PDF(4469 KB)
PDF(4469 KB)
PDF(4469 KB)
基于TCN和DLinear的光伏发电功率多步预测模型
Multistep Prediction Model for Photovoltaic Power Generation Based on Time Convolution and DLinear
【目的】光伏发电功率预测是提高太阳能利用效率和降低运营成本的关键技术。然而，传统模型在多步光伏发电功率预测中存在时间趋势学习能力不足和误差累积的问题，限制了预测精度的提升。【方法】文章提出了一种基于时间卷积神经网络（temporal convolutional networks，TCN）和DLinear的光伏发电功率多步预测模型。首先，通过改进的自适应白噪声完全集合经验模态分解（improved complete ensemble empirical mode decomposition with adaptive noise，ICEEMDAN）对多元气象序列进行分解，揭示其潜在特征，得到更易学习多尺度特征的多维子序列。其次，利用TCN对局部时序信息进行建模，挖掘短期内的时序特征。最后，利用DLinear将序列分解为趋势分量和残差分量，通过线性网络学习多尺度特征，并直接输出多步（每步15 min）光伏发电功率预测结果。【结果】实验结果表明，所提方法的每个模块均能显著提升模型的预测性能。与集合经验模态分解-卷积神经网络（convolutional neural networks， CNN）-双向长短期记忆网络（bidirectional long short-term memory， BiLSTM）（ICEEMDAN-CNN-BiLSTM）、Informer、Autoformer三个基准模型相比，归一化均方根误差（normalized root mean square error，NRMSE）平均分别下降了22.455%、6.139%、8.504%，具有明显优势。【结论】文章通过ICEEMDAN和TCN-DLinear组合模型，有效解决了传统方法在多步预测中的不足，显著提高了光伏发电功率预测的准确性和可靠性。研究结果为光伏发电功率的精准预测提供了新的技术路径，为太阳能发电的高效管理和运营提供了理论支持，为新型电力系统的安全稳定运行提供了数据支持。未来可进一步探索该模型在不同气象条件和地理环境下的泛化能力，以推动太阳能发电技术的发展。
[Objective] Photovoltaic power forecasting is a key technology to improve the efficiency of solar energy utilization and reduce operating costs. However， the traditional model has the problem of insufficient time-trend learning ability and error accumulation in multistep photovoltaic power prediction， which limits the improvement of prediction accuracy. [Methods] This paper presents a multistep prediction model for photovoltaic power generation based on a temporal convolutional network （TCN） and DLinear combined model. First， it improves the complete ensemble empirical mode decomposition with adaptive noise and ICEEMDAN decomposes multivariate meteorological sequences to reveal their potential features and obtain multidimensional subsequences that make it easier to learn multiscale features. Second， the TCN is used to model the local time sequence information and mine short-time sequence features. Finally， DLinear decomposes the sequence into trend and residual components， learns multiscale features through linear networks， and directly outputs multistep （four-step prediction， 15 min per step） photovoltaic power prediction results. [Results] Experimental results show that each module of the proposed method can significantly improve the prediction performance of the model. Compared with ICEEMDAN-CNN-BiLSTM， Informer， and Autoformer， the normalized root mean square error （NRMSE） decreased by 22.455%， 6.139%， and 8.504% on average， respectively， with obvious advantages. [Conclusions] Through the improved ICEEMDAN decomposition and TCN-DLinear combined model， this study effectively solves the shortcomings of traditional methods in multistep prediction and significantly improves the accuracy and reliability of photovoltaic power prediction. The research results provide a new technical path for the accurate prediction of photovoltaic power generation， theoretical support for the efficient management and operation of solar power generation， and data support for the safe and stable operation of new power systems. In the future， the generalizability of this model under different meteorological conditions and geographical environments should be further explored to promote the development of solar power generation technology.
改进的自适应白噪声完全集合经验模态分解 / 时间卷积神经网络 / DLinear / 光伏预测
improved complete ensemble empirical mode decomposition with adaptive noise / temporal convolutional networks / DLinear / photovoltaic forecast
| [1] |
黄莉, 甘恒玉, 刘兴举, 等. 基于Transformer编码器的超短期光伏发电功率预测[J]. 智慧电力, 2024, 52(5): 16-22, 59.
|
| [2] |
|
| [3] |
胡烜彬, 纪正森, 许晓敏. DPCA-POA-RF-Informer在多情景光伏多步预测中的应用[J]. 智慧电力, 2024, 52(1): 8-13, 22.
|
| [4] |
郑洪庆, 宋万清, 江月松, 等. 长相关随机模型FBM对光伏发电短期预测[J]. 电网与清洁能源, 2024, 40(4): 102-111.
|
| [5] |
|
| [6] |
刘开俊, 高艺, 宋福龙. 多适应性电网规划风险评估框架设计及应用展望[J]. 电力建设, 2015, 34(12): 1-5.
With the development of power grid, its structure is becoming more complicated. The potential operation risk level of power grid is growing with the increase of uncertain factors in power grid. This paper applied risk evaluation in power grid planning, and considered the internal and external uncertainty factors impacting the development of power grid, such as national economic development, the optimal allocation of resources, and the running environment of power grid, which could make power grid planning more economic, reasonable and stronger adaptability. The challenges including large-scale renewable energy integration, the change of power grid structure, the development of modern distribution network, etc., were analyzed at first. The research progress and work of risk evaluation applied in power grid planning at home and abroad were then studied. On this basis, the risk evaluation framework for flexibility power grid planning was proposed, which consisted of evaluation objects, evaluation hierarchy and constraint factors；the overall outline and flowchart of risk evaluation were presented；the future application prospect of risk evaluation technology in the field of power grid planning in China was discussed at last.
|
| [7] |
|
| [8] |
黄泽, 毕贵红, 谢旭, 等. 基于MBI-PBI-ResNet的超短期光伏功率预测[J]. 电力系统保护与控制, 2024, 52(2): 165-176.
|
| [9] |
|
| [10] |
雷柯松, 吐松江·卡日, 伊力哈木·亚尔买买提, 等. 基于WGAN-GP和CNN-LSTM-Attention的短期光伏功率预测[J]. 电力系统保护与控制, 2023, 51(9): 108-118.
|
| [11] |
陈文进, 陈菁伟, 钱建国, 等. 气象特征频繁变化区域的光伏功率预测方法[J]. 浙江电力, 2023, 42(3): 37-46.
|
| [12] |
|
| [13] |
宋玮琼, 赵成, 郭帅, 等. 考虑天气类型和历史相似日的短期光伏输出功率预测[J]. 电网与清洁能源, 2023, 39(2): 75-82.
|
| [14] |
|
| [15] |
陈海宏, 易永利, 黄珅, 等. 基于CatBoost算法的短期光伏功率预测方法[J]. 浙江电力, 2023, 42(2): 67-75.
|
| [16] |
|
| [17] |
|
| [18] |
|
| [19] |
|
| [20] |
|
| [21] |
王海军, 居蓉蓉, 董颖华. 基于时空关联特征与B-LSTM模型的分布式光伏功率区间预测[J]. 中国电力, 2024, 57(7): 74-80.
|
| [22] |
李翰章, 冯江涛, 王鹏程, 等. 基于TDE-SO-AWM-GRU的光伏发电功率预测模型研究[J]. 中国电力, 2024, 57(12): 41-49.
|
| [23] |
|
| [24] |
|
| [25] |
|
| [26] |
龙小慧, 秦际赟, 张青雷, 等. 基于相似日聚类及模态分解的短期光伏发电功率组合预测研究[J]. 电网技术. 2024, 48(7): 2948-2957.
|
| [27] |
李豪, 马刚, 李天宇, 等. 基于时空相关性的短期光伏出力预测混合模型[J]. 电力系统及其自动化学报, 2024, 36(5):121-129.
|
| [28] |
|
| [29] |
张琦, 谢丽蓉, 王威, 等. 计及补偿风电预测误差和平抑波动的混合储能分区优化控制策略[J]. 太阳能学报, 2023, 44(7): 7-13.
在“双碳”背景下,为实现补偿风电预测误差和平抑波动,提出一种计及补偿风电预测误差和平抑波动的控制策略。首先,制定比国家规定更严格的混合储能综合目标域,将目标域分为内、外两部分,采用自适应噪声集合经验模态分解（ICEEMDAN）和逼近理想解排序法（TOPSIS）求得超级电容作用域和蓄电池作用域,超级电容承担变化率较大的部分,蓄电池承担平滑部分。然后,将作用于目标域的混合储能电池组分为4组,根据充/放电参考功率进行状态切换。最后,以新疆某风电场为例进行仿真分析,综合对比多种分解方法和储能配置方法,证明所提策略的有效性。
Under the background of ‘dual carbon’, based on the comprehensive target domain of wind power forecasting error and smoothing fluctuation, a control strategy is proposed to compensate both the prediction error and smoothing fluctuation of wind power. Firstly, the allowable range of prediction error and wind power fluctuation is determined more strictly than that of the national standard, and the comprehensive target area is established. The target domain is divided into internal and external parts, using improved complete ensemble empirical mode decomposition with adaptive noise （ICEEMDAN） and technique for order preference by similarity to an ideal solution （TOPSIS）to calculate the scope of super capacitor and battery, which bear the part with the larger change rate and the smooth part respectively. And then, the hybrid energy storage batteries acting on the target domain are divided into four groups, and the states are switched according to the charge/discharge power. Finally, a wind farm in Xinjiang was taken as an example for simulation analysis, and a comprehensive comparison of various decomposition methods and energy storage configuration methods was made to prove the effectiveness of the proposed strategy.
|
| [30] |
毕贵红, 张梓睿, 赵四洪, 等. 基于多模式分解和多分支输入的光伏功率超短期预测[J]. 高电压技术, 2024, 50(9): 3837-3849.
|
| [31] |
马永光, 冯勇升. 基于IICEEMDAN-PCA-GRU的风电机组齿轮箱故障预警方法研究[J]. 太阳能学报, 2023, 44(4): 67-73.
为充分挖掘数据采集与监控（SCADA）数据的隐藏信息,减少特征间的冗余性,提升模型预测和预警的精度,提出一种双重改进的完全噪声辅助聚合经验模态分解（IICEEMDAN）、主成分分析（PCA）、门控循环网络（GRU）融合的风电机组齿轮箱故障预警方法。使用皮尔逊相关系数法作特征提取,采用IICEEMDAN对特征进行分解,得到特征在不同时间尺度上的连续性信号;利用PCA提取分解特征的关键因素作为网络训练输入;GRU网络对输入时间序列特征进行建模训练,实现对齿轮箱油池温度的预测,使用统计学方法分析油池温度预测值与实际值的误差,根据实际情况设定预警阈值;使用滑动窗口理论实现齿轮箱故障预警。采用华北某风场实际数据进行验证,结果验证了所提方法对齿轮箱早期故障预警的有效性。
In order to fully mine the hidden information of SCADA data, reduce the redundancy among features, and improve the prediction and warning accuracy of the model, a fault warning method for wind turbine gearbox was proposed based on the combination of fully noise-assisted aggregation empirical mode decomposition （IICEEMDAN）, principal component analysis （PCA） and gate recurrent unit（GRU）. Pearson correlation coefficient method was used for feature extraction, IICEEMDAN was used for feature decomposition to obtain the continuity signals of features in different time scales. PCA is used to extract the key factors of decomposition features as network training inputs. GRU network conducts modeling training on the input time series characteristics to predict the oil pool temperature of the gearbox. Statistical theory is used to analyze the error between the predicted value and the actual value of the oil pool temperature, and the early warning threshold is set according to the actual situation. Using sliding window theory to realize gearbox fault warning. A wind field in North China was used to verify the effectiveness of the proposed method for early fault warning of gearbox.
|
| [32] |
赵波, 张领先, 章雷其, 等. 基于TCN和AUKF联合迭代的PEMFC寿命融合预测方法[J/OL]. 中国电机工程学报, 2024. (2024-04-23)[2024-05-21]. https://doi.org/10.13334/j.0258-8013.pcsee.232377.
|
| [33] |
|
| [34] |
苏连成, 朱娇娇, 李英伟. 基于时间卷积网络残差校正的短期风电功率预测[J]. 太阳能学报, 2023, 44(7): 427-435.
为提高短期风电功率预测的准确性,提出一种基于时间卷积网络残差校正模型的短期风电功率预测方法。首先,采取自适应噪声完备集合经验模态分解算法分离出风电功率的局部特征信息,以网格搜索与交叉验证算法优化的支持向量回归模型对各分量进行预测。然后,构建时间卷积网络残差预测模型,并使用灰色关联度分析方法选择输入特征,对支持向量回归预测结果进行校正。最后,基于提出的模型对某风电场实际运行功率进行预测并与其他方法的预测精度进行比较,结果表明,该文所提方法提高了短期风电功率预测的精度。
A short-term wind power prediction method based on temporal convolutional network residual correction model is proposed to improve the accuracy of short-term wind power prediction. Firstly, using the complete ensemble empirical mode decomposition with adaptive noise algorithm to separate the local characteristic information of original wind power data, each component is predicted by the support vector regression model which is optimized by grid search and cross-validation algorithm. Secondly, a temporal convolutional network residual prediction model is constructed, and the gray correlation analysis method is used to select the input features of the residual prediction model to correct the support vector regression prediction results. Finally, based on the proposed model, the actual operating power of a wind farm is predicted and compared with the prediction accuracy of other methods. The results verify that the proposed method improves the accuracy of short-term wind power prediction.
|
AI小编
/
| 〈 |
|
〉 |
AI 导读
