堆叠隐空间模糊C回归算法及其在发酵数据多模型建模中的应用

摘要

The switching regression algorithm FCR is sensitive to noise data and outliers. The algorithm also has low levels of capability for dealing with complex data. In order to handle these problems, an improved fuzzy C⁃regres⁃sion algorithm is proposed based on cascaded hidden space. In our method, principal component analysis is com⁃bined with extreme machine learning feature mapping and multilayer feedforward neural networks. The experimental results show that our proposed method is more stable as regards noise data and outliers, and thus more suitable for handling complex data and multi⁃model modeling problems for the fermentation process.%切换回归算法FCR的性能容易受到噪声点以及离群点的影响，同时该算法对于复杂数据的处理能力较差。对此，文中提出一种基于堆叠隐空间的模糊C回归算法。该算法将基于ELM特征映射技术，利用主成分分析进行特征提取，再结合多层前馈神经网络学习结构对隐空间进行多次扩展和压缩。实验结果表明，该算法具有更好的抗噪性能，对模糊指数的变化不敏感，同时在处理复杂数据以及在多模型建模中更加精确、高效、稳定。