This research paper examines the vital role of reactive power in power systems and proposes a novel method based on deep learning to provide reliable reactive power forecasts. To encapsulate the complexity of various features of a multi-time-series data grid search-based gated recurrent unit (GRU) along with a correlation-based feature selection method is used. Owing to the ever-growing complexity of the power grid, the models are evaluated using real-world datasets of three different power plants in Rajasthan. The findings illustrate that GRU outperforms CNN, LSTM, and ConvLSTM based on MAE, MSE, and RMSE. The results indicate that GRU is a good fit for reliable reactive power forecasting, emphasizing its handling capacity of labeled data without overfitting. This paper aims to highlight how crucial reactive power forecasting is becoming, particularly when considering the integration of renewable energy and dynamic demand profiles. In light of ever-evolving power grid, the suggested models assist in improving grid stability, energy efficiency, and reliability.