Abstract: In chemical data test, the guarantee of data quality has great meanings, for unusual data needs accurate diagnosis and handling. This paper has enumerated the present common methods of some kinds of chemical unusual data diagnosis and handling. It analysed their basic principle and algorithm steps , and summed up the good and shortcomings of each one. Then, with the surplus oil of airplane as an example to measure observation data, we discusssed the method based on statistics how to come to handle this kind of problem. First, as handling data with general multivariate regression method, we discover such handling the data linear correlation coefficient that comes out very little. So, to make some improvements , we adopt step by step regression analysis method. Regress analysis step by step can will data in is away from crowd notable value remove , then again the value for surplus do regression handling; Discover the data of such handling linear related obvious improvement. But because these are all based on setting data in advance having the conclusion that reached on linear foundation, so this paper do improve further again for data processing technique. It have led into polynomial regression method to handle. So handling comes out that data have obviously very good relating linear, and the effect is also compared.
Keywords：unusual data; delayed coking; linear regression analysis；multivariate stepwise regression；polynomial regression
