Scrap steel foreign matter weight deduction learning method based on multi-task gain regression

The invention discloses a multi-task gain regression-based waste steel foreign matter deduction learning method, which adopts an MTSN prediction model and an Embedding module to realize data continuous vector representation, adopts a multi-layer sensing network under hierarchical conditions to realize data learning, and respectively generates cognitive data loss measurement and loss measurement of discrete waste steel grade data. Finally, an MTL multi-task automatic learning weight mechanism is adopted, variables with large correlation are concerned, variables with small correlation are weakened, and the weight coefficient is automatically adjusted; according to the multi-level regression network, training is carried out respectively according to different data types, feature influence factors are calculated independently for data with uncertain influence, the characterization capacity of the data is improved, and the problem that the characterization capacity of the data is poor can be solved; according to t