The invention discloses a prediction method for a fluctuation coefficient of power grid impedance identification errors of a photovoltaic grid-connected point. The method comprises the steps that (1) a time sequence of an evolution system for the fluctuation coefficient of the power grid impedance identification errors of the photovoltaic grid-connected point is established according to parameters obtained through real-time monitoring; (2) measured data of the photovoltaic grid-connected point is subjected to dynamic cluster partitioning according to the time sequence of the evolution system; and (3) a neural network prediction model for the fluctuation coefficient of the power grid impedance identification errors of the photovoltaic grid-connected point is calculated according to an optimization processing result of a genetic neural network. Through the method, the technical problem that assessment accuracy and photovoltaic utilization efficiency are not high is solved; the method has the advantages of improving the assessment accuracy of a photovoltaic inverter, avoiding problems concerning power quality caused by connection of a photovoltaic power station into a power distribution network system, increasing the photovoltaic utilization rate, improving the reliability of the power distribution network electric system and improving the economical efficiency of the power distribution network electric system in distributed photovoltaics.