The invention discloses a frequency spectral integration processing method and device as well as a storage medium. The method comprises the following steps: training a neural network model by using adeep learning method and on the basis of a source domain training sample, so that a trained source domain optical fiber transmission model is obtained; performing transfer learning training on the source domain optical fiber transmission model by using the deep learning method and on the basis of the source domain training sample, so that a target domain optical fiber transmission model is obtained; and receiving a sending path establishment request specific to a service flow, and if a fact that a resource can not be reserved for the service flow in a target domain space division multiplexingoptical network is judged, based on the target domain optical fiber transmission model, predicting frequency spectral integration predication time corresponding to the service flow, and performing corresponding processing on the basis of the frequency spectral integration predication time. The method and device as well as the storage medium which are disclosed by the invention can complete frequency spectrum resource optimization before a service start time and finally effectively bears a service about to be blocked, so that service blocking rate is reduced, frequency spectrum resource utilization rate is improved, and using experience degree of a user is improved.