Treffer: OPTIMAL DEEP NEURAL NETWORK PARAMETERS FOR POWER LOSS MINIMIZATION ANALYTICS.

Power loss remains a critical problem in power systems, leading to voltage instability, reduced transmission efficiency, equipment degradation, and significant financial losses for utility providers. These inefficiencies disrupt reliable electricity delivery, increase operational costs, and hinder the sustainability goals of smart grid infrastructure. Traditional analytical and optimization methods have often proven inadequate in accurately modeling the nonlinear and dynamic behaviors that characterize power networks. As a result, there is a growing need for advanced intelligent models capable of analyzing large-scale data, uncovering patterns, and predicting losses more effectively. This study addresses the power loss challenge by developing an optimized Deep Neural Network (DNN) framework for power loss minimization analytics in smart grids. Three separate power loss datasets were analyzed using the Orange data mining platform and the Python development environment. The DNN model was systematically tuned by optimizing key parameters such as activation functions, number of hidden layers, learning rates, and batch sizes. Among the configurations tested, the model employing the Rectified Linear Unit (ReLU) activation function with six hidden layers achieved the best performance. The optimized DNN produced a Mean Squared Error (MSE) of 1.0E-03, Root Mean Squared Error (RMSE) of 3.4E-02, Mean Absolute Error (MAE) of 1.9E-02, coefficient of determination (R²) of 0.94, and Mean Absolute Percentage Error (MAPE) of 4.8%. When compared with conventional models, the optimized DNN demonstrated a 20-25% improvement in predictive accuracy. These results confirm that optimizing DNN parameters significantly enhances power loss analytics in smart grid systems. The proposed model offers a robust and intelligent solution for minimizing losses, improving energy efficiency, and supporting informed decision-making in modern power networks. [ABSTRACT FROM AUTHOR]

Copyright of Nigerian Journal of Technology is the property of University of Nigeria, Faculty of Engineering and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)