| abstract |
The invention provides a method for calculating the airspace operation situation based on migration learning. Firstly, collect the airspace operation situation samples of the target sector and other non-target sectors, generate dimensionality reduction and decomposition of the training sample set based on the factor subset, and use the dimensionality-reduced sample subset to train the base classifier; During set training, the kernel conversion is performed first, so that the converted non-target sector sample subsets can minimize the sample distribution difference with the corresponding target sector sample subsets under the premise of retaining the situational calculation knowledge; scientifically integrate all base classifiers , calculate the classification confidence of each base classifier for each test sample, and then weight the output of each base classifier to obtain the final classification result. The invention realizes the full mining of knowledge in limited samples in the target airspace, and effectively transfers knowledge to samples in non-target airspace, reduces the dependence of model training on large samples, and can be better calculated in the case of lack of samples Operational situation of target airspace. |