Information computing load scheduling method and device in internet of things mode, equipment and medium

Abstract

The application relates to the technical field of artificial intelligence, and discloses an information computing load scheduling method, device and equipment under an Internet of Things mode and a medium, which comprises the following steps: receiving an information computing task, distributing the information computing task to a corresponding target execution container; generating a priority queue of the information computing task in each target execution container; controlling the computing resource allocation proportion of each target execution container through a pre-trained resource allocator; and adjusting the execution order of an execution task according to the state of the execution task in the priority queue through a pre-trained execution scheduler. The application realizes full utilization of platform computing resources, reduces the weighted average execution time of the task priority, and improves user satisfaction.

Description

Technical Field

[0001] This invention relates to the field of computer processing technology, and in particular to a method, apparatus, device and medium for scheduling information computing load under the Internet of Things (IoT) model. Background Technology

[0002] With the continuous development of IoT technology and the expansion of application scenarios, the collection of physiological information has moved from hospital laboratories to households. Thanks to the increasing miniaturization and intelligence of physiological information collection devices, the compact, portable, intelligent, and long-term wearable characteristics of various collection devices have led to numerous applications in collecting human health data such as heart rate, blood oxygen, body posture, body temperature, respiration, and blood pressure. Some physiological information can even be processed using intelligent devices alone. The widespread collection of physiological information strongly supports the development of telemedicine and home monitoring.

[0003] With the increasing ease of collecting physiological information, the scale of data to be processed has grown exponentially. Furthermore, as demands evolve, traditional offline data aggregation and simple real-time computation are no longer sufficient to meet the growing complexity of computational needs and the demand for more real-time monitoring. Therefore, it is necessary to improve the service architecture of cloud data computing centers to meet the complex computational processes, large volumes of parallel data, and strong real-time requirements. The significant increase in the scale and complexity of the data to be processed presents a considerable challenge to cloud data centers. Unlike user data in traditional internet models, physiological information has a long time span, diverse data types, and a large volume. Within the same time period, cloud data centers may need to process both real-time monitoring tasks and data collected and uploaded by various devices. The long time span, relatively complex computational scenarios, strong sequential dependencies in processing, and the general reliance on past data in real-time tasks make it difficult for traditional resource allocation algorithms designed for independent computational tasks to effectively coordinate the scheduling and efficient allocation of resources for physiological information computation. Therefore, it is also necessary to improve resource allocation and task scheduling algorithms to maximize the utilization of computing power and achieve higher efficiency in physiological information computation scenarios. Summary of the Invention

[0004] This invention provides a method, apparatus, computer equipment, and medium for scheduling information computing load under the Internet of Things (IoT) model, which can maximize the execution efficiency of computing tasks and reduce task waiting time while meeting priority requirements.

[0005] Firstly, a method for scheduling information computing load under the Internet of Things (IoT) model is provided, including:

[0006] Receive information computing tasks and assign the information computing tasks to the corresponding target execution containers;

[0007] Generate priority queues for the information computation tasks within each target execution container;

[0008] The computational resource allocation ratio of each target execution container is adjusted by a pre-trained resource allocator.

[0009] The execution order of tasks is adjusted based on the status of the currently executing tasks in the priority queue using a pre-trained execution scheduler.

[0010] In some embodiments, assigning the information computation task to the corresponding target execution container includes:

[0011] Based on the type of the information computing task and the load factor of each container in the resource allocator, the information computing task is assigned to the corresponding target execution container.

[0012] In some embodiments, generating priority queues for the information computation tasks within each target execution container includes:

[0013] The priority queue is generated based on the current task's waiting time, the current task's expected execution time, and its priority level.

[0014] In some embodiments, adjusting the computational resource allocation ratio of each target execution container through a pre-trained resource allocator includes:

[0015] The computing resource allocation ratio of each target execution container is adjusted according to the load index of the priority queue within each target execution container.

[0016] In some embodiments, the method further includes training the resource allocator, wherein training the resource allocator includes:

[0017] The current environmental state is determined through the actor network of the resource allocator;

[0018] The computing resources of the target container are allocated based on the decision results of the actor section;

[0019] Run for a preset time period and accumulate an experience pool based on environmental rewards until the size of the experience pool reaches a preset threshold.

[0020] The critic network is updated using the experience pool, and the actor network is updated based on the gradient obtained from the critic network.

[0021] The number of iterations for training the resource allocator is determined based on the convergence of the resource allocator.

[0022] In some embodiments, the execution scheduler to be trained includes multiple layers of topology, each layer of topology including two DQN structures, one of which is the decision model of this layer and the other is the time prediction model of the decision of the upper layer.

[0023] The method further includes training the execution scheduler, wherein training the resource allocator includes:

[0024] Determine whether the current training hierarchy in the execution scheduler to be trained is the top-level structure. If so, set the DQN structure parameters according to the execution scheduler to be trained.

[0025] Train the DQN structure using the training set;

[0026] Accumulate experience pools to train hierarchical DQN structures;

[0027] The number of iterations for training the DQN structure is determined based on whether the hierarchical DQN structure converges.

[0028] If not, then set the computing resource status and token quantity limit according to the model.

[0029] Randomly generate the number of initial environmental states within the constraints;

[0030] Select the initialization level DQN structure environment;

[0031] Accumulate an experience pool to train the DQN structure;

[0032] The number of iterations for training the DQN structure is determined based on whether the hierarchical DQN structure converges.

[0033] In some embodiments, the execution scheduler to be trained includes multiple levels of sub-schedulers, each level of sub-scheduler including a DQN structure, wherein the DQN structure is the decision model of that level.

[0034] The process of adjusting the execution order of tasks based on the current state of tasks in the priority queue using a pre-trained execution scheduler includes:

[0035] The computational task status of the currently executing task is input into a multi-layer DQN structure. For each layer of DQN, the task execution target of this layer is calculated. According to the task type, the task execution target of this layer is input into the corresponding lower layer DQN structure to obtain the lower layer task execution target. This process is iterated until the target is input into the bottom layer DQN structure to generate the actual task execution step sequence, thus obtaining the adjusted execution sequence of the executing tasks.

[0036] Secondly, an information computing load scheduling device under the Internet of Things (IoT) model is provided, comprising:

[0037] A receiving unit is used to receive information computing tasks and allocate the information computing tasks to the corresponding target execution containers.

[0038] A priority queue unit is used to generate priority queues for the information computation tasks within each target execution container;

[0039] The resource allocation unit is used to regulate the computational resource allocation ratio of each of the target execution containers through a pre-trained resource allocator.

[0040] The execution scheduling unit is used to adjust the execution order of tasks based on the current status of tasks in the priority queue using a pre-trained execution scheduler.

[0041] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the information computing load scheduling method under the above-described Internet of Things mode.

[0042] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the information computing load scheduling method described above under the Internet of Things (IoT) model.

[0043] In the aforementioned scheme implemented by the information computing load scheduling method, device, computer equipment, and storage medium under the Internet of Things (IoT) model, information computing tasks are received and allocated to corresponding target execution containers; priority queues of information computing tasks are generated within each target execution container; the allocation ratio of computing resources to each target execution container is adjusted through a pre-trained resource allocator; and the execution order of execution tasks is adjusted according to the current status of the tasks in the priority queues through a pre-trained execution scheduler. Therefore, task execution scheduling and resource allocation scheduling are optimized. The priority queue is used for coarse task scheduling, task execution scheduling performs fine-grained task scheduling based on the simplification of the optimization problem by optimizing the priority queue, and resource allocation scheduling is used to compensate for the inability of intra-container task scheduling to handle load differences between containers. The fine-grained task scheduling of the task execution scheduler is based on the coarse task scheduling of the priority queue, and the work of resource allocation scheduling is used to balance the load differences between node containers. Attached Figure Description

[0044] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention 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.

[0045] Figure 1 This is a flowchart illustrating an information computing load scheduling method under an Internet of Things (IoT) mode according to an embodiment of the present invention.

[0046] Figure 2 This is a scheduling strategy structure diagram of an information computing load scheduling method under the Internet of Things mode in one embodiment of the present invention;

[0047] Figure 3 This is a node computing resource allocation architecture for an information computing load scheduling method under the Internet of Things mode in one embodiment of the present invention;

[0048] Figure 4 This is a flowchart of the node computing resource model training process for an information computing load scheduling method under the Internet of Things mode in one embodiment of the present invention.

[0049] Figure 5 This is a hierarchical DQN structure diagram of the task execution scheduler in an information computing load scheduling method under the Internet of Things mode according to an embodiment of the present invention;

[0050] Figure 6 This is a flowchart of the hierarchical DQN training process of the information computing load scheduling method under the Internet of Things mode in one embodiment of the present invention;

[0051] Figure 7 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Please see Figure 1 As shown, Figure 1 A flowchart illustrating an information computing load scheduling method under the Internet of Things (IoT) model provided in this embodiment of the invention includes the following steps:

[0054] Step 10: Receive information computing tasks and assign them to the corresponding target execution containers;

[0055] Upon receiving a new physiological information calculation task, the system selects a target execution container based on the task type and the load coefficient of each container within the node resource allocator, and then sends the new task to the task queue of the target execution container to await processing.

[0056] In some embodiments, assigning information computation tasks to corresponding target execution containers includes:

[0057] Based on the type of information computing task and the load factor of each container within the resource allocator, the information computing task is assigned to the corresponding target execution container.

[0058] Step 20: Generate a priority queue for information computation tasks within each target execution container;

[0059] The task queue inside the container is a priority queue. Priority coefficients are calculated based on the expected execution time and priority of each task, and the tasks with the highest priority coefficients are selected for further processing. The task types and execution parameters of the highest priority tasks are then sent to the task scheduler, which determines the operations that the system needs to perform.

[0060] Figure 3 This is a priority queue structure diagram inside the container. Tasks are pre-ordered through the priority queue on the left. A fixed number of tasks that are most worth executing are selected from the queue and sent to the task scheduler for execution. The task scheduler's function is to execute tasks.

[0061] A single task in t i The penalty value at time step is calculated as follows:

[0062]

[0063] Where p i Indicates task priority, e i σ represents the expected runtime of the task. i The standard deviation coefficient of the task is used to control the reward decline curve, t i This indicates the task's execution time. The priority index is calculated using the following formula:

[0064] p(task) = [Penalty(t)] task +t c )-Penalty(t task )]

[0065] In the above formula, t i For the current time, t c =γ*e task e task γ is the expected execution time of the current task, and γ is a hyperparameter, typically set to γ∈[1,3) (expected execution time). Let the upper limit of the scheduling algorithm be N, sort the tasks according to their priority indices, and select the N tasks with the highest priority indices from the queue for processing.

[0066] Step 30: Adjust the computational resource allocation ratio of each target execution container using a pre-trained resource allocator;

[0067] Resource allocation scheduling is used to compensate for the inability of task scheduling within containers to handle load differences between containers.

[0068] Figure 3 As shown, the node-level resource allocator is a strategy algorithm based on an actor-critic architecture. It regulates the allocation of computing resources among PODs by collecting the load index of the priority queue within each POD container. The task execution scheduler is a task execution strategy algorithm based on hierarchical DQN. It determines the execution order of tasks based on the current status of the tasks, maximizing the execution efficiency of computational tasks while considering priority and reducing task waiting time. The priority queue is used to prioritize accumulated tasks to identify the few tasks that most need to be executed to reduce the overall task waiting time, and then hand them over to the task scheduler for processing.

[0069] The decision algorithm is based on the AC framework DDPG algorithm, and the decision function of the Actot part is a. t =π(s) t |θ μ ), where s t This represents the current state of the task queues in each container, a t This represents the model decision at the current moment, corresponding to the change in the resource allocation ratio of each container. The parameters of the Critic part are the value function Q(s). t a t |θ Q ) represents the current state action pair (s) t a t The expected penalty value is calculated as follows. The objective function is the total penalty value without considering the depreciation factor:

[0070]

[0071] The optimization objective is to minimize the objective function, from which we can obtain the Actor's partial parameters θ. μ The gradient formula is:

[0072]

[0073] Critic part parameter θ Q The loss function is:

[0074]

[0075] Similar to the DDPG algorithm, θ in the above formula μ- With θ Q- The parameter θ represents the policy network parameter in the target network part of DQN. μWith value network parameter 0 Q Let y = p i +Q(s i+1 ,π(s i+1 θ μ- )|θ Q- ) can be obtained as 0 Q The gradient formula is:

[0076]

[0077] Environmental state s t The calculation method involves calculating the expected increase of the current penalty value for each container under the node, which is then used as a vector to form the load index. The positions of the containers within the vector remain fixed, and the current global clock is T. The formula for calculating the load index is as follows:

[0078] load(pod) = ∑ t∈Task(pod) f(T-start(t); e(t), σ(t))

[0079] In the above formula, pod is the container whose load index needs to be calculated, Task(pod) is the set of tasks inside pod, start(t) is the creation timestamp of task t, e(t) is the expected running time of task t, and σ(t) is the standard deviation coefficient of task t. The function f(t; e, σ) is defined as a normal distribution, as shown in the following formula:

[0080]

[0081] Where t is the task execution time, e is the average expected execution time of the task, and σ is the standard deviation coefficient of the distribution.

[0082] Environmental state s t This can be represented as [load(pod0), load(pod1), ..., load(pod0)] n The newly created task is placed into a container queue based on the accumulated container load index from the upper levels.

[0083] In some embodiments, such as Figure 5 As shown, it also includes a training resource allocator, wherein the training resource allocator includes:

[0084] Decisions about the current environmental state are made through the actor network of the resource allocator;

[0085] The computing resources of the target container are allocated based on the decision results of the actor section;

[0086] Run for a preset time period and accumulate experience in the experience pool based on environmental rewards until the experience pool size reaches a preset threshold.

[0087] Update the critic network using the experience pool, and update the actor network based on the gradient obtained from the critic network;

[0088] The number of iterations for training the resource allocator is determined based on its convergence.

[0089] Step 40: Adjust the execution order of tasks according to the status of the current tasks in the priority queue using a pre-trained execution scheduler.

[0090] The task execution scheduler is a task execution strategy algorithm based on hierarchical DQN. It determines the execution order of tasks based on the current state of the tasks being executed, and maximizes the execution efficiency of the tasks while taking priority into account, thereby reducing the waiting time of the tasks.

[0091] Figure 6 To illustrate the internal structure and training method of a task execution scheduler based on reinforcement learning, the benefit of various task execution orders is determined by using reinforcement learning. The objective with the highest reinforcement learning benefit is changed to the one with the lowest penalty value, and the execution order with the lowest total penalty value is selected.

[0092] The computation process for physiological information is highly hierarchical. For example, the processing of a physiological signal includes two steps: preprocessing and computation. The preprocessing step can be further subdivided into data format preprocessing, data filling, data filtering, etc. Therefore, the computation process can be modeled as a multi-level topological structure, with each level's local topological structure corresponding to a type of computational operation at the level above.

[0093] For a local topological structure, the state transition sequence sample is (t i g i p i a i g i+1 ), where g i For decision time i, g represents the state of all unfinished tasks in the current hierarchical topology model. i+1 Let Q represent the environment state at the next decision step (i+1). For each level of topological logic, train two DQN structures: a policy model Q for the decision-making at that level and a policy model Q for the current level. c and the time estimation model Q used for upper-level decision-making t .

[0094] Decision Strategy Q c The training objective is to minimize the expected total penalty value of the local topology network, as shown in the following formula:

[0095]

[0096] Where π ag Given the execution strategy for the current local topology, for the i-th decision, gi For the target state of the local topology, a i The action decision is for the task to be executed in the current environment. Since the execution paths of all tasks in the system are fixed and there are no task loops, the overall penalty value is not affected by the task execution steps; therefore, no discount rate parameter is set in the above formula. i The method for calculating the local penalty value for an action is as follows:

[0097]

[0098] Similar to the training mode of the DQN model, the overall loss function of the local topology is the mean squared difference between the Q-value output of the current layer model and the Q-value output of the target value network. c The error function is as follows:

[0099]

[0100] For the parameter θ in the above equation c Taking the partial derivatives, we obtain the gradient formula as follows:

[0101]

[0102] Time evaluation model Q t Its function is to evaluate the local topology network in state g. i The total expected time consumption; the time evaluation model does not participate in the decision-making of this layer's model, but only evaluates the total time required for all tasks to complete under the current environmental conditions, Q. t The decision model is trained using a strategy, with the objective formula being:

[0103]

[0104] make The error function can be obtained as follows:

[0105]

[0106] Taking the partial derivative of the above equation with respect to the parameter θ, we obtain the gradient formula as follows:

[0107]

[0108] For the upper-level local topology, the lower-level local topology is treated as a type of operation, and the set of tasks participating in the operation is taken as the lower-level input target state g. The time evaluation model Q of the lower-level topology is then used. t Regarding the time consumption of this type of change:

[0109]

[0110] In some embodiments, such as Figure 7 As shown, the execution scheduler to be trained includes multiple layers of topology. Each layer of topology includes two DQN structures, one of which is the decision model of this layer and the other is the time prediction model of the decision of the upper layer.

[0111] The method also includes training an execution scheduler, wherein training a resource allocator includes:

[0112] Determine whether the current training hierarchy in the execution scheduler to be trained is the top-level structure. If so, set the DQN structure parameters according to the execution scheduler to be trained.

[0113] Train the DQN structure using the training set;

[0114] Accumulate experience pools to train hierarchical DQN structures;

[0115] The number of iterations for training the DQN structure is determined based on whether the hierarchical DQN structure converges.

[0116] If not, then set the computing resource status and token quantity limit in the root model.

[0117] Randomly generate the number of initial environmental states within the constraints;

[0118] Select the initialization level DQN structure environment;

[0119] Accumulate an experience pool to train the DQN structure;

[0120] The number of iterations for training the DQN structure is determined based on whether the hierarchical DQN structure converges.

[0121] In some embodiments, the execution scheduler to be trained includes multiple layers of topology, each layer of topology including one DQN structure, the DQN structure being the decision model for that layer;

[0122] The execution scheduler, through a pre-trained process, adjusts the execution order of tasks based on the current state of tasks in the priority queue, including:

[0123] The computational task status of the currently executing task is input into a multi-layer DQN structure. For each layer of DQN, the task execution target of this layer is calculated. According to the task type, the task execution target of this layer is input into the corresponding lower layer DQN structure to obtain the lower layer task execution target. This process is iterated until the target is input into the bottom layer DQN structure to generate the actual task execution step sequence, thus obtaining the adjusted execution sequence of the executing tasks.

[0124] In some embodiments, the information computing load scheduling device in the Internet of Things (IoT) mode provided by the present invention includes:

[0125] The receiving unit is used to receive information computing tasks and allocate the information computing tasks to the corresponding target execution containers.

[0126] Priority queue unit, used to generate priority queues for information computing tasks within each target execution container;

[0127] The resource allocation unit is used to regulate the computational resource allocation ratio of each target execution container through a pre-trained resource allocator.

[0128] The execution scheduling unit is used to adjust the execution order of tasks based on the current status of tasks in the priority queue using a pre-trained execution scheduler.

[0129] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 7 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the client-side functions or steps of an information computing load scheduling method under an Internet of Things (IoT) model.

[0130] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps:

[0131] Receive information computing tasks and assign them to the corresponding target execution containers;

[0132] Generate priority queues for information computation tasks within each target execution container;

[0133] The computational resource allocation ratio of each target execution container is adjusted through a pre-trained resource allocator.

[0134] The execution order of tasks is adjusted based on the status of the currently executing tasks in the priority queue using a pre-trained execution scheduler.

[0135] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0136] Receive information computing tasks and assign them to the corresponding target execution containers;

[0137] Generate priority queues for information computation tasks within each target execution container;

[0138] The computational resource allocation ratio of each target execution container is adjusted through a pre-trained resource allocator.

[0139] The execution order of tasks is adjusted based on the status of the currently executing tasks in the priority queue using a pre-trained execution scheduler.

[0140] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0141] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0142] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0143] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for scheduling information computing load under the Internet of Things (IoT) model, characterized in that, include: Receive information computing tasks and assign the information computing tasks to the corresponding target execution containers; Generate priority queues for the information computation tasks within each target execution container; The computational resource allocation ratio of each target execution container is adjusted by a pre-trained resource allocator. The execution order of tasks is adjusted based on the current status of tasks in the priority queue using a pre-trained execution scheduler. The pre-trained execution scheduler comprises multiple layers of topology, each layer containing two DQN structures: one DQN structure representing the decision model for that layer, and the other representing the time prediction model for the decision of the next higher layer. The time prediction model is used to evaluate the local topology network in state g. i The total expected time consumption; for the upper-level local topology, the lower-level local topology is regarded as a type of operation, and the set of tasks participating in the operation is taken as the lower-level input target state; The method further includes training the execution scheduler, wherein training the resource allocator includes: determining whether the currently trained hierarchical structure in the pre-trained execution scheduler is a top-level structure; if so, setting DQN structure parameters according to the pre-trained execution scheduler; training the DQN structure using a training set; accumulating an experience pool to train the hierarchical DQN structure; determining the number of iterations for training the DQN structure based on whether the hierarchical DQN structure has converged; if not, randomly generating the number of initial environment states within the limits set by the model for the computing resource state and token quantity; selecting an initialization environment for the hierarchical DQN structure; accumulating an experience pool to train the DQN structure; and determining the number of iterations for training the DQN structure based on whether the hierarchical DQN structure has converged.

2. The information computing load scheduling method under the Internet of Things (IoT) model as described in claim 1, characterized in that, The step of assigning the information computation task to the corresponding target execution container includes: Based on the type of the information computing task and the load factor of each container in the resource allocator, the information computing task is assigned to the corresponding target execution container.

3. The information computing load scheduling method under the Internet of Things (IoT) model as described in claim 1, characterized in that, The generation of priority queues for the information computation tasks within each target execution container includes: The priority queue is generated based on the current task's waiting time, the current task's expected execution time, and its priority level.

4. The information computing load scheduling method under the Internet of Things (IoT) model as described in claim 1, characterized in that, The step of adjusting the computational resource allocation ratio of each target execution container through a pre-trained resource allocator includes: The computing resource allocation ratio of each target execution container is adjusted according to the load index of the priority queue within each target execution container.

5. The information computing load scheduling method under the Internet of Things (IoT) model as described in claim 1, characterized in that, It also includes training the resource allocator, wherein training the resource allocator includes: The current environmental state is determined through the actor network of the resource allocator; The computing resources of the target container are allocated based on the decision results of the actor section; Run for a preset time period and accumulate an experience pool based on environmental rewards until the size of the experience pool reaches a preset threshold. The critic network is updated using the experience pool, and the actor network is updated based on the gradient obtained from the critic network. The number of iterations for training the resource allocator is determined based on the convergence of the resource allocator.

6. An information computing load scheduling device under the Internet of Things (IoT) model, characterized in that, include: A receiving unit is used to receive information computing tasks and allocate the information computing tasks to the corresponding target execution containers. A priority queue unit is used to generate priority queues for the information computation tasks within each target execution container; A resource allocation unit is used to regulate the computational resource allocation ratio of each target execution container through a pre-trained resource allocator. The pre-trained execution scheduler includes multiple layers of topology, each layer comprising two DQN structures: one DQN structure representing the decision model for that layer, and the other representing the time estimation model for the decision of the next higher layer. The time estimation model is used to evaluate the local topology network in state g. i The total expected time consumption; for the upper-level local topology, the lower-level local topology is regarded as a type of operation, and the set of tasks participating in the operation is taken as the lower-level input target state; it is also used to train the execution scheduler; the training of the resource allocator includes: determining whether the currently trained hierarchical structure in the pre-trained execution scheduler is the top-level structure, if so, setting the DQN structure parameters according to the pre-trained execution scheduler; training the DQN structure using the training set; accumulating experience pool to train the hierarchical DQN structure; determining the number of iterations for training the DQN structure according to whether the hierarchical DQN structure has converged; if not, randomly generating the number of initial environmental states within the limit of the set computing resource state and token quantity according to the model; selecting the initialization hierarchical DQN structure environment; accumulating experience pool to train the DQN structure; determining the number of iterations for training the DQN structure according to whether the hierarchical DQN structure has converged. The execution scheduling unit is used to adjust the execution order of tasks based on the current status of tasks in the priority queue using a pre-trained execution scheduler.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the information computing load scheduling method under the Internet of Things mode as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the information computing load scheduling method under the Internet of Things mode as described in any one of claims 1 to 5.

Images