Research on Continuous Variable Reasoning and Completion Method for Digital Twin Based on Improved Dynamic Discrete Bayesian Network

In the new generation of industrial intelligent systems, digital twin technology is a core method for simulating, predicting, and optimizing physical systems, relying on high- quality, continuous data streams to enable real-time mapping of physical entities. However, due to limitations such as insufficient device resources and communication failures, systems frequently face issues such as erroneous, missing, or delayed data, which undermine data integrity. Given the complexity of digital twin data, predominantly composed of continuous variables, the Dynamic Discrete Bayesian Network (DD-BN) has been introduced to infer and complete missing data. Yet, as data dimensionality increases, DD-BN encounters challenges like the curse of dimensionality and data sparsity, making it difficult to maintain real-time performance and accuracy. To address these issues, this paper proposes a High-Precision Data Inference and Completion Method for Continuous Variables based on an Improved Dynamic Discrete Bayesian Network (HP-DAD-BN). This approach combines Independent Component Analysis (ICA) for dimensionality reduction, spline function fitting, and Monte Carlo sampling to optimize the inference process. Simulation results show that this method significantly enhances inference accuracy, data completion performance, and real-time capabilities across both low- and high-dimensional data, demonstrating superior overall performance.