A transfer learning fusion model for semi-physical simulation and filed accuracy data

Abstract:
In order to solve the problem that the accuracy assessment of a single data source is subject to multiple constraints, data fusion reliability evaluation model based on transfer learning was established. Take the high-confidence but small sample field test data as the target domain, the semi-simulation test data as the source domain. Using the principal component analysis to extract the features separately, then embedding the feature distribution in glassman manifold space as two points, construct geodesic flow curve to attach them. So as to achieve data fusion, map the source and target domain to a common space through the geodesic flow curve. Accuracy evaluation can be calculated by constructing confidence region of semi-simulation data. The example verifies the feasibility and effectiveness of this model.