基于OVMD-CNN-BiLSTM的短期光伏发电预测

Proposes an OVMD-CNN-BiLSTM model optimized by Grey Wolf Optimizer (GWO) for short-term PV power forecasting. OVMD decomposes preprocessed PV data into frequency-separated modes; each mode plus meteorological features feeds into CNN-BiLSTM for training/testing; final prediction is reconstructed. Validation on Ningxia PV plant data shows superior accuracy and stability over benchmark models.

提出一种基于灰狼优化算法(GWO)改进变分模态分解(VMD)的卷积神经网络-双向长短期记忆网络(CNN-BiLSTM)预测模型以提高短期光伏发电功率的预测精度.首先,将预处理后的光伏发电数据通过最优变分模态分解(OVMD)算法分解为多个不同频率的模态分量;然后,将各模态信号及相关影响因素作为CNN-BiLSTM网络的输入进行训练、验证和测试;最后,将各模态分量的预测结果重构相加并分析预测误差.通过宁夏某光伏电站实测数据验证表明,该文提出的OVMD-CNN-BiLSTM模型相较于多种基准模型在预测精度和稳定性方面具有显著优势.