Microgrid Optimization for Rural Electrification in Myanmar

An advanced machine learning framework for sustainable rural electrification in Myanmar, combining renewable energy forecasting, demand modeling, and multi-objective capacity optimization.

Link to the main repository on DagsHub here

Project Overview

This project develops an intelligent microgrid optimization system specifically designed for Myanmar's rural electrification challenges. By leveraging AI-driven forecasting models and sophisticated optimization algorithms, the system determines optimal configurations for solar PV, battery storage, and diesel generators to provide reliable and cost-effective electricity access.

Key Achievements

Forecasting Models Performance

• Prophet Model: 91.8% accuracy (R² = 0.918) for 24-hour solar generation forecasting
• XGBoost Model: 91.3% accuracy (R² = 0.913) with excellent out-of-time performance
• LSTM Model: 70.22% accuracy with deep learning approach
• SARIMA Model: 49.6% accuracy with traditional time series methods

System Optimization Results

• 100% Load Coverage: Zero loss of load events in simulation
• Optimal Resource Allocation: Efficient PV-battery-diesel hybrid systems
• Cost Minimization: LCOE-optimized capacity planning
• Renewable Integration: High renewable energy penetration with minimal curtailment

Data Analysis Insights

• Population Coverage: Analysis of 303,758 population clusters nationwide
• Investment Analysis: Region-specific cost projections for 2020-2030
• Hydropower Potential: 45+ years of water discharge data across 7 river stations
• Seasonal Patterns: Comprehensive understanding of Myanmar's energy demand cycles

Technical Architecture

Core Components

• Data Collection: Supabase integration, Renewables Ninja API
• Demand Modeling: RAMP-based stochastic appliance usage simulation
• Generation Forecasting: Multi-model ensemble with XGBoost and Prophet
• Capacity Optimization: Differential evolution algorithm with energy balance constraints
• System Simulation: MPC-controlled microgrid operation

Key Features

• Myanmar-specific appliance usage patterns and demographics
• Weather-dependent seasonal demand modeling
• Real-time optimization with battery state-of-charge constraints
• Cost-optimized CAPEX/OPEX minimization
• Scalable village cluster analysis framework

Results & Impact

Model Performance Comparison

Model	RMSE (kW)	MAE (kW)	R² Score	Accuracy
Prophet	0.067	0.049	0.918	91.8%
XGBoost	0.070	0.037	0.913	91.3%
LSTM	3.90	2.48	-	70.2%
SARIMA	0.168	0.091	0.496	49.6%

Regional Electrification Analysis

• Highest Investment Regions: Bago, Ayeyawaddy, Yangon
• Average Demand: 308.8 kWh/capita/year
• Renewable Potential: Significant solar and hydropower resources identified

System Configuration Optimization

• Battery Efficiency: 95% round-trip efficiency modeling
• Diesel Backup: Cost-optimal sizing for reliability
• PV Capacity: Weather-adjusted generation profiles
• Load Forecasting: Village-scale demand prediction

Repository Structure

├── src/
│   ├── data_collection/          # External API integrations
│   ├── data_wrangling/           # Data preprocessing utilities
│   ├── model_development/        # Optimization and forecasting
│   │   └── optimization/         # Core optimization engine
│   ├── model_deployment/         # Production APIs
│   └── utils/                    # Shared utilities
├── notebooks/                    # Analysis and experimentation
│   ├── LSTM_FORECASTING.ipynb
│   ├── XGBoost_24hourprediction.ipynb
│   ├── PROPHET.ipynb
│   ├── SARIMA.ipynb
│   └── grdc_analysis.ipynb
├── docs/                         # Documentation

Quick Start

Prerequisites

pip install -r src/model_development/optimization/requirements.txt

Environment Variables

export SUPABASE_KEY="your_supabase_api_key"
export RENEWABLES_NINJA_API_TOKEN="your_renewables_ninja_token"

Run Optimization

from src.model_development.optimization.index import run

# Optimize microgrid for village cluster 308 over 2 days
optimal_pv, optimal_battery, optimal_diesel, dispatch = run(
    cluster_id=308, 
    num_days=2, 
    start_date="2024-09-12"
)

Key Findings

1. Renewable Energy Viability: High solar generation potential across Myanmar with predictable seasonal patterns
2. Optimal System Design: Hybrid PV-battery-diesel systems provide most cost-effective solution
3. Forecasting Accuracy: Machine learning models achieve >90% accuracy for generation prediction
4. Investment Priorities: Data-driven identification of high-impact regions for electrification
5. Technical Feasibility: MPC-controlled microgrids demonstrate reliable operation under varying conditions

Methodology

Demand Modeling

• Stochastic appliance usage simulation based on Myanmar household surveys
• Seasonal cooling demand integration from weather data
• Village-scale aggregation with realistic household compositions

Generation Forecasting

• Multi-model ensemble approach for robust predictions
• Feature engineering with weather variables and lag features
• Out-of-time validation for real-world performance assessment

Capacity Optimization

• Multi-objective optimization balancing cost and reliability
• Energy balance constraints ensuring 100% load coverage
• Battery state-of-charge modeling with efficiency losses

Impact & Applications

This framework provides a replicable methodology for:

• Policy Planning: Evidence-based electrification strategies
• Investment Decisions: Risk-adjusted capacity planning
• System Design: Optimal microgrid configurations
• Resource Assessment: Renewable energy potential mapping

Contributing

Contributions welcome! This project demonstrates sustainable electrification approaches applicable across Southeast Asia and similar developing regions.