Application of SPCA-LSSVM model in soft measurement of Gasoline dry point

A soft measurement model of least squares support vector machine based on data preprocessing for sparse principal component analysis(SPCA-LSSVM) is proposed to solve the problem that the gasoline dry point on the top of atmospheric pressure tower is difficult to measure online. This method combined sparse principal component analysis (SPCA) with the least squares support vector machine (LSSVM) method, principal component analysis (PCA) is transformed to solve the regression optimization problem with quadratic penalty by introducing the sparse constraint aiming at variable selection, in this way, the dimensions of the sparse eigenvector reduces dependency and becomes more independent, it also reduces the influence of measurement noise. The data sample processed by SPCA is used as the input of LSSVM model to establish the soft measurement model of dry point at atmospheric pressure tower top, the problem of lack of sparsity in LSSVM is solved. The simulation results showes that the SPCA-LSSVM model has higher prediction accuracy than the traditional LSSVM model, and the PCA-LSSVM model, reducing the complexity of the model, showing the superiority of the soft-sensing model.