Data Science in Energy and Environmental Systems With Multiple Data Sources

Abstract

The energy sector in modern society is undergoing a rapid transformation from fuel-based generation to renewable generation. Further, distributed supply and demand, grid-responsive demand management, and other complex technology increasingly rely on environmental data throughout the energy supply chain, from power production to end uses. However, data analysis in energy systems presents major technical challenges, including spatio-temporal heterogeneity, localized characteristics, and disparate data sources.

This dissertation study aims to design data science solutions that address some of these challenges and model the dynamics of ambient environmental conditions that are closely tied with energy system operations. Climate conditions are often temporally and spatially correlated and exhibit a non-stationary nature, constantly changing all the time. To fully characterize these characteristics, this study utilizes the rich data available from multiple sources, including data collected at spatially distributed locations and data generated from disparate sources, e.g., field meteorological data and physics-based numerical weather prediction data.

This dissertation initiates two major ideas: (a) making use of data collected from multiple spatially dispersed locations; and (b) integrating data generated from physics-based numerical weather prediction models with actual climate measurements through a linkage function. Specifically, the following three research topics are investigated.

The first study develops a probabilistic model for assessing wind resources at a target location by utilizing wind data collected at nearby meteorological stations. By quantifying daily and spatially correlated diurnal patterns of the wind speed at multiple locations, the developed integrative approach provides compelling capabilities for evaluating the wind variability at non-observational locations, while quantifying prediction uncertainties. The results will provide rich information for deciding suitable wind farm sites.

Next, we make use of temperature projections from physics-based global climate models for the purpose of long-term electricity load predictions. While the physics-based climate models can provide useful climate projections in the long run, they inevitably exhibit systematic discrepancies (also called `bias'), compared to actual climate conditions, because of incomplete characterization of local or regional variations. We calibrate the climate model projections to address possible biases and provide a long-term density prediction of peak electricity load. The results provide useful insights on how the daily peak demand densities would change over time, in response to climate change and other socio-economic factors.

Finally, we present another bias correction model that quantifies the spatially and temporally correlated bias from the physics-based numerical model output for urban temperature modeling. By combining both types of data, our approach can successfully characterize localized environmental variations over space and time and greatly improve the prediction accuracy compared to that of the original physics-based numerical model. hl{The proposed approach helps understanding temperature variations over dispersed locations, depending on urbanization intensity. Such results can be useful for predicting electricity demand and effectively managing power system operations such as demand response programs.}

The advantages of all proposed approaches are demonstrated with case studies using actual data. The results validate that the proposed approaches successfully address some of the challenges discussed above that arise in energy and environmental systems.