Traffic flow data processing method based on data word segmentation

The invention discloses a traffic flow data processing method based on data word segmentation, and the method comprises the following steps: 1, data processing and word segmentation: the data is traffic flow data and is acquired by a sensor deployed on a road, and a data set is defined as X; step 2, constructing a model; step 3, designing a preposed task and defining a pre-training loss function, and designing an MTP preposed task to train the model in the step 2 so as to learn internal characteristics of traffic flow data and obtain a representation vector of the data; 4, performing fine adjustment according to the downstream tasks, and in the fine adjustment stage, replacing the last full-connection layer with a corresponding output layer according to targets of different downstream tasks; according to the method, the model is trained in a self-supervised learning mode, the data does not need to be stored in a centralized mode and labeled by professionals, time and professional knowledge needed by data labe