Household electricity consumption prediction and appliance aging analysis based on VMD-BiLSTM model

With the rapid development of smart home and Internet of Things technologies, household appliance energy consumption monitoring and aging analysis have become important research areas for achieving smart energy management. Household electricity consumption data exhibits significant periodicity, nonlinearity, and high-frequency fluctuations. Traditional time-domain modeling methods struggle to effectively capture its complex signal structure. Furthermore, when data is limited or information is incomplete, the models are prone to overfitting or under-learning. To address the complexity of household appliance aging characteristics, this paper proposes a joint modeling approach based on improved deep learning—the variational mode decomposition combined with bidirectional long short-term memory (VMD-BiLSTM) model—to efficiently extract and model multi-scale non-stationary aging signals. This approach enhances feature extraction capabilities through a deep residual structure and reduces noise interference using VMD decomposition, thereby enhancing the model's prediction accuracy and robustness for appliance aging. Systematic experiments demonstrate that the improved model exhibits strong adaptability and stability in complex environments, significantly outperforming traditional single-deep network models.