Abstract: With the wide access of novel load and distributed generations (DGs), the harmonic problem in power system is becoming more and more prominent. Harmonic monitoring at grid connection points is the premise of responsibility division, source tracing and suppression of harmonic pollution. Aiming at the problem that harmonic signal is difficult to monitor due to its strong randomness and subtle characteristics, this paper proposes a dynamic harmonic monitoring method in power system based on sparse dictionary atoms sharing. First, the characteristics of harmonic in power system are analyzed, and a harmonic dynamic monitoring architecture based on sharing and multiplexing the over completed dictionary atoms of compressed sensing is proposed to realize continuous dynamic sampling of power grid operation data. Then, a residual energy-based sparsity adaptive matching pursuit (REB-SAMP) algorithm is proposed, which characterizes the degree of sparse decomposition of the original data by calculating the residual energy after each iteration. Based on this residual energy, an iterative termination criterion and a variable step-size strategy are formulated for the algorithm. In addition, by cascading Gabor over-complete sparse dictionary and Fourier sparse dictionary, the super-complete synthetic dictionary is constructed, which improves the sparse representation performance for harmonic monitoring data. Finally, a distributive generation system is built based on PSCAD/EMTDC simulation platform to verify the rationality and effectiveness of the proposed algorithm. The simulation shows that the proposed algorithm is easier to detect the harmonics at the points of common coupling (PCC), and has the advantages of higher reconstruction accuracy, better convergence and stronger anti-noise performance.