The invention relates to an on-line monitoring system for a dissolved gas in transformer oil based on an embedded platform. Aiming at the problems that an existing transformer online monitoring deviceis easily influenced by the phenomena, such as that the chromatographic base line jitter, the chromatographic base line shifts, the chromatographic base line has the abnormal points, the interferenceis large, the waveform is irregular, etc., in the chromatographic peak recognition, so that the misjudgment and the missed judgment of the chromatographic peak are caused and the existing device is lack of self-adaption and self-learning capabilities for misjudgment and the missed judgment of the chromatographic peak, and accordingly the measurement distortion of the gas component data is caused,and the accuracy of the fault diagnosis is further influenced. Under an embedded Vxworks operating system, the on-line monitoring system of the present invention adopts a differential threshold filtering method, a rough penalty method, the first-order and second-order derivatives, a PSO-SVM (particle swarm optimization support vector machine) and other algorithms to carry out the interference peak identification and the gas component classification on a series of chromatographic peaks in chromatographic curve sequence data after oil-gas separation, and calculates the gas concentration value.According to the system, the accuracy and stability of the chromatographic analysis are greatly enhanced, and the performance of an online monitoring device for the dissolved gas in the transformer oil is improved.