Improving time series forecasting performance with deep learning techniques
Time series are ubiquitous in nature and human society. Especially, the forecasting of time series could be instructive in meteorology, commerce, energy management, financial activities, and various fields. Hence, time series forecasting is the most important task under the topic of time series anal...
Saved in:
| 主要作者: | |
|---|---|
| 其他作者: | |
| 格式: | Thesis-Master by Coursework |
| 語言: | English |
| 出版: |
Nanyang Technological University
2021
|
| 主題: | |
| 在線閱讀: | https://hdl.handle.net/10356/150279 |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 總結: | Time series are ubiquitous in nature and human society. Especially, the forecasting of time series could be instructive in meteorology, commerce, energy management, financial activities, and various fields. Hence, time series forecasting is the most important task under the topic of time series analysis. Many models have been proved to be effective for time series forecasting such as ARIMA, LSTM, TCN, RNN, and RVFL. This dissertation is intended to improve the accuracy of time series forecasting by using TCN-based algorithms. In this dissertation, three novel hybrid algorithms: TCN-edRVFL algorithm, multi-receptive field TCN (MRF-TCN algorithm), and dynamic decision multi-receptive field TCN (DDM-TCN algorithm) are proposed. At last, experiments of electric load forecasting are implemented on Australian Energy Market Operator (AEMO) datasets. Comparisons and results analysis indicate that the proposed algorithms outperform the TCN baseline in different aspects. |
|---|