Research On The Timeliness Of Information-based Decision-making For The Internet Of Things

IoT has spawned a lot of applications with stringent delay requirements,such as the Internet of Vehicles(Io V),the smart agriculture,and the health monitoring systems.These applications need high demands on the timeliness of the received information.How to measure the timeliness of the received information has become a hot research topic at present.The Age of Information(AoI)was proposed to make up for the deficiency of delay and throughput in measuring the timeliness of the received information,and has been studied from the queuing model,the queuing strategy,the multi-source and the multi-path.For several important moments,the timeliness of information can be accurately represented by Age upon Decisions(AuD)rather than AoI.This paper studies the timeliness of the information used for making decisions in the IoT network through model construction,model optimization,numerical analysis and theoretical simulation.The innovative work of the thesis mainly includes the following aspects:1.Model the updata-and-decide system,analyze the sample path of AuD,calculate and compare the average AuD of each system for different service processes,i.e.,the uniform service process,the Poisson service process,and the periodic service process.The research proves that the average AuD of the system is the largest when the service time(e.g.,transmission time)is exponentially distributed and would be the smallest if it is deterministic.For different decision processes,i.e.,the Poisson decision process and the periodic decision process,the updateand-decide systems with the same arrival process and the same service process are studied.The research has proved that when the decisions are made periodically,the average AuD of the system is larger than,and decreases with decision rate to,the average AuD of the corresponding system with Poisson decision intervals.Finally,the Monte Carlo simulation verifies the correctness of the theoretical analysis results.2.For different service discipline,i.e.,the FCFS discipline and the LCFS discipline with preemption,study the performance of the same update-and-decide system under different service discipline.Research shows that when the LCFS discipline with preemption is used,the performance of the update-and-decide system with Poisson service process is optimal,while the update-and-decide system with deterministic service process performs the worst.When the update arrival rate is relatively large,the performance of the preemptive LCFS-based updateand-decide system is significantly better than the corresponding FCFS-based update-and-decide system.Finally,the Monte Carlo simulation verifies the correctness of the theoretical analysis results.3.The missing probability is proposed to indicate the utilization rate of received updates.Research shows that when the decision process is a Poisson decision process,although increasing the decision rate cannot reduce the average AuD of the update-and-decide system at the decision epochs,increasing the decision rate can help reduce the missing probability.At the same time,the update-and-decide system with a periodic service process has the lowest missing probability,that is,more update packets are used for making decision,which reflects the efficiency of the system.4.Based on the node wake-up control and the energy harvesting theory,a multi-source sensor network model is constructed by using the Markov decision process.Construct the weighted sum of the average AuD and the information distortion as the objective function.Obtain the optimal scheduling strategy through online power control and energy scheduling to minimize the objective function.Research shows that the optimal transmission power and the decision process scheduling are a monotonic and bi-valued function of current AuD and distortion,and simulation analysis proves the correctness of the results.