In the quest to harness the power of renewable energy sources, the journey begins with the collection of valuable data. In this pursuit, historical solar generation data was meticulously gathered, laying the foundation for our exploration. We didn't stop there; we recognized the importance of merging this solar generation data with weather information from a different source. This fusion of datasets allowed us to unlock deeper insights into the relationship between solar energy production and environmental conditions.
The project aims to predict solar power generation from solar panels using historical data. Beyond forecasting short-term (next-day) power generation, the objective is to extend predictions to encompass long-term (weekly or monthly) power generation trends, seasonal patterns and the factors that influence the forecasts. To achieve this goal, data manipulation was necessary to enable accurate long-term predictions.
In a comprehensive data analysis project, I collected historical solar generation data from the E.W. Brown Solar Farm, ensuring the data's cleanliness and reliability. To enrich the dataset, I integrated it with pertinent weather data obtained from the "Visual Crossing" website. This amalgamation allowed for a more holistic understanding of the factors influencing solar energy production. As part of data preprocessing, I meticulously cleaned and standardized the dataset, and further employed Principal Component Analysis (PCA) to effectively reduce the data's dimensions. Following this, I employed a range of predictive models including Random Forest (RF), XGBoost, a hybrid Convolutional Neural Network and Long Short-Term Memory model (CNN+LSTM), and Light Gradient Boosting Machine (LGBM). The performance of these models was evaluated using R-squared (R2) score, Mean Absolute Error (MAE), and Root Mean Square Error (RMSE). For a more engaging and informative exploration of the data, I created interactive visualizations and dashboards using Tableau, allowing stakeholders to gain valuable insights into solar energy generation trends.
Additionally, to facilitate user access and real-time predictions, I designed a user-friendly interface using Streamlit. This interactive Streamlit-based UI empowers users to input their own data and obtain power generation predictions from the deployed predictive model. This user-centric approach not only streamlines the process of generating solar power forecasts but also enhances accessibility for individuals interested in renewable energy planning. Overall, this project represents a comprehensive data analysis pipeline, encompassing data collection, preprocessing, integration, model building, evaluation, and interactive visualization, to offer a robust solution for understanding and forecasting solar energy generation while ensuring seamless user access to the predictive capabilities of the developed models.
The project relies on historical data, and all the models trained on this data are saved as pickle files. Due to the large volume of these files, uploading them to GitHub was not possible. The other notebook files encompass data preprocessing, standardization, feature reduction, modeling, hyperparameter tuning, evaluation, and visualization processes.