基于组合深度学习的光伏功率日前概率预测模型

doi:10.11930/j.issn.1004-9649.202306080

中国电力 ›› 2024, Vol. 57 ›› Issue (4): 100-110.DOI: 10.11930/j.issn.1004-9649.202306080

基于组合深度学习的光伏功率日前概率预测模型

高岩¹(), 吴汉斌¹(), 张纪欣¹(), 张华铭²(), 张沛³()

1. 国网河北省电力有限公司保定供电分公司，河北保定　071000
2. 北京清软创新科技股份有限公司，北京　102208
3. 北京交通大学电气工程学院，北京　100089

收稿日期:2023-06-23 出版日期:2024-04-28 发布日期:2024-04-26
作者简介:高岩（1982—），男，学士，高级工程师，从事新能源运行管理工作，E-mail：bd_gaoy@he.sgcc.com.cn
吴汉斌（1969—），男，学士，高级工程师，从事电网调度运行工作，E-mail：bd_wuhb@he.sgcc.com.cn
张纪欣（1990—），男，学士，工程师，从事新能源调度和功率预测工作，E-mail：bddkzy@126.com
张华铭（1967—），男，硕士，高级工程师，从事电力系统负荷预测、新能源发电功率预测、电力市场研究，E-mail：hmzhang@tsingsoft.com.cn
张沛（1972—），男，通信作者，博士，教授，博士生导师，从事新能源与电力信息化、电力系统规划技术研究，E-mail：pizhang166@qq.com
基金资助:
国网河北省电力公司科技项目（含高比例分布式光伏的多级电网负荷预测方法研究及应用，kj2022-051）。

Day-Ahead Probabilistic Prediction Model for Photovoltaic Power Based on Combined Deep Learning

Yan GAO¹(), Hanbin WU¹(), Jixin ZHANG¹(), Huaming ZHANG²(), Pei ZHANG³()

1. Baoding Power Supply Branch, State Grid Hebei Electric Power Co., Ltd., Baoding 071000, China
2. Beijing Qingsoft Innovation Technology Co., Ltd., Beijing 102208, China
3. School of Electrical Engineering, Beijing Jiaotong University, Beijing 100089, China

Received:2023-06-23 Online:2024-04-28 Published:2024-04-26
Supported by:
This work is supported by State Grid Hebei Electric Power Co., Ltd. (Load Forecasting Methods for Multiple Level Power Grid with High Penetration of Distributed Photovoltaic Generation, No.kj2022-051).

摘要/Abstract

摘要：

为准确量化复杂场景下光伏预测功率的不确定性，提出了一种基于时序卷积网络-注意力机制-长短期记忆网络组合的光伏功率短期概率预测方法。首先，基于多种相关性分析方法选出与光伏功率强相关的气象因素；然后，基于时序卷积网络的特征提取能力和长短期记忆网络的时序特征建模能力，并结合注意力机制和分位数回归，建立组合深度学习预测模型；最后，采用核密度估计方法生成连续概率密度函数。以实际集中式和分布式光伏电站为案例进行分析，结果表明：与长短期记忆网络、时序卷积网络、时序卷积网络-注意力机制和时序卷积网络-长短期记忆网络相比，所提方法在确保最优预测区间的同时，可以提升概率密度预测的性能。

关键词: 概率预测, 时序卷积网络, 长短期记忆网络, 注意力机制

Abstract:

To accurately quantify the uncertainty in the predicted photovoltaic (PV) power in complex scenarios, a short-term probabilistic prediction method for PV power based on a combination of temporal convolutional networks-attention mechanism-long short-term memory networks is proposed in this paper. Firstly, mete-orological factors strongly correlated with PV power are selected based on multiple correlation analysis methods. Then, a combined deep learning prediction model is built based on the feature extraction capability of the temporal convolutional network and the temporal feature modeling capability of the long and short-term memory network, combined with the attention mechanism and quantile regression. Finally, a kernel density estimation method is used to generate a continuous probability density function. The cases of actual centralized and distributed PV plants are analyzed, and the results show that compared with long short-term memory networks, temporal convolutional networks, temporal convolutional networks-attention mechanism, and temporal convolutional networks-long short-term memory networks, the proposed method can improve the performance of probability density prediction while ensuring the optimal prediction interval.

Key words: probabilistic prediction, temporal convolutional network, long short-term memory network, attention mechanism

高岩, 吴汉斌, 张纪欣, 张华铭, 张沛. 基于组合深度学习的光伏功率日前概率预测模型[J]. 中国电力, 2024, 57(4): 100-110.

Yan GAO, Hanbin WU, Jixin ZHANG, Huaming ZHANG, Pei ZHANG. Day-Ahead Probabilistic Prediction Model for Photovoltaic Power Based on Combined Deep Learning[J]. Electric Power, 2024, 57(4): 100-110.

导出引用管理器 EndNote|Ris|BibTeX

链接本文: https://www.electricpower.com.cn/CN/10.11930/j.issn.1004-9649.202306080

https://www.electricpower.com.cn/CN/Y2024/V57/I4/100

图/表 15

图 1 各变量相关关系分析结果

Fig.1 Results of correlation analysis of variables

图 2 TCN-Attention-LSTM预测流程

Fig.2 Probabilistic prediction process of TCN-Attention-LSTM

图 3 TCN的扩张因果卷积结构

Fig.3 Dilated causal convolution structure of TCN

图 4 自注意力机制

Fig.4 Attention mechanism

图 5 注意力机制TCN结构

Fig.5 Structure of TCN with attention mechanism

图 6 LSTM结构

Fig.6 Structure of LSTM

图 7 滑动窗口划分

Fig.7 Sliding window partitioning

图 8 光伏功率日前概率预测工作流程

Fig.8 Workflow of day-ahead probabilistic prediction of PV power

表 1 TCN-Attention-LSTM主要结构超参数设置

Table 1 Hyperparameters setting for main structure of TCN- Attention-LSTM

结构名称		超参数设置
TCN#1		i=2, o=10, p=3, d=1, k=4
TCN#2		i=10, o=10, p=9, d=3, k=4
TCN#3		i=10, o=10, p=18, d=6, k=4
LSTM		i_L=10, n_l=2, h_s=128
全连接层		i_N=128, o_N=58

表 2 模型共同超参数

Table 2 Common hyperparameters of model

参数		数值
学习率		0.001
批处理参数		32
优化器		Adam
训练集∶测试集		9∶1

表 3 90%置信区间各模型评价指标以及模型训练所需要的时间

Table 3 Evaluation metrics of each model within a 90% confidence interval and training time of different models

数据	模型	P_ICP/ %	P_INAW/ kW	S_W分数/ kW	训练时间/s
集中式	LSTM	93.79	0.2979	5.5948	1603.97
	TCN	96.57	0.3473	5.7086	121.89
	TCN-Attention	98.28	0.3693	5.9472	201.41
	TCN-LSTM	95.14	0.3259	5.7438	4260.87
	TCN-Attention-LSTM	97.18	0.3099	5.1763	5008.32
分布式	LSTM	88.68	0.3229	7.6214	912.94
	TCN	96.60	0.4497	10.2056	33.72
	TCN-Attention	92.08	0.3211	7.9540	122.67
	TCN-LSTM	94.34	0.3525	8.9185	1039.98
	TCN-Attention-LSTM	93.96	0.2674	7.2308	1168.41

图 9 晴天时90%置信区间各模型预测对比

Fig.9 Comparison of model predictions within a 90% confidence interval in sunny days

图 10 阴雨天时90%置信区间各模型预测对比

Fig.10 Comparison of model predictions within a 90% confidence interval in rainy days

表 4 各种模型CRPS评价指标

Table 4 CRPS evaluation metrics for various models

数据	模型	C_RPS/kW
集中式	LSTM	0.0589
	TCN	0.0539
	TCN-Attention	0.0507
	TCN-LSTM	0.0433
	TCN-Attention -LSTM	0.0421
分布式	LSTM	0.0478
	TCN	0.0565
	TCN-Attention	0.0505
	TCN-LSTM	0.0445
	TCN-Attention -LSTM	0.0441

图 11 TCN-Attention-LSTM概率密度预测结果

Fig.11 Probability density prediction results of TCN- Attention-LSTM

参考文献 26

1	国家能源局. 2022年光伏发电建设运行情况[EB/OL]. (2023-02-17)[2023-05-25]. http://www.nea.gov.cn/2023-02/17/c_1310698128.htm.
2	卢俊杰, 蔡涛, 郎建勋, 等. 基于集群划分的光伏电站集群发电功率短期预测方法[J]. 高电压技术, 2022, 48 (5): 1943- 1951.
	LU Junjie, CAI Tao, LANG Jianxun, et al. Short-term power output forecasting of clustered photovoltaic solar plants based on cluster partition[J]. High Voltage Engineering, 2022, 48 (5): 1943- 1951.
3	SANGRODY H, ZHOU N, ZHANG Z A. Similarity-based models for day-ahead solar PV generation forecasting[J]. IEEE Access, 2020, 8, 104469- 104478. DOI
4	杨茂, 闫琦, 苏欣, 等. 季节分型下一种面向风电功率日前预测的深度自适应滤波框架[J]. 南方电网技术, 2023, 17 (6): 62- 71. DOI
	YANG Mao, YAN Qi, SU Xin, et al. A depth self-adaptive filtering framework for wind power day-ahead prediction under seasonal classification[J]. Southern Power System Technology, 2023, 17 (6): 62- 71. DOI
5	LINDBERG O, LINGFORS D, ARNQVIST J, et al. Day-ahead probabilistic forecasting at a co-located wind and solar power park in Sweden: trading and forecast verification[J]. Advances in Applied Energy, 2023, 9, 100120. DOI
6	万灿, 崔文康, 宋永华. 新能源电力系统概率预测: 基本概念与数学原理[J]. 中国电机工程学报, 2021, 41 (19): 6493- 6509.
	WAN Can, CUI Wenkang, SONG Yonghua. Probabilistic forecasting for power systems with renewable energy sources: basic concepts and mathematical principles[J]. Proceedings of the CSEE, 2021, 41 (19): 6493- 6509.
7	GOLESTANEH F, PINSON P, GOOI H B. Very short-term nonparametric probabilistic forecasting of renewable energy generation—with application to solar energy[J]. IEEE Transactions on Power Systems, 2016, 31 (5): 3850- 3863. DOI
8	HUANG Q, WEI S Y. Improved quantile convolutional neural network with two-stage training for daily-ahead probabilistic forecasting of photovoltaic power[J]. Energy Conversion and Management, 2020, 220, 113085. DOI
9	项明俊. 基于地基云图的光伏发电功率概率预测[D]. 杭州: 浙江大学, 2022.
	XIANG Mingjun. Probabilistic forecasting of photovoltaic power generation based on ground cloud images [D]. Hangzhou: Zhejiang University, 2022.
10	GU B, SHEN H Q, LEI X H, et al. Forecasting and uncertainty analysis of day-ahead photovoltaic power using a novel forecasting method[J]. Applied Energy, 2021, 299, 117291. DOI
11	MEI F, GU J Q, LU J X, et al. Day-ahead nonparametric probabilistic forecasting of photovoltaic power generation based on the LSTM-QRA ensemble model[J]. IEEE Access, 2020, 8, 166138- 166149. DOI
12	姚程文, 杨苹, 刘泽健. 基于CNN-GRU混合神经网络的负荷预测方法[J]. 电网技术, 2020, 44 (9): 3416- 3424.
	YAO Chengwen, YANG Ping, LIU Zejian. Load forecasting method based on CNN-GRU hybrid neural network[J]. Power System Technology, 2020, 44 (9): 3416- 3424.
13	BAI S J, ZICO KOLTER J, KOLTUN V. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling[J]. ArXiv e-Prints, 2018: arXiv: 1803.01271.
14	宋绍剑, 姜屹远, 刘斌. 一种TCN的改进模型及其在短期光伏功率区间预测的应用[J]. 计算机应用研究, 2023, 40 (10): 3064- 3069.
	SONG Shaojian, JIANG Yiyuan, LIU Bin. Improved TCN model and its application in short-term photovoltaic power interval prediction[J]. Application Research of Computers, 2023, 40 (10): 3064- 3069.
15	庞昊, 高金峰, 杜耀恒. 基于时间卷积网络分位数回归的短期负荷概率密度预测方法[J]. 电网技术, 2020, 44 (4): 1343- 1350.
	PANG Hao, GAO Jinfeng, DU Yaoheng. A short-term load probability density prediction based on quantile regression of time convolution network[J]. Power System Technology, 2020, 44 (4): 1343- 1350.
16	陈禹帆, 温蜜, 张凯, 等. 基于相似日匹配及TCN-Attention的短期光伏出力预测[J]. 电测与仪表, 2022, 59 (10): 108- 116.
	CHEN Yufan, WEN Mi, ZHANG Kai, et al. Short-term photovoltaic output forecasting based on similar day matching and TCN-Attention[J]. Electrical Measurement & Instrumentation, 2022, 59 (10): 108- 116.
17	ZANG H X, CHENG L L, DING T, et al. Day-ahead photovoltaic power forecasting approach based on deep convolutional neural networks and meta learning[J]. International Journal of Electrical Power & Energy Systems, 2020, 118, 105790.
18	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.
	WANG Kaiyan, DU Haodong, JIA Rong, et al. Short-term interval probabilistic forecasting of photovoltaic power based on similar day clustering and QR-CNN-BiLSTM model[J]. High Voltage Engineering, 2022, 48 (11): 4372- 4388.
19	刘昳娟, 陈云龙, 刘继彦, 等. 基于集成学习的分布式光伏发电功率日前预测[J]. 中国电力, 2022, 55 (9): 38- 45.
	LIU Yijuan, CHEN Yunlong, LIU Jiyan, et al. Ensemble learning-based day-ahead power forecasting of distributed photovoltaic generation[J]. Electric Power, 2022, 55 (9): 38- 45.
20	黄书民, 蒋林高, 李志川, 等. 基于PSO寻优与DBN神经网络的电晕损耗预测[J]. 中国电力, 2022, 55 (6): 95- 102, 214.
	HUANG Shumin, JIANG Lingao, LI Zhichuan, et al. Corona loss prediction of UHV AC transmission line based on DBN neural network optimized by PSO[J]. Electric Power, 2022, 55 (6): 95- 102, 214.
21	冯裕祺, 李辉, 李利娟, 等. 基于CNN-GRU的光伏电站电压轨迹预测[J]. 中国电力, 2022, 55 (7): 163- 171.
	FENG Yuqi, LI Hui, LI Lijuan, et al. Voltage trajectory prediction of photovoltaic power station based on CNN-GRU[J]. Electric Power, 2022, 55 (7): 163- 171.
22	刘辉, 凌宁青, 罗志强, 等. 基于TCN-LSTM和气象相似日集的电网短期负荷预测方法[J]. 智慧电力, 2022, 50 (8): 30- 37. DOI
	LIU Hui, LING Ningqing, LUO Zhiqiang, et al. Power grid short-term load forecasting method based on TCN-LSTM and meteorological similar day sets[J]. Smart Power, 2022, 50 (8): 30- 37. DOI
23	符杨, 任子旭, 魏书荣, 等. 基于改进LSTM-TCN模型的海上风电超短期功率预测[J]. 中国电机工程学报, 2022, 42 (12): 4292- 4303.
	FU Yang, REN Zixu, WEI Shurong, et al. Ultra-short-term power prediction of offshore wind power based on improved LSTM-TCN model[J]. Proceedings of the CSEE, 2022, 42 (12): 4292- 4303.
24	赵雅雪, 王旭, 蒋传文, 等. 基于最大信息系数相关性分析和改进多层级门控LSTM的短期电价预测方法[J]. 中国电机工程学报, 2021, 41 (1): 135- 146, 404.
	ZHAO Yaxue, WANG Xu, JIANG Chuanwen, et al. A novel short-term electricity price forecasting method based on correlation analysis with the maximal information coefficient and modified multi-hierarchy gated LSTM[J]. Proceedings of the CSEE, 2021, 41 (1): 135- 146, 404.
25	HE Y Y, XU Q F, WAN J H, et al. Short-term power load probability density forecasting based on quantile regression neural network and triangle kernel function[J]. Energy, 2016, 114, 498- 512. DOI
26	DONG W C, SUN H X, TAN J X, et al. Regional wind power probabilistic forecasting based on an improved kernel density estimation, regular vine copulas, and ensemble learning[J]. Energy, 2022, 238, 122045. DOI

[1]	陶磊, 罗萍萍, 林济铿. 基于深度学习的直流微电网虚假数据注入攻击二阶段检测方法[J]. 中国电力, 2024, 57(9): 11-19.
[2]	杨国华, 祁鑫, 贾睿, 刘一峰, 蒙飞, 马鑫, 邢潇文. 基于CEEMD-SE的CNN&LSTM-GRU短期风电功率预测[J]. 中国电力, 2024, 57(2): 55-61.
[3]	陈璨, 苏紫诺, 马原, 刘佳林, 王玉庆, 王飞. 基于分层关联性建模的分布式光伏功率超短期概率预测[J]. 中国电力, 2024, 57(12): 50-59.
[4]	吴晓刚, 阎洁, 葛畅, 唐雅洁, 倪筹帷, 季青锋. 基于改进GRU-CNN的风光水一体化超短期功率预测方法[J]. 中国电力, 2023, 56(9): 178-186,205.
[5]	陆友文, 崔昊, 陈佳宁, 彭祥佳, 冯双, 刘栋. 基于RA-CNN和同步相量的风电场次/超同步振荡参数智能辨识方法[J]. 中国电力, 2023, 56(4): 46-55,67.
[6]	周震震, 宋云海, 何宇浩, 王黎伟, 黄和燕, 何珏, 朱志航, 闫云凤. 基于分组查询注意力的可扩展电力人员行为分类方法[J]. 中国电力, 2023, 56(11): 77-85.
[7]	苏向敬, 宇海波, 符杨, 田书欣, 李海瑜, 耿福海. 基于DALSTM和联合分位数损失的海上风电功率概率预测[J]. 中国电力, 2023, 56(11): 10-19.
[8]	李冰, 白云山, 赵宽, 郭聪彬, 翟永杰. 基于HSCA-YOLOv7的风电机组叶片表面缺陷检测算法[J]. 中国电力, 2023, 56(10): 43-52.
[9]	陈铁, 张治藩, 李咸善, 陈一夫, 李鸿鑫. 基于混合模态分解和LSTM-CNN的变压器油中溶解气体浓度预测[J]. 中国电力, 2023, 56(1): 132-141.
[10]	樊江川, 于昊正, 刘慧婷, 杨丽君, 安佳坤. 基于多分支门控残差卷积神经网络的短期电力负荷预测[J]. 中国电力, 2022, 55(11): 155-162,174.
[11]	赵会茹, 赵一航, 郭森. 基于互补集合经验模态分解和长短期记忆神经网络的短期电力负荷预测[J]. 中国电力, 2020, 53(6): 48-55.

基于组合深度学习的光伏功率日前概率预测模型

Day-Ahead Probabilistic Prediction Model for Photovoltaic Power Based on Combined Deep Learning

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 15

参考文献 26

相关文章 11

编辑推荐

Metrics

模态框（Modal）标题

基于组合深度学习的光伏功率日前概率预测模型

Day-Ahead Probabilistic Prediction Model for Photovoltaic Power Based on Combined Deep Learning

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 15

参考文献 26

相关文章 11

编辑推荐

Metrics