Skip to content
RT
Skip to main content

System Architecture

Five-layer architecture integrating Forecast (sources/Forecast) and Reinforcement (sources/Reinforcement) for Turkey's 2024–2050 energy transition.

Key Metrics

446
State dims
RL environment
119
Action dims
Hierarchical
0.9799
Forecast R²
ENSM
13
Algorithms
RL + evolutionary
7
Regions
Turkey
2024–2050
Horizon
26 years
26
Technologies
Per region
94.2%
PICP
Uncertainty

Hybrid AI Framework — Conceptual Overview

The diagram below summarises the data flow, computational modules, and validation processes of the hybrid AI framework developed for Turkey's 2050 energy transition. It shows the integration of physical constraints, neural networks, and reinforcement learning.

Hybrid Artificial Intelligence Framework for Turkey's 2050 Energy Transition — Data flow, ENSM forecasting, MAPPO optimization, bidirectional integration, validation, and web simulator

1. Data Sources and Physical Constraints

Data from TEİAŞ, IEA, IRENA, Ember and NREL ATB for 2000–2024. Energy balance (P=K×CF×8760), reserve margin and regional limits (7 TEİAŞ regions) applied as physical constraints.

2. Data Preparation and Feature Engineering

Raw data transformed into 106 features: temporal (year, season), technology-regional (26 technologies × 7 regions) and economic-policy variables (GDP, energy prices, regulations).

3. ENSM Forecast Module

Gompertz curves (technology diffusion) combined with GRU (time series). 2024–2050 capacity and generation forecasts. R²=0.9799, outperforms 17 benchmark models.

4. Uncertainty Layer

Monte Carlo Dropout (100 runs) quantifies forecast reliability. PICP (Prediction Interval Coverage Probability) and MPIW (Mean Prediction Interval Width) metrics. PICP=95.4%, MPIW=11.8 GW.

5. RL Optimization (MAPPO/PPO)

Uncertainty-aware forecasts fed to MAPPO (Multi-Agent Proximal Policy Optimization) / PPO. Optimal policy learned with 1000+ forward passes. 13 algorithms compared; MAPPO/PPO selected.

6. Policy and System Outputs

2050 portfolio (solar 32%, wind 28%, hydro 27%), emissions (152→14.30 MtCO₂), investment ($226.79B), LCOE ($42.8/MWh) and reliability (99.2%) computed.

7. Bidirectional Integration

Wright's law (learning curves) and CAPEX updates: policy decisions update forecasts; updated forecasts improve policy. Convergence in 4–7 iterations, 89% reduction in forecast-policy gap.

8. Validation and Analysis

17 model benchmarks, 13 RL algorithm benchmarks, 78 Diebold-Mariano tests, Bootstrap/ANOVA and SHAP analysis validate the framework. Explainable AI provided.

9. Web Platform Simulator

Interactive scenario analysis with forecast_original.json, forecast_projection.json, reinforcement_unified.json. Users explore transition pathways via map, charts and comparison tools.

10. Outcome: Turkey 2050 Energy Transition

97.34% renewable share, near net-zero electricity system (90.6% emission reduction), technical and economic feasibility (0.35% GDP investment). Evidence-based roadmap aligned with 2053 net-zero commitment.

Five-Layer Flow

Data Flow Sequence

Python Packages — Full List

All Python packages used in Forecast, Reinforcement, and scripts. Categorized, classified, and described in detail.

Python packages used in Forecast, Reinforcement, and scripts — Core, Deep Learning, ML, Visualization, Data I/O, Utilities

Core Scientific Computing

NumPy (≥1.21) — Numerical arrays, linear algebra. Used in Forecast, Reinforcement, scripts. Modules: numpy

pandas (≥1.3) — DataFrames, CSV/Excel/Parquet I/O, time series. Used in Forecast, Reinforcement, scripts. Modules: pandas

SciPy (≥1.7) — curve_fit (S-curve fitting), statistical tests, signal processing. Used in Forecast, Reinforcement. Modules: scipy.optimize, scipy.stats, scipy.signal

Deep Learning

PyTorch (≥1.12) — GRU, LSTM, Dense layers; PPO, SAC, DQN, A3C, TD3 agents; Monte Carlo Dropout; Adam optimizer. Used in Forecast, Reinforcement. Modules: torch, torch.nn, torch.optim, torch.nn.functional, torch.utils.data

Machine Learning & Preprocessing

scikit-learn (≥1.0) — RobustScaler, StandardScaler, MinMaxScaler; train_test_split, TimeSeriesSplit; SimpleImputer; PCA; mutual_info_regression. Used in Forecast, Reinforcement. Modules: sklearn.preprocessing, sklearn.model_selection, sklearn.impute, sklearn.decomposition, sklearn.feature_selection

Visualization

Matplotlib (≥3.5) — 18 figure categories, Q1 journal standards, 300 DPI. Used in Forecast, Reinforcement. Modules: matplotlib.pyplot, matplotlib.dates, matplotlib.patches, matplotlib.ticker, mpl_toolkits.axes_grid1

Seaborn (≥0.11) — Statistical plots, model comparison, heatmap, distribution plots. Used in Forecast, Reinforcement. Modules: seaborn

Data I/O

pyarrow — Parquet read (pd.read_parquet). Turkey_Energy_Technology_Metrics.parquet, unified_energy_technology_dataset.parquet. Used in Reinforcement, scripts. Modules: pyarrow

openpyxl — Excel (.xlsx) read. Turkey_Renewable_Energy_Annual_Dataset.xlsx backend for pandas. Used in Forecast. Modules: openpyxl

Utilities & Config

PyYAML — config/main.yaml. Environment, reward, agent, training parameters. Used in Reinforcement. Modules: yaml

emoji — Console emoji support (✅, ⚠️, 📊). Used in Forecast. Modules: emoji

psutil — CPU, memory monitoring. SystemMonitor (optional). Used in Reinforcement. Modules: psutil

Recommended requirements.txt
# Core
numpy>=1.21
pandas>=1.3
scipy>=1.7

# Deep Learning
torch>=1.12

# Machine Learning
scikit-learn>=1.0

# Visualization
matplotlib>=3.5
seaborn>=0.11

# Data I/O
pyarrow
openpyxl

# Config & Utilities
pyyaml
emoji
psutil
Standard library: sys, os, locale, argparse, logging, json, pathlib, typing, datetime, warnings, pickle, hashlib, random, copy, collections, threading, time, dataclasses, enum, glob, csv

Run Commands

Copy and run from the project root:

Forecast
cd sources/Forecast && python advanced_neural_physical_forecasting.py
Reinforcement (data)
cd sources/Reinforcement && python main.py --mode data
Reinforcement (train)
cd sources/Reinforcement && python main.py --mode train
Reinforcement (full)
cd sources/Reinforcement && python main.py --mode full
Data export
pnpm run data:export

Source Structure

sources/Forecast/
advanced_neural_physical_forecasting.py— Main entry — HNSM + ENSM
▼professional_visualization/— 18 visualizers
coordinator.py— Orchestrates all figures
main_figures.py— Methodology, dual approach
forecast_analysis.py— CI, trend, sensitivity
performance_analysis.py— Benchmarks, convergence
model_parameters.py— Hyperparams, activation
validation_results.py— Validation stats
literature_comparison.py— Random Forest, methodology
scientific_rigor.py— Power analysis
reproducibility.py— Data/experiment repro
data_quality.py— Data validation
model_selection.py— Model ranking
time_series_analysis.py— Decomposition, trend
target_compliance.py— Regulatory compliance
error_analysis.py— Error distribution
international_comparison.py— Regional, policy
economic_analysis.py— ROI, cost-benefit
advanced_statistics.py— Bootstrap, tests
reporting_documentation.py— Code/data docs
▼outputs/— Results
01_Figures/— 18 figure categories
04_Raw_Data/— Data dictionary, CSVs
comprehensive_results_*.json— Forecast results
advanced_neural_physical_2024_2050.csv— Projection CSV
sources/Reinforcement/
main.py— Main launcher — data, train, comparison
run_analysis_and_visualization.py— Post-training analysis
config/main.yaml— Environment, reward, agent, training
▼src/— Core modules
agent/— PPO, DQN, A3C, SAC, TD3, GA, PSO, DE, FuzzyQL, NeuroFuzzy, MARL, MAML, ProgressiveNN
data/— loader, processor, validator, structures
environment/energy_env.py— Turkey 2024–2050 RL env
training/trainer.py— PPO trainer
reward/— reward_calculator, technology_weights
cost/cost_calculator.py— Cost calculation
comparison/— comprehensive_comparison, statistical, sensitivity, monte_carlo
evaluation/— evaluator, detailed_reporter
visualization/— dashboard, portfolio_visualizer
monitoring/system_monitor.py— System monitor
▼data/— Input/output
raw/— Turkey_Energy_Technology_Metrics.parquet, capacity projection
processed/— integrated_data.parquet, state_space_data.parquet
▼outputs/— Results
models/— *.pth checkpoints
comparison/comprehensive_comparison.json— 13-algorithm comparison
2050_final_portfolio.csv— Final portfolio
yearly_evolution_summary.csv— Year-by-year evolution

System Code Processing Architecture

How the Forecast and Reinforcement modules execute: entry points, execution flow, core modules, and data integration.

Forecast (sources/Forecast)

Entry point
advanced_neural_physical_forecasting.py
Invocation
cd sources/Forecast && python advanced_neural_physical_forecasting.py
Execution flow
  1. Load data: pd.read_excel() → Turkey_Renewable_Energy_Annual_Dataset.xlsx (2000–2024, 106 vars)
  2. Feature engineering: make_enhanced_features(df) → lags, growth, CF, GDP, demand, RES share
  3. HNSM: main() → S-curve + CapacityForecastingNN + CapacityFactorNN, G = K × CF × 8760/1000
  4. ENSM: run_enhanced_neural_scurve_model() → S-curve fit with neural fallback
  5. Benchmarking: run_benchmarking_models() → ARIMA, LSTM, RF, XGBoost, Prophet, SARIMA, VAR, CNN-LSTM, Transformer (30+ models)
  6. Visualization: run_dual_model_visualization() → coordinator → 18 visualizers, 163 figures
Inputs
Turkey_Renewable_Energy_Annual_Dataset.xlsx
Outputs
outputs/comprehensive_results_*.json; outputs/advanced_neural_physical_2024_2050.csv; outputs/01_Figures/ (18 categories, 163 PNGs); outputs/04_Raw_Data/ (CSV, JSON)

Reinforcement (sources/Reinforcement)

Entry point
main.py
Invocation
python main.py --mode <mode>
Modes
data, train, full, comparison, test-all, flexible, dqn, a3c, sac, td3, ga, pso, de, fuzzyql, neurofuzzy, marl
Execution flow
  1. Data pipeline: DataLoader → integrated_data.parquet; DataProcessor → state_space_data.parquet; DataValidator → validation_report.json
  2. Environment: TurkeyEnergyTransitionEnv (state 446-dim, action 119-dim, 26 steps, 2024–2050)
  3. Cost & reward: CostCalculator (CAPEX/OPEX); RewardCalculator (multi-objective); TechnologyWeights
  4. Training: PPOTrainer or selected agent (PPO, DQN, A3C, SAC, TD3, GA, PSO, DE, FuzzyQL, NeuroFuzzy, MARL, MAML, ProgressiveNN)
  5. Comparison: ComprehensiveComparison → 13 algorithms benchmarked
  6. Analysis: run_analysis_and_visualization.py → statistical, sensitivity, Monte Carlo, reports, visualizations
Inputs
data/raw/Turkey_Energy_Technology_Metrics.parquet, data/raw/Turkey_Power_Installed_Capacity_Projection_2024_2050.parquet
Outputs
data/processed/integrated_data.parquet; data/processed/state_space_data.parquet; outputs/models/*.pth; outputs/comparison/comprehensive_comparison.json; outputs/2050_final_portfolio.csv; outputs/yearly_evolution_summary.csv; outputs/visualizations/

Core modules

Forecast
  • advanced_neural_physical_forecasting.py — Main script: HNSM + ENSM dual-model forecasting
  • professional_visualization/coordinator.py — Orchestrates all 18 visualization modules
  • professional_visualization/main_figures.py — Capacity, generation, methodology figures
  • professional_visualization/forecast_analysis.py — Forecast analysis plots
  • professional_visualization/performance_analysis.py — Benchmarks, convergence, metrics
  • professional_visualization/model_parameters.py — Hyperparams, SHAP, activation
  • professional_visualization/validation_results.py — Validation statistics
  • professional_visualization/literature_comparison.py — Random Forest, methodology comparison
  • professional_visualization/scientific_rigor.py — Power analysis, statistical tests
  • professional_visualization/data_quality.py — Data validation
  • professional_visualization/time_series_analysis.py — Decomposition, trend, ACF, PACF
  • professional_visualization/advanced_statistics.py — Bootstrap, Monte Carlo
Reinforcement
  • main.py — Main launcher, CLI, orchestration
  • src/data/loader.py — Load raw parquet, integrate datasets
  • src/data/processor.py — Build state space (446-dim)
  • src/data/validator.py — Validate data, report
  • src/data/structures.py — StateSpace, ActionSpace definitions
  • src/environment/energy_env.py — TurkeyEnergyTransitionEnv
  • src/agent/ppo_agent.py — PPO agent
  • src/agent/mappo_agent.py — MAPPO multi-agent
  • src/cost/cost_calculator.py — CAPEX, OPEX, LCOE
  • src/reward/reward_calculator.py — Multi-objective reward
  • src/training/trainer.py — PPOTrainer training loop
  • src/comparison/comprehensive_comparison.py — 13-algorithm comparison
  • src/evaluation/evaluator.py — Model evaluation
  • run_analysis_and_visualization.py — Post-training analysis, stats, sensitivity, Monte Carlo

Integration flow

FromToData
ForecastReinforcementCapacity/generation projections, 2024–2050
ReinforcementForecastPolicy-induced CAPEX updates, learning curves
ReinforcementWebforecast_original.json, forecast_projection.json, reinforcement_unified.json
scriptsWebexport-explorer-data.py → apps/web/data/explorer/*.json

Forecast ↔ RL Bidirectional Integration

Forecast provides projections → RL optimizes policy → Policy updates Forecast scenarios. 4–7 iterations to convergence, 89% gap reduction.

Technology Stack

ModulePythonPyTorchOther
Forecast3.9+✓NumPy, SciPy, scikit-learn, matplotlib, seaborn, emoji
Reinforcement3.x✓NumPy, pandas, scipy, scikit-learn, PyYAML, pyarrow, psutil
Web——Next.js 15, React 19, Recharts

Five-Layer Architecture (Details)

Layer 1: Data & Physical Constraints
Historical Data
Forecast/data, Turkey_Energy_Technology_Metrics
  • Generation & capacity data 2000–2024
  • TEİAŞ, MENR, IRENA, IEA sources
  • 106 columns, 25 annual records
Years: 2000–2024Columns: 106
Demand Trajectories
Thesis conceptual_architecture_enhanced.mmd
  • 327 TWh (2024) → 649 TWh (2050)
  • 98% growth over horizon
  • 3.5% GDP elasticity assumption
2024: 327 TWh2050: 649 TWh
Resource Potentials
Reinforcement/data, energy_env suitable_regions
  • Solar: 1200–2100 kWh/m²/yr
  • Wind: 5.5–8.5 m/s at 100m
  • Hydro: 58 GW · Geothermal: 4.5 GW
Solar: 380 GW techWind: 165 GW tech
Physical Constraints
Reinforcement/src/environment/energy_env.py
  • Capacity factor bounds [0, 1]
  • Ramping limits, energy balance
  • Grid & storage constraints
State dims: 446Action dims: 119
Layer 2: Physics-Informed Forecasting (ENSM)
Technology Diffusion
Forecast/outputs, HNSM S-curve
  • Sigmoid/Gompertz functional forms
  • Technology diffusion theory (Rogers)
  • K_max=239.41 GW, α=0.0972, t₀=2024.39
K_max: 239.41 GWα: 0.0972 yr⁻¹
Physical Constraints
Forecast, energy balance G=K×CF×8760
  • Energy balance: G = K × CF × 8760
  • CF ∈ [0, 1], monotonicity
  • <0.1% violation vs 8.2% LSTM
Balance error: <0.1%LSTM error: 8.2%
Long-term Forecasts
Forecast/outputs/comprehensive_results
  • 2024–2050 (26 years)
  • 4-layer GRU, 512 hidden units
  • R²=0.9799, outperforms 17 benchmarks
R²: 0.9799Horizon: 26 yr
Uncertainty Quantification
Forecast, Monte Carlo Dropout
  • Monte Carlo Dropout (100 runs)
  • 95% prediction interval
  • PICP=94.2%, MPIW=11.8 GW
PICP: 94.2%MPIW: 11.8 GW
Layer 3: Multi-Agent RL (MAPPO)
State Space
Reinforcement/config/main.yaml, energy_env
  • 446 dimensions total
  • Tech-region capacity: 182 dims
  • Gen/demand profiles: 91 dims
Total: 446 dimsMax steps: 26
Action Space
Reinforcement/src/data/structures
  • 119 dimensions, 6 hierarchical levels
  • L1: Regional investment allocation
  • L2: Technology portfolio selection
Action dims: 119Levels: 6
Multi-Objective Reward
Reinforcement/config/main.yaml, reward_calculator
  • Cost minimization: -0.35
  • Emission reduction: +0.30
  • RES maximization: +0.20
Objectives: 6Agents: 7 regions
CTDE Coordination
Reinforcement/src/agent/mappo_agent.py
  • 7 regional agents (Marmara, Ege, Akdeniz, İç Anadolu, Karadeniz, Doğu Anadolu, Güneydoğu)
  • Centralized Training, Decentralized Execution
  • Superior to 12 alternative algorithms
Agents: 7Technologies: 26
Layer 4: Policy & Implementation
Investment Pathways
Reinforcement/outputs, PPO Run 1
  • Total: $226.79B (USD 2023)
  • 0.35% of projected GDP
  • 90% CI: [$218.4B, $235.2B]
Total: $226.79BGDP share: 0.35%
Capacity Expansion
Reinforcement/outputs/2050_final_portfolio.csv
  • 220 GW total by 2050
  • Solar 32% · Wind 28% · Hydro 27%
  • 26 technologies × 7 regions
2050 GW: 180.76Renewable: 97.34%
Emission Trajectories
Reinforcement/outputs/yearly_evolution_summary
  • 152 MtCO₂ (2024) → 14.30 MtCO₂ (2050)
  • 90.6% reduction
  • Exceeds 87% target by 3.6 pp
Reduction: 90.6%2050: 14.30 Mt
Regional Transition Plans
Reinforcement, 7 regions
  • 7 Turkish electricity regions
  • Region-specific capacity allocation
  • 26 technologies per region
Regions: 7Technologies: 26
Layer 5: Feedback & Uncertainty
Policy-Induced Learning
Thesis conceptual_architecture
  • Technology cost evolution
  • Accelerated deployment → learning curves
  • Cost reduction feedback loop
Forecast–Policy Alignment
Thesis, bidirectional integration
  • Bidirectional integration
  • Fixed-point iteration
  • 4–7 iterations to convergence
Iterations: 4–7Gap reduction: 89%
Uncertainty Propagation
Forecast MC Dropout, Reinforcement
  • Epistemic: MC Dropout (100 runs)
  • Aleatory: ±30% tech costs, ±20% demand
  • >90% RES maintained in all scenarios
MC runs: 100Worst case: 88.7%

Forecast Module (sources/Forecast)

Entry pointadvanced_neural_physical_forecasting.py
ModelsHNSM (Hybrid Neural-Sigmoid), ENSM (Enhanced Neural S-curve)
Input dataTurkey_Renewable_Energy_Annual_Dataset.xlsx (2000–2024, 106 vars)
Outputscomprehensive_results_*.json, advanced_neural_physical_2024_2050.csv, 18 figure categories
Key equationsG = K × CF × 8760/1000; S-curve fit (K_max, α, t₀), Monte Carlo dropout
Visualizationprofessional_visualization/ — coordinator, 18 visualizers

Reinforcement Module (sources/Reinforcement)

Entry pointmain.py
Configconfig/main.yaml — environment, reward, agent, training
Input dataTurkey_Energy_Technology_Metrics.parquet, Turkey_Power_Installed_Capacity_Projection_2024_2050.parquet
Environmentsrc/environment/energy_env.py — Turkey 2024–2050, state 446, action 119, max_steps 26
AlgorithmsPPO, DQN, A3C, SAC, TD3, GA, PSO, DE, FuzzyQL, NeuroFuzzy, MARL, MAML, ProgressiveNN (13 total)
Outputsoutputs/comparison/comprehensive_comparison.json, 2050 portfolio, yearly evolution
Key modulessrc/agent/, src/data/, src/reward/, src/cost/, src/comparison/, src/visualization/

Output Preview

comprehensive_results.json (sample)
{
  "hnsm": { "K_max": 239.41, "alpha": 0.0972, "t0": 2024.39 },
  "ensm": { "R2": 0.9799, "horizon": "2024-2050" },
  "projections": { "2024": {...}, "2030": {...}, "2050": {...} }
}
2050_final_portfolio.csv (sample)
region,technology,capacity_gw,share
Marmara,Solar,12.4,0.32
Marmara,Wind,8.2,0.21
Ege,Solar,18.1,0.28
...
config/main.yaml (excerpt)
environment:
  state_dim: 446
  action_dim: 119
  max_steps: 26
  start_year: 2024
  end_year: 2050
reward_weights:
  capacity_target: 0.25
  cost_optimization: 0.20
  renewable_share: 0.20

Data Flow & Integration

sources/Data/
├── forecast_data/     → original_dataset, advanced_neural_physical_2024_2050_*.csv
├── reinforcement_data/→ raw/*.parquet, processed/integrated_data, state_space_data
└── atb_data/          → ATBe.csv, 2024_v3_Workbook.xlsx

Forecast → 2000–2024 training → 2024–2050 projections → comprehensive_results.json
Reinforcement → projections + constraints → 13 algorithms → 2050_final_portfolio.csv

Data Explorer (scripts/export-explorer-data.py) → apps/web/data/explorer/*.json

Forecast vs Reinforcement

AspectForecastReinforcement
PurposeLong-term capacity/generation forecasting (2024–2050)Policy optimization for energy transition
ModelsHNSM, ENSM (S-curve + NN)13 algorithms (RL, evolutionary, fuzzy, meta-learning)
InputTurkey_Renewable_Energy_Annual_Dataset.xlsxTurkey_Energy_Technology_Metrics.parquet, capacity projection
Outputcomprehensive_results.json, figures, projection CSVcomparison JSON, 2050 portfolio, yearly evolution
RelationProvides forecasts and uncertaintyConsumes constraints and projections; optimizes pathways

Python Packages — Full List (Detailed)

All Python packages used in Forecast, Reinforcement, and scripts. Categorized, classified, and described in detail.

Core Scientific Computing
NumPy>=1.21Used in: Forecast, Reinforcement, scripts
Numerical arrays, linear algebra, array operations. Used for data structures and mathematical operations across all models.
Modules: numpy
pandas>=1.3Used in: Forecast, Reinforcement, scripts
DataFrames, CSV/Excel/Parquet I/O, time series processing. Used throughout the data loading and preprocessing pipeline.
Modules: pandas
SciPy>=1.7Used in: Forecast, Reinforcement
curve_fit (S-curve fitting), statistical tests (entropy, bootstrap), signal processing (periodogram).
Modules: scipy.optimize, scipy.stats, scipy.signal
Deep Learning
PyTorch>=1.12Used in: Forecast, Reinforcement
GRU, LSTM, Dense layers; PPO, SAC, DQN, A3C, TD3 agents; Monte Carlo Dropout; Adam optimizer.
Modules: torch, torch.nn, torch.optim, torch.nn.functional, torch.utils.data
Machine Learning & Preprocessing
scikit-learn>=1.0Used in: Forecast, Reinforcement
RobustScaler, StandardScaler, MinMaxScaler; train_test_split, TimeSeriesSplit; SimpleImputer; PCA; mutual_info_regression.
Modules: sklearn.preprocessing, sklearn.model_selection, sklearn.impute, sklearn.decomposition, sklearn.feature_selection
Visualization
Matplotlib>=3.5Used in: Forecast, Reinforcement
18 figure categories, Q1 journal standards, 300 DPI, FancyBboxPatch, ConnectionPatch, inset_axes.
Modules: matplotlib.pyplot, matplotlib.dates, matplotlib.patches, matplotlib.ticker, mpl_toolkits.axes_grid1
Seaborn>=0.11Used in: Forecast, Reinforcement
Statistical plots, model comparison charts, heatmap, distribution plots.
Modules: seaborn
Data I/O
pyarrowUsed in: Reinforcement, scripts
Parquet file read (pd.read_parquet). Turkey_Energy_Technology_Metrics.parquet, unified_energy_technology_dataset.parquet.
Modules: pyarrow
openpyxlUsed in: Forecast
Excel (.xlsx) okuma. Turkey_Renewable_Energy_Annual_Dataset.xlsx için pandas arka ucu.
Modules: openpyxl
Utilities & Config
PyYAMLUsed in: Reinforcement
config/main.yaml read. Environment, reward, agent and training parameters.
Modules: yaml
emojiUsed in: Forecast
Console emoji support (✅, ⚠️, 📊). Unicode character compatibility.
Modules: emoji
psutilUsed in: Reinforcement
CPU, memory monitoring. SystemMonitor (optional). Resource usage during training.
Modules: psutil
Recommended requirements.txt
# Core
numpy>=1.21
pandas>=1.3
scipy>=1.7

# Deep Learning
torch>=1.12

# Machine Learning
scikit-learn>=1.0

# Visualization
matplotlib>=3.5
seaborn>=0.11

# Data I/O
pyarrow
openpyxl

# Config & Utilities
pyyaml
emoji
psutil
Standard library: sys, os, locale, argparse, logging, json, pathlib, typing, datetime, warnings, pickle, hashlib, random, copy, collections, threading, time, dataclasses, enum, glob, csv