Efficient Statistical Significance Approximation for Local Similarity Analysis of High-Throughput Time Series Data Using the Circular Moving Block Bootstrap

To address the challenges of statistical inference for non-stationary traffic flow, this paper proposed an improved block permutation framework tailored to the correlation analysis requirements of traffic volume time series, and developed a statistical significance assessment method for local similarity scores based on the Circular Moving Block Bootstrap (CMBBLSA). This method avoided the distortion of the statistical distribution caused by non-stationarity, thereby enabling the estimation of the statistical significance of local similarity scores. Simulation studies were conducted under different parameter settings in the AR(1) and ARMA(1,1) models, and the results demonstrated that the Type I error probability of CMBBLSA under the null hypothesis is closer to the preset significance level α. An empirical analysis was also carried out using traffic flow monitoring data from main roads in first-tier cities, and the results indicated that CMBBLSA can reduce more false positive relationships and more accurately capture real correlations.