Joint estimation method of state and parameter of digital twin model based on adaptive ensemble kalman particle filter

In simulation-based digital twin (DT) models, the physical models suffer from issues such as modeling errors, solution errors, model mismatch, and the inability to directly obtain necessary model parameters, leading to insufficient computational accuracy of the DT model and the accumulation of long-term calculation errors. In response to the aforementioned issues, this study proposes the noise adaptive ensemble kalman particle filter (AEnKPF). The integration of simulation and measurement data facilitates the estimation of both measurable and unmeasurable parameters within the physical model. State estimation has the capacity to denoise measurement data. Furthermore, parameter estimation can be employed to enhance the simulation accuracy of the model. In addition, based on DT's background, this study introduces a surrogate model into the joint estimation framework to enhance its real-time performance. The study takes the system-level simulation program in the nuclear energy field as the object, uses the core fluid temperature as the state estimation value, and the heat transfer coefficient as the parameter estimation value to verify the feasibility and effectiveness of the joint estimation framework. The results indicate that the MAPE between the simulated core fluid temperature values and the measured values decreased from 29.51% to 0.98%.