Prediction of Energy in Solar Powered Wireless Sensors Using Artificial Neural Network

Predicting of energy income into solar harvesters of a wireless sensor units is a necessary task to manage its operation. Thus, utilizing an accurate prediction algorithm leads finally to prolong the lifetime of that sensors, which is strongly required. The most common prediction algorithm, Exponentially Weighted Moving Average (EWMA), can be used only under consistent sequential days. This is attributed to the large prediction error that appears when weather changes unexpectedly from sunny to cloudy or vice versa. The other moving average algorithm, Weather Conditioned Moving Average (WCMA), still shows a considerable error at sunrise and sunset periods. Pro-Energy produces also a significant error. This paper presents Artificial Neural Network (ANN) as new prediction algorithm to reduce the prediction error caused. Additionally, it compares the results of applying this algorithm with the aforementioned conventional methods. For two weeks simulation periods, one in summer and another in winter, ANN shows respectively an average prediction error of (0.03% and 0.02%); while (EWMA, WCMA and Pro-Energy) show higher errors in overall. Thus, ANN is an accurate algorithm to be adopted for energy prediction in wireless sensors.