Evaluating predictive models for solar energy growth in the US states and identifying the key drivers

[摘要] Driven by a desire to control climate change and reduce the dependence on fossil fuels, governments around the world are increasing the adoption of renewable energy sources. However, among the US states, we observe a wide disparity in renewable penetration. In this study, we have identified and cleaned over a dozen datasets representing solar energy penetration in each US state, and the potentially relevant socioeconomic and other factors that may be driving the growth in solar. We have applied a number of predictive modeling approaches - including machine learning and regression - on these datasets over a 17-year period and evaluated the relative performance of the models. Our goals were: (1) identify the most important factors that are driving the growth in solar, (2) choose the most effective predictive modeling technique for solar growth, and (3) develop a model for predicting next year's solar growth using this year's data. We obtained very promising results with random forests (about 90% efficacy) and varying degrees of success with support vector machines and regression techniques (linear, polynomial, ridge). We also identified states with solar growth slower than expected and representing a potential for stronger growth in future.

[发布机构] Department of Mathematics and Computer Science, Lake Forest College, 555 North Sheridan Road, Lake Forest; IL; 60045, United States^1

[效力级别] 生态环境科学 [学科分类] 环境科学（综合）

[关键词] Energy growth;Predictive modeling;Predictive models;Random forests;Regression techniques;Relative performance;Renewable energy source