In order to achieve high-precision and rapid prediction of carbon and oxygen elements and active groups on the surface of biochar, a data set containing quantitative characterization information of midinfrared spectroscopy and surface carbon and oxygen elements and their occurrence forms was established based on 120 groups of biochar samples accumulated by the research group. Using support vector machine (SVM) and random forest (RF) machine learning intelligent modeling methods, combined with interval partial least squares (IPLS) and principal component analysis (PCA) and other feature selection strategies, four prediction models of IPLS+RF, IPLS+SVM, PCA+RF and PCA+SVM were constructed, and the quantitative and rapid prediction of surface carbon and oxygen content (S_C,S_O) and eight carbon and oxygen occurrence forms, a total of ten prediction targets, was realized. Among them, there were five forms of C=C, C—C, C—O, C=O, O=C—O from the C1s energy spectrum and three forms of C=O, C—O, O=C—O from the O1s energy spectrum. The results showed that the main occurrence forms of carbon on the surface of biochar were C_C=C and O_C—O, and the main occurrence forms of oxygen on the surface of biochar were C_C—O, C_C=O, C_O=C—O and O_C=O; the characteristic bands of 4000~3464cm-1 and 1588~650cm-1 both contained characteristic information highly related to the content and speciation of carbon and oxygen on the surface of biochar, but the information contained in 1588~650cm-1 was more abundant. From the perspective of model prediction accuracy, the four prediction models of IPLS+RF, IPLS+SVM, PCA+RF and PCA+SVM all had good prediction ability, especially IPLS+SVM and PCA+SVM. The coefficient of determination of the optimal model for ten prediction targets was above 0.93; however, from the perspective of model stability and generalization ability, C_C—C, C_O=C—O, O_C=O, O_C—O still needed to be further improved.