Deep Recurrent Neural Network for Time Series Modelling of Energy in Wireless Sensor Network Nodes
Zoren P. Mabunga, Jennifer C. Dela Cruz · 2024
This study addresses the challenge of predicting energy consumption in wireless sensor network (WSN) nodes using advanced AI techniques, specifically deep recurrent neural networks (DRNN). By collecting datasets from both static and dynamic sensor nodes, we aim to reflect diverse real-world energy consumption patterns. We developed and trained LSTM-based models, leveraging their proficiency in processing time-series data and learning temporal dependencies. The results demonstrated that the LSTM model significantly outperformed other models, including stacked LSTM, Prophet, and exponential smoothing. For static nodes, the LSTM model achieved an RMSE of 213.39 joules, MAE of 491.75 joules, and MAPE of 2.53%, while for dynamic nodes, it achieved an RMSE of 293.26 joules, MAE of 212.76 joules, and MAPE of 1.46%. In comparison, the stacked LSTM, Prophet, and exponential smoothing models showed higher error rates. The LSTM model's superior performance is attributed to its ability to balance complexity and generalization, making it well-suited for the relatively simple dataset used in this research.