to the economic decision, and consequently, there would be no variation in cost between the scenarios with increased inflexibility.

Figure 8. Energy Supply Criteria: (a) CVaR 10% Marginal Operating Cost (orange line = limit accepted by planning criteria); (b) CVaR 1% Energy Deficit (orange line = limit accepted by planning criteria).

Figure 9. Power Capacity Deficit Criteria: (a) CVaR 5% Power Capacity Deficit (orange line = limit accepted by planning criteria); (b) CVaR 1% Energy Deficit (orange line = limit accepted by planning criteria).

Additionally, it is important to highlight that all scenarios meet energy and power supply guarantee criteria [39], as shown in Figure 8 and Figure 9, respectively. These refer to the most critical scenario for service, which has greater demand and greater thermoelectric inflexibility (Scenario 12 GW gas, with reference demand: GDP + 2.7%). Although small violations of LOLP are observed in Figure 9, referring to power criteria, these are tolerated because they correspond to years with low depth of power deficit by monthly evaluation of CVaR (Conditional Value at Risk; a risk assessment measure that quantifies the amount of tail risk an investment portfolio has) 5% power capacity. The system also has reserve operating capacity equivalent to 5% of power demand. Considering an effective demand below that forecasted, limits of energy supply criteria for all evaluated scenarios have not been reached, according to Figure 10. This presents maximum CVaR 10% MOC for evaluated scenarios. This means that the system is not being operated in the optimal conditions (when criteria are at limit value in some years) established in the investment decision stage.

Figure 10. CVaR 10% (MOC) maximum between 2026 and 2029 (BRL/MWh) (black line = limit accepted by planning criteria).

Thus, an electricity matrix with a more flexible profile, that is, with less compulsory generation, proved to be more adequate to face unplanned adverse situations, such as economic, social, political, and health crises.

As in the previous section, the conclusion reached here refers to the specific configuration simulated in this case study and the sensitivities tested. However, it is understood that the level of participation of variable renewable resources in the matrix, of approximately 1/3, was decisive for the result obtained (Variable renewable sources, such as wind and solar photovoltaic, have compulsory generation just like the gas-fired thermal plants considered in the sensitivities in this study. However, unlike thermoelectric plants, the variable cost of these renewable sources is zero. Therefore, in situations where demand is lower than projected, for example, part of the available energy may not be used, or more expensive resources may be used than necessary due to the high level of inflexible generation in the Brazilian system). Therefore, unlike the previous case, considering that many countries have been looking to renew their energy matrices, driven, a large extent, by the decarbonization of the world economy, and the options for wind and photovoltaic plants have proven more attractive, which are highly variable resources, the conclusion reached can be generalized in most cases

Moreover, in this paper, natural gas-fired thermoelectric plants with compulsory dispatch were used as a proxy for inflexible supply due to the Brazilian government’s preference for contracting this source to guarantee the system’s reliability. However, similar rigidities exist across other technologies and contractual frameworks, such as coal-fired, nuclear (which must operate at base load due to technical and safety constraints), or other long-term power purchase agreements with fixed off-take commitments, which reduce operational flexibility.

Thus, the conclusions drawn here—namely, that greater dispatch flexibility reduces consumer costs under crisis scenarios—apply not only to gas plants but to any rigid supply commitments. In systems undergoing energy transitions, these rigidities may even grow in importance, making explicit consideration in expansion planning essential.

5.3. Final Discussions

The results presented in this paper extend the literature on electricity expansion planning by explicitly considering recurrent demand-contraction crises as a central uncertainty. Traditional planning models—whether deterministic least-cost optimization, stochastic programming, or robust optimization—typically focus on supply-side variability (hydrological conditions, renewable intermittency, or fuel price shocks) or on expected growth of electricity demand. In contrast, our complementary approach centers on persistent downside demand risks that have repeatedly materialized in Brazil and may emerge in other electricity markets facing political, economic, or environmental instability.

From a methodological perspective, our proposal complements advanced risk-aware techniques such as stochastic multi-stage optimization, CVaR-based methods, and bilevel formulations that model regulator–firm interactions under uncertainty. Recent works on holistic risk-aware bilevel optimization [39] illustrate the growing attention to embedding risk directly in expansion planning models. While such approaches improve robustness at the cost of mathematical complexity, our regret-based screening and flexibility stress test are deliberately simple and compatible with the official planning tools already in use. This enhances their practical applicability for agencies that must routinely deliver actionable expansion plans.

The advantages of the proposed complementary approach are threefold. First, it provides a clear decision criterion—minimax regret—to guide the choice between conservative and optimistic demand trajectories, ensuring that the risk of costly over-contracting is minimized. Second, it quantifies the tariff impact of rigid supply commitments (e.g., compulsory dispatch obligations in thermoelectric contracts), thereby highlighting the value of flexibility in generation portfolios. Third, because both analytical elements are implemented on Brazil’s official model (NEWAVE), the framework can be readily incorporated into existing planning studies without requiring a methodological overhaul.

Compared with stochastic and robust optimization methods that mainly target supply-side variability, our approach highlights the overlooked risk of recurrent demand downturns. The regret screen aligns with robust approaches but is easier to apply within existing planning workflows. Similarly, the flexibility stress test complements ongoing research on demand response and storage by quantifying the cost penalty of rigid baseload commitments. This situates our work as a practical, complementary decision-support tool that can be integrated without overhauling current models.

Taken together, these contributions demonstrate that crisis-robust expansion planning is not about replacing current least-cost optimization but about enriching it with additional decision-support tools. By systematically addressing recurrent crises, planners can provide consumers with greater protection against cost volatility and market distortions while preserving the integrity of established planning processes.

6. Conclusions and Policy Implications

Regret can be defined as the opportunity cost of failing to select the best decision for a given context. In light of the recurrent crises that have affected the Brazilian electricity sector (BES) over the past two decades, it is reasonable to assume a non-negligible probability that similar disruptions will occur within the ten-year expansion planning horizon. Such crises may result in abrupt reductions in electricity demand but may also involve supply shocks, sectoral reallocation of consumption, or unexpected booms in demand.

This study examined how to strengthen electricity expansion planning under recurrent crises that reduce demand. By applying regret minimization and flexibility stress tests with official Brazilian planning models, we identified two main results:

• Conservative demand projections reduce consumer-cost regret. Optimistic projections, while aligned with growth expectations, expose consumers to costly over-contracting when crises recur;
• Flexible supply portfolios lower costs and risks. Portfolios with smaller compulsory-dispatch shares adapt more effectively to demand shocks and hydrological variability, avoiding unnecessary tariff impacts.

Together, these elements form a crisis-robustness addendum to existing least-cost expansion studies. They do not replace current methods but complement them by highlighting downside risks often overlooked in standard optimization.

Beyond these specific findings, several broader implications emerge:

• Broadening the scope of “regret”: While this paper quantifies regret in terms of cost impacts for consumers, crises also entail broader regrets: social welfare losses, environmental damages, and geopolitical vulnerabilities. Future extensions of the framework could incorporate these dimensions through multi-objective optimization or social welfare functions;
• Accounting for both downside and upside risks: A crisis-proof system must be robust not only to recessions and demand drops but also to rapid industrialization, electrification booms, or severe supply disruptions. Planning tools should therefore test symmetric scenarios of contraction and expansion as well as shocks to supply chains and infrastructure reliability;
• Integrating policy and market design instruments: Historical responses to crises included rationing schemes, demand response pricing, and regulatory reforms. These instruments can be modeled quantitatively as flexible resources (e.g., demand response as “virtual generation” with activation costs). Incorporating such policy levers expands the scope of crisis resilience beyond physical generation and transmission assets;
• Guidance for conservative demand projections: While government planning traditionally assumes growth, our results highlight the importance of stress-testing lower-growth scenarios. Practical approaches include (i) using probabilistic forecasts with conservative quantiles, (ii) calibrating projections to historical forecast errors, and (iii) conducting stress tests for macroeconomic, political, or environmental shocks.

Policy implications are twofold. First, auction design and contracting should adopt conservative demand trajectories as baselines, reducing the financial burden of over-contracting on consumers. Second, planning agencies should limit rigid supply commitments by promoting contractual optionality or partial dispatch obligations, particularly for thermoelectric additions. These measures can be incorporated without altering the structure of official planning models, ensuring feasibility and immediate applicability.

By situating demand-contraction crises at the center of decision making, this complementary approach provides planners with simple, implementable tools to enhance resilience. The framework can also be adapted to other electricity markets where economic, political, or environmental crises pose recurring risks to demand growth.

In terms of practical use, planners should (1) run the standard least-cost plan; (2) apply the regret screen to select a conservative demand trajectory for contracting and (3) run the flexibility stress test to cap compulsory-dispatch shares or add contractual optionality. The outcome is a brief “crisis-robustness addendum” that accompanies the plan and guides auctions and contracting.

In summary, the findings suggest that expansion planning should prioritize conservative demand projections, generation portfolios with lower compulsory dispatch, and consideration of both market instruments and broader societal costs. Such configurations enhance adaptability, reduce tariff distortions, and promote a more resilient electricity system. Although the numerical application is grounded in the Brazilian context, the methodological contributions are generalizable to other jurisdictions facing uncertainty and crisis recurrence.