ENHANCED SHORT TERM LOAD FORECASTING IN SMART GRIDS USING A DEEP LEARNING MODEL WITH CNN-GRU

Abstract

Short Term load forecasting (STLF) helps in making necessary adjustments in the supply of electrical power against varying demands. It is therefore important in recent times for economic efficiency and operational reliability in the power systems. Traditional models are normally unable of effectively capture the complex temporal and spatial dependencies in the time-series data. This paper introduces a hybrid model including the features of Convolutional Neural Network (CNN) and Gated Recurrent Unit to promptly model the temporal dependencies and thus helps in performance improvement of load forecasting along with achieving lower values of Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and MAPE (Mean Absolute Percentage Error). This paper has successfully shown that the combination of CNN and GRU architecture is a better model for Short-Term Load Forecasting. The proposed model has a good precision, and large extensibility range. This model can be used to describe both the immediate spatial influence from the environment and the long-term temporal effects of human behavior, which makes it a helpful, reliable tool for the modern smart grid. This study's results provide an essential link between theoretical studies in artificial intelligence and the real-world requirements of the energy industry worldwide.  When evaluated against traditional machine learning networks available, the proposed model achieved high absolute accuracy with exceptionally low values of MAE and RMSE.  However, the MAPE percentage remained high because it was caused by near zero demand value during off-peak hours. The exceptionally low RMSE value of 0.5 shows that the model has successfully eliminated the dangerous forecast errors which comes costly during peak-load hours, making it an exceptionally reliable framework for real world grid scheduling.