This thesis is clear about the time, place, technology and methods it uses. By recognizing these limits and the limits associated with them, we increase transparency, promote accurate interpretation of findings, and define paths for future research.
| :------------------------- | :---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | :---------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------------------- |
| Temporal | Historical: 2000-2024 (25 years)` `Projection: 2024–2050 (26 years)` `Resolution: Annual timesteps | Sub-annual dynamics` `Hourly dispatch` `Seasonal variations | T1: Annual resolution misses intra-year variability` `T2: Assumes continuity of current paradigms |
| Geographic | Turkey (7 regions)` `783,562 km²` `Regional disaggregation | International trade` `Sub-regional detail (81 provinces) | G1: No-trade assumption may overestimate capacity` `G2: Aggregated sub-regional heterogeneity |
| Technological | 26 technologies:` `• Solar (5 types)` `• Wind (3 types)` `• Hydro (3 types)` `• Storage (4 types)` `• Geothermal/Biomass (3)` `• Nuclear (1)` `• Fossil (7 types) | Emerging technologies (perovskite, fusion)` `Demand-side management` `Sector coupling | Te1: Excluding DSM may overestimate capacity` `Te2: Ignoring sector coupling underestimates needs |
| Methodological | Hybrid AI framework` `Neural-physical forecasting` `Multi-agent RL` `Bidirectional integration | Endogenous policy optimization` `Full AC power flow` `Political economy dynamics | M1: Linear retirement assumptions` `M2: Exogenous policy treatment` `M3: Perfect capital markets |
| Computational | 100 Monte Carlo runs` `Grid search hyperparameters` `NVIDIA GeForce RTX 3070 Laptop GPU | 1,000+ Monte Carlo runs` `Bayesian optimization` `Neural architecture search | C1: Limited hyperparameter search` `C2: Data sparsity pre-2010 |
| Generalizability | Turkey-specific calibration` `Framework applicable to emerging economies | Direct transfer to developed economies` `Off-grid electrification | G3: Transfer requires modifications |
: Installed Capacity by Technology Category (2000-2024) {#tbl:installed-capacity-by-technology-category-2000-202}
| :---------------------------- | :------------------ | :------------------ | :------------------ | :------------------ | :---------------------- |
| Coal | 8.5 | 12.8 | 18.2 | 21.5 | 3.7 |
| Natural Gas | 4.2 | 15.8 | 25.6 | 29.8 | 7.9 |
| Hydroelectric | 11.8 | 16.2 | 28.5 | 35.8 | 4.1 |
| Wind | 0.0 | 1.3 | 8.7 | 12.8 | 26.8 |
| Solar PV | 0.0 | 0.0 | 6.8 | 13.2 | 45.2 |
| Geothermal | 0.0 | 0.1 | 1.7 | 2.1 | 21.4 |
| Biomass | 0.1 | 0.3 | 1.2 | 1.8 | 12.3 |
| Nuclear | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
| Total | 24.6 | 46.5 | 90.7 | 117.0 | 6.2 |
Temporal Scope
The forecast, which is based on a historical time series of energy data belonging to TEİAŞ, MENR, IRENA, IEA ranging from 2000-2024 (25 years). Between 2001 and 2013, the current installed capacity of Turkey's energy sector was around 21 GW, while in December 2024, the installed capacity was around 115.2 GW. During this period, the Turkish energy sector dealt with distinct energy policies and developments. The period and the policies include (i) privatization (2001-2013), (ii) support schemes for renewable energy (2011-2020), and (iii) liberalising the energy market (2015-2024).
Projection horizon refers capacity projections and optimization can be made until 2050 (26 years ahead of 2024) will be harmonized with Turkey's 2053 net-zero commitment (political target) and climate pledges (including Paris Agreement). The 30-40 year lifetimes of energy infrastructure make the 2050 endpoint a familiar planning horizon. Nevertheless, all forecasts beyond 2040 inherently become significantly more uncertain due to the compounding uncertainties of the parameters, technological breakthroughs (like fusion, advanced storage, etc.) and paradigm shifts benefiting (contraction circular economy, degrowth, etc.).
The annual timesteps of the RL optimization are deemed appropriate for making strategic choices in capacity expansion, as new projects typically require 2-5 years for planning, development and commissioning. This resolution doesn't record intra-year dynamics, such as seasonal swings in demand, fluctuating monthly hydroelectric generation, or the solar/wind resources' hourly intermittent nature. The hourly dispatch and real-time balancing that an operational planning, is seen as a different problem domain. Unit commitment and economic dispatch models deal with this area and can use the capacity expansion results of this dissertation as input.
Annual resolution limits simulation from capturing sub-annual variability critical for grid operations. For example, Turkey sees a rise of 40% in electricity demand in summer (air conditioning) than winter [worldbank2023indicators] affecting its adequacy planning. Hydroelectric output for a season may vary ±30%. Adding seasonal or monthly resolutions would increase the computational demand by 12 to 52 times and would need more granular input data. Such data were often not available for the period of 2000-2024. Future additions may use multi-level schemes in which yearly policy planning informs seasonal/hourly budgeting.
Limitation T2: This assumes no major technological paradigm shifts over the next 26 years. Disruptive innovations—perovskite solar (35% efficiency at $50/kW), solid-state batteries ($50/kWh), small modular reactors ($1,500/kW)—could materially alter optimal pathways.
The analysis shows incremental learning curves, estimating annual cost reductions of 5-8%, but not paradigm shifts. All breakthroughs in technology are inherently uncertain. Chapter 3 uses scenario analysis to integrate these uncertainties into so-called "breakthrough" scenarios, wherein performance improvements occur at an accelerated pace. These scenarios are rigorous, but can't cover everything.
Geographic Scope
The electricity sector analysis covers the Republic of Turkey as a whole, which is 783,562 km² of land and defined as the Marmara, Aegean, Mediterranean, Central Anatolia, Black Sea, Eastern Anatolia and Southeastern Anatolia region. A geographical breakdown allows us to include the differences regionally in renewable energy resources.
The model doesn't include Bulgaria, Greece, Georgia, Iran, Iraq, Syria electricity imports/exports as well as treat Turkey as an isolated system. Cross-border electricity trade, which contributed to the supply in 2024 with <2% [teias2024statistics], justifies the simplification. In the future, however, throughput integration within European and Middle Eastern power pools (especially if EU accession makes progress) could significantly change optimal capacity mixes by allowing one to either import cheap Balkan hydro or export solar/wind surpluses.
Limitation G1: Excluding international trade could lead to an overestimation of the required domestic capacity by 10-20 TWh per year, together with a possible "avoided investment" worth$5-10B. This is the maximum expected annual import if future interconnection allows for it (which would be 10-20 TWh per annum, which is about ~3% of the expected demand in 2050). On the other hand, with the possibility of becoming a regional renewable energy hub, Turkey can justify an extra 20-30 GW capacity. This could be for export possibilities as opposed to domestic ones. The no-trade assumption represents a conservative base case. Future analyses should relax it as regional markets are developed.
Limitation G2: Aggregating to seven regions masks sub-national heterogeneity. Mediterranean coasts have exceptional solar potential (~1,900 kWh/m²) versus inland areas (1,500 kWh/m²). Transmission costs between sub-regions are approximated. Finer spatial resolution (81 provinces) would improve siting but needs detailed grid topology data not publicly available for security reasons.
Technological Scope
The analysis utilizes 26 electricity generation and storage technologies that can be grouped into 7 categories:
- **Solar (5):** Utility-scale PV, distributed PV, rooftop PV, concentrated solar power (CSP), and PV+storage hybrids - **Wind (3):** Onshore, offshore, distributed - **Hydroelectric (3):** Large hydro (>50 MW), small hydro (<50 MW), pumped hydro storage (PHS) - **Storage (4):** Lithium-ion batteries, flow batteries, compressed air energy storage (CAES), green hydrogen - **Geothermal/Biomass (3):** Geothermal binary, biomass combustion, biogas anaerobic digestion - **Nuclear (1):** Akkuyu Nuclear Power Plant (4.8 GW, four VVER-1200 reactors) - **Fossil (7):** Coal (lignite, hard coal), natural gas (CCGT, OCGT), oil
Below are the excluded technologies as specified in the terms and conditions:
- **Emerging/Speculative Technologies:** Emerging technologies like perovskite solar power, floating wind/ wave/ tidal energy, advanced geothermal, and fusion and space-based solar power. These technologies haven't been commercially deployed at GW-scale, and cost/performance projections are highly uncertain. - **Demand-Side Technologies:** Demand-side technologies refer to energy-efficiency enhancement and demand-response innovation. Demand-side management, by reducing the required capacity on the supply side, has other behavioral and economic issues. Those issues are better addressed through integrated assessment models. - **Sector Coupling:** Hydrogen production through electrolysis (beyond green hydrogen storage), power-to-gas, power-to-heat, carbon capture utilization and storage (CCUS). Sector coupling is important for full decarbonization of the economy but brings complexities which require integrated energy systems models.
Limitation Te1: Excluding demand-side options may overestimate required generation. If efficiency programs cut 2050 demand by 15% (650 → 550 TWh), capacity needs fall by ~30 GW, saving ~$30-50B. Efficiency potential and implementation costs are uncertain; Chapter 3 addresses demand uncertainty via ±15% demand-rate scenarios.
Demand-Side Management Framework:
This thesis concentrates on supply-side capacity expansion; managing demand (DSM) is an alternative strategy stopping from generating capacity. Comprehensive energy efficiency programs related to buildings, industry, and appliances can all together contribute to reduction in 2050 demand by 10-15% (38-58 TWh/year) which can avoid further capacity addition of 25-35 GW and the related investment requirement of$25-35B (USD 2023) [iea2023efficiency]. Demand response tools, such as time-of-use pricing and automated controls, have the potential to reduce peak demand by 7-12 GW. That is a reduction of 13-22%. According to IEA [iea2023demandresponse] and EPİAŞ [epias2024flexibility], this will avoid a capacity addition of 5-8 GW.
The electric vehicle-to-grid (V2G) system offers 600 GWh of distributed storage, which is equivalent to a baseline battery that is 40× larger. This produces 30 GW of flexibility, allowing for a renewable penetration of 98.5% – versus 97.34% for the baseline [irena2024v2g]; [2g]. The total DSM impact is a capacity reduction of 25-35 GW, an investment saving of $25-35B (USD 2023) and an emission reduction of 91-93% (in comparison to a baseline value of 90.6%). Recommendations to guide the investments: building efficiency programs ($15-20B investment, USD 2023), demand response infrastructure ($2-3B, USD 2023), V2G integration (bidirectional chargers, market design). The plan for future work is an integrated supply-demand optimization framework simultaneously optimizing capacity expansion and DSM investments.