The Research On The Methods Of Soft-sensing And Its Industrial Application

As an emerging technology, soft-sensing technology has broad prospects for development, has become one of the main direction of process control and instrumentation research, with the soft measurement theory research and practice of continuous improvement and development, will play a vital role in the industrial control stage. This paper takes the actual project as the background, through the research of the soft measurement theory, put forward three kinds of soft measurement modeling method.One is soft sensor modeling method based on MKPLS. Engineering practice, the first use common methods partial least squares (PLS) to establish soft sensor model, but considering the PLS method of operation is slow, model abstraction, is a linear algorithm, less able to deal with nonlinear data, the introduction of kernel function, established the kernel Partial Least Squares (KPLS) soft sensor model, taking into account a single kernel function has some limitations, the introduction of mixed kernel compensate the lack of a single kernel function, the establishment of a hybrid kernel partial least squares (MKPLS) soft sensor model;The second is based on MKPLS and neural network soft sensor modeling method. Taking into account the previously modeled using a single method has its limitations, this hybrid method to establish soft measurement model to compensate for its shortcomings. The MKPLS respectively with BP neural network, RBF neural network are combined to create the corresponding MKPLS-BP and MKPLS-RBF two hybrid soft sensor model, and through industrial data using MATLAB software simulation results show hybrid model of learning ability and generalization ability with respect to the modeling of a single method improved;The third is based on least squares support vector machine and MKPLS soft sensor modeling method. Consider LSSVM has better nonlinear processing capacity, fast speed, etc., to establish a MKPLS-LSSVM soft measurement model; LSSVM model parameters taking into account the soft sensor model fitting accuracy and generalization ability plays an important role, simple and easy to implement and has a strong global optimization ability of particle swarm optimization (PSO) algorithm optimization LSSVM model parameters, build MKPLS-PSO-LSSVM soft sensor model, the simulation proved the model parameters using the PSO optimization, model accuracy greatly improved.