黄土高原集中式光伏电站功率预测模型适用性评价
马越 , 郭哲卓 , 李尚明 , 白静 , 许馨尹 , 杨红霞 , 杨丽平
延安大学学报(自然科学版) ›› 2026, Vol. 45 ›› Issue (01) : 3 -9.
黄土高原集中式光伏电站功率预测模型适用性评价
Evaluation of the applicability of power prediction models for centralized photovoltaic power stations in the Loess Plateau
针对黄土高原集中式光伏电站发电功率预测需求,本文基于陕西省延安市某集中式光伏电站2024年气象与光伏发电功率实测数据,系统对比了长短期记忆网络(LSTM)、门控循环单元(GRU)、卷积神经网络-长短期记忆网络混合模型(CNN-LSTM)、时间卷积网络(TCN)及时序融合Transformer(TFT)共5种深度学习模型的预测精度、稳定性、计算效率及适应性。结果表明,CNN-LSTM模型综合性能表现最优,其结合空间特征提取与时间依赖建模的混合架构更适用于该地区光伏功率预测;5种模型均表现出天气依赖性,晴天条件下LSTM模型表现最佳,阴天/多云天气下CNN-LSTM更具优势,雨天环境下GRU与TCN模型适应性更强。研究为黄土高原地区光伏功率预测的模型选择提供了实证参考。
To meet the power prediction requirements of centralized photovoltaic power stations in the Loess Plateau, this paper systematically compares the prediction accuracy, stability, computational efficiency, and adaptability of five deep learning models, namely Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) hybrid model, Temporal Convolutional Network (TCN), and Temporal Fusion Transformer (TFT), based on the measured meteorological and photovoltaic power generation data of a centralized photovoltaic power station in Yan’an City, Shaanxi Province in 2024. The results show that the CNN-LSTM model has the best overall performance, and its hybrid architecture combining spatial feature extraction and temporal dependency modeling is more suitable for photovoltaic power prediction in this region. All five models exhibit weather dependence, with the LSTM model performing best under clear weather conditions, the CNN-LSTM model having an advantage under cloudy and overcast conditions, and the GRU and TCN models showing stronger adaptability under rainy conditions. This study provides empirical references for the selection of power prediction models for photovoltaic power stations in the Loess Plateau region.
| [1] |
吕清泉,张珍珍,马彦宏, |
| [2] |
杨红霞,何春燕,郑海. 屋顶光伏与幕墙光伏在陕北地区建筑中的应用研究[J]. 延安大学学报(自然科学版),2025,44(1):14-19+28. |
| [3] |
|
| [4] |
张莉娅,王毅荣. 中国黄土高原近地面气候环境演变趋势[J]. 中国农学通报,2014,30(2):286-294. |
| [5] |
林文婷,李培强,荆志宇, |
| [6] |
|
| [7] |
|
| [8] |
|
| [9] |
|
| [10] |
赖昌伟,黎静华,陈博, |
| [11] |
王智,毛雪飞,冯英, |
| [12] |
蔡祯杨,李科,吴定会, |
| [13] |
殷钰朋,匡洪海,李星宇, |
| [14] |
夏鹏,黄磊,徐浩, |
| [15] |
沈炳华,曾智,陈清泉, |
| [16] |
米郝,高新宇,袁慧, |
| [17] |
郑涵. 基于改进的ELO模型和机器学习方法的NBA赛事预测[D]. 上海:上海师范大学,2024. |
| [18] |
徐崇斌,陈前,孙晓敏, |
| [19] |
丁钦钦. 分布式光伏系统中多源异构数据融合算法研究[D]. 南京:东南大学,2023. |
国家自然科学基金项目(52468029)
延能集团-陕西延安电业有限责任公司《绿色能源-零碳建筑实验室》建设项目(206021030)
/
| 〈 |
|
〉 |