基于神经网络的烟丝填充值预测模型研究

摘要

填充值是烟丝的一项重要物理指标。在卷烟制丝生产中，叶组配方和工艺流程一般较为固定，所以制丝工艺参数对烟丝填充值的影响更为直接、突出，但工艺参数对填充值的影响为非线性的，难以根据工艺参数直接推算出烟丝填充值。针对此问题，选取了7个影响较大的工艺参数，采用 BP 神经网络对7个工艺参数和烟丝填充值间的数量关系进行了初步建模。通过BP神经网络设计和大数据量的训练后，该模型具备了通过工艺参数预测烟丝填充值的能力，预测结果的相对误差为4%左右，这为工艺参数和填充值之间的相互调整提供了理论依据和仿真方法。%The filling property is a vitally important physical index of cut tobacco. Due to the similar cigarette blending formulation and fixed technological process in cigarette production, relevant technological parameters have direct and significant influence on the filling value of cut tobacco. However, the relationship between the parameters and filling value is nonlinear, which makes it difficult to calculate filling values of cut tobacco directly based on the parameters. Therefore, seven most important technological parameters were selected, and the quantitative relation model between various parameters and the filling value was established with the BP neural network. Further, the model was designed with the BP neural network and trained through a large amount of data, so it could forecast the filling value according to the seven technological parameters and the relative error of forecasting results was about 4%. The establishment of the forecasting model provides a theoretical basis and simulation method for the mutual adjustment between the technological parameters and the filling value of cut tobacco.