A Review on Time Series Aggregation Methods for Energy System Models

Abstract

Due to the high degree of intermittency of renewable energy sources (RES) and the growing interdependences amongst formerly separated energy pathways, the modeling of adequate energy systems is crucial to evaluate existing energy systems and to forecast viable future ones. However, this corresponds to the rising complexity of energy system models (ESMs) and often results in computationally intractable programs. To overcome this problem, time series aggregation (TSA) is frequently used to reduce ESM complexity. As these methods aim at the reduction of input data and preserving the main information about the time series, but are not based on mathematically equivalent transformations, the performance of each method depends on the justifiability of its assumptions. This review systematically categorizes the TSA methods applied in 130 different publications to highlight the underlying assumptions and to evaluate the impact of these on the respective case studies. Moreover, the review analyzes current trends in TSA and formulates subjects for future research. This analysis reveals that the future of TSA is clearly feature-based including clustering and other machine learning techniques which are capable of dealing with the growing amount of input data for ESMs. Further, a growing number of publications focus on bounding the TSA induced error of the ESM optimization result. Thus, this study can be used as both an introduction to the topic and for revealing remaining research gaps.

Efficient Quantification of the Impact of Demand and Weather Uncertainty in Power System Models

“…Finally, the method could leverage the use of subsampling for point estimates. Due to computational limitations, even point estimate PSM outputs are often determined using a subsampled time series such as a smaller number of "representative days" [41], [42]. Such approaches can be straightforwardly combined with the algorithm proposed in this paper by using subsampled data for the point estimate.…”

Importance subsampling for power system planning under multi-year demand and weather uncertainty

“…Power system modelers employ various schemes to reduce temporal complexity, as discussed in detail in [14]. A popular approach involves subsampling, often by clustering, into a smaller number of representative periods (typically days) that seek to encode longer time series in reduced form [15]- [19].…”

“…Most such reduction approaches are what the authors of [14] call a priori (or input-based) in that they depend only on the input time series but not on the underlying power system model. For example, clustering the same demand and weather data generates the same representative days (ignoring any randomness in the clustering algorithm) irrespective of whether these days are subsequently used in a highly renewable system or a system with only fossil fuel generation.…”

“…This "smoothes out" extremes and may lead to (1) overly optimistic predictions of system costs and (2) system design estimates that are not robust to such events. For example, such approaches may underestimate the amount of required backup generation capacity in events with high demand but low renewable generation, leading to systems with a high risk of blackouts [14]. For this reason, users sometimes make a priori heuristic adjustments such as including the maximum demand day [8].…”

“…As a benchmark, the model is also run across samples generated by k-medoids clustering, representing the current industry standard. The reason for choosing k-medoids clustering (where the real day closest to the cluster mean is used as the representative day, instead of the mean itself) is its widespread use in previous studies and that it typically performs better than k-means [14], which smooths out extremes by using the cluster mean. Two metrics indicating the "degree of suboptimality" of estimated system design (compared with the true optimum across the full time series) are also calculated:…”