The invention relates to a machine reading understanding method based on a pre-training model. The machine reading understanding method comprises the following steps: 1, preprocessing data, 2, performing advanced semantic fusion through an advanced semantic fusion network layer according to the output of the pre-training model, 3, further performing capability learning on the machine reading modelafter semantic fusion, and 4, calculating the mean square error loss of the named entity, and training the machine reading model. The method has the advantages that high-level semantic information isextracted from the text, higher-dimension information is provided for the model, and meanwhile, the high accuracy of the method has more reference significance compared with the mode that the model tries to extract the information in the training process, according to the method, through capability learning, under the condition that the scale of the model can be kept unchanged, the machine reading capability is improved, so that the model can quickly complete an inference task on the premise of relatively high performance.