Short-term PV Power Prediction Model Based on VMD-CNN-IBiLSTM

Abstract Because PV power’s generation is unpredictable, so timely and accurate prediction of power has great practical significance research value for daily grid dispatch system security. The combined forecast model presented in our study dependent on variational modal decomposition-convolutional neural network-improved bidirectional long short-term memory network (VMD-CNN-IBiLSTM) with as the object. uses VMD to decompose sequences, thus reducing non-smoothness sequences; CNN extract essential features information; BiLSTM peephole connections learn forward backward temporal data; genetic algorithm enhance model’s variables. For validation, actual data from Australian DKASC testbed are used. According experimental findings, study’s more precise than classical BP network, SVM model, LSTM at making predictions.