Conditional Variational Autoencoder Informed Probabilistic Wind Power Curve Modeling

Research output: Journal Publications and ReviewsRGC 21 - Publication in refereed journalpeer-review

2 Scopus Citations
View graph of relations

Related Research Unit(s)

Detail(s)

Original languageEnglish
Journal / PublicationIEEE Transactions on Sustainable Energy
Publication statusOnline published - 7 Jun 2023

Abstract

In this paper, a conditional variational autoencoder based method is proposed for the probabilistic wind power curve modeling task. To advance the modeling performance, the latent random variable is introduced to characterize underlying weather and wind turbine conditions. The infinite Gaussian mixture model is adopted to better model the asymmetric and heterogeneous conditional distribution of the wind power given the wind speed. The conditional variational autoencoder is composed of an encoder and a decoder network. The encoder infers the state of the latent random variable given the wind speed and wind power, while the decoder learns the observational conditional distribution of the wind power given the wind speed and latent variable. With a well-trained conditional variational autoencoder, the conditional probability density function of the wind power could be estimated through the decoder network by sampling the latent random variable from its prior distribution. Wind turbine supervisory control and data acquisition datasets are used in experiments to validate advantages of the proposed method. Experimental results show that the proposed method outperforms other benchmarking deterministic and probabilistic wind power curve models with the lower continuous ranked probability score and more reliable and sharper prediction intervals. Experiments also reflect the better robustness of the conditional variational autoencoder to data pre-processed using univariate or multivariate inputs, as well as its superiority and potential for the wind power estimation considering multivariate inputs. © 2023 IEEE.

Research Area(s)

  • Computational modeling, Data models, Data-driven models, deep neural networks, Predictive models, Probabilistic logic, probabilistic modeling, Random variables, wind power curve, Wind power generation, Wind speed