Research On Robustness Optimization Strategy For Internet Of Things With Sequence Predition

The Internet of Things(IoT)is deployed widely in modern society,such as fire warning,border anti-terrorism,and smart agriculture.These applications have stringent requirements on data delay and life cycle of IoT.The scale-free IoT model is resistant to random loss of nodes due to energy exhaustion and hardware failure,but it is fragile to malicious attacks on critical nodes in IoT.Therefore,it is necessary to improve IoT robustness and maintain network connectivity under malicious attacks.The neural network model has strong feature extraction capabilities for sequence modeling tasks.Constructing a sequence prediction task for IoT robustness optimization problem can improve optimization efficiency without losing accuracy.Therefore,this thesis proposes a novel robustness optimization strategy based on supervised learning task,which includes three stages: data set generation and processing,model training and predicted topology reconstruction.Firstly,the muti-population genetic algorithm is adopted to generate the topological sample data set,and the network connection relationship is converted into sequence models to adapt to feature extraction.To extract the mapping relationship between initial fragile topologies and target robust topologies,this strategy employs convolutional neural networks to learn the structural characteristics of robust IoT.In addition,the model learning strategy and evaluation indicator are designed in combination with the network connection relationship for model training and screening.Finally,combined with the predicted topology reconstruction algorithm,the trained neural network model can efficiently optimize the IoT robustness without changing the degree distribution.The experiment results show that the neural network model can converge with a few iterations and achieve an optimization effect similar to the target algorithm.Meanwhile,the predicted topology also has an extremely robust onion-like structure.Compared with Hill-Climbing algorithm and Simulated-Annealing algorithm,this strategy not only improves the optimization effect,but also reduces the optimization time.Therefore,the proposed strategy can optimize the resistance of scale-free IoT to malicious attacks with less time cost.