**作者：**PAN-CHIO TUAN AND MING-CHING JU**中文摘要：**This work presents a novel on-line inverse methodology, the adaptive input estimation algorithm, that efficiently estimates time-varying boundary unknowns in a thermal system. The algorithm includes use of the Kalman filter to derive a regression equation between the biased residual innovation and thermal unknown. Based on this regression model, the Recursive Least Squares Estimator (RLSE) weighted by the adaptive forgetting factor is proposed to robustly extract unknowns. The robust weighting technique is essential in RLSE since the estimation process encounters, in the course of system measurement, noise, process uncertainties and unpredictable changes in the status of time-varying unknowns. In this paper, the maximum likelihood type estimator (M-estimator) combined with the Huber psi-function is used to construct the weighting forgetting factor as a reciprocal function of biased innovation at each time step, thereby allowing us to estimate an unknown in a system with global uncertainties. The superior capabilities of the proposed algorithm are demonstrated in three time-varying estimation cases using the 1-D cylindrical inverse heat conduction model. In addition, the efficiency of the proposed weighting technique is validated through the bias and variance test. Finally, the proposed algorithm is shown to be an improvement over the conventional input estimation approach and to be appropriate for implementation.**英文摘要：**--**中文關鍵字：**adaptive, robust, on-line input estimation, M-estimator**英文關鍵字：**--