Online testing method and device for deep learning system based on MAPE-D annular structure

The invention provides a deep learning system online test method and device based on an MAPE-D annular structure, and aims at dynamically changing data to realize deep learning system online test through four steps of monitoring, analysis, planning and execution. The method includes: monitoring: performing adversarial generation on the static data set part, generating three similar but different data sets to train three label models, and labeling unsupervised samples according to most principles so as to monitor the accuracy of the operation model; analyzing: analyzing the unsupervised sampleinput into the operation model, and regarding the sample with inconsistent results of the operation model and the label model as a negative sample; planning: enhancing the negative sample data obtained by analysis, mixing the enhanced negative sample data with the static data set part to form a retraining data set, retraining an operation model, and storing weight data; and executing: judging whether to update the operati