Resampling Methods that Generate Time Series Data to Enable Sensitivity and Model Analysis in Energy Modeling

TL;DR

This paper introduces two non-parametric bootstrapping methods for generating multiple realistic time series from a single observed series, enabling sensitivity and robustness analysis in energy system modeling, especially under climate change scenarios.

Contribution

It presents and evaluates two simple, efficient methods for generating synthetic time series data that closely resemble original data and can be systematically altered for robustness testing.

Findings

Generated series closely match original data statistically and visually.

Synthetic series induce variability in key energy modeling properties.

Methods are suitable for sensitivity and robustness analysis in energy planning.

Abstract

Energy systems modeling frequently relies on time series data, whether observed or forecast. This is particularly the case, for example, in capacity planning models that use hourly production and load data forecast to occur over the coming several decades. This paper addresses the attendant problem of performing sensitivity, robustness, and other post-solution analyses using time series data. We explore two efficient and relatively simple, non-parametric, bootstrapping methods for generating arbitrary numbers of time series from a single observed or forecast series. The paper presents and assesses each method. We find that the generated series are both visually and by statistical summary measures close to the original observational data. In consequence these series are credibly taken as stochastic instances from a common distribution, that of the original series of observations. With climate change in mind, the paper further proposes and explores two general techniques for systematically altering (increasing or decreasing) time series. Both for the perturbed and unperturbed synthetic series data, we find that the generated series induce variability in properties of the series that are important for energy modeling, in particular periods of under- and over-production, and periods of increased ramping rates. In consequence, series produced in this way are apt for use in robustness, sensitivity, and in general post-solution analysis of energy planning models. These validity factors auger well for applications beyond energy modeling.