China Mechanical Engineering ›› 2021, Vol. 32 ›› Issue (17): 2082-2089.

### Short Term Wind Speed Prediction of Wind Turbine Hubs Based on Combined Neural Network

MA Junyan;YUAN Yiping;CHAI Tong;ZHAO Qin

1. School of Mechanical Engineering，Xinjiang University，Urumqi，830047
• Online:2021-09-10 Published:2021-09-28

### 基于组合神经网络的风机轮毂处短期风速预测

1. 新疆大学机械工程学院，乌鲁木齐，830047
• 通讯作者: 袁逸萍（通信作者），女，1973年生，教授、博士研究生导师。研究方向为计算机集成制造系统、工业工程、质量与可靠性等。发表论文50余篇。E-mail：yipingyuan@163.com。
• 作者简介:马军岩，女，1990年生，博士研究生。研究方向为风速预测、风力发电机的故障预测与健康管理。E-mail：majunyanbyr@163.com。
• 基金资助:
国家自然科学基金（71961029）；
新疆维吾尔自治区研究生科研创新项目（XJ2020G015）

Abstract: The wind speed less than 6 hours at the wind turbine hubs was predicted by comprehensively considering the historical wind speed rules and wind speed changes in recent numerical weather forecasts. In order to improve the accuracy of short-term wind speed prediction under the condition of sudden changes of wind speeds， a new combined neural network was proposed. According to the characteristics of different frequency subsequences obtained by signal decomposition， the deep convolution neural network and the gated cyclic recursive unit were used to predict the low-frequency subsequence， and Elman recurrent neural network was established to predict the high-frequency sequence. The downscaling numerical weather prediction wind speed was used to identify inflection points， and the generalized autoregressive conditional heteroscedasticity model was used to modify the non inflection point of wind speed. Finally， the actual wind speed data of a wind farm in Xinjiang were used for the experiments. The mean absolute errors， mean square errors and mean absolute percentage errors were used to calculate the deterministic prediction accuracy， and the coverage rate and interval width were used to calculate the uncertainty prediction accuracy， which verified the effectiveness of the proposed algorithm.

CLC Number: