Prediction of Power Loss in Grid Using Neural Network

Abstract

This Paper proposes a data-driven  approach for grid loss prediction in power systems. It  utilizes a comprehensive dataset with relevant features  such as grid load, temperature forecasts, and calendar data.  The dataset is pre-processed by handling missing values,  normalizing features, and encoding cyclic calendar  features. A Long Short-Term Memory (LSTM) recurrent  neural network is employed for the prediction model,  capturing temporal dependencies and generating forecasts  of grid loss two hours ahead. The model is trained using  mean absolute error (MAE) as the loss function and  optimized through hyperparameter tuning. Evaluation  metrics like MAE and root mean squared error (RMSE)  assess the model's accuracy. Visualization techniques  compare predicted and actual grid loss values. The Paper  concludes with analysis, future research suggestions, and  highlights the potential of the Prophet data-driven  approach for efficient and reliable power distribution.