Akademik Veri Yönetim Sistemi
Necmettin Erbakan University Journal of Science and Engineering, cilt.7, sa.1, ss.31-47, 2025 (Hakemli Dergi)
The acceleration of industrialisation has resulted in a corresponding increase in the demand for energy supplies, driven by the growing use of new generation electronic equipment in residential settings. The advent of renewable energy, or green energy, has prompted a shift away from traditional methods of energy production, such as the use of natural gas and fossil fuels. However, this transition has given rise to a number of challenges. It is of great importance to monitor data in order to integrate the obtained electricity into the system and to monitor it. The acquisition of energy data on a large scale is made possible by the implementation of dynamic relay communication within micro and macro scale smart grids, which have been specifically designed for this purpose. Deep learning algorithms and machine learning methods are employed for the processing and analysis of data obtained in the context of the Internet of Things (IoT). The implementation of these methods enables smart grids to operate with reduced loss and enhanced efficiency. The estimation of energy consumption at the smallest scale facilitates the implementation of optimised energy management strategies. By ensuring the flow of electricity in the optimal amount (power), the potential for waste can be mitigated. The rapid and sophisticated responses of machine and deep learning algorithms facilitate more structured and sustainable energy management for both users and producers. In this study, four years' worth of electrical energy data from residential sources was analysed using techniques such as Convolutional Neural Network, Long Short-Term Memory, Random Forest and K-Nearest Neighbours Regression. The resulting analyses enabled the estimation of energy consumption. To assess the efficacy of learning algorithms in the study across varying training and test data ratios, the dataset was partitioned using three distinct division methods: hold-out (90% training - 10% testing), hold-out (80% training - 20% testing), and a 67% training - 33% testing split. Additionally, a 10-fold cross-validation approach was employed for further evaluation. Comparative analysis revealed that the LSTM model emerged as the top-performing model, boasting the lowest MSE value of 0.0054 for daily forecasts.