Energy Prediction Models for Harvesting Enabled Sensor Networks: Performance Evaluation and Enhancements

Abstract

Energy-Harvesting is a technique which is playing a vital role to improve the lifetime of wireless sensor networks. Di erent prediction models have pro- posed so far to predict the solar or wind energy with maximum prediction accuracy and minimum prediction error rate. This dissertation is present- ing the proposed energy prediction model as an enhancement of Pro-Energy prediction model and then a fair comparison with existing statistical and stochastic energy prediction models (i.e. Pro-Energy and ASIM) over the short, medium and long-term prediction horizons. For the fair analysis, Pro- Energy short-term prediction model is extended to the long-term predictions and the ASIM model is extended to the short-term predictions. Results show that the overall performance of statistical models is better than stochastic prediction model. Quantitative performance analysis based on eight di erent datasets shows that the proposed model outperforms the candidate models in terms of prediction accuracy up to 74% and 55% for the short and medium- term prediction horizons respectively but in case of long-term prediction all models are performing relatively close to each other. On the other hand, ASIM model signi cantly consumes less execution time for energy prediction. Through simulations, it is demonstrated that the proposed energy prediction model results in better network lifetime.