An Effective Improved Dropout-Boosted Convolutional Autoencoder for Energy Consumption Prediction in Electrical Construction Industry

To reduce prediction error, energy data from several sources, such as distributed energy supplies, requires powerful data predictive frameworks. An efficient deep learning (DL) model for smart building energy use prediction is proposed by this framework. Start with min-max normalising the public dataset's raw data to remove missing values to reduce overfitting. The improved dropout-assisted convolutional autoencoder (IDCAE) technique predicts the electrical construction industry's energy usage last. Python is used for experiments and the Individual Household Electric Power Consumption Data Set for training. We also used UK and Canadian datasets to test the framework's generalis ability to underfitting and overfitting. Real user smart boxes were utilised to test the system with these datasets. The framework's resilience to varied building types, locations, weather, and load distributions is shown by an average MSE of 0.09 across all datasets, suggesting it generalises well to changing data and limits.