卡尔曼滤波优化的高斯过程回归模型

To solve the problem, that whole modeling can not be carried out with data from multiple information sources, an optimization Gaussian process regression model based on Kalman filtering (KF-GPR) was proposed. Firstly, according to discrete sample data obtained from multiple sensors, the model carried through GPR, respectively predicted mean value and variance of the key parameters, regarding them as measurements and noise of soft sensor outputs. Then, fusing the outputs of soft sensors based on Kalman filtering algorithm, the model carried out fuse optimization for the outputs of multiple GPR under the rule of the least mean square error, achieving the outputs of the optimized model. And then, some simulation experiments were carried out to compare KF-GPR with other average value fusion methods. The results show that KF-GPR can obtain prediction curves with higher fitting accuracy, verifying the validity of the model. Finally, KF-GPR was applied to the case analysis of temperature variation with latitude, presenting the latitude-temperature prediction curves according to season.