A comprehensive wind speed forecast correction strategy with an artificial intelligence algorithm
Xueliang Zhao, Qilong Sun, Wanru Tang, Shuang Yu, Boyu Wang
Abstract
Wind speed forecasting is critical to renewable energy generation, agriculture, and disaster prevention. Due to the uncertainty and intermittence of wind, conventional forecasting methods with numerical weather prediction (NWP) models fall short of achieving satisfactorily high accuracy. Post-processing of the predicted results is necessary for enhancing the prediction accuracy. The industry generally employs time-series prediction (TSP) methods for error correction, yet it is time-consuming since repeated modeling is needed if the location changes. Aiming at addressing this problem, this paper discusses the application of a deep learning algorithm in the post-processing period of wind speed prediction. NWP results are utilized as the forecasting basis, and deep learning algorithms are used for minimizing errors. An experimental study is conducted with industrial data. The functionality and performance of TSP-based algorithms including rolling mean, exponential smoothing, and autoregressive integrated moving average algorithms are compared with deep learning-based algorithms, including long-short term memory and convolutional neural network. From the numerical results, both TSP and deep-learning error-correction methods can effectively increase the accuracy of day-level NWP model prediction results, while deep-learning methods are data-driven, and no modeling process is needed. This work also poses an insight into the future development of wind speed prediction in meteorology.