| 引用本文: | 杨京渝,罗隆福,阳同光,彭丽,田飞扬.基于气象特征挖掘和改进深度学习模型的风电功率短期预测[J].电力自动化设备,2023,43(3): |
| YANG Jingyu,LUO Longfu,YANG Tongguang,PENG Li,TIAN Feiyang.Wind power short-term forecasting based on meteorological feature exploring and improved deep learning model[J].Electric Power Automation Equipment,2023,43(3): |
|
| 摘要: |
| 为提高风电功率短期预测的准确度，需进一步挖掘气象特征，为此，提出一种基于贝叶斯优化调参的特征挖掘改进深度学习模型。对气象因素提取多时间尺度下的统计特征、组合特征和类别特征；构建包含长短时记忆神经网络与注意力机制结合模块、Embedding模块和输出模块的深度学习模型，将连续数值特征输入长短时记忆神经网络与注意力机制结合模块，将类别特征输入Embedding模块；由贝叶斯优化调参进行特征组合选择，找出最优特征组合，得到最终的风电功率预测结果。与某风电场历史数据的对比分析表明，所提方法能有效提高风电功率的预测精度。 |
| 关键词: 风电功率预测 气象特征 深度学习 特征构造 贝叶斯优化 |
| DOI：10.16081/j.epae.202208019 |
| 分类号:TM614 |
| 基金项目:湖南省科技重大专项 (2020GK1013)；湖南省自然科学基金面上项目(2021JJ30079)；湖南省教育厅科研项目(20C0358)；益阳市社科课题(2022YS016) |
|
| Wind power short-term forecasting based on meteorological feature exploring and improved deep learning model |
|
YANG Jingyu1,2, LUO Longfu1, YANG Tongguang2, PENG Li2, TIAN Feiyang2
|
|
1.School of Electrical & Information Engineering, Hunan University, Changsha 410000, China;2.Key Laboratory Energy Monitoring and Edge Computing for Smart City of Hunan Province, Hunan City University, Yiyang 413000, China
|
| Abstract: |
| In order to improve the accuracy of wind power short-term forecasting, further mining of meteorological features is needed, for which, an improved deep learning model for feature mining is proposed based on Bayesian optimal parameter tuning. The statistical features, combinatorial features and category features are extracted for meteorological factors under multiple time scales. A deep learning model consisting of module combining long short-term memory neural network and Attention mechanism, Embedding module and output module is constructed, the continuous numerical values are input into the module combining long short-term memory neural network and Attention mechanism, and the categorical features are input into the Embedding module. The Bayesian optimal parameter tuning is used for feature combination selection to find the optimal feature combination, and the final wind power forecasting results is obtain. The comparative analysis with historical data of a wind farm shows that the proposed method can effectively improve the forecasting accuracy of wind power. |
| Key words: wind power forecasting meteorological feature deep learning feature exploring Bayesian optimization |
