Enhancing Power Grid Stability through Reactive Power Demand Forecasting Using Deep Learning

This paper examines deep learning models for reactive power forecasting in contemporary power systems, emphasizing the Temporal Convolutional Network (TCN), the Transformer, and the models that use both TCN and the Transformer. Reactive power plays a critical role in the stability of power grids, and this paper presents TCN and Transformer models to enhance the forecasting precision of this parameter. Four models, including TCN, Transformer, TCN-Transformer, and Transformer-TCN, were trained and tested based on performance indicators such as Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). Among all the models, the Transformer-TCN hybrid model achieved the best results in all the metrics and successfully learned the short-term and long-term dependencies in the data. The results of the hybrid models were superior to the single models, and the accuracy and stability were improved. The model was validated by checking residual plots and error distributions, and most errors were observed near zero. There were still some prediction challenges in extreme cases, while the Transformer-TCN model was the most accurate for most cases among the architectures analyzed. This work focuses on the prospect of deep learning models for reactive power forecasting and offers a useful tool for managing the power grid. Other research could be directed towards further optimizing hybrid architectures and other methods for enhancing predictive capability in power systems.