Utilization of machine‐learning algorithms for wind turbine site suitability modeling in Iowa, USA

Abstract

AbstractBecause of the current shift away from fossil fuels and toward renewable energy sources, it is necessary to plan for the installation of new infrastructure to meet the demand for clean energy. Traditional methods for determining wind turbine site suitability suffer from the selection of arbitrary criteria and model parameters by experts, which may lead to a degree of uncertainty in the models produced. An alternative empirically based methodology for building a wind turbine siting model for the state of Iowa is presented in the study. We employ ‘ecological niche’ principles traditionally utilized to model species allocation to develop a new multicriteria, spatially explicit framework for wind turbine placement. Using information on suitability conditions at existing turbine locations, we incorporate seven variables (wind speed, elevation, slope, land cover, distance of infrastructure and settlements, and population density) into two machine‐learning algorithms [maximum entropy method (Maxent) and Genetic Algorithm for Rule Set Prediction (GARP)] to model suitable areas for installation of wind turbines. The performance of this method is tested at the statewide level and a six‐county region in western Iowa. Maxent and GARP identified areas in the Northwest and North Central regions of Iowa as the optimum location for new wind turbines. Information on variable contributions from Maxent illuminates the relative importance of environmental variables and its scale‐dependent nature. It also allows validating existing assumptions about the relationship between variables and wind turbine suitability. The resultant models demonstrate high levels of accuracy and suggest that the presented approach is a possible methodology for developing wind turbine siting applications. Copyright © 2014 John Wiley & Sons, Ltd.