基于注意力机制的CNN-LSTM-XGBoost台风暴雨电力气象混合预测模型

Attention Mechanism Based CNN-LSTM-XGBoost Electric Power Meteorological Hybrid Forecasting Model of Typhoon Rainstorm

侯慧 ¹吴文杰 ¹魏瑞增 ²何浣 ²王磊 ²李正天 ³林湘宁³

扫码查看

作者信息

1. 武汉理工大学自动化学院,湖北武汉 430070
2. 广东省电力装备可靠性重点实验室,广东广州 510080
3. 华中科技大学电气与电子工程学院,湖北武汉 430074
折叠

摘要

极端台风暴雨灾害具有非线性、极差大以及多峰值等特点.为使电网及时获取预警信息,提出一种基于注意力机制的CNN-LSTM-XGBoost台风暴雨电力气象混合预测模型.首先,利用基于注意力机制的卷积神经网络(CNN)辨识关键台风暴雨灾害特征;然后,利用长短期记忆网络(LSTM)训练时间序列预测模型以挖掘台风暴雨时序特征,使用极限梯度提升算法替换模型输出层以缓解过拟合问题;最后,以2023年台风泰利为例验证所提方法的有效性.算例分析表明,所提模型具有较高的准确性,对预测精度的提升可达40.84%以上.

Abstract

Major disasters such as typhoon rainstorm disasters have the characteristics of nonlinear,large range and multi-peak.To make the power grid obtain early warning information in time,the paper proposes a attention mechanism based CNN-LSTM-XGBoost electric power meteorological hybrid forecasting model of the typhoon rainstorm.Firstly,the attention mechanism based convolutional neural network is used to extract the key disaster characteristics of the typhoon rainstorm.Then the long short-term memory network(LSTM)is utilized to train time series prediction model and mine the time series feature information.To solve the problem of overfitting,the extreme gradient boosting algorithm is applied to replace model's output layer.Finally,typhoon Talim in 2023 is taken as a case study to verify the effectiveness of the proposed method.The results show that the proposed model has better performance,and its prediction accuracy is improved by more than 40.84%.

关键词

台风灾害/暴雨预测/神经网络/混合模型/电网预警

Key words

typhoon disaster/rainstorm forecast/neural network/hybrid model/power grid early warning

引用本文复制引用

基金项目

国家自然科学基金资助项目(U22B20106)

中国南方电网有限责任公司科技项目(GDKJXM20231426)

出版年

2024

智慧电力

陕西省电力公司

智慧电力

CSTPCDCSCD北大核心

影响因子：0.831

ISSN：1673-7598

参考文献量18

段落导航