We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
Forecasting a short‐term wind speed using a deep belief network combined with a local predictor.
- Authors
Yu, Y.; Chen, Z. M.; Li, M. S.; Ji, T. Y.; Wu, Q. H.
- Abstract
This article proposes a novel wind speed forecast algorithm framework, which combines a deep belief network (DBN) and an improved Local Predictor (LP). The DBN adopted is a multiple hidden layers deep neural network consisting of restricted Boltzmann machines (RBMs) with superior performances on classification and prediction applications of a large‐scale dataset, such as wind speed data. However, it is known that the time series of wind speed under certain weather conditions have different patterns. Therefore, it is difficult to represent these patterns using a single model. Thus, a data‐filtering technique based on phase space reconstruction, LP, is used to embed one‐dimension high‐variance time series to a high dimensional space, and to select the training samples having the same pattern as the forecast sample. Therefore, a DBN with a Local Predictor (DBNLP) can be applied to forecast the wind speed using large‐scale training samples, which provides more reliable forecast results. In the experimental studies, the proposed algorithm is evaluated on the wind speed data collected from AIMS weather station in Australia. The forecast performance is compared with the commonly used predictors, including Auto‐Regressive Moving Average (ARMA), Support Vector Machines (SVMs), and Artificial Neural Networks (ANNs), respectively. The results demonstrate that the proposed algorithm exhibits a better performance than the other algorithms under various forecast steps. © 2018 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.
- Subjects
WIND power; ARTIFICIAL neural networks; WIND turbines; FUZZY logic; RENEWABLE energy sources
- Publication
IEEJ Transactions on Electrical & Electronic Engineering, 2019, Vol 14, Issue 2, p238
- ISSN
1931-4973
- Publication type
Article
- DOI
10.1002/tee.22802