GRU-ATT high-precision short-term PV power combination prediction model combining improved MVMD and INFO

The objective of this study is to address the issue of inadequate prediction accuracy in short-term photovoltaic (PV) power forecasting. To this end, we propose a novel combined prediction model that integrates the gated recurrent unit (GRU) with an enhanced multivariate variational modal decomposition (MVMD) and a vector-weighted averaging algorithm (INFO), along with an attention mechanism (ATT). Initially, the improved multivariate variational mode decomposition (iMVMD), or iMVMD, is employed to decompose the original data, thereby reducing the volatility of historical photovoltaic (PV) power generation data. Secondly, the improved INFO (imINFO) is employed to optimize the GRU and ascertain its optimal parameters, thereby enhancing the performance of the GRU model. Additionally, the attention mechanism is integrated to accentuate the more pivotal influencing factors, thereby enabling the GRU neural network to render more precise assessments. Ultimately, the dataset of photovoltaic power plants in the Ningxia region of China is selected for predictive analysis. The findings demonstrate that the combined prediction model proposed in this paper can markedly enhance the prediction accuracy of PV power.