Provided is a method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument. The electronic nose instrument includes a gas sensor array module (I), a capillary gas chromatographic column module (II), an automatic headspace sampling module (III), a computer control and data analysis module (IV), an automatic lifter (V) for headspace sampling, a large-volume headspace vapor generation device (VI) and two auxiliary gas sources (VII-1, VII-2). In the gas sampling period of T0-300-600 s, the gas sensor array module and the gas chromatography module have different flow rates, volumes and staring sampling time points of gas sampling, but have synchronous selection and analysis time points of multiple sensitive information. The electronic nose instrument obtains a 69-dimensional combined pattern, including steady-state response peak values, corresponding peak time points as well as under-curve areas, through each on-site real-time detection to a tested sample. The electronic nose instrument detects a large number of odorous samples to establish a big odor data. On this basis, the normalization fusion preprocessing is done, and the cascade machine learning model realizes both an on-site recognition of many foods, condiments, fragrances and flavors, and petroleum waxes and a real-time quantitative prediction of their odor quality grades and many key component concentrations.