Research on Photovoltaic Power Prediction Method Based on Dynamic Similar Selection and Bidirectional Gated Recurrent Unit

Abstract

Photovoltaic (PV) power generation is vital for sustainable energy development, yet its inherent randomness and volatility challenge grid stability. Accurate short-term PV power prediction is essential for reliable operation. This paper proposes an integrated prediction method combining dynamic similar selection (DSS), variational mode decomposition (VMD), bidirectional gated recurrent unit (BiGRU), and an improved sparrow search algorithm (ISSA). First, DSS selects training data based on local meteorological similarity, reducing randomness interference. VMD then decomposes PV power data into smooth components, mitigating volatility. The Pearson correlation coefficient is used to filter highly relevant meteorological variables, enhancing input quality. BiGRU captures temporal evolution patterns, with ISSA optimizing key parameters for robust forecasting. Validated on historical Australian PV data under diverse weather conditions, the proposed method effectively reduces randomness and volatility, significantly improving prediction accuracy and reliability. These advancements support stable PV power supply and efficient grid operation.