Noise Unknown EM Based Transfer Identification for Dynamic Systems With Limited Target Data

This letter develops a noise unknown expectation- maximization based transfer identification method for dynamic systems with limited target data. Different from existing transfer identification methods that usually require known or pre-calibrated noise variances, the proposed method treats the inter system parameter discrepancy as a latent variable and jointly estimates the target parameters and target noise variance within a unified probabilistic framework. In the E step, the posterior distribution of the parameter discrepancy is inferred from the source data and the current target parameter estimate. In the M step, closed form updates are derived for the target parameters and the unknown noise variance. The resulting algorithm adaptively balances source and target information according to the estimated uncertainty. Simulations on a mass spring damper system demonstrate that the proposed method improves identification accuracy under limited data and unknown noise conditions, and remains competitive with variance known transfer methods.