Welding trajectory real-time detecting method in welding process of welding robot

A welding trajectory real-time detecting method in the welding process of a welding robot includes the steps that welding trajectory data in the welding process of a normal welding robot are acquired; dimensionality reduction analysis is performed on the acquired welding trajectory data, and the cluster number of distribution dense clusters of the data is determined; according to the cluster data, all data points are accurately classified, and center point coordinates of each category are obtained; most TCP data point coordinate values of each category are substituted into a BP neutral network algorithm for network training; remaining TCP coordinate values are substituted into a neutral network algorithm for network authentication success; welding trajectory data of a certain point are acquired in real time; TCP-X and TCP-Y coordinate values in actual TCP coordinate values are substituted into the neutral network algorithm so that a predicted TCP-Z coordinate value can be obtained; and a difference between the predicted TCP-Z coordinate value and an actual TCP-Z coordinate value are is calculated, the point is a normal point if the difference falls into a preset confidence interval, and the point is an abnormal point if the difference does not fall into the preset confidence interval, and predicting alarming is performed.