DSP: A Deep Learning Based Approach To Extend the Lifetime of Wireless Sensor Networks

Session: Poster Session

Jack Press, Wayne State University, [email protected]
Suzan Arslanturk, Wayne State University, [email protected]

Abstract

Wireless Sensor Networks (WSNs) equipped with batteries and solar panels enabled applications in various areas such as environmental monitoring, agricultural, military, and medical systems. Research has shown that batteries often fail earlier than their projected lifetime due to external parameters affecting battery life. Sensor-nodes with solar panels placed in areas with sufficient sunlight can have their batteries recharged and can stay online for longer periods. However, sensor-nodes placed in areas with insufficient sunlight may need to adjust how often they send data in order to stay online for longer periods. In this study, we present a Dynamic Sleep Protocol (DSP) to forecast the lifetime of a sensor-node by dynamically adjusting the sleep period between transmissions. We have used a deep recurrent neural network with Long Short Term Memory (LSTM) units to forecast the lifetime of the batteries and have discussed potential optimization functions to adjust the sleep period. Our results have shown that accurate identification of the battery life with accurate adjustments help us obtain longer operating hours without sacrificing the system performance.