基于时间戳特征提取和CatBoost-LSTM模型的光伏短期发电功率预测
SHORT-TERM PHOTOVOLTAIC POWER PREDICTION BASED ON TIMESTAMP FEATURE EXTRATION AND CatBoost-LSTM MODEL
徐恒山 1莫汝乔 1薛飞 2秦子健 3潘鹏程1
作者信息
- 1. 三峡大学电气与新能源学院,宜昌 443002
- 2. 国网宁夏电力有限公司电力科学研究院,银川 750001
- 3. 国网山东省电力有限公司莱芜供电公司,莱芜 271100
- 折叠
摘要
为解决预测模型输入特征维度不足以及单一模型预测精度不高而导致的短期功率预测效果较差的问题,提出一种对时间戳进行特征提取(FE)的CatBoost和长短期记忆(LSTM)神经网络组合的光伏短期发电功率预测模型.首先,利用信息熵加权的方式对传统灰色关联分析进行改进,并采用改进方法对辐照度、温度、降雨量等气象特征与发电功率特征进行关联性分析,选择关键特征作为输入特征;然后,从时间戳和功率特征中提取年、月、日、时、分、秒、时间戳-功率等新时序特征;在此基础上,将关键气象特征与提取的新时序特征用于组合模型训练;最后,利用光伏电站的真实运行数据对所提方法和组合模型进行算例分析.结果表明:提取的新时序特征和组合模型均有助于提高预测精度,在非晴天工况下组合模型的预测误差较单一模型可降低12～23个百分点,且与其他组合模型相比具有更高的预测精度.
Abstract
To solve the problem that the short-term power prediction effect is poor due to the insufficient input feature dimensions of the prediction model and the low prediction accuracy of a single model,a short-term prediction model of photovoltaic power generation combining with timestamp feature extraction and based on CatBoost and long short-term memory(LSTM)neural network is proposed.Firstly,the traditional grey correlation analysis is improved by using the information entropy weighting,and the improved method is used to analyze the correlation between the meteorological features such as irradiance,temperature,rainfall and power,and select the key features as the input features;Then,new temporal features such as year,month,day,hour,minute,second,timestamp-power are extracted from timestamp and power;On this basis,key meteorological features and extracted new temporal features are used for combined model training;Finally,an example analysis was conducted on the proposed method and combination model using real operating data of photovoltaic power plants.The results show that both the extracted new time series feature and the combined model can help improve the prediction accuracy.Under the non-sunny conditions,the prediction error of the combined model can be reduced by about 12%-23%compared with the single model,and it has higher prediction accuracy compared with other combined models.
关键词
光伏发电/特征提取/预测/长短期记忆神经网络/时间戳/灰色关联分析Key words
PV power/feature extraction/prediction/long short-term memory/timestamp/grey correlation analysis引用本文复制引用
出版年
2024
NETL
NSTL
万方数据