Fuzzy Multi-model Soft Sensor And Recursive PLS Algorithm

Although the technique of process modeling and system identification has been successfully applied in modern industrial such as oil-refinery, petrol-chemistry, metallurgy, paper making, and pharmacy, there are still many cases that many variables are difficult to be measured using the physical sensors. In order to solve this problem, the soft sensor technique is proposed which mainly focuses on the variable estimations through the calculation of mathematical models. With the development of modern industry which will confront more and more complex process,the technique of soft sensor has a widely applicable prospects and has been one of the most focuses in process instrument and measurement technique. Soft sensor modeling methods and on-line updating technique based on traditional PLS model are investigated in detail, and have been applied to estimate the dry point of aviation kerosene Oil of the Dushanzi Oil-field. The main work is described as follows:Firstly, a survey of soft sensor technique is reviewed; the backgrounds, application fields and prospects of soft sensor are introduced. The concept of soft sensor is demonstrated and engineering design steps are described, which includes secondary variables selection based on the mechanism analysis, data collection and pretreatment, soft sensor model establishment, on-line correction and implementation.Secondly, As process modeling is the core problem of soft sensor, All kinds of traditional methods of soft sensor modeling methods are discussed and summarized in details in this paper, while the characteristic of each approach is described respectively. One improved soft sensor approach based on the fuzzy clustering multi-model is proposed. Simulations on the Dry Point of Aviation Kerosene Oil process show the improved method is effective, more accurate and better widely applicable prospects.Thirdly, considering the model mismatch problem of traditional PLS modeling method with long-term operation online, a discounted-measurement block recursive PLS is proposed to update model, which combines with classical recursive PLS algorithm. The result of simulation and estimation of the Dry Point of Aviation Kerosene Oil shows that this method can solve model mismatch problem. The drawbacks of classical recursive PLS on data saturation and storage can be overcome, and the computational complexity can be reduced greatly.Fourthly, a convenient and intuitive simulation platform of soft sensor modeling is developed, which is based on the graphic user interface (GUI) of MatLab environment. This platform provides the possibility of good interaction and extensibility.Finally, based on summarization of this paper, several problems relative with theory and applications for further research are discussed.