The cigarette draw resistance monitoring method is incomplete and single, and the lacks correlation analysis and preventive modeling, resulting in substandard cigarettes in the market. To address this problem without increasing the hardware cost, in this paper, multi-indicator correlation analysis is used to predict cigarette draw resistance. First, the monitoring process of draw resistance is analyzed based on the existing quality control framework, and optimization ideas are proposed. In addition, for the three production units, the cut tobacco supply (VE), the tobacco rolling (SE), and the cigarette-forming (MAX), direct and potential factors associated with draw resistance are explored, based on the linear and non-linear correlation analysis. Then, the correlates of draw resistance are used as inputs for the machine learning model, and the predicted values of draw resistance are used as outputs. Finally, this research also innovatively verifies the practical application value of draw resistance prediction: the distribution characteristics of substandard cigarettes are analyzed based on the prediction results, the time interval of substandard cigarettes being produced is determined, the probability model of substandard cigarettes being sampled is derived, and the reliability of the prediction result is further verified by the example. The results show that the prediction model based on correlation analysis has good performance in three months of actual production.

翻译：目前卷烟吸阻监测方法单一且不完善，缺乏相关性分析与预防性建模，导致市场中出现不合格卷烟。针对该问题，在不增加硬件成本的前提下，本文采用多指标相关性分析预测卷烟吸阻。首先，基于现有质量控制框架分析吸阻监测流程，提出优化思路；其次，针对切丝供给（VE）、卷制（SE）和成型（MAX）三个生产单元，通过线性和非线性相关性分析探索与吸阻直接或潜在相关的因素；再次，将吸阻的相关因素作为机器学习模型输入，以吸阻预测值作为输出；最后，本研究还创新性地验证了吸阻预测的实际应用价值：基于预测结果分析不合格卷烟分布特征，确定不合格卷烟产生的时间间隔，推导不合格卷烟被抽检的概率模型，并通过实例进一步验证预测结果的可靠性。结果表明，基于相关性分析的预测模型在实际生产三个月内表现良好。