准确而可靠的光伏发电功率预测对于确保电力系统的安全、稳定和经济运行至关重要。然而，由于光伏发电功率自身随机性和波动性较强的特点，很难建立令人十分满意的预测模型，从而也难以满足高精度预测的需求。鉴于此，本文结合变分模态分解、相空间重构以及松鼠觅食算法，提出了一种基于深度多核极限学习机的短期光伏发电功率组合预测模型。本文具体研究内容如下：

（1）针对传统人工神经网络存在特征表达能力不足、预测精度低的问题，构造了一种基于深度多核极限学习机（DMKELM）的短期光伏发电功率预测模型。以深度极限学习机（DELM）为基础模型，引入核函数理论，选择典型的局部核函数（改进高斯核函数)和全局核函数（多项式核函数），构造一种线性组合的多核函数，将其融合至深度极限学习机中，得到深度多核极限学习机，以此实现预测精度提升的目的。通过实例仿真表明，相较于其他传统的模型，本文所提模型具有更好的预测精度。

（2）针对DMKELM预测性能受参数选择的影响，提出一种改进的松鼠觅食算法（ISSA）用来对DMKELM关键参数进行优化。其中，通过引入捕食者概率自适应机制、反向学习机制和t分布变异机制对标准松鼠觅食算法（SSA）进行改进，并选取典型的测试函数进行测试，验证ISSA在收敛速度和寻优性能方面的显著优势。通过实例仿真表明，相较于其他传统模型及智能优化算法，本文所提的ISSA能够更大限度的提高DMKELM的预测性能。

（3）针对光伏发电功率自身随机性和波动性较强的特点，构造一种变分模态分解-相空间重构（VMD-PSR）的数据处理方法，并与ISSA优化DMKELM的单一模型结合，构建短期光伏发电功率组合预测模型。利用VMD算法得到的子序列能够完整的表现光伏发电功率的局部特征，而PSR方法能够对每个子序列内含的混沌特征呈现出来，二者结合能够更好的对光伏发电功率进行特征挖掘，以建立更为可靠的数据集。通过实例仿真表明，所提组合模型在不同天气类型下能够做出精准的预测，且相较于其他组合模型具有更高的稳定性和预测性能。

Accurate and reliable photovoltaic power prediction is essential to ensure the safety, stability, and economic operation of the power system. However, due to the strong randomness and volatility characteristics of photovoltaic power itself, it is diffificult to build satisfactory photovoltaic power prediction models, and thus also difficult to meet the demand for high accuracy prediction. In view of this, this thesis proposes a short-term combined photovoltaic power prediction model based on deep multi-kernel extreme learning machine using variational mode decomposition, phase space reconstruction and squirrel search algorithm. The specific research of this thesis is as follows:

(1) A short-term photovoltaic power prediction model based on deep multi-kernel extreme learning machine (DMKELM) is constructed to address the problems of insufficient feature expression capability and low prediction accuracy of traditional artificial neural networks. Using the deep extreme learning machine (DELM) as the base model, the kernel function theory is introduced, typical local kernel functions (improved Gaussian kernel function) and global kernel functions (polynomial kernel function) are selected, and a linear combination of multiple kernel functions is constructed and fused into the DELM to obtain the DMKELM for the purpose of improving the prediction accuracy. Through example simulation, the results show that the proposed model has better prediction accuracy compared with other traditional models.

(2) An improved squirrel search algorithm (ISSA) is proposed to optimize the key parameters of DMKELM for the influence of DMKELM prediction performance by parameter selection. Among them, this thesis improves the standard squirrel search algorithm (SSA) by introducing a predator probability adaptive mechanism, a opposition-based learning mechanism and a t-distribution variation mechanism, and selects typical test functions for testing to verify the significant advantages of ISSA in terms of convergence speed and optimization performance. Through example simulation, the results show that the ISSA proposed can improve the prediction performance of DMKELM to a greater extent compared with other traditional models and intelligent optimization algorithms.

(3) A variational mode decomposition-phase space reconstruction (VMD-PSR) data processing method is constructed for the characteristics of photovoltaic power itself with strong randomness and volatility, and combined with the single model of ISSA optimized DMKELM to construct a combined short-term photovoltaic power prediction model. The subseries obtained by using the VMD algorithm can fully represent the local features of photovoltaic power, and the PSR method can present the chaotic features embedded in each subseries, and the combination of the two can better feature mining of photovoltaic power to build a more reliable datasets. Through example simulation, the results show that the proposed combined model can make accurate predictions under different weather types and has higher stability and prediction performance than other combined models. 

[1]段光正. 能源革命: 本质探究及中国的选择方向[D]. 河南大学, 2016.

[2]李坤泽. 全球能源变革与能源安全新特征[J]. 国际石油经济, 2023, 31(01): 42-48.

[3]李昕蕾. 全球气候危机中的能源安全韧性治理[J]. 国家治理, 2022(17): 20-25.

[4]姚亚军. 太阳能光伏发电技术的应用研究[J]. 科技创新与应用, 2022, 12(36): 181-184.

[5]董雪, 赵宏伟, 赵生校, 等. 基于SOM聚类和二次分解的BiGRU超短期光伏功率预测[J]. 太阳能学报, 2022, 43(11): 85-93.

[6]尚丽萍. 全球光伏产业发展形势及预测[J]. 太阳能, 2020(05): 16-22.

[7]Li P, Zhou K, Lu X, et al. A hybrid deep learning model for short-term PV power forecasting[J]. Applied Energy, 2020, 259.

[8]赖昌伟, 黎静华, 陈博, 等. 光伏发电出力预测技术研究综述[J]. 电工技术学报, 2019, 34(06): 1201-1217.

[9]闫钇汛, 王丽婕, 郭洪武, 等. 基于多特征分析和提取的短期光伏功率预测[J]. 高电压技术, 2022, 48(09): 3734-3743.

[10]Zhen Z, Liu J, Zhang Z, et al. Deep learning based surface irradiance mapping model for solar PV power forecasting using sky image[J]. IEEE Transactions on Industry Applications, 2020, 56(04): 3385-3396.

[11]Mishra M, Dash P B, Nayak J, et al. Deep learning and wavelet transform integrated approach for short-term solar PV power prediction[J]. Measurement, 2020, 166(04): 108250.

[12]Hajjaj C, Merrouni A A, Bouaichi A, et al. Evaluation, comparison and experimental validation of different PV power prediction models under semi-arid climate[J]. Energy Conversion and Management, 2018, 173(01): 476-488.

[13]赵滨滨, 王莹, 王彬, 等. 基于ARIMA时间序列的分布式光伏系统输出功率预测方法研究[J]. 可再生能源, 2019, 37(06): 820-823.

[14]Li Y, Yan S, Shu L. An ARMAX model for forecasting the power output of a grid connected photovoltaic system[J]. Renewable Energy, 2014, 66: 78-89.

[15]侯伟, 肖健, 牛利勇. 基于灰色理论的光伏发电系统出力预测方法[J]. 电气技术, 2016(04): 53-58.

[16]Li Y, Luan R, Niu J. Forecast of power generation for grid-connected photovoltaic system based on grey model and Markov chain[C]// IEEE Conference on Industrial Electronics & Applications, Singapore: IEEE, 2008: 1729-1733.

[17]Shi J, Lee W J, Liu Y Q, et al. Forecasting power output of photovoltaic systems based on weather classification and support vector machines[J]. IEEE Transactions on Industry Applications, 2012, 48(03): 1064-1069.

[18]Jang H S, Bae K Y, Park H S, et al. Solar power prediction based on satellite images and support vector machine[J]. IEEE Transactions on Sustainable Energy, 2016, 7(3): 1255-1263.

[19]李勇杰, 秦艳辉. 基于混沌理论及PSO-LS-SVM的新能源出力短期预测方法[J]. 水电能源科学, 2019, 37(12): 194-197.

[20]王育飞, 杨启星, 薛花. 考虑混沌特征的增强型大脑情绪神经网络光伏发电功率超短期预测模型[J]. 高电压技术, 2021, 47(04): 1165-1177.

[21]丁明, 虞海彪, 刘练, 等. 基于多变量相空间重构和RBF神经网络的光伏功率预测方法[J]. 电子测量与仪器学报, 2020, 34(08): 1-7.

[22]蒋小波, 徐小艳, 刘乐平, 等. 天气分类和BP神经网络的光伏短期出力预测[J].电源技术, 2020, 44(12): 1809-1813.

[23]饶宇飞, 刘阳, 李玲玲, 等. 基于改进鸟群算法和极限学习机模型的光伏发电系统输出功率预测研究[J].可再生能源, 2020, 38(10): 1318-1325.

[24]刘佳林, 李刚, 王腾飞. 基于Elman神经网络模型的短期光伏发电功率预测[J]. 科技与创新, 2019(02): 43-45.

[25]Jiang Y, Zhang Q, Zhong F. Ultra-short-term PV power forecasting based on FA-DELM model considering weather classification and vmd noise reduction[C]//2022 IEEE 4th International Conference on Civil Aviation Safety and Information Technology (ICCASIT). Dali, China: IEEE, 2022: 1229-1239.

[26]王振浩, 王翀, 成龙, 等. 基于集合经验模态分解和深度学习的光伏功率组合预测[J]. 高电压技术, 2022, 48(10): 4133-4142.

[27]李贞, 崔丽艳, 陶颍军, 等. 基于博弈论赋权的光伏功率组合预测模型[J]. 电气技术, 2017(05): 24-29.

[28]叶斌. 基于LSTM模型的大坝安全监测数据异常值检测[D]. 长江科学院, 2020.

[29]文立, 钟永. 光伏发电功率预测中数据预处理方法[J]. 中国科技信息, 2022(19): 132-134.

[30]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.

[31]李鹏钦. 基于PWVRSM算法的光伏发电功率预测研究[D].昆明理工大学, 2021.

[32]高华兴. 南疆地区光伏功率预测及农业大棚负荷可时移的研究[D]. 塔里木大学, 2021.

[33]赵莉, 孙娜, 李丽萍, 等. 拉格朗日插值法在数据清洗中的应用[J]. 辽宁工业大学学报(自然科学版), 2022, 42(02): 102-105+117.

[34]吕伟杰, 方一帆, 程泽. 基于模糊C均值聚类和样本加权卷积神经网络的日前光伏出力预测研究[J]. 电网技术, 2022, 46(01): 231-238.

[35]Vandeventer W, Jamei E, Thirunavukkarasu G S, et al. Short-term PV power forecasting using hybrid GASVM technique[J]. Renewable energy, 2019, 140: 367-379.

[36]廖尔泰. 基于增强循环神经网络的短期光伏发电功率区间预测方法研究[D]. 广东工业大学, 2022.

[37]刘佳林. 基于相似日理论和PCA-IPSO-Elman模型的短期光伏发电功率预测研究[D].兰州交通大学, 2019.

[38]李晶, 许洪华, 赵海翔, 等.并网光伏电站动态建模及仿真分析[J]. 电力系统自动化, 2008, 32(24): 83-87.

[39]商立群, 朱伟伟. 基于全局学习自适应细菌觅食算法的光伏系统全局最大功率点跟踪方法[J]. 电工技术学报, 2019, 34(12): 2606-2614.

[40]徐宏锐. 分层考虑气象因素的电力系统短期负荷预测[D]. 华北电力大学(河北), 2009.

[41]孟安波, 陈嘉铭, 黎湛联, 等. 基于相似日理论和CSO-WGPR的短期光伏发电功率预测[J]. 高电压技术, 2021, 47(04): 1176-1184.

[42]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.

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

[44]Ahmed R, Sreeram V, Mishra Y, et al. A review and evaluation of the state-of-the-art insolar power forecasting: techniques and optimization[J]. Renewable and Sustainable Energy Reviews, 2020, 124: 109792.

[45]张美霞, 李丽, 杨秀, 等. 基于高斯混合模型聚类和多维尺度分析的负荷分类方法[J].电网技术, 2020, 44(11): 4283-4296.

[46]Li K, Ma Z, Robinson D, et a1. Identification of typical building daily electricity usage profiles using Gaussian mixture model-based clustering and hierarchical clustering[J]. Applied Energy, 2018, 231: 331-342.

[47]Huang G, Zhu Q, Siew C. Extreme learning machine: a new learning scheme of feedforward neural networks[C]//2004 IEEE Intermational Joint Conference on Neural Networks. Budapest, Hungary: IEEE, 2004: 985-990.

[48]商立群, 李洪波, 侯亚东, 等. 基于特征选择和优化极限学习机的短期电力负荷预测[J]. 西安交通大学学报, 2022, 56(04): 165-175.

[49]何金, 曹梦, 王伟, 等. 基于稀疏自编码和极限学习机的局部放电模式识别[J].高压电器, 2018, 54(11): 295-300+306.

[50]颜学龙, 马润平. 基于深度极限学习机的模拟电路故障诊断[J]. 计算机工程与科学, 2019, 41(11): 1911-1918.

[51]曾亮, 雷舒敏, 王珊珊, 等. 基于OVMD-SSA-DELM-GM模型的超短期风电功率预测方法[J]. 电网技术, 2021, 45(12): 4701-4712.

[52]桑发文. 深度核极限学习机和相似性LSSVM时间序列预测模型研究[D]. 兰州大学, 2019.

[53]钱立恒. 基于合成核支持向量机的股指期货价格预测研究[D]. 东南大学, 2020.

[54]江晓苏. 鱼群优化的多核支持向量机在软件缺陷预测中的应用[D]. 重庆师范大学, 2016.

[55]Jain M, Singh V, Rani A. A novel nature-inspired algorithm for optimization: squirrel search algorithm[J]. Swarm and Evolutionary Computation, 2019, 44(02): 148-175.

[56]Sakthivel V. P., Suman M., Sathya P. D. Combined economic and emission power dispatch problems through multi-objective squirrel search algorithm[J]. Applied Soft Computing, 2021,100(1): 106950.

[57]Fares D, Fathi M, Shams I, et al. A novel global MPPT technique based on squirrel search algorithm for PV module under partial shading conditions[J]. Energy Conversion and Management, 2021, 230: 113773,

[58]王秋萍, 王梦娜, 王晓峰. 改进收敛因子和比例权重的灰狼优化算法[J]. 计算机工程与应用, 2019, 55(21): 60-65+98.

[59]郑婷婷, 刘升, 叶旭. 自适应t分布与动态边界策略改进的算术优化算法[J].计算机应用研究, 2022, 39(05): 1410-1414.

[60]何庆, 林杰, 徐航. 混合柯西变异和均匀分布的蝗虫优化算法[J]. 控制与决策, 2021, 36(07): 1558-1568.

[61]Desert Knowledge Australia Centre. Download Data. Location (DKASC, Alice Springs)[2022-02-01]. http:// dkasolarcentre.com.au/download.

[62]Dutta S, Li Y, Venkataraman A, et al. Load and renewable energy forecasting for a microgrid using persistence technique [J]. Energy Procedia, 2017, 143: 617-622.

[63]Dragomiretskiy K, Zosso D. Variational mode decomposition[J]. IEEE Transactions on Signal Processing, 2014, 62(03): 531-544.

[64]Li H, Liu T, Wu X, et al. An optimized VMD method and its applications in bearing fault diagnosis[J]. Measurement, 2020, 166: 108185.

[65]Lahmiri, Salim. A variational mode decompoisition approach for analysis and forecasting of economic and financial time series[J]. Expert Systems with Application, 2016, 55(08): 268-273.

[66]张亚超, 刘开培, 秦亮. 基于VMD-SE和机器学习算法的短期风电功率多层级综合预测模型[J]. 电网技术, 2016, 40(05): 1334-1340.

[67]Rostaghi M, Azami H. Dispersion entropy: a measure for time-series analysis[J]. IEEE Signal Processing Letters, 2016, 23(05): 610-614.

[68]Yan X ,Tang G ,Wang X . A chaotic feature extraction based on SMMF and CMMFD for early fault diagnosis of rolling bearing[J]. IEEE Access, 2020, 8: 179497-179515.

[69]Pourasad Y, Ranjbarzadeh R, Mardani A. A new algorithm for digital image encryption based on chaos theory[J]. Entropy, 2021, 23(03): 341.

[70]杨茂, 孙志博, 苏欣. 基于混沌理论的风电功率超短期预测模式改进研究[J]. 中国电机工程学报, 2022, 42(22): 8117-8129.

[71]刘子玥. 基于混沌理论和支持向量机的短期电力负荷预测[D]. 燕山大学, 2017.

[72]虞海彪. 基于混沌时间序列的光伏功率预测方法研究[D]. 合肥工业大学, 2020.

[73]Rosenstein M T, Collins J J, Luca C J D. A practical method for calculating largest Lyapunov exponents from small data sets[J]. Physica D: Nonlinear Phenomena, 1993, 65(1-2): 117-134.