针对光伏功率预测中气象特征冗余、模型多时间尺度动态响应不足、组合预测方法缺乏动态适应性等问题,提出一种基于因果特征提取与改进强化学习（Q-L）算法的超短期预测模型。首先,采用基于熵值计算的因果特征提取方法筛选关键气象因子,该方法能反映非线性耦合关系,提升特征选择质量;其次,构建由双向长短期记忆网络（BiLSTM）与TCN-Transformer组成的并行预测模型,分别捕捉短期时序波动、局部模式与长期依赖关系,实现多尺度特征融合;然后,提出一种基于改进动态奖惩机制的Q-L算法,通过迭代反馈动态调节各模型预测权重并加权输出最终预测结果;最后,在3种典型天气下进行多次仿真验证,对比实验与结果显示：该模型预测性能稳定,雨天时与单一模型BiLSTM相比ERMSE和EMAE平均下降49.3%与51.9%,预测精度与泛化能力得到提升。
Abstract
To address the issues of redundant meteorological features, insufficient multi-timescale dynamic response, and lack of dynamic adaptability in ensemble forecasting methods for photovoltaic power prediction, this paper proposes an ultra-short-term prediction model based on causal feature extraction and an improved Q-L algorithm. First, a causal feature extraction method based on entropy calculation is adopted to screen key meteorological factors. This method can reflect nonlinear coupling relationships and improve feature selection quality. Second, a parallel prediction model combining a bidirectional long short-term memory network （BiLSTM） and a TCN-Transformer is constructed to capture short-term temporal fluctuations, local patterns, and long-term dependencies, achieving multi-scale feature fusion. Third, an improved Q-L algorithm with a dynamic reward-punishment mechanism is proposed to adaptively adjust the prediction weights of each model through iterative feedback and produce the final weighted prediction result. Finally, multiple simulation validations are carried out under three typical weather conditions. Comparative experimental results show that the proposed model achieves stable prediction performance. Under rainy conditions, compared with a single BiLSTM model, the ERMSE and EMAE metrics decrease by an average of 49.3% and 51.9%, respectively, indicating improved prediction accuracy and generalization ability.
关键词
光伏功率 /
特征提取 /
强化学习 /
长短期记忆网络 /
TCN-Transformer /
Q-L算法
Key words
photovoltaic power /
feature extraction /
reinforcement learning /
long short-term memory network /
TCN-Transformer /
Q-L algorithm
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 贾凌云, 云斯宁, 赵泽妮, 等. 神经网络短期光伏发电预测的应用研究进展[J]. 太阳能学报, 2022, 43(12): 88-97.
Jia L Y, Yun S N, Zhao Z N, et al.Recent progress of short-term forecasting of photovoltaic generation based on artificial neural networks[J]. Acta Energiae Solaris Sinica, 2022, 43(12): 88-97.
[2] Yang X D, Xu C B, Zhang Y B, et al.Real-time coordinated scheduling for ADNs with soft open points and charging stations[J]. IEEE Transactions on Power Systems, 2021, 36(6): 5486-5499.
[3] 吉兴全, 赵国航, 叶平峰, 等. 基于QMD-HBiGRU的短期光伏功率预测方法[J]. 高电压技术, 2024, 50(9): 3850-3859.
Ji X Q, Zhao G H, Ye P F, et al.Short-term PV forecasting method based on the QMD-HBiGRU model[J]. High Voltage Engineering, 2024, 50(9): 3850-3859.
[4] 叶林, 裴铭, 路朋, 等. 基于天气分型的短期光伏功率组合预测方法[J]. 电力系统自动化, 2021, 45(1): 44-54.
Ye L, Pei M, Lu P, et al.Combination forecasting method of short-term photovoltaic power based on weather classification[J]. Automation of Electric Power Systems, 2021, 45(1): 44-54.
[5] 唐雅洁, 林达, 倪筹帷, 等. 基于XGBoost的双层协同实时校正超短期光伏预测[J]. 电力系统自动化, 2021, 45(7): 18-27.
Tang Y J, Lin D, Ni C W, et al.XGBoost based bi-layer collaborative real-time calibration for ultra-short-term photovoltaic prediction[J]. Automation of Electric Power Systems, 2021, 45(7): 18-27.
[6] Sun J, Bollt E M.Causation entropy identifies indirect influences, dominance of neighbors and anticipatory couplings[J]. Physica D: Nonlinear Phenomena, 2014, 267: 49-57.
[7] 陈凯, 王小荷, 时欣利, 等. 基于因果推断的图神经网络沥青路面车辙预测方法[J]. 南通大学学报(自然科学版), 2025, 24(1): 18-27, 50.
Chen K, Wang X H, Shi X L, et al.Causal inference-based graph neural network method for predicting asphalt pavement performance[J]. Journal of Nantong University(Natural Science Edition), 2025, 24(1): 18-27, 50.
[8] 崔建勋, 要甲, 赵泊媛. 基于深度学习的短期交通流预测方法综述[J]. 交通运输工程学报, 2024, 24(2): 50-64.
Cui J X, Yao J, Zhao B Y.Review on short-term traffic flow prediction methods based on deep learning[J]. Journal of Traffic and Transportation Engineering, 2024, 24(2): 50-64.
[9] 汪鸿, 朱正甲, 陈建华, 等. 基于人工智能技术与物理方法结合的新能源功率预测研究[J]. 高电压技术, 2023, 49(S1): 111-117.
Wang H, Zhu Z J, Chen J H, et al.Research on new energy power prediction based on artificial intelligence technology and physical method[J]. High Voltage Engineering, 2023, 49(S1): 111-117.
[10] 刘源延, 孔小兵, 马乐乐, 等. 基于小波包变换与深度学习的超短期光伏功率预测[J]. 太阳能学报, 2024, 45(5): 537-546.
Liu Y Y, Kong X B, Ma L L, et al.Ultra-short-term photovoltaic power forecasting based on wavelet packet transform and deep learning[J]. Acta Energiae Solaris Sinica, 2024, 45(5): 537-546.
[11] 谢小瑜, 周俊煌, 张勇军, 等. 基于W-BiLSTM的可再生能源超短期发电功率预测方法[J]. 电力系统自动化, 2021, 45(8): 175-184.
Xie X Y, Zhou J H, Zhang Y J, et al.W-BiLSTM based ultra-short-term generation power prediction method of renewable energy[J]. Automation of Electric Power Systems, 2021, 45(8): 175-184.
[12] 吴珺玥, 赵二刚, 郭增良, 等. 基于Spearman系数和TCN的光伏出力超短期多步预测[J]. 太阳能学报, 2023, 44(9): 180-186.
Wu J Y, Zhao E G, Guo Z L, et al.Ultra-short-term photovoltaic power multi-step prediction based on Spearman coefficient and tcn[J]. Acta Energiae Solaris Sinica, 2023, 44(9): 180-186.
[13] 董俊, 刘瑞, 束洪春, 等. 基于BIRCH聚类的L-Transformer分布式光伏短期发电功率预测[J]. 高电压技术, 2024, 50(9): 3883-3893.
Dong J, Liu R, Shu H C, et al.Short-term distributed photovoltaic power generation prediction based on BIRCH cluster-ing and L-Transformer[J]. High Voltage Engineering, 2024, 50(9): 3883-3893.
[14] 陈振宇, 刘金波, 李晨, 等. 基于LSTM与XGBoost组合模型的超短期电力负荷预测[J]. 电网技术, 2020, 44(2): 614-620.
Chen Z Y, Liu J B, Li C, et al.Ultra short-term power load forecasting based on combined LSTM-XGBoost model[J]. Power System Technology, 2020, 44(2): 614-620.
[15] LI Z Y, LIU X, CHEN L Y.Load interval forecasting methods based on an ensemble of Extreme Learning Machines[C]//2015 IEEE Power & Energy Society General Meeting. Denver, CO, 2015: 1-5.
[16] 李波, 杨永标. 基于改进强化学习的超短期光伏集群功率预测[J]. 信息技术, 2024(5): 86-90, 97.
Li B, Yang Y B.Ultra-short-term photovoltaic cluster power prediction based on improved reinforcement learning[J]. Information Technology, 2024(5): 86-90, 97.
[17] 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.
Meng A B, Xu X C, Chen J M, et al.Ultra short term photovoltaic power prediction based on reinforcement learning and combined deep learning model[J]. Power System Technology, 2021, 45(12): 4721-4728.
[18] 王小凯, 朱小文. 计量检定中3种判别和剔除异常值的统计方法[J]. 中国测试, 2018, 44(S1): 41-44.
Wang X K, Zhu X W.Three statistical methods for distinguishing and eliminating outliers in metrological verification[J]. China Measurement & Test, 2018, 44(S1): 41-44.
[19] 高云龙, 李建鹏, 郑兴莘, 等. 基于自适应近邻信息的模糊C均值聚类算法[J]. 光学精密工程, 2024, 32(7): 1045-1058.
Gao Y L, Li J P, Zheng X S, et al.Fuzzy C-means clustering algorithm based on adaptive neighbors information[J]. Optics and Precision Engineering, 2024, 32(7): 1045-1058.
[20] An G Q, Chen L B, Tan J X, et al.Ultra-short-term wind power prediction based on PVMD-ESMA-DELM[J]. Energy Reports, 2022, 8: 8574-8588.
基金
国家自然科学基金（52177039）; 河南省科技攻关项目（242102241027; 242102210185）; 河南省高等学校重点科研项目（24A470006）
PDF(2388 KB)
