The broadband properties of impedance in power line communication (PLC) networks would enable smart PLC systems to implement fault detection and channel health monitoring in smart grid. In this paper, a novel real-time impedance estimation method for power line communication networks is proposed. We study the relationship between the channel frequency response (CFR) and the impedance behavior, CFR can be treated as the already known variable which is normally calculated by channel estimation algorithms in PLC devices, and find that the variation of certain key factor of the CFR curves including the frequency characteristics and values of Peak-Valley Difference and so on, could be used to tracking the impedance. A VMD (Variational Mode Decomposition) algorithm is used to obtain the useful frequency properties information in CFR waveform and machine learning method is used to estimate the parameters. Proposed impedance estimation method is verified by the computer simulations and the results show that it is a feasible solution for impedance estimation without any additional hardware and costs.