Research On The Detection Methods Of Gross Errors For Process Data

In the industry process, there are inevitable errors in the data gathered in the industry scene, which contain not only random errors but also gross errors. Gross error is defined as the significant difference between the measured data and the real data, a major distortion which is caused by the measurement failure of the instrument, the transmission error of measuring instrument data, the operational instability and so on. Generally, gross errors contain system errors and significant likelihood errors, which are virtually an umbrella of all the outliers not conformed to the normal distribution. The existence of gross errors makes the measured data unable to manifest the true working condition. Therefore, it is important to detect gross errors and make amendment.This paper concerns on gross error detection and estimation in state linear and nonlinear process. The main contents of this paper are outlined as follows,(1) Based on the classic Measurement Test (MT) Node Test (NT) combined method, the proposed method adds some constrained conditions to improve the NT and proposes synchro-compensation to ensure the accuracy of the gross error detection problem which contains more than one gross error.(2) The transfer belief model (TBM) is proposed to be applied in the gross error detection and it uses the traditional statistical testing methods as the basis of generating evidence. It fills in gaps in the traditional statistical testing methods that they only use statistical testing to make diagnosis in some aspect. Considering that it is possible there is more than one maximum pignistic probability, the paper puts forward the idea of sequence selection of pignistic probabilities and synchro-compensation to make sure that all the gross errors can be detected as much as possible.(3) Owing to the fact the former two methods can only be applied in gross error detection problems in which the data are subjected to linear constraint equations rather than the nonlinear constraint equations. This paper proposes the multiplied integer nonlinear programming (MINLP) synchro-compensation gross error detection method. Based on the AIC framework, this method combines the multiplied integer nonlinear programming and least square estimate model to detect the gross errors and estimates the value of the gross errors.