受到复杂多变的气象因素影响，光伏发电输出功率具有随机性、波动性和间歇性等特点，导致光伏并网后电力系统的稳定运行出现安全风险，而短期光伏发电功率的精确预测是解决此问题的关键。

由于气象因素与光伏发电功率不确定性存在明显关联，本文通过选择最优相似日来削弱气象数据不平稳对预测精度的影响。首先，针对光伏功率历史数据和气象历史数据存在缺失值、异常值及量纲差距大的问题，采用3原则对样本数据实行预处理操作。其次，采用皮尔逊相关系数筛选出与光伏发电功率有较强相关性的气象因素。最后，采用K-memoids聚类方法利用筛选出的气象因素和光伏发电功率数据将预测日天气划分为晴天和非晴空（多云、雨天）场景。

针对非晴空条件下短期光伏功率点预测精度不高的问题，提出了基于双阶段注意力机制的长短期记忆神经网络光伏发电短期功率预测方法。该方法基于非晴空场景下的相似日数据样本，第一阶段根据多维气象因素的不同影响，通过注意力机制将输入量特征自适应赋权，强化了对非晴空条件下复杂多维气象因素的特征提取能力；第二阶段通过注意力机制给不同隐藏层单元自适应赋权，动态调整不同隐藏层对模型输出的影响。改进后的长短期记忆神经网络预测模型具有良好的鲁棒性，提高了非晴空条件下短期光伏功率点预测精度。

针对短期光伏功率点预测无法评估预测结果的不确定问题，在上述基于双阶段注意力机制的长短期记忆神经网络点预测基础上，采用分位数回归和核密度估计法对短期光伏功率进行概率预测，并采用分位数损失函数作为概率预测模型的目标函数，优化与调整模型训练过程。

通过宁夏同心光伏电站2020-2022年实际运行数据对本文方法进行验证，在非晴空条件下进行短期光伏功率点预测和概率预测，验证本文预测方法的先进性与适用性。

Affected by complex and variable meteorological factors, the output power of photovoltaic power generation is characterised by randomness, fluctuation and intermittency, leading to security risks for the stable operation of the power system after grid connection. The accurate prediction of short-term PV power generation is the key to solving this problem.

As meteorological factors are significantly associated with PV power uncertainty, this paper weakens the impact of unstable meteorological data on prediction accuracy by selecting the optimal similarity day. Firstly, to address the problems of missing values, outliers and large discrepancies between the PV power history and the meteorological history, the 3δ principle was used to pre-process the sample data. Secondly, Pearson correlation coefficients were used to screen out meteorological factors that have a strong correlation with PV power generation. Finally, the K-memoids clustering method was used to classify the forecast day weather into sunny and non-sunny sky (cloudy, rainy) scenarios using the filtered meteorological factors and PV power data.

To address the problem of low accuracy of short-term PV power point prediction under non-clear sky conditions, a two-stage attention mechanism based on a long and short-term memory neural network for short-term power prediction of PV power generation is proposed. The method is based on similar daily data samples under non-clear sky scenarios, and in the first stage, the input quantity features are adaptively weighted according to the different effects of multidimensional meteorological factors through an attention mechanism, which enhances the feature extraction capability of complex multidimensional meteorological factors under non-clear sky conditions; The second stage adaptively assigns weights to different hidden layer units through an attention mechanism to dynamically adjust the influence of different hidden layers on the model output. The improved long and short-term memory neural network prediction model has good robustness and improves the short-term PV power point prediction accuracy under non-clear sky conditions.

To address the uncertainty of short-term PV power point prediction which cannot assess the prediction results, probabilistic prediction of short-term PV power using quantile regression and kernel density estimation based on the above two-stage attention mechanism based on long- and short-term memory neural network point prediction. The quantile loss function is also used as the objective function of the probabilistic prediction model to optimize and adjust the model training process.

The method of this paper is validated by the actual operation data of Ningxia Concentric PV power plant from 2020 to 2022. Short-term PV power point prediction and probabilistic prediction are carried out under non-clear sky conditions to verify the advanced and applicability of the prediction method of this paper.

[1]张洁, 郝倩男. 基于烟花算法优化BP神经网络的光伏功率预测[J]. 计算机技术与发展, 2021, 31(10): 146-153.

[2]Blaga R, Sabadus A, Stefu N, et al. A current perspective on the accuracy of incoming solar energy forecasting[J]. Progress in energy and combustion science, 2019, 70: 119-144.

[3]于佳弘, 包哲静, 李志杰. 基于LSTM的用户负荷区间预测方法[J]. 工业控制计算机,2018, 31(04): 100-102.

[4]中国能源转型“十四五”电力规划研究报告[EB/OL]. 全球能源互联网发展合作组, 2020-07.

[5]Zervos A, Lins C, Muth J. RE-thinking 2050: a 100% renewable energy vision for the European Union[M]. Erec, 2010.

[6]PVPS I E A. Snapshot of Global PV Markets 2021[J]. International Energy Agency (IEA): Paris, France, 2020, 14: 169-181.

[7]Osório G J, Lujano-Rojas J M, Matias J C O, et al. A fast method for the unit scheduling problem with significant renewable power generation[J]. Energy Conversion and Management, 2015, 94: 178-189.

[8]马原, 张雪敏, 甄钊, 等. 基于修正晴空模型的超短期光伏功率预测方法[J]. 电力系统自动化, 2021, 45(11): 44-51.

[9]陈丽霞. 基于交叉熵理论的光伏发电功率组合预测方法[J]. 电工电气, 2022(04): 17-20+48.

[10]杨延勇, 孟祥剑, 高峰, 等. 一种基于双层人工神经网络的多时间尺度区域光伏功率预测方法[J]. 华北电力大学学报(自然科学版), 2021, 48(02): 55-63.

[11]赵亮, 黎嘉明, 艾小猛, 等. 光伏出力随机性分量的提取和统计特性分析[J]. 电力系统自动化, 2017, 41(01): 48-56.

[12]左远龙, 黄玉水, 杨晓辉, 等. 基于PFA-MBAS-BP神经网络模型的光伏发电短期预测[J]. 电力系统保护与控制, 2020, 48(15): 84-91.

[13]Lim N C, Ranaweera I U, Norum L, et al. A real-time energy management system for smart grid integrated photovoltaic generation with battery storage[J]. Renewable Energy, 2019, 130: 774-785.

[14]陈文进, 朱峰, 张童彦, 等. 基于AFSA-BP神经网络的光伏功率预测方法[J]. 浙江电力, 2022, 41(04): 7-13.

[15]Kudo M, Takeuchi A, Nozaki Y, et al. Forecasting electric power generation in a photovoltaic power system for an energy network[J]. Electrical Engineering in Japan, 2009, 167(4): 16-23.

[16]Das U K, Tey K S, Seyedmahmoudian M, et al. Forecasting of photovoltaic power generation and model optimization: A review[J]. Renewable and Sustainable Energy Reviews, 2018, 81: 912- 928.

[17]杨延勇, 孟祥剑, 高峰, 等. 种基于双层人工神经网络的多时间尺度区域光伏功率预测方法[J]. 北电力大学学报(自然科学版), 2021, 48(02): 55-63.

[18]Mellit A, Massi Pavan A, Ogliari E, et al. Advanced methods for photovoltaic output power forecasting: A review[J]. Applied Sciences, 2020, 10(2): 487.

[19]王书芹, 华钢, 郝国生, 等. 基于灰狼优化算法的长短期记忆网络在时间序列预测中的应用[J]. 中国科技论文, 2017, 12(20): 2309-2314.

[20]万灿, 宋永华. 新能源电力系统概率预测理论与方法及其应用[J]. 电力系统自动化, 2021, 45(01): 2-16.

[21]黎静华, 骆怡辰, 杨舒惠, 等. 可再生能源电力不确定性预测方法综述[J]. 高电压技2021, 47(04): 1144-1157.

[22]Utpal K D, Kok S T, Mehdi S, et al. Forecasting of photovoltaic power generation and model optimization: A review[J]. Renewable and Sustainable Energy Reviews, 2018, 81.

[23]Akhter M N, Mekhilef S, Mokhlis H, et al. Review on forecasting of photovoltaic power

generation based on machine learning and metaheuristic techniques[J]. IET Renewable Power Generation, 2019, 13(7): 1009-1023.

[24]Böök H, Lindfors A V. Site-specific adjustment of a NWP-based photovoltaic production forecast[J]. Solar Energy, 2020, 211: 779-788.

[25]Larson D P, Coimbra C F. M. Direct Power Output Forecasts From Remote Sensing Image Processing[J]. Journal of Solar Energy Engineering, 2018, 140(2): 021011.

[26]Kong W, Jia Y, Dong Z , et al. Hybrid approaches based ondeep wholesky-image learning to photovoltaic generation forecasting[J]. Applied Energy, 2020, 280: 115875.

[27]Voyant C, Motte F, Notton G, et al. Prediction intervals for irradiation forecasting using regression trees methods[J]. Renewable Energy, 2018, 126(OCT): 332-340.

[28]Lorenz E, Heinemann D. 1.13 - Prediction of Solar Irradiance and Photovoltaic Power[J]. Comprehensive Renewable Energy, 2012, 1: 239-292.

[29]Alsharif M H, Younes M K, Kim J. Time Series ARIMA Model for Prediction of Daily and Monthly Average Global Solar Radiation: The Case Study of Seoul, South Korea[J]. Symmetry, 2019, 11(2): 240.

[30]李燕斌, 张久菊, 肖俊明，等.基于指数平滑法的灰色预测模型[J]. 中原工学院学报, 2015, 26(04): 1-4.

[31]蒋峰, 王宗耀, 张鹏. 基于灰色-加权马尔可夫链的光伏发电量预测[J]. 电力系统保护, 2019, 47(15): 55-60.

[32]姚宁, 周力民, 陈城.基于回归分析法的光伏发电功率预测模型研究[J]. 机械工程师, 2020(02): 82-84.

[33]刘鑫, 滕欢, 宫毓斌, 等. 基于改进卡尔曼滤波算法的短期负荷预测[J]. 电测与仪表, 2019, 56(03): 42-46.

[34]Wang H, Liu Y, Zhou B, et al. Taxonomy research of artificial intelligence for deterministic solar power forecasting[J]. Energy Conversion and Management, 2020, 214: 112909.

[35]张雨金, 杨凌帆, 葛双冶, 等.基于Kmeans-SVM的短期光伏发电功率预测[J]. 电力系统保护与控制, 2018, 46(21): 118-124.

[36]Pan M, Li C, Gao R, et al. Photovoltaic power forecasting based on a support vector machine with improved ant colony optimization[J]. Journal of Cleaner Production, 2020, 277: 123948.

[37]Wang H Z , Li G Q , Wang G B , et al. Deep learning based ensemble approach for probabilistic wind power forecasting[J]. Applied Energy, 2017, 188(FEB.15):56-70.

[38]谭津, 邓长虹, 杨威, 等. 微电网光伏发电的Adaboost天气聚类超短期预测方法[J].电力系统自动化, 2017, 41(21): 33-39.

[39]Lee J, Wang W, Harrou F, et al. Reliable solar irradiance prediction using ensemble learningbased models: A comparative study[J]. Energy Conversion and Management, 2020, 208: 112582.

[40]蒋建东,余沣,董存,等. 基于PSO与ELM组合算法的短期光伏发电功率预测模型[J].郑州大学学报(理学版), 2019, 51(03): 120-126.

[41]Zhou Y, Zhou N , Gong L , et al. Prediction of photovoltaic power output based on similar day analysis, genetic algorithm and extreme learning machine[J]. Energy, 2020, 204: 117894.

[42]Zang H, Cheng L, Ding T, et al. Hybrid method for short-term photovoltaic power forecasting based on deep convolutional neural network[J]. IET Generation, Transmission & Distribution, 2018, 12(20): 4557-4567.

[43]宋绍剑, 李博涵. 基于LSTM网络的光伏发电功率短期预测方法的研究[J]. 可再生能源, 2021, 39(05): 594-602.

[44]朱坤, 付青. 基于EEMD-Kmeans-ALO-LSTM的短期光伏功率预测[J]. 电源技术, 2023, 47(01): 103-107.

[45]Zhou H, Zhang Y, Yang L, et al. Short-term Photovoltaic Power Forecasting based on Long Short Term Memory Neural Network and Attention Mechanism[J]. IEEE Access, 2019, 7(99): 78063-78074.

[46]张进, 刘运, 彭曙蓉. 基于特征挖掘的GRU-A光伏发电功率预测[J]. 实验室研究与探索, 2020, 39(05): 25-30+49.

[47]孙荣富, 张涛, 和青, 等. 风电功率预测关键技术及应用综述[J]. 高电压技术, 2021, 47(04): 1129-1143.

[48]史如新, 王德顺, 余涛, 等. 基于NARX神经网络-小波分解光伏发电功率预测[J].郑州大学学报(工学版), 2020, 41(06): 79-84.

[49]Hu L, Zhen Z, Li K, et al. An Ultra-Short-Term PV Power Prediction Model Based on Path Space Distance Cross-Similar Clustering and STL Decomposition[C]//2019 IEEE Sustainable Power and Energy Conference (iSPEC). IEEE, 2019: 1353-1358.

[50]Lan H, Zhang C, Hong Y Y, et al. Day-ahead spatiotemporal solar irradiation forecasting using frequency-based hybrid principal component analysis and neural network[J]. Applied Energy, 2019, 247: 389-402.

[51]Behera M K, Nayak N. A comparative study on short-term PV power forecasting using decomposition based optimized extreme learning machine algorithm[J]. Engineering Science and Technology, an International Journal, 2020, 23(1): 156-167.

[52]Liu L, Zhan M, Bai Y. A recursive ensemble model for forecasting the power output of photovoltaic systems[J]. Solar Energy, 2019, 189: 291-298.

[53]杨茂, 孟玲建, 李大勇, 等. 基于混合t Location-Scale分布模型的光伏功率随机性分量波动性分析[J]. 可再生能源, 2017, 35(10): 1494-1499.

[54]孟安波, 许炫淙, 陈嘉铭, 等. 基于强化学习和组合式深度学习模型的超短期光伏功率预测[J]. 电网技术, 2021, 45(12): 4721-4728.

[55]AlKandari M, Ahmad I. Solar power generation forecasting using ensemble approach based on deep learning and statistical methods[J]. Applied Computing and Informatics, 2020(ahead-of-print).

[56]万灿, 宋永华. 新能源电力系统概率预测理论与方法及其应用[J]. 电力系统自动化, 2021, 45(01): 2-16.

[57]Ramakrishna R, Scaglione A, Vittal V, et al. A model for joint probabilistic forecast of solar photovoltaic power and outdoor temperature[J]. IEEE Transactions onSignal Processing, 2019,67(24): 6368-6383.

[58]赵铁军, 孙玲玲, 牛益国, 等. 基于改进非参数核密度估计的光伏出力概率分布建模方法[J]. 燕山大学学报, 2021, 45(05): 430-437+448.

[59]杨召, 徐姣新. 基于分位数回归平均的电力负荷统计建模与预测[J]. 计算机应用与软件, 2021, 38(11): 98-103+204.

[60]孙东磊, 王艳, 于一潇, 等. 基于BP神经网络的短期光伏集群功率区间预测[J]. 山东大学学报(工学版), 2020, 50(05): 70-76.

[61]丁学辉, 许海林, 罗颖婷, 等. 基于CNN特征选择与QRGRU的电力负荷概率密度预测方法[J]. 电力信息与通信技术, 2021, 19(06): 32-38.

[62]赵康宁, 蒲天骄, 王新迎, 等. 基于改进贝叶斯神经网络的光伏出力概率预测[J].电网技术, 2019, 43(12): 4377-4386.

[63]韩家炜, 裴健, 范明, 等. 数据挖掘概念与技术[M]. 北京: 机械工业出版社, 2012.

[64]杨茂, 孟玲建, 李大勇, 等. 基于类3σ准则的光伏功率异常数据识别[J]. 可再生能源, 2018, 36(10): 1443-1448.

[65]王宁, 王恩路, 韩则胤, 等. 基于BP神经网络的光伏发电设备故障检测方法研究[J].自动化仪表, 2023, 44(03): 88-90+97.

[66]肖勇, 赵云, 涂治东, 等. 基于改进的皮尔逊相关系数的低压配电网拓扑结构校验方法[J]. 电力系统保护与控制, 2019, 47(11): 37-43.

[67]王千, 王成, 冯振元, 等. K-means聚类算法研究综述[J]. 电子设计工程, 2012, 20(07): 21-24.

[68]夏宁霞, 苏一丹, 覃希. 一种高效的K-medoids聚类算法[J]. 计算机应用研究, 2010, 27(12): 4517-4519.

[69]朱乔木, 李弘毅, 王子琪, 等. 基于长短期记忆网络的风电场发电功率超短期预测[J]. 电网技术, 2017, 41(12): 3797-3802.

[70]王开艳, 杜浩东, 贾嵘, 等. 基于相似日聚类和QR-CNN-BiLSTM模型的光伏功率短期区间概率预测[J]. 高电压技术, 2022, 48(11): 4372-4388.

[71]Han Y, Wang N, Ma M, et al. A PV power interval forecasting based on seasonal model and non-parametric estimation algorithm[J]. Solar Energy, 2019, 184: 515-526.

[72]张虹, 孟庆尧, 王明晨, 等. 考虑火电机组参与绿证购买交易的含氢综合能源系统经济低碳调度策略[J]. 电力系统保护与控制, 2023, 51(03): 26-35.

[73]宋雨桐, 陈涛, 高赐威, 等. 基于深度强化学习技术的光伏-固体氧化物燃料电池混合能源系统多场景控制[J]. 中国电机工程学报, 2022, 42(22): 8129-8140.