基于功率特征的K-ISSA-LSTM光伏功率预测
K-ISSA-LSTM PHOTOVOLTAIC POWER PREDICTION BASED ON POWER CHARACTERISTIC
金伟勇 1卢丽娜 2赖欢欢 3张森林1
作者信息
- 1. 浙江大学电气工程学院,杭州 310058
- 2. 杭州电子科技大学信息工程学院,杭州 310000
- 3. 国网温州供电公司,温州 325028
- 折叠
摘要
历史功率特征能反映一段时间内光伏功率的波动情况,结合聚类算法对原始数据进行聚类,利用长短期记忆神经网络实现对光伏发电功率的预测.同时使用改进的麻雀搜索算法进行神经网络超参数寻优,实现对不同功率特征场景的超参数优化.采用华东地区某光伏电站的实测数据进行验证,预测模型功率波动情况下较传统预测方法对该组数据有更高的预测精度.
Abstract
Improving the accuracy of photovoltaic power prediction is of great value to the stable operation of the power system.The historical power characteristics can reflect the fluctuation of photovoltaic power over a period of time,using clustering algorithms to cluster the raw data,and the long-term short-term memory neural network is used to predict the photovoltaic power generation.At the same time,the improved sparrow search algorithm is used to optimize the hyperparameters of neural networks to realize the hyperparameter optimization of different power feature scenarios.Using the measured data of a photovoltaic power station in East China for verification,the prediction model has higher prediction accuracy than the traditional prediction method in the case of power fluctuation.
关键词
光伏功率/预测/聚类算法/长短期记忆/麻雀搜索算法Key words
photovoltaic power/forecasting/clustering algorithms/long short-term memory/sparrow search algorithm引用本文复制引用
基金项目
国网浙江省电力有限公司科技项目(5211WZ220002)
出版年
2024
NETL
NSTL
维普
万方数据