Research on scenario recognition for THz channels based on mRMR-GA

To address the challenges of excessive feature parameter redundancy and insufficient scene correlation in terahertz (THz) channel scenario recognition, a recognition algorithm integrating the minimal redundancy maximal relevance (mRMR) criterion with genetic algorithm (GA) optimization was constructed based on feature selection theory and evolutionary computation principles. The crossover and mutation operations of channel characteristics were executed by the genetic algorithm (GA), and the optimal feature parameters with high scenario relevance were selected using the minimum redundancy maximum relevance (mRMR) criterion. These parameters were then inputed into a backpropagation neural network model. To validate the method, a dataset containing 12 channel features was constructed with 1 745 groups of terahertz channel simulation data collected from indoor scenarios, and the model was trained and rigorously validated based on this dataset. The results demonstrate that the proposed algorithm improves accuracy and efficiency by 8% and 38.8%, respectively, and outperforms traditional algorithms in terms of convergence and transfer generalization capabilities.