An energy efficient prediction based data collection approach using deep LSTM in wireless sensor network

Abstract

Wireless Sensor Networks (WSNs) are essential for real-time monitoring in environments, healthcare, and industrial settings, because they facilitate effective data gathering. A significant issue in WSNs is the high-energy usage owing to the transmission of redundant data, which shortens the lifespan of the network. Current research has not adequately resolved this problem, resulting in poor energy management and decreased data accuracy during aggregation. This study seeks to address these shortcomings by introducing a new energy-efficient data collection framework using a Deep Long short-term memory-based Prediction Model (DP-LSTM). The goal of this approach is to reduce redundant data transmissions while ensuring high data accuracy, thereby prolonging the operational life of WSNs. The hypothesis is that predictions based on deep learning can significantly reduce unnecessary transmissions while preserving data quality. Initially, Cluster Heads (CHs), chosen based on their proximity to the centroid and energy levels, and the DP-LSTM model was used to make predictions and communicate them to cluster nodes. Each node then compares its observed data with the predicted value and transmits data only if the difference exceeds a set threshold, thereby minimizing unnecessary transmissions. The proposed methodology was evaluated in MATLAB, and its performance was compared with that of existing algorithms, such as Extended Linear Regression (ELR), Hierarchical Least Mean Square (HLMS), and Temporal Data Prediction-based Aggregation (TDPA). The findings show that the proposed DP-LSTM model surpasses traditional methods in terms of energy efficiency, prediction accuracy, positive prediction, and transmission overhead.The proposed method achieved energy consumption reductions of 71.54%, 53.90%, and 27.15% compared to ELR, HLMS, and TDPA, respectively.