Imputation and Prediction of IoT Sensor Data Using Multi-layered Fused Deep Autoencoder

碩士 === 國立臺北科技大學 === 電機工程系 === 107 === Nowadays, with the application of the Internet of Things, the number of sensors and nodes are rising, the transmission of data is more susceptible due to sensor failure, wireless network disconnection, low battery energy, etc., resulting in data collection by th...

Full description

Bibliographic Details
Main Authors: YEN, YU-HSIANG, 顏于翔
Other Authors: WANG, SHUN-YUAN
Format: Others
Language:zh-TW
Published: 2019
Online Access:http://ndltd.ncl.edu.tw/handle/tgv8qk
id ndltd-TW-107TIT00441066
record_format oai_dc
spelling ndltd-TW-107TIT004410662019-11-10T05:31:20Z http://ndltd.ncl.edu.tw/handle/tgv8qk Imputation and Prediction of IoT Sensor Data Using Multi-layered Fused Deep Autoencoder 應用多層融合式深度自動編碼器於物聯網感測器資料之修補與預測 YEN, YU-HSIANG 顏于翔 碩士 國立臺北科技大學 電機工程系 107 Nowadays, with the application of the Internet of Things, the number of sensors and nodes are rising, the transmission of data is more susceptible due to sensor failure, wireless network disconnection, low battery energy, etc., resulting in data collection by the database incomplete. This partially lost data has a great impact on the accuracy of the decision analysis of data. Consequently, the missing data should be corrected properly for future use. This study uses a model called multi-layered fused deep autoencoder, which is trained by the relationship between the sensors and the sensor data, to compute the best data results as the sensor data lost. This study uses a fused model, which is trained in consideration of the relationship between rainfall, temperature and humidity, compared with fused deep autoencoder trained only considering the relationship between temperature and humidityity, and deep auto-encoder trained only considering temperature. This study also proposes an iterative testing method designed for deep autoencoders. Considering dataset with high missing rate may be hard to learn the relationship between data, iterative testing method can make missing data gradually learn the model, achieving better results than the traditional one-time testing method did. After the data was imputed, this study will put missing data, data imputed through the fused deep autoencoder and data imputed through multi-layered fused deep autoencoder, into another multi-layered fused deep autoencoder to predict future data. Comparing the results of these three kinds of data predicting the future, it proves that data in this paper is effective for data decision after imputed. Experiment results show that data imputed by the multi-layered fused deep auto-encoder provides better imputation error than other imputation methods, whether in the traditional one-time or iterative testing method, and the iterative method is better than the traditional one. A stable and good error can be provided in the data with 60% of missing rate or less, and it can also be directly proportional to the error of imputation in predicting the future, thus provides an excellent decision. WANG, SHUN-YUAN 王順源 2019 學位論文 ; thesis 72 zh-TW
collection NDLTD
language zh-TW
format Others
sources NDLTD
description 碩士 === 國立臺北科技大學 === 電機工程系 === 107 === Nowadays, with the application of the Internet of Things, the number of sensors and nodes are rising, the transmission of data is more susceptible due to sensor failure, wireless network disconnection, low battery energy, etc., resulting in data collection by the database incomplete. This partially lost data has a great impact on the accuracy of the decision analysis of data. Consequently, the missing data should be corrected properly for future use. This study uses a model called multi-layered fused deep autoencoder, which is trained by the relationship between the sensors and the sensor data, to compute the best data results as the sensor data lost. This study uses a fused model, which is trained in consideration of the relationship between rainfall, temperature and humidity, compared with fused deep autoencoder trained only considering the relationship between temperature and humidityity, and deep auto-encoder trained only considering temperature. This study also proposes an iterative testing method designed for deep autoencoders. Considering dataset with high missing rate may be hard to learn the relationship between data, iterative testing method can make missing data gradually learn the model, achieving better results than the traditional one-time testing method did. After the data was imputed, this study will put missing data, data imputed through the fused deep autoencoder and data imputed through multi-layered fused deep autoencoder, into another multi-layered fused deep autoencoder to predict future data. Comparing the results of these three kinds of data predicting the future, it proves that data in this paper is effective for data decision after imputed. Experiment results show that data imputed by the multi-layered fused deep auto-encoder provides better imputation error than other imputation methods, whether in the traditional one-time or iterative testing method, and the iterative method is better than the traditional one. A stable and good error can be provided in the data with 60% of missing rate or less, and it can also be directly proportional to the error of imputation in predicting the future, thus provides an excellent decision.
author2 WANG, SHUN-YUAN
author_facet WANG, SHUN-YUAN
YEN, YU-HSIANG
顏于翔
author YEN, YU-HSIANG
顏于翔
spellingShingle YEN, YU-HSIANG
顏于翔
Imputation and Prediction of IoT Sensor Data Using Multi-layered Fused Deep Autoencoder
author_sort YEN, YU-HSIANG
title Imputation and Prediction of IoT Sensor Data Using Multi-layered Fused Deep Autoencoder
title_short Imputation and Prediction of IoT Sensor Data Using Multi-layered Fused Deep Autoencoder
title_full Imputation and Prediction of IoT Sensor Data Using Multi-layered Fused Deep Autoencoder
title_fullStr Imputation and Prediction of IoT Sensor Data Using Multi-layered Fused Deep Autoencoder
title_full_unstemmed Imputation and Prediction of IoT Sensor Data Using Multi-layered Fused Deep Autoencoder
title_sort imputation and prediction of iot sensor data using multi-layered fused deep autoencoder
publishDate 2019
url http://ndltd.ncl.edu.tw/handle/tgv8qk
work_keys_str_mv AT yenyuhsiang imputationandpredictionofiotsensordatausingmultilayeredfuseddeepautoencoder
AT yányúxiáng imputationandpredictionofiotsensordatausingmultilayeredfuseddeepautoencoder
AT yenyuhsiang yīngyòngduōcéngrónghéshìshēndùzìdòngbiānmǎqìyúwùliánwǎnggǎncèqìzīliàozhīxiūbǔyǔyùcè
AT yányúxiáng yīngyòngduōcéngrónghéshìshēndùzìdòngbiānmǎqìyúwùliánwǎnggǎncèqìzīliàozhīxiūbǔyǔyùcè
_version_ 1719289404164407296