Improvement of Nonembedded EMC Uncertainty Analysis Methods Based on Data Fusion Technique

The nonembedded uncertainty analysis method is one of the popular research topics in the field of electromagnetic compatibility. The simulation theory system built around it has been initially completed. The essence of the nonembedded uncertainty analysis method is to construct a surrogate model, like a "black-box", to accurately describe the deterministic electromagnetic compatibility simulation process. Therefore, the key lies in how to train an accurate surrogate model. However, no matter how the existing nonembedded uncertainty analysis methods are improved, there is no escape from the fact that the more deterministic simulations that are performed, the more accurate the uncertainty analysis results are. When a single electromagnetic compatibility simulation is computationally costly (high-frequency problems and finite element numerical modeling), the number of deterministic simulations used is limited (high-precision simulation data has limited availability), so the accuracy of the uncertainty analysis method cannot be intrinsically improved, which is a bottleneck problem that is difficult to break through. In this article, an improved nonembedded uncertainty analysis method based on data fusion is proposed. It requires large amounts of low precision simulation data through low time cost solvers such as approximate formula method. Applying machine learning to introduce the useful information from the low-precision simulation data into the high-precision simulation data results in constructing a more accurate surrogate model without changing the cost of the simulation time, to achieve the purpose of essentially improving the accuracy of the nonembedded uncertainty analysis method.