Abstract

The importance of incorporating underlying theory and domain knowledge while building artificial intelligence-based predictive models is examined. Using the context of predicting intraday solar radiation, we show that a theoretically grounded predictive model yields better performance and offers more interpretability and generalizability than a model that relies solely on other variables. Inclusion of theoretically-guided variables in data-driven predictive models is proposed as a means to mitigate overfitting and reduce potential bias.

Share

COinS
 

Theory-Guided AI for Intraday Solar Radiation Prediction

The importance of incorporating underlying theory and domain knowledge while building artificial intelligence-based predictive models is examined. Using the context of predicting intraday solar radiation, we show that a theoretically grounded predictive model yields better performance and offers more interpretability and generalizability than a model that relies solely on other variables. Inclusion of theoretically-guided variables in data-driven predictive models is proposed as a means to mitigate overfitting and reduce potential bias.