关键词: 柴油馏分, 烃类组成, 碳数分布, 预测, 最近邻回归, 过采样

Abstract: For the purpose of obtaining the detailed carbon number distribution of diesel distillate from its bulk properties and hydrocarbon group compositions，a new method was proposed based on the k-nearest neighbor regression algorithm （KNR） and over-sampling. With the standard methods，bulk properties，group compositions and carbon number distribution of the representative samples were obtained to build the data bank. As to a new sample to be measured，the nearest 6 samples were confirmed according to the bulk properties and group compositions, a massive virtual samples were obtained around this new sample using over-sampling technique on the 6 samples. With the KNR，the carbon number distribution of the new sample can be determined by linear weighted summing of the 4 virtual neighbors. A prediction model was established for the straight-run diesel，the contents of 312 carbon number lumps can be calculated simultaneously from the content of sulfur，nitrogen，acid number and the compositions of 11 type hydrocarbons（paraffin，monocycloalkane，bicycloalkane，tricycloalkane，alkylbenzene，indan/tetrahydronaphthalene，indene，naphthalene，acenaphthene，acenaphthlene and tricyclic aromatic hydrocarbon）. The model was accurate，fast and easy to maintain，which made it more useful and valuable in practical implement.

Key words: diesel distilate, hydrocarbon group composition, carbon number distribution, prediction, k-nearest neighbor regression algorithm, over-sampling