A soft sensor based on kernel PCA and composite kernel support vector regression for a flotation circuit

A soft sensor was developed to estimate the concentrate grade and recovery rate of a flotation circuit. The algorithm uses kernel principal component analysis (KPCA) and composite kernel support vector regression (CK-SVR) to perform the estimation. Firstly, the flotation prior knowledge and KPCA are employed to reduce the dimension of input vector of CK-SVR. Then, considering that the characteristics of kernels have great impacts on learning and predictive results of SVR, a composite kernel SVR modeling method based on polynomial kernel and RBF kernel is adopted which hyperparameters are adaptively evolved by the particle swarm optimization (PSO) algorithm. Simulations using real operating data show that the soft sensor provides the necessary accuracy for a flotation circuit.