Approach for the evaluation of indirect measurement uncertainty based on neural networks

Abstract

A new approach, in which a nonparametric measurement model is built on the radial basis functional neural networks to evaluate the indirect measurement uncertainty, is proposed in this paper to solve the difficult problem of evaluating the uncertainty of indirect measurement with no measurement model. By determining the center of basis functions based on the clustering result of training samples, neural networks can still be secured a high model building accuracy even when there are relatively fewer training samples. By using the measurement model built to approximately compute the sensitivity coefficient that reflects the uncertainty propagating law of each influence quantity, it is possible to evaluate the result of indirect measurement and its uncertainties. As is demonstrated in simulation results, the method of indirect measurement model building based on neural networks requires no prior knowledge of the measuring process, enjoys a relatively higher modeling accuracy, effectively secures the high accuracy in evaluating the indirect measurement uncertainty and can serve as a beneficial complement to Guide to the Expression of Uncertainty in Measurement [1] .