Modelling complex investment decisions in Germany for renewables with different machine learning algorithms. (August 2019)

• Record Type: Journal Article
• Authors: Frey, Ulrich J.; Klein, Martin; Deissenroth, Marc
• Is Part Of: Environmental modelling & software. Volume 118(2019)
• Page Start: 61
• Page End: 75
• Publication Date: 2019-08
• Subjects: Investment decisions -- Solar installation -- Renewable energy -- Deep learning -- Machine learning -- Germany; Environmental monitoring -- Computer programs -- Periodicals; Ecology -- Computer simulation -- Periodicals; Digital computer simulation -- Periodicals; Computer software -- Periodicals; Environmental Monitoring -- Periodicals; Computer Simulation -- Periodicals; Environnement -- Surveillance -- Logiciels -- Périodiques; Écologie -- Simulation, Méthodes de -- Périodiques; Simulation par ordinateur -- Périodiques; Logiciels -- Périodiques; Computer software; Digital computer simulation; Ecology -- Computer simulation; Environmental monitoring -- Computer programs; Periodicals; Electronic journals; 363.70015118
• DOI: 10.1016/j.envsoft.2019.03.006
• Languages: English
• ISSNs: 1364-8152

Abstract

Abstract: Investment decisions in renewable energies are known to be influenced by many diverse drivers, e.g. social, political, geographic, economic and psychological. Non-comprehensive models are problematic since missed interactions might introduce bias. We implement a robust modelling approach by (1) using a large data set with 1.4 million solar installations and (2) three different machine learning algorithms (deep neural networks, gradient boosting, random forests). Generalized linear models serve as baseline and comparison. A high prediction accuracy can be achieved on the county level with deep neural networks (adjusted R 2 = 0.86) and gradient boosting (adjusted R 2 = 0.87). The most important drivers are population per county, followed by type of urbanisation and social variables like unemployment, with varying degree of importance for the different machine-learning algorithms. Our approach points out both differences and agreements across methods and therefore a higher confidence in their interpretation. Highlights: One of the first comprehensive models for investment decisions in photovoltaics. Applies gradient boosting achieving very high explanatory power (adj. R2 = 0.87). Provides robust estimates by using neural networks, random forests as comparison. Most important drivers are demographic.