Reliable photovoltaic power prediction is crucial to power dispatching and power grid management. Aiming at the problems that the existing photovoltaic power prediction has low accuracy and cannot reflect the power variation range, this paper proposes a multi-factor photovoltaic power prediction model based on an improved fuzzy C-means (IFCM), variational mode decomposition (VMD) optimized by modified cosine similarity (MCS), least squares support vector machine (LSSVM) optimized by golden jackal optimization (GJO), long-short term memory (LSTM) optimized by weighted mean of vectors algorithm (INFO) and multi-objective hummingbird optimization algorithm (MOAHA), named MVMD-GLSSVM-ILSTM-MOAHA, is proposed. Firstly, propose an improved fuzzy C-means, named IFCM, screen the major meteorological factors influencing photovoltaic power by combined feature determination method, and aggregate the daily photovoltaic power with similar meteorological characteristics into the same category. Secondly, propose using MCS to determine the decomposition level of VMD, named MVMD, and use it to decompose all kinds of photovoltaic power daily data to get intrinsic mode functions (IMFs). Thirdly, propose LSSVM optimized by GJO and LSTM optimized by INFO, named GLSVM and ILSTM, train and test IMFs respectively, and obtain the prediction results of IMFs. Finally, introduce MOAHA to dynamically weight the prediction results of IMFs to obtain the final prediction result, and insert evaluation index to quantitatively evaluate it. Collect the data of Yulala photovoltaic power station as experimental data, and establish 11 comparison models to compare with the proposed model. Verify its superiority by statistical methods such as DM test, Taylor chart and scatter chart. The results show that RMSE, MAE, MAAPE and R are 1.1716, 0.7509, 0.2674 and 0.9986, respectively, and the correlation coefficient reaches 0.99, which shows that the proposed model has excellent prediction effect. After that, use kernel density estimation (KDE) to make known the variation range of photovoltaic power, and achieve the perfect combination of deterministic prediction and uncertain prediction.

中文翻译：

---

基于多目标优化和数据集成策略的光伏发电功率精准预测方法


可靠的光伏功率预测对于电力调度和电网管理至关重要。针对现有光伏功率预测精度不高、不能反映功率变化范围的问题，提出一种基于改进模糊C均值(IFCM)、变分模态分解(VMD)的多因素光伏功率预测模型通过修正余弦相似度（MCS）优化、通过金豺优化（GJO）优化的最小二乘支持向量机（LSSVM）、通过向量加权平均算法（INFO）优化的长短期记忆（LSTM）和多目标蜂鸟优化提出了一种MOAHA算法，命名为MVMD-GLSSVM-ILSTM-MOAHA。首先，提出一种改进的模糊C均值IFCM，通过组合特征判定方法筛选影响光伏发电的主要气象因素，将具有相似气象特征的日光伏发电量归为同一类别。其次，提出利用MCS确定VMD的分解层次，命名为MVMD，并用其分解各类光伏发电日数据，得到本征模态函数(IMF)。第三，提出GJO优化的LSSVM和INFO优化的LSTM，命名为GLSVM和ILSTM，分别训练和测试IMF，得到IMF的预测结果。最后引入MOAHA对IMF的预测结果进行动态加权得到最终的预测结果，并插入评价指标进行定量评价。收集尤拉拉光伏电站数据作为实验数据，建立11个对比模型与本文提出的模型进行比较。通过DM检验、泰勒图、散点图等统计方法验证其优越性。 结果表明，RMSE、MAE、MAAPE和R分别为1.1716、0.7509、0.2674和0.9986，相关系数达到0.99，表明所提模型具有优异的预测效果。然后利用核密度估计（KDE）获知光伏发电功率的变化范围，实现确定性预测与不确定性预测的完美结合。