| With the rapid development of science and technology, a wide range of industries such as chemical, biological and nuclear power have higher requirements about the accuracy and reliability of the measurement of the value of the variable in the production process. Soft sensing techniques have gone through the process from linear to nonlinear, offline correction to online correction. Coupled with the advantages of high-precision and low-cost, soft sensing techniques have been valued by more and more experts. In order to solve problems of high dimension, data coupling with each other and nonlinearity in soft sensing of industrial processes, a new soft sensing method based on principal curves was proposed. Data from actual industry was used to validate the method, resulting in a good prediction. Above all, the main content and innovative work of this paper were as follows:(1)According to the study with high dimension, data coupling with each other and nonlinearity in the industrial processes, a regression model based on principal curves with stronger correlation was proposed. The model adopted correlation coefficient weighted method to strengthen the correlation between independent variable matrix X and dependent variable matrix Y, achieved greatly correlation between X and Y indirectly. After a correction of the data matrix of independent variables and the dependent variable, the correlation between dependent variables and independent variables was considered simultaneously, when extracting latent variables by principal curves. A nonlinear function was used to fit nonlinear relations between latent variables in latent space. So the method could extract the nonlinear characteristics of the object well.(2)To validate the accuracy of the nonlinear regression model based on principal curves, this model was used in the distillation unit of vinyl chloride monomer process of production and the power control system of the nuclear reactor to achieve the soft sensing of the water content of the bottom product of the rival tower and the power of the reactor of nuclear power. It was also made an efficiency comparison with the nonlinear model, modeled by traditional partial least squares method, polynomial partial least squares method and BP neural network method. The comparison result showed that the soft sensing method based on nonlinear regression model has a more accurate prediction of the dominant measured variables.(3)For the problem of a large number of single-loop and cascade loop control system in nuclear power industry and the special importance of the operation of nuclear power, a simulation tool to achieve the simulation of the subsystems of nuclear power was designed. This tool has achieved the system configuration, the dynamic trend of drawing functions, and it was simulated and proved in nuclear steam emission control systems and the regulator level control system. It was easy and intuitive to be used, and had value of promotion and application. |
