Digital twin (DT) has emerged as a powerful tool for optimizing design, control, and maintenance in power electronic systems. This paper proposes a machine learning-based DT framework for the non-invasive parameter identification of converter systems. A mathematical DT model that can operate independently is first established. Subsequently, the DT model serves as a virtual representation of the physical system, generating operational data covering diverse conditions. The data is then used to train an artificial neural network (ANN) tool for parameter estimation. Unlike traditional metaheuristic optimization based on offline iterative optimization, the ANN can directly estimate parameters from available measurements offline and in real time. Eventually, the obtained parameters update the DT model, improving its accuracy. The effectiveness of the framework and accuracy of the DT model are validated via simulations and experiments.