Time series data often have inherent periodicity. Therefore, accurate analysis of the extracted data’s periodicity is very important for the accuracy of the subsequent model’s prediction. Despite this, some of the existing methods for extracting the periodicity often fail to accurately extract the data’s potential periodicity due to the large number of repetitions of the data itself within a short period of time. Therefore, this paper proposes the Weight Fast Fourier Transform (WFFT) module for the preprocessing stage of the data, which firstly accurately extracts the cycles of the input time series data and then further adopts the sliding window averaging and cycle splitting averaging for the extracted cycles to reduce the amount of data and reduce the computational cost. The final experimental results demonstrate how well our algorithm predicts the dataset. This module can be used in the data preprocessing stage of any other prediction model, which offers a fresh perspective on forecasting time series data.