The increasing integration of renewable energy sources and smart appliances in IoT underscores the importance of accurate short-term load forecasting. This paper presents an innovative Attention-based CNN-BiLSTM model to address the non-linear nature, randomness, volatility, and limited size of residential electricity consumption data. By incorporating data augmentation techniques, the model demonstrates improved robustness and performance on diverse datasets. Tested on IHEPC and AMPds datasets, our model shows significant accuracy improvements, with reductions in MSE to 28% and 34%, MAE to 17% and 24%, and RMSE to 16.5% and 17%, respectively. These results highlight the potential of our approach in enhancing energy management strategies, especially for small, sparse, or complex datasets driven by IoT devices.
Enhancing Short-Term Load Forecasting in Internet of Things: A Hybrid Attention-based CNN-BiLSTM with Data Augmentation Approach
07.10.2024
3007986 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch