In recent years, in photovoltaic (PV) power forecasting research, there are certain limitations in single forecasting methods. In traditional combined forecasting methods, such as the average weight combined method and the fixed weight combined method, the determination of the weight value cannot guarantee that the forecasting error at each moment is the smallest. In order to reduce the PV power forecasting error, this paper proposes a short‐term PV power dynamic weighted combination forecasting based on the least squares (LS) method. First, the random components of the PV power are extracted using the periodogram method, and then the dynamic weight value of each method is determined with the LS method. The combined model forecasts the random components of PV power and superimposes them with the periodic component to obtain the final PV power forecast. Using the data from a PV power plant in Ashland, USA, we verify the effectiveness of the proposed method proposed. © 2019 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.