CN-122028078-A - RIS-UAV-based auxiliary Internet of things computing power resource optimization and scheduling method

CN122028078ACN 122028078 ACN122028078 ACN 122028078ACN-122028078-A

Abstract

The invention discloses an RIS-UAV (remote radio system-unmanned aerial vehicle) -assisted internet of things computing power resource optimization and scheduling method, in particular to a cloud-side-end joint resource allocation method assisted by a reconfigurable intelligent surface unmanned aerial vehicle (Reconfigurable Intelligent Surface Unmanned AERIAL VEHICLE, RIS-UAV) under a terahertz (Terahertz, THz) communication link. The method comprehensively utilizes the large bandwidth transmission capability of THz and the enhancement capability of RIS-UAV to shield sensitive and cover limited links in CPN architecture to realize the cooperative configuration of task unloading and calculation/communication resources, and performs joint optimization on task unloading decision, bandwidth Allocation, edge calculation Resource Allocation, RIS phase control, RIS-UAV motion control and the like through Resource Allocation Multi-Agent Soft Actor evaluation (RA-MASAC) strategy of Multi-Agent reinforcement learning, thereby reducing comprehensive consumption of system time delay, energy consumption and the like. Simulation results show that the method has lower total system consumption under the conditions of different equipment numbers, different edge calculation forces, different task complexity and the like, and is obviously superior to a comparison baseline method.

Inventors

LI MENG
PAN KAIWEN
LI QI
LV SUYU
LIU CHANG
AI YUAN

Assignees

北京工业大学

Dates

Publication Date: 20260512
Application Date: 20260212

Claims (7)

1. An RIS-UAV-based auxiliary Internet of things computing power resource optimization and scheduling method comprises the following steps: step one, setting the number of equipment nodes of the Internet of things, the number of nodes of an edge server, cloud server configuration and RIS-UAV (radio service-UAV) number system information, and modeling equipment computing capacity, edge computing capacity, cloud computing capacity and base station available bandwidth infrastructure parameters; step two, a terahertz communication and RIS-UAV reconfigurable intelligent reflection enhancement technology is introduced, a wireless communication process from the Internet of things equipment to an edge server is modeled, equivalent transmission capacity of a direct link and an RIS auxiliary cascade link is described, and communication performance characterization is provided for bandwidth scheduling and phase configuration; Modeling a system task execution process, respectively quantifying task time delay and energy consumption of a task in three modes of local execution, edge unloading execution and cloud unloading execution, and incorporating mobile expenses introduced by RIS-UAV participating in communication enhancement into system comprehensive consumption to form a unified task consumption evaluation index and establishing a joint optimization problem; modeling a multi-agent Markov decision process, setting a state space to reflect the task arrival and resource allowance system state, defining an action space to cover joint decision variables such as unloading, bandwidth, calculation power, phase and track, and constructing a reward function taking the comprehensive consumption of the system as a core; Fifthly, finishing execution strategy and deep neural network training parameter setting by adopting a multi-agent soft actor commentator algorithm; Step six, generating and selecting an optimal joint action under each state according to the strategy network trained in the step five, issuing the action as an optimal execution instruction under the current state to a communication and calculation entity for continuous execution, and rolling and updating to the next state according to environmental feedback until a preset termination condition is reached or task execution is finished.
2. The method for optimizing and scheduling computing power network resources oriented to the Internet of things according to claim 1, wherein the method is characterized by setting the system information of the number of nodes of the equipment of the Internet of things, the number of nodes of an edge server, the configuration of a cloud server and the number of RIS-UAVs, and modeling the computing power of the equipment, the computing power of the edge, the computing power of a cloud and the bandwidth infrastructure parameters for a base station, and comprises the following specific steps: The system comprises an IoT device, an unmanned plane carrying a reconfigurable intelligent surface RIS, an edge computing node and a cloud computing node, wherein the modeled system comprises Individual cells, the set of which is denoted as Each cell is internally provided with 1 base station and 1 edge computing node, each base station adopts terahertz frequency band to communicate with equipment in the cell, the base station is provided with a single antenna, wherein the antenna height of the base station in the nth cell is The location of the base station antenna is expressed as: the base stations in different cells establish communication through wired connection; IoT device aggregation Representation of the device Which is in time slot The position of (2) is expressed as For each device, there may be a task to be performed in each time slot, and the task of device m in time slot t is Computing task information uses triplets Representation of wherein Representing the size of the data necessary to perform the task, Representing the computational resources required for the task, Representing a maximum acceptable execution time for the task; In-system sharing And a stand RIS-UAV for dynamically adjusting the position of the UAV and the phase of a reflection unit mounted on the RIS.
3. The method for optimizing and scheduling computing power network resources oriented to the Internet of things according to claim 2, wherein in the second step, a terahertz communication and RIS-UAV reconfigurable intelligent reflection enhancement technology is introduced, a wireless communication process from an Internet of things device to an edge server is modeled, equivalent transmission capacities of a direct link and an RIS auxiliary cascade link are described, and communication performance characterization is provided for bandwidth scheduling and phase configuration, wherein the method comprises the following steps: the direct-connection terahertz communication link is a direct communication channel between the base station and the equipment, and the equipment With a base station The distance between the antennas is noted as ; First, the Individual base station antenna and apparatus At the moment of time Terahertz direct channel gain is noted as ; The cascade terahertz communication link is a channel formed after information is reflected by the RIS-UAV, and the moment In this case, the coordinates of the RIS-mounted unmanned aerial vehicle are defined as ; If the device Determining to upload its data to the base station for performing its tasks on the edge computing network node ECN, the received signal of the base station being determined jointly by the direct channel condition, the concatenated channel condition, the transmit power and the noise, the uplink transmission rate being recorded as Calculation was performed by shannon's formula.
4. The method for optimizing and scheduling the computing power network resources oriented to the Internet of things according to claim 3, wherein in the third step, modeling is carried out on a system task execution process, respectively quantifying task time delay and energy consumption of the task in three modes of local execution, edge unloading execution and cloud unloading execution, and incorporating mobile expenses introduced by RIS-UAV participation communication enhancement into system comprehensive consumption to form a unified task consumption evaluation index, and establishing a joint optimization problem; Equipment for setting up Executing tasks locally Time consuming as follows The energy consumption of locally executing the task is recorded as Wherein, the method comprises the steps of, Calculating time and time slots for tasks Device at the beginning The sum of the residual execution time of the task is expressed as follows: ; In the formula, Representation device Is used for the calculation of the calculation capacity of (a); Representing time slots Device at the beginning The residual execution time of the task; Device m performs tasks locally Is related to the hardware characteristics of the device m, expressed as: ; In the formula, Representing the energy consumption coefficient of the device m, the parameter being related to the device Hardware characteristic correlation of (a) to execute a task The overall consumption of (2) is: ; In the formula, Is a time delay preference coefficient; The device offloads its task to any ECN through the communication network, the device m offloads the task Offloading to a target edge computing node ECNk; if the forwarding is required to the edge node of the non-affiliated base station, the time delay and the energy consumption are as follows: ; ; Wherein, the For base stations And (3) with The transmission rate of the wired link between the two; For base stations Is a forwarding power consumption of (1); For base stations Received power consumption of (a) After being forwarded to the target edge node, the time delay and energy consumption of the target edge node for calculating the task are as follows: ; In the formula, For time slots Tasks The number of CPU cycles required; for time slots Edge computing node Distribution to devices Is a computing resource of (a); Is an edge node Is a coefficient of energy consumption of (2); is a device Idle power during waiting for the calculation result; the comprehensive consumption of the edge computing task is as follows: In the formula, Is a device In time slot Unloading the total delay to edge execution; Is the corresponding total energy consumption; comprehensive consumption for edge execution; When the device In time slot When the task is unloaded to the cloud end for execution, the task completion process comprises uplink transmission from the equipment to the base station, feedback transmission from the base station to the cloud server and cloud end calculation execution; When the equipment transmits uplink to the base station, the time delay and the energy consumption are respectively recorded as And (3) with : After receiving the task data of the equipment, the base station forwards the data to a cloud server, and calculates time delay and energy consumption; the cloud end calculates and returns the task, and calculates time delay and energy consumption; The overall consumption of the cloud computing task is expressed as a weighted sum of the total latency and the total energy consumption.
5. The method for optimizing and scheduling resources of an Internet of things-oriented computing power network of claim 4, wherein in the fourth step, a multi-agent Markov decision process is modeled, a state space is set to reflect task arrival and resource allowance system states, an action space is defined to cover unloading, bandwidth, computing power, phase and track joint decision variables, and a reward function with system comprehensive consumption as a core is constructed, and the method is specifically as follows: Setting a state space, an action space and a reward function of an algorithm by combining an environment and an optimization target, and modeling the process as a Markov process due to the Markov characteristic of the task unloading calculation process; five cooperative agents are arranged in total, and each agent is divided into the following steps: Track agent-based adjustment of motion vector matrix of unmanned aerial vehicle based on observed system state The optimization of the unmanned aerial vehicle position is realized; Unloading agent generating an unloading decision matrix based on observed system state Determining a task processing mode, and selecting an optimal scheme from the three modes of local calculation, edge calculation and cloud calculation according to task requirements; bandwidth agent-allocation of bandwidth resource matrix based on observed system state On the premise of not exceeding the upper limit of the bandwidth of the base station, reasonably distributing bandwidth resources; phase agent-obtaining a reflective phase matrix for adjusting the intelligent supersurface based on observed system conditions The cascade channel gain is improved; an agent for computing power allocates computing power resources for each task offloaded to an edge node based on observed system state The efficiency is maximized, and the task timeout rate is as low as possible when the tasks are more.
6. The method for optimizing and scheduling computing power network resources oriented to the Internet of things according to claim 5, wherein in the fifth step, an execution strategy and a deep neural network training parameter setting are completed by adopting a multi-agent soft actor commentator algorithm, and the method is specifically as follows: Training the combined resource allocation strategy by adopting a multi-agent soft actor reviewer MASAC algorithm according to the state space, the action space and the rewarding function constructed in the step four, and completing the structure and training parameter setting of the strategy network actor and the value network critic; The network is divided into two types, namely a commentator network and an actor network, wherein the commentator network consists of two Q networks, adopts minimum mean square error loss and gradient descent, and is updated at the end of each training step; the actor network is updated by maximizing the Q value.
7. The method for optimizing and scheduling resources of an Internet of things-oriented computing network according to claim 6, wherein in the step six, according to the policy network trained in the step five, an optimal joint action is generated and selected in each state, the action is issued to a communication and computing entity as an optimal execution instruction in a current state for continuous execution, and the action is updated to a next state according to environmental feedback rolling until a preset termination condition is reached or task execution is ended.

Description

RIS-UAV-based auxiliary Internet of things computing power resource optimization and scheduling method Technical Field The invention relates to an RIS-UAV (remote radio system-unmanned aerial vehicle) -assisted internet of things computing power resource optimization and scheduling method, in particular to a cloud-side-end joint resource allocation method assisted by a reconfigurable intelligent surface unmanned aerial vehicle (Reconfigurable Intelligent Surface Unmanned AERIAL VEHICLE, RIS-UAV) under a terahertz (Terahertz, THz) communication link, and belongs to the technical field of communication networks. Background In recent years, the internet of things has a large-scale expansion trend, the number of terminals is rapidly increased, and application scenes are continuously popularized from intelligent transportation, remote medical treatment, intelligent agriculture and the like. The new applications generally have stricter requirements for quality of service (Quality of Service, qoS), especially for communication capacity and computing power, whereas conventional "relying only on terminal local computing" approaches often have difficulty in meeting the processing requirements of low latency tasks. In order to reduce the energy consumption of a terminal and improve the task processing efficiency, the computing network is used as a novel network architecture, heterogeneous computing resources are connected in a networked mode, so that the heterogeneous computing resources can be cooperated and flexibly called, and the task needing stronger computing power is supported to be unloaded to an edge or cloud computing node for execution. However, existing CPN related researches often focus on traditional communication means such as millimeter waves, and in a scenario facing high-density access and large data task transmission, the communication efficiency still has difficulty in meeting the continuously growing demands. Terahertz communication utilizes higher frequency band resources, has the potential of significantly improving transmission performance, and is considered as one of important candidate technologies meeting the future high-throughput requirements. At the same time, however, THz base stations have less coverage and are highly sensitive to occlusion, resulting in limited link reliability and coverage available, thereby affecting the continuous service capability of THz-based CPN systems in dynamic scenarios. The reconfigurable intelligent surface is composed of a large number of reflection units capable of independently adjusting phase shift, and can improve propagation environment and transmission rate by adjusting reflection phase, and is considered as one of important means for overcoming THz shielding and coverage problems. The existing work focuses on the fixed deployed RIS, and the fixed RIS is difficult to adapt to the characteristics of a time-varying system such as terminal movement, service dynamic change and the like, so that the requirements of a CPN system based on THz communication on the dynamic property and the real-time property are difficult to meet. With the development of Unmanned AERIAL VEHICLE, UAV technology, RIS is carried on UAVs to form RIS-UAVs, the transmission performance can be further enhanced by dynamically adjusting the spatial position of the RIS and combining with phase control, and a new technical path is provided for solving the problem of shielding and coverage limitation in THz scenes. Based on this, fusing RIS-UAV, THz communication and CPN architecture is expected to promote network transmission and task computing performance simultaneously to adapt to evolving real-time and complexity requirements in the IoT scenario. However, in the above fusion system, variables such as task offloading decision, bandwidth allocation, edge computing resource allocation, RIS reflection phase control, RIS-UAV trajectory control, etc. are strongly coupled to each other, and the decision space is high-dimensional, non-convex and limited by multiple types of constraints, so that it is difficult to implement stable and real-time joint decision in dynamic environments by using conventional analytic optimization or step-by-step strategies. Under the background, multi-agent reinforcement learning (Multi-Agent Reinforcement Learning, MARL) is considered to be a feasible solution to the problem of joint resource allocation due to the fact that the Multi-agent reinforcement learning has Multi-agent distributed decision capability and can learn effective strategies in a complex high-dimensional non-convex optimization space. Disclosure of Invention The invention aims at a dynamic and changeable computing power network system of the Internet of things, which is characterized in that the system has large terminal scale, the task arrival and the terminal position change along with time, the communication and the computing resources are heterogeneous and limited, and a cloud-side-end th