非线性部分最小二乘方法用于推断估计器设计

针对工业过程中某些变量难于在线测量的问题，提出一种基于人工神经网络（ANN）的非线性部分最小二乘的推断估计策略。首先对历史数据进行部分最小二乘回归分析，提取特征信息。然后用神经网络建立主元之间的内部非线性关系，得到产品质量的非参数模型。仿真结果表明：该推断估计器具有良好的跟踪速率和较高的估计精度，其性能优于基于原始人工神经网络推断估计器的性能。

Nonlinear Partial Least Square for Designing Inferential Estimation

In order to overcome difficulty of some variables online measurement in industry processes, a inferential estimation strategy based on ANN and nonlinear partial least square is proposed. Firstly partial least square is used to analyze historical data. Then RBFN is used to establish inner relations between principal components. Simulations show that this inferential estimator possesses features of fast tracking speed and high estimation accuracy, with performance much better than that of inferential estimators based on primal radial basis function networks.