A multi-model fusion soft sensor modelling method and its application in rotary kiln calcination zone temperature prediction

Abstract

A soft sensor is necessary for industrial process control and analysis, and the core problem is how to construct an appropriate model having a fast convergence speed and a good generalization performance. A multi-model fusion soft sensor modelling method is proposed. Firstly, kernel principal component analysis is applied to choose the non-linear principal component of the model input data space, and then the least-squares support vector machine applied to regression modelling, which could not only reduce the complexity of the calculation but improve the generalization ability. For a least-squares support vector machine, prediction model parameters are difficult to determine; an improved particle swarm optimization algorithm is proposed to optimize the least-squares support vector machine prediction model parameters, and prediction accuracy is improved. The calcination zone temperature of a rotary kiln is chosen as a simulation object because it is difficult to measure directly. Simulation results indicate the proposed soft sensor modelling method has a high learning speed, good approximation and generalization ability, and proved an efficient modelling method.