Intelligent Adaptive On-line Modeling With The Ability To Cope With Input Exception

When it comes to the engineering applications and economy problems analysis, system recognition and modeling is the indispensable premise of the applications and analysis. Most of the documents cast the focus on the aspect of modeling on the real system based on the presupposition of that the system inputs are absolutely reliable. Few documents can be found on the topic of assuring the system in the right track when there is something wrong with the input variables. Such topic has been researched into and discussed in the dissertation.The problem of multi-co-linearity was analyzed on the process of setting up the model to measure the fuel volume in the airplane tank. Methods of transforming the input variables'domain were applied to solve the multi-co-linearity problem. Backword stepwise regression was applied to the sample data for choosing the input variables. Partial least- squares regression was applied to pick up the principal components. Different ANFIS models based on the dealed data were set up after implementing substractive clustering to obttain the fuzzy rules. The modeling methods proposed above were applied to measure the aircraft fuel volumn during the flight. There was multicollinearity among the sample data of the aircraft fuel volumn. The modeling method which combined the PLS, Substractive Clustering and ANFIS was proposed. The complexity and precision of various types of models including multiple linear regression, backword stepwise regression SCANFIS (Substractive Clustering Adaptive-Network-based Fuzzy Inference Systems) and partial least-squares regression SCANFIS were compared.The reliability of the model in the operating process was discussed. Several kinds of diagnosis algorithm of judging whether there is any exception with the input variables and whether the exception of the input variables disappear was proposed. Then the respective advantage and disadvantage of such types of algorithm was compared. Various solutions was brought forward to assure the models to attain the right result even if there is something wrong with the input variables. And these solutions were applied to the measuring system for calculating the fuel volume of the airplane.