An interpretable statistical approach to photovoltaic power forecasting using factor analysis and ridge regression

Area of Science:

• Renewable Energy Systems
• Data Science and Analytics
• Meteorology

Background:

• Accurate solar energy forecasting is crucial for grid stability and renewable integration.
• Existing machine learning models, like LSTMs, often lack transparency and interpretability.
• Complex multivariate models pose challenges in development and understanding their predictive drivers.

Purpose of the Study:

• To present an interpretable by design photovoltaic (PV) forecasting framework.
• To couple Hierarchical Factor Analysis (HFA) with ridge regression for enhanced transparency.
• To simplify complex forecasting models and identify key drivers of solar power generation.

Main Methods:

• Utilized Hierarchical Factor Analysis (HFA) to reduce high-dimensional meteorological data into three key factors.
• Employed a single-parameter ridge regression model for coefficient-level transparency and regularization.
• Evaluated the framework using 15-minute solar power measurements from a PV plant in Adıyaman, Türkiye.

Main Results:

• The proposed framework achieved strong generalization capabilities in PV forecasting.
• The model provided clear, verifiable insights into how meteorological variables influence solar power output.
• Demonstrated the effectiveness of regression-based methods as explainable alternatives to deep learning.

Conclusions:

• The developed framework offers a transparent and explainable approach to photovoltaic forecasting.
• Regression-based methods can serve as robust and understandable alternatives to complex deep learning models.
• The study simplifies solar power forecasting by identifying critical meteorological drivers.