光伏发电功率受气象因素的影响，呈现出不稳定性和间歇性，准确预测光伏功率有助于实现大规模并网并保障电网的稳定运行。本文以DKASC Solar Centre光伏电站数据为研究对象，提出一种基于气象相似日的变分模态分解算法、长鼻浣熊算法和双向长短期记忆神经网络(VMD-COA-BiLSTM)的光伏功率短期预测模型。针对光伏数据的复杂非线性特征、噪声干扰以及高维特征等问题，通过K均值聚类将数据划分为三种天气类型，增强模型映射能力；利用VMD将聚类之后的原始信号分解，采用中心频率法确定最佳模态数，充分提取集合中的输入因素信息，提高数据质量；将分解后的各分量分别输入BiLSTM网络进行预测，采用COA优化BiLSTM的超参数配置，实现不同天气类型下的光伏功率的准确预测。实验结果表明：本文提出的模型相比于COA-BiLSTM模型，在晴天、多云和阴雨天的RMSE分别降低了35.24%、45.54%和42.88%。

Photovoltaic power is affected by meteorological factors, showing instability and intermittency, and accurate prediction of photovoltaic power can help to realize large-scale and ensure the stable operation of the power grid. This paper focuses on the DKASC Solar Centre photovoltaic plant data and proposes a short-term photovoltaic power forecasting model based on a meteorological similarity day-based Variational Mode Decomposition (VMD) algorithm, Coati Optimization Algorithm (COA), and Bidirectional Long Short-Term Memory Network (BiLSTM) (VMD-COA-BiLSTM). To address the complex nonlinear characteristics, noise interference, and high-dimensional nature of photovoltaic data, K-means clustering is applied to categorize the data into three weather types, enhancing the model's ability to map these variations. VMD is then used to decompose the clustered raw signals, with the optimal number of modes determined using the center frequency method, allowing for the extraction of input factor information and improving data quality. The decomposed components are fed into the BiLSTM network for prediction, and COA is used to optimize BiLSTM's hyperparameters to achieve accurate power forecasting under different weather conditions. Experimental results show that the proposed model outperforms the COA-BiLSTM model, with RMSE reductions of 35.24%, 45.54%, and 42.88% for sunny, cloudy, and rainy days, respectively.

国家自然科学基金资助项目(51877161)