Spacecraft assembly requires high precision and safety, and the error and precision of the six-dimensional force sensor is a key factor. Subsequently, this paper, in order to specifically explore the specific process of dynamic calibration, through the design of the experimental device, calibration experiments, through the multi-dimensional recursive least squares algorithm and the improved Bitterling Fish Optimized BP neural network fitting algorithm to decouple the experimental data and the model, according to the experimental validation and data comparison and analysis, to determine the calibration algorithms applicable to the field of aerospace flexible assembly, and reduce the error of the decoupling process of the sensor calibration to the0.6% or less, which meets the error requirements in the aerospace field.