【摘要】 针对复杂流程工业中固有的非线性将导致传统线性降维方法性能降低的问题,以及传统核方法的性能严重依赖于所选取的核函数形式的问题,提出了一种改进的最大方差展开投影非线性过程建模法。该方法利用流形学习中最大方差展开对核函数进行学习,同时保留了输入数据空间中的边界特性。利用最大角回归学习出一种映射,避免了传统最大方差展开法只能提供训练样本的低维嵌入,使得输入空间能够最大程度地接近于这种低维空间。数值仿真和化工流程仿真模型上的实验表明了该算法的有效性。更多还原

【Abstract】 An improved maximum variance unfolding projection method is proposed in this paper to address the following two problems. The one is that performance of the tranditional linear dimension reduction methods can be influenced by the inherent nonlinear characteristics in the process,and the other is that kernel methods have high dependence on the empirical selection of its kernel function. The kernel can be learned by the maximum variance unfolding in the proposed method,and simultaneously the boundary in the input space can be preserved. Tranditionally,the maximum variance unfolding only provides a lower dimensional embedding of training samples,a least angle regress-based functional mapping is employed in this paper. It makes the input space approximate the lower dimensional space at most. The experiments on numerical simulation and chemical plant simulation shows that the proposed method is more efficient.更多还原