Skip to content
RT
Skip to main content

Start Here

Understanding Turkey's Energy Transition Framework

Hybrid AI • Forecasting • Reinforcement Learning • Policy Optimization

Your onboarding guide. Light, focused, and designed to prepare you without overload.

Research TeamProject SnapshotGo to Overview →
Data→ Forecast→ RL→ Policy→ Feedback

Research Team

This doctoral research is conducted at Erciyes University under the supervision of Prof. Dr. Faik Bilgili. The team brings together expertise in energy economics, econometrics, and AI-driven energy system modelling.

Prof. Dr. Faik Bilgili

Prof. Dr. Faik Bilgili

Thesis Supervisor · T.R. Erciyes University

Education

  • · BA Economics — Dokuz Eylül Univ. (1982–86)
  • · MA Economics — Hacettepe Univ. (1986–89)
  • · Ph.D. Economics — CUNY, USA (1991–97)

Citation & Impact

  • · Publications: 161 (WoS 84, Scopus 86)
  • · Citations: Scholar 6,825 · WoS 4,190 · Scopus 4,708
  • · H-Index: Scholar 37 · WoS 32 · Scopus 32
  • · Projects: 25 · Thesis Supervision: 29 · Open Access: 47

Work & Expertise Areas

  • · Editorial: Journal editor, peer review (650+ manuscript evaluations)
  • · Advisory: Energy policy, environmental economics, panel data econometrics
  • · Expertise: Energy economics, macroeconomics, machine learning applications
  • · 100+ articles, 16 books/chapters; Environmental Kuznets Curve, renewable energy–growth nexus
View profile →faikbilgili.com ↗
Seyit Önderol

Seyit Önderol

Ph.D. Candidate · Erciyes University · University of Vienna

Education

  • · BA Economics — Vienna Univ.
  • · BA Mathematics (double degree) — Vienna Univ.
  • · MSc Quantitative Economics and Finance — Vienna Univ.
  • · Ph.D. Wirtschaftsinformatik — Vienna Univ.
  • · Ph.D. Economics (current) — Erciyes Univ.

Citation & Impact

  • · Publications: 6 (all since 2021)
  • · Citations: Scholar 207
  • · H-Index: 4 · i10-Index: 4
  • · Strong 2023–2025 momentum; rapid impact growth

Work & Expertise Areas

  • · Platform: Energy transition modelling, data visualisation, web applications
  • · Energy: Renewable energy economics, Turkey electricity system, long-horizon forecasting
  • · AI: Reinforcement learning, deep learning, time-series forecasting
  • · Software engineering: Full-stack development, data pipelines, open-source tooling
View profile →seyitonderol.com ↗
Next section →

Project Snapshot

2024–2050
Time horizon
7
Regions
13
Algorithms tested
MAPPO
Best result
97.34%
Renewable share
90.6%
CO₂ reduction
$226.79B
Investment
← Previous sectionNext section →

What This Project Is

This platform presents a hybrid artificial intelligence framework designed to analyze Turkey's renewable energy transition from 2024 to 2050, aligned with the 2053 net-zero commitment pledged at COP26. It integrates physics-informed forecasting with reinforcement learning–based policy optimization.

← Previous sectionNext section →

Why This Research Matters

Turkey's electricity demand is projected to nearly double by 2050. The power sector generates 152 MtCO₂ (27% of national emissions), maintains 54.2% renewable share, but faces 70% energy import dependency ($60B+ annually).

This research investigates whether a renewable-dominant electricity system is feasible by 2050, aligned with the 2053 net-zero pledge, and how investment and policy decisions should be optimized under uncertainty.

The platform provides an interactive environment to analyze these transition pathways using hybrid artificial intelligence models.

← Previous sectionNext section →

Key Concepts

Energy Transition — Shift from fossil fuels to renewables. Turkey target: ≥90% renewable electricity by 2050.
Energy System Planning — Long-term optimization of generation capacity, investment, and policy decisions under economic and environmental constraints.
Hybrid AI Framework — Physics constraints + deep learning (ENSM) + reinforcement learning (MAPPO). Bidirectional integration converges in 4–7 iterations, reducing forecast–policy gap from 8 GW to <1 GW.
Physics-Informed Forecasting — S-curve and G=K×CF×8760 energy balance. <0.1% physics violation vs pure ML.
Reinforcement Learning (RL) — Policy optimization via reward signal. 13 algorithms compared (PPO, DQN, SAC, MAPPO, etc.).
TurkeyEnergyTransitionEnv — RL environment: 26 steps (2024–2050), 7 regions, 26 technologies, 446 state dims, 119 action dims.
2053 Net-Zero — Turkey&apos;s political commitment at COP26. This framework delivers the first internally consistent optimization pathway toward that target.

Literature gaps and risk layers → Motivation

← Previous sectionNext section →

Research Contributions

  1. A hybrid artificial intelligence framework integrating physics-informed forecasting with multi-agent reinforcement learning for long-term energy transition planning.
  2. A regional multi-agent RL model for Turkey's energy system (7 agents, CTDE).
  3. Bidirectional forecast–policy integration loop reducing forecast–policy gap by 89%.
  4. Comprehensive benchmark comparing 13 optimization algorithms.
  5. Physics-consistent forecasting with <0.1% energy balance violation.
← Previous sectionNext section →

Platform Map

Transition section (left sidebar). Also: Portal, Simulator, Trust.

Overview

Executive summary and key numbers.

Go →Motivation

Problem statement, literature gaps, and risk layers.

Go →Data Explorer

Six datasets with filtering and export.

Go →Architecture

Five-layer system diagram and technical structure.

Go →Methodology

ENSM, MAPPO, reward design, and 13 algorithms.

Go →Results

2050 pathway, algorithm comparison, and policy insights.

Go →Thesis Viewer

Full dissertation (7 chapters) with search and navigation.

Go →Experiments

Reproducible training runs and model evaluation.

Go →Simulator

Interactive scenario builder for energy transition analysis.

Go →
← Previous sectionNext section →

Targets

2024–2050
Time Horizon
26 steps
≥90%
Renewable Target
by 2050
$226.79B
Investment
optimal (range $200–450B)
26
Technologies
capacity types
327 → 649 TWh
Demand Growth
2024–2050
446
State Space
dimensions
119
Action Space
6 levels
7
Regions
Turkey

Full parameters → Methodology · Results → Results

← Previous sectionNext section →

System Overview

High-level flow with forecast–policy feedback. Technical details are on the Architecture page.

Data↓Physics Forecast↓RL Optimization↓Policy Pathway↓Forecast Update

↻ Bidirectional forecast–policy integration loop

Full 5-layer diagram and technical structure →

← Previous sectionNext section →

Acronyms

Forecast Models

ENSM — Enhanced Neural S-curve Model
HNSM — Hybrid Neural-Sigmoid Model
GRU — Gated Recurrent Unit
PICP — Prediction Interval Coverage Probability
MPIW — Mean Prediction Interval Width
MC Dropout — Monte Carlo Dropout

Reinforcement Learning

MARL — Multi-Agent Reinforcement Learning
MAPPO — Multi-Agent Proximal Policy Optimization
PPO — Proximal Policy Optimization
CTDE — Centralized Training, Decentralized Execution

Energy Economics

LCOE — Levelized Cost of Energy
OCC — Overnight Capital Cost
CF — Capacity Factor
CBAM — Carbon Border Adjustment Mechanism

More terms: Glossary

← Previous sectionNext section →

Reading Paths

General Reader

Overview → Results → Policy Insights

Start →

Researcher

Motivation → Methodology → Results

Start →

Policymaker

Overview → Results → Policy Insights → Data Sources

Start →

Technical Deep Dive

Architecture → Forecasting → RL → Experiments

Start →

Full Thesis

Thesis Viewer (7 chapters, search, figures)

Start →
← Previous sectionNext section →

Data Sources

TEİAŞ — National grid operator
MENR — Ministry of Energy
IRENA — International renewable database
IEA — International Energy Agency
Eurostat — European statistics
NREL ATB — Technology costs
World Bank — GDP, population
← Previous sectionNext section →

Key Numbers

Essential metrics at a glance.

0.9799
Forecast R² (gen)
0.9131
Forecast R² (cap)
13
Algorithms
7
Regional agents
26
Time steps
218 GW
Total capacity
446
State variables
119
Action variables
97.34%
Renewable share
$226.79B
Investment
90.6%
CO₂ reduction
99.2%
Reliability
+3.22 pp
Integration gain
78
DM tests
95.4%
PICP
← Previous sectionNext section →

Reproducibility

All datasets, code, and experiments used in this study are accessible through the platform modules and experiment logs.

The experiment module provides:

  • Model configurations
  • Training logs
  • Hyperparameters
  • Reproducible runs
  • Experiment lineage

This ensures full transparency and reproducibility of the research results.

View Experiments →
← Previous section