Soft Sensing Technology And Application In Industrial Coal Gasifier

There are increasing demands for process control systems in industry engineering field today.But many variables still exist in some control situations，which are very difficult to measure online,such as variables of control system quality. These factors always lead to a series of consequencessuch as higher measuring error. This paper focuses on several soft-sensor models based on the PLSand neural network of RBF, which can effectively solve the problem of estimation and control ofthese variables. The organization of this paper is as follows:1. Introduce the basic principles of soft sensor technology, and its development and presentsituation. Then, introduce the modeling method and its research status in particularly.2. Introduce the method of principal component analysis simply. Then elaborate the basicprinciples, fundamental properties of the PLS in detail, as well as research situation of its algorithm.At the same time, the model structure of RBF neural network, the training algorithm andperformance are presented, which are base of applications in following chapters.3. RPLS algorithm is proposed by a new derivation method. The advantage of this algorithm isthat it can update the models due to new data and former PLS parameters instead of whole data. Thesimulation results show that this algorithm is able to update models efficiently, and behaves better atprediction accuracy.4. Propose new methods that combines quadratic polynomial, RBF neural network, PLS andrecursive PLS algorithm. Then establish QPLS model, RBF-PLS model and RBF-RPLS. Test eachmodel with two nonlinear model object. Experiment compare and analysis is presented at last of thischapter.5. Introduce the industrial coal gasification technology and its basic principle. Then, thesoft-sensor model of coal gasifier’s synthesis gas compositions is established by using RBF-RPLSmodeling method. The operation results of this model prove its satisfying performance aboutaverage prediction errors of the models. So it can satisfy the requirement of industrial production.6. Summarizes the research results of this total paper and discusses the prospect of the futureresearch work.