Applying Predictive Modeling to Enhancing Solar Energy

Though analysis has been provided on how factors like temperature or humidity impact panel efficiency, there has not been as much research conducted on how the various environmental conditions all relate with each other to affect solar energy output real-time. Having a model that correlates several environmental predictor variables known to impact solar energy can help determine what adjustments need to be made to optimize solar panel performance. In this project, prediction models like artificial neural networks (ANN), multiple linear regression (MLR), elastic, ridge, and lasso were used for relating environmental variables like high temperature, outside humidity, or rain rate to the total solar energy output produced. Data containing thirty-three different weather measurements and their respective solar energy outputs was obtained from the UK Power Networks and will be used as the principal dataset for the models. To check linear model assumptions like normality of residuals or heteroscedasticity for models like MLR, several functions provided from a model performance package in R verified whether the assumptions for the model were met. Predictive model selection was based on cross validation with k = 10 and Mean Squared Error (MSE). Neural networks were the highest performing model, but lasso and elastic net were the most interpretable models in terms of how conditions affected energy output.