Development of data-driven insights using energy system models: A systematic scoping review

The prevalence of “What-if” scenario analyses has limited the types of insights that can be produced with energy system models, resulting in insufficient exploration of uncertainty and pathway diversity in the integrated energy system design space. Today, novel data science methods allow modellers to develop valuable insights from complex high-dimensional datasets; this form of analysis is appropriate for the complex questions increasingly posed by stakeholders. However, these methods have yet to be widely adopted, likely due to visibility challenges and perceived high computational cost of producing large results datasets. Identifying and systematizing the extant methods and the resources they require is necessary to promote their adoption. We conducted a systematic scoping review of studies that conduct a data-driven analysis of energy system model outputs. We identified 62 papers that met the inclusion criteria. Of the included manuscripts, there was substantial heterogeneity in modelling framework, analysis approach, and resource requirement, but the breadth of related subdomains indicates a growing role for data scientists in evaluating energy futures. We identified three major application areas: exploration of configurations and trade-offs, distribution of key outcomes under uncertainty, and advancement of modelling methodologies. Finally, we proposed a framework for scoping future data-driven analyses, including the potential role of surrogate models for reducing the computational requirement of high solution volume analyses. Inconsistent reporting practices still weaken the current body of literature; however, standardized reporting and further experimentation will enhance the utility of data-driven analyses, ultimately providing relevant and timely insights to stakeholders. • Data-driven analyses of large-scale energy system modelling results were reviewed. • Uncertainty and sensitivity analyses were the most common in included works. • Articles explored configurations, quantified uncertainty, or advanced model methods. • We propose a resource-based framework for scoping future results analyses. • Surrogate methods are emerging and may aid in producing robust sensitivity analyses.