| 引用本文： |
-
马愈昭,史亚鑫,王大蕴,等.一种基于DSConvGRU-Transformer-ConvLSTM模型的风速预测方法[J].电讯技术，2026，66（10）： - . [点击复制]
- MA Yuzhaoa,SHI Yaxin a,WANG Dayun b,et al.A Wind Speed Prediction Method Based on DSConvGRU-Transformer-ConvLSTM Model[J].，2026，66（10）： - . [点击复制]
|
|
| 摘要: |
| 为提升机场测风雷达对剧烈风速变化的预测能力，构建了一种融合深度可分离卷积门控循环单元（Depthwise Separable Convolutional Gated Recurrent Unit，DSConvGRU）、Transformer空间注意力机制与卷积长短期记忆（Convolutional Long Short-term Memory，ConvLSTM）网络的多步时空风速预测模型。该结构结合DSConvGRU对局部特征的高效提取能力、Transformer对风场全局空间依赖的自适应建模能力，以及ConvLSTM对时序动态演化的连续表达能力，实现了时空特征的高质量协同融合建模。基于乌鲁木齐地窝堡国际机场激光测风雷达构建高分辨率风场图像数据集，开展了多组实验验证。结果显示，在剧烈风场预测任务中，该模型的平均绝对误差和均方根误差最多分别降低18.1%和13.2%；在单点风速预测任务中，误差亦显著低于Informer、双向长短期记忆网络和卷积神经网络-长短期记忆网络等主流模型。 |
| 关键词: 激光雷达 风速预测 深度可分离卷积门控循环单元 自注意力机制 卷积长短期记忆网络 |
| DOI：10.20079/j.issn.1001-893x.250519006 |
|
| 基金项目:民航局安全能力建设项目(KJZ44420240061) |
|
| A Wind Speed Prediction Method Based on DSConvGRU-Transformer-ConvLSTM Model |
| MA Yuzhaoa,SHI Yaxin a,WANG Dayun b,WANG Nan |
| (1a.College of Electronic Information and Automation;1b.Institute of Civil Aircraft Safety and Airworthiness，Civil Aviation University of China，Tianjin 300300，China;2.Meteorological Center of Xinjiang Air Traffic Management Bureau，Civil Aviation Administration of China，Urumqi 830016，China) |
| Abstract: |
| To enhance the capability of light detection and ranging(Lidar) to predict abrupt wind speed changes at airports，a multi-step spatiotemporal wind speed prediction model is developed by integrating Depthwise Separable Convolutional Gated Recurrent Unit(DSConvGRU)，Transformer-based spatial attention mechanism，and Convolutional Long Short-term Memory(ConvLSTM) network.The proposed framework leverages the local feature extraction efficiency of DSConvGRU，the global spatial dependency modeling capability of the Transformer module，and the temporal continuity modeling strength of ConvLSTM to achieve high-quality spatiotemporal feature fusion.A high-resolution wind field image dataset is constructed using Lidar data collected at Urumqi Diwopu International Airport for experimental validation.Results demonstrate that under severe wind scenarios，the proposed model reduces mean absolute error and root mean square error by up to 18.1% and 13.2%，respectively.In the single-point wind speed prediction task，the model also outperforms mainstream models such as Informer，Bidirectional Long Short-term Memory(BiLSTM)，and Convolutional Neural Network-Long Short-term Memory(CNN-LSTM). |
| Key words: Lidar wind speed prediction DSConvGRU Transformer attention mechanism ConvLSTM |
《电讯技术》编辑部 