The Performance Analysis of Deep Belief Networks: A Case Study of PM2.5
碩士 === 國立暨南國際大學 === 資訊管理學系 === 107 === This study mainly discusses whether data preprocessing is helpful for training of Deep Belief Networks(DBN). Using air pollution data, we first lower the noise by preprocessing and fill the missing values. Then we use logarithm(LOG), verification, stepwise r...
Main Authors: | , |
---|---|
Other Authors: | |
Format: | Others |
Language: | zh-TW |
Published: |
2019
|
Online Access: | http://ndltd.ncl.edu.tw/handle/rax6j4 |
id |
ndltd-TW-107NCNU0396006 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-TW-107NCNU03960062019-09-07T03:30:33Z http://ndltd.ncl.edu.tw/handle/rax6j4 The Performance Analysis of Deep Belief Networks: A Case Study of PM2.5 深度信念網路的研究與觀察:以空氣污染資料PM2.5為例 LAI,MENG-NAN 賴孟南 碩士 國立暨南國際大學 資訊管理學系 107 This study mainly discusses whether data preprocessing is helpful for training of Deep Belief Networks(DBN). Using air pollution data, we first lower the noise by preprocessing and fill the missing values. Then we use logarithm(LOG), verification, stepwise regression, wavelet analysis to change the data structure, or eliminate some of the independent variables to lower the noises. We use one of metaheuristic algorithms, namely Genetic Algorithm(GA), to help searching parameters, decide network structure to minimize human involve, and to prevent local minimum solution. We discuss whether data preprocessing is helpful for forecasting in terms of DBN using 12-hour and 24-hour air pollution forecasting models. But massive missing values in a dataset is not trustable neither in multivariate regression nor time series forecasting in terms of performance. The proposed preprocessing methods combined with DBN improves the performances of forecasting. PAI,PING-FENG 白炳豐 2019 學位論文 ; thesis 39 zh-TW |
collection |
NDLTD |
language |
zh-TW |
format |
Others
|
sources |
NDLTD |
description |
碩士 === 國立暨南國際大學 === 資訊管理學系 === 107 === This study mainly discusses whether data preprocessing is helpful for training of Deep Belief Networks(DBN). Using air pollution data, we first lower the noise by preprocessing and fill the missing values. Then we use logarithm(LOG), verification, stepwise regression, wavelet analysis to change the data structure, or eliminate some of the independent variables to lower the noises. We use one of metaheuristic algorithms, namely Genetic Algorithm(GA), to help searching parameters, decide network structure to minimize human involve, and to prevent local minimum solution. We discuss whether data preprocessing is helpful for forecasting in terms of DBN using 12-hour and 24-hour air pollution forecasting models. But massive missing values in a dataset is not trustable neither in multivariate regression nor time series forecasting in terms of performance. The proposed preprocessing methods combined with DBN improves the performances of forecasting.
|
author2 |
PAI,PING-FENG |
author_facet |
PAI,PING-FENG LAI,MENG-NAN 賴孟南 |
author |
LAI,MENG-NAN 賴孟南 |
spellingShingle |
LAI,MENG-NAN 賴孟南 The Performance Analysis of Deep Belief Networks: A Case Study of PM2.5 |
author_sort |
LAI,MENG-NAN |
title |
The Performance Analysis of Deep Belief Networks: A Case Study of PM2.5 |
title_short |
The Performance Analysis of Deep Belief Networks: A Case Study of PM2.5 |
title_full |
The Performance Analysis of Deep Belief Networks: A Case Study of PM2.5 |
title_fullStr |
The Performance Analysis of Deep Belief Networks: A Case Study of PM2.5 |
title_full_unstemmed |
The Performance Analysis of Deep Belief Networks: A Case Study of PM2.5 |
title_sort |
performance analysis of deep belief networks: a case study of pm2.5 |
publishDate |
2019 |
url |
http://ndltd.ncl.edu.tw/handle/rax6j4 |
work_keys_str_mv |
AT laimengnan theperformanceanalysisofdeepbeliefnetworksacasestudyofpm25 AT làimèngnán theperformanceanalysisofdeepbeliefnetworksacasestudyofpm25 AT laimengnan shēndùxìnniànwǎnglùdeyánjiūyǔguāncháyǐkōngqìwūrǎnzīliàopm25wèilì AT làimèngnán shēndùxìnniànwǎnglùdeyánjiūyǔguāncháyǐkōngqìwūrǎnzīliàopm25wèilì AT laimengnan performanceanalysisofdeepbeliefnetworksacasestudyofpm25 AT làimèngnán performanceanalysisofdeepbeliefnetworksacasestudyofpm25 |
_version_ |
1719244068005871616 |