Next Article in Journal
Effect of Nanoparticle and Carbon Nanotube Additives on Thermal Stability of Hydrocarbon-Based Drilling Fluids
Previous Article in Journal
Prediction of Battery Return Volumes for 3R: Remanufacturing, Reuse, and Recycling
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:

Long-Term Forecasting Framework for Renewable Energy Technologies’ Installed Capacity and Costs for 2050

Faculty of Engineering, Stellenbosch University, Stellenbosch 7602, South Africa
Author to whom correspondence should be addressed.
Energies 2023, 16(19), 6874;
Submission received: 28 August 2023 / Revised: 20 September 2023 / Accepted: 26 September 2023 / Published: 29 September 2023
(This article belongs to the Special Issue Energy Market Analysis, Perspectives and Policy Simulation)


Published forecasts underestimate renewable energy capacity growth and potential cost reductions, creating uncertainty around investment decisions and slowing progress. Scenario-based projections diverge widely, driven by variations in modelling techniques and underlying assumptions, with policy-based models typically being overly conservative. With historical generation capacity and cost data readily available, this research demonstrates that data-driven approaches can be leveraged to improve long-term capacity and cost forecasts of solar, wind, and battery storage technologies. Unlike exponential growth models prevailing over shorter time scales, logistic curves requiring asymptotic limits, or machine learning algorithms dependent on extensive datasets, this analysis demonstrates that temporal quadratic regressions are a better starting point to represent capacity growth trends over two to three decades. When coupled with published learning rates, trend-based capacity forecasts provided tighter and lower capital and levelized cost of energy outlooks than most reviewed scenarios, with photovoltaics global average levelized cost of energy reducing from 0.057/kWh to below USD 0.03/kWh by 2030 and below USD 0.02/kWh by 2040. Greater transparency on manufacturing ecosystems is proposed so that more advanced analytical techniques can be utilized. This analysis indicates that without direct interventions to accelerate the growth in wind power generation, global renewable energy technology deployment will fall short of the generation capacities required to meet climate change objectives.

1. Introduction

Global power generation is irreversibly shifting with the rapid deployment of renewable energy technologies and the rapid cost reductions in the sector. In 2020, solar photovoltaic (PV) and onshore wind energy each had more than 700 GW of installed power generation capacity globally [1]. Combined, this represents nearly 80 times the installed solar and wind generation capacity that prevailed in 2000 [1]. While this exploding growth in renewable energy generation capacity was initially nurtured by policy and investment incentives [2], the favorable economics of solar and wind renewable energy solutions are increasingly driving adoption [3]. Hydropower, geothermal, and bioenergy capacities have also grown; however, projects are resource-dependent and have not benefited from the same capacity expansion and associated cost reductions [1,3]
Industry bodies, such as the International Renewable Energy Agency (IRENA) and the International Energy Agency (IEA), have been building global databases with increasingly comprehensive analyses when it comes to electricity generation [1,4]. The levelized cost of energy (LCOE) information in Figure 1 shows significant deflationary trends for most technologies over the last decade or two. Countries favored by steady coastal winds or better than average irradiation have LCOEs rivalling the marginal cost of operating fossil fuel plants [5], notwithstanding the complex dynamics, subsidies, and taxes in fossil fuel markets [6,7]. In the Middle East, new large-scale PV projects are being announced with LCOE below USD 0.02/kWh [5]. While LCOE has limitations and is not the only metric to track the economics of renewable energy technologies, it is widely used [8].
Over the last decade, renewable energy technologies’ LCOE improvements have primarily been derived from steadily reducing total installed costs or capital expenditures (Capex) and lowering financing and operational costs [5]. While most analyst projects sustained cost reductions in real terms across all significant renewable power-generating technologies, but forecasts vary substantially [9,10]. Conversely, traditional fossil fuel power plants face volatile and uncertain input costs, with instability in energy markets highlighting risks to economic activity [11,12].
According to the technical report published by the European Commission’s Joint Research Centre (JRC), with an extensive reference list of 52 publications and a review of some 20 sets of forecasts, installed capacity projections for major renewable energy technologies can vary by as much as a factor of six with the Capex interquartile range by a factor of two [10]. A subsequent study by Sens et al. [13] uses published generation capacity forecasts for solar and wind technologies to develop long-term Capex and LCOE projections; however, the range of forecasts used for the cost predictions remains relatively wide. The Capex projections in Figure 2 are derived from the IEA and JRC analyses for major solar and wind technologies. LCOE predictions are more sporadic, typically available for a specific country or region, namely, from the National Renewable Energy Laboratory (NREL) for the United States; the IEA for the European Union, China, and India; and the Fraunhofer Institute for Germany [9,14].
Furthermore, forecast accuracy for renewable energy technologies has proven challenging [15]. Sens et al. [13] determined that only some studies have produced forecasts for Capex that remained valid by the time of publication due to dynamic cost developments. Additionally, the exact method and assumptions for Capex estimates are rarely fully described, making the forecasting approach challenging to duplicate in most instances. The more recent study by Xiao et al. [16] is believed to be the first to have carried out a systematic cost assumption comparison in scholarly literature. Their findings show that almost all future energy scenarios underestimate the trend of substantial cost decreases and that even the most current studies use out-of-date or extremely cautious numbers. Metayer et al. [17] reached the same conclusion when reviewing the International Energy Agency World Energy Outlook by highlighting the chronic upward capacity projection revisions over time for renewable energy technologies. In 2016, the United States Energy Information Administration conducted a self-assessment of its past wind and solar energy forecasts, highlighting chronic underestimation of penetration and cost reductions [18]. While several corrective actions were proposed, the outlook of the Energy Information Administration is usually mainly driven by policy impact assumptions. Other approaches, using survey-based forecasts compiling hundreds of experts’ views, have underestimated life cycle cost reductions. For example, in 2021, Wiser et al. [19] expected the cost of wind technologies in 2050 to be half what they predicted in 2015, a mere six years before. These wide-ranging cost forecasts, coupled with the general underestimation of technology adoption and deployment, tend to underestimate the future role of renewable energy, overestimate the costs of the energy transition, create uncertainty, and delay investments [16].
As a result, aspirational scenarios and roadmaps have been proposed, with current monitoring with the IEA pointing to large gaps in installed generation capacity that are expected to persist [20]. Within the context of ambitious renewable energy targets set in the United States in 2009, Daim et al. [21] reviewed extensive work on logistic growth or S-curves and proposed capacity growth projections. Using an asymptotic maximum of 10% in electricity production for any specific technology, the negligible contribution from solar PV has proven erroneous. In 2015, Metayer et al. [17] developed projections for renewable energy technologies, using global population projections and European average electricity usage to establish an asymptotic limit for the year 2100 for global electricity consumption. As a result, their capacity projections for PV and wind electricity generation were double the aspirational IRENA “REmap” scenario for 2050, with growth only slowing after 2070. The logistic curve model developed by Metayer et al. [17] also had unexplained deviations from known renewable energy generation values, with a forecast of 31%, too low for 2000.
With transparent generation capacity market data available, it is possible to move away from scenario-based forecasting techniques to analyze and extrapolate market dynamics. Over the last decade, forecasting techniques within the renewable energy sector have noticeably evolved [22], including long-term power infrastructure planning for a given country [23]. However, whether forecasts look at electricity demand or variable wind and solar generation, long-term perspectives tend to be limited to days or weeks [24,25,26]. More recently, machine learning algorithms have been shown to forecast variable energy production for up to thirty months with relatively high accuracy, including generation growth, by combining shorter-term seasonal trends with long-short-term memory dynamics [27]. Other research proposed long-term forecasts of carbon emissions for the building sector [28]. However, these models tend to require more extended historical datasets, which are not always available, such as in the case of more novel technologies, namely, for battery energy storage systems.
This research proposes a trend-based renewable energy capacity extrapolation approach combined with technological learning rates to provide decadal cost forecasts. The validity and utility of the proposed approach are demonstrated through application in solar energy, wind energy, and battery energy storage technologies. Contributions from this research include the following:
  • Best-fit temporal quadratic regressions on available annual capacity data provide narrower decadal forecasts than those published for wind and solar energy technologies.
  • New long-term forecasts for capacity and cost are provided for the more novel battery energy storage systems.
  • Insights into the manufacturing infrastructure required for renewable technologies’ continued growth are shown as among the most influential drivers of capacity expansion.
The methodology is based on the premise that the market behavior is now primarily determined by favorable economics and, to a lesser extent, by policy frameworks (i.e., the technologies are established and self-sustaining). In this respect, the approach differs from previously published work. A further advantage of this approach is that it can be replicated at a global, national, or regional level by accounting for differences in market conditions. This methodology is self-correcting such that any significant disturbances for a specific technology are captured in the latest capacity or learning rate data, with the next iteration adjusting accordingly. Developing more comprehensive databases around each technology manufacturing ecosystem should become a priority to enable further advances in modelling algorithms to forecast decade-long growth. The findings from this work should help stakeholders gain greater confidence in the capacity and cost outlook essential to accelerate renewable energy uptake, especially in markets that have been slow to transform, and to develop insights on the growth dynamics for each technology.

2. Materials and Methods

A data-driven approach is proposed to narrow capacity and cost forecasts for existing technologies and develop an outlook on more novel technologies. While many technical, environmental, economic, and policy factors may impact scenario-based cost outlooks, the temporal extrapolation methodology presented in this research relies on existing and underlying growth trends. As such, this approach embeds drivers of growth and their constraints. This system-thinking approach assumes that progress in global renewable energy deployment over the next two to three decades will be driven by similar dynamics to those that have prevailed over the last decade. Discontinuities that have occurred over the last decade are also considered and discussed. The building blocks presented in Figure 3 describe the research methodology based on the principles of technological learning, as initially proposed by Wright [29].

2.1. Global Installed Electricity Generation Capacity Scenarios

Global capacity deployment and forecasts are critical in establishing cost predictions for a given technology. However, forecasts can vary significantly given social, political, and economic factors and assumptions built into specific models [10]. In a 2018 report, the JRC described three capacity scenarios encapsulating most other scenarios proposed across several industry bodies [10]. The first scenario was a low-end “Baseline” scenario, which assumes no significant interventions. The second “Diversified” scenario was carbon-neutral, per the 2017 International Energy Agency outlook. The third was a “proRES” scenario, which assumed nuclear energy was replaced with renewable energy and excluded non-economic carbon capture and storage solutions. The JRC capacity forecasts in Table 1 are compared with IRENA’s “REmap” projections, which define the required renewable energy capacity necessary to reduce greenhouse gas emissions and limit global temperature increases to less than 2 °C above pre-industrial levels [30]. As can be seen, IRENA’s “REmap” scenario is mostly in line with the JRC’s “proRES”, with a higher penetration of solar PV but a lower capacity for concentrating solar power (CSP). The IEA’s “Net zero” 2050 scenario, also included in Table 1, calls for significantly greater renewable energy generation capacity, despite similar limits on temperature increases. Results-driven scenarios such as IRENA’s “REmap” and the IEA’s “Net zero” help define what is required to achieve specific outcomes. However, these aspirational projections may not realistically represent what can be achieved when considering growth dynamics, as shown in this analysis.

2.2. Global Installed Electricity Generation Capacity Growth Profile of Solar and Wind Technologies

Renewable energy generation capacity growth is usually considered exponential, characterized by compound annual growth rates [31,32] or subject to asymptotic limits associated with logistic or S-curves [17,21]. This research shows that global electricity generation capacity across all major renewable energy technologies has accelerated over time but has not achieved exponential growth over decadal periods. This section analyses and presents best-fit regressions on historical capacity data based on the least sum of square deviation method.
The coefficients of determination (R2) included in Table 2 are particularly low for linear regressions, which do not reflect the generation capacity acceleration over time for solar and wind technologies. While this analysis determined high R2 (greater than 0.9) for exponential regressions, the divergence in early and later years, which can be seen in Figure 4, led to large divergence for periods longer than 20 years. Third-degree polynomials were found to have the highest R2; however, distortions tended to be amplified when extrapolating over long periods. For example, PV generation capacity for 2050 was determined to be twice the level of IRENA’s REmap scenario, while the result for concentrating solar power was negative. Quadratic extrapolations (second-order polynomials) were found to model reality more accurately for 20 years or more with R2 of more than 0.95 for the 2000–2020 IRENA dataset. Given the use of annual capacity data, the multi-year extrapolation is not subject to seasonality as would be electricity generation data.
The quality of fit for the quadratic regression for onshore wind shown in Figure 4 can also be observed for PV, offshore wind, and concentrating solar power CSP, as represented in Figure 5, albeit with slightly lower correlation coefficients. These high levels of correlation were surprising and are investigated further in a theoretical underpinning presented in Section 2.4. Note that for PV, the regression curve is shown from 2010, given the low number of utility-scale installations existing before this period.
While PV and offshore wind technologies only gained scale from 2010 onwards, onshore wind has a longer track record, which was used to test various regression timeframes and forecasts. When extrapolating the 5-year, 10-year, and 15-year regressions from 2000 onwards to predict the onshore wind power generation capacity in 2020, the deviation obtained from the actual 2020 generation capacity of 698 GW inevitably reduced with longer time series. A time series of at least ten years appeared necessary to minimize deviations between forecasted and realized growth. The generation capacity extrapolations in Table 3 confirm that quadratic regressions on historical capacity were more accurate than exponential regressions (or compounded growth models), which consistently overestimated generation capacity [31,32].
Despite an apparent deviation of 7% for 2020, closer examination of the 10-year 2000–2010 best-fit quadratic regression extrapolated over the next decade revealed a tight correlation to actual capacity with an average annual deviation of less than 5%. For the last three years (2018–2020), the average deviation was merely 3%, with deviations of 2% in 2018 (eight years out) and 1% in 2019 (nine years out). This research consequently proposes using temporal quadratic regressions to model the deployed generation capacity of renewable technologies.

2.3. Learning Rates as a Means to Determine Cost Trends

Modelling approaches leveraging selected interrelated variables have become increasingly accepted in understanding future cost development of technologies, specifically when transposing cost reduction trends from one industrial sector to another. However, this is often challenging, given the need for large datasets to build such models [33]. McDonald and Schrattenholzer [34], in an extensive review of more than 50 publications, reported a simple and consistent relationship between the evolving cost of a given technology and its productive capacity progression. This relationship was shown to have a high level of correlation over decadal periods. Most often referred to as the one-factor learning rate (LR), this “experience curve” is represented by the rate of reduction in cost from a doubling of cumulative capacity [34]. For many technologies, namely, in power production, empirical relationships between the unit cost of the technology (Y) and the cumulative installed capacity are found to be linear on a double logarithmic scale. The relationship is defined as follows:
Y (X) = a Xb
where X is the cumulative capacity, a is the cost of the first unit, and b is the learning elasticity and is negative for reducing cost trends [34].
By assigning doubling = 2X0 for a doubling in installed capacity and defining X0 as the reference capacity, Y0 as the reference unit cost, and Ydoubling as the unit cost at the time of the doubling of capacity, the learning rate can be derived as follows [34]:
LR = (Y0 − Ydoubling)/Y0 = 1 − (Ydoubling/Y0) = 1 − (2X0)b/X0b = 1 − 2b
Rubin et al. [13] provided qualitative explanations for the relationship between cost and experience with the following three drivers:
  • Improvements to production processes such as productivity, process innovations, standardization, and economies of scale;
  • Changes in product design, namely, redesign and the use of novel materials;
  • Real-term reductions in the input price of materials and labor.
These improvements were argued to co-occur in varying degrees. More advanced cost models looked at both installed capacity and research and development expenditure, splitting learning rates into “doing” and “research” components [33].
While learning rates may start tapering down as a technology matures and may not be applicable over a century, historical learning rates coupled with capacity forecasts are an established way to anticipate energy cost developments over the coming decades [10]. In their review, McDonald and Schrattenholzer [34] analyzed the learning rates of 42 energy-related technologies, including solar and wind power generation. They established a median of 16%, similar to that found more broadly for many diverse manufacturing technologies in the 1980s. Yao et al. [35] extensively reviewed published learning rate analyses for renewable energy technologies. They concluded that capacity factor effects, reductions in the cost of capital, and specific technology learning effects were the three major drivers enabling cost reductions across renewable energy technologies [32]. In the 2021 cost report, IRENA has included, for the first time, learning rates (averaged over the 2010 to 2020 period) for solar and wind electricity-generating technologies [5].
Published learning rates for electricity generation renewable energy technologies are summarized in Table 4. Capex-related learning rates have been studied for decades, while LCOE learning rates have only recently been analyzed. The onshore wind Capex learning rate of around 15% has been stable since the 1980s. In contrast, the Capex learning rate for PV has accelerated over time. Using a 1968–1998 data set, McDonald and Schrattenholzer calculated a 20% Capex learning rate for PV [34]. A subsequent study analyzing German installations between 1990–2019 showed a Capex learning rate of 24% for PV [36]. IRENA’s calculations [3] show a higher rate of 34% for the Capex of PV plants between 2010–2020. CSP has experienced a similar trend over the last two decades, with the Capex learning rate increasing from around 10% to 20% in the last decade. IRENA published 2010–2020 learning rates for LCOE, included in Table 4, and are higher than Capex learning rates for the same technology. The higher learning rate for LCOE is consistent with the conclusion of Xiao et al. [16] that declines in discount rates accounted for a sizeable portion of the levelized cost decrease in renewable energy technologies. They found that reducing discount rates stemmed from the experience effect in lending, brought on by lower anticipated project risk due to technology maturity and increased competition [16].

2.4. Empirical and Theoretical Underpinnings

Considering a particular renewable energy technology with G0 [GW] of installed generating capacity and P0 [GW/annum] in manufacturing capacity for a given reference year t0. Assuming a constant rate of change in manufacturing capacity P ˙ over a relatively stable period, then the increase in manufacturing capacity as a function of time would be as follows:
P t = i n t e g r a l ( P ) ˙ = P ( ˙ t ) + P 0
The total installed capacity over time would then be calculated from the following:
G t = i n t e g r a l ( P t ) = 1 2 P ˙ ( t ) 2 + P 0 ( t ) + G 0
As can be seen from the coefficients presented in Table 5, unconstrained quadratic regressions yielded negative P0 for the installed manufacturing capacity in the year 2000 for all the technologies. Except for the onshore wind, these regressions did not yield fitted G0 values consistent with the known installed generating capacity in the reference year. Constrained regressions were then obtained by forcing P0 and G0 values for the reference year (the year 2000 in the case of the 20-year regressions and the year 2010 in the case of the 10-year regressions). The installed manufacturing capacity coefficients P0 values were estimated using a three-year average growth in generation capacity from the reference year. Global electricity generation capacity coefficients G0 values were set to the installed generation capacity for the reference year.
Coefficients of determination remained above 0.9 for all four renewable energy technologies considered, even when imposing manufacturing and generation capacity-related constraints. There were, however, differences in the nature of the best-fit regressions.
  • Onshore wind showed the highest level of coherence over the 10- and 20-year time frames, with a relatively constant rate of increase in manufacturing capacity. Whether unconstrained or forced, the 20-year regressions yielded R2 greater than 0.994.
  • For both PV and offshore wind, the correlation weakened with R2 reducing from 0.98 to 0.92 when forcing reference manufacturing and generation capacity for the year 2000. The 2010–2020 regression better predicted future trends for both technologies. This result can be attributed to the scaling of manufacturing infrastructure over the last decade. There have been notable shifts for PV since 2010, with Chinese PV production aggressively coming onstream, which is accommodated by a rebasing in manufacturing capacity.
  • CSP showed a robust correlation between actual generation capacity and the 10- and 20-year regressions. However, the erratic capacity addition provided distortions to the shorter 10-year time series, an unlikely predictor of future trends.
While the onshore wind had a stable increase in manufacturing capacity over time, both PV and offshore wind are in an acceleration phase. For more dynamic technologies, it is understandable that the second-degree coefficient, representing the rate of increase in manufacturing capacity, would rise as technologies become more universally accepted. Once technologies mature, global generation capacity becomes more linear, with the second-degree coefficient becoming a lesser contributor. This progressive transition would be catered for in the quadratic coefficients as regressions are updated over time.
For onshore wind, fixing the initial manufacturing capacity reduced the impact of the second-order term. Over the shorter 2000–2010 period, fixing P0 and G0 measurably reduced the accuracy of generation capacity forecasts. The 10-year 2000 2010 best-fit regression capacity forecasts for the subsequent ten years presented in Figure 6 highlight that the unconstrained regression better predicted global generation capacity evolution for the subsequent decade.

3. Results

This section compares forecasts for global electricity generation capacity and wind and solar technologies’ cost against published scenarios. Furthermore, the same forecasting approach is used for more novel technologies, such as battery energy storage systems, where few scenarios exist.

3.1. A Narrower 2050 Global Electricity Generation Capacity Outlook

The global electricity generation capacity forecasts in Figure 7 confirm that the 20-year quadratic unconstrained regressions provide credible 2050 capacity outlooks for solar and wind technologies, with predictions falling within the range of published scenarios. Figure 7 is explained below:
  • The more representative forecast using the unconstrained 20-year quadratic extrapolation is the primary marker on the graph (yellow bar). The constrained quadratic provides more conservative extrapolations of generation capacity for PV and onshore wind, reflected in the lower value marked with crosses (yellow X).
  • PV estimates using the 2010–2020 ten-year regressions are also shown. The forecasts using the shorter, later time series point to an acceleration in global installations reaching or exceeding the upper end of reviewed ranges, in line with IRENA’s 2050 “REmap” scenario [29]. The 133GW or 19% increase in PV power generation capacity in 2021 [3] aligns with this higher forecast.
  • Concentrating solar power with a projected 40–50 GW capacity for 2050 is below published scenarios, at less than half of the JRC’s baseline expectation. The marked reduction in capacity addition in later years drives this lower level of adoption.
Constrained regression-based forecasts, using initial manufacturing and generation capacity conditions, appeared conservative and were excluded from Figure 7. Therefore, the unconstrained quadratic regressions are used to predict capacity trends when forecasting costs.
This research and associated forecasts confirm that PV and onshore wind technologies are expected to continue to dominate new power generation installations over the coming decades, with PV set to lead the growth. PV capacity growth prediction of 9600 GW for 2050, falling at the upper end of the range, is likely to provide consistent overperformance against industry outlooks [17,37]. It would align with IRENA’s REmap projection of 8520 GW, as shown in Table 1. Onshore wind predictions of 4200 GW falling in the middle of the published range appear reasonable but would fall short of the aspirational targets proposed by IRENA of 5440 GW and the IEA at 7439 GW included in Table 1.

3.2. Cost Forecasts of Established Renewable Energy Technologies

As the introduction highlights, this research aims to provide narrower cost forecasts for solar and wind power generation technologies. The global average LCOE in Figure 8 are plotted against capacity on a double logarithmic scale. The chart includes the progress over the last two decades, from the year when IRENA’s dataset started to provide global averages and offers annual predictions up to the year 2050. Each technology LCOE outlook is based on the 2000–2020 20-year unconstrained best-fit quadratic capacity model combined with IRENA’s 2010–2020 10-year LCOE learning rates presented in Table 4. For CSP, the three-year average LCOE for 2018–2020 was used as a reference, given the high variations in reported LCOE leading up to 2020.
Based on this analysis, solar PV is expected to be increasingly competitive and have a similar global average LCOE compared to onshore wind by 2035. Concentrating solar power and offshore wind technologies are expected to continue to lag. However, both should become increasingly competitive against traditional fossil fuel power plants with life-cycle costs of between USD 0.06–0.15/kWh [5]. The same approach can be used to extrapolate costs using the relevant national/local starting point LCOE for a given geography with specific wind and irradiation conditions. Locations with favorable solar and wind conditions and lower costs would yield substantially lower forecasts than presented in this analysis.
The LCOE forecasts presented in Table 6 were derived from sensitivity analyses around IRENA 2010–2020 10-year published learning rates and include values using the 10-year regression for PV and offshore wind. Global average LCOE predictions from this analysis tend to be more aggressive than reviewed scenario estimates, also included in Table 6. For 2050, the projected PV utility scale global average LCOE at USD 0.009–0.015/kWh is at the lower end of published forecasts, with a cost decrease of 74–86% in real terms. Onshore wind 2050 LCOE predictions of USD 0.013–0.014/KWh are less than half of the published estimates of USD 0.03–0.04/kWh, driven partly by the higher learning rates calculated for the last ten years. For offshore wind, 2050 projected LCOE of USD 0.036–0.051/kWh is comparable with the published range. For CSP, Fraunhofer indicated a value of USD 0.06/kWh by 2050 [9], consistent with the USD 0.044–0.069/kWh from this analysis.
Capex forecasts presented in Figure 9 align with the lower end of published scenarios reviewed by the European Commission JRC [10], but for offshore wind, they are nearer the median. Forecasts from this research, using 20-year unconstrained capacity regressions, were compared to the JRC’s interquartile range using the most relevant technologies (the JRC provided broader ranges from an extensive set of technology options). The magnitude of Capex reductions over the next thirty years is also shown in Figure 9 and was found to be in line with more aggressive published forecasts [13].

3.3. Cost Forecasts of Nascent Renewable Energy Technologies

Electricity storage is seen as a critical enabler in lowering the transportation sector’s carbon footprint, balancing national power grids, and supporting the distributed generation of electricity [38]. Battery energy storage systems (BESS) are increasingly being adopted for electricity with variable renewable energy generation from PV and wind [38] and have been used in this analysis to test the methodology. For BESS, this analysis combined capacity forecasts with published learning rates to forecast Capex and costs for PV + BESS systems.
While several battery energy storage technologies exist, lithium-ion (Li-ion) batteries have led their growth and will be used as a basis for capacity extrapolations and LCOE forecasts. This growth has resulted from rapid consumer electronic and transport application advances [39]. Since 2014, Li-ion storage systems have been introduced in utility-scale renewable energy generation installations [40]. The global Li-ion installed capacity progressions expressed in gigawatts in Figure 10 provide an overview of the total market since 2011 and for stationary applications since 2014.
Quadratic regression coefficients in Table 7 were derived using best-fit unconstrained and constrained temporal regressions on installed capacity. With R2 of 0.992, quadratic regressions were found to be well suited to model the growth of the total lithium-ion battery market since 2011. The quadratic best-fit is also strongly correlated with stationary application growth since 2014, with an R2 of 0.999. When forcing P0 and G0 values for the estimated installed manufacturing and storage capacity for reference years 2011 and 2014, the second-degree coefficient and the R2 marginally decreased, as was the case for PV and wind. For capacity extrapolation, the unconstrained best-fit coefficients were again adopted.
While a relatively short time frame is available, growth trends for the total Li-ion market and stationary applications are emerging. The quadratic extrapolation would point to approximately 110 GW of stationary storage capacity installed by 2030, a six-fold increase from the 17 GW reported in 2020. By 2050, installed capacity could exceed 500 GW based on this analysis. While this points to spectacular growth, the IEA’s “Net zero” emission 2050 scenario would require six times more storage, making this outlook relatively conservative [14].
Using the quadratic regression to extrapolate installed capacity and a learning rate of 15–20% for Li-ion battery energy storage systems [41,42], applied to a 2020 cost of USD 430/kWh for a four-hour system [43], this analysis found Capex reducing by more than a third by 2030 to USD 210–290/kWh and reducing by 50–70% to USD 125–210/kWh by 2050. These cost reductions, summarized in Table 8, align with the NREL annual technology baseline forecast of a 60% reduction by 2050 in their moderate forecast [43]. However, the 2050 forecasts fall short of more aggressive capacity growth scenarios as modelled by Vartiainen et al. [37], with resulting four-hour system capex of USD 56–113/kWh by 2050 (using a 2020 average exchange rate of USD 1.13/EUR).
Setting an LCOE baseline for PV power generation and battery energy storage systems is challenging as calculation methodologies vary. In addition to the direct benefit of using the excess energy stored during solar or wind power generation hours, the ability of the system to store low-cost power from regional or national grids and provide ancillary services would also contribute positively to the system’s return and reduce the cost of storage [38,44]. Representative values were therefore selected for LCOE forecasts for PV + BESS systems. For illustration purposes, cost extrapolations presented in Table 8 were based on 2020 estimates for Capex and LCOE for a combined system. The Capex of USD 2044/kW and LCOE of USD 0.065/kWh represent the average for a 4 h BESS utility-scale PV system, with the benefit of storing low-cost electricity, as determined by the National Renewable Energy Laboratory [43]. The learning rates for combined PV + BESS systems are expected to be somewhat lower than PV alone but higher than batteries. Indicatively, the LCOE of this illustrative combined PV + BESS system would be lower than the cost of new fossil fuel plants by 2030, with the NREL expecting an average of USD 0.035/kWh [43] under moderate scenario assumptions, in line with this analysis. By 2050, the NREL forecast of USD 0.028/kWh [43] falls in line with the more conservative forecast from this analysis. By comparison, Vartiainen et al. [37], using a value of USD 0.055/kWh for a 5 h-BESS utility-scale PV system located in Malaga (Spain) in 2020, projected LCOE of USD 0.02/kWh by 2050. With improved tracking of combined PV + BESS projects in the coming years, more robust forecasting of Capex and LCOE global averages and regional values should be possible.

3.4. Limitations

When taking into consideration physical growth drivers by forcing coefficients to reflect the initial installed global manufacturing and generation capacities of a given technology, high levels of correlation were observed. However, the associated quadratic regressions lost some of their apparent accuracy. The bias introduced by fixing values of two of the three regression coefficients reduced growth outlooks. This deviation indicates that other dynamics may be involved, pointing to interactions between global growth in manufacturing capacity and the baseline manufacturing capacity. Detailed, verified, and consistent information on the manufacturing infrastructure and other possible growth drivers would be required to derive more accurate representations for each regression coefficient through more advanced modelling algorithms, possibly using machine learning-based models, which goes beyond the scope of this work. Greater transparency on market growth drivers could also be used to understand the relative impact of investments, research and development interventions, growth-shaping incentives, and policy influences on each technology’s generation capacity and manufacturing ecosystem [45].
While the temporal regression-based models were relatively consistent across periods for onshore wind, shifts were identified for PV and offshore wind technologies. Some studies have proposed steeper growth trajectories for solar PV and associated cost reductions in the coming years [37,46]. The acceleration of PV-generating capacity is also evident in the updated generation capacities for 2021 and 2022 presented in Table 9, which for 2022 were 7% ahead of forecasts using the more aggressive 10-year regression. Therefore, a deeper understanding of the PV supply chains would be required to understand the progressive acceleration in the growth rate in both generation and manufacturing. One observation, for example, is that both reported global manufacturing capacity for PV and installed electricity generation capacity is accelerating in tandem.
Finally, material shortages, supply chain disruptions, and a change to the inflationary environment, as observed in 2021 and 2022, could lead to short-term cost increases and, as a result, impact forward-looking learning rates. In the 2020–2023 period, the global COVID pandemic disrupted supply chains and project timelines, while unpredictable regional conflicts had far-reaching impacts on energy markets; nonetheless, LCOE reductions continued into 2021 in line with projections using the higher learning rates presented in this study. Furthermore, an unprecedented capacity increase occurred for offshore wind in 2021, as seen in Table 9, with China adding 50% to the global installed electricity generation capacity in one year [3]. This surge was followed by an increase in capacity of 7 GW in 2022, in line with expectations. Care should, therefore, be taken in using the results of this analysis directly. Instead, the focus should be on the insights developed from the data-driven methodology, the apparent links to manufacturing capacity, and the rate of increase in manufacturing investments. Forecasts can then be updated regularly, automatically embedding emerging trends and smoothing aberrations.

4. Discussion

The narrower capacity outlook and associated cost forecasts for wind and solar power generation technologies were found to be more aggressive than scenarios reviewed or developed by leading renewable energy agencies and research groups. These results were consistent with recent assessments that published scenarios and forecasts tend to underestimate solar and wind energy cost declines [16,37]. This trend indicates that the estimated costs for the transition away from fossil fuel to renewable energy for electricity generation are likely overstated.
While exponential regressions are believed to provide a good fit in modelling generation capacity over 5–10 years, this research shows that temporal quadratic regressions are more effective over decadal periods. Uncharacteristically high correlations were observed for unconstrained quadratic regressions and even when linking coefficients to initial manufacturing and generation capacity estimates (R2 > 0.9). The forecasts presented in this research are, therefore, based on historical growth drivers and constraints, which may be subject to change over time. Therefore, regular regression updates should be conducted as longer time series become available to ensure that discontinuities or shifts in manufacturing capacity and capabilities are built into future modelling. Furthermore, coherent datasets on generation and manufacturing capacity should enable more advanced analytical techniques to validate the assumption of relatively constant growth in manufacturing capacity. Piecewise linear approximation of quadratic regressions could be an additional tool to address discontinuities observed for solar thermal technologies since 2015 and for the step function increase in offshore wind generating capacity in 2021.
A key advantage of the proposed data-driven approach is that it can be replicated and adapted to cater to regional, national, or sectorial realities without complex forecasting and modelling techniques. The prevailing levelized cost of energy or capital expenditures for a given geography can be used as a basis for extrapolation. Should regional or national learning rates vary from global trends due to local dynamics, localized learning rates could be substituted to yield more representative forecasts for a given context. Further work would be required to validate the approach for geographies with already very low LCOE for solar and wind and for emerging technologies with limited time series, which typically need more meaningful forecasts.

5. Conclusions

Improving forecast accuracy for the cost and deployment of renewable energy technologies is critical to quantify the investment required to facilitate the transition away from fossil fuels. While learning rates are increasingly being used to project the cost of renewable technologies, they can only be effective with accurate capacity forecasts. Unfortunately, published reviews have shown that capacity forecasts tend to have significant deviations driven by underlying scenario assumptions. The approach developed in this research to predict the cost evolution of renewable energy technologies provides a transparent and practical forecasting technique.
  • Quadratic best-fit regressions on historical global electricity generation capacity help develop narrower, data-driven growth forecasts.
  • The multi-year forecasting approach on global installed capacity was shown to apply to solar and wind renewable energy technologies and high-growth novel technologies such as battery energy storage systems.
  • By relying on embedded growth dynamics, shown to be strongly correlated to the manufacturing infrastructure of each technology, the approach helps reduce assumption-driven biases.
A more comprehensive database capturing critical information about the supply chain and manufacturing ecosystems, namely, the historical manufacturing capacity and rate of manufacturing capacity expansion of each renewable energy technology, should be developed to enable more advanced algorithms to be utilized. Future work should include the development of machine learning-based modelling using the expended dataset, including manufacturing capacity. Specifically for photovoltaic installations, which will likely meet ambitious penetration scenarios, further analysis would be required to better understand growth drivers and limits to growth. Future work should include technical and physical limits on material availability and costs to validate forecasts, specifically where favorable conditions allow for significantly lower-than-average LCOE.
The narrower predictions of global electricity generation capacity growth and cost reductions developed in this analysis should better equip stakeholders regarding long-term planning. For policymakers, more accurate capacity forecasts should serve as a reality check on whether aspirational targets for decarbonization are likely to be achieved and at what costs. Under current growth dynamics, the global capacity outlook should serve as a warning that global electricity generation capacity from renewable sources will not be sufficient to fulfil the requirements necessary to meet climate change objectives, as modelled by the IEA and IRENA. When it comes to wind, fiscal and regulatory interventions (such as feed-in tariffs, tax credits, exemptions, and preferential interest rates) will continue to be required to remove constraints, stimulate investments, and broaden the manufacturing footprint at global and local level. Finally, better coordination will be required to develop more accurate and aggressive consensus capacity and cost outlook across all renewable energy technologies, especially for those less researched, such as battery energy storage systems.

Author Contributions

Conceptualization, F.R. and C.M.; methodology, F.R.; validation, F.R.; formal analysis, F.R.; writing—original draft preparation, F.R.; writing—review and editing, C.M. and M.O.; supervision, C.M. and M.O. All authors have read and agreed to the published version of the manuscript.


This research received no specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data Availability Statement

Analyses data available on request.

Conflicts of Interest

The authors declare no conflict of interest.


  1. International Renewable Energy Agency. Renewable Energy Statistics 2021. Available online: (accessed on 3 November 2021).
  2. Timilsina, G.R.; Kurdgelashvili, L.; Narbel, P.A. Solar energy: Markets, economics and policies. Renew Sustain. Energy Rev. 2012, 16, 449–465. [Google Scholar] [CrossRef]
  3. International Renewable Energy Agency. Renewable Power Generation Costs in 2021; International Renewable Energy Agency: Abu Dhabi, United Arab Emirates, 2022. [Google Scholar]
  4. International Energy Agency. Renewable Energy Market Update—Outlook for 2020 and 2021; International Energy Agency: Paris, France, 2020. [Google Scholar]
  5. International Renewable Energy Agency. Renewable Power Generation Costs in 2020; International Renewable Energy Agency: Abu Dhabi, United Arab Emirates, 2021. [Google Scholar]
  6. Antimiani, A.; Costantini, V.; Paglialunga, E. Fossil fuels subsidy removal and the EU carbon neutrality policy. Energy Econ. 2023, 119, 106524. [Google Scholar] [CrossRef]
  7. Hansen, T.A. Stranded assets and reduced profits: Analyzing the economic underpinnings of the fossil fuel industry’s resistance to climate stabilization. Renew. Sustain. Energy Rev. 2022, 158, 112144. [Google Scholar] [CrossRef]
  8. Loth, E.; Qin, C.; Simpson, J.G.; Dykes, K. Why we must move beyond LCOE for renewable energy design. Adv. Appl. Energy 2022, 8, 100112. [Google Scholar] [CrossRef]
  9. Kost, C.; Shammugam, S.; Fluri, V.; Peper, D.; Memar, A.D.; Schlegel, T. Levelized Cost of Electricity—Renewable Energy Technologies; Fraunhofer ISE: Freiburg, Germany, 2021. [Google Scholar]
  10. Tsiropoulos, I.; Tarvydas, D.; Zucker, A. Cost Development of Low Carbon Energy Technologies—Scenario-Based Cost Trajectories to 2050; Publications Office of the European Union: Luxembourg, 2018. [Google Scholar] [CrossRef]
  11. Yu, Y.; Guo, S.; Chang, X. Oil prices volatility and economic performance during COVID-19 and financial crises of 2007–2008. Resour. Policy 2022, 75, 102531. [Google Scholar] [CrossRef]
  12. Kuik, F.; Feveile Adolfsen, J.; Magdalena Lis, E.; Meyler, A. Energy price developments in and out of the COVID-19 pandemic—From commodity prices to consumer prices. Econ. Bull. Artic. 2022, 4. [Google Scholar]
  13. Sens, L.; Neuling, U.; Kaltschmitt, M. Capital expenditure and levelized cost of electricity of photovoltaic plants and wind turbines—Development by 2050. Renew. Energy 2022, 185, 525–537. [Google Scholar] [CrossRef]
  14. Bouckaert, S.; Pales, A.F.; McGlade, C.; Remme, U.; Wanner, B.; Varro, L.; D’Ambrosio, D.; Spencer, T. Net Zero by 2050: A Roadmap for the Global Energy Sector; International Energy Agency: Paris, France, 2021. [Google Scholar]
  15. Mauleón, I. Assessing PV and wind roadmaps: Learning rates, risk, and social discounting. Renew. Sustain. Energy Rev. 2019, 100, 71–89. [Google Scholar] [CrossRef]
  16. Xiao, M.; Junne, T.; Haas, J.; Klein, M. Plummeting costs of renewables—Are energy scenarios lagging? Energy Strategy Rev. 2021, 35, 100636. [Google Scholar] [CrossRef]
  17. Metayer, M.; Breyer, C.; Fell, H.-J. The projections for the future and quality in the past of the World Energy Outlook for solar PV and other renewable energy technologies. In Proceedings of the 31st EU PVSEC, Hamburg, Germany, 14–18 September 2015. [Google Scholar]
  18. United States Energy Information Administration. Wind and Solar Data and Projections; United States Energy Information Administration: Washington, DC, USA, 2016. [Google Scholar]
  19. Wiser, R.; Rand, J.; Seel, J.; Beiter, P.; Baker, E.; Lantz, E.; Gilman, P. Expert elicitation survey predicts 37% to 49% declines in wind energy costs by 2050. Nat. Energy 2021, 6, 555–565. [Google Scholar] [CrossRef]
  20. International Energy Agency. World Energy Outlook; International Energy Agency: Paris, France, 2022. [Google Scholar]
  21. Daim, T.; Harell, G.; Hogabaum, L. Forecasting renewable energy production in the US. Foresight 2012, 14, 225–241. [Google Scholar] [CrossRef]
  22. Croonenbroeck, C.; Stadtmann, G. Renewable generation forecast studies—Review and good practice guidance. Renew. Sustain. Energy Rev. 2019, 108, 312–322. [Google Scholar] [CrossRef]
  23. Giannelos, S.; Jain, A.; Borozan, S.; Falugi, P.; Moreira, A.; Bhakar, R.; Mathur, J.; Strbac, G. Long-term expansion planning of the transmission network in India under multi-dimensional uncertainty. Energies 2021, 14, 7813. [Google Scholar] [CrossRef]
  24. Soman, S.S.; Zareipour, H.; Member, S.; Malik, O.; Fellow, L. A Review of Wind Power and Wind Speed Forecasting Methods with Different Time Horizons. In Proceedings of the North American Power Symposium 2010, Arlington, TX, USA, 26–28 September 2010. [Google Scholar] [CrossRef]
  25. Dudek, G. Multilayer perceptron for GEFCom2014 probabilistic electricity price forecasting. Int. J. Forecast. 2016, 32, 1057–1060. [Google Scholar] [CrossRef]
  26. Zhang, Z.; Zhang, Y.; Huang, Q.; Lee, W.-J. Market-oriented optimal dispatching strategy for a wind farm with a multiple stage hybrid energy storage system. CSEE J. Power Energy Syst. 2018, 4, 417–424. [Google Scholar] [CrossRef]
  27. Ding, S.; Zhang, H.; Tao, Z.; Li, R. Integrating data decomposition and machine learning methods: An empirical proposition and analysis for renewable energy generation forecasting. Expert Syst. Appl. 2022, 204, 117635. [Google Scholar] [CrossRef]
  28. Giannelos, S.; Moreira, A.; Papadaskalopoulos, D.; Borozan, S.; Pudjianto, D.; Konstantelos, I.; Sun, M.; Strbac, G. A machine learning approach for generating and evaluating forecasts on the environmental impact of the buildings sector. Energies 2023, 16, 2915. [Google Scholar] [CrossRef]
  29. Wright, T.P. Factors affecting the cost of airplanes. J. Aeronaut Sci. 1936, 3, 122–128. [Google Scholar] [CrossRef]
  30. International Renewable Energy Agency. Global Energy Transformation: A Roadmap to 2050; International Renewable Energy Agency: Abu Dhabi, United Arab Emirates, 2019. [Google Scholar]
  31. Nagy, B.; Farmer, J.D.; Bui, Q.M.; Trancik, J.E. Statistical basis for predicting technological progress. PLoS ONE 2013, 8, e52669. [Google Scholar] [CrossRef] [PubMed]
  32. Farmer, J.D. How predictable is technological progress? Res. Policy 2016, 45, 647–665. [Google Scholar] [CrossRef]
  33. Rubin, E.S.; Azevedo, I.M.L.; Jaramillo, P.; Yeh, S. A review of learning rates for electricity supply technologies. Energy Policy 2015, 86, 198–218. [Google Scholar] [CrossRef]
  34. McDonald, A.; Schrattenholzer, L. Learning curves and technology assessment. Int. J. Technol. Manag. 2003, 23, 718–745. [Google Scholar] [CrossRef]
  35. Yao, Y.; Xu, J.; Sun, D. Untangling global levelised cost of electricity based on multi-factor learning curve for renewable energy: Wind, solar, geothermal, hydropower and bioenergy. J. Clean. Prod. 2021, 285, 124827. [Google Scholar] [CrossRef]
  36. Wirth, H. Recent facts about photovoltaics in Germany. Fraunhofer Inst. Sol. Energy Syst. 2021, 1, 92. [Google Scholar]
  37. Vartiainen, E.; Masson, G.; Breyer, C.; Moser, D.; Román Medina, E. Impact of weighted average cost of capital, capital expenditure, and other parameters on future utility-scale PV levelised cost of electricity. Prog. Photovolt. Res. Appl. 2020, 28, 439–453. [Google Scholar] [CrossRef]
  38. International Renewable Energy Agency. Electricity Storage and Renewables: Costs and Markets to 2030; International Renewable Energy Agency: Abu Dhabi, United Arab Emirates, 2017. [Google Scholar]
  39. Bloomberg New Energy Finance. Electric Vehicle Outlook 2020; Bloomberg New Energy Finance: New York, NY, USA, 2020. [Google Scholar]
  40. International Energy Agency. Energy Storage 2021. Available online: (accessed on 9 March 2022).
  41. Schmidt, O.; Hawkes, A.; Gambhir, A. The future cost of electrical energy storage based on experience rates. Nat. Energy 2017, 2, 17110. [Google Scholar] [CrossRef]
  42. Glenk, G.; Meier, R.; Reichelstein, S. Cost dynamics of clean energy technologies. Schmalenbach J. Bus. Res. 2021, 73, 179–206. [Google Scholar] [CrossRef] [PubMed]
  43. National Renewable Energy Laboratory. Annual Technology Baseline 2022. Available online: (accessed on 21 July 2022).
  44. Denholm, P.; Eichman, J.; Margolis, R. Evaluating the Technical and Economic Performance of PV Plus Storage Power Plants; National Renewable Energy Lab. (NREL): Golden, CO, USA, 2017. [Google Scholar]
  45. Lee, J.; Yang, J.S. Government R&D investment decision-making in the energy sector: LCOE foresight model reveals what regression analysis cannot. Energy Strategy Rev. 2018, 21, 1–15. [Google Scholar] [CrossRef]
  46. Haegel, N.M.; Atwater, H., Jr.; Barnes, T.; Breyer, C.; Burrell, A.; Chiang, Y.M.; De Wolf, S.; Dimmler, B.; Feldman, D.; Glunz, S.; et al. Terawatt-scale photovoltaics: Transform global energy. Science 2019, 364, 836–838. [Google Scholar] [CrossRef] [PubMed]
Figure 1. Global average levelized cost of energy for solar and wind energy technologies [5].
Figure 1. Global average levelized cost of energy for solar and wind energy technologies [5].
Energies 16 06874 g001
Figure 2. Global average Capex projections for major renewable energy technologies [5,10,14] (IEA net zero scenario using US, EU, and China average for PV and wind; JRC utility-scale technology interquartile maximum forecasts for PV and wind adjusted for exchange rate and inflation; and CSP from JRC range only).
Figure 2. Global average Capex projections for major renewable energy technologies [5,10,14] (IEA net zero scenario using US, EU, and China average for PV and wind; JRC utility-scale technology interquartile maximum forecasts for PV and wind adjusted for exchange rate and inflation; and CSP from JRC range only).
Energies 16 06874 g002
Figure 3. Schematic of approach to improve forecasts through best-fit regression modelling (dark-grey-shaded boxes are new contributions from this research).
Figure 3. Schematic of approach to improve forecasts through best-fit regression modelling (dark-grey-shaded boxes are new contributions from this research).
Energies 16 06874 g003
Figure 4. Best-fit regressions for the global installed generation capacity of onshore wind [1].
Figure 4. Best-fit regressions for the global installed generation capacity of onshore wind [1].
Energies 16 06874 g004
Figure 5. Global installed generation capacity quadratic best-fit regressions 2000–2020. (a) Onshore wind and solar photovoltaic. (b) Offshore wind and concentrating solar power.
Figure 5. Global installed generation capacity quadratic best-fit regressions 2000–2020. (a) Onshore wind and solar photovoltaic. (b) Offshore wind and concentrating solar power.
Energies 16 06874 g005
Figure 6. Onshore wind global installed generation capacity forecasts using 2000–2010 best-fit regressions.
Figure 6. Onshore wind global installed generation capacity forecasts using 2000–2010 best-fit regressions.
Energies 16 06874 g006
Figure 7. Global electricity generation capacity—2050 scenario ranges and quadratic predictions (The low end of the forecast range is from the JRC “baseline” scenario [10], and the high end of the range is from the IRENA “REmap” scenario [29]).
Figure 7. Global electricity generation capacity—2050 scenario ranges and quadratic predictions (The low end of the forecast range is from the JRC “baseline” scenario [10], and the high end of the range is from the IRENA “REmap” scenario [29]).
Energies 16 06874 g007
Figure 8. Global average levelized cost of energy trends and forecasts (log–log scale).
Figure 8. Global average levelized cost of energy trends and forecasts (log–log scale).
Energies 16 06874 g008
Figure 9. Global average Capex forecasts from this analysis compared to current and published scenarios [10] (PV: one axis tracking; Onshore: medium specific capacity, medium hub height; Offshore: monopole, medium distance; and CSP: tower with storage).
Figure 9. Global average Capex forecasts from this analysis compared to current and published scenarios [10] (PV: one axis tracking; Onshore: medium specific capacity, medium hub height; Offshore: monopole, medium distance; and CSP: tower with storage).
Energies 16 06874 g009
Figure 10. Global installed lithium-ion nominal capacity in operation [39,40].
Figure 10. Global installed lithium-ion nominal capacity in operation [39,40].
Energies 16 06874 g010
Table 1. Global installed electricity generation capacity for major renewable energy technologies—current base and scenarios.
Table 1. Global installed electricity generation capacity for major renewable energy technologies—current base and scenarios.
Global Installed Electricity Generation Capacity (GW)
Base [5]
EU Joint Research Centre (JRC)
2050 Outlook [10]
REmap [30]
Net Zero [14]
ActualBaselineDiversifiedProRES<2 °C<1.5 °C
Solar photovoltaic707146444246745852014,458
Onshore wind69815213037444454407439
Offshore wind341084371131600827
Concentrating solar61349391473300426
Table 2. Coefficient of determination from best-fit regressions for IRENA’s 2000–2020 global electricity generation capacity dataset.
Table 2. Coefficient of determination from best-fit regressions for IRENA’s 2000–2020 global electricity generation capacity dataset.
R2 from Regressions on Global Installed Electricity Generation Capacity
a 0 ( 1 + C A G R ) ( t t 0 )
c1 (tt0) + c0
c2 (tt0)2 + c1 (tt0) + c0
c3 (tt0)3 + c2 (tt0)2 +c1 (tt0) + c0
Solar photovoltaic0.9930.7460.9810.999
Onshore wind0.9830.9270.9980.998
Offshore wind0.9960.7600.9780.999
Concentrating solar0.9160.8800.9560.976
CAGR: compounded annual growth rate; t is the year of the forecast, and t0 is the year 2000 (the baseline year).
Table 3. The 2020 Onshore wind generation capacity predictions using 5-, 10-, and 15-year regressions (t0: year 2000).
Table 3. The 2020 Onshore wind generation capacity predictions using 5-, 10-, and 15-year regressions (t0: year 2000).
PeriodRegression CoefficientsR22020 Capacity
Extrapolation (GW)
2020 Actual 698
2000–2015 (15 y)1.8 (tt0)2 − 3.1 (tt0) + 250.9987071%
2000–2015 (15 y)27 ( 1.200 ) ( t t 0 ) 0.993103849%
2000–2010 (10 y)1.6 (tt0)2 − 0.8 (tt0) + 220.995650−7%
2000–2010 (10 y)18.9 ( 1.253 ) ( t t 0 ) 0.9991713145%
2000–2005 (5 y)0.5 (tt0)2 + 5.6 (tt0) + 170.999325−54%
2000–2005 (5 y)18.9 ( 1.254 ) ( t t 0 ) 0.9941752151%
Table 4. Overview of learning rates for solar and wind renewable energy technologies.
Table 4. Overview of learning rates for solar and wind renewable energy technologies.
Total Installed Costs or Capex
(USD/kW Improvement)
(USD/kWh Improvement)
LR from Compiled
Literature Review [10]
IRENA Reported
LR [5]
IRENA Reported
LR [5]
Solar photovoltaic10%35%34%39%
Onshore wind8%20%17%32%
Offshore wind8%15%9%15%
Concentrating solar5%20%13–22%36%
Literature review data exclude LR outliers (observations that are outside the cluster of reported learning rates). For concentrating solar power, few projects exhibit year-on-year LR distortions with a lower value of 13% when analyzing the 2010–2019 IRENA Capex dataset.
Table 5. Constrained and unconstrained quadratic regression coefficients and R2.
Table 5. Constrained and unconstrained quadratic regression coefficients and R2.
Solar PhotovoltaicOnshore Wind
P ˙ P0G0R2 P ˙ P0G0R2
(GW)(GW)(GW) (GW)(GW)(GW)
20 y
P0 and G0 2000 estimates3.
10 y
P0 and G0 2010 estimates6.932400.9892.3381780.995
Offshore windConcentrating solar
P ˙ P0G0R2 P ˙ P0G0R2
(GW)(GW)(GW) (GW)(GW)(GW)
20 y
P0 and G0 2000 estimates0.
10 y
P0 and G0 2010 estimates0.−
Shaded regression coefficients are recommended for extrapolations; t0 uses year 2000 for 20 year regressions and year 2010 for 10 year regressions.
Table 6. Global average levelized cost of energy forecasts and sensitivity to learning rates.
Table 6. Global average levelized cost of energy forecasts and sensitivity to learning rates.
Learning Rate
2050 LCOE Estimates
2000–2020 Capacity Forecasts
20 y Regression
2050 LCOE Estimates
2010–2020 Capacity Forecasts
10 y Regression
2050 LCOE Published Forecast and Scenario Range
% of 2020
39% *0.012−79%0.009−84%
Onshore wind32% *0.014−63%10 y regression outcome consistent with 20y0.03–0.04
Offshore wind15% *0.051−39%0.045−46%0.035–0.040
25%0.069−36%10 y regression distorted with the slowdown0.06
36% *0.044−59%
* IRENA 2010–2020 published learning rates [5]; 2050 published scenarios from the IEA [14] and from Fraunhofer [9].
Table 7. Best-fit regression for lithium-ion global capacity in operation (Total t0: 2011; Stationary t0: 2014).
Table 7. Best-fit regression for lithium-ion global capacity in operation (Total t0: 2011; Stationary t0: 2014).
Li-Ion TotalLi-Ion Stationary
P ˙ P0G0R2 P ˙ P0G0R2
(GW)(GW)(GW) (GW)(GW)(GW)
8 y
P0 and G0 2011 estimates3.66.5350.981
6 y
Unconstrained 0.820.141.00.999
P0 and G0 2014 estimates 0.650.700.90.996
Table 8. Cost outlook for BESS and PV + BESS systems with 4–5 h storage.
Table 8. Cost outlook for BESS and PV + BESS systems with 4–5 h storage.
Learning Rate202020302050
MinMax MinMaxMinMax
BESS Capex
2020 USD/kWh
PV + BESS Capex
2020 USD/kW
2020 USD/kWh
The USD 430/kWh BESS and the USD 2044/kW PV + 4 h BESS 2020 Capex based on the 2022 NREL annual technology baseline R&D report [43]; 2030 and 2050 BESS Capex forecasts minimum (Min) using BESS quadratic 2014–2020 regression 0.41(tt0)2 + 0.14(tt0) + 0.98 with t0: 2014 and 20% LR; maximum (Max) calculated using Li-ion quadratic 2011–2019 regression 2.74(tt0)2 − 1.76(tt0) + 37.5 with t0: 2010 and 15% LR; PV + BESS outlook derived from quadratic regression PV capacity extrapolation presented in Table 5 using for Min 5.8(tt0)2 + 7.1(tt0) + 53 with t0: 2010 and a 30% LR, and for Max 3.1(tt0)2 – 32(tt0) + 54 with t0: 2000 and a 20% LR for Capex and a 25% LR for LCOE.
Table 9. The 2021 and 2022 generation capacity forecasts against actual forecasts.
Table 9. The 2021 and 2022 generation capacity forecasts against actual forecasts.
Generation Capacity
PV: 10-year regression 5.8 (tt0)2 + 7.1 (tt0) + 53. Onshore 20-year regression 1.7 (tt0)2 − 1.0 (tt0) + 21. Offshore: 10-year regression 0.29 (tt0)2 + 0.21 (tt0) + 3.4. CSP: 10-year regression 0.035 (tt0)2 + 0.86 (tt0) + 1.2.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Rozon, F.; McGregor, C.; Owen, M. Long-Term Forecasting Framework for Renewable Energy Technologies’ Installed Capacity and Costs for 2050. Energies 2023, 16, 6874.

AMA Style

Rozon F, McGregor C, Owen M. Long-Term Forecasting Framework for Renewable Energy Technologies’ Installed Capacity and Costs for 2050. Energies. 2023; 16(19):6874.

Chicago/Turabian Style

Rozon, Francois, Craig McGregor, and Michael Owen. 2023. "Long-Term Forecasting Framework for Renewable Energy Technologies’ Installed Capacity and Costs for 2050" Energies 16, no. 19: 6874.

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop