Research On Improved Data Classification And Gross Error Detection

The large number of process data, flow、temperature、pressure、density、composition and so on, generated by modern industrial processes is the foundation of the process modeling, control and optimization and is to protect the safety of industrial processes carried out. Therefore, we must ensure the accuracy and reliability of the process data. However, the process data will deviate from its real value due to inevitably be affected by the error generated in the measurement process or in the treating process, so that the process data no longer satisfy the certain physical and chemical laws or other process constraints. The data correction is a technique for dealing with these process measurement data, which makes the coordinate values close to the true value of the process data on the maximum degree and to satisfy the certain physical and chemical laws or other process constraints, by eliminating or compensating the gross errors and by eliminating the effect of random errors. This paper mainly does researches on the data classification and gross error detection of industrial steady-state system and the gross error detection of power system, as follows:(1) Propose the method of data classification based of elementary transformation, uses the related properties of matrix to classify data based on the theory of graph theory. The new method has the features of the process-oriented and equation-oriented. When classifying the unmeasured data, the method only need to do a row elementary transformation of the unmeasured data factor matrix. When classifying the measured data, the method reduces the unmeasured data factor matrix to its set of vector basis. The study found that, compared to the method of projection matrix, the new method is more thorough in classifying data and has a more simple calculation process. Compared to the linear correlation method, the new method simplifies the size of matrix, has less calculation steps and smaller amount of calculation and is more applied to the data classification of complex process system.(2) Propose the method of gross error detection based on improved residual, due to the defect that the measurement error obtained by the least squres state estimation has missing part. The new method decomposes the measurement error into the detectable part and undetectable part by the projection matrix. The ratio, which is named Undetectable Index (UI), of the latter to the former represents the degree of difficulty of the measurement data error detection. Then the UI is used to calculate the residuals which were not included in the state estimation residuals and were used to modify the data coordination residuals. The simulation results show that the improved residuals can improve the accuracy of gross error detection of power system.(3) Conventional gross error detection methods are mainly based on statistic test, but these methods can not display the credibility of the test results. This paper proposes a new method for gross error detection based on Transferable Belief Model (TBM). This method calculates the confidence level of each measured value by combining the evidences obtained from the generalized likelihood ratio test (GLR) and node test (NT). Then the pignistic probability of each measured value is figured out which is the criterion for diagnosis. The strategy of sequence compensation is adopted in this new method, so it’s considered to differentiate the degree of confidence of synthesis evidence to avoid the equation of the degree of confidence of different evidence by multiplying the degree of confidence and the ratio of the current statistic to the max statistic. The simulation results indicate the validity of the new method to crack the system also with measurement error and system leak.