Multi-objective Optimization of Tobacco Leaf-Threshing Process Parameters Based on Optimal BP Neural Network

In response to the current problem that the process parameters of the leaf threshing process are mainly set empirically and the optimal combination of many process parameters cannot be accurately obtained to improve the leaf threshing effect, this paper proposes a multi-objective optimisation method for the leaf threshing process parameters by combining PSO-BP neural network with NSGA-II. With the rate of large and medium pieces as the research objectives, the data sets of the leaf threshing process parameters and the post-threshing leaf structure were established through the method of uniform test design. Compared to the unoptimised BP neural network, the PSO optimisation method resulted in a 13% and 17% better fit and a 6.1 and 0.9 reduction in mean square error for the neutral test set. The optimal combination of process parameters was verified by combining the PSO-BP neural network as a fitness function and the NSGA-II algorithm to optimise the threshing parameters. Comparing the model predictions with the experimental validation results, the relative errors of the large and median values were only 8.3% and 7.9%, indicating the accuracy of the model, which can provide some reference for the optimisation of the threshing process parameters.