Feature Selection and Hyperparameter Optimization for Combustion Status Recognition Model of MSWI Process Based on Parallel Differential Evolution

Accurately identifying flame combustion status during municipal solid waste incineration (MSWI) is a crucial prerequisite for intelligent combustion control. Previous research in constructing identification models often relied on expert knowledge and extensive repetitive experiments to determine hyperparameters and feature selection parameters. The process of parameter determination was resourceintensive, demanding significant computational power. To tackle these challenges, a method using parallel differential evolution (PDE) algorithms for optimizing the hyperparameters of combustion state recognition models is proposed. Initially, feature parameters and model hyperparameters are encoded as chromosome groups for the differential evolution algorithm, followed by random parameter initialization. Then, the population is randomly divided into several subpopulations, and each undergoing parallel evolution. Evolution halts upon meeting specific stagnation conditions. Finally, the hyperparameters enabling the recognition model based on vision transformer and improved deep forest classification to achieve optimal performance. The effective of the proposed method is validated by using the actual data of an MSWI plant.