| 引用本文: | 胡纪元,任奕铭,张晨浩,林红帆,张明轩,宋国兵.基于新型时空预测模型STICA的超短期海上风电出力预测[J].电力自动化设备,2025,45(12):58-65. |
| HU Jiyuan,REN Yiming,ZHANG Chenhao,LIN Hongfan,ZHANG Mingxuan,SONG Guobing.Ultra-short-term offshore wind power output prediction based on new spatiotemporal prediction model STICA[J].Electric Power Automation Equipment,2025,45(12):58-65. |
|
| 摘要: |
| 传统海上风电出力预测方法难以充分捕捉海上风电场的复杂时空相关性，导致预测精度受限。为此，设计一种具有优秀时空信息挖掘能力的时空交互因果注意力神经网络(STICA)，并用于超短期海上风电功率预测。因果注意力模块用于构建变量间的依赖关系，自适应学习风机功率与相关变量之间的相关性；动态图卷积网络用于动态生成风电场之间的空间联系；结合经相关性学习后的序列及动态图卷积网络生成的结果，层次化交互式时间序列分解模块用于建立多维时间数据的时序关系，实现超短期海上风电功率的有效预测。使用海上风电机组实测数据进行分析验证，结果表明所提方法具备优良的超短期海上风电功率预测能力，与常规预测模型相比，该方法在预测准确度和效率上均有提升。 |
| 关键词: 海上风电 超短期功率预测 时空相关性 因果注意力 动态图卷积网络 交互式 |
| DOI：10.16081/j.epae.202510025 |
| 分类号:TM614 |
| 基金项目:博士后创新人才支持计划(BX20230287) |
|
| Ultra-short-term offshore wind power output prediction based on new spatiotemporal prediction model STICA |
|
HU Jiyuan1, REN Yiming1, ZHANG Chenhao1, LIN Hongfan1, ZHANG Mingxuan1, SONG Guobing1,2
|
|
1.School of Electrical Engineering, Xi’an Jiaotong University, Xi’an 710049, China;2.State Key Laboratory of Electrical Insulation and Power Equipment, Xi’an Jiaotong University, Xi’an 710049, China
|
| Abstract: |
| The traditional offshore wind power prediction methods are difficult to fully capture the complex spatiotemporal correlation of offshore wind farm, which results in limited prediction accuracy. Therefore, a spatiotemporal interactive causal attention neural network(STICA) with excellent spatiotemporal information mining ability is designed for ultra-short-term offshore wind power prediction. The causal attention module is used to construct dependency relationship between variables and adaptively learn the correlation between wind turbine power and relevant variables. The dynamic graph convolutional network is used to dynamically generate the spatial connection between wind farms. Combining the sequence after correlation learning and the results generated by dynamic graph convolutional network, the hierarchical interactive time series decomposition module is used for establishing the temporal relationship of multi-dimensional time data, which realizes effective ultra-short-term offshore wind power prediction. The measured data from offshore wind turbines is used for analysis and verification, and the results show that the proposed method has superior ultra-short-term offshore wind power prediction ability, and it has improved the accuracy and efficiency compared with the conventional prediction models. |
| Key words: offshore wind power ultra-short-term power prediction spatiotemporal correlation causal attention dynamic graph convolutional network interaction |
