Photovoltaic Power Prediction Based on VMD-MBKA-CNN-LSTM

For effective power system planning and scheduling, precision in photovoltaic power prediction is crucial as it allows for optimal resource allocation and operational efficiency. The research unveils a cutting-edge photovoltaic power prediction model established through the synergy of Variational Mode Decomposition (VMD) and a Multi-Strategy Enhanced Black-Winged Kite Algorithm (MBKA), which performs hierarchical optimization of the hyperparameters for a Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM) network model. Initially, VMD is utilized for signal dimensionality reduction. Subsequently, the MBKA, integrated into the model, employs an elite opposite learning strategy to refine the initial population quality. It enhances search efficiency by deploying spiral search and balances global and local searches via Levy flight. Moreover, to optimize the CNN-LSTM network model, the longitudinal and transversal crossover technique is incorporated into the model to boost accuracy and convergence speed. Finally, the MBKA is employed to optimize several key hyperparameters of the CNN-LSTM model, including the number of hidden units, training cycles, and learning rate. Specifically, compared to the VMD-BKA-CNN-LSTM model, the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) are reduced by 22.69%, 25.39%, and 20.96%, respectively.