| 引用本文: | 李 嵩,夏成军,刘译夫,汪愉康,刘育成.基于KAN和误差时间序列自相关性特征的多时间尺度光伏功率预测[J].电力系统保护与控制,2025,53(24):176-187.[点击复制] |
| LI Song,XIA Chengjun,LIU Yifu,WANG Yukang,LIU Yucheng.Multi-time scale PV power forecasting based on KAN and error time series autocorrelation features[J].Power System Protection and Control,2025,53(24):176-187[点击复制] |
|
| 本文已被：浏览 218次 下载 10次 |
码上扫一扫！ |
| 基于KAN和误差时间序列自相关性特征的多时间尺度光伏功率预测 |
| 李嵩1,2,夏成军1,2,刘译夫1,2,汪愉康1,2,刘育成1,2 |
|
|
| (1华南理工大学电力学院，广东 广州 510640;2.广东省新能源电力系统智能运行与
控制企业重点实验室，广东 广州 510663) |
|
| 摘要: |
| 针对日前预测模型在非线性特征提取能力上的不足，以及日内预测时功率输出波动特征复杂导致模型学习困难的问题，提出了基于科尔莫戈洛夫-阿诺尔德网络(Kolmogorov-Arnold networks, KAN)和误差时间序列自相关性特征的多时间尺度光伏功率预测方法。首先，在日前预测阶段，设计了以KAN为基本单元的预测模型，通过残差连接改进的深层KAN提取数据空间特征，并结合多头注意力机制提取数据时序特征，显著提升了模型对各种气候特征的提取能力。然后，在日内预测阶段，基于日前预测结果，结合误差时间序列的自相关性特征进行间接预测，显著降低了预测序列的波动幅度，降低了模型的学习难度。最后，基于某光伏功率预测大赛提供的数据进行实验，结果表明，所设计的日前预测模型相较于长短时记忆(long short-term memory, LSTM)网络、Transformer等模型，预测结果的均方误差(mean square error, MSE)至少降低3.8%。所提出的日内间接预测方法相比于直接预测方法，预测结果的MSE降低38.3%。 |
| 关键词: 光伏功率预测 多时间尺度 KAN 误差时间序列的自相关性特征 |
| DOI：10.19783/j.cnki.pspc.250170 |
| 投稿时间：2025-02-24修订日期：2025-07-01 |
| 基金项目:国家重点研发计划项目资助(2022YFB2403500) |
|
| Multi-time scale PV power forecasting based on KAN and error time series autocorrelation features |
| LI Song1,2,XIA Chengjun1,2,LIU Yifu1,2,WANG Yukang1,2,LIU Yucheng1,2 |
| (1. School of Electric Power, South China University of Technology, Guangzhou 510640, China; 2. Key Laboratory for Guangdong
Provincial New Energy Power System Intelligent Operation and Control Enterprise, Guangzhou 510663, China) |
| Abstract: |
| To address the limited nonlinear feature extraction capability of day-ahead forecasting models and the difficulty of model learning due to complex power output fluctuations in intraday forecasting, a multi-time scale photovoltaic (PV) power forecasting method based on Kolmogorov-Arnold networks (KAN) and the autocorrelation features of error time series is proposed. First, in the day-ahead forecasting stage, a forecasting model using KAN as the basic building block is designed. A deep KAN architecture enhanced with residual connections is used to extract spatial features, while a multi-head attention mechanism is employed to extract temporal features, significantly improving the model’s ability to capture diverse climatic characteristics. Then, in the intraday forecasting stage, based on the day-ahead forecasting results, indirect prediction is performed by incorporating the autocorrelation features of the error time series. This approach significantly reduces the fluctuation range of the predicted sequence and lowers the learning difficulty of the model. Finally, experiments conducted using data provided by a PV power forecasting competition demonstrate that the proposed day-ahead model reduces the mean square error (MSE) by at least 3.8% compared with long-short-term memory (LSTM) and Transformer models. Compared with direct forecasting, the MSE of the forecasting results is reduced by 38.3%. |
| Key words: PV power forecasting multi-time scale KAN autocorrelation features of error time series |
