Abstract: The online monitoring has been a main challenge for the development of advanced micro reactors due to the constrained number of ex-core detectors that can be deployed. A method for core power reconstruction and control rod position prediction was proposed in this study to tackle this problem. It was based on the principal component analysis (PCA) of the core power distributions at various state conditions, with the ex-core detector response functions (DRFs) mapping the core power distribution to the ex-core detector counting. The application of the method on a gas-cooled micro reactor was reported in this paper. The nodal-level power distribution database covering different operating conditions has been constructed using the nuclear design code, followed by the PCA that gives the principal components (PCs). The reduced order was achieved by transforming the determination of the nodal power to determination of the coefficients of the PCs. It is confirmed that PCs of a number less than the number of ex-core detectors can produce the nodal power recovery accuracy of within 2.5%, as long as the power distribution database covers enough range of the core operation conditions. On the other hand, the DRFs show strong spatial dependence, which implies tight coupling of the core power and the ex-core detector counting. Specifically, the axial DRFs are significantly affected by the thickness of the front and rear reflector, and the radial DRFs confirms that no fuel assembly contributes a negligible fraction. Overall, the nodal-level core power can be reconstructed from the ex-core detector signals, with reconstruction errors within 2.9%. The axial pellet-level power distribution of a fuel assembly can be reconstructed through polynomial fitting with the axial nodal-level power distribution, which gives the accurate peak power factor of error within 0.2%. Furthermore, the prediction of the control rod position can be achieved by constructing a function approximately mapping the coefficients of PCs to the rod position, with the prediction error of within 0.5 cm. This online monitoring method has high accuracy for power reconstruction and control rod position prediction, and is applicable to various advanced micro reactors at various operating conditions.