Abstract: Owing to its pleasant flavor and rich nutrition, peanut oil is popular among consumers.However, as the price of peanut oil increases year by year, peanut oil is prone to adulteration by the unscrupulous dealers.The adulteration of oil not only infringes upon the rights and interests of consumers, food processors and industries, but also leads to potential health risks.Therefore, it is particularly important to establish a simple and rapid method to detect adulteration in peanut oil.In this study, using the low-field nuclear magnetic resonance(LF-NMR)technology and Chemometrics, a new approach was proposed for the adulteration identification of peanut oil.At first, based on the differences in the attenuation curve of the LF-NMR Carr-Purcell-Meiboom-Gill(CPMG)echo of oils, principal component analysis(PCA)algorithm was used to distinguish peanut oil from adulterated vegetable oils(soybean oil and corn oil).Then, the corresponding discriminant models were developed using cluster analysis(CA)and support vector machine(SVM)algorithm respectively.The parameters and kernel functions of SVM-based model were optimized by grid search method and particle swarm algorithm.At last, the versatility and prediction performances of the SVM-based global model and the SVM-based specific models were compared.Experimental results showed that all samples met the national standard requirements, and all kinds of vegetable oil species can be distinguished by LF-NMR CPMG echo attenuation data combined with PCA algorithm.Peanut oil resides on the left side of the principal component score graph, while the other vegetable oils(soybean oil and corn oil)reside on the right side of the 3D score plot of principal components.The optimal identification rate of the CA-based models was 55.3%, and the detection limit of adulteration identification of peanut oil(soybean oil and corn oil)was 60%.Compared with CA-based models, the SVM-based model using LF-NMR CPMG echo attenuation data can achieve better performances in adulteration identification of peanut oil.When radical basis funcition(RBF)was selected as the optimal kernel functions, the identification rate of the global model and the specific model was equal to 100% with the detection limit of 5%.Moreover, for the SVM-based models, the generality of the global model was better than that of the specific model, but the accuracy of these models were not much different.In addition, it is necessary to continue to add new samples to expand the library, so as to further improve the applicability and accuracy of this model.