A method and apparatus for computing power assessment and task scheduling

Abstract

This invention relates to the field of communication technology, and in particular to a method and apparatus for computing power assessment and task scheduling. The invention first collects task and device status information, constructs a model including total latency and total energy consumption, then quantifies and calculates the total latency weight and total energy consumption weight, and finally trains a strategy model based on the state feature vector and weights. The optimal task scheduling strategy is then output through the trained strategy model. This invention, by quantifying and evaluating various indicators of computing resources and combining them with the strategy model to optimize task scheduling, makes resource scheduling more accurate and efficient. It effectively solves the problems of low resource utilization, unintelligent scheduling, and poor dynamic adaptability of traditional methods. It can adapt to changes in system load and task requirements, significantly improves the efficiency of computing resource utilization, reduces latency and energy consumption, and thus optimizes the overall performance of the system.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a method and apparatus for computing power assessment and task scheduling. Background Technology

[0002] With the continuous development of information technology, the computing power demand of modern computing systems is constantly increasing, especially in the fields of big data processing, cloud computing, and artificial intelligence, making task scheduling problems increasingly complex.

[0003] Traditional task scheduling methods mostly rely on static rules or simple priority ordering, lacking the ability to dynamically adapt to resource requirements and task characteristics, which can easily lead to resource waste or system performance bottlenecks. Therefore, how to achieve efficient task scheduling in dynamic and complex computing environments has become a hot research topic.

[0004] Therefore, overcoming the shortcomings of the existing technology is an urgent problem to be solved in this technical field. Summary of the Invention

[0005] The technical problem to be solved by this invention is how to address the issues of low resource utilization, insufficient intelligence of scheduling strategies, and poor adaptability to dynamic environments in existing task scheduling methods.

[0006] The present invention adopts the following technical solution: Firstly, a method for computing power assessment and task scheduling is provided, including: Collect status information of tasks and equipment in the unmanned command system, construct status feature vectors, and establish total delay and total energy consumption models; The total delay data output by the total delay model and the total energy consumption data output by the total energy consumption model are quantitatively analyzed to obtain the total delay weight and the total energy consumption weight, respectively. The strategy model is trained based on the state feature vector, the total delay weight, and the total energy consumption weight, and the optimal task scheduling strategy is output based on the trained strategy model.

[0007] Preferably, the process of collecting the status information of tasks and equipment in the unmanned command system and constructing a status feature vector specifically includes: Based on the collected task and device status information, a status feature vector is constructed. The status information includes the task's computational requirements and data volume, as well as the device's bandwidth capacity, computing power, computational energy consumption constant, communication efficiency constant, and transmission power. ; in, For the task The computational requirements; For the task The amount of data; For the device's bandwidth capability; For computing power; To calculate the energy consumption constant; Let be the communication efficiency constant; Send power to the device, The number of terminal devices. This refers to the number of server devices.

[0008] Preferably, the total delay model includes computation delay and communication delay; ; ; ; in, For the total delay model, For the first The terminal device in the first The computational latency of each server For transmission to the first Communication latency of individual server devices; The total energy consumption model includes computational energy consumption and communication energy consumption; ; ; ; in, For the total energy consumption model, For the first The terminal device in the first The computing power consumption of a single server; For transmission to the first The communication energy consumption of each server; The effective switching capacitor coefficient of the terminal equipment; This refers to the transmission power of the terminal device.

[0009] Preferably, the step of quantitatively analyzing the total delay data output by the total delay model and the total energy consumption data output by the total energy consumption model to obtain the total delay weight and the total energy consumption weight respectively includes: Outlier removal and standardization were performed on total delay and total energy consumption to obtain standardized data. ; ; Calculate the proportion of each standardized data point in the corresponding indicator. ; ; The information entropy of total delay and total energy consumption is calculated based on the aforementioned proportion. ; ; ; Calculate the total delay weight based on information entropy. Total energy consumption weight ; ; in, Indicates the first The first of the indicators The result after data standardization; For the first in the original data The first of the indicators One data point; and These represent the first and second parts of the original data. Minimum and maximum values ​​of each indicator For the first The data in the first The proportion of each indicator For the first Information entropy of each indicator For the first The weight of each indicator, Total number of indicators, total delay weight = Total energy consumption weight = .

[0010] Preferably, training the policy model based on the state feature vector, the total delay weight, and the total energy consumption weight specifically includes: The state feature vector is input into the policy network, and the probability distribution of task allocation to each device is output. Based on the total latency weight, the total energy consumption weight, and the latency and energy consumption changes before and after the task execution action, an instant reward function is constructed. With the goal of maximizing cumulative rewards, the policy network is optimized through the clip mechanism and advantage function of the policy model; The value network is optimized based on mean squared error and discounted reward. The policy network and value network are iteratively updated until convergence is achieved to complete the training of the policy model.

[0011] Preferably, the probability distribution of task allocation satisfies: ; ; in, Indicates the action of task assignment; This represents the state feature vector at the current moment; Tasks are assigned to devices The probability of; This represents the total number of optional devices. Indicates the first The original score value corresponding to each device; and The first The device and the first The rating value corresponding to each device; Network weights; For bias; For summation index variables.

[0012] Preferably, the instant reward function satisfy: ; ; ; in, To perform the action The delay; To perform the action The delay; To perform the action Energy consumption; To perform the action Energy consumption; The total delay weight, Weighted by total energy consumption; The change in energy consumption before and after the action is performed; This refers to the change in delay before and after the action is performed.

[0013] Preferably, the policy network satisfy: ; ; ; ; in, This represents the probability ratio between the new strategy and the old strategy. Let be the policy function to be optimized at the current time step, when the model parameters are . At that time, based on the input status Output the probability distribution of the execution probability of each action; For the old policy function before the update, when the model parameters are At that time, based on the input status Output the probability distribution of the execution probability of each action; The advantage function is used to measure the action. The advantages and disadvantages; Used to limit the update range and prevent excessive policy changes; For editing parameters, As a discount factor, For value networks, As a discount reward; Indicates the first The reward value at any given moment.

[0014] Secondly, a computing power assessment and task scheduling device is provided, the computing power assessment and task scheduling device comprising: a processor and a memory for storing processor-executable instructions; The processor is configured to execute the computing power assessment and task scheduling method.

[0015] Thirdly, a non-volatile computer storage medium is provided, the computer storage medium storing computer-executable instructions, which are executed by one or more processors to perform the computing power assessment and task scheduling method described in the first aspect.

[0016] Fourthly, a chip is provided, comprising: a processor and an interface for calling and running a computer program stored in a memory, and executing the computing power assessment and task scheduling method as described in the first aspect.

[0017] Fifthly, a computer program product containing instructions is provided, which, when executed on a computer or processor, causes the computer or processor to perform the computing power assessment and task scheduling method as described in the first aspect.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention first collects task and device status information, constructs a total latency model and a total energy consumption model, then quantifies and calculates the total latency weight and total energy consumption weight, and finally trains a strategy model based on the state feature vector and weights. The optimal task scheduling strategy is then output through the trained strategy model. This invention optimizes task scheduling by quantitatively evaluating various indicators of computing resources and combining them with the strategy model, making resource scheduling more accurate and efficient. It effectively solves the problems of low resource utilization, unintelligent scheduling, and poor dynamic adaptability of traditional methods. It can adapt to changes in system load and task requirements, significantly improving computing resource utilization efficiency, reducing latency and energy consumption, and thus optimizing the overall system performance. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating a computing power assessment and task scheduling method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the training process of a strategy model provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a cumulative reward provided in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the changes in total delay and total energy consumption provided by an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a computing power assessment and task scheduling device provided in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0022] Unless the context otherwise requires, throughout the specification and claims, the term "comprising" is interpreted as openly inclusive, meaning "including, but not limited to." In the description of the specification, terms such as "one embodiment," "some embodiments," "exemplary embodiment," "example," "specific example," or "some examples" are intended to indicate that a particular feature, structure, material, or characteristic associated with that embodiment or example is included in at least one embodiment or example of this disclosure. The illustrative representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics mentioned may be included in any suitable manner in any one or more embodiments or examples; that is, although they may be incorporated into embodiments or examples using the above terms for reasons such as order and position, it does not limit them to be incorporated in combination by a single embodiment or example.

[0023] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of embodiments of this disclosure, unless otherwise stated, "a plurality of" means two or more. Furthermore, for example, the description may use the prefix "A" or "B" to describe the same type of nouns as two independent entities. In this case, the corresponding features defined with "A" and "B" are used only to distinguish between similar entities and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features.

[0024] In describing some embodiments, the terms "coupled," "coupled," and "connected," and their derivative expressions, may be used. For example, the term "connected" may be used in describing some embodiments to indicate that two or more components have direct physical or electrical contact with each other. Similarly, the term "coupled" may be used in describing some embodiments to indicate that two or more components have direct physical or electrical contact. However, the terms "connected" or "coupled" may also refer to two or more components that do not have direct contact with each other but still cooperate or interact with each other, such as "optical coupling," "wireless connection," etc. The embodiments disclosed herein are not necessarily limited to the scope of this invention.

[0025] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0026] Example 1: To address the problems of existing technologies, this embodiment proposes a computing power assessment and task scheduling method. In one embodiment, such as... Figure 1 As shown, it includes: Step 101: Collect the status information of tasks and equipment in the unmanned command system, construct the status feature vector, and establish the total delay model and total energy consumption model.

[0027] In this embodiment, an unmanned command system is used as an example. In other embodiments, this embodiment can also be applied to fields including but not limited to big data processing systems, cloud computing systems, and artificial intelligence systems.

[0028] In unmanned command system scenarios, comprehensive status information of tasks and equipment is collected. This status information includes the task's computational requirements, data volume, and core parameters of the equipment such as bandwidth capacity, computing power, computational energy consumption constant, communication efficiency constant, and transmission power. Based on this multi-dimensional information, a status feature vector that can accurately reflect the current resource status and task requirements of the system is constructed.

[0029] Meanwhile, based on the actual scenarios of task execution and data transmission, a total latency model and a total energy consumption model are established respectively. The total latency model is used to integrate the computation latency and communication transmission latency in the computation process, while the total energy consumption model is used to statistically analyze the computation energy consumption in the task execution phase and the communication energy consumption in the data transmission process, providing a quantitative analysis basis for subsequent computing power evaluation and scheduling optimization.

[0030] Step 102: Perform quantitative analysis on the total delay data output by the total delay model and the total energy consumption data output by the total energy consumption model to obtain the total delay weight and the total energy consumption weight, respectively.

[0031] Specifically, the total delay data output by the total delay model and the total energy consumption data output by the total energy consumption model in step 101 are first subjected to outlier removal and standardization preprocessing to eliminate the impact of differences in the units of different indicators and ensure that the data are comparable.

[0032] Quantitative analysis is then performed. By calculating the information entropy of total delay and total energy consumption, the influence of the two indicators on scheduling decisions is determined based on the magnitude of the information entropy. The smaller the information entropy, the higher the importance of the indicator. Then, the total delay weight and total energy consumption weight are automatically assigned based on the calculation results of the information entropy, avoiding the subjectivity of manually setting weights and making the weight allocation more in line with the actual operating characteristics of the system.

[0033] Step 103: Train the policy model based on the state feature vector, the total delay weight, and the total energy consumption weight, and output the optimal task scheduling policy based on the trained policy model.

[0034] In this process, the state feature vector constructed in step 101, along with the total delay weight and total energy consumption weight obtained in step 102, are used as core inputs and imported into a preset policy model for targeted training.

[0035] During training, a reasonable reward function is constructed by combining the total latency weight and the total energy consumption weight. A reward and penalty mechanism guides the policy model's learning process, while simultaneously optimizing the objective function and value network of the policy model to continuously reduce prediction errors. Once the policy model training converges, it can fully integrate the real-time system status and the importance of various indicators, automatically outputting the optimal task scheduling strategy that adapts to the current task requirements and system load, achieving efficient task allocation among devices.

[0036] The following section will provide a detailed explanation of the computing power assessment and task scheduling methods.

[0037] In one embodiment, real-time status information is collected from task and device nodes in the unmanned command system, and status feature vectors of tasks and devices are constructed based on this information, providing an input basis for subsequent task scheduling optimization. This step involves the collection and modeling of multiple parameters, including task computational requirements, data volume, device bandwidth capacity, and computing power, to ensure that the system can dynamically and accurately reflect the current resource status and task requirements.

[0038] In one embodiment, a state feature vector is constructed based on the collected task and device state information. The status information includes the task's computational requirements and data volume, as well as the device's bandwidth capacity, computing power, computational energy consumption constant, communication efficiency constant, and transmission power.

[0039] in, For the task The computational requirements; For the task The amount of data; For the device's bandwidth capability; For computing power; To calculate the energy consumption constant; Let be the communication efficiency constant; Send power to the device, The number of terminal devices. This refers to the number of server devices.

[0040] In one embodiment, the unmanned command system includes 50 terminal devices (i.e., =50) and 4 server devices (i.e., =4). These state feature vectors can comprehensively describe the computing requirements, resource capabilities, and status of each task and device, thus providing sufficient information support for subsequent task scheduling.

[0041] In one embodiment, the total latency model includes computational latency and communication latency; ; ; ; in, For the total delay model, For the first The terminal device in the first The computational latency of each server For transmission to the first Communication latency of individual server devices; The total energy consumption model includes computational energy consumption and communication energy consumption; ; ; ; in, For the total energy consumption model, For the first The terminal device in the first The computing power consumption of a single server; For transmission to the first The communication energy consumption of each server; The effective switching capacitor coefficient of the terminal equipment; This refers to the transmission power of the terminal device.

[0042] The state feature vector is obtained through the above steps. After establishing the total energy consumption model and the total delay model, the total delay weight and total energy consumption weight in task scheduling are then calculated to provide appropriate reward function weights for subsequent task scheduling optimization of the strategy model. When calculating the total delay weight and total energy consumption weight, an algorithm that can automatically calculate weights based on the data distribution characteristics of each indicator can be selected to avoid the subjectivity of manually setting weights, thereby improving the scientific nature and accuracy of scheduling decisions.

[0043] Before calculating the total delay weight and total energy consumption weight, it is necessary to adjust the initial total delay. Total energy consumption Normalization is performed to ensure consistency of numerical units among different features; that is, outlier removal and standardization are performed on total delay and total energy consumption to obtain standardized data. ; ; Then, calculate the proportion of each standardized data point in the corresponding indicator. This is used to measure the contribution of each data point to the overall metric, and the formula is as follows: ; Information entropy is a measure of the uncertainty and information content of an indicator. The smaller the information entropy, the greater the impact of changes in the indicator on decision-making. Subsequently, the information entropy of total delay and total energy consumption is calculated based on the stated proportion. Information entropy The formula is as follows: ; ; Based on the information entropy calculation results, the weights of each indicator are calculated, and the total delay weight α and total energy consumption weight β are automatically assigned. The total delay weight is calculated based on the information entropy. Total energy consumption weight ; ; in, Indicates the first The first of the indicators The result after data standardization; For the first in the original data The first of the indicators One data point; and These represent the first and second parts of the original data. Minimum and maximum values ​​of each indicator For the first The data in the first The proportion of each indicator For the first Information entropy of each indicator For the first The weight of each indicator, Total number of indicators, total delay weight = Total energy consumption weight = .

[0044] After obtaining the total latency weight and total energy consumption weight, a policy optimization algorithm is used to optimize task scheduling. The policy optimization algorithm takes the state feature vector as input to the policy model and generates a task scheduling policy. By continuously optimizing the policy, the algorithm can automatically adjust the task scheduling strategy to minimize latency and energy consumption, achieving the optimal task allocation effect.

[0045] In one embodiment, such as Figure 2 As shown, training the policy model based on the state feature vector, the total delay weight, and the total energy consumption weight specifically includes: Step 1031: Input the state feature vector into the policy network and output the probability distribution of task allocation to each device.

[0046] The state feature vector constructed in step 101 is used as the input data of the policy network. The state feature vector contains multi-dimensional core information about tasks and devices, which can comprehensively reflect the current resource supply capacity and task execution requirements of the system.

[0047] The policy network performs in-depth analysis of the input state feature vector through internal weight calculation and feature mapping, ultimately outputting a probability distribution of task allocation to various devices. This probability distribution intuitively presents the suitability of each device for undertaking the current task, providing a quantitative reference for subsequent task allocation decisions and ensuring the scientific and rational nature of task allocation actions.

[0048] In one embodiment, the probability distribution of task assignment satisfies: ; ; in, Indicates the action of task assignment; This represents the state feature vector at the current moment; Tasks are assigned to devices The probability of; This represents the total number of optional devices. Indicates the first The original score value corresponding to each device; and The first The device and the first The rating value corresponding to each device; Network weights; For bias; For summation index variables.

[0049] Step 1032: Construct an instant reward function based on the total delay weight, the total energy consumption weight, and the delay and energy consumption changes before and after the task execution action.

[0050] Based on the total delay weight and total energy consumption weight obtained in step 102, and combined with the actual data changes before and after the execution of different actions, namely the delay difference between the action switching and the delay before the switching (delay change) and the energy consumption difference between the action switching and the energy consumption before the switching (energy consumption change), an instant reward function is constructed.

[0051] The instant reward function assigns different decision priorities to total latency and total energy consumption by assigning weights to total latency and total energy consumption respectively. This allows for precise quantification of the merits of each task allocation action. When an action reduces latency and energy consumption, the function outputs a positive reward; conversely, it outputs a negative penalty, providing clear optimization guidance for the training of the policy model.

[0052] In one embodiment, to ensure that the policy optimization algorithm can optimize the task scheduling policy, an instant reward function needs to be designed. This is used to evaluate the merits of each task allocation scheme. The immediate reward function comprehensively considers the latency and energy consumption changes after task scheduling to encourage the policy optimization algorithm to allocate tasks. The task allocation scheme generated by the policy network provides optimization feedback. In one embodiment, the immediate reward function... satisfy: ; ; ; in, To perform the action The delay; To perform the action The delay; To perform the action Energy consumption; To perform the action Energy consumption; The total delay weight, Weighted by total energy consumption; The change in energy consumption before and after the action is performed; This refers to the change in delay before and after the action is performed.

[0053] Step 1033: With the goal of maximizing cumulative reward, optimize the policy network through the clip mechanism and advantage function of the policy model.

[0054] The optimization objective of the policy network is clearly defined as maximizing cumulative reward. A clip mechanism and a dominance function are introduced into the policy model. The dominance function measures the relative advantage of the current task assignment action compared to other possible actions, accurately capturing the actual value of the action. The clip mechanism, by setting a reasonable parameter range, limits the magnitude of policy updates, preventing model training instability due to sudden policy changes. Through the synergistic effect of these two mechanisms, the parameters of the policy network are iteratively optimized, gradually improving the network's ability to output the optimal task assignment probability distribution.

[0055] In one embodiment, the goal of the policy network is to maximize cumulative reward to improve task scheduling efficiency. The following objective function optimizes the policy network: In one embodiment, the policy network satisfy: ; ; ; ; in, This represents the probability ratio between the new strategy and the old strategy. Let be the policy function to be optimized at the current time step, when the model parameters are . At that time, based on the input status Output the probability distribution of the execution probability of each action; For the old policy function before the update, when the model parameters are At that time, based on the input status Output the probability distribution of the execution probability of each action; The advantage function is used to measure the action. The advantages and disadvantages; Used to limit the update range and prevent excessive policy changes; For editing parameters, As a discount factor, For value networks, As a discount reward; Indicates the first The instantaneous reward value can be referenced. .

[0056] Step 1034: Optimize the value network based on mean squared error and discounted reward, and iteratively update the policy network and value network until convergence to complete the training of the policy model.

[0057] The optimization objective of the value network is constructed using mean squared error as the loss function and combined with discounted rewards. The value network is responsible for estimating the long-term reward of the task scheduling state. It calculates the error by comparing the estimated value of the current state with the discounted reward, and then backpropagates the error, continuously adjusting the network parameters to reduce the estimation error. Subsequently, the process of optimizing the policy network and updating the value network is iteratively executed, continuously optimizing the performance of both networks until convergence. This ensures that the trained policy model can stably output the optimal scheduling policy that adapts to the system state and task requirements, completing the entire policy model training process.

[0058] In one embodiment, the policy optimization algorithm also applies to the value network. Optimization is performed to improve the value estimation of task scheduling states. The value network measures the difference between the current state value estimate and the target value using mean squared error, and rewards are given based on a discount. Update.

[0059] Each time an update is performed, the discount reward is calculated using the parameters of the current policy network. This optimizes the policy and value networks, thereby reducing estimation errors. The policy network is updated cyclically, outputting the reward, total latency, and total energy consumption each time. Through cyclic updates, the policy and value networks gradually converge to the optimal state, thus completing the training of the policy model.

[0060] In one embodiment, the above steps are implemented to construct a communication system comprising 50 terminal devices and 4 server devices. After initializing the state information, state information collection and feature construction, calculation of reward function weights, and task scheduling optimization based on a policy model are performed sequentially. Figure 3 It can be seen that the rewards gradually increase in the first 40 rounds of training and then maintain a stable trend in the subsequent 60 rounds; from Figure 4As can be seen, both total latency and total energy consumption are effectively reduced, decreasing rapidly in the first 40 rounds and maintaining a stable trend in the subsequent 60 rounds of training, indicating that the optimization method proposed in this embodiment can effectively optimize the overall cost of the system.

[0061] In summary, this embodiment first collects task and device status information, constructs a model including total latency and total energy consumption, then quantifies and calculates the total latency weight and total energy consumption weight, and finally trains a strategy model based on the state feature vector and weights. The optimal task scheduling strategy is then output through the trained strategy model. This invention, by quantitatively evaluating various indicators of computing resources and combining them with the strategy model to optimize task scheduling, makes resource scheduling more accurate and efficient. It effectively solves the problems of low resource utilization, unintelligent scheduling, and poor dynamic adaptability of traditional methods. It can adapt to changes in system load and task requirements, significantly improves the efficiency of computing resource utilization, reduces latency and energy consumption, and thus optimizes the overall performance of the system.

[0062] Example 2: In Embodiment 1, a computing power assessment and task scheduling method is provided. In this embodiment, a computing power assessment and task scheduling device is proposed. The computing power assessment and task scheduling device includes: a processor and a memory for storing processor-executable instructions; wherein, the processor is configured to execute the computing power assessment and task scheduling method described in Embodiment 1.

[0063] like Figure 5 As shown, the computing power assessment and task scheduling device includes a processor 21 and a memory 22, wherein the processor 21 and the memory 22 can be connected by a bus or other means.

[0064] Processor 21 can be a Central Processing Unit (CPU). Processor 21 can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0065] The memory 22, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the computing power assessment and task scheduling method in Embodiment 1 of this invention. The processor executes various functional applications and training processes by running the non-transitory software programs, instructions, and modules stored in the memory.

[0066] The memory 22 may include a program storage area and a training storage area. The program storage area may store the operating system and applications required for at least one function; the training storage area may store training data created by the processor. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory, or other non-transitory solid-state storage device. In some embodiments, the memory 22 may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof. The one or more modules stored in the memory 22, when executed by the processor 21, perform functions such as... Figure 1 The computing power assessment and task scheduling method in Example 1 is shown. For specific details of the above computing power assessment and task scheduling method, please refer to the relevant documentation. Figure 1 , Figure 2 and Figure 3 The relevant descriptions and effects in the embodiments shown are for reference only and will not be repeated here.

[0067] This embodiment also provides a computer storage medium storing a computer program that can be executed by a processor to complete the computing power assessment and task scheduling method described in Embodiment 1.

[0068] The computer storage medium stores computer-executable instructions, which can execute the computing power assessment and task scheduling methods in any of the above method embodiments. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium may also include combinations of the above types of memory.

[0069] The specific steps of the computing power assessment and task scheduling method are described in Example 1, and will not be repeated in this example.

[0070] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for computing power assessment and task scheduling, characterized in that, include: Collect status information of tasks and equipment in the unmanned command system, construct status feature vectors, and establish total delay and total energy consumption models; The total delay data output by the total delay model and the total energy consumption data output by the total energy consumption model are quantitatively analyzed to obtain the total delay weight and the total energy consumption weight, respectively. The strategy model is trained based on the state feature vector, the total delay weight, and the total energy consumption weight, and the optimal task scheduling strategy is output based on the trained strategy model.

2. The computing power assessment and task scheduling method according to claim 1, characterized in that, The process of collecting status information of tasks and equipment in the unmanned command system and constructing a status feature vector specifically includes: Based on the collected task and device status information, a status feature vector is constructed. The status information includes the task's computational requirements and data volume, as well as the device's bandwidth capacity, computing power, computational energy consumption constant, communication efficiency constant, and transmission power. ； in, For the task The computational requirements; For the task The amount of data; For the device's bandwidth capability; For computing power; To calculate the energy consumption constant; Let be the communication efficiency constant; Send power to the device, The number of terminal devices. This refers to the number of server devices.

3. The computing power assessment and task scheduling method according to claim 2, characterized in that, The total delay model includes computational delay and communication delay; ； ； ； in, For the total delay model, For the first The terminal device in the first The computational latency of each server For transmission to the first Communication latency of individual server devices; The total energy consumption model includes computational energy consumption and communication energy consumption; ； ； ； in, For the total energy consumption model, For the first The terminal device in the first The computing power consumption of a single server; For transmission to the first The communication energy consumption of each server; The effective switching capacitor coefficient of the terminal equipment; This refers to the transmission power of the terminal device.

4. The computing power assessment and task scheduling method according to claim 1, characterized in that, The quantitative analysis of the total delay data output by the total delay model and the total energy consumption data output by the total energy consumption model, to obtain the total delay weight and the total energy consumption weight respectively, specifically includes: Outlier removal and standardization were performed on total delay and total energy consumption to obtain standardized data. ; ； Calculate the proportion of each standardized data point in the corresponding indicator. ; ； The information entropy of total delay and total energy consumption is calculated based on the aforementioned proportion. ; ； ； Calculate the total delay weight based on information entropy. Total energy consumption weight ; ； in, Indicates the first The first of the indicators The result after data standardization; For the first in the original data The first of the indicators One data point; and These represent the first and second parts of the original data. Minimum and maximum values ​​of each indicator For the first The data in the first The proportion of each indicator For the first Information entropy of each indicator For the first The weight of each indicator, Total number of indicators, total delay weight = Total energy consumption weight = .

5. The computing power assessment and task scheduling method according to claim 1, characterized in that, The step of training the policy model based on the state feature vector, the total delay weight, and the total energy consumption weight specifically includes: The state feature vector is input into the policy network, and the probability distribution of task allocation to each device is output. Based on the total latency weight, the total energy consumption weight, and the latency and energy consumption changes before and after the task execution action, an instant reward function is constructed. With the goal of maximizing cumulative rewards, the policy network is optimized through the clip mechanism and advantage function of the policy model; The value network is optimized based on mean squared error and discounted reward. The policy network and value network are iteratively updated until convergence is achieved to complete the training of the policy model.

6. The computing power assessment and task scheduling method according to claim 5, characterized in that, The probability distribution of task assignment satisfies: ； ； in, Indicates the action of task assignment; This represents the state feature vector at the current moment; Tasks are assigned to devices The probability of; This represents the total number of optional devices. Indicates the first The original score value corresponding to each device; and The first The device and the first The rating value corresponding to each device; Network weights; For bias; For summation index variables.

7. The computing power assessment and task scheduling method according to claim 5, characterized in that, The instant reward function satisfy: ； ； ； in, To perform the action The delay; To perform the action The delay; To perform the action Energy consumption; To perform the action Energy consumption; The total delay weight, Weighted by total energy consumption; The change in energy consumption before and after the action is performed; This refers to the change in delay before and after the action is performed.

8. The computing power assessment and task scheduling method according to claim 5, characterized in that, The policy network satisfy: ； ； ； ； in, This represents the probability ratio between the new strategy and the old strategy. Let be the policy function to be optimized at the current time step, when the model parameters are . At that time, based on the input status Output the probability distribution of the execution probability of each action; For the old policy function before the update, when the model parameters are At that time, based on the input status Output the probability distribution of the execution probability of each action; The advantage function is used to measure the action. The advantages and disadvantages; Used to limit the update range and prevent excessive policy changes; For editing parameters, As a discount factor, For value networks, As a discount reward; Indicates the first The reward value at any given moment.

9. A computing power assessment and task scheduling device, characterized in that, The computing power assessment and task scheduling device includes: a processor and a memory for storing processor-executable instructions; The processor is configured to execute the computing power assessment and task scheduling method according to any one of claims 1-8.

10. A non-volatile computer storage medium, characterized in that, The computer storage medium stores computer-executable instructions, which are executed by one or more processors to perform the computing power assessment and task scheduling method according to any one of claims 1-8.

Images