In recent years, with the continuous development of technology in the field of artificial intelligence, brain-computer interface (BCI) has attracted more scholars' attention. Real-time monitoring of the mental load level of high-intensity mental workers and making dynamic adjustment according to their tasks is an important means to protect national property and the safety of operators. Studies have shown that the characteristic power spectrum density extracted by electroencephalogram (EEG) is sensitive to the change of mental load, but its dimension is too high, which is prone to data disaster. The traditional principal component analysis (PCA) algorithm will lose some of the nonlinear characteristics of the data. Locally Linear Embedding (LLE) is a commonly used nonlinear dimensionality reduction algorithm, but the algorithm is highly sensitive to noise, and the dimensionality reduction results are greatly affected by parameters. RLLE (robust locally linear embedding) adds regular term optimization when LLE optimizes the weight matrix, which not only enhances the anti-noise ability of the model, but also solves the problem of matrix sickness and singularity that may occur in the process of understanding the model. The experimental results show that the classification accuracy of the data reduced by RLLE using support vector machine (SVM) is generally higher than that of the data reduced by PCA and LLE, and has stronger anti-interference ability.