MLBAM (Machine Learning-Based Albedo Modeling) is a comprehensive project that integrates advanced machine learning techniques to predict the impact of Earth’s albedo on solar cell performance, particularly focusing on Isc (short-circuit current). This repository builds on SPOOPA, a project that uses comprehensive data science and domain knowledge integration to model the performance of solar cells in space. While SPOOPA focuses on space-based solar cell modeling, it does not consider the effect of Earth’s albedo. MLBAM fills this gap by employing machine learning models to estimate the albedo effect on Isc.
- Data Preprocessing: Handles solar panel data, including creating lagged features for historical data points.
- Machine Learning Models: Incorporates Random Forest, LightGBM, and other models to predict Isc and improve albedo modeling.
- Visualization and Evaluation: Provides detailed plots comparing predicted and observed Isc values, with latitude and longitude influencing marker color and size.
- Performance Metrics: Evaluates model performance using R², MAE, and MAPE for accurate predictions.
- Time Series Prediction: Includes functionality to forecast Isc over time using past performance data and machine learning models.
- Random Forest Regressor
- LightGBM (with GPU support)
- Support Vector Regression (SVR)
- K-Nearest Neighbors (KNN)
The implementation of MLBAM significantly improves the accuracy of Isc predictions by incorporating data-driven learning about albedo impact. The error in prediction of Isc reduces from 1.69% to 0.29% with MLBAM, demonstrating a notable improvement in the model's performance.
As shown in the plot above, the MLBAM model accurately predicts Isc values, with the majority of predictions closely matching the observed values.