Micro-motion exploration calculation-intensive dynamic allocation method and imaging system
By decomposing the micro-motion exploration task into sub-tasks and dynamically allocating them to sensor nodes for processing, the problems of high transmission delay and high resource cost in traditional micro-motion exploration methods are solved, achieving real-time imaging with low latency and low power consumption, and improving the robustness and resource utilization of the system.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING UNIV OF INFORMATION SCI & TECH
- Filing Date
- 2025-12-09
- Publication Date
- 2026-05-05
AI Technical Summary
Traditional micro-motion exploration methods suffer from high transmission delays, high resource costs, and unstable system performance in imaging tasks. Existing task allocation methods are difficult to adapt to dynamic environments and the energy limitations of sensor nodes.
The real-time imaging task is decomposed into multiple sub-tasks. Based on the dependencies between sub-tasks and the real-time resource status of sensor nodes, tasks are dynamically allocated to sensor nodes for processing. The total transmission delay and energy consumption are optimized through topology sorting and single-objective cost function, and dynamic adjustments are made in conjunction with the historical path database.
It enables real-time imaging with low transmission latency and low power consumption in dynamic environments, improves the resource utilization of sensor nodes and the robustness of the system, and adapts to changes in sensor node speed and increased task complexity.
Smart Images
Figure CN121979652A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of micro-motion exploration calculation technology, specifically a micro-motion exploration calculation-intensive dynamic allocation method and imaging system. Background Technology
[0002] Microkertic exploration, utilizing subsurface noise data for shear wave velocity structural imaging, has become a cutting-edge field and key direction in seismological research and application. Traditional microkertic exploration methods employ sensor nodes to collect noise data, which is then transmitted back to a central server for centralized processing and imaging. This cluster architecture and centralized management model not only increases resource costs and operational complexity but also leads to severe transmission delays in imaging tasks due to long exploration and construction cycles and cumbersome large-scale data recovery work, making it impossible to obtain subsurface structural information in real time. The lack of real-time imaging methods has become a key bottleneck restricting the development of microkertic exploration in the field of geophysical exploration.
[0003] Edge computing technology, by deploying lightweight edge servers near sensor nodes, provides highly resilient and responsive computing services to meet the low transmission latency requirements of real-time imaging. This physical proximity enables edge servers to respond quickly to imaging task requests, effectively solving the inherent imaging transmission latency problem in traditional micro-motion exploration methods. Furthermore, using edge computing for real-time imaging avoids tedious data recovery work and expensive cluster architecture construction. Applying edge computing to micro-motion exploration systems aims to build a micro-motion exploration edge network with low transmission latency and high reliability. However, the cost constraints and limited computing power of edge servers in micro-motion exploration edge networks are major challenges. Task allocation methods, by distributing computing tasks to sensor nodes for processing, ensure the rational scheduling of resources in the micro-motion exploration edge network. This method not only improves the resource utilization of sensor nodes but also reduces the computational burden on edge servers. Therefore, optimizing task allocation strategies in micro-motion exploration edge networks is crucial, as it can significantly reduce the transmission latency of real-time imaging and improve the overall reliability of the system.
[0004] Existing research typically employs intelligent search algorithms, metaheuristic algorithms, or artificial intelligence to optimize task allocation. While intelligent search algorithms can find the optimal task allocation scheme through iteration, they are prone to getting trapped in local optima. Therefore, in practical applications, the performance of intelligent search algorithms is volatile, and system performance is difficult to guarantee. Artificial intelligence methods require extensive model training, increasing computational costs and being limited by hardware conditions. Furthermore, real-time imaging tasks involve numerous dependency-aware subtasks, but most existing research neglects the impact of these dependencies on task allocation. Similarly, most methods focus on optimization strategies in static environments, with very little research addressing dynamic scenarios. In micro-motion exploration edge network tasks, sensor nodes are limited by environmental interference and residual energy, resulting in stringent operating conditions that traditional static task allocation methods cannot meet. Summary of the Invention
[0005] The purpose of this application is to overcome the current situation where sensor nodes are subject to many constraints in their operation due to environmental interference and residual energy limitations. It proposes a computationally intensive dynamic allocation method and imaging system for micro-motion exploration.
[0006] A computationally intensive dynamic allocation method for micro-motion exploration according to an embodiment of the first aspect of this application includes: Break down the real-time imaging task in the edge server into multiple sub-tasks; Based on the dependencies between subtasks, subtasks are divided into subtasks of serial tasks and subtasks of parallel tasks. Determine the optimal execution order based on the dependency relationship between subtasks; The real-time resource status of each sensor node is evaluated to determine the set of sensor nodes suitable for undertaking the task. From the set of sensor nodes suitable for undertaking the task, the optimal task allocation scheme is obtained by combining the transmission delay and transmission energy consumption required by the sub-task and minimizing the total transmission delay and total transmission energy consumption. Following the optimal execution order and based on the optimal task allocation scheme, subtasks of serial tasks are offloaded to sensor nodes, and subtasks of parallel tasks with no dependencies are offloaded to multiple sensor nodes for simultaneous processing.
[0007] Furthermore, the optimal execution order is determined based on the dependency relationship between subtasks, including: Calculate the in-degree of the subtask; Select a subtask with an in-degree of 0 as the starting task; During the sorting process, whenever a subtask... Once completed, all subsequent subtasks will be updated. in-degree; if the subtask is updated If the in-degree becomes 0, then the subtask will be... Add to queue; Repeat the operation until the in-degree of all subtasks is 0.
[0008] Furthermore, to minimize the total transmission delay and total transmission energy consumption, the optimal task allocation scheme is determined, including: A single-objective cost function is set using total transmission delay and total transmission energy consumption; The constraints are: task offloading uniqueness constraint, computing resource constraint, link bandwidth constraint, energy constraint, and deadline constraint. Minimize the single-objective cost function under constraints to obtain the optimal decision variable. The task allocation scheme corresponding to the optimal decision variable is the optimal task allocation scheme.
[0009] Furthermore, the single-objective cost function is: .in As a weighting factor for the total transmission delay, As a weighting factor for total transmission energy consumption, satisfying , This refers to the transmission delay used for normalization under current network conditions. It is a reference value for transmission energy consumption used for normalization under current network conditions.
[0010] Furthermore, the task unloading uniqueness constraint is that each subtask can only be scheduled onto one sensor node at any given time; The computational resource constraint is: the computational cost of a subtask cannot exceed the capacity of the sensor node itself; The link bandwidth constraint is: the total link transmission rate must not exceed the wireless transmission bandwidth; The energy constraint is: the transmission power consumption of the subtask cannot exceed the remaining energy of the sensor node; The deadline constraint is: subtasks with set deadlines must be completed before the deadline.
[0011] Furthermore, the method also includes, when a triggering condition is met, triggering a dynamic adjustment strategy for the task unloading path to adjust the unloading path of the subtask; the triggering condition includes: The remaining energy of the sensor node is less than the energy trigger threshold; The real-time transmission delay exceeds the transmission delay trigger threshold; The average transmission delay is greater than the target transmission delay, and the average transmission energy consumption is greater than the target transmission energy consumption.
[0012] Furthermore, the strategy of dynamically adjusting the task uninstallation path is triggered to adjust the uninstallation path of the subtask, including: Build a historical task path database to store and continuously update historical task uninstallation path data; After each new subtask is unloaded, the historical average transmission latency and historical average transmission energy consumption of the new subtask are updated using a weighted average method. The updated historical average transmission delay and historical average transmission power consumption are normalized. Each uninstallation path is scored based on the normalization results; The uninstallation path of the subtask is updated by selecting the uninstallation path with the lowest score as the task uninstallation path.
[0013] Furthermore, the total transmission delay is the sum of the transmission delays of all serial task groups and the transmission delays of all parallel task groups; The transmission delay of a serial task group includes: calculating the offloading transmission delay when a subtask needs to be offloaded to a sensor node for processing, and calculating the sum of the transmission delay and the queuing transmission delay to obtain the transmission delay of the subtask. The transmission delay of the serial task is obtained by adding the transmission delays of the subtask and all its predecessor subtasks. The transmission delay of a parallel task group includes: calculating the transmission delay of the subtasks, and taking the transmission delay of the largest subtask in the parallel task group as the transmission delay of the parallel task group.
[0014] Furthermore, the calculation of total transmission energy consumption includes: The sum of the offloading transmission energy consumption and the calculation transmission energy consumption when the subtask needs to be offloaded to the sensor node for processing is used to obtain the transmission energy consumption of the subtask. The total transmission energy consumption is obtained by summing the transmission energy consumption of all subtasks.
[0015] A computationally intensive imaging system for micro-motion exploration according to a second aspect of this application includes an edge server for decomposing real-time imaging tasks into multiple sub-tasks. Based on the dependencies between subtasks, subtasks are divided into subtasks of serial tasks and subtasks of parallel tasks. Determine the optimal execution order based on the dependency relationship between subtasks; The real-time resource status of each sensor node is evaluated to determine the set of sensor nodes suitable for undertaking the task. From the set of sensor nodes suitable for undertaking the task, the optimal task allocation scheme is obtained by combining the transmission delay and transmission energy consumption required by the sub-task and minimizing the total transmission delay and total transmission energy consumption. According to the optimal execution order and the optimal task allocation scheme, the subtasks of serial tasks are offloaded to sensor nodes, and the subtasks of parallel tasks with no dependencies are offloaded to multiple sensor nodes for simultaneous processing. It receives the processing results transmitted back from the sensor nodes and performs imaging based on the processing results.
[0016] The embodiments of this application have at least the following beneficial effects: This application has strong environmental adaptability and robustness. It can adapt to different scenarios such as changes in sensor node processing speed, fluctuations in the number of tasks, and increases in task complexity. It can also improve the utilization rate of sensor node resources by reasonably unloading tasks and alleviate the dependence on edge server resources. It can maintain stable and excellent performance under various conditions. Attached Figure Description
[0017] Figure 1 This is a task allocation architecture diagram of the micro-motion exploration edge network system of this application; Figure 2 Here is a flowchart of the method described in this application; Figure 3 Here is a task model diagram for this application: (a) is a serial task, (b) is a diagram showing a "one-to-many" connection relationship between subtasks, and (c) is a diagram showing a "many-to-many" connection relationship. Figure 4 The diagrams show the total transmission delay and total transmission energy consumption for different numbers of tasks in this application, where (a) is the total transmission delay diagram and (b) is the total transmission energy consumption diagram. Figure 5 The following diagrams show the transmission delay and transmission energy consumption at different sensor node processing speeds in this application: (a) is the total transmission delay diagram, and (b) is the total transmission energy consumption diagram. Figure 6 The diagrams show the transmission delay and transmission energy consumption under different task complexities in this application, where (a) is the total transmission delay diagram and (b) is the total transmission energy consumption diagram. Detailed Implementation
[0018] To better understand the above-mentioned objectives, features, and advantages of this application, the application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of this application.
[0019] This application provides a computationally intensive dynamic task allocation method for micro-motion exploration. By dividing subtasks into serial and parallel tasks, a topology sorting algorithm is used to determine the optimal execution order based on the dependencies between subtasks. For constraints in the micro-motion exploration edge network system, such as deadlines and remaining energy, the optimal task allocation scheme can be continuously and dynamically updated by combining historical path information with the current sensor node status to meet the constraints, effectively reducing system transmission latency and stabilizing transmission energy consumption.
[0020] See Figure 1This paper describes a computationally intensive imaging system for micro-motion exploration. The primary purpose of the edge servers deployed in the micro-motion exploration edge network system is to effectively manage distributed sensor nodes and provide real-time imaging services. Such real-time imaging tasks are typically highly sensitive to transmission latency because they require rapid processing and result return. These tasks usually consist of multiple interdependent subtasks, each of which needs to be executed precisely to ensure the accuracy and real-time nature of the final imaging. If all subtasks are concentrated on the edge server, the limited processing power and resources of the edge server may increase the overall transmission latency, thus affecting the quality of the imaging service. Therefore, to reduce transmission latency and improve efficiency, these subtasks need to be rationally distributed to the various sensor nodes, which possess sufficient computing resources to handle the tasks assigned to them. For example, the edge server offloads different subtasks to the sensor nodes based on the imaging task. In one example, the subtasks could be sensing, preprocessing, and computation, and the subtasks are offloaded to the sensor nodes by the edge server in the optimal execution order.
[0021] During the subtask allocation process, the edge server first determines which sensor nodes are best suited to perform the subtasks based on factors such as the sensor nodes' computing power, load, and network transmission latency. Once the sensor nodes responsible for executing the subtasks are determined, the edge server delegates these subtasks to them. These sensor nodes can process these subtasks in parallel or sequentially according to the specific dependencies between the subtasks. After receiving the subtask, the sensor node uses its computing resources to process it and returns the processing results to the edge server. After receiving the processing results of all subtasks, the edge server integrates and processes them to ultimately generate the required real-time imaging results. In this way, the micro-motion exploration edge network system can effectively utilize the computing resources of distributed sensor nodes to provide efficient, low-latency real-time imaging services.
[0022] The edge server deployment in the micro-motion exploration edge network system aims to manage sensor nodes and provide services. Due to the latency-sensitive nature of the tasks, the overall process can be decomposed into multiple sub-tasks with interdependent relationships, closely linked in data input / output and execution timing. Executing these tasks entirely on the edge server would result in high latency; therefore, sub-tasks need to be distributed to sensor nodes with sufficient computing resources. During task allocation, the edge server first assesses the real-time resource status of each sensor node to determine the suitable set of nodes for the task. This assessment considers factors such as the sensor nodes' computing power, load, and network latency. Subsequently, based on the dependencies and serial / parallel characteristics of the sub-tasks, the sub-tasks are appropriately delegated to the selected nodes. For sub-tasks without dependencies, sensor nodes use parallel processing to improve overall efficiency; for sub-tasks with sequential dependencies, they are executed serially. Finally, the sensor nodes return the calculation results to the edge server to complete real-time imaging.
[0023] See Figure 2 As shown, this application provides a computationally intensive dynamic allocation method for micro-motion exploration based on the aforementioned imaging system, comprising: S1 breaks down the real-time imaging task in the edge server into multiple sub-tasks; S2 categorizes subtasks into subtasks of serial tasks and subtasks of parallel tasks based on the dependencies between them. S3 determines the optimal execution order based on the dependency of subtasks; S4 evaluates the real-time resource status of each sensor node and determines the set of sensor nodes suitable for undertaking the task. S5 selects the optimal task allocation scheme from the set of sensor nodes suitable for undertaking the task, taking into account the transmission delay and transmission energy consumption required by the sub-task, and minimizing the total transmission delay and total transmission energy consumption. S6 follows the optimal execution order and the optimal task allocation scheme to offload subtasks of serial tasks to sensor nodes, and offloads subtasks of parallel tasks with no dependencies to multiple sensor nodes for simultaneous processing.
[0024] In S1, assume the real-time imaging task in the edge server. It can be divided into Each subtask can be represented as a set. , Indicates the first These subtasks, which have dependencies, are modeled as a directed acyclic graph (DAG). ,in This represents a directed edge used to describe the dependencies between subtasks. When a directed edge exists... At that time, subtask Execution time depends on subtask The completion time.
[0025] Dependency-aware subtasks can be divided into three different categories. Figure 3 (a) in the diagram shows a serial task, indicating that there is only a one-to-one connection between subtasks; Figure 2 (b) shows a one-to-many connection between subtasks, while Figure 2 (c) in the example represents a "many-to-many" connection. The latter two types of subtasks are categorized as parallel subtasks. To identify the type of subtask: Indicates a series of subtasks. This indicates a parallel subtask. Figure 3 The serial task shown in (a) consists of multiple one-to-one subtasks. The output of subtask 1 must be used as the input of subtask 2; therefore, subtask 2 must wait for subtask 1 to complete its computation before it can begin execution. In this relationship, subtask 1 is called the predecessor subtask of subtask 2. Regarding... Figure 3 (b) and Figure 3 The parallel tasks shown in (c) require attention to the fact that a subtask will only begin execution after all its predecessor subtasks have completed. For example, in... Figure 3 In (c), subtask 6 will not execute until all its predecessor subtasks (i.e., subtasks 3, 4, and 5) have completed. Therefore, in a parallel relationship, the execution time of a subtask depends on the completion time of the latest of its predecessor subtasks.
[0026] In one embodiment, S2 determines the optimal execution order based on the dependency of subtasks, including: Calculate the in-degree of the subtask; Select a subtask with an in-degree of 0 as the starting task; During the sorting process, whenever a subtask... Once completed, all subsequent subtasks will be updated. in-degree; if the subtask is updated If the in-degree becomes 0, then the subtask will be... Add to queue; Repeat the operation until the in-degree of all subtasks is 0.
[0027] The in-degree metric is introduced to quantify the dependence of a subtask on other subtasks, and an in-degree-based topological sorting is used to determine the execution order of dependent subtasks.
[0028] In the dependency relationships between serial and parallel tasks, in-degree is a core metric describing the degree of dependency between subtasks, used to quantify the strength of a subtask's dependence on other subtasks. Definition: For instantaneous imaging tasks... Subtasks in Its in-degree This indicates the number of predecessor subtasks that the subtask directly depends on, i.e., in the subtask... The total number of subtasks that must be completed before execution can begin. Quantization formula: In real-time imaging tasks... In this context, dependencies are represented by the dependency matrix. Indicates. If This indicates a subtask. Depends on subtasks Subtask It is a subtask The preceding sub-task. Therefore, the sub-task... The in-degree is: , Subtasks It is a subtask The predecessor subtask. For example, if the subtask Kazuko's Mission All are subtasks The precursor sub-task, then ,therefore .
[0029] Topological sorting is used to determine a reasonable execution order in a set of dependent subtasks, ensuring that all dependent tasks are executed only after all their predecessor subtasks have been completed.
[0030] The sorting process is as follows: First, select the subtask with an in-degree of 0 as the starting task; during the sorting process, whenever a subtask... Once completed, all its successor subtasks are updated. in-degree; if the subtask is updated If the in-degree of a task becomes 0, add it to the queue. Repeat this process until the in-degree of all subtasks is 0.
[0031] In one embodiment, minimizing the total transmission delay and total transmission power consumption to obtain the optimal task allocation scheme includes: A single-objective cost function is set using total transmission delay and total transmission energy consumption; The constraints are: task offloading uniqueness constraint, computing resource constraint, link bandwidth constraint, energy constraint, and deadline constraint. Minimize the single-objective cost function under constraints to obtain the optimal decision variable. The task allocation scheme corresponding to the optimal decision variable is the optimal task allocation scheme.
[0032] The total transmission delay is the sum of the transmission delays of all serial task groups and the transmission delays of all parallel task groups; The transmission delay of a serial task group includes: calculating the offloading transmission delay when a subtask needs to be offloaded to a sensor node for processing, and calculating the sum of the transmission delay and the queuing transmission delay to obtain the transmission delay of the subtask. The transmission delay of the serial task is obtained by adding the transmission delays of the subtask and all its predecessor subtasks. The transmission delay of a parallel task group includes: calculating the transmission delay of the subtasks, and taking the transmission delay of the largest subtask in the parallel task group as the transmission delay of the parallel task group.
[0033] When subtask Need to uninstall to the 1st When individual sensor nodes process data, the transmission delay of subtasks It can be decomposed into offloading transmission delay and computation transmission delay. The transmission delay of the subtask. The calculation formula is: ,in Subtasks Uninstall to the The offload transmission delay required for each sensor node Subtasks In the The computational transmission latency required to process data in each sensor node.
[0034] Unload transmission delay This depends on the amount of task data and the link bandwidth between the edge server and the sensor nodes. Offloading transmission latency. It can be represented as: .in, Subtasks The amount of data, Indicates the link bandwidth.
[0035] Calculate transmission delay The computational load of the subtask and the first The processing speed of each sensor node is determined by the speed of the sensor nodes, and its expression is: In the formula, Subtasks The computational load (number of CPU cycles required). Indicates the first Processing speed (CPU frequency) of each sensor node.
[0036] Furthermore, transmission latency is also affected by queue congestion. When subtasks... Scheduled to the When a sensor node is processing data, if that sensor node still has unfinished tasks, a queuing transmission delay will occur. The task arrival rate of each sensor node is Task processing rate Then the queuing transmission delay The calculation formula is: .in, For sensor nodes The load rate is calculated as follows: .
[0037] Therefore, when considering queuing transmission delay After that, subtasks In the Transmission delay on each sensor node It should be expressed as: ; Furthermore, for dependent subtasks, the transmission latency also depends on the type of the subtask. (Definition) For subtasks Completion time. Record subtasks. The set of all precursor subtasks is ,Right now: , It is a set of predecessor-successor dependencies between tasks, defined For subtasks The latest completion time among all precursor sub-tasks, i.e.: If subtask Without any precursor tasks, then .
[0038] Therefore, for serial tasks, the formula for calculating the transmission delay is: . N represents the number of subtasks and N represents the number of sensor nodes. Unload decision variables for subtasks: If This indicates that the subtask has been unloaded to the first... One sensor node; otherwise For parallel tasks, the transmission delay is: The transmission delay of a parallel task group includes: calculating the transmission delay of each subtask, and taking the transmission delay of the largest subtask in the parallel task group as the transmission delay of the parallel task group.
[0039] Real-time imaging tasks involve both serial and parallel tasks; therefore, the total transmission delay must take into account the transmission delays of all serial and parallel tasks. Define a set. Indicates all serial task groups. Indicates the first Grouped serial tasks, each Contains a set of sequential tasks; a set Represents all parallel task groups. Indicates the first Group parallel tasks, each It comprises a set of parallel tasks. Therefore, the total transmission delay of the instantaneous imaging task is the sum of the transmission delays of all serial task groups and all parallel task groups: .in, and These represent the total transmission delay for the serial subtask group and the parallel subtask group, respectively.
[0040] In one embodiment, the calculation of total transmission power consumption includes: The sum of the offloading transmission energy consumption and the calculation transmission energy consumption when the subtask needs to be offloaded to the sensor node for processing is used to obtain the transmission energy consumption of the subtask. The total transmission energy consumption is obtained by summing the transmission energy consumption of all subtasks.
[0041] The task consumes energy during both the transmission and computation phases, including offloading transmission energy and computation transmission energy. Offloading transmission energy... Depends on transmission power With transmission delay The calculation formula is as follows: Transmission power With transmission distance Related, can be expressed as: .in, It is the first Minimum detectable power of each sensor node, It is related to communication frequency The relevant wavelength, This refers to the gain during task transmission.
[0042] Calculate transmission power consumption With the Computational power in each sensor node and calculate transmission delay The relevant calculation formula is as follows: Calculate power It is proportional to the square of the frequency, and can be expressed as: .in These are hardware coefficients. Therefore, the subtask occurs at the sensor node. The computational transmission energy consumption can also be expressed as: .
[0043] Therefore, the total transmission energy consumption is: Since transmission energy consumption depends only on transmission power and transmission delay, and computation transmission energy consumption depends only on computation power and computation transmission delay, the impact of task type (serial or parallel) on transmission energy consumption can be ignored. Therefore, the total transmission energy consumption of the entire real-time imaging task can be simplified as follows: .
[0044] Based on total transmission delay and total transmission energy consumption, the task offloading problem in the micro-motion exploration edge network is defined as a multi-objective optimization problem, with the objective being the same set of decision variables. Simultaneously minimize total transmission delay and total transmission power consumption.
[0045] Since there is often a trade-off between total transmission delay and total transmission energy consumption during task offloading, and different applications have different focuses on optimization objectives, this multi-objective problem is transformed into a single-objective cost function for solution, which takes the following form: .in , These are the weighting factors for total transmission delay and total transmission energy consumption, respectively, satisfying... . , This represents the total transmission delay and total transmission power consumption. , These are reference values for total transmission delay and total transmission energy consumption used for normalization under current network conditions, in order to eliminate differences in dimensions.
[0046] In one embodiment, the task offloading uniqueness constraint is that each subtask can only be scheduled onto one sensor node at any given time; The computational resource constraint is: the computational cost of a subtask cannot exceed the capacity of the sensor node itself; The link bandwidth constraint is: the total link transmission rate must not exceed the wireless transmission bandwidth; The energy constraint is: the transmission power consumption of the subtask cannot exceed the remaining energy of the sensor node; The deadline constraint is: subtasks with set deadlines must be completed before the deadline.
[0047] Since each sensor node has limited computing resources, and the edge server needs to manage the sensor nodes in addition to imaging, the task offloading must meet the following constraints.
[0048] C1: Task offloading uniqueness constraint. Each subtask can only be scheduled onto one sensor node at a time. The constraint expression is: .in Subtasks Unloaded to sensor node .
[0049] C2: Computational resource constraint: The computational cost of a subtask cannot exceed the capacity of the sensor node itself. The sensor node's computational resources are limited; its available computational cost cannot exceed its own capacity. The constraint expression is: Its own capacity , This represents the maximum allowed computation time.
[0050] C3: Link bandwidth constraint. The total link transmission rate must not exceed the wireless transmission bandwidth. Since the wireless transmission bandwidth is inherently limited, exceeding this limit will significantly increase the total transmission latency and total transmission power consumption. Therefore, the total link transmission rate must not exceed the wireless bandwidth. The constraint expression is: , This indicates the wireless transmission bandwidth.
[0051] C4: Energy Constraint: The transmission power consumption of the subtasks cannot exceed the remaining energy of the sensor node. The sensor node has finite energy, and the total transmission power consumption of the subtasks cannot exceed the remaining energy of the sensor node. The constraint expression is: .in, For sensor nodes The remaining energy is equal to the initial energy minus the historical cumulative transmission energy consumption, that is: . For sensor nodes The initial energy (determined by hardware parameters). Let be the total transmission energy consumption of the tasks processed by the sensor node. Therefore, the energy constraint can also be written as: .
[0052] C5: Deadline Constraint. Subtasks with set deadlines must be completed before the deadline. Some urgent subtasks sensitive to transmission delays must also be completed before their deadlines. The constraint expression is as follows: , This represents the deadline. For computational subtasks, their completion time depends only on the processing time required by the corresponding sensor node. However, the completion time of communication tasks is affected by both processing time and transmission latency. Therefore, the deadline calculation formula is: .in, Subtasks Processing time For the duration of the received transmission delay, Let be the transmission delay duration. Therefore, the deadline constraint formula can also be expressed as: .
[0053] Under constraints, the single-objective cost function is solved to obtain the optimal task allocation scheme.
[0054] Subtasks are offloaded to sensor nodes according to the optimal task allocation scheme.
[0055] However, sensor nodes in wireless sensor edge networks have limited energy and experience severe fluctuations in wireless links, making static task offloading methods unable to guarantee the stability and robustness of the system.
[0056] In one embodiment, the current uninstallation path is scored and evaluated, and trigger conditions are set. When the trigger conditions are met, a dynamic adjustment strategy for the task uninstallation path is triggered to adjust the uninstallation path of the subtask. The trigger conditions include: The remaining energy of the sensor node is less than the energy trigger threshold; The real-time transmission delay exceeds the transmission delay trigger threshold; The average transmission delay is greater than the target transmission delay, and the average transmission energy consumption is greater than the target transmission energy consumption.
[0057] Specifically, the dynamic adjustment of the task uninstallation path is triggered when any one of the conditions is met. The triggering conditions include: S1: When the sensor node has remaining energy Triggered at time For sensor nodes The initial energy (determined by hardware parameters). This is achieved by adjusting the parameters. The value can be changed to dynamically adjust the trigger threshold, preventing tasks from failing due to insufficient remaining node energy. This strategy also ensures that the task offloading method always satisfies constraint C4, ensuring that the total transmission energy consumption does not exceed the node's remaining energy.
[0058] S2: When real-time transmission delay Triggered at time This represents the historical average transmission delay. Adjusting parameters... The trigger criteria can be dynamically adjusted to optimize and prevent task timeouts. This strategy ensures that tasks are completed within the deadline, thus satisfying constraint C5. S3: If , and For target transmission delay and target transmission power consumption, If the historical average transmission energy consumption is used, it indicates that the current offloading path score does not match the current environment.
[0059] Furthermore, the strategy of dynamically adjusting the task uninstallation path is triggered to adjust the uninstallation path of the subtask, including: Build a historical task path database to store and continuously update historical task uninstallation path data; After each new subtask is unloaded, the historical average transmission latency and historical average transmission energy consumption of the new subtask are updated using a weighted average method. The updated historical average transmission delay and historical average transmission power consumption are normalized. Each uninstallation path is scored based on the normalization results; The uninstallation path of the subtask is updated by selecting the uninstallation path with the lowest score as the task uninstallation path.
[0060] For serial tasks, the output of a preceding subtask must be used as the input of the next subtask. Offloading subtasks of a serial task to the sensor node with the lowest transmission latency and power consumption can effectively reduce overall transmission latency and power consumption. For parallel subtasks, a subtask must wait for all its preceding subtasks to complete before it can begin execution. Offloading subtasks of a parallel task to multiple sensor nodes for simultaneous processing can reduce computational transmission latency; its completion time depends on the last subtask to complete.
[0061] By combining historical task execution paths with the real-time status of sensor nodes, the algorithm dynamically optimizes task offloading, thereby achieving a balance between transmission latency and transmission power consumption. The core logic of the algorithm is to dynamically adjust offloading decisions by acquiring performance data of task execution paths in similar past scenarios and combining this with the current sensor node status perception, in order to adapt to network changes.
[0062] A historical task path database (HistoryTPD) is constructed to store and continuously update historical task unloading path data to support dynamic unloading decisions. Each path record includes: path identifier (subtask ID, target sensor node ID), link information (link bandwidth). Transmission distance Sensor node status (processing frequency) Remaining energy ), performance metrics (transmission delay) Transmission energy consumption To ensure the timeliness of the data, existing data will be updated accordingly. Sub-task uninstall path After each new subtask is unloaded, its historical average transmission latency and historical average transmission energy consumption are updated using a weighted average. The calculation formula is: , , in This represents the historical number of times the path has been uninstalled. , This represents the transmission latency and power consumption of the current subtask. For the first Historical average transmission delay of the secondary task unloading path For the first Historical average transmission power consumption of the secondary task unloading path For the first Historical average transmission delay of the secondary task unloading path For the first The historical average transmission energy consumption of the secondary task offloading path. Furthermore, to avoid data redundancy, the historical task path database is periodically cleaned up of records that have not been used for a long time or whose performance is significantly worse than the optimal offloading path. To eliminate the difference in units of transmission latency and transmission energy consumption, the updated performance metrics are normalized: , . and These represent the maximum transmission delay and maximum transmission power consumption among all current unloading paths. Based on the normalization results, each unloading path is scored using the following formula: .
[0063] The scoring weights are adjustable and can be flexibly set according to actual needs. Finally, the path with the lowest score is selected as the task unloading path, which is the path that satisfies the optimal combination of total transmission latency and total transmission energy consumption.
[0064] For adjustable rating weights When the current uninstallation path score does not match the current environment, the adjustable score weights can be adjusted. This can effectively improve the adaptability of the offloading scheme. For example, it can increase the adaptability of latency-sensitive tasks. The value is reduced for sensor nodes with insufficient power. value.
[0065] Finding the optimal task unloading path: A path-aware dynamic task unloading method. By combining historical task unloading paths with sensor node states, a quantitative scoring mechanism is used to select the optimal task unloading path. Furthermore, to address dynamic network changes, a feedback optimization mechanism is implemented to dynamically adjust the task unloading path.
[0066] For serial tasks, the output of a preceding subtask must be used as the input of the next subtask. Offloading serial subtasks to sensor nodes with minimal transmission latency and power consumption effectively reduces overall transmission latency and power consumption. For parallel tasks, a subtask must wait for all its preceding subtasks to complete before it can begin execution. Offloading parallel subtasks to multiple sensor nodes for simultaneous processing reduces computational transmission latency; its completion time depends on the last subtask to complete.
[0067] Sensor nodes in wireless sensor edge networks have limited energy and experience significant wireless link fluctuations, making static task offloading methods insufficient to guarantee system stability and robustness. Therefore, this paper proposes a method that combines historical task execution paths with the real-time status of sensor nodes to dynamically optimize task offloading, thereby achieving a balance between transmission latency and energy consumption. The core logic of the algorithm is to dynamically adjust offloading decisions by acquiring performance data of task execution paths in similar past scenarios and combining this with the current sensor node status perception, in order to adapt to network changes.
[0068] In finding the optimal task unloading path, a historical task path database (HistoryTPD) was constructed as the foundation for the path-aware task unloading strategy. This database stores and continuously updates historical task unloading path data to support dynamic unloading decisions. Each path record includes: path identifier (subtask ID, target sensor node ID), link information (bandwidth), etc. Transmission distance Sensor node status (processing frequency) Remaining energy ), performance metrics (transmission delay) Transmission energy consumption To ensure data timeliness, existing paths... After each new subtask is unloaded, its historical average transmission latency and transmission energy consumption are updated using a weighted average. The calculation formula is: , , in This represents the historical number of times this path has been uninstalled. For the transmission delay of the current task, The transmission power consumption for the current task. For average transmission delay, For average transmission power consumption, The historical average transmission delay This represents the historical average transmission energy consumption. Furthermore, to avoid data redundancy, the historical task path database is periodically cleaned of records that have not been used for a long time or whose performance is significantly worse than the optimal path. To eliminate the difference in units between transmission latency and transmission energy consumption, the updated performance metrics are normalized: , . The maximum transmission delay across all current paths. This represents the maximum transmission energy consumption across all current paths. Based on the normalization results, each offloading path is scored using the following formula: The scoring weights are adjustable and can be flexibly set according to actual needs. Ultimately, the path with the lowest score is selected as the task unloading path, which is the path with the optimal combination of transmission latency and transmission energy consumption.
[0069] This application possesses multi-dimensional core advantages in task allocation scenarios. It models task allocation as a multi-objective optimization problem of "transmission delay - transmission energy consumption." Relying on historical task paths and path awareness mechanisms, it can continuously update the optimal task allocation scheme, dynamically select the allocation path with the lowest transmission delay and the richest computing resources, and simultaneously perform topological sorting based on task in-degree to determine the optimal execution order that conforms to task dependencies, effectively optimizing the overall transmission delay. In terms of transmission energy consumption control, this application can sense the path status through historical transmission delay and prioritize sensor nodes with sufficient computing resources. When transmission energy consumption reaches a set threshold, it can dynamically update the allocation path to achieve transmission energy consumption balance, fundamentally avoiding additional iterative transmission energy consumption and energy waste.
[0070] This application verifies its effectiveness through comparative experiments.
[0071] See Figure 4 As shown, the trends in transmission latency and energy consumption are illustrated when the number of tasks increases. With the increase in the number of tasks, the cost of completing subtasks also rises, which is reflected in the increase in computational and transmission latency, leading to a continuous increase in total transmission latency. Furthermore, the system requires more resources to complete task computation and transmission, thus significantly increasing transmission energy consumption. Figure 4 As shown in (a), the transmission latency of the Local algorithm is significantly higher than other methods. This is because the method completes all tasks in the edge server but ignores the utilization of computing resources in the sensor nodes. The QPSO algorithm (Quantum Particle Swarm Optimization) is generally considered a suitable solution for handling finite task scenarios. As the number of tasks increases, the number of iterations required by the QPSO algorithm to determine the optimal task allocation scheme also increases, ultimately leading to an increase in transmission latency. When the number of tasks is between 50 and 200, the transmission latency of this application is slightly lower than that of the MESON algorithm (Mobile Sensing Dependency Offloading Algorithm) and the DATA algorithm (Dependency Sensing Task Allocation Algorithm), which is due to the fact that this application continuously updates the optimal task allocation path through historical task paths. In the task scale stage of 200-400, the increase in transmission latency of the MESON algorithm and the DATA algorithm is significantly higher than that of this application. This phenomenon can be attributed to the dynamic adjustment strategy adopted in this application, which can maintain the optimal performance level in large-scale tasks. Therefore, compared with the MESON algorithm (mobile-aware dependency offloading algorithm) and the DATA algorithm (dependency-aware task allocation algorithm), this application achieves a transmission latency reduction of 11.7% and 19.42% respectively, fully demonstrating its significant advantages in transmission latency optimization. Figure 4As shown in (b), when the number of tasks is small, the transmission energy consumption of the Local algorithm is slightly lower than that of the QPSO algorithm. This is because the Local algorithm only performs computations at the edge server stage and does not need to consider transmission energy consumption. However, as the number of tasks increases, relying solely on the edge server can no longer complete all tasks, and its transmission energy consumption increases significantly. Obviously, compared with other methods, the QPSO algorithm always requires more energy, mainly due to the additional transmission energy consumption generated during the iteration process. Although the MESON and DATA algorithms can reduce transmission latency to some extent, they ignore transmission energy consumption optimization. This application models task allocation as a multi-objective optimization problem, aiming to achieve a dynamic balance between transmission latency and transmission energy consumption. Experimental results show that regardless of the number of tasks, this application consistently maintains lower transmission energy consumption performance in various comparative tests.
[0072] Combination Figure 5 This demonstrates the impact of different sensor node processing speeds on transmission latency and power consumption. The local process executes tasks only on the edge server and is unaffected by variations in sensor node performance. Figure 5 In this context, "Local" is always a horizontal line. For example... Figure 5 As shown in (a), the transmission latency of all allocation methods decreases with the increase in sensor node processing speed, because faster processing speed results in lower computational transmission latency. When the sensor node processing speed increases to 5 kHz, the amplitude attenuation rate decreases significantly. At this point, the transmission latency is mainly attributed to the transmission delay. This application introduces a path-aware mechanism, which can dynamically select the allocation path with the lowest transmission latency and the richest computational resources. This improves the utilization of sensor node computational resources, ensuring that its transmission latency is always kept at a minimum. Figure 5 As shown in (b), the increased processing speed of sensor nodes leads to increased transmission energy consumption, with the transmission energy consumption of the MESON, DATA, and QPSO algorithms showing a linear growth trend. Since the historical transmission delay of this application can be adjusted according to the path status to optimize the allocation scheme, the system can always select sensor nodes with sufficient computing resources. Therefore, the problem of surging energy consumption is avoided by rationally utilizing computing resources. Furthermore, the dynamic adjustment strategy updates the selected allocation path to achieve balance after the transmission energy consumption reaches a set threshold. At a processing speed of 6 kHz, compared with the MESON, DATA, and QPSO algorithms, this application reduces transmission energy consumption by 16%, 24.1%, and 32.01%, respectively. The effectiveness of the proposed allocation method in stabilizing transmission energy consumption has been fully verified.
[0073] Combination Figure 6This demonstrates the impact of task complexity on transmission energy consumption and latency. Task complexity is measured by the task dependency metric, which is quantified by the in-degree value. A higher in-degree value indicates greater task complexity. When handling complex tasks, local algorithms and QPSO algorithms suffer from higher latency and transmission energy consumption due to ignoring inter-task dependencies, and this deficiency becomes increasingly pronounced with increasing task complexity. Figure 6 As shown in (a), although this application, the MESON algorithm, and the DATA algorithm all consider the dependencies between tasks, this application reduces latency by 8.93% and 19.8% respectively compared to the MESON and DATA algorithms. The DATA algorithm uses a static allocation method to distribute parallel tasks to different sensor nodes to reduce parallel latency, but it does not consider the impact of transmission latency. The MESON algorithm shortens queue time through a priority determination algorithm, but it should be noted that this is only one of the factors to consider in reducing overall latency. This application first implements topological sorting based on in-degree, determines the optimal processing order according to task complexity, and then dynamically adjusts the optimal unloading path through a path-aware mechanism. Clearly, this application exhibits more significant advantages compared to other methods. Figure 6 As shown in (b), transmission energy consumption increases with increasing task complexity. This is because executing complex tasks requires more computational resources, naturally consuming more energy. In terms of transmission energy consumption, this application still maintains the best optimization performance, with transmission energy consumption being 19.73% and 26.57% lower than the MESON algorithm and DATA algorithm, respectively.
[0074] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. A computationally intensive dynamic allocation method for micro-motion exploration, characterized in that, include: Break down the real-time imaging task in the edge server into multiple sub-tasks; Based on the dependencies between subtasks, subtasks are divided into subtasks of serial tasks and subtasks of parallel tasks. Determine the optimal execution order based on the dependency relationship between subtasks; The real-time resource status of each sensor node is evaluated to determine the set of sensor nodes suitable for undertaking the task. From the set of sensor nodes suitable for undertaking the task, the optimal task allocation scheme is obtained by combining the transmission delay and transmission energy consumption required by the sub-task and minimizing the total transmission delay and total transmission energy consumption. Following the optimal execution order and based on the optimal task allocation scheme, subtasks of serial tasks are offloaded to sensor nodes, and subtasks of parallel tasks with no dependencies are offloaded to multiple sensor nodes for simultaneous processing.
2. The computationally intensive dynamic allocation method for micro-motion exploration according to claim 1, characterized in that, Determining the optimal execution order based on the dependencies between subtasks includes: Calculate the in-degree of the subtask; Select a subtask with an in-degree of 0 as the starting task; During the sorting process, whenever a subtask... Once completed, all subsequent subtasks will be updated. in-degree; if the subtask is updated If the in-degree becomes 0, then the subtask will be... Add to queue; Repeat the operation until the in-degree of all subtasks is 0.
3. The computationally intensive dynamic allocation method for micro-motion exploration according to claim 1, characterized in that, To minimize total transmission delay and total transmission energy consumption, the optimal task allocation scheme is determined, including: A single-objective cost function is set using total transmission delay and total transmission energy consumption; The constraints are: task offloading uniqueness constraint, computing resource constraint, link bandwidth constraint, energy constraint, and deadline constraint. Minimize the single-objective cost function under constraints to obtain the optimal decision variable. The task allocation scheme corresponding to the optimal decision variable is the optimal task allocation scheme.
4. The computationally intensive dynamic allocation method for micro-motion exploration according to claim 3, characterized in that, The single-objective cost function is: .in As a weighting factor for the total transmission delay, As a weighting factor for total transmission energy consumption, satisfying , This refers to the transmission delay used for normalization under current network conditions. It is a reference value for transmission energy consumption used for normalization under current network conditions.
5. The computationally intensive dynamic allocation method for micro-motion exploration according to claim 3, characterized in that, The task unloading uniqueness constraint is that each subtask can only be scheduled onto one sensor node at any given time. The computational resource constraint is: the computational cost of a subtask cannot exceed the capacity of the sensor node itself; The link bandwidth constraint is: the total link transmission rate must not exceed the wireless transmission bandwidth; The energy constraint is: the transmission power consumption of the subtask cannot exceed the remaining energy of the sensor node; The deadline constraint is: subtasks with set deadlines must be completed before the deadline.
6. The computationally intensive dynamic allocation method for micro-motion exploration according to claim 1, characterized in that, The method further includes, when a triggering condition is met, triggering a dynamic adjustment strategy for the task unloading path to adjust the unloading path of the subtask; the triggering condition includes: The remaining energy of the sensor node is less than the energy trigger threshold; The real-time transmission delay exceeds the transmission delay trigger threshold; The average transmission delay is greater than the target transmission delay, and the average transmission energy consumption is greater than the target transmission energy consumption.
7. The computationally intensive dynamic allocation method for micro-motion exploration according to claim 6, characterized in that, Triggering a dynamic task uninstallation path adjustment strategy adjusts the uninstallation path of subtasks, including: Build a historical task path database to store and continuously update historical task uninstallation path data; After each new subtask is unloaded, the historical average transmission latency and historical average transmission energy consumption of the new subtask are updated using a weighted average method. The updated historical average transmission delay and historical average transmission power consumption are normalized. Each uninstallation path is scored based on the normalization results; The uninstallation path of the subtask is updated by selecting the uninstallation path with the lowest score as the task uninstallation path.
8. The computationally intensive dynamic allocation method for micro-motion exploration according to claim 3, characterized in that, The total transmission delay is the sum of the transmission delays of all serial task groups and the transmission delays of all parallel task groups; The transmission delay of a serial task group includes: calculating the offloading transmission delay when a subtask needs to be offloaded to a sensor node for processing, and calculating the sum of the transmission delay and the queuing transmission delay to obtain the transmission delay of the subtask. The transmission delay of the serial task is obtained by adding the transmission delays of the subtask and all its predecessor subtasks. The transmission delay of a parallel task group includes: calculating the transmission delay of the subtasks, and taking the transmission delay of the largest subtask in the parallel task group as the transmission delay of the parallel task group.
9. The computationally intensive dynamic allocation method for micro-motion exploration according to claim 3, characterized in that, The calculation of total transmission energy consumption includes: The sum of the offloading transmission energy consumption and the calculation transmission energy consumption when the subtask needs to be offloaded to the sensor node for processing is used to obtain the transmission energy consumption of the subtask. The total transmission energy consumption is obtained by summing the transmission energy consumption of all subtasks.
10. A computationally intensive imaging system for micro-motion exploration, used to execute the dynamic allocation method according to any one of claims 1-9, characterized in that, This includes edge servers for breaking down real-time imaging tasks into multiple subtasks; Based on the dependencies between subtasks, subtasks are divided into subtasks of serial tasks and subtasks of parallel tasks. Determine the optimal execution order based on the dependency relationship between subtasks; The real-time resource status of each sensor node is evaluated to determine the set of sensor nodes suitable for undertaking the task. From the set of sensor nodes suitable for undertaking the task, the optimal task allocation scheme is obtained by combining the transmission delay and transmission energy consumption required by the sub-task and minimizing the total transmission delay and total transmission energy consumption. According to the optimal execution order and the optimal task allocation scheme, the subtasks of serial tasks are offloaded to sensor nodes, and the subtasks of parallel tasks with no dependencies are offloaded to multiple sensor nodes for simultaneous processing. It receives the processing results transmitted back from the sensor nodes and performs imaging based on the processing results.