Joint scenario generation method for meteorology-wind-solar power output based on retrieval-augmented probabilistic downscaling
Received date: 2026-04-27
Revised date: 2026-06-12
Online published: 2026-09-03
The operation and planning of power systems with high penetrations of renewable energy highly depend on high- temporal-resolution meteorological data and wind-solar power output scenarios.However,most existing climate models only provide daily-scale meteorological information,which is insufficient to directly support the hourly-level operational analysis of power systems.Therefore,this paper proposes a joint scenario generation method for meteorology-wind- solar power outputs based on retrieval-augmented probabilistic downscaling(RAPD).First,a similar-day retrieval mechanism based on the rank-sum distance is constructed to extract historical high-resolution meteorological baseline sequences. Second,a Transformer-based conditional variational autoencoder(T-CVAE)learning network is built to realize the probabilistic generation of meteorological fluctuation residuals. Finally,a physical mapping model from meteorological variables to wind-solar power outputs is established to generate continuous hourly wind-solar power output scenarios for a whole year.Validations based on historical meteorological data demonstrate that the mean absolute error(MAE)values for temperature and wind speed are reduced by 30.17%and 28.84% ,respectively,compared with the traditional statistical analog method.Furthermore,taking the daily-scale data from climate models as input,the corresponding hourly-level meteorology-wind-solar power output scenarios are generated through downscaling,and the consistency between the downscaled results and the daily-scale inputs is verified.The results indicate that the proposed method can effectively accomplish temporal downscaling and scenario generation, providing a high-temporal-resolution scenario foundation for renewable energy output analysis under future scenarios.
Yihe ZHAO , Jian CHEN , Yiran WANG , Wen ZHANG , Lei DING . Joint scenario generation method for meteorology-wind-solar power output based on retrieval-augmented probabilistic downscaling[J]. Electric Power, 2026 , 59(8) : 78 -91 . DOI: 10.11930/j.issn.1004-9649.202604075
表 1 T—CVAE 网络结构参数Table 1 Parameters of T-CVAE network structure |
| 参数类别 | 参数名称 | 局部候选范围 | 设置值 |
|---|---|---|---|
| T-CVAE网络 | 主特征维度 | [32,64,128,256] | 128 |
| 潜变量维度 | [16,32,64,128] | 32 | |
| 注意力头数 | [2,4,6,8] | 4 | |
| 前馈层宽度 | [32,64,128,256] | 256 | |
| 训练超参数 | 优化器 | Adam | |
| 初始学习率 | | 1×10-3 | |
| 批大小 | [16,32,64,128] | 64 | |
| 训练轮数 | [100,150,200,250,300,350,400] | 300 | |
| 边界约束权重 | [0.5,1,1.5,2,2.5,3] | 2 | |
| KL散度权重 | [0.05,0.1,0.15,0.2] | 0.1 |
表 2 生成场景评估指标对比Table 2 Comparison of evaluation indicators for generated scenarios |
| 变量 | 算法 | CRPS | MAE | WD | ACF/% | PACF/% |
|---|---|---|---|---|---|---|
| 温度 | SAD | 1.170 | 0.284 | 4.130 | 10.246 | |
| T-CVAE | 0.749 | 1.031 | 0.375 | 3.688 | 7.477 | |
| RAPD | 0.579 | 0.817 | 0.252 | 3.042 | 5.876 | |
| 比湿 | SAD | 0.677 | 0.084 | 4.691 | 4.426 | |
| T-CVAE | 0.370 | 0.510 | 0.156 | 3.372 | 4.475 | |
| RAPD | 0.337 | 0.475 | 0.063 | 1.679 | 3.390 | |
| 辐照度 | SAD | 14.579 | 1.079 | 0.047 | 3.168 | |
| T-CVAE | 11.234 | 14.739 | 1.347 | 0.053 | 3.961 | |
| RAPD | 9.091 | 12.391 | 0.936 | 0.049 | 3.185 | |
| 风速 | SAD | 0.690 | 0.038 | 3.744 | 3.470 | |
| T-CVAE | 0.375 | 0.532 | 0.047 | 1.587 | 2.729 | |
| RAPD | 0.352 | 0.491 | 0.034 | 1.238 | 2.382 |
图6 生成场景95%预测区间性能对比Fig. 6 Performance comparison of 95%prediction interval for generated scenarios |
图8 气象变量历史与生成场景散点对比Fig. 8 Comparison of historical scenarios and generated scenarios of meteorological variables |
表3 变量间相关性评估结果Table 3 Correlation evaluation results of variables |
| 变量 | 历史场景 | SAD | T-CVAE | RAPD |
|---|---|---|---|---|
| 温-湿 | 0.6823 | 0.5934 | 0.5983 | 0.6264 |
| 温-辐照 | 0.4064 | 0.4799 | 0.4614 | 0.4387 |
| 温-风 | 0.1151 | 0.1427 | 0.1374 | 0.1053 |
| 湿-辐照 | -0.0485 | -0.1089 | -0.1058 | -0.1016 |
| 湿-风 | -0.1199 | -0.1209 | -0.1359 | -0.1213 |
| 辐照-风 | 0.2603 | 0.2485 | 0.2839 | 0.2586 |
图10 生成数据与 CMIP6 日均值时序叠加序列Fig. 10 Superimposed time series of generated data and CMIP6 daily averages |
表4 风机与光伏模型参数Table 4 Parameters of PV and wind turbine models |
| 设备 | 参数名称 | 数值 |
|---|---|---|
| 风机 | 平均海拔 Z/m | 108 |
| 轮毂高度 h/m | 80 | |
| 风速幂律指数 α | 0.14 | |
| 综合效率 ηw/\% | 94 | |
| 光伏 | 光伏板高度 hpv/m | 2 |
| 组件基础散热系数 u0/(W⋅m-2⋅ K-1) | 25 | |
| 风速相关对流散热系数 u1/(W⋅s⋅m-3⋅ K-1) | 6.84 | |
| 材料功率温度系数 γ/(\%⋅∘C-1) | 0.4 | |
| 综合效率 ηpv/\% | 94 |
| [1] |
石宇龙, 彭乔, 刘天琪, 等. 基于相似日和双层校正 LSTM 的光伏功率短期预测方法[J]. 电力工程技术, 2026, 45(3):85-94.
|
| [2] |
皮晟源, 刘牧阳, 陈雨田, 等. 基于 IPOA-VMD 的风储联合发电系统混合储能容量配置[J]. 电力系统及其自动化学报, 2025, 37(8): 87-96.
|
| [3] |
李夫刚, 李夫海, 琚承乾, 等. 新能源不确定性表征方法及关键技术问题研究现状、挑战及展望[J]. 电力系统保护与控制, 2025, 53(15):172-187.
|
| [4] |
李文武, 万梓幸, 何睦, 等. 基于概率调控的风光联合出力极端场景生成方法[J]. 电力系统自动化, 2026, 50(5):197-208.
|
| [5] |
黄学勤, 杨鹏举, 赵耀, 等. 考虑多维气象数据与时间影响的风电功率区间预测[J]. 浙江电力, 2026, 45(1):66-77.
|
| [6] |
姜通海, 王峰, 刘子琪, 等. 基于改进生成对抗网络的风光气象资源联合场景生成方法[J]. 中国电力, 2025,58(3): 183-192. JIANG Tonghai,WANG Feng,LIU Ziqi,et al. A joint scenario generation method for wind-solar meteorological resources based on improved generative adversarial network[J]. Electric Power,2025, 58(3):183-192.
|
| [7] |
王成磊, 匡熠, 濮永现, 等. 基于改进生成对抗网络场景生成的长-短期储能优化配置[J]. 高电压技术, 2024, 50(3):1131-1144, I0016-I0020.
|
| [8] |
|
| [9] |
徐聪, 徐静静, 江婷, 等. 数据驱动的综合能源系统运行优化研究[J]. 综合智慧能源, 2026, 48(1):34-42.
|
| [10] |
袁家海, 牟琪林, 张浩楠, 等. 考虑气候风险耦合影响的中长期电力资源规划[J]. 气候变化研究进展, 2024, 20(5):624-635.
|
| [11] |
郭红霞, 陈凌轩, 张启, 等. 电力电量平衡视角下新型电力系统极端场景研究及应对综述[J]. 电网技术, 2024, 48(10):3975-3991, I0002,I0004,I0005.
|
| [12] |
|
| [13] |
|
| [14] |
|
| [15] |
|
| [16] |
|
| [17] |
李丹, 梁云嫣, 缪书唯, 等. 基于高斯混合聚类和改进条件变分自编码的多风电场功率日场景生成方法[J]. 中国电力, 2024, 57(12): 17-29.
|
| [18] |
|
| [19] |
|
| [20] |
|
| [21] |
|
| [22] |
|
| [23] |
石铖, 高红均, 罗龙波, 等. 极端高温场景下计及需求响应的配电网动态重构[J]. 电网技术, 2025, 49(1):136-145,I0055-I0058.
|
| [24] |
|
| [25] |
高红均, 郭明浩, 刘俊勇, 等. 从四川高温干旱限电事件看新型电力系统保供挑战与应对展望[J]. 中国电机工程学报, 2023, 43(12):4517-4537. GAO Hongjun,GUO Minghao,LIU Junyong,et al. Power supply
challenges and prospects in new power system from Sichuan electricity curtailment events caused by high-temperature drought weather[J]. Proceedings of the CSEE, 2023, 43(12):4517-4537.
|
| [26] |
吴晓刚, 阎洁, 葛畅, 等. 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. WU Xiaogang, YAN Jie,GE Chang,et al. Ultra-short-term power forecasting method for wind-solar-hydro integration based on improved GRU-CNN[J]. Electric Power,2023, 56(9):178-186, 205.
|
| [27] |
|
| [28] |
|
| [29] |
|
| [30] |
|
/
| 〈 |
|
〉 |