For time series forecasting problems, this report proposes a novel forecasting mechanism that predicts values with multistep-ahead, and builds a real forecasting system to verify the applicability and effectiveness of the proposed mechanism. The forecasting techniques require on-line learning in order to avoid the concept shift and concept drift, and the key challenges to on-line learning approach include fast and memory efficient, adaptation, and robust to noise. Besides, the challenges to multistep-ahead forecasting result from increased uncertainty from longer forecasting steps. Based on the last year’s results, including the solar irradiance information platform, the proposed prediction mechanism, and so forth, this report presents a very short-term solar irradiance intelligent forecasting system with 5 to 30 minutes ahead and its performance comparison.
|Effective start/end date||2017/08/01 → 2018/10/31|
- Neural networks
- time series
Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.