The invention discloses an alarm correlation analysis method based on a graph data model. The alarm correlation analysis method comprises the following steps: step 1, constructing the graph data model: converting a relational database model into the graph data model; Step 2,performing fault marking: when a network fault event or a user complaint event is received, marking a corresponding node inthe graph data model as a fault; Step 3, performing alarm pre-association: taking out the node with the lowest hierarchy from the graph data according to a fault time window, and storing the node in apre-association result set; Step 4, carrying out root alarm association judgment to determine a fault root source: in a pre-association result, when a further association relationship does not exist,determining the corresponding node as the fault root source; And when the further association relationship exists, the node involved in the association relationship is the fault source. According tothe method, the relation model is converted into the graph model, the processing efficiency is improved, the service request can be quickly responded, and the scene of real-time processing and analysis of the network data is effectively supported.