A Dynamic Weighted Edge Computing Scheduling Method Based on 5G RedCap Module

Abstract

The present invention belongs to the technical field of edge computing, and particularly relates to a dynamic weighted edge computing scheduling method based on a 5G RedCap module. The method includes: Step 1: Perform node resource perception on each RedCap module node in the network, and calculate the effective computing power of each RedCap module node at each moment; Step 2: By analyzing the dynamic characteristics of the RedCap links of each RedCap module node in the network, extract the RedCap link feature vectors of each RedCap module node; Step 3: Model the resource requirements of each task and estimate the execution time on different RedCap module nodes; Step 4: Construct a multi-dimensional decision matrix to quantify the execution benefits of each task on each RedCap module node; Step 5: Based on the multi-dimensional decision matrix, use the Lagrangian relaxation method to solve the optimal task allocation scheme. The present invention ensures that tasks can be scheduled in an optimal manner and improves the task execution success rate.

Description

Technical Field

[0001] The present invention belongs to the technical field of edge computing, and particularly relates to a dynamic weighted edge computing scheduling method based on a 5G RedCap module. Background Art

[0002] With the rapid development of 5G communication technology, edge computing, as a technology capable of executing computing tasks at the network edge close to users, has gradually become an important means to solve the computing pressure of cloud computing centers and reduce data transmission latency. The 5G network features high bandwidth, low latency, and large-scale connection, enabling edge computing to support more applications with high real-time and high computing requirements, such as intelligent transportation, industrial automation, augmented reality (AR) / virtual reality (VR), etc. In this context, the 5G RedCap module, as an important technology in the 5G standard, has been optimized for scenarios with low-power and lightweight computing requirements such as the Internet of Things (IoT) and intelligent terminals, enabling 5G to better support the access of a large number of low-power devices. However, there are still many challenges in the edge computing task scheduling of existing 5G RedCap modules, especially in aspects such as computing resource management for task execution, communication overhead optimization, energy consumption control, and dynamic environment adaptability. The existing technologies have not provided an efficient task scheduling scheme.

[0003] Currently, the research on edge computing task scheduling mainly focuses on two types of methods: static task scheduling based on computing resource awareness and task offloading strategies based on dynamic network states. The first type of method mainly focuses on the computing capabilities of edge computing nodes. For example, based on the shortest task execution time or minimum completion time strategy, tasks are assigned to the node with the strongest computing power to minimize the task execution time. However, this type of method usually assumes that the computing capabilities of edge computing nodes are stable, ignoring the dynamic changes in the computing capabilities of RedCap modules due to factors such as CPU frequency adjustment, temperature change, and load fluctuation. Therefore, in practical applications, static task scheduling strategies often cannot adapt to the volatility of RedCap module computing resources, resulting in some nodes being overloaded while the computing resource utilization of other nodes is relatively low, thus reducing the computing efficiency of the entire system.

[0004] The second type of methods mainly considers the dynamic characteristics of wireless communication networks and adopts task offloading strategies based on factors such as channel state, bandwidth occupancy, packet error rate, etc. Since the communication capabilities of RedCap modules are limited compared to traditional 5G devices, task offloading usually needs to consider the communication link quality to reduce data transmission delay and energy consumption. For example, existing task offloading strategies usually use signal-to-noise ratio, channel quality indicator (CQI), or round-trip delay (RTT) to evaluate the network link quality and decide whether a task should be executed locally or offloaded to other computing nodes. However, such methods often have the following problems: (1) lack of collaborative optimization of computing resources and communication resources, only focusing on the communication link quality while ignoring the computing capabilities and energy consumption characteristics of RedCap modules, which may cause tasks to be offloaded to nodes with relatively weak computing capabilities, resulting in reduced task execution efficiency; (2) do not consider the characteristics of the tasks themselves, such as the computing complexity, data input / output requirements, etc. Different types of tasks have different requirements for computing resources and network resources, so a single task offloading strategy cannot meet all application scenarios. Summary of the Invention

[0005] In view of this, the main object of the present invention is to provide a dynamic weighted edge computing scheduling method based on 5G RedCap modules. The dynamic weighted edge computing scheduling method of the present invention ensures that tasks can be scheduled in an optimal manner and improves the task execution success rate through accurate computing capacity modeling, network quality analysis, task execution time prediction, energy consumption optimization, load balancing, and intelligent optimization solution.

[0006] The technical solution adopted by the present invention is as follows:

[0007] A dynamic weighted edge computing scheduling method based on 5G RedCap modules, the method comprising:

[0008] Step 1: Perform node resource awareness on each RedCap module node in the network, and calculate the effective computing capacity of each RedCap module node at each moment;

[0009] Step 2: By analyzing the dynamic characteristics of the RedCap links of each RedCap module node in the network, extract the RedCap link feature vectors of each RedCap module node;

[0010] Step 3: Model the resource requirements of each task and estimate the execution time on different RedCap module nodes;

[0011] Step 4: Construct a multi-dimensional decision matrix to quantify the execution benefits of each task on each RedCap module node;

[0012] Step 5: Based on the multi-dimensional decision matrix, use the Lagrangian relaxation method to solve the optimal task allocation scheme.

[0013] Furthermore, in Step 1, the effective computing power of each RedCap module node at each moment is calculated through the following formula:

[0014] ;

[0015] where, is the effective computing power of RedCap module node at moment ; is the CPU frequency of RedCap module node ; is the CPU utilization rate of RedCap module node at moment , and its value range is (0, 1); is the system overhead coefficient of RedCap module node at moment , indicating the proportion of computing resources consumed by system maintenance; is the temperature of RedCap module node at moment ; is the temperature threshold, and the computing performance begins to decline significantly when it exceeds this value, and its value range is 70 to 85; is the temperature scaling factor, which controls the steepness of the impact of temperature on performance; is the available memory of RedCap module node at moment ; is the total memory capacity of RedCap module node .

[0016] Furthermore, in Step 2, the RedCap link feature vector of each RedCap module node is extracted through the following formula:

[0017] ;

[0018] where, is the RedCap link feature vector of RedCap module node at moment , which contains 4 dimensions; is the effective bandwidth of RedCap module node at moment ; is the jitter of RedCap module node at moment ; is the reference jitter value; is the RedCap module node at time round-trip delay; is the maximum round-trip delay acceptable to the system; is the RedCap module node at time physical moving speed. For a fixed RedCap module node, this value is 0; is the reference physical moving speed value; is the RedCap module node at time signal-to-noise ratio; is the system noise floor signal-to-noise ratio; is the RedCap module node at time channel quality indicator, with a value range of 0 to 15; is the RedCap module node at time packet error rate; is the RedCap module node at time channel switching frequency; is the packet error impact coefficient, with a value range of 5 to 100; is the packet error switching frequency impact coefficient, with a value range of 2 to 50.

[0019] Furthermore, in step 3, the execution time on different RedCap module nodes is estimated through the following formula:

[0020] ;

[0021] where, is the execution time of task on RedCap module node at time ; is the number of computing operations required for task ; is the computing complexity coefficient of task ; is the input data volume of task ; is the output data volume of task ;

[0022] Furthermore, the computing complexity coefficient of task is calculated using the following formula:

[0023] ;

[0024] wherein, is the basic complexity coefficient of the task, determined by the asymptotic complexity of the task: for tasks with O(1) complexity, ; for tasks with O(logn) complexity, ; for tasks with O(n) complexity, ; for tasks with O(nlogn) complexity, ; for tasks with O(n²) complexity, ; for tasks with O(n³) complexity, ; ; is the working set memory size of the task, that is, the amount of memory frequently accessed during the execution of the task; is the total memory requirement of the task; is the memory impact factor, with a value ranging from 0.2 to 0.5, reflecting the impact of memory access on computing performance; is the data dependency of the task, with a value of 0 or 1, indicating the degree of dependency between the internal computing units of the task: 0 means fully parallel and no dependency; 1 means fully serial and strong dependency; is the dependency impact factor, with a value range of 0.5 to 1.5; is the instruction mode coefficient, determined according to the instruction characteristics of the task: for compute-intensive tasks, ranges from 0.8 to 1.0; for memory-intensive tasks, ranges from 1.2 to 1.5. Furthermore, in step 4, a multi-dimensional decision matrix is constructed through the following formula to quantify the execution benefit of each task on each RedCap module node: ; ;

[0025] wherein,

[0026] ;

[0027] wherein, is the execution benefit of task on RedCap module node at time ; is the priority weight of task ; is the deadline constraint of task ; is the energy required for the calculation of task on RedCap module node ; For the task At the RedCap module node At the moment Energy required for communication; For the RedCap module node At the moment The load variance, reflecting the load stability; For the influence factor of the dimension of the link quality eigenvector;

[0028] Furthermore, It is calculated using the following formula:

[0029] ;

[0030] Wherein, For the RedCap module node The energy efficiency coefficient.

[0031] Furthermore, It is calculated using the following formula:

[0032]

[0033] Wherein, For the RedCap module node The energy efficiency coefficient; For the RedCap module node The transmission power; For the RedCap module node The reception power.

[0034] Furthermore, in step 5, based on the multi-dimensional decision matrix, the Lagrangian relaxation method is used to solve the optimal task allocation scheme through the following formula:

[0035] ;

[0036] Wherein, Is the objective function; Is the task allocation matrix, Indicates that the task Is allocated to the RedCap module node , otherwise ; For the RedCap module node The Lagrangian multiplier of the computational resource capacity constraint of, Is the Lagrangian multiplier vector of the resource capacity constraint; For the task The Lagrangian multiplier of the allocation constraint of, is the Lagrange multiplier vector for the allocation constraint; is the total number of tasks; is the total number of RedCap module nodes; The final scheduling decision is obtained by solving the following optimization problem:

[0037] ;

[0038] where, and are the Lagrange multiplier vectors for the optimal computing resource capacity constraint and the optimal allocation constraint iteratively solved by the subgradient method, respectively; is the optimal task allocation matrix.

[0039] With the above technical solutions, the present invention has the following beneficial effects: In traditional edge computing scheduling methods, the evaluation of the computing power of tasks usually only depends on the CPU frequency or the idle state of computing nodes, and fails to fully consider the dynamic changes of computing resources. By constructing a computing power model for RedCap modules, the present invention introduces multiple factors such as CPU utilization, system overhead, temperature impact, and memory occupancy to accurately model the computing power of RedCap modules, enabling task scheduling to accurately evaluate the computing power of each computing node. For example, when the temperature is high, the computing power of the RedCap module will decline, but the traditional method cannot perceive this, which may lead to tasks being assigned to high-temperature nodes, thus affecting the execution efficiency. The computing power model of the present invention can dynamically adjust the computing power evaluation, enabling task scheduling to avoid nodes with declining computing power, thereby improving the utilization rate of computing resources. The computing tasks of RedCap modules often involve data transmission. Most traditional task scheduling methods only consider computing power and ignore the dynamic changes in network status, resulting in problems such as task failure or excessive execution time due to insufficient bandwidth, high latency, and increased packet error rate during task execution. By constructing a RedCap link quality feature vector, the present invention extracts multiple key network features such as bandwidth, latency, jitter, signal-to-noise ratio, packet error rate, and channel switching frequency, and uses mathematical modeling methods such as exponential decay, normalization, and logarithmic transformation to enable the task scheduling system to accurately evaluate the communication quality of different computing nodes, ensuring that tasks are preferentially assigned to RedCap module nodes with better communication conditions, thereby reducing the task execution failure rate and improving the stability of task execution. In traditional task scheduling methods, tasks tend to be assigned to nodes with the strongest computing power, which may cause some RedCap modules to be overloaded while the computing resources of other nodes are not fully utilized, affecting the overall throughput of the system. By constructing a load variance model for computing nodes, the present invention adjusts the task scheduling scheme to minimize the load variance during task assignment, ensuring that computing tasks are evenly distributed among different RedCap modules and avoiding system computing power decline caused by overload of certain nodes. In this way, the scheduling method of the present invention can improve the stability of the entire computing system and reduce the failure rate caused by overload during task scheduling. Description of the Drawings

[0040] Figure 1 It is a schematic flowchart of a method for a dynamic weighted edge computing scheduling method based on 5G RedCap modules provided by an embodiment of the present invention. Detailed Embodiments

[0041] All features disclosed in this specification, or all steps in the disclosed methods or processes, except for mutually exclusive features and / or steps, can be combined in any manner.

[0042] Any feature disclosed in this specification (including any additional claims, abstract) can be replaced by other equivalent or alternative features with similar purposes, unless specifically stated. That is, unless specifically stated, each feature is only an example of a series of equivalent or similar features.

[0043] Example 1: Refer to Figure 1 , a dynamic weighted edge computing scheduling method based on a 5G RedCap module, the method comprising:

[0044] Step 1: Perform node resource awareness on each RedCap module node in the network, and calculate the effective computing power of each RedCap module node at each moment;

[0045] In the process of evaluating the computing power of the RedCap module, multiple influencing factors need to be considered, including the CPU clock frequency, the current CPU utilization rate, system maintenance overhead, node temperature, and available memory, etc. First of all, the CPU clock frequency determines the basic computing power of the computing node. However, since the RedCap module usually adopts a low-power architecture and its clock frequency is relatively low, it is inaccurate to measure the computing power only by the clock frequency. The computing power is also affected by the current CPU utilization rate. The higher the utilization rate, the fewer idle computing resources of the node, so the computing power will decrease accordingly. At the same time, during the actual operation process, system maintenance tasks (such as background processes of the operating system, network protocol stack management, etc.) will also consume a certain amount of computing resources. Therefore, it is necessary to introduce a system maintenance overhead parameter to reflect the impact of these overheads on the computing power. In addition, temperature is also a key factor. Since the hardware design of the RedCap module usually emphasizes low power consumption and high energy efficiency, its heat dissipation ability is weaker than that of traditional edge computing devices during high-load operation. When the module temperature rises, the chip may start dynamic voltage and frequency scaling (DVFS) to reduce power consumption and temperature, which directly affects the computing power. For this reason, the present invention introduces a temperature impact factor in the computing power modeling to quantify the impact of temperature changes on the computing power, so as to avoid the problem of reduced computing performance caused by ignoring the temperature factor during the scheduling process. Specifically, when the temperature exceeds a certain threshold, the computing power will decrease significantly. Therefore, a non-linear function is used in the computing model to simulate this effect, so that as the temperature rises, the decrease in computing power shows a progressive characteristic. In addition to temperature, the availability of memory is also an important factor affecting the computing power. Since tasks in edge computing often involve a large amount of data processing and the memory resources of the RedCap module are limited, if the available memory of the node is insufficient, task execution may be restricted or even cause task failure. Therefore, in the computing power modeling of the present invention, the ratio of available memory to total memory is introduced as a correction factor to reflect the impact of the storage resources of the current node on the computing power. When the available memory is more, the computing power is relatively stronger; on the contrary, when the available memory is insufficient, the computing power will decrease accordingly, so that the scheduling algorithm can avoid possible memory bottleneck problems during task allocation.

[0046] Step 2: By analyzing the dynamic characteristics of the RedCap links of each RedCap module node in the network, extract the RedCap link feature vectors of each RedCap module node;

[0047] To accurately measure the network link quality of the RedCap module, it is first necessary to focus on the core indicator of effective bandwidth. Effective bandwidth is a key parameter that determines the data transmission rate and directly affects the transmission efficiency of task data in the network. However, the bandwidth is not constant but is affected by network load, interference, and changes in channel quality. Therefore, in actual measurement, one cannot rely solely on the current bandwidth value but needs to combine historical data for smoothing to obtain a more representative network state. At the same time, considering that RedCap module devices are mostly used in low-power scenarios and there may be certain limitations in their bandwidth usage patterns, when measuring the effective bandwidth, it is also necessary to combine the power management strategy of RedCap devices to ensure that the measurement results can reflect the transmission capabilities of the devices during actual operation. In addition to bandwidth, network jitter is another factor that cannot be ignored. Jitter refers to the variation in packet transmission delay, which determines the smoothness of the task data stream. When the jitter is large, even if the average delay is low, the actual task execution may be affected, especially in latency-sensitive application scenarios such as industrial control or remote medical treatment. Therefore, in the link quality modeling of the present invention, jitter is introduced as an important characteristic indicator and is processed using an exponential decay method to avoid drastic fluctuations in the calculation results due to short-term jitter anomalies. In addition, to more intuitively measure the impact of jitter on the data stream, the system also introduces a reference jitter threshold, enabling the impact of jitter to be dynamically adjusted within a certain range to adapt to different types of computing tasks.

[0048] Another key factor of the network link is the round-trip delay, which is an important criterion for measuring whether task scheduling can meet the low-latency requirements. For edge computing tasks, low latency is the basis for ensuring real-time performance. Although the RedCap module adopts 5G communication technology, due to its simplified design, its round-trip delay may be affected by greater uncertainties. Therefore, in the process of network feature extraction of the present invention, it is necessary to normalize the round-trip delay of the RedCap module and perform dynamic comparison in combination with the maximum acceptable delay, so as to effectively avoid high-latency nodes during task scheduling. In addition, considering the mobility of the RedCap module, the network delay may also be affected by the moving speed of the device. Therefore, when calculating the delay, a speed parameter is also introduced, so that the system can dynamically adjust the task allocation strategy according to the moving state of the device, thereby reducing the delay fluctuation caused by the device movement. The signal-to-interference ratio is also an important indicator affecting the link quality. In a wireless communication environment, the ratio of the signal strength to the environmental interference directly determines the stability of data transmission. For the RedCap module, since it mainly operates in the low-power mode, its signal power is usually low, making the signal-to-interference ratio more susceptible to external interference. In the method of the present invention, a logarithmic function is used to process the signal-to-interference ratio, so that the change of the signal quality can be more intuitively reflected during task scheduling. In addition, in order to further enhance the accuracy of channel quality assessment, a channel quality indication parameter is also introduced. This parameter is provided by the 5G base station and can reflect the current channel environment of the RedCap module in real time. By combining the signal-to-interference ratio with the channel quality indication, the system can more accurately evaluate the communication stability of the node, so as to preferentially select nodes with better communication quality during task allocation to reduce the risk of task execution failure. In the actual communication process, the packet error rate and the channel switching frequency are also key factors affecting the stability of data transmission. The packet error rate refers to the probability of packet loss during data transmission. An excessively high packet error rate will lead to data retransmission, thereby increasing the network delay and energy consumption. The channel switching frequency reflects the frequency at which the device switches between different channels. When the switching frequency is too high, the stability of data transmission will be affected. Therefore, in the network link quality assessment model of the present invention, the packet error rate and the channel switching frequency are combined into a comprehensive factor and normalized through a non-linear function, so that their influence can better meet the actual needs of the edge computing scenario. Specifically, during the task scheduling process, when the packet error rate or the channel switching frequency of a certain RedCap module is too high, the link quality score of this node will be correspondingly reduced, thereby reducing the probability of task allocation to this node.

[0049] Step 3: Model the resource requirements of each task and estimate the execution time on different RedCap module nodes;

[0050] In the process of modeling the resource requirements of tasks, multiple factors such as the computational volume of the task, the data input and output sizes, and the computational complexity need to be considered. The computational volume of a task is usually measured by the number of floating-point operations required for the task to execute, which is a basic parameter determining the task execution time. The computational complexity is determined by the algorithm characteristics of the task and reflects the execution efficiency of the task on different computing architectures. Due to the limited computing power of the RedCap module, the impact of computational complexity on the task execution time is particularly obvious, and there may be significant differences in the execution times of different tasks on the same computing resources. Therefore, in the method of the present invention, the task computational complexity is modeled separately so that different task computing requirements can be distinguished during the scheduling process and more accurate allocation decisions can be made based on the computing power of the RedCap module. In addition to the computing requirements, the execution time of a task is also affected by the data transmission delay. Since the RedCap module is usually deployed in an edge computing environment, the data input and output of tasks often need to be transmitted through a wireless network, and there are certain limitations in the communication ability of the RedCap module compared with full-featured 5G devices. Therefore, the proportion of data transmission overhead in the task execution time may be relatively high. To more accurately estimate the task execution time, the data input size, data output size of the task, and the network bandwidth situation of the RedCap module need to be considered. At the same time, the packet error rate is an important factor affecting the data transmission time. When the packet error rate is relatively high, data retransmission will cause additional transmission delays. Therefore, when calculating the task transmission delay, the packet error rate needs to be corrected to reflect its actual impact on the task execution time.

[0051] In addition, the data transmission of the task is also affected by the round-trip delay. In the task scenario with a large amount of data, a high round-trip delay will significantly increase the transmission time of the task. Therefore, in the method of the present invention, the estimation of the task execution time not only considers the computing overhead, but also combines the dynamic characteristics of network transmission to ensure the accuracy of the estimation result. In the process of calculating the task execution time, it is necessary to comprehensively calculate the computing time and the data transmission time to obtain the complete execution delay of the task on different RedCap modules. First of all, the computing time is directly related to the task computing volume and the computing power of the RedCap module. When the computing power of the RedCap module is strong, the computing time is short, and vice versa, the computing time will be long. Therefore, when scheduling tasks, it is necessary to preferentially select RedCap modules with strong computing power and low current load to reduce the computing time. On the other hand, the data transmission time is affected by factors such as bandwidth, packet error rate, and round-trip delay. In order to reduce the data transmission time, the system needs to select RedCap modules with higher bandwidth, lower packet error rate, and shorter round-trip delay to execute tasks as much as possible. Especially in large-scale data processing tasks, the data transmission time often occupies an important part of the total task execution time. Therefore, in the task scheduling process, it is necessary to dynamically adjust the weights of computing and communication to ensure the balance between computing and communication overheads, so that the task can be completed within the optimal time range.

[0052] Step 4: Construct a multi-dimensional decision matrix to quantify the execution benefits of each task on each RedCap module node;

[0053] In the process of task scheduling, it is first necessary to clarify the core factors affecting the execution efficiency of tasks, including task execution latency, computing energy consumption, communication energy consumption, load balance of nodes, and network link quality, etc. Execution latency is one of the key indicators affecting task scheduling. In the RedCap module, due to its relatively limited computing power, the task execution latency may be restricted by both computing resources and network bandwidth. Therefore, in the process of constructing the decision matrix, the task execution latency is modeled as an exponential decay function to ensure that scheduling schemes with shorter latency obtain higher benefit values. At the same time, in order to reduce the energy consumption of the system while ensuring computing efficiency, the decision matrix also needs to consider the impact of computing energy consumption and communication energy consumption. In the RedCap module, the computing energy consumption is proportional to the CPU frequency, while the communication energy consumption is closely related to the data transmission time and transmission power. Therefore, when constructing the decision matrix, it is necessary to normalize the computing energy consumption and communication energy consumption so that they can have the same dimension as indicators such as execution latency, thus realizing fair comparison between different indicators. In addition, the load balance of the RedCap module is also an important factor affecting task scheduling. Since the computing resources of the RedCap module are limited, if the load of some nodes is too high, it may lead to an increase in the task queuing waiting time and even resource overflow. Therefore, in the process of constructing the decision matrix, it is necessary to model the load variance of nodes to ensure that tasks can be evenly distributed among different RedCap module nodes, thereby avoiding uneven use of computing resources and improving the overall computing efficiency of the system. At the same time, in order to further improve the accuracy of task scheduling, the method of the present invention also introduces influencing factors of network link quality, including parameters such as bandwidth, latency, packet error rate, and channel switching frequency, and integrates these network characteristics into the decision matrix by means of weighted product, so that RedCap module nodes with better network conditions can obtain higher scores in the scheduling process to improve the task execution success rate and data transmission efficiency.

[0054] Step 5: Based on the multi-dimensional decision matrix, use the Lagrangian relaxation method to solve the optimal task allocation scheme.

[0055] In the optimization process of task scheduling, it is first necessary to define the allocation relationship between tasks and nodes and construct an optimization objective to maximize the computational efficiency of the overall system while ensuring the reasonable use of computing resources and communication resources. Since the computing power of RedCap modules is relatively limited, if all tasks are scheduled to the node with the strongest computing power, this node may reduce the overall task throughput due to overload. Therefore, a computing resource constraint needs to be introduced in the optimization objective to ensure that tasks can be evenly distributed among different RedCap module nodes. In addition, since the execution of tasks is affected not only by computing power but also by network factors such as network bandwidth, round-trip delay, and packet error rate, a communication resource constraint also needs to be added to the optimization objective to ensure that task scheduling can meet the requirements of both computing and communication, thereby improving the execution success rate of tasks and reducing task delays and failures caused by network bottlenecks.

[0056] In the actual solution process, since the task scheduling problem is usually a discrete optimization problem, directly solving it may lead to too high computational complexity. Especially when the number of tasks is large, it is difficult for traditional exhaustive search methods to obtain the optimal solution within a reasonable time. To solve this problem, the present invention adopts the Lagrangian relaxation method. This method relaxes the constraint conditions of the task scheduling problem into an unconstrained optimization problem by introducing Lagrange multipliers, enabling the task allocation decision to gradually converge to the optimal solution through iterative solution. During the solution process, the Lagrange multipliers will be dynamically adjusted according to the current task allocation situation to ensure that the computing resource and communication resource constraints of tasks can be met while maximizing the computing benefits of the system. Compared with traditional integer programming solution methods, the Lagrangian relaxation method can effectively reduce the computational complexity, making the optimization process of task scheduling more efficient and applicable to edge computing environments with limited computing power such as RedCap modules.

[0057] During the optimization process, the priority issue of tasks also needs to be considered. Since the urgency of different tasks is different, in some scenarios, high-priority tasks need to be preferentially scheduled to RedCap module nodes with stronger computing capabilities and better network conditions, while low-priority tasks can be appropriately delayed. Therefore, in the optimization objective of the Lagrangian relaxation method, the present invention adopts a dynamic weight adjustment mechanism, that is, according to the priority of tasks, the benefit values of different tasks are weighted, so that high-priority tasks can obtain greater weights during the scheduling process, thus ensuring that critical tasks can be completed in the shortest time. At the same time, in order to prevent low-priority tasks from waiting for a long time, the method of the present invention also introduces a priority promotion strategy based on the waiting time of tasks, that is, when the waiting time of a certain task exceeds a certain threshold, its priority will gradually increase over time, so as to ensure that all tasks can be scheduled and executed within a reasonable time range, and prevent some tasks from not being processed for a long time due to low priority. During the optimization process of task scheduling, the load balance of the system also needs to be considered. If the computing load of some RedCap module nodes is too high while the computing resources of other nodes are still idle, the computing efficiency of the overall system will be affected. Therefore, in the optimization process of the Lagrangian relaxation method, the present invention adopts a load balance constraint, that is, during the task scheduling process, the load variance of computing nodes needs to be kept within a certain range to ensure that computing tasks can be evenly distributed among different RedCap module nodes, thereby avoiding the reduction of computing efficiency due to overload of some nodes and improving the computing throughput of the entire system. In addition, in order to enhance the stability of task scheduling, the method of the present invention also adopts an iterative optimization strategy based on the subgradient method, that is, during the optimization process, the system will dynamically adjust the task allocation scheme according to the current task allocation situation, so that the utilization rates of computing resources and communication resources of tasks can be continuously optimized, thereby improving the overall efficiency of task scheduling.

[0058] Embodiment 2: In step 1, the effective computing capacity of each RedCap module node at each moment is calculated through the following formula:

[0059] ;

[0060] where is the effective computing capacity of RedCap module node at moment ; is the CPU frequency of RedCap module node ; is the CPU utilization rate of RedCap module node at moment , and the value range is (0, 1); Is the RedCap module node At the moment The system overhead coefficient, which represents the proportion of computing resources consumed for system maintenance; Is the RedCap module node At the moment The temperature; Is the temperature threshold. When the temperature exceeds this value, the computing performance begins to decline significantly, and its value range is from 70 to 85; Is the temperature scaling factor, which controls the steepness of the impact of temperature on performance; Is the RedCap module node At the moment The available memory; Is the RedCap module node The total memory capacity.

[0061] Specifically, the numerator part of the formula Is mainly used to describe the theoretical computing power of the node. Among them, Represents the CPU frequency, usually in Hz, which determines the number of computing instructions that can be executed per unit time. In most traditional computing power modeling methods, the CPU frequency is one of the most core parameters determining computing power. However, in the RedCap device environment, only considering the CPU frequency is not enough because RedCap devices are often in a state of dynamic load changes, and some computing resources may be occupied by system processes. Therefore, this formula introduces , that is, the CPU utilization rate, whose value range is between , which represents the available proportion of CPU resources at the current moment. A higher CPU utilization rate means that the computing resources of the node have been occupied by more tasks, so the actual available computing power will be reduced accordingly. In addition, RedCap devices usually need to perform system maintenance tasks, such as network protocol management, data cache update, etc., and these tasks will consume a part of computing resources. Therefore, the formula also introduces , that is, the proportion of computing resources occupied by system maintenance. When Is higher, it means that the computing resources of this node are occupied by more system tasks, thus reducing the actual computing power available for task calculation. Therefore, the formula uses As a correction coefficient to reflect the impact of system maintenance tasks on computing power.

[0062] Secondly, the denominator part of the formula Mainly used for temperature decay modeling. During the operation of an actual RedCap device, temperature is one of the key factors affecting computing power. High temperature will lead to an increase in the power consumption of the chip, which in turn triggers the dynamic frequency scaling (DVFS) mechanism, causing the CPU frequency to decrease. In extreme cases, it may even trigger the thermal protection mechanism, resulting in a sharp decline in computing power. Therefore, in the computing power modeling of the present invention, a temperature impact factor based on the Logistic function is adopted, such that the computing power is hardly affected at low temperatures, while when the temperature exceeds a certain threshold (usually between and degrees), the computing power will decline exponentially with the increase in temperature. The parameter controls the steepness of this temperature impact. The smaller the value, the more sensitive the computing power is to temperature changes. The larger the value, the gentler the temperature impact. In this way, the system can accurately evaluate the computing power of RedCap devices under different temperature conditions and avoid high-temperature nodes during task scheduling to improve the overall system stability. Finally, the last term of the formula is used to characterize the impact of memory availability on computing power. In the computing environment of RedCap devices, tasks usually involve data access, and the availability of memory directly affects the execution efficiency of tasks. When the available memory is scarce, the computing process of tasks may be affected by memory swapping (swap), and may even lead to task failure. Therefore, in the computing power modeling of the present invention, the memory occupancy ratio is introduced as a correction factor. To avoid drastic fluctuations in computing power with the memory occupancy situation, the square root function is used for smoothing in this formula, such that the computing power slightly increases under high memory availability conditions, while significantly decreases under low memory availability conditions, thereby ensuring that task scheduling can preferentially select RedCap nodes with more available memory to improve the successful execution rate of tasks.

[0063] Embodiment 3: In step 2, the RedCap link feature vectors of each RedCap module node are extracted through the following formula:

[0064] ;

[0065] wherein, is the RedCap link feature vector of RedCap module node at time , which contains 4 dimensions; is the effective bandwidth of RedCap module node at time ; is the RedCap link feature vector of RedCap module node at time Jitter; is the reference jitter value; is the RedCap module node at time round-trip delay; is the maximum round-trip delay acceptable by the system; is the RedCap module node at time physical movement speed. For a fixed RedCap module node, this value is 0; is the reference physical movement speed value; is the RedCap module node at time signal-to-noise ratio; is the system noise floor signal-to-noise ratio; is the RedCap module node at time channel quality indicator, with a value range of 0 to 15; is the RedCap module node at time packet error rate; is the RedCap module node at time channel switching frequency; is the packet error impact coefficient, with a value range of 5 to 100; is the packet error switching frequency impact coefficient, with a value range of 2 to 50.

[0066] Specifically, the first item of the link feature vector is mainly used to describe the bandwidth status of the RedCap device and its jitter impact. Among them, represents the device at time effective bandwidth, which determines how fast the device can transmit data. However, in a wireless communication environment, only considering bandwidth is not enough, because network jitter may cause instability in the packet transmission time, thereby affecting the task execution time. Therefore, this formula introduces an exponential decay factor to describe the impact of jitter on bandwidth. When the jitter is small, the attenuation impact of this item on bandwidth is small, and when the jitter is large, the exponential term approaches zero, thereby reducing the link quality score of this node. In this way, the formula can more accurately reflect the effectiveness of bandwidth at different jitter levels and improve the preference of task scheduling for high-stability communication paths. Secondly, the second item of the link feature vector is mainly used to describe the impact of round-trip delay (RTT) and device physical movement speed (V). Round-trip delay Represents the round-trip time of data packets between the RedCap device and the remote server, which is a key metric for measuring network latency. To make this metric more adaptable, the formula uses normalization to make it relative to the maximum acceptable round-trip latency of the system for comparison, thus ensuring that tasks preferentially select low-latency links. At the same time, considering that the RedCap device may be in a mobile state, the physical movement speed will also affect the stability of data transmission. For example, when the device is in a high-speed motion state, the channel quality will become unstable, and the volatility of the round-trip latency will increase. Therefore, this formula introduces an exponential decay factor to ensure that devices with faster movement speeds are given lower link scores during task scheduling to reduce the impact of high-dynamic environments on task execution.

[0067] The third dimension is mainly used to measure the impact of signal quality on link performance. In a wireless communication environment, the signal-to-interference-plus-noise ratio (SINR) is an important factor determining communication quality, which reflects the ratio of the signal strength to the environmental noise. When the SINR is too low, the error rate of data transmission will increase, resulting in retransmissions and increased latency. Therefore, this formula uses as the normalized signal quality metric, enabling the SINR to be reasonably quantified under different background noises. In addition, the channel quality indicator (CQI) is also an important parameter for measuring wireless communication quality, and its value range is from 0 to 15. The higher the value, the better the channel state. Therefore, this formula normalizes to between so that this term can accurately reflect the signal quality of the device and preferentially select RedCap devices with better signal quality for data transmission during the task scheduling process. Finally, the fourth term of the link feature vector is mainly used to characterize the stability of data transmission, that is, the impact of the packet error rate (PER) and the channel switching frequency (H). The packet error rate reflects the probability of a device losing data packets during wireless communication. A higher packet error rate means that more data retransmissions are required, which increases the task execution time and the energy consumption of the system. Therefore, the formula uses this non-linear decay function, so that when the packet error rate is low, this term is close to 1, and when the packet error rate is high, this term drops rapidly, thereby reducing the link score of this device. Similarly, the channel switching frequency reflects the number of times a device switches between different channels. A higher switching frequency usually means the instability of the communication link, which is likely to cause packet loss or delay. Therefore, the formula uses as a correction term to ensure that the task scheduling process is more inclined to select devices with stable channels to execute computing tasks.

[0068] Example 4: In step 3, the execution time on different RedCap module nodes is estimated through the following formula:

[0069] ;

[0070] where is the execution time of task on RedCap module node at time ; is the number of computational operations required for task ; is the computational complexity coefficient of task ; is the input data volume of task ; is the output data volume of task ;

[0071] Specifically, the first term of the formula calculates the pure computational time of the task on the RedCap device. Among them, represents the number of computational operations required for task , usually measured in floating-point operation counts (FLOP), reflecting the computational volume of the task itself. Tasks with a higher number of computational operations usually require longer computational time. Therefore, plays a role in linear growth in the computational time. To more accurately characterize the computational requirements of different types of tasks, the computational complexity coefficient is introduced in the formula to describe the algorithmic complexity of the task on the RedCap device. For some tasks, even if the number of computational operations is the same, due to complex data dependencies, matrix operations, or non-linear transformations, there may be significant differences in computational efficiency. Therefore, as an adjustable parameter enables the estimation of computational time to be more in line with the actual computational overhead of the task. The computational power of the RedCap device is a dynamic value affected by factors such as CPU frequency, system load, temperature, and memory occupancy. Therefore, this formula uses as the computational power indicator, which represents the actual available computational power of the RedCap device at time . In the estimation of computational time, is used as the denominator, enabling RedCap devices with higher computational power to provide shorter computational times for the same computational task, thereby improving the efficiency of task execution. This computational power value has been accurately defined in the computational power modeling in step 1. Therefore, in the estimation of task execution time, it can ensure the reasonable utilization of computational resources, enabling task scheduling to fully consider the differences in computational power of different nodes and avoiding overloaded nodes from affecting computational efficiency.

[0072] Next, the second part of the formula calculates the data transmission time of the task on the RedCap device. In the edge computing scenario, tasks usually need to transfer data between the RedCap device and the cloud or other computing nodes. Therefore, the impact of communication resources on the task execution time cannot be ignored. Here,[[]] represents the input data volume of the task , represents the output data volume of the task . The sum of the two reflects the total data transmission requirement of the task. In 5G RedCap devices, due to limited bandwidth, data transmission may become a bottleneck in task execution. Therefore, when estimating the execution time, it is necessary to comprehensively consider the bandwidth and packet error rate 's impact on the data transmission rate. The first term of the data transmission time represents the theoretical data transmission duration of the task, where represents the current moment RedCap device 's available bandwidth, which determines the rate at which data can be transmitted. However, in a wireless communication environment, errors and packet losses may occur during data transmission, and data retransmission is required. Therefore, this formula introduces as a correction factor in the bandwidth term to ensure that RedCap devices with a higher packet error rate exhibit a higher communication delay in the task execution time estimation, so that the task scheduling can automatically avoid links with a high error rate to improve the task execution success rate. The second term of the data transmission time is used to correct the impact of network round-trip delay (RTT) on data transmission. During the actual task execution process, the data transmission time not only depends on the bandwidth but is also affected by network delay. The network link quality of the RedCap device may be affected by factors such as channel interference, device mobility, and network load, resulting in a longer round-trip delay of data packets and thus reducing the overall efficiency of data transmission. Therefore, when calculating the data transmission time in this formula, an RTT correction term is introduced to describe the impact of the round-trip time of data packets between the RedCap device and the server on the data transmission rate. When is small, this correction term is close to 1 and has little impact on the transmission time; while when is large, this correction term will increase significantly, making the task execution time calculation more in line with the actual network environment.[[]]

[0073] Example 5: The computational complexity coefficient of the task is calculated using the following formula:[[]]

[0074] ;

[0075] in, For the task The basic complexity coefficient is determined by the asymptotic complexity of the task: for O(1) complexity tasks, ; For O(logn) complexity tasks, ; For O(n) complexity tasks, ; For O(nlogn) complexity tasks, ; For O(n²) complexity tasks, ; For O(n³) complexity tasks, ; For the task The working set memory size, that is, the amount of memory frequently accessed during task execution; For the task Total memory requirements; is the memory impact factor, with a value of 0.2 to 0.5, reflecting the impact of memory access on computing performance; For the task The data dependency of is 0 or 1, which indicates the dependency between computing units within the task: 0 indicates complete parallelism and no dependency; 1 indicates complete serialization and strong dependency; is the dependency impact factor, with a value range of 0.5 to 1.5; is the instruction mode coefficient, which is determined according to the instruction characteristics of the task: for computationally intensive tasks, The value range is 0.8 to 1.0; for memory-intensive tasks, The value range is 1.2 to 1.5.

[0076] Specifically, the basic part of the computational complexity is given by The parameter is set according to the asymptotic time complexity of the task. Since the computational complexity of the task hardly increases with the input size, we set , that is, the computational complexity is the lowest. As for the logarithmic time complexity For tasks such as binary search or hash lookup, the computational complexity is slightly higher than constant time, so we set To appropriately improve the computational complexity estimate. For linear time complexity For tasks such as simple traversal or linear search, the computational complexity grows proportionally with the input size, so setting For linear logarithmic time complexity For tasks such as quick sort or merge sort, the computational complexity is between linear and quadratic complexity, so set For higher order complexity, such as quadratic complexity , cubic complexity and higher-order complexity tasks. Since the computational amount increases sharply with the increase of the input scale, they are respectively set to 3.0 and 5.0 to more accurately reflect the consumption of these tasks on computing resources.

[0077] However, relying solely on the asymptotic complexity of the algorithm to estimate the computational complexity of tasks is insufficient because during actual execution, the computational complexity of tasks is also affected by the memory access pattern. Therefore, the present invention introduces a memory impact factor, and the second term of the formula quantifies the impact of the memory access overhead of the task on the computational complexity. Among them, represents the working set size of the task, that is, the size of the memory area frequently accessed during the calculation, while represents the total memory requirement of the task. If is close to , it means that the calculation of this task involves a large amount of memory access, which may lead to cache misses and thus increase the calculation time. Therefore, this formula adopts the normalization ratio as a correction factor for the computational complexity and introduces a parameter for weight adjustment, with a value range of 0.2 to 0.5, to ensure that the impact of the memory access pattern on the computational complexity is not too large but can sufficiently reflect the differences in the computational efficiency of different tasks at the CPU-cache-memory level.

[0078] In addition to the memory access pattern, the data dependence within the task is also an important factor affecting the computational complexity. Some tasks can be fully parallelized during the calculation, while others have strong dependencies and need to be executed in a strict order, which directly affects the computational efficiency of the task. Therefore, the third term of the formula quantifies the impact of the parallelism of the task on the computational complexity. The data dependence degree takes a value of 0 or 1, where 0 indicates no data dependence and can be fully parallel calculated, such as tasks like matrix addition and vector operations, while 1 indicates that the task is completely serial and has strong data dependence, such as matrix Gaussian elimination and some dynamic programming tasks. In the estimation of computational complexity, if the dependence degree of the task is high, its computational complexity will increase accordingly. Therefore, this term is adjusted by (value range 0.5 to 1.5) so that tasks with stronger data dependence have higher computational complexity, while fully parallel tasks have lower computational complexity, thereby optimizing the execution efficiency of tasks on the RedCap device.

[0079] Finally, the fourth term of the formula Then, it is used to consider the instruction mode of the task, that is, how the calculation type of the task affects the calculation complexity. Different tasks may be more computationally intensive or memory-intensive, and there are significant differences in the consumption of computing resources when these tasks are actually executed. Computationally intensive tasks mainly rely on the CPU to perform floating-point operations or integer operations, and have a high instruction execution efficiency. Therefore, the instruction mode coefficient ranges from 0.8 to 1.0. For memory-intensive tasks, such as database queries and large-scale data sorting, the execution time of the task is mainly affected by the memory access overhead, and the instruction execution efficiency is low. Therefore, it has a higher value range (1.2 to 1.5) to accurately reflect the calculation complexity characteristics of such tasks.

[0080] Example 6: In step 4, a multi-dimensional decision matrix is constructed through the following formula to quantify the execution benefit of each task on each RedCap module node:

[0081] ;

[0082] where, is the execution benefit of task on RedCap module node at time ; is the priority weight of task ; is the deadline constraint of task ; is the energy required for task to be calculated on RedCap module node ; is the energy required for task to communicate on RedCap module node at time ; is the load variance of RedCap module node at time , reflecting the load stability; is the influence factor of the th dimension of the link quality eigenvector.

[0083] Specifically, the numerator part of the formula is mainly used to describe the time sensitivity of the task and the influence of the task priority weight on the execution benefit. Among them, represents task The priority weight is used to distinguish the importance of different tasks. For example, in the edge computing scenario, some tasks (such as telemedicine and driverless control instructions) have high requirements for real-time performance, while some tasks (such as data batch processing and log analysis) have lower requirements for execution latency. Therefore, this formula allows the scheduling system to set different priority weights for different tasks, so that high-priority tasks can obtain more computing resource support during the scheduling process. The execution time of the task is jointly determined by the computing time and the data transmission time, and it is one of the key factors affecting the execution efficiency of the task. In order to make the deadline constraint of the task have an appropriate impact on the scheduling decision, this formula adopts an exponential decay function , which is used to quantify the impact of the task execution time on the efficiency. When the execution time of the task is short (i.e., ), this term is close to 1, indicating that the task can be completed within a sufficient time range without affecting the overall execution efficiency; while when the execution time of the task approaches or even exceeds its deadline (i.e., ), this term approaches 0, meaning that the value of executing this task on the current node is low, thereby reducing the selection probability of this node in task allocation. Through this time decay mechanism, the present invention can effectively avoid the execution failure of timeout tasks and improve the overall completion rate of tasks.

[0084] Secondly, the denominator part of the formula is mainly used to describe the energy consumption overhead of the task and the load stability of the node, ensuring the low energy consumption and load balancing of task scheduling. Among them, represents the computing energy consumption of task on the RedCap device . Usually, the computing energy consumption is proportional to the square of the CPU frequency, that is, devices with high computing power tend to consume more energy. Therefore, during the task scheduling process, if the computing energy consumption of a certain RedCap device is too high, the system will reduce the scheduling of tasks on this device to reduce the energy consumption of the entire system and improve the sustainability of computing. At the same time, represents the communication energy consumption of the task, which is mainly affected by bandwidth, data volume, packet error rate, and transmission distance. In the edge computing environment of 5G RedCap devices, the data transmission cost of tasks cannot be ignored. Especially in the case of limited bandwidth, the communication energy consumption may be much higher than the computing energy consumption. Therefore, this formula comprehensively considers the computing energy consumption and communication energy consumption to ensure that the execution efficiency of tasks not only depends on the computing power but also is restricted by communication conditions. In addition, reflects the load balancing state of the RedCap device. Among them, represents the device at time The load variance, which is used to measure the load stability of the current RedCap device. A higher load variance means that the usage of the device's computing resources fluctuates greatly, and there may be resource shortages at certain moments, resulting in task execution failures. Therefore, in this formula, is used as the load balancing penalty term, so that devices with larger load fluctuations are given lower execution benefits during task scheduling, thus preferentially selecting RedCap devices with more stable loads to execute tasks. This method can effectively improve the global load balancing of task scheduling and avoid individual devices being overloaded and affecting the overall computing efficiency of the system. Finally, the last part of the formula quantifies the impact of the link quality of the RedCap device on the task execution benefit. Among them, represents the RedCap device at time link feature vector, which includes multiple key indicators such as bandwidth, delay, signal quality, and packet error rate. Different link quality characteristics have different degrees of influence on task scheduling. Therefore, a weight factor is introduced in the formula to adjust the contribution degree of different network parameters to the task execution benefit. For example, in some tasks (such as high-throughput data transmission tasks), bandwidth may be the most important factor, so the value of bandwidth-related parameters is relatively high; while for low-latency tasks (such as real-time video analysis and remote control tasks), round-trip delay may be the decisive factor, so the value of delay-related parameters is relatively large. Through the weighted product of the link feature vector, the present invention can ensure that the impact of the communication network is fully considered during task scheduling, and preferentially select RedCap devices with low latency, high bandwidth, low packet error rate, and good signal quality, thereby improving the task execution success rate and the communication efficiency of the overall system.

[0085] Embodiment 7: Calculated using the following formula:

[0086] ;

[0087] Among them, is the energy efficiency coefficient of the RedCap module node .

[0088] Specifically, the core part of this formula is , which describes the computing energy consumption characteristics of the RedCap device. Among them, represents the energy efficiency coefficient of the RedCap device , which is a parameter related to the hardware characteristics of the device and is used to describe the energy consumption sensitivity of the device. Different RedCap devices may have different processor architectures and power management strategies. Therefore, It may vary depending on the device design. For example, high-performance RedCap devices may adopt more advanced low-power processors, making their relatively small, while some older or less energy-efficient devices may have a higher , meaning higher energy consumption when performing the same tasks. During the task scheduling process, the system can optimize and adjust the energy consumption according to differences, preferentially select nodes with lower energy consumption to execute computing tasks, thereby reducing the energy consumption of the entire computing network. In addition to , the in the formula reflects the impact of CPU frequency on computing energy consumption. In most computing devices (including RedCap devices), the computing energy consumption is usually proportional to the square of the CPU frequency, that is . This is due to the action of the dynamic voltage and frequency regulation mechanism. When the CPU runs at a high frequency, not only does the computing performance improve, but its power consumption also increases exponentially. Therefore, during the task scheduling process, if you want to reduce the computing energy consumption, the system can select a RedCap device with a low frequency to execute the task to reduce energy consumption. However, too low a CPU frequency may cause the task execution time to be too long. Therefore, the method of the present invention needs to comprehensively weigh the impact of CPU frequency and computing energy consumption during scheduling to ensure that the task can be executed with low energy consumption and will not fail due to too low a CPU frequency.

[0089] In addition to the computing characteristics of the device itself, the computing requirements of the task are also a decisive factor in computing energy consumption. Therefore, the latter part of this formula quantifies the computing volume of the task. Here, represents the computing operand of the task , usually measured in floating-point operation counts (FLOP), indicating the basic computing volume required during the task execution. The larger the computing operand, the more computing steps the task needs to execute, so its computing energy consumption is also higher. And represents the computing complexity coefficient of the task, which is determined by factors such as the algorithm complexity, data dependence, and memory access pattern of the task, and has been defined in detail in Embodiment 5. The computing complexity of different tasks is different. Even if the computing operands are the same, the computing energy consumption of the tasks may vary greatly. For example, compute-intensive tasks (such as deep learning inference, matrix operations) usually consume more computing energy than I / O-intensive tasks (such as database queries, log processing). Therefore, the value may vary significantly in different types of tasks. By incorporating into the computing energy consumption model, the present invention makes the energy consumption estimation more accurate, which helps the task scheduling system make more intelligent decisions when allocating tasks.

[0090] Example 8: It is calculated using the following formula:

[0091]

[0092] where is the energy efficiency coefficient of the RedCap module node ; is the transmission power of the RedCap module node ; is the reception power of the RedCap module node ;

[0093] Specifically, the formula consists of two parts. The first term calculates the uplink transmission energy consumption of the task, that is, the energy consumed when the data output of the task is transmitted from the RedCap device to the remote server or other computing nodes; the second term calculates the downlink transmission energy consumption of the task, that is, the energy consumed when the data input of the task is transmitted from the external server or other computing nodes to the RedCap device. The sum of the two parts is the total communication energy consumption of the task. The core parts and of the formula represent the transmission power and reception power of the RedCap device respectively, usually in watts (W). The transmission power of the RedCap device determines the energy consumption required for the device to transmit data on the wireless link, while the reception power reflects the power consumption of the device when receiving data transmitted from the remote server. Since the RedCap device usually adopts a low-power wireless communication protocol, and are relatively low compared to traditional 5G devices, but during long-term operation or large-scale data transmission, their cumulative energy consumption may still be high. Therefore, during the task scheduling process, it is necessary to accurately evaluate the data transmission requirements of the task and ensure that the task can be preferentially allocated to devices with lower communication energy consumption for execution to reduce the overall energy consumption of the system. Secondly, and in the formula are used to calculate the uplink and downlink data transmission times of the task respectively. Among them, and represent the output data volume and input data volume of the task respectively, which determine the amount of data that needs to be transmitted between the RedCap device and the remote server or other computing nodes. The bandwidth represents the available communication bandwidth of the RedCap device at time , which determines the rate at which data can be transmitted, and Denoted as the packet error rate, it is used to characterize the reliability of the wireless channel. Since there is a certain packet loss rate in data transmission in the wireless communication environment and retransmission is required, a correction factor is introduced in the bandwidth term in this formula , which is used to compensate for the impact of packet errors on the transmission time. When the packet error rate is high, the actual data throughput decreases, resulting in an increase in the data transmission time, thereby increasing the communication energy consumption. Therefore, during the task scheduling process, the system should preferentially select devices with a lower packet error rate for task scheduling to reduce the energy consumption overhead caused by data retransmission.

[0094] Example 9: In step 5, based on the multi-dimensional decision matrix, the Lagrangian relaxation method is used to solve the optimal task allocation scheme through the following formula:

[0095] ;

[0096] where, is the objective function; is the task allocation matrix, represents the task assigned to the RedCap module node otherwise ; is the Lagrange multiplier of the computational resource capacity constraint of the RedCap module node , is the Lagrange multiplier vector of the resource capacity constraint; is the Lagrange multiplier of the allocation constraint of the task , is the Lagrange multiplier vector of the allocation constraint; is the total number of tasks; is the total number of RedCap module nodes; the final scheduling decision is obtained by solving the following optimization problem:

[0097] ;

[0098] where, and are respectively the Lagrange multiplier vector of the optimal computational resource capacity constraint and the Lagrange multiplier vector of the optimal allocation constraint obtained by iterative solution using the subgradient method; is the optimal task allocation matrix.

[0099] Specifically, the core objective of the method of the present invention is to maximize the execution benefit of tasks, that is, to ensure that tasks are assigned to the most suitable RedCap module nodes, enabling them to be completed in the shortest time and minimizing computational energy consumption and communication overhead as much as possible. When constructing the optimization objective, the execution benefit obtained when each task is executed on different nodes is first considered. The execution benefit is a comprehensive indicator that measures the execution efficiency of a task on a specific node. It not only depends on the priority of the task but is also affected by the execution time, computational energy consumption, communication energy consumption, and device load balance of the task. To enable task scheduling to adapt to the characteristics of different tasks, the present invention introduces a task execution benefit function into the optimization objective, such that task schemes with high priority, low execution latency, and low energy consumption obtain greater weights during the optimization process. However, task scheduling must not only pursue the highest execution benefit but also meet the constraints of computational resources. Since the computational capacity of each RedCap module node is limited, when tasks are assigned, no node can be burdened with tasks exceeding its computational capacity, otherwise, task execution failures or computational resource congestion will occur. Therefore, the present invention introduces a computational resource capacity constraint to ensure that the computational requirements of the tasks assigned to each node do not exceed its currently available computational capacity. This constraint is related to the computational operation count, computational complexity coefficient of the task, and the available computational capacity of the node, ensuring that the task scheduling system always considers the constraints of computational resources when optimizing the task assignment scheme, thereby avoiding the unbalanced use of computational resources. To solve the optimization problem under the constraint conditions, the present invention adopts the Lagrange multiplier method, introducing the computational resource constraint into the optimization objective in a relaxed form, enabling dynamic adjustment of the task load distribution during the optimization process and ensuring the rational use of computational resources.

[0100] In addition to the limitation of computing resources, task scheduling must also satisfy the constraint of unique task allocation, that is, each task must and can only be allocated to one RedCap module node for execution. This constraint is crucial in the task scheduling process because if a task is not allocated to any node, the task execution fails, and if a task is allocated to multiple nodes, it will cause waste of computing resources. Therefore, the method of the present invention introduces the constraint of unique task allocation into the optimization objective and relaxes it through the Lagrangian multiplier method, enabling the task scheduling to dynamically adjust the task allocation scheme during the optimization process to ensure that each task can obtain a unique computing node for execution while maximizing the execution benefit of the overall task. When solving the optimization problem, the present invention uses the subgradient method for iterative solution. Since the task scheduling problem is a discrete optimization problem, the traditional integer programming method has a high computational complexity and is difficult to efficiently solve under the limited computing power of the RedCap module. The advantage of the Lagrangian relaxation method is that it can transform the original discrete optimization problem into a continuous optimization problem by introducing Lagrangian multipliers, making the optimization calculation more efficient. The subgradient method gradually approaches the optimal solution by continuously adjusting the value of the Lagrangian multiplier and finally converges to the optimal task allocation scheme that satisfies the computing resource and unique task allocation constraints. During the optimization process, the Lagrangian multiplier will be dynamically adjusted according to the computing requirements of the current task, the available computing resources of the node, and the priority of the task, so as to ensure that the optimization result can not only guarantee the reasonable allocation of computing resources but also make the task scheduling strategy conform to the overall optimization objective of the system. Finally, through the construction of the optimization objective, the introduction of the Lagrangian relaxation method, and the solution of the subgradient method, the method of the present invention can achieve the optimal task scheduling scheme on the premise of ensuring the reasonable allocation of computing resources. Compared with the traditional task scheduling method, the advantage of this method is that it can optimize multiple objectives simultaneously, enabling the task scheduling scheme to not only ensure the real-time performance of task execution but also achieve a balance among computing resources, communication resources, and energy consumption. In addition, since this method adopts a dynamic optimization strategy, it can be dynamically adjusted according to the computing state of the RedCap module, the network environment, and the priority of the task, thereby ensuring that the task scheduling strategy can adapt to different computing environments and improving the overall throughput and computing efficiency of the system.

[0101] Although the specific implementation manners of the present invention are described above, those skilled in the art should understand that these specific implementation manners are only examples. Without departing from the principles and essence of the present invention, those skilled in the art can make various omissions, substitutions, and changes to the details of the above methods and systems. For example, combining the above method steps so as to perform substantially the same function in a substantially the same manner to achieve substantially the same result falls within the scope of the present invention. Therefore, the scope of the present invention is only defined by the appended claims.

Claims

1. A dynamic weighted edge computing scheduling method based on 5G RedCap module, characterized in that: The method comprises: Step 1: Perform node resource perception on each RedCap module node in the network and calculate the effective computing capacity of each RedCap module node at each moment; Step 2: By analyzing the dynamic characteristics of the RedCap link of each RedCap module node in the network, extract the RedCap link feature vector of each RedCap module node; Step 3: Model the resource requirements of each task and estimate the execution time on different RedCap module nodes; Step 4: Construct a multi-dimensional decision matrix to quantify the execution benefits of each task on each RedCap module node; Step 5: Based on the multi-dimensional decision matrix, use the Lagrangian relaxation method to solve the optimal task allocation solution; In step 1, the effective computing capacity of each RedCap module node at each moment is calculated using the following formula: ； in, For RedCap module nodes At the moment Effective computing power; For RedCap module nodes CPU frequency; For RedCap module nodes At the moment The CPU utilization rate is in the range of (0,1). For RedCap module nodes At the moment The system overhead factor represents the proportion of computing resources consumed by system maintenance; For RedCap module nodes At the moment Temperature; The temperature threshold is the value above which computing performance begins to decline significantly. The value range is 70 to 85. is the temperature scaling factor, which controls the steepness of the effect of temperature on performance; For RedCap module nodes At the moment Available memory; For RedCap module nodes Total memory capacity; In step 2, the RedCap link feature vector of each RedCap module node is extracted using the following formula: ； in, For RedCap module nodes At the moment The RedCap link feature vector contains 4 dimensions; For RedCap module nodes At the moment The effective bandwidth of For RedCap module nodes At the moment jitter; is the reference jitter value; For RedCap module nodes At the moment The round trip delay of is the maximum round-trip delay acceptable to the system; For RedCap module nodes At the moment The physical movement speed of the RedCap module node is 0. is the reference physical movement speed value; For RedCap module nodes At the moment signal-to-noise ratio; is the system noise floor signal-to-noise ratio; For RedCap module nodes At the moment The channel quality index ranges from 0 to 15; For RedCap module nodes At the moment Packet error rate; For RedCap module nodes At the moment Channel switching frequency; is the packet error impact coefficient, ranging from 5 to 100; is the error packet switching frequency influence coefficient, ranging from 2 to 50; In step 3, the execution time on different RedCap module nodes is estimated using the following formula: ； in, For the task In the RedCap module node At the moment Execution time; For the task The number of computational operations required; For the task The computational complexity coefficient of ; For the task The amount of input data; For the task The amount of output data; Task The computational complexity coefficient of Calculated using the following formula: ； in, For the task The basic complexity coefficient is determined by the asymptotic complexity of the task: for O(1) complexity tasks, ; For O(logn) complexity tasks, ; For O(n) complexity tasks, ; For O(nlogn) complexity tasks, ; For O(n²) complexity tasks, ; For O(n³) complexity tasks, ; For the task The working set memory size, that is, the amount of memory frequently accessed during task execution; For the task Total memory requirements; is the memory impact factor, with a value ranging from 0.2 to 0.5, reflecting the impact of memory access on computing performance; For the task The data dependency of is 0 or 1, which indicates the dependency between computing units within the task: 0 indicates complete parallelism and no dependency; 1 indicates complete serialization and strong dependency; is the dependency impact factor, with a value range of 0.5 to 1.5; is the instruction mode coefficient, which is determined according to the instruction characteristics of the task: for computationally intensive tasks, The value range is 0.8 to 1.0; for memory-intensive tasks, The value range of is 1.2 to 1.5; In step 4, a multi-dimensional decision matrix is ​​constructed using the following formula to quantify the execution benefit of each task on each RedCap module node: ； in, For the task In the RedCap module node At the moment implementation benefits; For the task The priority weight of For the task Deadline constraints; For the task In the RedCap module node Calculate the energy required; For the task In the RedCap module node At the moment Energy required for communication; For RedCap module nodes At the moment The load variance reflects the load stability; is the link quality feature vector The impact factor of dimension; Calculated using the following formula: ； in, For RedCap module nodes Energy efficiency coefficient; Calculated using the following formula: in, For RedCap module nodes Energy efficiency coefficient; For RedCap module nodes The transmission power of For RedCap module nodes The received power; In step 5, the optimal task allocation solution is solved using the Lagrangian relaxation method based on the multi-dimensional decision matrix using the following formula: ； in, is the objective function; Assign the matrix to the task, Indicates the task Assigned to RedCap module nodes ,otherwise ; For RedCap module nodes The Lagrange multiplier constrained by the computing resource capacity, is the Lagrange multiplier vector of resource capacity constraints; For the task The Lagrange multiplier of the distribution constraint, is the Lagrange multiplier vector of the allocation constraint; is the total number of tasks; is the total number of RedCap module nodes; the final scheduling decision is obtained by solving the following optimization problem: ； in, and are the Lagrange multiplier vector of the optimal computing resource capacity constraint and the Lagrange multiplier vector of the optimal allocation constraint solved iteratively by the subgradient method, respectively; is the optimal task assignment matrix.

Images