单次脑电分类实验中,采用基于logistic回归的正则化方法来提高分类准确率.首先,提出一种新算法——局部保持投影稀疏logistic回归,将局部保持投影正则项加入到稀疏logistic回归中.该算法旨在保留原始特征空间邻域信息的同时保证结果的稀疏性.然后,利用边界优化法和逐分量迭代算法在训练集上求解权重向量,克服了牛顿-拉夫森法和迭代重加权最小二乘法的局限性.最后,在自步调手指运动数据集上采用十重交叉验证法得到80%的分类准确率,并与稀疏logistic回归的实验结果进行对比,说明局部保持投影正则项有效地保留了对脑电分类有用的信息. In a single EEG classification experiment, the regularization method based on logistic regression is used to improve the classification accuracy.Firstly, a new algorithm, local preserving projection sparse logistic regression, is proposed to add the locally preserved projection regular item to the sparse logistic regression The algorithm aims to preserve the sparseness of the original information while preserving the information of the original feature space.Then, the boundary vector optimization and the component-by-component iterative algorithm are used to solve the weight vector on the training set, which overcomes the Newton-Raphson method and the iterative re-weighting The least squares method.Finally, the classification accuracy of 80% was obtained by using the ten-fold cross-validation method on the self-paced finger motion data set, and compared with the experimental results of sparse logistic regression, which shows that the local preserving projection regular item effectively Retains useful information about EEG classification.