Information Freshness Optimization For Correlated Data In Energy Harvesting Enabled IoT Networks

For the real-time applications of the Internet of Things(Io T),it is necessary to improve the timeliness of sensors' perception and transmission to the environment to ensure the freshness of the information obtained by data consumers and the effectiveness of related decisions.However,on the one hand,the limited energy of sensors restricts the frequency of collecting and updating environmental information;on the other hand,the correlation of data generated by different sensors brings new challenges to the design of dynamic data update strategies and the optimization of information freshness.At present,Energy Harvesting(EH)technology is considered to be an emerging technology that can effectively alleviate the energy limitation of sensors.However,how to design an efficient data dynamic update strategy and optimize the freshness of information in the Io T system while considering the correlation of the data generated by the sensors,while taking into account the causality and randomness introduced by the energy arrival and consumption process,which still is an important issue that needs to be addressed urgently.In view of this,for the energy harvesting Io T system,this thesis uses the Age of Correlated Information(Ao CI)as the optimization indicator,and uses the Deep Reinforcement Learning(DRL)technology to study the data dynamics with the best information freshness update strategy.The specific research contents are as follows:(1)A data dynamic update strategy based on perfect information is designed,and its effectiveness in improving information freshness is verified.First,when the data fusion center can know the real state of the system in real-time(that is,grasp the perfect information),the dynamic update of sensor data in the Io T system is modeled as a dynamic optimization problem with the goal of minimizing the long-term average Ao CI.Second,The optimization problem is modeled as a Markov Decision Process,and the invalid action elimination mechanism is embedded into the standard Deep Q-Network(DQN)algorithm by considering the coupling relationship between states and valid actions.A DRL-based state update algorithm is used to solve the model.Finally,simulation experiments are used to verify that the designed algorithm has better convergence and stability,and can effectively reduce Ao CI,that is,effectively improve the freshness of information.(2)A data dynamic update strategy based on imperfect information is designed,and its effectiveness in improving information freshness is verified.For practical Io T systems,the data fusion center usually cannot know the real-time power of sensors in real-time(i.e.,have imperfect information),in this case,first,the problem of dynamic update of sensor data is modeled with the goal of minimizing the long-term average Ao CI.Second,in order to cope with the challenges posed by energy causal constraints,unknown environmental dynamics,unknown real energy states of sensors,and large-scale discrete action spaces,the dynamic update optimization problem is modeled as a partially observable Markov decision process,and further designed a DRL-based dynamic update algorithm to solve it.The algorithm combines the advantages of the DRL algorithm Soft Actor-Critic(SAC)and long short-term memory network technology.At the same time,in order to avoid the challenges brought by large-scale discrete action spaces,an action decomposition,and mapping mechanism is introduced.Finally,through a large number of simulation experiments,it is verified that the designed algorithm has better convergence,scalability,and stability,and can effectively reduce Ao CI,that is,effectively improve the freshness of information.