> \pJava Excel API v2.6 Ba==h\:#8X@"1Arial1Arial1Arial1Arial + ) , * `DC,title[*]contributor[author]contributor[advisor]keywords[*]date[issued] publisher citationsidentifier[uri]identifier[doi]abstractrelation[journal]relation[volume]relation[no]relation[page]m 8< ̸ \ T| xȌ$\2023. 2\ՑYPxhttp://hanyang.dcollection.net/common/orgView/200000654875;
https://repository.hanyang.ac.kr/handle/20.500.11754/179747;l |8 2019D 1ƀ0 4L m 280 !ΌX |ļ PM10 ̸| tǩX D ɉX. !Ό ļD X\ X p L 3X !ΌX D (| 3<\ $D L 0 9X t X | tǩX. \ spatio-temporal vine copula D tǩX X 1D 0<\ ɷX i U` | TX J PM10D !X.
0 ٳHX PM10̸ $x<\ , 4 | Ǵ t | $X U` | , |T \ ܭ | tǩX. \ \ ĳh \| LX AIC, BICX DP| t p| 0<\ X D 10@ 50\ D L| <\ X. U` ļ\ vine copula Dmlp| ǔ DtTH T|\ l1 ň.
spatio-temporal vine copula X ! \ 1D UxX0 t T|| X JŔ DP| X. 3 U` P ȴ \ ! T|X 1t JX. 췘 ǩX L$ T|| tǩƈ0 L8 X \ ! DP 1t XLD UxX.|In this study, analysis was conducted using daily average PM10 data from 280 observation stations nationwide from January to April 2019. Considering each observation station as the central location, the information of 9 space-time neighbors generated when the value of 3 observation stations close to each other and the time lag of 3 was considered was used. In addition, the PM10 value was predicted without modeling the multivariate joint distribution based on the correlation of each variable using the spatio-temporal vine copula model.
The PM10 data during the analysis period had a distribution with a long tail to the right. Considering this, the gumbel distribution, generalized extreme value distribution, and lognormal distribution were used for the
marginal distribution. In addition, through comparison of the value when the log-likelihood function is maximum, AIC, and BIC, analysis was conducted focusing on when the space was divided into 10 and 50 spaces based on distance. Vine copulas for each marginal distribution are composed of Archimedian copulas with an asymmetric structure.
For performance comparison on prediction, we considered a spatio-temporal model that does not use copula, and all three marginal distributions perform poorly in prediction of the entire value. But because copula, which is known to be useful for extreme values, is used, it was confirmed that all of them performed better than the comparison model.&HQsj&)KLng
<
dMbP?_*+%" ,,??U
>@
Root EntryWorkbookSummaryInformation(DocumentSummaryInformation8