Aiming at the problems of lack of Industrial Control System （ICS） data and poor detection of unknown attacks by industrial control intrusion detection systems， an unknown attack intrusion detection method for industrial control systems based on Generative Adversarial Transfer Learning network （GATL） was proposed. Firstly， causal inference and cross-domain feature mapping relations were introduced to reconstruct the data to improve its understandability and reliability. Secondly， due to the data imbalance between source domain and target domain， domain confusion-based conditional Generative Adversarial Network （GAN） was used to increase the size and diversity of the target domain dataset. Finally， the differences and commonalities of the data were fused through domain adversarial transfer learning to improve the detection and generalization capabilities of the industrial control intrusion detection model for unknown attacks in the target domain. The experimental results show that on the standard dataset of industrial control network， GATL has an average F1-score of 81.59% in detecting unknown attacks in the target domain while maintaining a high detection rate of known attacks， which is 63.21 and 64.04 percentage points higher than the average F1-score of Dynamic Adversarial Adaptation Network （DAAN） and Information-enhanced Adversarial Domain Adaptation （IADA） method， respectively.

After the introduction of federated learning technology in intrusion detection scenarios， there is a problem that the traffic data between nodes is non-independent and identically distributed （non-iid）， which makes it difficult for models to aggregate and obtain a high recognition rate. To solve this problem， an efficient federated learning algorithm named H?E?Fed was constructed， and a network intrusion detection model based on this algorithm was proposed. Firstly， a global model for traffic data was designed by the coordinator and was sent to the intrusion detection nodes for model training. Then， by the coordinator， the local models were collected and the skewness of the covariance matrix of the local models between nodes was evaluated， so as to measure the correlation of models between nodes， thereby reassigning model aggregation parameters and generating a new global model. Finally， multiple rounds of interactions between the coordinator and the nodes were carried out until the global model converged. Experimental results show that compared with the models based on FedAvg （Federated Averaging） algorithm and FedProx algorithm， under data non-iid phenomenon between nodes， the proposed model has the communication consumption relatively low. And on KDDCup99 dataset and CICIDS2017 dataset， compared with baseline models， the proposed model has the accuracy improved by 10.39%， 8.14% and 4.40%， 5.98% respectively.

Aiming at the problem that the existing adversarial example generation methods require a lot of queries to the target model， which leads to poor attack effects， a Text Adversarial Examples Generation Method based on BERT （Bidirectional Encoder Representations from Transformers） model （TAEGM） was proposed. Firstly， the attention mechanism was adopted to locate the keywords that significantly influence the classification results without query of the target model. Secondly， word-level perturbation of keywords was performed by BERT model to generate candidate adversarial examples. Finally， the candidate examples were clustered， and the adversarial examples were selected from the clusters that have more influence on the classification results. Experimental results on Yelp Reviews， AG News， and IMDB Review datasets show that compared to the suboptimal adversarial example generation method CLARE （ContextuaLized AdversaRial Example generation model） on Success Rate （SR）， TAEGM can reduce the Query Counts （QC） to the target model by 62.3% and time consumption by 68.6% averagely while ensuring the SR of adversarial attacks. Based on the above， further experimental results verify that the adversarial examples generated by TAEGM not only have good transferability， but also improve the robustness of the model through adversarial training.

Most of the existing target detection algorithms rely on large?scale annotation datasets to ensure the accuracy of detection， however， it is difficult for some scenes to obtain a large number of annotation data and it consums a lot of human and material resources. In order to resolve this problem， a Few?Shot Target Detection method based on Negative Margin loss （NM?FSTD） was proposed. The negative margin loss method belonging to metric learning in Few?Shot Learning （FSL） was introduced into target detection， which could avoid mistakenly mapping the samples of the same novel classes to multiple peaks or clusters and helping to the classification of novel classes in few?shot target detection. Firstly， a large number of training samples and the target detection framework based on negative margin loss were used to train the model with good generalization performance. Then， the model was finetuned through a small number of labeled target category samples. Finally， the finetuned model was used to detect the new sample of target category. To verify the detection effect of NM?FSTD， MS COCO was used for training and evaluation. Experimental results show that the AP₅₀ of NM?FSTD reaches 22.8%； compared with Meta R?CNN （Meta Regions with CNN features） and MPSR （Multi?Scale Positive Sample Refinement）， the accuracies are improved by 3.7 and 4.9 percentage points， respectively. NM?FSTD can effectively improve the detection performance of target categories in the case of few?shot， and solve the problem of insufficient data in the field of target detection.