Intelligent reflecting surface assistance-based task unloading and resource allocation method for unmanned aerial vehicle mobile edge computing network system

Abstract

An intelligent reflecting surface assistance-based task unloading and resource allocation method for an unmanned aerial vehicle mobile edge computing network system, comprising: constructing an intelligent reflecting surface-assisted unmanned aerial vehicle mobile edge computing network system model; constructing an intelligent reflecting surface-assisted unmanned aerial vehicle mobile edge computing network communication model ; according to the communication model, obtaining a time delay model and an energy consumption model; establishing a multivariable joint optimization model so as to minimize the energy consumption of an unmanned aerial vehicle; and using a deep reinforcement learning method to solve the objective optimization model. By means of intelligent reflecting surface assistance, joint optimized unloading decision and resource allocation, the present application effectively reduces the energy consumption of the unmanned aerial vehicle and improves the performance of the unmanned aerial vehicle network system.

Description

A task offloading resource allocation method for UAV mobile edge computing network system based on intelligent reflective surface assistance

[0001] This application claims priority to the Chinese patent application filed with the Patent Office of China on July 24, 2023, with application number 2023109132575 and invention name “A task offloading resource allocation method for a UAV mobile edge computing network system based on intelligent reflective surface assistance”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present application belongs to the technical field of drone task offloading and resource allocation, and specifically relates to a task offloading resource allocation method for a drone mobile edge computing network system based on intelligent reflective surface assistance. Background Art

[0003] Mobile edge computing networks are considered a promising key technology for improving the quality of service for computationally intensive and latency-critical tasks. Mobile edge computing networks equipped with drones can leverage the flexible deployment of drones to provide extended communication coverage and reliable connectivity to mobile users in a timely manner, especially when ground-based communication infrastructure is destroyed or when a large number of users surge. However, due to the limited energy and computing resources of drones, deploying more drones in mobile edge computing networks is costly and power-intensive. Fortunately, a new technology called smart reflective surfaces offers an energy-efficient alternative to increase network capacity.

[0004] Smart reflective surfaces are low-cost, energy-efficient, high-gain metasurfaces that can reconfigure the wireless propagation environment, thereby improving communication capabilities, reducing energy consumption in mobile edge computing networks, and increasing transmission reliability in MEC networks and expanding wireless coverage.

[0005] Users share the computing and communication resources of a drone. Due to the limited resources and energy of a drone, users' computing tasks cannot be fully completed simultaneously on the drone. Therefore, it is necessary to jointly optimize task offloading and resource allocation. However, the joint optimization problem is a complex non-convex problem, and traditional methods have difficulty in quickly obtaining the optimal solution of the multidimensional decision matrix. Fortunately, deep reinforcement learning has recently been used as an emerging tool to effectively solve various optimization problems in wireless communication networks.

[0006] Summary of the Invention

[0007] To solve the problems existing in the above-mentioned prior art, this application proposes a task offloading resource allocation method for a UAV mobile edge computing network system based on intelligent reflective surface assistance, which includes:

[0008] S1: Construct a UAV mobile edge computing network system model assisted by intelligent reflective surfaces;

[0009] S2: Construct communication model, delay model and energy consumption model based on the UAV mobile edge computing network system model assisted by intelligent reflective surface;

[0010] S3: Construct the minimum UAV energy consumption problem based on the communication model, delay model and energy consumption model;

[0011] S4: Optimize the minimum UAV energy consumption problem based on deep reinforcement learning methods; solve the optimized problem to obtain the optimal task offloading decision and resource allocation results.

[0012] Preferably, constructing a communication model includes: obtaining the channel gain of the link between the user and the drone based on the position information of the user and the drone; obtaining the channel gain between the link between the drone and the smart reflecting surface based on the position information of the drone and the smart reflecting surface; obtaining the channel gain between the link between the network access point and the smart reflecting surface based on the position information of the smart reflecting surface and the network access point; obtaining the channel gain between the link between the drone and the network access point based on the position information of the network access point and the drone; and obtaining the phase shift matrix of the smart reflecting surface based on the amplitude coefficient and phase shift coefficient of the smart reflecting surface element.

[0013] Furthermore, the expression of the communication model is: Θ=[θ1(n),θ2(n),...,θ M (n)] T

[0014] Preferably, the expression of the delay model is:

[0015] Preferably, the energy consumption model expression is:

[0016] Preferably, the expression of the minimum UAV energy consumption problem is:

[0017] Preferably, the expression of the minimum UAV energy consumption optimization problem is:

[0018] Beneficial effects of this application:

[0019] This application proposes a method for joint task offloading and resource allocation of a UAV mobile edge computing network system assisted by an intelligent reflective surface to reduce the energy consumption of UAVs. This application minimizes the energy consumption of UAVs by jointly optimizing UAV computing resource allocation and UAV offloading decisions, intelligent reflective surface phase shifting, and transmission power allocation. By solving the problem of minimum UAV energy consumption, the optimal task offloading and resource allocation scheme is obtained; compared with the existing baseline method, the method proposed in this application effectively saves UAV energy consumption and improves the performance of the system. This application uses UAVs as aerial base stations and aerial relay stations, and considers applying intelligent reflective surfaces to UAV network systems, rationally scheduling UAV resources, and making real-time optimal offloading decisions and resource allocation schemes. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] FIG1 is a flow chart of a method for task offloading and resource allocation of UAV mobile edge computing assisted by intelligent reflective surfaces in this application;

[0021] FIG2 is a schematic diagram of a UAV mobile edge computing network system model assisted by a smart reflective surface in this application;

[0022] Figure 3 shows the relationship between the total energy consumption of drones and the number of users;

[0023] Figure 4 shows the relationship between the total delay of user computing task processing and the number of users. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0025] A task offloading resource allocation method for a UAV mobile edge computing network system based on intelligent reflective surface assistance, as shown in FIG1 , includes:

[0026] S1: Construct a UAV mobile edge computing network system model assisted by intelligent reflective surfaces;

[0027] S2: Construct communication model, delay model and energy consumption model based on the UAV mobile edge computing network system model assisted by intelligent reflective surface;

[0028] S3: Construct the minimum UAV energy consumption problem based on the communication model, delay model and energy consumption model;

[0029] S4: Optimize the minimum UAV energy consumption problem based on deep reinforcement learning methods; solve the optimized problem to obtain the optimal task offloading decision and resource allocation results.

[0030] As shown in Figure 2, the proposed smart reflective surface-assisted drone mobile edge computing network system model includes a drone, a smart reflective panel, and multiple users. The drone, smart reflective surface, and network access point are located at [600, 0, 50], [400, 300, 30], and [0, 600, 0], respectively. Users are randomly distributed within a circle with a radius of 100 meters centered at [0, 600, 0].

[0031] Building the communication model involves: obtaining the channel gain of the link between the user and the drone based on their respective location information; obtaining the channel gain between the drone and the smart reflector based on their respective location information; obtaining the channel gain between the network access point and the smart reflector based on their respective location information; obtaining the channel gain between the drone and the network access point based on their respective location information; and obtaining the phase shift matrix of the smart reflector based on the amplitude coefficients and phase shift coefficients of its components. The communication model is expressed as: Θ = [θ1(n),θ2(n),...,θ M (n)] T

[0032] in, is the reflection coefficient of the mth reflection unit of RIS in the nth time slot, where φ m [n]∈[0,2π) and λ m ∈[0,1] represents the phase shift and amplitude reflection coefficient respectively; G k (n), are the channel gains from user k to UAV, from UAV to RIS, from RIS to AP, and from UAV to AP in the nth time slot, respectively; β0 represents the path loss with a reference distance of 1 meter; d is the element distance of RIS, is the cosine of the angle of arrival (AOA) of the signal; β1 is the Rician factor, which determines the LoS component is the cosine of the angle of departure (AOD); the NLoS component is a Gaussian distribution with zero mean and unit variance of 1.

[0033] The process of constructing the delay model includes: obtaining the task offloading delay based on the channel gain of the transmission link from the user to the drone; obtaining the drone-side task calculation delay based on the drone's computing resources; and obtaining the task relay transmission delay based on the transmission link between the drone and the network access point. The delay model is expressed as:

[0034] in, is the offloading delay of the kth user to offload the computing task to the UAV in the nth time slot; is the computational delay of the computational task of user K at the UAV end in the nth time slot; Calculate the relay transmission delay of the task from the UAV to user k in the nth time slot; The transmission rate of user k offloading tasks to the UAV; α k (n) is the drone unloading decision; L k (n) represents the total length of the computation task, C k (n) represents the number of CPU cycles required to calculate 1 bit; represents the computing resources allocated by the UAV to user k in the nth time slot.

[0035] The process of building the energy consumption model includes: calculating the delay on the UAV side based on the task data to obtain the computing energy consumption; and calculating the relay transmission delay of the task based on the UAV relay transmission user to obtain the relay transmission energy consumption. The energy consumption model expression is:

[0036] in, represents the computing energy consumption of the nth time slot task data on the drone side; β is the energy consumption of task data relay transmission; UAV is the effective switched capacitance; represents the computing resources allocated to user k by the UAV in the nth time slot. The process of constructing the minimum UAV energy consumption problem includes: constructing the minimum UAV energy consumption objective function based on the UAV computing energy consumption and the UAV relay transmission energy consumption; constructing the constraints based on the RIS phase shift matrix, UAV transmission power allocation, UAV task transmission energy consumption, and UAV computing resource allocation; and constructing the minimum UAV energy consumption problem based on the minimum UAV energy consumption objective function and the constraints. The UAV energy consumption formula can be defined as:

[0037] in, represents the computing energy consumption of the nth time slot task data on the drone side, α k (n) making decisions for drone unloading; Energy consumption for task data relay transmission.

[0038] The expression of the minimum UAV energy consumption optimization problem is:

[0039] Among them, C1 represents the offloading decision of the UAV; C2, C3, C4 and C5 represent the constraints of the UAV computing resources and relay transmission power, respectively; C6 is to ensure that the tasks calculated on the UAV side in each time slot can be completely calculated; C7 represents the processing delay constraint of the computing task; C8 is the phase shift constraint of the reflective element of each smart reflector.

[0040] Preferably, the drone interacts with the environment to obtain state information and finds the optimal strategy to maximize the cumulative reward. The state, action, and reward of the formulated MDP are defined as follows:

[0041] (1) State space: The state space of the nth time slot consists of the task data bits L(n) = {L1(n), L2(n), ..., L K (n)} and the task processing delay of the previous time slot T(n-1) = {T1(n-1), T2(n-1), ..., T(n-1)}, that is: S(n) = {L(n), T(n-1)}.

[0042] (2) Action space: contains task offloading decisions α(n) = {α1(n), α2(n), …, α K (n)}, UAV computing resource allocation f(n)={f1(n),f2(n),…,f K (n)}, UAV relay transmission power p(n)={p1(n),p2(n),…,p K (n)} and the smart reflector phase shift matrix φ(n)={φ1(n),φ2(n),…,φ M (n)}. The action space can be expressed as: A n ={p(n),α(n),f(n),Θ}

[0043] (3) Reward function: The reward obtained by taking action a in the state of the nth time slot can be defined as:

[0044] Where z is the corresponding penalty, which is a constant greater than zero.

[0045] The DDQN algorithm process includes: the drone interacts with the environment to obtain the system state s n , find the action a corresponding to the maximum value function value through the learning network n , perform action a n After that, the state will enter the next state s n+1 , and will also receive rewards r from environmental feedback n+1 . {s n ,a n ,r n+1 ,s n+1 Such an experience will be stored in the experience pool D, and then d experiences will be randomly sampled to train the neural network. The purpose of training the neural network is to minimize the error function L(ω), which can be expressed as:

[0046] Among them, Q(s n ,a n ,ω) is the Q-value predicted by Q-Network, y i The reward obtained by the target network after the network transfer:

[0047] Finally, the parameters ω of Q-Network are updated through gradient descent backpropagation.

[0048] Figure 3 shows the total energy consumption of the drone for four different algorithms as the number of users increases. The results show that the total energy consumption of all four algorithms increases with the number of users. Compared to the full-computation, full-relay with smart reflective surfaces, and full-relay without smart reflective surfaces, the proposed DTORA algorithm reduces energy consumption by 16%, 30%, and 49%, respectively. The advantage of this algorithm is that it coordinates computational and communication resources to make appropriate offloading decisions based on different tasks. Furthermore, we note that the total energy consumption of the full-relay with smart reflective surfaces algorithm is 19% less than that of the algorithm without smart reflective surfaces. This demonstrates that the assistance of smart reflective surfaces can effectively reduce drone energy consumption.

[0049] Figure 4 analyzes the relationship between the number of users and latency. The results show that the total latency of all algorithms increases with the number of users, with the latency of the full-computation algorithm, in particular, showing a sharp upward trend. This is due to the limited computing resources of the drone. As the number of users increases, the computing resources allocated to each user gradually decrease. This results in computational tasks requiring more processing time, which increases latency. Furthermore, the latency rate of this algorithm is lower than that of the full-relay algorithm without intelligent reflectors, and slightly higher than that of the full-relay algorithm with intelligent reflectors. While this solution incurs a certain cost in terms of latency, the drone's energy consumption is minimized within an acceptable latency range.

[0050] The above embodiments further illustrate the purpose, technical solutions and advantages of the present application. It should be understood that the above embodiments are only preferred implementation methods of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made to the present application within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A task offloading resource allocation method for a UAV mobile edge computing network system based on intelligent reflective surface assistance, characterized in that: include: S1: Construct a UAV mobile edge computing network system model assisted by intelligent reflective surfaces; S2: Construct communication model, delay model and energy consumption model based on the UAV mobile edge computing network system model assisted by intelligent reflective surface; S3: Construct the minimum UAV energy consumption problem based on the communication model, delay model and energy consumption model; S4: Optimize the minimum UAV energy consumption problem based on the deep reinforcement learning method; solve the optimized problem to obtain the optimal task offloading decision and resource allocation results.

2. According to claim 1, a method for allocating task offloading resources of a UAV mobile edge computing network system based on intelligent reflective surface assistance is characterized in that: The model of the UAV mobile edge computing network system assisted by the intelligent reflective surface includes: a UAV, an intelligent reflective panel and a user terminal; wherein the user terminal is used to generate computing tasks; and the computing tasks are uploaded to the UAV through the intelligent reflective panel.

3. According to claim 1, a method for allocating task offloading resources of a UAV mobile edge computing network system based on intelligent reflective surface assistance is characterized in that: Building a communication model includes: obtaining the channel gain of the link between the user and the UAV according to the position information of the user and the UAV; obtaining the channel gain of the link between the UAV and the intelligent reflecting surface according to the position information of the UAV and the intelligent reflecting surface; obtaining the channel gain of the link between the network access point and the intelligent reflecting surface according to the position information of the intelligent reflecting surface and the network access point; obtaining the channel gain of the link between the UAV and the network access point according to the position information of the network access point and the UAV; obtaining the phase shift matrix of the intelligent reflecting surface according to the amplitude coefficient and phase shift coefficient of the intelligent reflecting surface element.

4. According to claim 3, a method for allocating task offloading resources of a UAV mobile edge computing network system based on intelligent reflective surface assistance is characterized in that: The expression of the communication model is: θ=[θ1(n),θ2(n),...,θ M (n)] T Among them, θ m (n) is the reflection coefficient of the mth reflection unit of RIS in the nth time slot; G k (n) is the channel gain from user k to UAV in the nth time slot, is the channel gain from UAV to RIS in the nth time slot, is the channel gain from RIS to AP in the nth time slot, is the channel gain from UAV to AP in the nth time slot; β0 represents the path loss with a reference distance of 1 meter; d is the element distance of RIS, is the cosine value of the signal arrival angle; β1 is the Rician factor, is the LoS component; the NLoS component is a Gaussian distribution with zero mean and unit variance of 1.

5. According to claim 1, a method for allocating task offloading resources of a UAV mobile edge computing network system based on intelligent reflective surface assistance is characterized in that: The expression of the delay model is: in, is the offloading delay of the kth user to offload the computing task to the UAV in the nth time slot; is the computational delay of the computational task of user K at the UAV end in the nth time slot; Calculate the relay transmission delay of the task from the drone to user k in the nth time slot; is the transmission rate of user k offloading tasks to the UAV; α k (n) Unloading the solution for the drone Policy; L k (n) represents the total length of the computation task bit, C k (n) represents the number of CPU cycles required to calculate 1 bit; represents the computing resources allocated by the UAV to user k in the nth time slot; Represents the transmission power of the drone relay in the nth time slot.

6. The task offloading resource allocation method of the UAV mobile edge computing network system based on intelligent reflective surface assistance according to claim 1 is characterized in that: The energy consumption model expression is: in, It represents the computing energy consumption of the nth time slot task data on the UAV side; β is the energy consumption of task data relay transmission; uAV is the effective switched capacitance; It represents the computing resources allocated by the UAV to user k in the nth time slot.

7. The task offloading resource allocation method of the UAV mobile edge computing network system based on intelligent reflective surface assistance according to claim 1 is characterized in that: The expression of the minimum UAV energy consumption problem is: in, represents the computing energy consumption of the nth time slot task data on the drone side, α k (n) making decisions for drone offloading; Energy consumption for task data relay transmission.

8. The task offloading resource allocation method of the UAV mobile edge computing network system based on intelligent reflective surface assistance according to claim 1 is characterized in that: The expression of the minimum UAV energy consumption optimization problem is: Among them, C1 represents the unloading decision of the drone; C2, C3, C4 and C5 represent the constraint range of the drone computing resources and relay transmission power respectively; C6 is to ensure that the tasks calculated on the drone side in each time slot can be completely calculated; C7 represents the processing delay constraint of the computing task; C8 is the phase shift constraint of each smart reflective surface reflective element.

9. The method for allocating task offloading resources of a UAV mobile edge computing network system based on intelligent reflective surface assistance according to claim 1 is characterized in that: The process of solving the optimized problem includes: modeling the optimization problem with a Markov decision process, determining the state space, action space, and reward function; collecting data sets, training the network using a deep reinforcement learning algorithm, and updating the network using a gradient descent method.

Images