Wind power scenario generation with non-separable spatio-temporal covariance function and fluctuation-based clustering

Tan, Jin; Wu, Qiuwei; Zhang, Menglin; Wei, Wei; Hatziargyriou, Nikos D.; Liu, Feng; Konstantinou, Theodoros

doi:10.1016/j.ijepes.2021.106955

Cited by 10 publications

(1 citation statement)

References 36 publications

(76 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The support vector classifier (SVC) predicts data labels in this approach. The wind power scenarios are generated by integrated non-separable spatiotemporal covariance function and fluctuation-based clustering [14]. The historical data is grouped into clusters with different fluctuations using the K-means clustering algorithm to estimate the covariance matrix precisely.…”

Section: Introductionmentioning

confidence: 99%

Wind Power Scenario Generation Considering Spatiotemporal Correlations: A Distribution Free Hybrid VARMA-Copula Approach

Sharma,

Gupta,

Bhakar

2023

Smart Grids and Energy

View full text Add to dashboard Cite

Wind power scenario forecast is a primary step for probabilistic modelling of power systems' operation and planning problems in stochastic programming framework considering uncertainties. Several models have been proposed in the literature to generate wind power scenarios using statistical and machine learning approaches. Most of these models are univariate and do not consider dependency between the Wind Farms (WFs), resulting in generated scenarios lacking spatiotemporal correlation. Furthermore, most scenario generation approaches assume a parametric distribution (i.e., Normal or Weibull distribution) for wind power uncertainty, which is unrealistic. This paper proposes a novel distribution-free hybrid approach that combines multivariate Vector Autoregressive Moving Average (VARMA) and Copula models to generate wind power scenarios. The VARMA model is used to forecast wind power considering spatiotemporal correlation, and then a regular vine (R-Vine) copula is applied on WFs' residuals to retain spatial correlation. The proposed approach is independent of the distribution type because the R-Vine copula can decompose residuals into a copula function and a marginal distribution. The proposed model is illustrated through a realistic case study based on nine Australian WFs. The results obtained are compared with benchmark models that show the efficacy of the proposed model for the generation of wind power scenarios. Minimum energy score, nearly accurate Kendall's correlation, and cross correlation plots show that the proposed method can produce high-quality wind power scenarios without sacrificing spatiotemporal correlation or making distribution assumptions. Generated scenarios using the proposed approach can help WFs and system operators improve decisions in the stochastic programming framework.

show abstract

Section: Introductionmentioning

confidence: 99%