Neural Network Based Short Term Forecasting Engine to Optimize Energy and Big Data Storage Resources of Wireless Sensor Networks

Abstract

Energy efficient wireless networks is the primary research goal for evolving billion device applications like IoT, smart grids and CPS. Monitoring of multiple physical events using sensors and data collection at central gateways is the general architecture followed by most commercial, residential and test bed implementations. Most of the events monitored at regular intervals are largely redundant/minor variations leading to large wastage of data storage resources in Big data servers and communication energy at relay and sensor nodes. In this paper a novel architecture of Neural Network (NN) based day ahead steady state forecasting engine is implemented at the gateway using historical database. Gateway generates an optimal transmit schedules based on NN outputs thereby reducing the redundant sensor data when there is minor variations in the respective predicted sensor estimates. It is observed that NN based load forecasting for power monitoring system predicts load with less than 3% Mean Absolute Percentage Error (MAPE). Gateway forward transmit schedules to all power sensing nodes day ahead to reduce sensor and relay nodes communication energy. Matlab based simulation for evaluating the benefits of proposed model for extending the wireless network life time is developed and confirmed with an emulation scenario of our testbed. Network life time is improved by 43% from the observed results using proposed model.