Monitoring and early warning of dynamic roof hazards in mines is crucial for mining production and personnel safety. Traditional multi-point displacement monitoring and early warning methods rely on a single absolute threshold based on monitoring data, resulting in poor sensitivity and reliability of the warning. In response to this issue, a new early warning method for dynamic roof hazards in mines was proposed based on multi-point displacement data and the Autoregressive Integrated Moving Average (ARIMA) model. In this method, superposition techniques were employed to enhance the signal-to-noise ratio of multi-point displacement time series data, and a graded early warning method combining enhanced windowed waveform characteristic function ratio and kurtosis was introduced, which addressed the inadequacies of traditional threshold warnings. Furthermore, a dynamic weighted parameter ARIMA model was constructed to predict displacement data, enabling the integration of monitoring and prediction data for early warning of dynamic roof hazards in mines. The theoretical testing of this method shows a false alarm rate of 13.3%, with an average error of 15.4% for secondary or higher warning characteristic values. The test results of engineering actual data of kurtosis over two periods show that the warning false alarm rates are 11.8% and 6.7%, respectively, and the average errors of the secondary or higher warning characteristic values are 15.1% and 14.4%, respectively. In summary, the proposed early warning method for dynamic roof hazards in mine has a high application value.