Note! Please note that this article is currently in the "Article in Press" stage and is not the final "Version of record". While it has been accepted, copy-edited, and formatted, however, it is still undergoing proofreading and corrections by the authors. Therefore, the text may still change before the final publication. Although "Articles in Press" may not have all bibliographic details available, the DOI and the year of online publication can still be used to cite them. The article title, DOI, publication year, and author(s) should all be included in the citation format. Once the final "Version of record" becomes available the "Article in Press" will be replaced by that.
Abstract
Background: Accurate prediction of offshore wind power is the basis for promoting the safe and economic operation of offshore wind farms.
Objective: This paper proposes an uncertainty prediction learning model based on outlier processing, synchronous wavelet denoising and attention mechanism optimization to achieve accurate prediction of offshore wind power.
Method: Firstly, the isolated forest is adopted to filter the outliers of offshore wind power data and delete the error data caused by equipment or humans. Secondly, a syn-chrosqueezing wavelet neural network (SWT) is applied to denoise historical wind power data, improve data quality, and lay a foundation for accurate prediction. Next, the offshore wind power prediction method based on IP-SO-LSTM-Attention is constructed to realize offshore wind power prediction, in which the attention mechanism is applied to focus on the influence of important features on the output of offshore wind power, and the improved particle swarm optimization algorithm is adopted to find the best network structure of LSTM-Attention to optimize the prediction effect. After predicting the point prediction results based on the SWT-IPSO-LSTM-Attention model, this paper sets MAPE, RMSE, MAE and other indicators to evaluate the prediction effect.
Results: The prediction error MAPE of the proposed model is 4.12%. It is 63.21% higher than the benchmark model (SWT-BP) and 56.35% higher than the benchmark model (SWT-LSTM).
Conclusion: Based on the point prediction results, this paper uses KDE (Gaussian) prediction results to calculate the out-put curve under different confidence levels and provides decisionmaking reference information for accurate decision-making.
Related Journals
Recent Patents on Engineering
Related Books
Voltammetry for Sensing Applications
