Key Technologies For Intelligent Networking Of Heterogeneous Integrated Networks In Complex Environments

In recent years,natural disasters have occurred frequently worldwide,and public security has become more complex and severe.According to the United Nations Office for Disaster Risk Reduction,there have been 7,348 major natural disaster events recorded globally in the past 20 years,caused 1.23 million deaths and US$2.97 trillion in economic losses.To cope with increasingly severe natural disasters,terrorist attacks,and other complex scenarios,there is an urgent need for reliable,stable,and efficient communication networks.Traditional communication network technology is limited by a single networking method,and network capabilities are limited,making them less reliable.Therefore,it is necessary to construct a heterogeneous integrated network that combines multiple communication systems,technologies,and equipment,using different types of communication networking methods to complement each other,therefore improving the robustness and reliability of communication networks.It can satisfy diverse communication tasks and plays a significant role in safeguarding people’s lives,health,and property security.However,compared to traditional communication networks,heterogeneous integrated networks face three major challenges in complex environments:1)network nodes are vulnerable,links are easily interrupted,and therefore reliable information transmission is difficult;2)there are many types of equipment,heterogeneous resources,and therefore resource collaborative control is challenging;3)the network business requirements is great differentiation and strong emergence,and therefore resource ondemand scheduling is difficult.In order to address these issues,this paper focuses on the key technologies of intelligent networking of heterogeneous integrated networks in complex scenarios,and proposes following methods,traffic control and reliable transmission in non-stationary environments,efficient network resource game method in non-cooperative modes,and differentiated business and resource on-demand adaptation method under emergency conditions.The main innovations are described in detail below.1)Traffic Control and Reliable Transmission in Non-stationary Environments:In response to the challenge of reliable information transmission in environments where network nodes and links are susceptible to damage and interruption,this paper proposes a method for traffic control and reliable transmission in non-stationary environments.Firstly,we design a centralized and distributed hybrid network traffic control architecture,propose a hierarchical flow scheduling mechanism based on multiple time scales,and construct a centralized learning and distributed execution network traffic control model.These methods achieve global optimal control and local quick response to network traffic.On this basis,we propose a network load balancing method for multipath transmission,and construct a multipath transmission balancing model.Then,we design a maximum entropy reinforcement learning based network load balancing algorithm.Experimental results show that,the proposed methods can effectively reduce the link load,and can quickly converge to the global load balancing point under network traffic fluctuations and random network link damage.In addition,we propose a queue coordination-based network congestion control method,and construct a multi-queue coordination model.Then,we design a multi-agent policy gradient based congestion control algorithm.Experimental results show that,the proposed methods can effectively reduce network packet loss rates,extend network survival time,and improve network reliability.2)Efficient Network Resource Game Method in Non-Cooperative Modes:In response to the challenge of network resource coordination and control in environments with a variety of devices and heterogeneous resources,this paper proposes an efficient network resource game method in non-cooperative modes.Firstly,we propose an incentive mechanismbased device-to-device resource auction game method.We set up a distributed intelligent computing scenario between IoT devices and base station nodes.We construct a double resource auction model between IoT devices and base station nodes.Then,we design an iterative double auction algorithm and an experience-weighted attraction-double auction algorithm.Experimental results show that,the proposed methods can quickly converge to the system maximizing social welfare.In addition,we propose a shared resource pools-based network resource evolutionary game method.We set up a trustworthy data transmission scenario between IoT devices.We construct a resource shared pool model among IoT devices.Then,we design an evolutionary game-based resource pool selection algorithm and a win or learn fast policy hill-climbing based pool selection algorithm.Experimental results show that,the system can quickly converge to the evolutionary stable strategy point.The methods achieve distributed,fair,and efficient resource game and sharing among devices.3)Differentiated Network Business and Resource On-Demand Adaptation Method under Emergency Conditions:In response to the challenge of on-demand adaption between business and resources in scenarios with high variance of business demand and frequent emergencies of business arrive,this paper proposes a differentiated network business and resource on-demand adaptation method under emergency conditions.Firstly,we propose a dynamically mapping method between differentiated businesses and heterogeneous intelligences is propose,and construct a resource graph representation model.Then,we adopt a two-stage node-link mapping mechanism,and design a graph attention network-based network resource intelligent mapping algorithm.Experimental results show that,the methods outperform baseline algorithms in terms of request accept ratio and cost-revenue ratio.In addition,we propose a multi-gateway slicing strategy optimization method for emergency business.We use LoRa network as the case study.Then,we construct a network resource representation model,and design a transfer reinforcement learning based multi-gateway slicing strategy optimization algorithm.Experimental results show that,the proposed methods effectively shorten the subnet strategy generation time and improve the slicing subnet activation speed.