Prediction of Residential Electricity Consumption Using Long Short-Term Memory Based Framework

Antara Mahanta Barua

Abstract: Accurate prediction of electricity consumption is vital for the planning and operation of electric utility companies to ensure a balanced supply and demand of power. This study presents a Long Short-Term Memory (LSTM) based neural network model for predicting residential electricity consumption using smart meter data. The dataset consists of three months of electricity usage records collected from 300 households. The proposed model leverages the ability of LSTM networks to capture temporal dependencies and nonlinear patterns inherent in time-series data. To evaluate performance, different activation functions, such as Sigmoid and Hyperbolic Tangent (tanh), were implemented and compared. Model accuracy was assessed using standard evaluation metrics: Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). The results demonstrate that the LSTM-based approach provides reliable and accurate predictions of residential electricity consumption, highlighting its suitability for short-term load forecasting and smart grid applications.

Keywords: Electricity Consumption Prediction, Long Short-Term Memory, Smart Meter Data, Forecast Accuracy Metrics