Analysis and comparison of household energy consumption based on deep learning

Abstract

Effective prediction of energy consumption can greatly assist in the planning and allocation of energy resources and thus contribute to energy conservation efforts. To effectively predict energy consumption, this paper applies deep learning techniques to construct analytical models. Convolutional neural networks (CNN), long and short-term memory (LSTM) and recurrent neural networks (RNN) are simultaneously constructed as evaluation models for analysis. Each model was trained and tested on a dataset representative of the energy consumption of the device or system. The training and validation loss curves of the models were observed over a series of epochs to assess their learning ability. The results showed that the LSTM models outperformed the CNN and RNN models in all performance metrics. The superior learning ability is demonstrated due to the LSTM exhibiting a substantial learning curve and stable validation loss. Findings derived from empirical studies reveal that LSTM models exhibit proficiency in predicting energy usage in a reliable manner. Such an insight is instrumental in progressing the design and evolution of devices and systems that demonstrate improved energy efficiency. Looking ahead, there is scope to further refine hybrid LSTM models. This not only offers the potential for heightened prediction accuracy, but also extends significant contributions to the discipline of energy conservation and efficiency.