Machine learning reduces soft costs for residential solar photovoltaics

Area of Science:

• Energy Policy
• Environmental Science
• Machine Learning Applications

Background:

• Rooftop solar photovoltaic (PV) deployment is hindered by high soft costs, particularly customer acquisition expenses.
• Current technology adoption studies predominantly use significance-based methods like logistic regression.
• Reducing soft costs is crucial for expanding solar energy accessibility and market penetration.

Purpose of the Study:

• To evaluate the effectiveness of prediction-oriented machine learning models versus significance-based methods for identifying solar PV adopters.
• To quantify the impact of improved adoption prediction on reducing customer acquisition costs for solar companies.
• To explore the potential for machine learning to uncover new market opportunities in the clean energy sector.

Main Methods:

• Employed machine learning algorithms to predict solar photovoltaic (PV) adoption.
• Compared the predictive performance of machine learning models against logistic regression.
• Analyzed the impact of enhanced prediction accuracy on customer acquisition costs.

Main Results:

• Machine learning models demonstrated substantial improvements in adoption prediction accuracy.
• True positive rate for predicting adopters increased from 66% to 87%.
• True negative rate for predicting non-adopters rose from 75% to 88%, reducing customer acquisition costs by 15% ($0.07/Watt).

Conclusions:

• Machine learning offers a superior approach to predicting solar PV adoption compared to traditional methods.
• Accurate adoption prediction via machine learning can significantly lower customer acquisition costs and expand solar markets.
• Findings have implications for the adoption of other clean energy technologies and related policy challenges.