Low-orbit satellite network multi-dimensional resource scheduling method for IoT task offloading

Abstract

The application discloses a low-orbit satellite network multi-dimensional resource scheduling method for IoT task offloading, and relates to the technical field of space-air communication networks. Firstly, satellite ephemeris is calculated according to constellation information of the satellite network, and then visibility and distance of the satellite relative to the Internet of Things device at the observation time are obtained, and communication channel gain and communication time constraints are calculated according to the relative position and distance of the satellite and the Internet of Things device. Based on the communication channel gain and communication time constraints obtained by the above calculation and the network to be deployed service demand condition, a heuristic algorithm is used to iteratively solve the joint task offloading and resource allocation problem. Finally, network resource allocation is carried out according to the decision of the joint task offloading and resource allocation problem. The method can ensure that a suboptimal solution of the joint task offloading and resource allocation optimization problem of the satellite Internet of Things edge computing space-ground integrated network is obtained within a polynomial calculation time of the algorithm, and has the effect of improving the offloading decision and resource allocation efficiency of the space-ground integrated network system.

Description

Technical Field

[0001] This invention relates to the field of aerospace communication network technology, specifically to a multi-dimensional resource scheduling method for low-Earth orbit satellite networks oriented towards IoT task offloading. Background Technology

[0002] With the continuous development of technology and the expansion of application scenarios, the demand for space-air communication networks is growing. Edge computing networks, as a new type of communication network architecture, are attracting increasing attention due to their characteristics of resource dispersion and strong computing power. Space-air converged network systems, which integrate the characteristics of space-air communication networks and edge computing networks, have become a popular research area. In a typical scenario of this system, satellites and IoT devices jointly form an edge computing network, which provides task offloading capabilities for IoT devices. Figure 1 This demonstrates a typical satellite IoT network architecture. In this architecture, IoT devices can choose to offload their services to satellites with computing capabilities based on mission efficiency.

[0003] Academic research has extensively studied task offloading and resource allocation in edge computing networks, establishing optimization problems with energy consumption, latency, and network performance of user devices as optimization objectives. Satellite IoT, building upon edge computing networks, deploys server nodes with computing capabilities onto low-Earth orbit (LEO) satellites. However, due to the time-varying topology of LEO satellite networks, the impact of link changes on communication needs to be considered when designing optimization problems. While existing papers have investigated offloading and resource allocation in satellite IoT edge computing networks, modeling the time-varying characteristics of communication links in dynamic LEO satellite network topologies is still lacking. To more accurately describe the characteristics of actual satellite networks, time-slotting of the system is necessary, and communication performance needs to be analyzed based on the visible link status of network nodes in each time slot.

[0004] For task offloading and network resource allocation decisions in satellite IoT networks, a joint task offloading and resource allocation problem needs to be constructed first. This problem is a mixed-integer linear programming problem, and its closed-form solution cannot be obtained. To solve this problem, it needs to be decomposed into independent subproblems, and algorithms need to be designed to solve them separately.

[0005] On the other hand, in space-ground integrated networks, resource allocation and scheduling become a complex and challenging problem due to the constantly changing locations and numbers of satellites and IoT devices. Traditional resource allocation methods, such as static or random allocation, cannot effectively handle such changes. Therefore, resource allocation methods that address dynamic changes in network topology are needed to solve this problem. The exhaustive offloading decision method can obtain the optimal solution for network resource allocation, but its computational complexity is exponential, making it impractical. Greedy algorithms targeting channel gain parameters can complete the decision within constant computational complexity, but their allocation performance is poor. Therefore, it is necessary to design an algorithm with acceptable computational complexity and suboptimal performance to solve the resource sharing problem in satellite IoT networks. Summary of the Invention

[0006] In view of this, the present invention provides a multi-dimensional resource scheduling method for low-Earth orbit satellite networks for IoT task offloading, which can solve the resource sharing problem of satellite IoT edge computing networks and the resource allocation problem in the above-mentioned network scenarios.

[0007] To achieve the above objectives, the technical solution of this invention is: a multi-dimensional resource scheduling method for low-Earth orbit satellite networks oriented towards IoT task offloading, wherein nodes in the low-Earth orbit satellite network include satellite nodes and IoT device nodes; wherein the initial time slot is set to 1, and the following steps are performed in each time slot:

[0008] Step 1: Initialize the current time slot network state and construct the satellite server computing power matrix, user local computing power matrix, task characteristics, user priority parameters, user optimization preference parameters, user maximum uplink transmission power, system noise variance, system transmission bandwidth, and chip power consumption coefficient.

[0009] Step 2: Collect satellite constellation information and location information of IoT devices with mission access requirements in the network, and calculate the visibility and distance between nodes based on the relative positions of satellites and IoT devices;

[0010] Furthermore, the communication channel gain and communication link hold time between nodes are calculated using the free space attenuation equation based on the distance between nodes.

[0011] Step 3: Based on the network user's task computation volume, local computing resources, and chip power consumption parameters, calculate the user's local computing time and local computing power consumption, and calculate the task computation characteristics.

[0012] Step 4: Generate the initial unloading decision, and calculate the objective function value, resource allocation, and uplink power allocation based on the initial unloading decision using the resource allocation algorithm. Set the initial unloading decision as the current unloading decision and set its corresponding objective function value as the current objective function value.

[0013] Step 5: Based on the current unloading decision, generate a set of neighborhood solutions for the unloading decision; traverse all neighborhood solutions for the unloading decision, and allocate resources based on the neighborhood solutions for the unloading decision to obtain a set of objective functions for the neighborhood solutions.

[0014] Step Six: Compare the objective function set of neighboring solutions with the current objective function value. If there exists a neighboring solution A whose objective function is greater than the current objective function value, replace the current unloading decision with the unloading decision corresponding to neighboring solution A and return to Step Five. Otherwise, if the objective functions of all neighboring solutions are not greater than the current objective function value, use the current unloading decision as the unloading decision for the current time slot, allocate resources in the current time slot according to the unloading decision of the current time slot, and then statistically analyze the service energy consumption and service latency of the current time slot.

[0015] Further, in step one, the current time-slot network state is initialized, and the following parameters are constructed: satellite server computing power matrix, user local computing power matrix, task characteristics, user priority parameters, user optimization preference parameters, user maximum uplink transmission power, system noise variance, system transmission bandwidth, and chip power consumption coefficient. Specifically, these parameters include the following:

[0016] The satellite server computing power matrix is ​​as follows Where S is the total number of satellite nodes in the low-Earth orbit satellite network. The server chip frequency for satellite nodes 1 to S;

[0017] The user's local computing power matrix is ​​as follows: This indicates the chip frequency of IoT device nodes 1 to U;

[0018] Task characteristics include the amount of task data d u And the computational cost c required for the task u Using the binary T u = <d u ,c u > indicates task characteristics;

[0019] The user priority parameter is λ. u ;

[0020] User optimization preferences are This represents the preference parameters for IoT devices, where This indicates that the equipment is energy-saving; otherwise... This indicates that the device is a preference for delay reduction; when the two are equal, it means that the device is in balance between a preference for energy saving and a preference for delay reduction.

[0021] The user's maximum uplink transmission power is

[0022] The system noise variance is σ;

[0023] The system transmission bandwidth is B;

[0024] The chip's energy efficiency coefficient is κ. u .

[0025] Furthermore, in step three, based on the network user's task computation volume, local computing resources, and chip power consumption parameters, the user's local computing time and local computing power consumption are calculated, and the task computation characteristics are calculated, specifically:

[0026] The computational workload c based on network user u u Local computing resources And chip power consumption parameters κ u The user's local computing time is calculated as follows:

[0027] Local computing power consumption is κ l These are the power consumption parameters for IoT device chips.

[0028] And calculate the task computational features Where λ u Indicates the priority of user u. This indicates the user's device's latency preference; This indicates the local chip frequency of user u.

[0029] Furthermore, in step four, the objective function value, resource allocation, and uplink power allocation are calculated using a resource allocation algorithm based on the initial unloading decision. The specific resource allocation algorithm is as follows:

[0030]

[0031] in This indicates a fixed initial unloading decision; For uplink power allocation; Allocate computing resources for the system; This represents the user's device preference parameters. Greater than A time value indicates that the user u device prefers energy conservation; otherwise, it indicates that the user u device prefers latency reduction. u U represents the priority of user u; S is the set of satellites s, U s Let u be the set of users; Let Λ(F) represent the uplink power allocation (UPA) problem of the network, and Λ(F) represent the computing resource allocation (CRA) problem of the system. The optimal solutions for uplink power allocation obtained by solving the UPA and CRA problems respectively are: The optimal solution for system computing resource allocation is The objective function value is

[0032] Furthermore, the optimal solution for uplink power allocation obtained by solving the UPA and CRA problems is: The optimal solution for system computing resource allocation is Specifically, the CRA problem is a convex optimization problem, which is solved using KKT conditions;

[0033] The UPA problem is a quasi-convex optimization problem, which is solved using the bisection method. The steps of the bisection method are as follows:

[0034] Regarding the current offloading decision and user transmit power constraints: Where p u For user u's device transmit power, Let p be the maximum transmit power of user u's device, where U is the set of all users. tol The minimum transmit power required to meet communication latency requirements;

[0035] Iterate through all uninstalled users and perform the following steps:

[0036] S1: Judgment Or Γ′(p tol Does ≤0 hold true? If so, then... The optimal solution for the uplink power allocation is... And further obtain the optimal set of transmit powers P u ; where Γ(*) represents the uplink power allocation (UPA) problem in the network, and Γ′(*) is the p-partial derivative of Γ(*);

[0037] like Or Γ′(p tol If )≤0 is not true, then initialize p. u The upper and lower bounds are respectively: upper bound The Lower World

[0038] S2 sets the maximum number of iterations and executes the following iteration process: [Judgment / Determination] If true, update the lower bound. Otherwise update the upper bound. Determine if the preset maximum number of iterations has been reached; if so, proceed accordingly. As the optimal solution for uplink power allocation And further obtain the optimal set of transmit powers P u ;

[0039] If the preset maximum number of iterations is not reached, the iteration process of S2 will be re-entered with the updated upper and lower bounds.

[0040] Furthermore, in step five, based on the current unloading decision, a neighborhood solution set for the unloading decision is generated. Specifically, based on the current unloading decision X, all decision variables are traversed:

[0041] The neighborhood solution is obtained by performing a zeroing operation: select one decision variable from X. Set it to zero to obtain a neighborhood solution; iterate through all decision variables in X and perform the zeroing operation on each one, thereby generating a neighborhood solution set x. remove Among them, decision variables u represents the user, s represents the satellite, i represents the time slot, and j represents the subcarrier of the multi-carrier system;

[0042] To obtain a neighborhood solution, perform a value swap operation: select a pair of decision variables from X. and Swap the values ​​of the two variables to obtain a neighborhood solution; iterate through all decision variables in X and perform the value swap operation on each variable to generate a neighborhood solution set x. exchange .

[0043] Beneficial effects:

[0044] The method of this invention first calculates the satellite ephemeris based on the constellation information of the satellite network, thereby obtaining the visibility and distance of the satellite relative to IoT devices at the observation time. Then, based on the relative positions and distances between the satellite and the IoT devices, the communication channel gain and communication time constraints are calculated. Next, based on the calculated communication channel gain and communication time constraints, and the service requirements to be deployed in the network, a heuristic algorithm is used to iteratively solve the joint task offloading and resource allocation problem. Finally, network resources are allocated based on the decisions made in the joint task offloading and resource allocation problem. This method can guarantee that a suboptimal solution to the joint task offloading and resource allocation optimization problem of the satellite IoT edge computing space-ground integrated network is obtained within the polynomial computation time of the algorithm, thus improving the efficiency of offloading decisions and resource allocation in the space-ground integrated network system. Attached Figure Description

[0045] Figure 1 Schematic diagram of a satellite IoT edge computing space-ground integrated network architecture scenario;

[0046] Figure 2 Flowchart for solving the UPA problem using the bisection method;

[0047] Figure 3 Flowchart of a multi-dimensional resource scheduling method for low-Earth orbit satellite networks for IoT task offloading;

[0048] Figure 4 Initial unloading decision generation flowchart;

[0049] Figure 5 This is a flowchart illustrating the heuristic algorithm for generating neighborhood solutions. Detailed Implementation

[0050] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0051] Firstly, for the scenario of computational offloading of IoT devices, a joint task offloading and resource allocation optimization problem for satellite IoT networks is constructed, and an objective function is set with system performance as the optimization objective. This optimization problem is a mixed-integer linear programming problem. To reduce the solution complexity, the objective function is decomposed into two subproblems: a task offloading problem and a resource allocation problem with fixed offloading decisions. The resource allocation problem is further decomposed into a computing resource allocation (CRA) problem and an uploading power allocation (UPA) problem. The CRA problem is a convex optimization problem and can be solved using KKT conditions; the UPA problem is a quasi-convex optimization problem, and this invention designs a bisection method for solving it. Based on the network resource allocation results under fixed decisions, the task offloading problem can be further solved. This invention designs a heuristic-based task offloading algorithm that can solve for suboptimal network solutions with polynomial computational complexity.

[0052] For edge computing scenarios in satellite IoT terrestrial-space integrated networks, the following model addresses the joint task offloading and resource allocation problem:

[0053] like Figure 1 As shown, a classic satellite IoT network consists of satellite nodes with edge computing capabilities and various ground / air nodes across multiple domains, including vehicles, drones, and mobile IoT devices. Consider the following MEC scenario: When IoT user devices need to offload their services to satellites, offloading to ground-based data centers wastes significant energy and time due to the large computational load and small results. Furthermore, for scenarios requiring exploration in oceans, deserts, and post-disaster environments, the coverage of ground infrastructure is often insufficient, while low-Earth orbit (LEO) satellite MEC can provide more reliable communication and computing support.

[0054] Considering the limited energy storage of devices and the latency requirements of services in satellite IoT, network performance is set as the optimization objective. Network performance includes two components: node energy consumption and service latency. These two components are weighted together to form the objective function. Therefore, it is necessary to model network node energy consumption and service latency separately.

[0055] Elements in the system are represented by sets, including sets of user nodes. Satellite node set and satellite available subcarrier set Use the binary T u = <d u ,c u > represents the task, where d u Indicates the amount of data (bits) for the task, c u This indicates the computational load (CPU cycles) required for the task.

[0056] The network performance objective function can be decomposed into the sum of the energy efficiency of all network tasks. Task energy efficiency includes task energy consumption and task latency. Task energy consumption can be divided into task computing energy consumption and communication energy consumption, and task latency can be divided into computing latency and communication latency.

[0057] Therefore, the formula for calculating task energy consumption is:

[0058]

[0059] Where E u Indicates the energy consumption of the task. Indicates the energy consumption of task transmission. This indicates the energy consumption for task calculation.

[0060] The formula for calculating task latency is:

[0061]

[0062] Where T u Indicates task delay. Indicates the task computation latency. This indicates the task transmission delay.

[0063] On the other hand, considering the high dynamism of LEO satellite network nodes, it is necessary to model the visibility between satellites and IoT devices, as well as the temporal relationship of established communication links. Based on the relative positions of the satellite and the user, the time T during which the satellite and user equipment can maintain a communication link can be calculated. tol =T end -T now T end T indicates the moment when a chain cannot be established. now This indicates the current time. To ensure user service data can be transmitted to the satellite, the communication time must not exceed the link hold-up time, i.e.

[0064]

[0065] In this optimization problem, service deployment is set up in two scenarios: a user-local computing task scenario and a computing scenario where the user offloads the task to the satellite. In the user-local computing task scenario, service computation is performed on the user's device, and a user computing capability matrix is ​​set. The energy consumption of the computable task is

[0066]

[0067] in κ represents the energy consumption of user u's business computing. l This indicates the power consumption parameters of IoT device chips. This indicates the chip frequency of device u. The task latency is the device's local computation latency.

[0068]

[0069] in This represents the local computation latency of user u. On the other hand, in scenarios where users offload tasks to communication satellites for service deployment, the user's energy consumption can be derived.

[0070]

[0071] in This indicates whether user u offloads its service to satellite s via subcarrier j. p represents the energy consumption of service transmission. u Indicates the uplink transmit power of the service. R represents the business communication time. us This represents the communication rate from user u to satellite s. Service latency can be derived from this.

[0072]

[0073] Where c u f represents the computing resources required by user u, i.e., the number of CPU cycles. us This represents the computing resources allocated by satellite s to user u. The communication rate R from user u to satellite s is also represented. us It can be calculated using Shannon's formula.

[0074]

[0075] in This represents the uplink SINR from user u to satellite s on subcarrier j.

[0076] The network operates in time slots. At the beginning of each time slot, the network controller collects network information and makes decisions on task offloading and network resource allocation. Therefore, the total network energy efficiency is determined by the sum of the energy efficiency of each time slot. The energy efficiency function of user equipment u in time slot i is...

[0077]

[0078] in This indicates the user's device preference parameters, specifically whether the device prefers energy-saving or latency-reducing features.

[0079] Total energy efficiency of the system

[0080]

[0081] Where λ u Indicates the priority of user u. These represent the user offload decision set, the user uplink power set, and the network computing resource allocation set, respectively.

[0082] Based on the above derivation, a mathematical model for the joint optimization problem of satellite IoT offloading decision and resource allocation can be obtained.

[0083]

[0084]

[0085]

[0086]

[0087]

[0088]

[0089]

[0090]

[0091] Constraints (a) and (b) are unloading decision constraints. (a) describes the unloading decision... The constraints are as follows: (b) Constraint (c) is a communication resource constraint, which means that a subcarrier can be allocated to a maximum of one user equipment in a time slot; (d) is an energy consumption constraint, which limits the maximum uplink transmit power of the user equipment; (e) and (f) are computing resource constraints, which mean that the computing resource allocation in a time slot should be greater than zero and the total computing resources allocated by the satellite server are limited; (g) is a delay constraint, which requires that the communication delay of user u is not greater than the keep-alive link time.

[0092] Based on the above derivation, the joint optimization problem of system unloading decision and resource allocation was modeled. This problem belongs to mixed integer programming and is an NP-hard problem, therefore a closed-form optimal solution cannot be obtained directly. To solve this optimization problem, it is necessary to first decompose the joint optimization problem into subproblems, solve the subproblems separately, and then combine them to solve the original optimization problem.

[0093] To address the issue that the constructed joint task unloading and resource allocation problem model cannot find a closed-form solution and suffers from high computational complexity when solved directly, the solution complexity is reduced by decomposing the problem into independent subproblems. Since the unloading decision and resource allocation decision are independent of each other in the above equation, the objective function of the joint unloading and resource allocation problem can be decomposed as follows:

[0094]

[0095] in For the Task Offloading (TO) problem, This is a resource allocation (RA) problem. The RA problem can be further decomposed into fixed offloading decisions. In the case of, there are

[0096]

[0097] The optimization objective function after extracting the constants from the function V() is...

[0098]

[0099] Therefore, the RA problem can be transformed into

[0100]

[0101] Based on the problem model described above, we can obtain

[0102]

[0103] in Indicates the user's transmission latency characteristics. Indicates user energy consumption characteristics, This represents the computational characteristics of the task. The function V(·) can be further decomposed.

[0104]

[0105] in This indicates an uplink power allocation (UPA) problem in the network. This refers to the system's Computing Resource Allocation (CRA) problem.

[0106] On the other hand, it is necessary to transform the time delay constraint (g) of the original optimization problem.

[0107] T u ≤T tol

[0108] It can be decomposed into

[0109]

[0110] The upper part of the formula represents the maximum power used by IoT user devices. The transmission can meet the latency requirements to offload the decision-making process; the lower part requires the device to allocate a transmit power greater than a certain value so that the communication rate is high enough to transmit data within the latency constraints. Therefore

[0111]

[0112]

[0113] therefore

[0114]

[0115] The CRA problem, which is further decomposed from the resource allocation problem.

[0116]

[0117]

[0118]

[0119] This problem is a convex optimization problem. We can construct the KKT conditions and find its closed-form solution to obtain the optimal allocation of computational resources and the optimal value.

[0120]

[0121]

[0122] For the UPA problem derived from the resource allocation problem Since this is a quasi-convex optimization problem, the optimal value of Γ(P) occurs at the transmit power p. u The present invention uses a binary search method to solve for the extreme values ​​or points where the derivative equals 0, determined by constraints. The solution process is as follows: Figure 2 As shown.

[0123] Based on the resource allocation method derived above under the fixed unloading decision, we can obtain...

[0124]

[0125] Λ(F) represents the uplink power allocation (UPA) problem in the network, and Λ(F) represents the computing resource allocation (CRA) problem in the system. The optimal solution for uplink power allocation; The optimal solution for system computational resource allocation;

[0126] This indicates the user's device preference parameters, i.e., whether the device prefers energy saving or latency reduction.

[0127] λ u U represents the priority of user u; S is the set of satellites s, U s Let u be the set of users;

[0128] Therefore, the TO problem can be transformed into

[0129]

[0130]

[0131]

[0132]

[0133]

[0134] To address the joint task offloading and resource allocation problem, this invention further designs a task offloading decision-making method based on a heuristic learning algorithm. This method can generate task offloading decisions in polynomial computation time and obtain suboptimal solutions.

[0135] A multi-dimensional resource scheduling method for low-Earth orbit (LEO) satellite networks oriented towards IoT task offloading is proposed. The LEO satellite network includes satellite nodes and IoT device nodes. The basic implementation process is as follows: First, the satellite ephemeris is calculated based on the constellation information of the satellite network, thereby obtaining the satellite's visibility and distance relative to the IoT devices at the observation time. Then, the communication channel gain and communication time constraints are calculated based on the relative positions and distances between the satellites and IoT devices. Next, based on the calculated communication channel gain and communication time constraints, and the service requirements to be deployed in the network, a heuristic algorithm is used to iteratively solve the joint task offloading and resource allocation problem. Finally, network resources are allocated based on the decisions made in the joint task offloading and resource allocation problem. Figure 3 A flowchart illustrating a multi-dimensional resource scheduling method for low-Earth orbit satellite networks aimed at IoT task offloading. Firstly, based on... Figure 4 The method generates an initial feasible offloading decision X, and then performs resource allocation based on the offloading decision to obtain the transmit power allocation P and calculate the resource allocation F. The results are judged to determine whether they have reached the local optimum. If they have not reached the local optimum, the offloading decision neighborhood solution is generated and the decision is replaced. X, P and F are regenerated. If they have reached the local optimum, the offloading decision X, resource allocation P and F and objective function value J are output.

[0136] In this embodiment, the initial time slot is set to 1, and the following steps are performed in each time slot:

[0137] Step 1: Initialize the current time-slot network state, and construct the satellite server computing power matrix, user local computing power matrix, task characteristics, user priority parameters, user optimization preference parameters, user maximum uplink transmission power, system noise variance, system transmission bandwidth, and chip power consumption coefficient. In this embodiment, the specific parameters include:

[0138] The satellite server computing power matrix is ​​as follows Where S is the total number of satellite nodes in the low-Earth orbit satellite network. The server chip frequency for satellite nodes 1 to S;

[0139] The user's local computing power matrix is ​​as follows: This indicates the chip frequency of IoT device nodes 1 to U;

[0140] Task characteristics include the amount of task data d u And the computational cost c required for the task u Using the binary T u = <d u ,c u > indicates task characteristics;

[0141] The user priority parameter is λ. u ;

[0142] User optimization preferences are This represents the preference parameters for IoT devices, where This indicates that the equipment is energy-saving; otherwise... This indicates that the device is a preference for delay reduction; when the two are equal, it means that the device is in balance between a preference for energy saving and a preference for delay reduction.

[0143] The user's maximum uplink transmission power is

[0144] The system noise variance is σ;

[0145] The system transmission bandwidth is B;

[0146] The chip's energy efficiency coefficient is κ. u .

[0147] Step 2: Collect satellite constellation information and location information of IoT devices with mission access requirements in the network, and calculate the visibility and distance between nodes based on the relative positions of satellites and IoT devices;

[0148] Furthermore, the communication channel gain and communication link hold time between nodes are calculated using the free space attenuation equation based on the distance between nodes;

[0149] Step 3: Based on the network user's task computation volume, local computing resources, and chip power consumption parameters, calculate the user's local computing time and local computing power consumption, and calculate the task computation characteristics; in this embodiment, specifically:

[0150] The computational workload c based on network user u u Local computing resources And chip power consumption parameters κ u The user's local computing time is calculated as follows:

[0151] Local computing power consumption is κ l These are the power consumption parameters for IoT device chips.

[0152] And calculate the task computational features Where λ u Indicates the priority of user u. This indicates the user's device's latency preference; This indicates the local chip frequency of user u.

[0153] Step 4: Generate the initial unloading decision, and calculate the objective function value, resource allocation, and uplink power allocation based on the initial unloading decision using the resource allocation algorithm. Set the initial unloading decision as the current unloading decision and set its corresponding objective function value as the current objective function value.

[0154] In this embodiment, the specific resource allocation algorithm is as follows:

[0155]

[0156] in This indicates a fixed initial unloading decision. For uplink power allocation; For system computing resource allocation, Λ(F) represents the uplink power allocation (UPA) problem of the network, and Λ(F) represents the computing resource allocation (CRA) problem of the system. This is the optimal solution for uplink power allocation; The optimal solution for allocating system computing resources; This represents the user's device preference parameters. Greater than A time value indicates that the user u device prefers energy conservation; otherwise, it indicates that the user u device prefers latency reduction. u U represents the priority of user u; S is the set of satellites s, U s Let u be the set of users.

[0157] The CRA problem, derived from the resource allocation problem, is a convex optimization problem and can be solved using the KKT conditions; the UPA problem, derived from the resource allocation problem... Since this is a quasi-convex optimization problem, the optimal value of Γ(P) occurs at the transmit power p. u The present invention uses a binary search method to solve for the extreme values ​​or points where the derivative equals 0, determined by constraints. The solution process is as follows: Figure 2 As shown.

[0158] Regarding the current offloading decision and user transmit power constraints: Where p u For user u's device transmit power, Let p be the maximum transmit power of user u's device, where U is the set of all users. tol The minimum transmit power required to meet communication latency requirements;

[0159] Iterate through all uninstalled users and perform the following steps:

[0160] S1: Judgment Or Γ′(p tol Does ≤0 hold true? If so, then... The optimal solution for the uplink power allocation is... And further obtain the optimal set of transmit powers P u ; where Γ(*) represents the uplink power allocation (UPA) problem in the network, and Γ′(*) is the p-partial derivative of Γ(*);

[0161] like Or Γ′(p tol If )≤0 is not true, then initialize p. u The upper and lower bounds are respectively: upper bound The Lower World

[0162] S2 sets the maximum number of iterations and executes the following iteration process: [Judgment / Determination] If true, update the lower bound. Otherwise update the upper bound. Determine if the preset maximum number of iterations has been reached; if so, proceed accordingly. As the optimal solution for uplink power allocation And further obtain the optimal set of transmit powers P u ;

[0163] If the preset maximum number of iterations is not reached, the iteration process of S2 will be re-entered with the updated upper and lower bounds.

[0164] Step 5: Based on the current unloading decision, generate a neighborhood solution set for the unloading decision; traverse all neighborhood solutions for the unloading decision, and perform resource allocation based on the neighborhood solutions to obtain a set of objective functions for the neighborhood solutions; the specific method for generating the neighborhood solution set for the unloading decision is as follows... Figure 5 As shown, the process is as follows: Based on the current unloading decision X, iterate through all decision variables:

[0165] The neighborhood solution is obtained by performing a zeroing operation: select one decision variable from X. Set it to zero to obtain a neighborhood solution; iterate through all decision variables in X and perform the zeroing operation on each one, thereby generating a neighborhood solution set x. remove Among them, decision variables u represents the user, s represents the satellite, i represents the time slot, and j represents the subcarrier of the multi-carrier system;

[0166] To obtain a neighborhood solution, perform a value swap operation: select a pair of decision variables from X. and Swap the values ​​of the two variables to obtain a neighborhood solution; iterate through all decision variables in X and perform the value swap operation on each variable to generate a neighborhood solution set x. exchange .

[0167] Step Six: Compare the objective function set of neighboring solutions with the current objective function value. The specific resource allocation method under the fixed unloading decision is as follows:

[0168]

[0169] in For neighborhood solutions The objective function; Λ(F) represents the uplink power allocation (UPA) problem of the network, and Λ(F) represents the computing resource allocation (CRA) problem of the system. The optimal solution for uplink power allocation; The optimal solution for system computational resource allocation; This represents the user device's preference parameter, i.e., whether the device prefers energy saving or latency reduction; λ u U represents the priority of user u; S is the set of satellites s, U s Let u be the set of users;

[0170] If there exists a neighboring solution A whose objective function is greater than the current objective function value, then replace the current unloading decision with the unloading decision corresponding to the neighboring solution A, and return to step five; otherwise, if the objective functions of all neighboring solutions are not greater than the current objective function value, then use the current unloading decision as the unloading decision for the current time slot, and allocate resources in the current time slot according to the unloading decision of the current time slot, and then count the service energy consumption and service latency of the current time slot.

[0171] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-dimensional resource scheduling method for low-Earth orbit satellite networks for IoT task offloading, characterized in that, The nodes in the low-Earth orbit satellite network include satellite nodes and Internet of Things (IoT) device nodes. The initial time slot is set to 1, and the following steps are performed in each time slot: Step 1: Initialize the current time slot network state and construct the satellite server computing power matrix, user local computing power matrix, task characteristics, user priority parameters, user optimization preference parameters, user maximum uplink transmission power, system noise variance, system transmission bandwidth, and chip power consumption coefficient. Step 2: Collect satellite constellation information and location information of IoT devices with mission access requirements in the network, and calculate the visibility and distance between nodes based on the relative positions of satellites and IoT devices; Furthermore, the communication channel gain and communication link hold time between nodes are calculated using the free space attenuation equation based on the distance between nodes; Step 3: Based on the network user's task computation volume, local computing resources, and chip power consumption parameters, calculate the user's local computing time and local computing power consumption, and calculate the task computation characteristics; Step 4: Generate the initial unloading decision, and calculate the objective function value, resource allocation, and uplink power allocation based on the initial unloading decision using the resource allocation algorithm. Set the initial unloading decision as the current unloading decision and set its corresponding objective function value as the current objective function value. Step 5: Based on the current unloading decision, generate a set of neighborhood solutions for the unloading decision; traverse all neighborhood solutions for the unloading decision, and allocate resources based on the neighborhood solutions for the unloading decision to obtain a set of objective functions for the neighborhood solutions; Step Six: Compare the objective function set of neighboring solutions with the current objective function value. If there exists a neighboring solution A whose objective function is greater than the current objective function value, replace the current unloading decision with the unloading decision corresponding to neighboring solution A and return to Step Five. Otherwise, if none of the neighboring solutions' objective functions are greater than the current objective function value, use the current unloading decision as the unloading decision for the current time slot, allocate resources in the current time slot according to the unloading decision, and then statistically analyze the service energy consumption and service latency of the current time slot.

2. The multi-dimensional resource scheduling method for low-Earth orbit satellite networks for IoT task offloading as described in claim 1, characterized in that, In step one, the current time slot network state is initialized, and the following parameters are constructed: satellite server computing power matrix, user local computing power matrix, task characteristics, user priority parameters, user optimization preference parameters, user maximum uplink transmission power, system noise variance, system transmission bandwidth, and chip power consumption coefficient. Specifically, these parameters include the following: The satellite server computing power matrix is ​​as follows Where S is the total number of satellite nodes in the low-Earth orbit satellite network. The server chip frequency for satellite nodes 1 to S; The user's local computing power matrix is ​​as follows: This indicates the chip frequency of IoT device nodes 1 to U; Task characteristics include the amount of task data d u And the computational cost c required for the task u Using the binary T u = <d u ,c u > indicates task characteristics; The user priority parameter is λ. u ; User optimization preferences are This represents the preference parameters for IoT devices, where This indicates that the equipment is energy-saving; otherwise... This indicates that the device is a preference for delay reduction; when the two are equal, it means that the device is in balance between a preference for energy saving and a preference for delay reduction. The user's maximum uplink transmission power is The system noise variance is σ; The system transmission bandwidth is B; The chip's energy efficiency coefficient is κ. u .

3. The multi-dimensional resource scheduling method for low-Earth orbit satellite networks for IoT task offloading as described in claim 2, characterized in that, In step three, based on the network user's task computation volume, local computing resources, and chip power consumption parameters, the user's local computing time and local computing power consumption are calculated, and the task computation characteristics are calculated, specifically: The computational workload c based on network user u u Local computing resources And chip power consumption parameters κ u The user's local computing time is calculated as follows: Local computing power consumption is κ l These are the power consumption parameters for IoT device chips. And calculate the task computational features Where λ u Indicates the priority of user u. This indicates the user's device's latency preference; This indicates the local chip frequency of user u.

4. The multi-dimensional resource scheduling method for low-Earth orbit satellite networks for IoT task offloading as described in claim 3, characterized in that, In step four, the objective function value, resource allocation, and uplink power allocation are calculated using a resource allocation algorithm based on the initial unloading decision. The specific resource allocation algorithm is as follows: in This indicates a fixed initial unloading decision; For uplink power allocation; Allocate computing resources for the system; This represents the user's device preference parameters. Greater than A time value indicates that the user u device prefers energy conservation; otherwise, it indicates that the user u device prefers latency reduction. u U represents the priority of user u; S is the set of satellites s, U s Let u be the set of users; Let Λ(F) represent the uplink power allocation (UPA) problem of the network, and Λ(F) represent the computing resource allocation (CRA) problem of the system. The optimal solutions for uplink power allocation obtained by solving the UPA and CRA problems respectively are: The optimal solution for system computing resource allocation is The objective function value is 5. The multi-dimensional resource scheduling method for low-Earth orbit satellite networks for IoT task offloading as described in claim 4, characterized in that, The optimal solution for uplink power allocation obtained by solving the UPA and CRA problems is: The optimal solution for system computing resource allocation is Specifically, the CRA problem is a convex optimization problem, which is solved using KKT conditions; The UPA problem is a quasi-convex optimization problem, which is solved using the bisection method. The steps of the bisection method are as follows: Regarding the current offloading decision and user transmit power constraints: Where p u For user u's device transmit power, Let p be the maximum transmit power of user u's device, where U is the set of all users. tol The minimum transmit power required to meet communication latency requirements; Iterate through all uninstalled users and perform the following steps: S1: Judgment Or Γ′(p tol Does ≤0 hold true? If so, then... The optimal solution for the uplink power allocation is... And further obtain the optimal set of transmit powers P u ; where Γ(*) represents the uplink power allocation (UPA) problem in the network, and Γ′(*) is the p-partial derivative of Γ(*); like Or Γ′(p tol If )≤0 is not true, then initialize p. u The upper and lower bounds are respectively: upper bound The Lower World S2 sets the maximum number of iterations and executes the following iteration process: [Judgment / Determination] If true, update the lower bound. Otherwise update the upper bound. Determine if the preset maximum number of iterations has been reached; if so, proceed accordingly. As the optimal solution for uplink power allocation And further obtain the optimal set of transmit powers P u ; If the preset maximum number of iterations is not reached, the iteration process of S2 will be re-entered with the updated upper and lower bounds.

6. The multi-dimensional resource scheduling method for low-Earth orbit satellite networks for IoT task offloading as described in any one of claims 1 to 4, characterized in that, In step five, a neighborhood solution set for the current unloading decision is generated. Specifically, based on the current unloading decision X, all decision variables are iterated through. The neighborhood solution is obtained by performing a zeroing operation: select one decision variable from X. Set it to zero to obtain a neighborhood solution; iterate through all decision variables in X and perform the zeroing operation on each one, thereby generating a neighborhood solution set x. remove Among them, decision variables u represents the user, s represents the satellite, i represents the time slot, and j represents the subcarrier of the multi-carrier system; To obtain a neighborhood solution, perform a value swap operation: select a pair of decision variables from X. and Swap the values ​​of the two variables to obtain a neighborhood solution; iterate through all decision variables in X and perform the value swap operation on each variable to generate a neighborhood solution set x. exchange .

Images