Data Correction Technology Research

Every day modern enterprises are collecting a large amount of measurements. These measurements can reflect the operating conditions of production devices, and is also the basis of various work. Process measurements are often inaccurate, and even contain gross errors. If these measurements with gross errors are directly used, the enterprises can not effectively carry out the work, and even have a great economic loss. So it is necessary to correct measurements to obtain high quality and reliability data for process industries. In this dissertation, some methods of gross error detection are analyzed and studied. Then the methods with some defects are improved. The main contributions of this dissertation are described as follows:(1) Some problems of MT-NT method are discussed, then an improved method is proposed. After a gross error is detected, its variance instead of its value is corrected, and then its original variance is replaced with corrected value. The improved method can overcome the decrease of coefficient matrix rank and also reduce the probability of misidentification. The simulation results show that this method is effective.(2) To solve the misjudgment of Generalized likelihood ratio (GLR) method, nodal test method is introduced, and then a combined method is proposed. After a gross error is detected by GLR method, nodal test method will test the gross error detected again. The combined method can avoid the misjudgment to a certain extent. And the simulation results show its effectiveness.(3) The problems of serial identification with collective compensation method are analyzed and discussed. After that, the linear combination technique is introduced, and then an improved method is presented. The method can reduce the number of variables in the set of candidate gross errors, so that it can reduce the probability of misidentification. Besides, this method can also avoid the computation of project matrix.(4) Several common methods of data classification are introduced, and then these methods are studied and compared by simulation.