Solarwise Hybrid Solar Forecasting Model
Solarwise: Advanced PV Forecasting Model Project
Objective: Enhance the accuracy of short-term Photovoltaic (PV) forecasts to support solar power operators in grid integration, driving toward a more sustainable energy future.
Methodology:
Research Focus: Recognized the significant impact of cloud conditions on PV output. Aimed to produce precise 10-minute-ahead forecasts tailored to varying sky conditions.
Hybrid CNN/ResNet Model: Leveraged the image processing prowess of Convolutional Neural Networks (CNN) combined with the time-sensitive learning of Residual Neural Networks (ResNet). This synergy birthed condition-specific sub-models adept at predicting PV output based on sky images and past data.
Data & Training: Utilized the open-source Stanford Solar Forecasting dataset. The models trained on 600 sky images across four distinct cloud conditions: clear, partly-cloudy, mostly-cloudy, and overcast.
Key Outcomes:
Achieved Root Mean Squared Error (RMSE) values as follows: Clear (1.91), Partly-cloudy (1.72), Mostly-cloudy (1.13), and Overcast (4.48). The model excelled with mostly-cloudy conditions but faced challenges with overcast scenarios, likely due to the unpredictability of dense cloud cover.
Contribution & Future Scope: Solarwise’s hybrid approach aids PV system operators in optimizing energy performance. Future iterations could integrate more meteorological data to refine forecasting accuracy, priming the model for real-time applications.