Deep Transfer Learning-Enabled Energy Management Strategy for Smart Home Sensor Networks

The applications of wireless sensor networks are extensively used to detect and control home residents' activities in smart homes. However, the sensors are battery-powered, so keeping them in active mode consumes tremendous energy. In this regard, we propose a solution to activate the smart home sensors based on detecting the upcoming activities using a Deep Long-Short Term Memory (DLSTM) model. The pre-trained model is then transferred to the same and different Target Domains (TDs) to reduce the time for training. The proposed system applies to preprocess and feature mapping steps to both the source and target data to make grounds for efficient transfer. Further, applying the trained model to the TD may miss the essential activities. Therefore, a reinforcement learning model is applied in the TD. To handle unusual activities in real-time, guard sensors are appointed among the idle sensors. The performance evaluation shows that the proposed scheme detects the activities with an accuracy of 96.1%. Additionally, the proposed scheme outperforms the sentry and prediction-based schemes in energy consumption of the sensors and network lifetime.