Accurate harvested energy prediction of the energy harvesting Internet-of-things (IoT) nodes is the basis of the proper power management and should be low-overhead. A new multi-algorithm fusion framework, which merges results of multiple prediction algorithms to achieve a higher accuracy, has been proposed. A three-algorithm fusion solar radiation predictor was implemented. The experiments using the real solar radiation data show that it improves the percentage prediction error by 10%-26% for different prediction intervals. Its complexity is low enough to run on the embedded systems in real-time.