In the process of frequency adjustment in photovoltaic power grids, relying on conventional Smith predictive controllers to achieve grid frequency adjustment control has a strong dependence on the accuracy of the model, and the maximum rate of change of frequency (RoCoF) after the implementation of the control strategy is large. Therefore, a stepwise inertial control method for photovoltaic grid frequency adjustment based on an improved Smith predictor and big data is proposed. First, the historical meteorological data and photovoltaic power grid operation data are collected, and the density peak clustering algorithm in the field of big data analysis is applied for partition processing, and then the data of similar days are screened into the long and short term memory network to predict the power change of photovoltaic power generation in the future. Then, based on the idea of stepwise inertia control, the frequency adjustment control strategy of power grid is designed, including short-time overdrive, speed recovery and other stages. The fuzzy adaptive proportion-integration-differentiation (PID) controller is integrated into the conventional Smith predictor. Thus, the optimized Smith predictor is upgraded. Finally, under the condition of not being affected by the change of the controlled model, the stepwise inertial frequency adjustment control is completed according to the predictive compensation principle, and the sparrow search algorithm is used to solve the optimal control parameters. The experimental results show that after the implementation of the control method, the maximum RoCoF during the operation of the photovoltaic power grid is only 0.086 Hz/s, which effectively reduces the dependence on the accuracy of the model and ensures the stable operation of the power system.