Imaging time-series with features to enable visual recognition of regional energy consumption by bio-inspired optimization of deep learning

Abstract

To increase the efficiency of energy use, ensure the stability of the power supply, and achieve balance in the energy supply, power management units have proposed plans that integrate energy-saving with intelligent systems, in which smart grids are used to distribute power and to manage power consumption. Imagery deep learning technology is proposed to address the knowledge gap, and highly accurate energy consumption predictions can be made by converting the 1-D time-series and features to 2-D images for visual recognition. Models based on machine learning and convolutional neural networks (CNNs) were used to predict future power consumption. Performance indicators were evaluated to determine the prediction accuracy and identify the best model for predicting power consumption. A metaheuristic—Jellyfish Search (JS)—is incorporated into the best model to optimize its hyperparameters to ensure model accuracy and stability. After the hybrid JS-CNNs model was constructed, validation was carried out. The analytical results provide insights into the formulation of energy policy for management units and can help power supply agencies to distribute regional power in a way that minimizes unnecessary energy loss. This study contributes to the prediction of future energy consumption trends, reveals power consumption patterns in cities and counties across a nation.