Time-series based prediction for energy consumption of smart home data using hybrid convolution-recurrent neural network

The rapid increase in technological development has led to the rise in usage of IoT devices for monitoring Electrical Energy Consumption. As countries around the world are committing to United Nations Sustainable Development Goals, reducing carbon footprint has become an eminent priority for policymakers, businesses, and the public. Clean and green energy in the form of electricity has emerged as an alternative to fossil fuel. Since electricity is scarce and in high demand it has become an important problem for identifying robust energy consumption predictive models for powered smart residential homes. In our research we compare SVR, LSTM, GRU, CNN-LSTM, CNN-GRU models for predictive energy consumption data of smart residential homes. Empirical results indicate that with increase in the amount of data the performance of machine learning SVR degraded significantly more as compared to Deep Learning Techniques, which provides conclusive evidence that machine learning techniques are not suitable for the task. Whereas, our proposed CNN-GRU architecture performs 17.4% better in terms of Mean Absolute Error (MAE) with a value of 0.151 compared to the LSTM which has a value of MAE equals to 0.183 for days granularity of data and is only bested by the LSTM by 0.4% in terms of MAE for hour granularity data, where the CNN-GRU has MAE of 0.229 and LSTM achieves the MAE of 0.228. Additionally, CNN-LSTM and LSTM architectures were found effective in identifying outliers.