Evaluation of Power Grid Investment Capacity Using GRU Neural Network to Promote Sustainable Asset Growth

This study introduces a novel power grid investment capacity evaluation model called the Dual-Skip Improved GRU Network (DS-IGRU), which is based on the Gated Recurrent Unit (GRU). The primary objective is to enhance the precision of power grid investment forecasts. The DS-IGRU model leverages the memory capabilities of GRU, integrated with Skip layers and an autoregressive component, to effectively process both periodic and non-periodic data. The model’ s effectiveness has been confirmed through comparisons with various traditional and contemporary forecasting techniques. The experimental outcomes demonstrate that the DS-IGRU model achieves superior prediction accuracy, with a root mean square error (RMSE) of 1.91E-5, a mean absolute error (MAE) of 0.0023, and a mean absolute percentage error (MAPE) of 0.933%. In contrast, the DS-IGRU model without the autoregressive module exhibits reduced prediction accuracy, highlighting the significant contribution of the autoregressive layer to the model’s forecasting capabilities. This study offers a fresh perspective on power grid investment forecasting and provides more accurate references for investment decision-making in the electricity sector.