Akademik Veri Yönetim Sistemi
Journal of the Faculty of Engineering and Architecture of Gazi University, cilt.40, sa.3, ss.1637-1646, 2025 (SCI-Expanded, Scopus, TRDizin)
In this study, a novel method based on Convolutional Neural Network (CNN) deep learning has been proposed for identifying residential loads, and the feasibility and performance of the method have been investigated. Initially, six different loads commonly found in residences are identified, and the load datas are collected individually and in various combinations. These data are utilized to develop machine learning models including the proposed model, as well as Random Forest, Decision Trees, and K-Nearest Neighbor models, which are frequently employed for similar purposes in the literature. Subsequently, the identification results of residential loads for all models are compared using performance metrics, determining that the CNN model proposed in this study achieved the most successful identifications. Figure A illustrates the flowchart of the proposed CNN-based load identification approach. Figure A. Proposed CNN deep learning model flow chart Purpose: Responding to the rapidly changing energy needs of the modern era has made the management of residential energy increasingly critical. This study aims to develop a new Convolutional Neural Network (CNN) based deep learning model to achieve more successful identification of residential loads. The objective is to demonstrate that the proposed deep learning-based model outperforms traditional machine learning models and showcases its applicability. Theory and Methods: With the proliferation of smart home technologies, the efficient identification and management of residential loads are gaining prominence. Within this framework, a novel CNN-based deep learning model is introduced for residential load identification. The study incorporates data measurements from six distinct loads conducted at the laboratories of Afyon Kocatepe University. The efficacy and suitability of the proposed model are evaluated through comparisons with various machine learning methodologies including Decision Tree Regression, Random Forest Regression, and K-Nearest Neighbour. Results: The identification performance of the proposed CNN model has been compared with other machine learning methods using four different statistical evaluation criteria (accuracy, precision, recall, F-score). The results indicate that the CNN model outperforms other comparative models in terms of all statistical evaluation criteria, establishing itself as the most successful predictive model. Conclusion: In this study, a novel deep learning model based on CNN for the identification of residential loads has been proposed and its performance and applicability have been investigated. The results demonstrate that the proposed CNN-based deep learning model outperforms machine learning algorithms in all accuracy, precision, recall, and F-score statistical evaluation criteria, indicating its effectiveness in the identification of residential loads.