Typical solar radiation year construction using k-means clustering and discrete-time Markov chain

Shuai Li, Hongjie Ma, Weiyi Li

Research output: Contribution to journalArticlepeer-review

196 Downloads (Pure)

Abstract

Daily solar radiation (DSR) fluctuation and transition rules affect the design of the energy storage system and online control strategy of solar energy utilisation systems. However, the current synthesis methods for the typical meteorological year do not emphasise such features of DSR. To overcome this shortcoming, this study presents an innovative synthesis method for a typical solar radiation year (TSRY) based on k-means clustering and discrete-time Markov chain (DTMC). The historical distributions of clear-sky ratio (CSR) in four representative regions were analysed, and a six-dimensional feature vector that represents the DSR fluctuations based on CSR was defined. Then, based on the feature vector, k-means clustering was used to cluster the historical DSR into four types. Subsequently, a DTMC-based model was built for transition rule estimation among the four types of solar radiation. Finally, the TSRY was established based on the clustering categories and transition rules among them. The innovative synthesis method was also verified in this study. Results for the four regions showed that the average error of the synthesised TSRY has maximum and minimum values of 10% and 6% in all seasons, respectively, compared with historical data. The proposed method could represent DSR fluctuation and transition characteristics of certain regions and could also be extended to other regions.
Original languageEnglish
Pages (from-to)720–731
JournalApplied Energy
Volume205
Early online date18 Aug 2017
DOIs
Publication statusPublished - 1 Nov 2017

Keywords

  • DSR
  • Typical solar radiation year
  • Clear-sky ratio
  • k-means clustering
  • Discrete-time Markov chain
  • Typical meteorological year

Fingerprint

Dive into the research topics of 'Typical solar radiation year construction using k-means clustering and discrete-time Markov chain'. Together they form a unique fingerprint.

Cite this