Ya-Hui Sun, Xiaoqing Ye
Abstract Energy harvesting systems capture ambient energy and convert it into electrical power to sustain low-power electronic devices. However, they face challenges such as regular maintenance requirements and low energy conversion efficiency. Digital twin technology, which enables predictive analysis without direct physical intervention, offers a promising approach for real-time monitoring and optimization of harvesting system performance. Over recent decades, methods such as stochastic averaging and stochastic perturbation have been widely used to derive approximate analytical solutions for dynamical systems. These methods, however, are generally suited for offline theoretical analysis and do not readily support real-time forecasting or control. In contrast, few studies have explored the use of digital twins for real-time simulation and prediction of nonlinear dynamic responses in energy harvesting contexts. To address this gap, this paper develops a digital twin–based framework to simulate and predict the dynamic behavior of energy harvesting systems, with the aim of improving energy conversion efficiency and enabling predictive maintenance. Firstly, by integrating machine learning with Bayesian filtering algorithms within a data-driven and physics-based framework, a Bayesian grey-box model is established for simulating system states. The consistency between real and simulated values demonstrates the effectiveness of the proposed framework, allowing for real-time monitoring of the system. Subsequently, by refining the Gaussian process and incorporating a particle optimization algorithm, it is proven that digital twin can predict data within a certain period of time. Finally, two distinct systems are investigated to analyze their stochastic responses under Gaussian white noise and Gaussian colored noise excitations. Results demonstrate that the theoretical framework proposed in this paper consistently achieves high accuracy and precision across both noise conditions.