Due to poor measurement environment

expensive analytical instruments and large measurement delay

some key variables of power generating units are difficult to realize on-line measurement. To solve this problem

a feature selection method based on the Pearson correlation coefficient and the least angle regression algorithm(LARS) is presented. On this basis

a soft sensing model of oxygen content in flue gas of power plants is established. Firstly

the initial variables of oxygen content in flue gas are determined according to the mechanism study

and the Pearson correlation coefficient is used to fuse the data of multiple measurement points. Then

the least angle regression algorithm is used to select the input variables

and the soft sensor model of oxygen content in flue gas is established by introducing the Gauss process regression(GPR) model. Finally

this model is applied to simulate the actual steady condition data of an ultra supercritical 1 000 MW unit. The results show that

the predicted results have little error with the actually measured values

and the precision of the model is higher. Applying this feature selection method can effectively reduce the number of the secondary variables used in modeling

decrease the input complexity of the model and improve the calculation efficiency. Moreover

it can enhance the static accuracy and generalization ability of the flue gas oxygen content model.