MITIGATION OF LOSSES IN POWER SYSTEM THROUGH LOAD FORECASTING USING DEEP NEURAL NETWORK ARCHITECTURE

Subject Area: POWER

This study presents a Deep Neural Network (DNN) model for short-term load forecasting for mitigation of losses in power system. The model developed in this study was trained using historical power consumption data from the Emene Injection Substation of the Enugu Electricity Distribution Company (EEDC) which underwent pre-processing phase that comprised of cleaning, normalization and splitting into training (80%), testing (10%) and validation (10%) sets. Various DNN architectures were evaluated through different numbers of hidden layers and activation functions to determine the optimal model. The model training process employed the Adam optimizer and Mean Squared Error (MSE) loss function, with dropout regularization to prevent overfitting. Implementation was carried out on Google Colab using the cloud-based GPU acceleration for computational execution. The results of the implementation demonstrated that the 5-layer DNN model achieved superior performance by yielding an MSE of 0.0021 and an R² score of 0.92. The study identified that deep learning-based forecasting models can significantly enhance power grid management by improving loss mitigation through load forecasting.