keithpaulkato/solar-radiation-drift-correction-ml

Machine learning and deep learning approach for reconstructing incoming shortwave solar radiation under sensor drift conditions in African weather stations. # Solar Radiation Drift Correction Using Machine Learning and Deep Learning ## 📌 Project Overview This project addresses the challenge of reconstructing accurate **incoming shortwave solar radiation measurements** in the presence of sensor drift. Using data from the **Trans-African Hydro-Meteorological Observatory (TAHMO)** network, the study applies both **machine learning** and **deep learning techniques** to predict solar radiation values at 15-minute intervals, particularly for periods where sensor readings are unreliable or missing. --- ## 🎯 Problem Statement Solar radiation sensors deployed in weather stations degrade over time due to environmental exposure, resulting in **nonlinear drift and measurement bias**. This affects: * Climate monitoring * Agricultural planning * Renewable energy forecasting The objective of this project is to develop a **data-driven model** capable of reconstructing accurate solar radiation values using available meteorological variables. --- ## 🎯 Objectives ### General Objective To develop a predictive system for reconstructing solar radiation data under sensor drift conditions. ### Specific Objectives * Perform exploratory data analysis on time-series weather data * Preprocess and clean multi-station meteorological datasets * Engineer temporal and environmental features * Implement baseline and advanced regression models * Develop deep learning models for time-series prediction * Evaluate and compare model performance --- ## 🧪 Dataset Description The dataset is obtained from the **TAHMO Solar Radiation Prediction Challenge (Zindi)**. ### Key Characteristics: * Data recorded at **15-minute intervals** * Collected from **50 weather stations** * Includes: * Solar radiation (target variable) * Temperature * Relative humidity * Rainfall * Training data: **odd months** * Test data: **even months (to be predicted)** --- ## 🔁 Machine Learning Workflow 1. Data Collection 2. Exploratory Data Analysis (EDA) 3. Data Cle …