Combinatorial Prediction Method Based on Meteorological Irradiation Data and GSA-optimized VMD-BiLSTM

To enhance the prediction accuracy of photovoltaic power generation and ensure the safe scheduling and stable operation of the power system, a new combinatorial prediction method is proposed based on variational mode decomposition (VMD) and long short term memory (BiLSTM) optimized by gravitational search algorithm (GSA). To address the issue that the nonlinear and non-stationary characteristics of the temporal signal affect the photovoltaic power generation, meteorological data such as tilt irradiance, horizontal irradiance, temperature and humidity, are selected as the characteristic input variables. These data are subsequently decomposed into several different modes by the VMD algorithm, and each component is modeled individually. To address the uncertainty of parameter selection and slow convergence of traditional BILSTM model, the GSA algorithm is introduced to optimize model parameters. This approach not only shortens the time of manual parameter modulation, but also improves the accuracy and efficiency of hyperparameter configuration. The performance of the proposed model is verified using the actual PV power generation data set of a park in South China. The results indicate that, in comparison to the traditional neural network prediction model and fixed parameter combination prediction model, the proposed method exhibits superior prediction accuracy and stability.