To address the challenges of low prediction accuracy and poor stability caused by the complexity of data,noise interference in signal processing,and difficulty in extracting nonlinear features in current photovoltaic power prediction models,a hybrid prediction model combining second-order mode decomposition(SMD)and CNN-LSTM neural network optimized by Walrus Optimization Algorithm(WaOA)is proposed.Firstly,the photovoltaic data is decomposed using Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)and multiple sub-sequences are reconstructed into low-frequency,medium-frequency,and high-frequency sequences by combining with K-means clustering algorithm(K-means).Secondly,the high-frequency sequences containing residuals are subjected to a second-order decomposition using Variational Mode Decomposition(VMD).Finally,CNN-LSTM models are constructed for each component separately.The prediction results of each component are then superimposed to obtain the final prediction result.The SMD processing method addresses the issues of mode aliasing,excessive low-frequency components,and residual noise in high-frequency components commonly found in traditional data processing methods.The CNN-LSTM model can capture spatial relationships and long-term dependencies in the data,while the WaOA algorithm improves the model performance and efficiency by optimizing the model parameters.The experimental results on the data from a photovoltaic power station in Shaanxi Province demonstrate the higher prediction accuracy of the proposed method through multiple comparative experiments.
关键词
二次模态分解/短期光伏功率预测/海象优化算法/深度学习
Key words
second-order modal decomposition/short-term photovoltaic power prediction/walrus optimization algorithm/deep learning
维普
万方数据