Solar Power Prediction through AI Models
Duration
Place of Work
Project Type
Project Overlook
The paper explores the prediction of solar power generation using machine learning algorithms and real-time weather data. Accurate solar power prediction is essential for effective planning and integration of photovoltaic (PV) systems in the electricity grid. The study implements various machine learning models to identify the best algorithm for solar power forecasting.
Objective
To develop a real-time solar power prediction system using weather data and machine learning algorithms, improving the accuracy and reliability of solar power output forecasting.
Methodology
- Data Collection: Real-time weather data (temperature, humidity, pressure, irradiance) was collected using sensors and OpenWeatherMap API.
- Machine Learning Models: Several machine learning algorithms were tested, including Linear Regression, Artificial Neural Networks (ANN), Random Forest, and Long Short-Term Memory (LSTM).
- Real-Time System:
- Hardware setup with NodeMCU and sensors for data acquisition.
- Data stored in Firebase and displayed in a mobile application using Blynk and Android Studio.
- Predictions made in Google Colab for better computational efficiency.
Key Findings
- Random Forest provided the highest accuracy (81.31%) with the lowest mean squared error (MSE: 0.0138), making it the most suitable algorithm for real-time solar power prediction.
- LSTM Networks improved prediction accuracy compared to standard ANN models, avoiding the vanishing gradient problem.
- Real-Time Forecasting: The real-time data acquisition system successfully logged weather parameters and PV data, allowing accurate predictions and visual representation on a mobile app.
Results
- Random Forest outperformed other algorithms, achieving the best accuracy for solar power forecasting.
- The mobile app provided users with real-time data visualization and future solar power predictions, enhancing usability and accessibility.
Conclusion
The study concludes that Random Forest Regression is the most reliable algorithm for solar power prediction in real-time systems. Integrating this model with a real-time data acquisition setup and mobile application offers a practical solution for better solar power planning and utilization.
DISCLAIMER
All projects showcased on this website are academic projects completed as part of coursework, research, or personal learning. These are not commercial projects affiliated with any company or organization.
Softwares Used
Excel (Advanced)
Python
MATLAB
Collaborators
Rohit Kumar Mahto
Harish Ragaavendra