| 引用本文: | 栗然,丁星,孙帆,韩怡,刘会兰,严敬汝.基于Wide & Deep-XGB2LSTM模型的超短期光伏功率预测[J].电力自动化设备,2021,41(7): |
| LI Ran,DING Xing,SUN Fan,HAN Yi,LIU Huilan,YAN Jingru.Ultra-short-term photovoltaic power prediction based on Wide & Deep-XGB2LSTM model[J].Electric Power Automation Equipment,2021,41(7): |
|
| 摘要: |
| 为了充分利用电网自身的海量历史数据进行光伏功率预测，提出一种宽度&深度(Wide & Deep)框架下融合极限梯度提升(XGBoost)算法和长短时记忆网络(LSTM)的Wide & Deep-XGB2LSTM超短期光伏功率预测模型。对历史数据进行特征提取，获得时间、辐照度、温度等原始特征，在此基础上进行特征重构，通过交叉组合和挖掘统计特征构造辐照度 × 辐照度、均值、标准差等组合特征，并通过Filter法和Embedded法进行特征选择。在TensorFlow框架下通过算例对比验证了所提模型及特征工程工作对光伏功率预测性能的提升效果。 |
| 关键词: 光伏功率预测 宽度&深度模型 极限梯度提升 长短时记忆网络 特征工程 模型融合 |
| DOI：10.16081/j.epae.202103020 |
| 分类号:TM615 |
| 基金项目:中央高校基本科研业务费专项资金资助项目(2017MS093) |
|
| Ultra-short-term photovoltaic power prediction based on Wide & Deep-XGB2LSTM model |
|
LI Ran1, DING Xing1, SUN Fan1, HAN Yi1, LIU Huilan1, YAN Jingru2
|
|
1.School of Electrical and Electronic Engineering, North China Electric Power University, Baoding 071003, China;2.Electric Power Research Institute of State Grid Hebei Electric Power Company, Shijiazhuang 050021, China
|
| Abstract: |
| In order to fully use massive historical data of power grid for photovoltaic power prediction, XGBoost(eXtreme Gradient Boosting) algorithm and LSTM(Long Short-Term Memory network) are fused under Wide & Deep framework, and a ultra-short-term photovoltaic power prediction model based on Wide & Deep-XGB2LSTM is proposed. The feature extraction is performed on historical data to obtain primitive features of time, irradiance, temperature and so on, on this basis, feature reconstruction is carried out and combination features such as irradiance*irradiance, mean value, standard deviation are constructed by cross combination and statistical feature mining, and Filter method and Embedded method are used for feature selection. Case comparison under TensorFlow framework verifies the promotion effect of the proposed model and feature engineering work on prediction performance of photovoltaic power. |
| Key words: photovoltaic power prediction Wide & Deep model XGBoost LSTM feature engineering model fusion |
