This study aims to investigate the chaotic characteristics of subway passenger flow during the COVID pandemic in order to improve the prediction accuracy of passenger flow. The research focuses on the daily traffic and passenger flow growth at Chengdu subway stations, using the maximum Lyapunov index to determine the temporal characteristics of chaos. Additionally, the study compares the prediction accuracy of passenger flow growth using an ARIMA model and a Volterra model. Furthermore, a Granger causality test is conducted to examine the relationship between chaos and the prediction results. The findings reveal that chaos accounts for 96.85% of the time series of daily passenger flow growth during COVID, and the overall prediction error of the Volterra model is 23.37% lower than that of the ARIMA model. Causal analysis demonstrates that chaos is an important factor impacting prediction accuracy.