A DISTRIBUTED SELF-SUPERVISED INTRUSION DETECTION METHOD FOR MANETs

With the widespread use of MANETs, the intrusion phenomenon is becoming more and more serious. Most of the existing schemes are difficult to meet their requirements of accuracy and real-time. Therefore, we design a distributed self-supervised network intrusion detection model by integrating the deep sequential structure TabNet and gated recurrent unit improved by genetic algorithm (Tab-GAGRU). The model was pre-trained through TabNet which provided efficient and fine-grained representation information for the classification model. The GRU was constructed to capture the time dependence between feature vectors, and network parameters of the model were automatically optimized by GA to ensure the anomaly detection accuracy of network traffic. Spark was used to optimize resources and reduce the processing time of model training classification. The experimental results show that the accuracy of the proposed method is up to 99.95%, and the model can reach 22.4s in detection time.