akwesalem-sk/Solar-Irradiance-forecasting-Minna-

Domain:

environment and energy

Record type:

model

Creator:

akw

Host:

ML model predicting solar irradiance in Minna, Nigeria using NASA POWER weather data — Random Forest, R² 0.706 # Solar Irradiance Forecasting — Minna, Niger State, Nigeria Machine learning model predicting daily solar irradiance in Minna, Nigeria using historical weather data, to support solar energy planning for off-grid and mini-grid operators. ## Problem Solar power output is highly variable due to weather. Off-grid and mini-grid solar operators in Nigeria need reliable irradiance estimates to plan storage management, diesel backup usage, and system sizing. This model predicts daily surface solar irradiance (ALLSKY_SFC_SW_DWN) from same-day weather conditions for Minna, Niger State. **Note on scope:** this model predicts irradiance given known same-day weather inputs (e.g. cloud cover, humidity). It is a conditional prediction model, not a future-forecasting system — true forecasting would require predicted (not historical/actual) weather inputs. This scope was a deliberate choice to keep the project hardware-agnostic and focused on the resource-estimation problem. ## Data - **Source:** NASA POWER API, daily temporal data - **Location:** Minna, Niger State, Nigeria (9.6178°N, 6.5569°E) - **Period:** 2015–2024 (3,653 daily records) - **Features:** clear-sky irradiance, temperature, relative humidity, cloud amount, wind speed, precipitation - **Target:** ALLSKY_SFC_SW_DWN (actual surface solar irradiance, kWh/m²/day) ## Method - Time-based train/test split (train: 2015–2022, test: 2023–2024) to respect temporal order - Baseline model: Random Forest Regressor - Comparison model: XGBoost Regressor ## Results | Model | MAE | RMSE | R² | |---|---|---|---| | Random Forest | 0.336 | 0.539 | 0.706 | | XGBoost | 0.343 | 0.547 | 0.698 | Random Forest performed marginally better as a baseline. **Feature importance (Random Forest):** 1. Cloud Amount — 49.8% 2. Clear-sky irradiance — 15.3% 3. Temperature — 12.4% 4. Relative Humidity — 7.6% 5. Day of year — 6.4% This aligns with the physical drivers of solar irradiance — cloud cover dominating makes physical sense and valid …

Visit

github.com

PROFBEN10/Physics-Informed-Neural-Network-for-Solar-Irradiance-Forecasting-in-Nigeria

Embedding thermodynamic laws into a neural network's loss function to produce accurate and physicall

Outperforming Self-Attention Mechanisms in Solar Irradiance Forecasting via Physics-Guided Neural Networks

Accurate Global Horizontal Irradiance (GHI) forecasting is critical for grid stability, particularly

Physics-Informed State Space Models for Reliable Solar Irradiance Forecasting in Off-Grid Systems

The stable operation of off-grid photovoltaic systems requires accurate, computationally efficient s

Machine Learning-Based Hourly Solar Irradiance Forecasting for Jimma City, Ethiopia: A Seasonal Performance Analysis

Python scripts (preprocessing.py and modeling.py) for the paper titled 'Machine Learning-Ba

Short-term and long-term solar irradiance forecasting with advanced machine learning techniques in Zafarana, Egypt

Abstract The increasing demand for renewable energy sources has positioned solar