The invention relates to a system for monitoring the working state of a boring rod in real time. The system comprises a signal detection device, an upper computer and an intermediate station, wherein the upper computer analyzes and processes a vibration signal by a method comprising the following steps of: establishing an eigenvector for each group of historical vibration signal; constructing a learning sample by using each group of eigenvector and the corresponding state information as an input parameter and an output parameter; selecting a kernel function, inputting a training sample into a support vector machine (SVM), and performing sample training to obtain an SVM-based classification model; calculating an eigenvector of a real-time vibration signal, inputting the eigenvector into the classification model, performing pattern recognition, and judging whether the boring rod is in a stable state or a flutter state currently; and adding a sample which is to be recognized and is used as a new learning sample to the SVM to improve the recognition capability of the classification model. The system has the advantages of high response speed and high antijamming capability and is convenient to install.