• Àüü
  • ÀüÀÚ/Àü±â
  • Åë½Å
  • ÄÄÇ»ÅÍ
´Ý±â

»çÀÌÆ®¸Ê

Loading..

Please wait....

±¹³» ÇÐȸÁö

Ȩ Ȩ > ¿¬±¸¹®Çå > ±¹³» ÇÐȸÁö > µ¥ÀÌÅͺ£À̽º ¿¬±¸È¸Áö(SIGDB)

µ¥ÀÌÅͺ£À̽º ¿¬±¸È¸Áö(SIGDB)

Current Result Document :

ÇѱÛÁ¦¸ñ(Korean Title) µö·¯´× ±â¹Ý ´Ùº¯·® ½Ã°è¿­ µ¥ÀÌÅÍ º¸Á¤±â¹ý
¿µ¹®Á¦¸ñ(English Title) Deep Learning-based Multivariate Time Series Data Correction Method
ÀúÀÚ(Author) Á¤ÇѼ®   ±èÇÑÁØ   Hanseok Jeong   Han-joon Kim  
¿ø¹®¼ö·Ïó(Citation) VOL 39 NO. 02 PP. 0047 ~ 0062 (2023. 08)
Çѱ۳»¿ë
(Korean Abstract)
º» ³í¹®Àº µö·¯´× ±â¹Ý ÀÌ»óŽÁö ¸ðµ¨°ú º¯¼öµé °£ÀÇ °ü°è¸¦ °í·ÁÇÏ´Â ¿¹Ãø¸ðµ¨À» ÀÌ¿ëÇÏ¿© ´Ùº¯·® ½Ã°è¿­ µ¥ÀÌÅÍÀÇ Ç°ÁúÀ» °³¼±ÇÏ´Â º¸Á¤±â¹ýÀ» Á¦¾ÈÇÑ´Ù. ´Ùº¯·® ½Ã°è¿­ µ¥ÀÌÅÍ´Â ½Ã°£Àû Ư¼º»Ó¸¸ ¾Æ´Ï¶ó º¯¼öµé °£ÀÇ °ü°è°¡ µ¿½Ã¿¡ °í·ÁµÇ¾î¾ß ÇÑ´Ù. ÀÌ·¯ÇÑ Æ¯¼ºÀ» °í·ÁÇϱâ À§ÇØ º» ³í¹®¿¡¼­´Â ¸ðµ¨ ÇнÀ °úÁ¤¿¡¼­ º¯¼öµé °£ÀÇ ÇнÀÀ» À§ÇØ »ó°ü°è¼ö Çà·ÄÀ» ÇнÀÇÏ´Â Attention ±â¹Ý LSTM ¿¹Ãø¸ðµ¨À» Á¦¾ÈÇÏ°í, ÀÌ ¸ðµ¨À» ÀÌ¿ëÇÏ¿© ´Ùº¯·® ½Ã°è¿­ µ¥ÀÌÅÍ º¸Á¤À» ¼öÇàÇÑ´Ù. ¸ÕÀú ±âÁ¸ ¿¬±¸ÀÎ LSTM ±â¹Ý VAE-GAN ÀÌ»óŽÁö ¸ðµ¨À» ÀÌ¿ëÇÏ¿© µ¥ÀÌÅÍ ³»¿¡ Á¸ÀçÇÏ´Â ÀÌ»ó°ªÀ» ŽÁöÇÏ°í, º» ³í¹®¿¡¼­ Á¦¾ÈÇÏ´Â ¿¹Ãø¸ðµ¨·Î ÀÌ»óÀ¸·Î ŽÁöµÈ ÇØ´ç À©µµ¿ì¸¦ ¿¹ÃøÇÑ´Ù. ±× ´ÙÀ½, ¿¹ÃøµÈ À©µµ¿ì¸¦ Á¤»ó À©µµ¿ì¸¦ Àß »ý¼ºÇϵµ·Ï ÇнÀµÈ ÀÌ»óŽÁö ¸ðµ¨ÀÇ »ý¼ºÀÚ¿¡ Àü´ÞÇÏ¿© Àç»ý¼ºÇÑ À©µµ¿ì·Î ÀÌ»ó À©µµ¿ì¸¦ ´ëüÇÔÀ¸·Î½á º¸Á¤À» ¼öÇàÇÑ´Ù.
¿µ¹®³»¿ë
(English Abstract)
This paper proposes a correction method to improve the quality of multivariate time series data by using a deep learning-based anomaly detection model and a prediction model that considers the relationship between variables. Multivariate time series data should be considered not only for its temporal characteristics but also for the relationships between variables. In order to consider these characteristics, this paper proposes an attention-based LSTM prediction model that trains a correlation matrix to consider the relationship between variables in the model training process, and uses this model to correct multivariate time series data. First, we detect anomalies in the data using an existing study, the LSTM-based VAE-GAN anomaly detection model, and predict the corresponding windows that are detected as anomalies with the prediction model proposed in this paper. Then, we forward the predicted windows to the generator of the anomaly detection model, which is trained to generate normal-like windows, and perform the correction by replacing the abnormal windows with the generated windows.
Å°¿öµå(Keyword) µö·¯´×   ´Ùº¯·® ½Ã°è¿­ µ¥ÀÌÅÍ   µ¥ÀÌÅÍ º¸Á¤   µ¥ÀÌÅÍ Ç°Áú   ÀÌ»óŽÁö   GAN   Deep Learning   Multivariate Time-Series Data   Data Correction   Data Quality   Anomaly Detection   GAN  
ÆÄÀÏ÷ºÎ PDF ´Ù¿î·Îµå