Using Long Short-Term Memory Deep Learning in PM2.5 Multivariate Time Series Data Prediction research
| Autor: | HUANG, MAN-TING, 黃曼婷 |
|---|---|
| Rok vydání: | 2019 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 107 This research builds a Long Short-Term Memory (LSTM) model to forecast univariate and multivariate Time Series Data by using TensorFlow, which is an open source platform for machine learning. The subject of research is the PM2.5 air quality index (AQI) per hour per day at Taoyuan District of Taoyuan City, Taiwan (R.O.C), with 1704 records in total. The data period is ranged from 2017/04/01 to 2017/06/30. The input features are the measured PM2.5, temperature, humidity, wind speed, rainfall. Finally, the purpose of this research is to generate the best forecast results by parameter adjustment and analysis of the average and the short-term variability of root-mean-square error index. The forecast results generated by using the best parameters combinations data of both LSTM predict models are as follows: In Univariate Time Series Forecasting, the average of root-mean-square errors are 6.176 (Train Score) and 5.163 (Test Score), the short-term variability are 0.086839234 (Train Score) and 0.134755139 (Test Score). In Multivariate Time Series Forecasting, the average of root-mean-square errors are 6.1068 (Train Score) and 4.45025 (Test Score), the short-term variability are 0.01156492 (Train Score) and 0.016189259 (Test Score). |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|