BiLSTM Model for Electricity Consumption Prediction: A Comparative Study with Traditional Machine Learning Approaches

The reliability of power systems depends on accurate electricity load forecasting, as the demand for residential electricity rises annually as a result of infrastructure expansion and population expansion. While some traditional models based on machine learning have shown improvement in many areas, their accuracy in predicting energy consumption is still not up to par. Residential electricity consumption patterns are irregular and uneven, making precise power load assessment more difficult. U sing methods such as gated recurrent unit (GRU), convolutional neural network (CNN), eXtreme Gradient Boosting (XGB) regressor, bidirectional long short-term memory (BiLSTM), and Convolutional LSTM (ConvLSTM), the research focuses on building and assessing energy consumption prediction models. Four performance metrics namely, mean absolute error, root mean square error, mean square error and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{R}^{2}$</tex> are considered to check the potency of different machine learning algorithms but after the data preprocessing step i.e. filling missing values and removing outliers in the dataset. In this research, among all short-term prediction algorithms, BiLSTM performance surpassed good results in energy consumption prediction with values 0.0281, 0.0163, 0.0007 and 0.9197 of RMSE, MAE, MSE and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{R}^{2}$</tex> respectively.