Application of Kmeans-IPSO Optimized RBFNN in the Prediction of Tobacco Leaf Drying Process Parameters

This paper introduces a prediction model for tobacco drying process parameters using a radial basis neural network (RBFNN). The model establishes a nonlinear relationship between varying tobacco leaf characteristics and drying process parameters. To enhance prediction accuracy, the initial centroid is selected using the Kmeans algorithm. Additionally, an improved particle swarm optimization (IPSO) algorithm optimizes the training center, width, and weights of the RBFNN. Experiments show the model minimizes mean square error with 56 hidden layer centroids. The optimized model closely predicts actual values, reducing manual labor reliance and increasing automation in tobacco drying production. To the best of our knowledge, we are the first to apply this specific hybrid approach to an industrial environment.