Machine learning based optimal renewable energy allocation in sustained wireless sensor networks

Abstract

The environmental energy harvesting is adjudged as a reliable solution to power the wireless nodes for infinite time and assuring uninterrupted operation of deployed network nodes. But uncertain energy availability initiates an important research issue of energy management in rechargeable sensor nodes. An integrated approach of energy assignment principles with adaptive duty cycling has been proposed to efficiently utilize the available energy and to maximize the node performance. The R interface based machine learning ensemble approach has been used for solar irradiance prediction to pre-estimate the node duty cycle. Dynamic programming based optimization problem has been used for real time adaption of pre-computed node duty cycle. The effectiveness of proposed work has been validated using MATLAB interface by extensive simulations on real time solar energy profiles in terms of magnitude and stability of sensors average duty cycle. The proposed algorithm achieves an average duty cycle of 65% to 69% with a limit of 70% maximum duty cycle irrespective of irregular radiation patterns throughout the day as well as for different forecasting horizons. The results shows minimum variation in estimated and real time energy profiles in stable weather conditions and optimize the duty cycle in irregular weather conditions. The results also shows minimum variation ( $$>2\%$$ ) in estimated and real time energy profiles in stable weather conditions and optimize the duty cycle in irregular weather conditions.