CBLSTM-AE: A Hybrid Deep Learning Framework for Predicting Energy Consumption

Multisource energy data, including from distributed resources and its multivariate nature, necessitate the integration of robust data predictive frameworks to minimise prediction error. This work presents a hybrid deep learning framework accurately predict consumption different building types, both commercial domestic, spanning countries, Canada UK. Specifically, we propose architectures comprising convolutional neural network (CNN), an autoencoder (AE) with bidirectional long short-term memory (LSTM), LSTM BLSTM). The CNN layer extracts important features dataset AE-BLSTM layers are used for prediction. We use individual household electric power University California, Irvine compare skillfulness proposed state-of-the-art frameworks. Results show performance improvement in computation time 56% 75.2%, mean squared error (MSE) 80% 98.7% comparison BLSTM-based (EECP-CBL) vanilla LSTM, respectively. In addition, various datasets UK further validate generalisation ability underfitting overfitting, which was tested on real consumers’ smart boxes. results that generalises well varying constraints, giving average MSE ?0.09 across all datasets, demonstrating robustness locations, weather, load distributions.