A Real-Time Charge Predictive Model for Intelligent Networks

A smart grid is a modern electrical system that uses information technology, including sensors, measurement tools and communication devices, to monitor and improve the efficiency of the power system. However, real-time forecasting remains a challenge due to its complexity. This paper presents a forecasting framework that combines Convolutional Neural Networks (CNN) and Bidirectional Long-Short Term Memory (BiLSTM) for real-time load forecasting in smart grids. Compared to traditional methods like ARMA and Decision Trees (DTs), the proposed CNN-BiLSTM model demonstrates superior performance in terms of prediction accuracy, reaching up to 99% - higher than Long-Short Term Memory (LSTM) (93%) and Support Vector Machine (SVM) (84%). Additionally, the CNN-BiLSTM model requires fewer computational resources, with 90 Gigaflops (G) and 94 Million (M) parameters, compared to 151 (G) and 120 (G) for ARIMA and CNN-LSTM, respectively. These results indicate the proposed model's ability to accurately predict power system loads in real time with high computational efficiency.