Iot node energy consumption optimization method and system based on edge computing

Abstract

The application discloses an edge-computing-based Internet of Things node energy consumption optimization method and system, and relates to the technical field of edge computing. The method comprises the following steps: an edge gateway dynamically connects Internet of Things nodes to construct an energy consumption portrait; the real-time state of the nodes and to-be-executed task data are regularly collected; energy consumption balance optimization is performed to generate an optimized data set; the data set is used to train a deep reinforcement learning decision maker to generate a task offloading decision; the decision is mapped into a dynamic power control instruction, which is jointly issued to the nodes for execution to realize energy consumption optimization. The technical problems of uneven energy consumption distribution of existing Internet of Things nodes in complex scenarios, lack of cooperation between task offloading and power control, shortened node life and low overall energy efficiency of the system are solved, and the technical effects of realizing adaptive joint optimization of Internet of Things node task offloading and power control, balancing node energy consumption, reducing overall energy consumption and prolonging the operation life of the Internet of Things system are achieved through the combination of edge computing and deep reinforcement learning.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, and more specifically to a method and system for optimizing the energy consumption of IoT nodes based on edge computing. Background Technology

[0002] With the rapid development of IoT technology, massive heterogeneous IoT nodes are widely deployed in smart cities, industrial internet, environmental monitoring, intelligent transportation, and other application scenarios. These IoT nodes are typically characterized by their large number, wide distribution, limited computing and storage capabilities, and constrained power supply methods (such as battery power or energy harvesting). Their long-term stable operation highly depends on energy management and optimization strategies. However, in practical applications, IoT nodes need to continuously perform tasks such as data acquisition, sensing computing, and communication transmission. Significant differences in task load and energy consumption among nodes can easily lead to premature failure of some nodes, thus affecting the overall network lifespan and service quality.

[0003] Traditional IoT energy consumption optimization methods often rely on static parameter configuration or routing-layer-based energy control mechanisms, making it difficult to fully perceive real-time node status changes and lacking the ability to coordinate task offloading and power control. With the development of edge computing technology, edge gateways are gradually acquiring certain computing and intelligent decision-making capabilities, providing IoT nodes with local computing and collaborative control support. However, existing edge computing solutions primarily focus on reducing latency or increasing throughput, and still fall short in terms of refined and dynamic optimization of node energy consumption, especially in complex scenarios with strong node heterogeneity and frequent dynamic changes in tasks, making it difficult to achieve energy balance and maximize network lifetime. Summary of the Invention

[0004] This application provides a method and system for optimizing the energy consumption of IoT nodes based on edge computing, which solves the technical problems of uneven energy consumption distribution, lack of coordination between task offloading and power control, and shortened node lifespan and low overall system energy efficiency of existing IoT nodes in complex scenarios.

[0005] The first aspect of this application provides a method for optimizing the energy consumption of IoT nodes based on edge computing, the method comprising: The system dynamically connects the edge gateway to IoT nodes to construct a node energy consumption profile. A data collection cycle is set, and the IoT nodes are monitored in real-time according to this cycle to obtain real-time status data and data on tasks to be performed. Based on the node energy consumption profile, the real-time status data, and the data on tasks to be performed, the system optimizes node energy consumption balancing to generate an optimized dataset. An edge gateway learning decision-maker is constructed based on the optimized dataset to perform deep reinforcement learning and generate task offloading decisions. These task offloading decisions are mapped to the IoT nodes to generate dynamic power control commands. The task offloading decisions, combined with the dynamic power control commands, are then sent to the IoT nodes for execution to optimize the energy consumption of the IoT nodes.

[0006] A second aspect of this application provides an energy consumption optimization system for IoT nodes based on edge computing, the system comprising: Dynamic Connection Module: Establishes dynamic connections between the edge gateway and IoT nodes, constructing a node energy consumption profile; Real-time Sensing Module: Sets a data collection cycle and performs real-time sensing of IoT nodes according to the cycle, obtaining real-time status data and pending sensing task data of the IoT nodes; Node Energy Consumption Balancing Optimization Module: Optimizes node energy consumption balancing based on the node energy consumption profile, real-time status data, and pending sensing task data, generating an optimized dataset; Deep Reinforcement Learning Module: Constructs an edge gateway learning decision engine based on the optimized dataset and performs deep reinforcement learning to generate task offloading decisions; Energy Consumption Optimization Module: Maps the task offloading decisions to IoT nodes, generates dynamic power control commands, and sends the task offloading decisions in conjunction with the dynamic power control commands to the IoT nodes for execution, optimizing the energy consumption of the IoT nodes.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, IoT nodes are dynamically identified and connected via an edge gateway, establishing a connection between the nodes and the gateway. A corresponding energy consumption profile is constructed by combining the node's static attributes and operational behavior. Then, a reasonable data acquisition cycle is set based on the node's energy status to continuously sense the IoT nodes, acquiring their real-time operational status and pending sensing tasks. Next, by integrating the node's energy consumption profile, real-time status, and task requirements, energy balance analysis and optimization are performed on the nodes, forming an optimized dataset for decision-making. Then, using this optimized dataset, a learning decision model is built on the edge gateway side, generating the optimal task offloading scheme through deep reinforcement learning. Finally, the task offloading results are converted into specific power control commands, which are sent to the IoT nodes along with the offloading decisions for execution. This achieves dynamic control over the node's task processing methods and energy consumption levels, thereby achieving overall energy consumption optimization. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0009] Figure 1 A schematic diagram of the process for optimizing the energy consumption of IoT nodes based on edge computing, provided in an embodiment of this application.

[0010] Figure 2 A schematic diagram of the structure of an IoT node energy consumption optimization system based on edge computing provided in this application embodiment.

[0011] Figure labeling: Dynamic connection module 11, real-time perception module 12, node energy consumption balance optimization module 13, deep reinforcement learning module 14, energy consumption optimization module 15. Detailed Implementation

[0012] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0013] Example 1, as Figure 1 As shown, this application provides a method for optimizing the energy consumption of IoT nodes based on edge computing. The method includes: The edge gateway is positioned to dynamically connect with IoT nodes, and a node energy consumption profile is constructed.

[0014] In this embodiment, the edge gateway first continuously monitors the wireless communication environment within its coverage area, detecting the access status of IoT nodes through signal scanning, identity recognition, or registration mechanisms. When an IoT node enters the communication coverage area of ​​the edge gateway, the edge gateway locates and confirms the IoT node and establishes a communication link, thereby forming a dynamic connection between the node and the edge gateway. During the connection establishment process, the edge gateway collects static attribute information such as node type and hardware configuration from the IoT node, while continuously collecting dynamic behavioral data such as data acquisition frequency, computing load, and communication power during node operation. Subsequently, the edge gateway performs correlation analysis on the static attribute information and dynamic behavioral data to identify the energy consumption characteristics of the IoT node under different operating states, and parametrically describes the energy consumption characteristics to form a node energy consumption profile that reflects the long-term operating energy consumption characteristics of the node. This node energy consumption profile is used to characterize the computing power, remaining energy level, and communication link quality of the IoT node, providing a basis for subsequent data acquisition cycle adjustments, task offloading decisions, and power control, thereby ensuring the feasibility and adaptability of the subsequent energy consumption optimization process.

[0015] Furthermore, methods for dynamically connecting edge gateways and IoT nodes to construct node energy consumption profiles include: The communication range of the edge gateway is identified, and it is determined whether an IoT node enters the communication range. When an IoT node enters the communication range, its static attribute information is extracted, a connection request is generated, and the connection request is synchronized to the edge gateway for response, thus establishing a dynamic connection relationship. The edge gateway collects data on the IoT node according to the dynamic connection relationship to obtain dynamic behavior data. The static attribute information and the dynamic behavior data are correlated and analyzed, and the energy consumption pattern characteristics of the IoT node are identified based on the analysis results. The energy consumption pattern characteristics are mapped to the IoT node for training to construct a node energy consumption profile.

[0016] Preferably, communication parameters are first pre-configured on the edge gateway side, including communication standard, communication channel, maximum transmit power, and communication radius threshold, and the effective communication range of the edge gateway is determined based on these parameters. Subsequently, the edge gateway continuously monitors the communication range according to a preset scanning cycle, determining whether an IoT node has entered the communication range by receiving heartbeat signals or access beacon signals sent by the IoT node. When the edge gateway detects an IoT node entering the communication range, it sends an attribute request command to the IoT node. Upon receiving the attribute request command, the IoT node reads and encapsulates its locally stored static attribute information and returns it. This static attribute information includes at least the node processor model, clock speed parameters, available storage capacity, battery rated capacity, and power supply method. After receiving this static attribute information, the edge gateway generates a corresponding connection request for the IoT node, completes node identity verification, and establishes a dynamic connection relationship between the IoT node and the edge gateway.

[0017] After the dynamic connection is established, the edge gateway sends operation status collection commands to the IoT nodes at preset sampling intervals according to the dynamic connection. Upon receiving the operation status collection commands, the IoT nodes read the current processor utilization, task queue length, instantaneous transmit power, data transmission rate, and remaining battery percentage in real time, and return this data as dynamic behavior data to the edge gateway. The edge gateway then performs correlation analysis between the static attribute information and the dynamic behavior data. Specifically, based on the node processor's clock frequency and processor utilization, the node's executable computation per unit time is calculated by multiplying them, thus quantifying the node's computing power. Based on the node's rated battery capacity, current remaining battery percentage, and energy consumption per unit time, the node's remaining energy value is calculated by multiplying the rated battery capacity by the current remaining battery percentage. This remaining energy value is then divided by the energy consumption per unit time to obtain the node's sustainable operating time. By comparing this sustainable operating time with a preset critical time threshold, the node's remaining energy is categorized into levels, such as low, medium, and high energy levels. Based on the node's received signal strength, data transmission rate, and packet loss rate, the received signal strength is converted to a standard received signal strength through maximum-minimum normalization. The standard data transmission rate is obtained by dividing the data transmission rate by the maximum data transmission rate. The reliability is obtained by subtracting the packet loss rate from 1. Finally, the calculated standard received signal strength, standard data transmission rate, and reliability are weighted to obtain a wireless link stability index, thus quantifying the node's communication link quality.

[0018] After completing the above calculations, the edge gateway uniformly encodes the node's computing power, remaining energy level, and communication link quality to form a node energy consumption pattern feature vector. This feature vector is then mapped to the corresponding IoT node and stored as the node's historical energy consumption record. During continuous node operation, the edge gateway repeatedly acquires the node's dynamic behavior data according to a preset update cycle and recalculates the corresponding energy consumption pattern feature vector based on the latest dynamic behavior data. The newly generated energy consumption pattern feature vector is appended to the node's historical energy consumption record, thus forming a node energy consumption feature sequence. Next, based on the node's historical energy consumption record, the edge gateway performs statistical analysis on the node's energy consumption change trend. By calculating the mean, magnitude, and rate of change of the energy consumption pattern feature vector over time, it extracts the node's stable energy consumption characteristics under different operating stages. These stable energy consumption characteristics reflect the node's average computing power level, energy consumption trend, and communication stability level under long-term operating conditions. Finally, this stable energy consumption characteristic constitutes the node energy consumption profile after training. This node energy consumption profile is stored in the edge gateway in the form of structured data as a training result and is dynamically updated as the node's operating status changes. This enables continuous training and dynamic evolution of the node energy consumption profile, providing a data foundation for subsequent data collection cycle adjustments, node energy consumption balancing optimization, and task offloading and power control decisions.

[0019] Set a data collection cycle, and perform real-time sensing on the IoT nodes according to the data collection cycle to obtain real-time status data and data of sensing tasks to be performed on the IoT nodes.

[0020] In one embodiment, the edge gateway first assesses the energy status and operational capabilities of IoT nodes based on a pre-built node energy consumption profile, determining the node category according to its remaining energy level, computing power, and communication link quality. For different categories of IoT nodes, corresponding data acquisition cycle parameters are configured. Subsequently, the edge gateway generates a sensing schedule table according to the data acquisition cycle for each IoT node and sends sensing request commands to the IoT nodes at the corresponding time nodes based on this schedule table. Upon receiving the sensing request command, the IoT node triggers a local status acquisition process, reading in real time its current remaining energy value, processor load, current operating mode, and communication status, and encapsulates this information as real-time status data before returning it to the edge gateway. Meanwhile, during real-time status acquisition, IoT nodes parse locally cached or newly generated sensing data, identify the sensing tasks that need to be executed according to preset task triggering rules, and extract task attribute information such as task type, task scale, data volume, and latency requirements to generate sensing task data to be executed. This sensing task data to be executed, along with the real-time status data, is uploaded to the edge gateway for subsequent node energy consumption balance analysis, task offloading decisions, and power control decisions. This ensures that the edge gateway can perform dynamic and refined energy consumption optimization control of IoT nodes based on the real-time status and task requirements of the nodes.

[0021] Furthermore, the process and methods for setting the data collection cycle include: Based on the node energy consumption profile, IoT nodes are classified by energy consumption, and multiple energy levels are set. The type parameter of the IoT node is introduced and combined with the multiple energy levels to traverse the IoT nodes for periodic data analysis, setting multiple initial collection cycles. An energy threshold is set, and the multiple energy levels are divided according to the energy threshold. Energy levels below the energy threshold are marked as low energy, identifying the first IoT node. Energy levels greater than or equal to the energy threshold are marked as sufficient energy, identifying the second IoT node. The multiple initial collection cycles are arranged in ascending order to determine the collection cycle sequence. Based on the collection cycle sequence, the first IoT node is traversed in reverse order, and the first matching cycle is associated with the first IoT node as a long data collection cycle. Based on the collection cycle sequence, the second IoT node is traversed in forward order, and the second matching cycle is associated with the second IoT node as a short data collection cycle.

[0022] Preferably, the edge gateway reads the node energy consumption profiles corresponding to each constructed IoT node, and evaluates the current energy status of the IoT node based on the remaining energy parameters recorded in the node energy consumption profiles. According to the proportion of remaining energy to its rated energy, the node energy status is divided into multiple energy levels. These energy levels are set sequentially from low to high according to the proportion of remaining energy, reflecting the energy sufficiency of the IoT node, thus completing the energy consumption classification of the IoT node. After completing the energy consumption classification, the IoT node type parameter is introduced to further analyze the multiple energy levels. This node type parameter includes at least the node's application type, deployment location, and sensing task characteristics. The edge gateway groups the IoT nodes according to the node type parameter, and within each group, classifies the nodes according to their energy levels, forming node subsets based on a combination of node type and energy level. For each node subset, the status reporting records of the nodes during historical operation are retrieved, and the frequency of changes in the node status data is analyzed periodically. In this analysis, the edge gateway uses a preset time window as the analysis unit to perform time-series statistics on the status data reported by nodes within the time window, calculates the time interval between two adjacent status changes, and counts the frequency of status changes. Simultaneously, based on the perception task attributes corresponding to the node type parameters, the edge gateway determines the degree of impact of the frequency of status changes on the task execution results, used to judge the sensitivity of node status perception to the data acquisition cycle. Subsequently, the edge gateway comprehensively analyzes the frequency of status changes in conjunction with the node's current energy level. When a node belongs to a high real-time task type and has a high frequency of status changes, a shorter data acquisition cycle is set for it; when a node belongs to a low real-time task type or has a low frequency of status changes, a longer data acquisition cycle is set for it. Based on the above analysis results, the edge gateway generates corresponding initial data acquisition cycle parameters for different combinations of node types and energy levels, forming multiple initial acquisition cycles. These initial acquisition cycles serve as candidate cycles for subsequent acquisition cycle sequence construction and long / short cycle matching, supporting differentiated data acquisition scheduling strategies.

[0023] Next, a preset energy threshold is obtained. This energy threshold is used to distinguish between nodes with insufficient energy and those with sufficient energy. The edge gateway divides multiple energy levels according to the energy threshold. Energy levels below the energy threshold are marked as low energy levels, and IoT nodes in these low energy levels are identified as the first IoT node. At the same time, energy levels above or equal to the energy threshold are marked as sufficient energy levels, and IoT nodes in these sufficient energy levels are identified as the second IoT node.

[0024] After classifying the nodes, the edge gateway sorts multiple initial data acquisition cycles in ascending order according to their cycle length, generating a data acquisition cycle sequence. For the first IoT node, the edge gateway iterates backwards from the longest cycle in the data acquisition cycle sequence, matching each cycle with the energy level of the first IoT node. When a cycle that meets the energy consumption constraints of the first IoT node is matched, it is designated as a long data acquisition cycle and associated with the corresponding first IoT node to reduce its data acquisition frequency and energy consumption. For the second IoT node, the edge gateway iterates forwards from the shortest cycle in the data acquisition cycle sequence, matching each cycle with the energy level and task requirements of the second IoT node. When a cycle that meets the real-time requirements is matched, it is designated as a short data acquisition cycle and associated with the corresponding second IoT node to improve the real-time performance of node status awareness. By adopting the above methods, we can configure differentiated data collection cycles based on node energy consumption profiles and node types, providing a refined and executable scheduling foundation for subsequent real-time sensing and energy consumption optimization.

[0025] Furthermore, the method for performing real-time sensing of IoT nodes according to the aforementioned data acquisition cycle to obtain real-time status data and data of sensing tasks to be performed by the IoT nodes includes: The edge gateway sends a first sensing request command to the first IoT node according to the long data acquisition cycle; based on the IoT node's response to the first sensing request command, the edge gateway triggers a first sensing request event to report first real-time status data; the edge gateway sends a second sensing request command to the second IoT node according to the short data acquisition cycle; based on the IoT node's response to the second sensing request command, the edge gateway triggers a second sensing request event to report second real-time status data; the first IoT node identifies the first real-time status data to generate first sensing task data to be executed; the second IoT node identifies the second real-time status data to generate second sensing task data to be executed.

[0026] Optionally, firstly, the edge gateway performs perception scheduling on low-energy nodes identified as the first IoT node based on the configured data acquisition cycle parameters. When the trigger time point corresponding to the long data acquisition cycle is reached, the edge gateway sends a first perception request command to the first IoT node, which instructs the first IoT node to perform a status acquisition operation. After receiving the first perception request command, the first IoT node triggers its local status perception module to collect the node's current operating status, which includes at least the remaining energy value, current working mode, processor load, and communication status. The first IoT node encapsulates the collected operating status data and sends it back to the edge gateway. After receiving the operating status data, the edge gateway triggers the first perception request event and records and reports the operating status data as the first real-time status data. Simultaneously, the edge gateway performs high-frequency perception scheduling on high-energy nodes identified as the second IoT node based on the configured short data acquisition cycle parameters. When the trigger time point corresponding to the short data acquisition cycle is reached, the edge gateway sends a second perception request command to the second IoT node, instructing the second IoT node to perform real-time status acquisition. Upon receiving the second sensing request command, the second IoT node immediately performs a local status acquisition operation, collecting information such as its current remaining energy, computing resource usage, task queue status, and wireless communication link quality. The acquired results are then encapsulated and returned to the edge gateway. Upon receiving the status acquisition results, the edge gateway triggers a second sensing request event and records and reports the status acquisition results as second real-time status data.

[0027] After completing the real-time status data acquisition, the first IoT node analyzes the first real-time status data based on preset task triggering rules. These rules include at least state-change-based triggering conditions and time-based triggering conditions. The state-change-based triggering conditions determine whether the first real-time status data has changed beyond a preset threshold compared to the previous acquisition cycle. When a significant change is detected, a sensing task that needs to be executed is determined. The time-based triggering conditions determine whether the latest execution time for the sensing task has been reached. When the latest execution time is reached, a corresponding sensing task needs to be generated. After determining that a sensing task needs to be executed, the first IoT node further identifies the type of the triggering task. This task type includes at least environmental parameter acquisition tasks, status reporting tasks, or anomaly detection tasks. Simultaneously, the first IoT node filters the identified sensing tasks based on the remaining energy and processing capacity of the current node reflected in the first real-time status data, retaining only those sensing tasks that can be executed under the current energy constraints. Subsequently, the first IoT node structures the selected sensing tasks, extracting task attribute information such as task identifier, required computation, expected data volume, and latency requirements for each task. This task attribute information is then encapsulated to form first-stage sensing task data. This first-stage sensing task data characterizes the sensing tasks that the first IoT node needs to execute but has not yet completed within the current acquisition cycle and is sent to the edge gateway as input data for subsequent task offloading decisions and energy consumption optimization control. Similarly, the second IoT node analyzes its current state based on second real-time state data, combining the node's current sensing data generation and task triggering conditions to identify the sensing tasks to be processed, generating second-stage sensing task data. This first-stage and second-stage sensing task data are then uploaded to the edge gateway for subsequent node energy consumption balancing optimization and task offloading decisions, thereby achieving hierarchical sensing and task scheduling based on differences in node energy states.

[0028] Based on the node energy consumption profile, the real-time status data, and the data of the sensing tasks to be executed, node energy consumption balancing optimization is performed to generate an optimized dataset.

[0029] In one embodiment, the edge gateway first acquires the node energy consumption profile of each IoT node and, combined with real-time collected node status data, comprehensively evaluates the current energy status and operational capabilities of the nodes. Subsequently, the edge gateway reads the data of sensing tasks to be executed uploaded by each IoT node, analyzes the task type, task scale, required computation, data transmission volume, and latency constraints corresponding to each sensing task, and matches the task requirements with the current energy status and computing power of the nodes to determine the task carrying capacity of each IoT node in its current state, identifying nodes at risk of insufficient energy or excessive load. Based on this, the edge gateway, aiming at balanced node energy consumption, jointly evaluates and optimizes the network lifetime and energy consumption of each IoT node to achieve a balance in energy consumption among nodes. After completing the above energy consumption balance analysis, the edge gateway organizes the different task allocation methods and their corresponding node energy consumption results, and encapsulates the energy state parameters, task allocation parameters and energy consumption evaluation results of each node in a structured manner to form multiple sample data that can be used for decision training and evaluation. These sample data together constitute the optimization dataset, which is used for the subsequent training and inference of the edge gateway's learning decision-maker to support task offloading and power control decisions based on deep reinforcement learning, thereby achieving dynamic balance optimization of IoT node energy consumption.

[0030] Furthermore, based on the node energy consumption profile, the real-time status data, and the data of the sensing tasks to be executed, node energy consumption balancing optimization is performed to generate an optimized dataset. The method includes: Based on the real-time status data, energy analysis is performed on the IoT nodes to extract their remaining energy values. These remaining energy values ​​are then mapped to the node energy consumption profile for calculation, yielding an average energy consumption rate. Based on the remaining energy values ​​and the average energy consumption rate, an energy sufficiency index for the IoT nodes is set. The data of the sensing tasks to be executed is analyzed according to the energy sufficiency index to generate task execution capability parameters. Energy consumption is calculated based on the task execution capability parameters, and low-energy nodes are identified according to the first calculation result, defining a network lifetime sub-objective. Based on the remaining energy values, a balance calculation is performed on the IoT nodes, and failed nodes are identified according to the second calculation result, defining a node energy consumption sub-objective. The network lifetime sub-objective and the node energy consumption sub-objective are used as a joint optimization objective to optimize node energy consumption balance, generating the optimized dataset.

[0031] Preferably, the edge gateway first receives and parses the real-time status data uploaded by the IoT nodes, extracting energy parameters reflecting the node's current energy status. These parameters include at least the node's remaining battery power, which is used as the node's remaining energy value to characterize its current available energy level. Subsequently, the edge gateway maps the node's remaining energy value to the node's energy consumption profile, performing correlation calculations with the energy consumption characteristics recorded in the profile. By statistically analyzing the node's energy consumption per unit time during historical operation, the average energy consumption rate is calculated. This average energy consumption rate reflects the node's energy consumption rate under normal operating conditions. After obtaining the node's remaining energy value and average energy consumption rate, the edge gateway calculates the expected sustainable operating time based on these parameters and divides this expected sustainable operating time by a standard operating time parameter to construct an energy sufficiency index. This index measures the node's energy sustainability in its current state, reflecting its ability to continue undertaking sensing tasks. Next, the edge gateway parses the data of the sensing tasks to be executed uploaded by the IoT nodes, extracts the task attribute information corresponding to each sensing task, and matches this task attribute information with the node's energy sufficiency index. Specifically, the edge gateway classifies the sensing tasks to be executed based on the node's energy sufficiency index. When the node's energy sufficiency index is in the high energy range, it is determined that the node has the ability to execute multiple sensing tasks simultaneously; when the node's energy sufficiency index is in the medium range, it is determined that the node only has the ability to execute some sensing tasks; when the node's energy sufficiency index is in the low energy range, it is determined that the node only has the ability to execute basic or necessary sensing tasks. Based on the above classification results, the edge gateway, combined with the computational and communication requirements of each sensing task to be executed, selects a set of sensing tasks that can be executed locally by the node under the current energy sufficiency index. It then correlates the selected set of sensing tasks with the node's currently available computing resources to generate a task execution capability parameter characterizing the node's task processing capability. This task execution capability parameter includes at least the number of tasks the node can execute, the total computational load it can handle, and the supported data transmission scale, reflecting the node's comprehensive processing capability for the sensing tasks to be executed under the current energy state.

[0032] After calculating the task execution capability parameters, the edge gateway estimates the energy consumption required for the node to execute the current sensing task based on these parameters. Specifically, it maps the node's executable computational load to corresponding computational energy consumption, and the node's supported data transmission scale to corresponding communication energy consumption. The computational and communication energy consumption are then summed to obtain the node's estimated energy consumption under the current task execution conditions, which serves as the node's first energy consumption calculation result. After obtaining this first energy consumption calculation result, the edge gateway compares it with the node's current remaining energy value. If the node's first energy consumption calculation result is close to or exceeds its remaining energy value, or if it causes the node's remaining energy to fall below a preset safety threshold after completing the current task, the edge gateway determines that the node is at risk of rapid energy depletion under the current task load and identifies the node as a low-energy node. After identifying low-energy nodes, the edge gateway optimizes by reducing the number of low-energy nodes and delaying premature node failure. It defines a network lifetime sub-objective with the goal of extending the overall operating time of the Internet of Things. This network lifetime sub-objective constrains the subsequent node energy consumption balancing optimization process, so that task allocation and execution strategies prioritize avoiding excessive load on low-energy nodes, thereby reducing the probability of premature failure of key nodes in the network and improving the overall network lifetime.

[0033] Simultaneously, the edge gateway performs energy balancing calculations on each IoT node based on its remaining energy value. By comparing the remaining energy distribution among different nodes, it identifies nodes with remaining energy below a preset failure threshold and marks these nodes as failure-risk nodes, thus defining a node energy consumption sub-objective aimed at reducing excessive node energy consumption. Finally, the edge gateway uses the network lifetime sub-objective and the node energy consumption sub-objective as a joint optimization objective to comprehensively optimize and evaluate the task allocation and energy consumption of IoT nodes. The optimized task allocation scheme and node power adjustment strategy are then structured and encapsulated to form an optimized dataset for subsequent decision training and inference.

[0034] Furthermore, the optimization dataset is generated by using the network lifetime sub-objective and the node energy consumption sub-objective as a joint optimization objective to optimize node energy consumption balance, and the method includes: Based on the data of the sensing tasks to be executed, task enumeration is performed to determine multiple task execution locations; the multiple task execution locations are traversed and allocated to construct a task allocation scheme; based on the task allocation scheme, data transmission power analysis is performed to generate multiple candidate node power adjustment strategies, which correspond to the task allocation scheme; the task allocation scheme and the multiple candidate node power adjustment strategies are combined to construct a decision solution space; multiple constraints are preset, and the decision solution space is filtered according to the multiple constraints to obtain decision solutions that satisfy the multiple constraints; the decision solutions are used as feasible solutions for optimization evaluation to obtain the contribution value of the joint optimization objective; based on the contribution value of the joint optimization objective, feasible solutions are filtered to construct the optimization dataset.

[0035] Optionally, the edge gateway first receives and parses the sensing task data uploaded by the IoT nodes, and performs task enumeration for each sensing task. During task enumeration, for each sensing task, its possible task execution positions are determined. These execution positions include at least execution on the original IoT node that generated the sensing task, and offloading the entire sensing task to the edge gateway for execution. For sensing tasks with divisible characteristics, further offloading execution positions are determined, where some tasks are executed locally on the IoT node and the rest are offloaded to the edge gateway, thus generating multiple candidate task execution positions for each sensing task. After determining multiple task execution positions, the edge gateway reads the determined multiple task execution positions for each sensing task and, according to the task number order, combines and traverses the various task execution positions for each sensing task, arranging and combining the execution position selection results of different sensing tasks to construct multiple complete task allocation schemes. Each task allocation scheme is used to explicitly describe which IoT node executes each sensing task under that scheme. After constructing each task allocation scheme, the edge gateway analyzes the relevant data transmission process based on the task allocation scheme. Specifically, the edge gateway identifies the set of tasks requiring data transmission under the task allocation scheme and determines the corresponding data transmission direction and volume. When a task is assigned to local execution, no task offloading data transmission occurs. When a task is assigned to offloading or partial offloading execution, the edge gateway determines the data scale that needs to be transmitted from the IoT node to the edge gateway based on the task data volume recorded in the data of the sensing task to be executed. Subsequently, the edge gateway performs communication analysis on the data transmission process by combining the real-time status data of the IoT node and the communication link quality parameters, and calculates the minimum transmit power required to complete the data transmission under the current wireless link conditions. After completing the above calculation, the edge gateway generates a corresponding node power adjustment strategy for the current task allocation scheme. This node power adjustment strategy is used to indicate how the IoT node adjusts its communication power when executing the task allocation scheme, including the transmit power value or power range of each node during the task offloading process. For different task allocation schemes, the edge gateway generates a one-to-one corresponding node power adjustment strategy, thereby forming a set of multiple candidate node power adjustment strategies.

[0036] Next, the edge gateway combines each task allocation scheme with its corresponding node power adjustment strategy to construct a decision solution space describing different combinations of task execution and power control. Each decision solution in this space contains a set of task execution location selection results and a matching node power adjustment strategy, used to fully characterize an optional task offloading and power control scheme. After constructing the decision solution space, the edge gateway presets multiple constraints, including at least node remaining energy constraints, task latency constraints, communication link quality constraints, and node computing power constraints. The edge gateway filters each decision solution in the decision solution space according to these constraints, eliminating those that do not meet any of the constraints, and retaining those that simultaneously meet multiple constraints as feasible solutions. For the retained feasible solutions, the edge gateway performs optimization evaluation based on the defined network lifetime sub-objective and node energy consumption sub-objective. In this process, the edge gateway calculates the change in remaining energy of each IoT node after executing the current task combination based on the expected energy consumption of each IoT node under the feasible solution, and assesses whether there is a risk of rapid energy decline or premature failure of the node, in order to quantify the contribution of the feasible solution to extending the overall network uptime. During the evaluation of the node energy consumption sub-objective, the edge gateway statistically analyzes the energy consumption distribution of each IoT node under the feasible solution, calculates the energy consumption differences between nodes, and uses this to evaluate the effectiveness of the feasible solution in suppressing excessive node energy consumption and improving energy balance. After completing the independent evaluation of the two sub-objectives, the edge gateway merges the evaluation results of the network lifetime sub-objective and the node energy consumption sub-objective, generating a unified joint optimization objective contribution value through weighted or normalized methods. This joint optimization objective contribution value is used to comprehensively reflect the overall optimization effect of the feasible solution in extending network lifetime and reducing node energy consumption imbalance. Finally, the edge gateway filters and organizes feasible solutions based on the calculated contribution value of the joint optimization objective, and structurally encapsulates the feasible solution with the largest contribution value and its corresponding node state parameters, task allocation schemes, and power adjustment strategies to construct an optimized dataset for subsequent training and inference of the learning decision-maker, thereby providing high-quality training samples for subsequent deep reinforcement learning decision models.

[0037] Based on the optimized dataset, an edge gateway learning decision engine is constructed to perform deep reinforcement learning and generate task unloading decisions.

[0038] In one embodiment, the resulting optimized dataset is first loaded into the edge gateway to initialize the learning decision-maker, enabling it to learn the energy consumption status of nodes and the results of task execution. Subsequently, the edge gateway iterates through the currently connected IoT nodes, summarizing key operational information and analyzing the node's energy status by combining node energy consumption profiles and real-time status data. This yields energy features reflecting the node's energy availability, and simultaneously extracts the scale information of the tasks to be processed from the data of the sensing tasks to be executed. Based on this, the edge gateway performs unified processing of the energy features and task information to form a state representation reflecting the overall state of the current network. This state representation is then input into the learning decision-maker for decision exploration. The learning decision-maker outputs corresponding decision results based on a deep reinforcement learning strategy. By parsing the output results, the execution location selection for each sensing task is obtained, determining whether each sensing task should be executed locally by the IoT node or offloaded to the edge gateway, forming the final task offloading decision to guide subsequent task scheduling and energy consumption optimization.

[0039] Furthermore, based on the optimized dataset, an edge gateway learning decision engine is constructed for deep reinforcement learning to generate task offloading decisions. The method includes: The optimized dataset is transmitted to the edge gateway for initialization, and an edge gateway learning decision-maker is constructed. Key node information of IoT nodes is traversed and aggregated to generate network status information. Energy analysis is performed based on the node energy consumption profile to extract energy sufficiency. Energy analysis is also performed based on the real-time status data to extract the remaining energy percentage. Task analysis is performed based on the data of the sensing tasks to be executed to extract the number of tasks. The energy sufficiency, remaining energy percentage, and number of tasks are normalized based on the network status information to generate normalized information. The normalized information is concatenated sequentially according to the IoT nodes to construct a network status vector. This network status vector is synchronized to the edge gateway learning decision-maker for action exploration to generate network action parameters. The network action parameters are then decoded to construct the task unloading decision.

[0040] Preferably, firstly, the optimized dataset that has been constructed is transmitted to the edge gateway. The edge gateway uses the optimized dataset to initialize and configure the learning decision-maker, enabling the learning decision-maker to have the basic ability to learn and make decisions based on different network states and task execution results.

[0041] Taking a Deep Q-Network (DQN) learning decision-maker as an example, the edge gateway uses an optimized dataset as its training sample set. The network state information of each sample in the set is used as the model input data. This network state information is represented as a network state vector, containing normalized features such as the energy sufficiency, remaining energy percentage, and number of tasks for each IoT node. The network state vector is first input to the input layer of the learning decision-maker. The input layer only receives and transmits network state features and does not perform any computational processing. Subsequently, the network state vector is passed to the first hidden layer. The first hidden layer performs preliminary feature extraction on the input network state features through weighted summation and nonlinear transformation, used to learn the relationship between the overall network energy distribution and task load. The output of the first hidden layer is further passed to the second hidden layer. The second hidden layer performs high-dimensional combination processing on the features from the first hidden layer to characterize the potential correlation between the energy states of different nodes and task offloading selection, thus forming a more discriminative intermediate feature representation. After completing multi-layer feature extraction, the output of the second hidden layer is sent to the output layer. The output layer corresponds to multiple task offloading action options, and each output node represents the evaluation value corresponding to selecting a certain task offloading method in the current network state. The results generated by the output layer are used to characterize the merits of different task offloading decisions under the current network state.

[0042] During training, the edge gateway compares the task offloading decisions generated by the output layer with the actual decision effects recorded in the optimization dataset. It constructs a reward signal based on the contribution value of the joint optimization objective. This reward signal measures the overall effect of the current decision on extending network lifetime and reducing node energy consumption. The edge gateway updates the learning decision-maker in reverse based on the reward signal, propagating the error signal layer by layer from the output layer forward and adjusting the connection weights of each layer. This gradually enhances the model's ability to select decisions with high contribution values. Through multiple rounds of iterative training on samples in the optimization dataset, the learning decision-maker continuously corrects the parameters of each layer, gradually establishing a mapping relationship between network state features and task offloading decisions. After training, the edge gateway's learning decision-maker can quickly output task offloading decisions for each sensing task upon receiving a new network state vector, thereby achieving intelligent and dynamic optimization control of IoT node energy consumption.

[0043] Next, the edge gateway iterates through the currently connected IoT nodes, extracting key node information for each node. This key information includes at least the node identifier, node energy consumption profile, real-time energy status, and current pending tasks. The edge gateway centrally summarizes and correlates this key node information to generate network status information describing the overall operational status of the current IoT system. After obtaining the network status information, the edge gateway performs energy analysis on the IoT nodes based on their energy consumption profiles, extracting energy sufficiency reflecting the long-term energy sustainability of the nodes; it also performs energy analysis on the IoT nodes based on real-time status data, extracting the current remaining energy percentage of the nodes; and simultaneously, it analyzes the current task load of the nodes based on pending sensing task data, extracting the number of pending tasks for each node. After obtaining these features, the edge gateway, based on the network status information, performs unified dimensional processing on energy sufficiency, remaining energy percentage, and task quantity. It converts different types of features into comparable numerical ranges using a maximum-minimum normalization method, generating corresponding normalized information. This normalized information is used to eliminate scale differences between different features, facilitating state recognition and policy learning by the learning decision-maker. Then, the edge gateway sequentially concatenates the normalized information corresponding to each IoT node according to a preset order, constructing a network state vector that comprehensively reflects the overall state of the current network. The edge gateway inputs this network state vector into its learning decision-making unit, which explores actions based on the current policy and outputs network action parameters corresponding to the current state. Finally, the edge gateway parses and decodes these network action parameters, converting them into specific task execution location selection results. This constructs a task offloading decision, indicating whether each pending sensing task should be executed locally by an IoT node or offloaded to the edge gateway, providing direct control for subsequent task scheduling and node energy consumption optimization.

[0044] Furthermore, the method for decoding the network action parameters and constructing the task unloading decision includes: The network action parameters are decoded to generate a task unloading instruction, which is used as a preliminary decision. The execution of the preliminary decision is simulated to monitor energy consumption and generate energy consumption simulation effect data. A reward signal is generated based on the energy consumption simulation effect data. The network state vector, the network action parameters, and the reward signal are combined to construct a data pair. Iterative training of the preliminary decision is performed based on the data pair to construct the task unloading decision.

[0045] Optionally, after the learning decision-maker completes an action exploration, the edge gateway first obtains the corresponding network action parameters. These parameters represent the action results output by the learning decision-maker under the current network state vector input, and are used to characterize the offloading preference for each sensing task to be executed. The edge gateway decodes the network action parameters, converting continuous or discrete action parameters into executable control instructions, generating task offloading instructions for each sensing task. These instructions explicitly indicate whether the corresponding sensing task should be executed locally by the IoT node or offloaded to the edge gateway, and are used as preliminary decisions. After generating the preliminary decisions, the edge gateway does not immediately issue execution orders, but instead performs simulated execution based on the preliminary decisions. Specifically, the edge gateway simulates the computation and data transmission processes of the task at the corresponding execution location based on the task unloading instruction, the current energy status of the nodes, and communication link parameters. It monitors and statistically analyzes the potential computational and communication energy consumption of the nodes, generating energy consumption simulation data reflecting the effectiveness of the initial decision. This data includes at least the computational and communication energy consumption values ​​of each IoT node under the initial decision, as well as the change in remaining energy after the task completion, reflecting the impact of the initial decision on node energy consumption and network operation. Subsequently, the edge gateway calculates multiple evaluation indicators based on the energy consumption simulation data. These indicators include at least node energy consumption indicators and network lifetime impact indicators. The node energy consumption indicator is obtained by weighted summation of the computational and communication energy consumption generated by each IoT node during the simulation, characterizing the overall energy consumption level brought about by the initial decision. The network lifetime impact indicator analyzes the distribution of remaining energy of each node after the simulation, assessing whether there is a significant decrease in remaining energy or a situation approaching the failure threshold, characterizing the degree of impact of the initial decision on extending the overall network operating time.

[0046] After calculating the aforementioned evaluation metrics, the edge gateway normalizes the node energy consumption metrics and network lifetime impact metrics to eliminate dimensional differences between different metrics. Based on preset weight parameters, it weights and fuses each evaluation metric to obtain an energy consumption evaluation result that comprehensively reflects the effectiveness of the initial decision. This energy consumption evaluation result measures the overall performance of the initial decision in reducing energy consumption and improving network stability. Next, the edge gateway generates reward signals based on the energy consumption evaluation results. A positive reward signal is generated when the energy consumption evaluation results indicate that the initial decision effectively reduces node energy consumption, prevents premature failure of low-energy nodes, and helps extend network lifetime. A negative reward signal is generated when the energy consumption evaluation results indicate that the initial decision leads to excessive node energy consumption or increases the risk of node failure. This reward signal serves as feedback information for the subsequent iterative training of the learning decision-maker, guiding it to gradually learn better task offloading strategies.

[0047] Next, the edge gateway combines the current network state vector, the corresponding network action parameters, and the generated reward signal to construct a data pair for training the learning decision-maker. The network state vector describes the network environment at the time of the decision, the network action parameters represent the chosen decision action in that environment, and the reward signal provides feedback on the energy efficiency optimization effect of the action. After constructing the data pair, the edge gateway iteratively trains the learning decision-maker based on iterative training. Specifically, the edge gateway inputs the constructed data pair as training samples into the learning decision-maker. In one training iteration, the edge gateway inputs the network state vector into the input layer of the learning decision-maker. The network state vector sequentially passes through hidden layers for feature extraction and generates prediction results for each task unloading action at the output layer. The edge gateway compares the prediction results generated at the output layer with the corresponding network action parameters in the data pair and calculates the evaluation error of the current decision based on the reward signal. This evaluation error measures the deviation between the current output of the learning decision-maker and the expected decision effect. Then, based on the evaluation error, the edge gateway triggers the reverse update process of the learning decision-maker, propagating the error signal backward from the output layer to the input layer and adjusting the connection weights between layers. During weight adjustment, the weights of unloading actions that generate higher reward signals are increased in the model, while the weights of unloading actions that generate lower reward signals are decreased. This guides the learning decision-maker to gradually favor task unloading strategies with better energy efficiency. After each weight update, the edge gateway applies the updated learning decision-maker to the next set of data pairs, repeating the process of network state input, action prediction, error calculation, and weight updates to complete one training cycle. In multiple training cycles, the learning decision-maker continuously adjusts its parameters using different network states and reward feedback, enabling the model to gradually learn the mapping relationship between network state features and task unloading decisions. As iterative training continues, when the reward signal corresponding to the task unloading decision output by the learning decision-maker in multiple consecutive training cycles tends to stabilize, or the evaluation error falls below a preset threshold, the learning decision-maker is considered to have converged. At this point, the decision strategy formed in the learning decision-maker is solidified as the final task unloading decision. This task unloading decision is used to directly guide the selection of task execution locations between IoT nodes and the edge gateway during actual operation, thereby achieving intelligent optimization and control of IoT node energy consumption.

[0048] The task offloading decision is mapped to the IoT node, a dynamic power control command is generated, and the task offloading decision, together with the dynamic power control command, is sent to the IoT node for execution to optimize the energy consumption of the IoT node.

[0049] In one embodiment, the edge gateway first parses the task offloading decisions output by the learning decision-maker, mapping these decisions to the task execution relationships between each IoT node and the edge gateway. It then matches and processes sensing tasks that need to be offloaded to the edge gateway, extracting multiple offloading tasks. Each offloading task corresponds to a source node that generated it. For the source node, the edge gateway analyzes the wireless communication conditions between the node and the edge gateway using its real-time state data, obtaining wireless link quality parameters reflecting channel quality, transmission stability, and signal strength. Based on these parameters, it determines the minimum communication power level required to ensure reliable transmission of task data, thereby generating a first-type power control command to adjust the transmit power of the node's wireless communication module. Simultaneously, the edge gateway maps the task offloading decisions to the local execution side of the IoT node, matching and processing sensing tasks that are not offloaded and are executed locally by the node, extracting multiple local tasks. For each local task, the edge gateway determines the corresponding local computation cycle based on the task size and the node's computing power, and analyzes the node's computational load under different local computation cycles. The edge gateway assesses task execution complexity based on local computing cycles and generates a second type of power control instruction to adjust node workloads and processor operating states, reducing unnecessary energy consumption during local computation. After generating these two types of power control instructions, the edge gateway integrates them into a dynamic power control instruction that matches the current task offloading decision. The edge gateway then encapsulates the task offloading decision and the corresponding dynamic power control instruction and distributes it to the corresponding IoT node. Upon receiving the joint instruction, the IoT node executes local processing or offloading of the sensing task according to the task offloading decision, and simultaneously adjusts the transmit power and workload of the wireless communication module according to the dynamic power control instruction. By dynamically controlling communication power and workload during task execution, the synergistic optimization between task processing methods and node energy consumption is achieved, thereby reducing the overall energy consumption of the IoT node, delaying node energy depletion, and improving the overall operating efficiency and stability of the IoT system.

[0050] Furthermore, the method for mapping the task offloading decision to IoT nodes and generating dynamic power control commands includes: The task offloading decision is mapped to IoT nodes for matching with edge gateways, extracting multiple offloading tasks, each containing multiple source nodes, with a corresponding relationship between the source nodes and the offloading tasks; based on the multiple source nodes and the real-time status data, wireless link quality parameters are calculated for communication analysis, generating a first type of power control command; the task offloading decision is then mapped to IoT nodes for matching with local nodes, extracting multiple local tasks, each containing multiple local computation cycles, with a corresponding relationship between the local computation cycles and the local tasks; based on the multiple local tasks, the complexity of the task offloading decision is calculated, generating a second type of power control command; the first type of power control execution and the second type of power control command are integrated to generate the dynamic power control command.

[0051] Preferably, the edge gateway first acquires the task offloading decision output by the learning decision unit and maps the task execution relationship between the IoT nodes and the edge gateway based on the task offloading decision. During the mapping process, the edge gateway identifies the sensing tasks that need to be offloaded to the edge gateway for execution, extracting multiple offloading tasks. These offloading tasks are all generated by the IoT nodes and need to be transmitted to the edge gateway for execution via wireless communication links. Simultaneously, the edge gateway determines multiple source nodes corresponding to the multiple offloading tasks. These source nodes are the IoT nodes that generated the corresponding offloading tasks, and each offloading task corresponds one-to-one with its corresponding source node. After determining the multiple source nodes, it extracts wireless communication-related state parameters from the real-time state data uploaded by each source node. These state parameters include at least the node's current remaining energy value, current transmit power level, received signal strength, data transmission rate, and communication packet loss rate, reflecting the real-time communication status between the source nodes and the edge gateway. Subsequently, the edge gateway performs wireless link quality calculations for each source node based on these parameters. That is, it evaluates the stability of the wireless channel by comprehensively analyzing the received signal strength and communication packet loss rate; and it evaluates the link's transmission capacity by comparing the current data transmission rate with the minimum transmission rate required by the task. The edge gateway weights and fuses the above evaluation results to generate corresponding wireless link quality parameters, which are used to characterize the communication reliability and transmission efficiency of the source node when performing task offloading in the current environment.

[0052] After obtaining the wireless link quality parameters, the edge gateway analyzes the communication power of the source node based on these parameters. When the wireless link quality parameters are at a high level, it is determined that the source node can meet the data transmission requirements of task offloading at a lower transmit power. When the wireless link quality parameters are at a low level, it is determined that the transmit power needs to be appropriately increased to ensure the complete transmission of task data. Based on the above determination results, the edge gateway generates a corresponding transmit power adjustment strategy for each source node, which indicates the transmit power value or adjustment range of the node when performing the offloading task. By encapsulating the transmit power adjustment strategies generated by the edge gateway for each source node, a first type of power control instruction is formed. This first type of power control instruction is specifically used to control the wireless communication transmit power of the source node during the task offloading communication process, so as to minimize the node's communication energy consumption while ensuring communication quality and task offloading success rate, thereby achieving fine-grained control of energy consumption during the communication phase of the offloading task.

[0053] Simultaneously, the edge gateway matches the locally executed tasks of IoT nodes based on task offloading decisions, extracting multiple local tasks. These local tasks are sensing tasks determined by the task offloading decisions to be executed locally by the IoT nodes, and each local task corresponds to multiple local computing cycles. Each local computing cycle describes the computing time and processing rhythm of the corresponding local task on the node side, and there is a one-to-one correspondence between local computing cycles and local tasks. After extracting multiple local computing cycles, the edge gateway analyzes the complexity of the task offloading decision based on the computing cycles of multiple local tasks. Specifically, the edge gateway calculates the overall complexity of the current task offloading decision based on multiple local computing cycles. By summarizing and analyzing each local computing cycle, the cumulative local computing time of the computing node under the current task combination is used to reflect the size of the node's local computing load. At the same time, combined with the node's current computing power and real-time status, the cumulative local computing time is corrected to obtain a decision complexity parameter used to characterize the complexity of the task offloading decision.

[0054] After obtaining the decision complexity parameters, the edge gateway analyzes the local operating status of the IoT nodes based on these parameters. When the decision complexity parameters indicate a high local computing load, the gateway determines that the node needs to maintain high processor activity and a short sleep cycle to ensure timely task completion. Conversely, when the decision complexity parameters indicate a low local computing load, the gateway determines that the node can appropriately reduce processor activity or extend the sleep cycle to reduce energy consumption. Based on these analysis results, the edge gateway generates a second type of power control instruction for the IoT nodes executing local tasks. This second type of power control instruction is used to adjust the node's workload and computing module operating status, including strategies for adjusting processor operating frequency, active time percentage, or sleep time length. This reduces energy consumption during local computing while meeting the requirements of local task execution, achieving dynamic optimization control of energy consumption during the local task execution phase.

[0055] Finally, the edge gateway integrates the first type of power control command with the second type of power control command to form a dynamic power control command for each IoT node. This dynamic power control command includes both the wireless communication transmission power adjustment strategy and the node duty cycle adjustment strategy, and is sent to the corresponding IoT node for execution along with the task offloading decision. This forms a collaborative optimization mechanism between task execution location selection and node power control, enabling dynamic and refined optimization control of IoT node energy consumption.

[0056] In summary, the embodiments of this application have at least the following technical effects: First, an edge gateway is dynamically connected to IoT nodes to construct a node energy consumption profile. Next, a data collection cycle is set, and the IoT nodes are monitored in real-time according to this cycle to obtain real-time status data and data on tasks to be performed. Then, based on the node energy consumption profile, the real-time status data, and the data on tasks to be performed, node energy consumption balancing optimization is performed to generate an optimized dataset. Then, based on the optimized dataset, an edge gateway learning decision engine is constructed and deep reinforcement learning is performed to generate task offloading decisions. Finally, the task offloading decisions are mapped to the IoT nodes to generate dynamic power control commands. The task offloading decisions, combined with the dynamic power control commands, are then sent to the IoT nodes for execution, optimizing the energy consumption of the IoT nodes. This approach solves the technical problems of uneven energy consumption distribution and lack of coordination between task offloading and power control in complex scenarios, leading to shortened node lifespan and low overall system energy efficiency. It achieves the technical effect of combining edge computing and deep reinforcement learning to realize adaptive joint optimization of task offloading and power control in IoT nodes, balancing node energy consumption, reducing overall energy consumption, and extending the operational lifespan of the IoT system.

[0057] Example 2, based on the same inventive concept as the edge computing-based IoT node energy consumption optimization method in the previous examples, such as... Figure 2 As shown, this application provides an IoT node energy consumption optimization system based on edge computing, the system comprising: Dynamic Connection Module 11: Locates the edge gateway and dynamically connects to the IoT node to construct a node energy consumption profile; Real-time Sensing Module 12: Sets a data collection cycle and performs real-time sensing of the IoT node according to the data collection cycle to obtain real-time status data and data of sensing tasks to be executed; Node Energy Consumption Balancing Optimization Module 13: Optimizes node energy consumption balancing based on the node energy consumption profile, the real-time status data, and the data of sensing tasks to be executed, generating an optimized dataset; Deep Reinforcement Learning Module 14: Constructs an edge gateway learning decision-maker based on the optimized dataset to perform deep reinforcement learning and generate task offloading decisions; Energy Consumption Optimization Module 15: Maps the task offloading decisions to the IoT node, generates dynamic power control instructions, and sends the task offloading decisions in conjunction with the dynamic power control instructions to the IoT node for execution, thereby optimizing the energy consumption of the IoT node.

[0058] Furthermore, the dynamic connection module 11 is used to perform the following methods: The communication range of the edge gateway is identified, and it is determined whether an IoT node enters the communication range. When an IoT node enters the communication range, its static attribute information is extracted, a connection request is generated, and the connection request is synchronized to the edge gateway for response, thus establishing a dynamic connection relationship. The edge gateway collects data on the IoT node according to the dynamic connection relationship to obtain dynamic behavior data. The static attribute information and the dynamic behavior data are correlated and analyzed, and the energy consumption pattern characteristics of the IoT node are identified based on the analysis results. The energy consumption pattern characteristics are mapped to the IoT node for training to construct a node energy consumption profile.

[0059] Furthermore, the real-time sensing module 12 is used to perform the following methods: Based on the node energy consumption profile, IoT nodes are classified by energy consumption, and multiple energy levels are set. The type parameter of the IoT node is introduced and combined with the multiple energy levels to traverse the IoT nodes for periodic data analysis, setting multiple initial collection cycles. An energy threshold is set, and the multiple energy levels are divided according to the energy threshold. Energy levels below the energy threshold are marked as low energy, identifying the first IoT node. Energy levels greater than or equal to the energy threshold are marked as sufficient energy, identifying the second IoT node. The multiple initial collection cycles are arranged in ascending order to determine the collection cycle sequence. Based on the collection cycle sequence, the first IoT node is traversed in reverse order, and the first matching cycle is associated with the first IoT node as a long data collection cycle. Based on the collection cycle sequence, the second IoT node is traversed in forward order, and the second matching cycle is associated with the second IoT node as a short data collection cycle.

[0060] Furthermore, the real-time sensing module 12 is used to perform the following methods: The edge gateway sends a first sensing request command to the first IoT node according to the long data acquisition cycle; based on the IoT node's response to the first sensing request command, the edge gateway triggers a first sensing request event to report first real-time status data; the edge gateway sends a second sensing request command to the second IoT node according to the short data acquisition cycle; based on the IoT node's response to the second sensing request command, the edge gateway triggers a second sensing request event to report second real-time status data; the first IoT node identifies the first real-time status data to generate first sensing task data to be executed; the second IoT node identifies the second real-time status data to generate second sensing task data to be executed.

[0061] Furthermore, the node energy consumption balancing optimization module 13 is used to perform the following method: Based on the real-time status data, energy analysis is performed on the IoT nodes to extract their remaining energy values. These remaining energy values ​​are then mapped to the node energy consumption profile for calculation, yielding an average energy consumption rate. Based on the remaining energy values ​​and the average energy consumption rate, an energy sufficiency index for the IoT nodes is set. The data of the sensing tasks to be executed is analyzed according to the energy sufficiency index to generate task execution capability parameters. Energy consumption is calculated based on the task execution capability parameters, and low-energy nodes are identified according to the first calculation result, defining a network lifetime sub-objective. Based on the remaining energy values, a balance calculation is performed on the IoT nodes, and failed nodes are identified according to the second calculation result, defining a node energy consumption sub-objective. The network lifetime sub-objective and the node energy consumption sub-objective are used as a joint optimization objective to optimize node energy consumption balance, generating the optimized dataset.

[0062] Furthermore, the node energy consumption balancing optimization module 13 is used to perform the following method: Based on the data of the sensing tasks to be executed, task enumeration is performed to determine multiple task execution locations; the multiple task execution locations are traversed and allocated to construct a task allocation scheme; based on the task allocation scheme, data transmission power analysis is performed to generate multiple candidate node power adjustment strategies, which correspond to the task allocation scheme; the task allocation scheme and the multiple candidate node power adjustment strategies are combined to construct a decision solution space; multiple constraints are preset, and the decision solution space is filtered according to the multiple constraints to obtain decision solutions that satisfy the multiple constraints; the decision solutions are used as feasible solutions for optimization evaluation to obtain the contribution value of the joint optimization objective; based on the contribution value of the joint optimization objective, feasible solutions are filtered to construct the optimization dataset.

[0063] Furthermore, the deep reinforcement learning module 14 is used to perform the following methods: The optimized dataset is transmitted to the edge gateway for initialization, and an edge gateway learning decision-maker is constructed. Key node information of IoT nodes is traversed and aggregated to generate network status information. Energy analysis is performed based on the node energy consumption profile to extract energy sufficiency. Energy analysis is also performed based on the real-time status data to extract the remaining energy percentage. Task analysis is performed based on the data of the sensing tasks to be executed to extract the number of tasks. The energy sufficiency, remaining energy percentage, and number of tasks are normalized based on the network status information to generate normalized information. The normalized information is concatenated sequentially according to the IoT nodes to construct a network status vector. This network status vector is synchronized to the edge gateway learning decision-maker for action exploration to generate network action parameters. The network action parameters are then decoded to construct the task unloading decision.

[0064] Furthermore, the deep reinforcement learning module 14 is used to perform the following methods: The network action parameters are decoded to generate a task unloading instruction, which is used as a preliminary decision. The execution of the preliminary decision is simulated to monitor energy consumption and generate energy consumption simulation effect data. A reward signal is generated based on the energy consumption simulation effect data. The network state vector, the network action parameters, and the reward signal are combined to construct a data pair. Iterative training of the preliminary decision is performed based on the data pair to construct the task unloading decision.

[0065] Furthermore, the energy consumption optimization module 15 is used to perform the following method: The task offloading decision is mapped to IoT nodes for matching with edge gateways, extracting multiple offloading tasks, each containing multiple source nodes, with a corresponding relationship between the source nodes and the offloading tasks; based on the multiple source nodes and the real-time status data, wireless link quality parameters are calculated for communication analysis, generating a first type of power control command; the task offloading decision is then mapped to IoT nodes for matching with local nodes, extracting multiple local tasks, each containing multiple local computation cycles, with a corresponding relationship between the local computation cycles and the local tasks; based on the multiple local tasks, the complexity of the task offloading decision is calculated, generating a second type of power control command; the first type of power control execution and the second type of power control command are integrated to generate the dynamic power control command.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for optimizing the energy consumption of IoT nodes based on edge computing, characterized in that, The method includes: Locate and dynamically connect edge gateways and IoT nodes to build node energy consumption profiles; Set a data collection cycle, and perform real-time sensing on the IoT nodes according to the data collection cycle to obtain real-time status data and sensing task data to be executed of the IoT nodes; Based on the node energy consumption profile, the real-time status data, and the data of the sensing tasks to be executed, the node energy consumption balance is optimized to generate an optimized dataset. Based on the optimized dataset, an edge gateway learning decision engine is constructed for deep reinforcement learning to generate task unloading decisions. The task offloading decision is mapped to the IoT node, a dynamic power control command is generated, and the task offloading decision is combined with the dynamic power control command and sent to the IoT node for execution to optimize the energy consumption of the IoT node. The positioning edge gateway dynamically connects to IoT nodes and constructs node energy consumption profiles, the method of which includes: Identify the communication range of the edge gateway and determine whether an IoT node has entered the communication range; When an IoT node enters the communication range, the static attribute information of the IoT node is extracted, a connection request is generated, the connection request is synchronized to the edge gateway for response, and a dynamic connection relationship is established. The edge gateway collects data on the operation of IoT nodes according to the dynamic connection relationship to obtain dynamic behavior data; The static attribute information is correlated with the dynamic behavior data, and the energy consumption pattern characteristics of the IoT node are identified based on the analysis results. The energy consumption pattern features are mapped to IoT nodes for training to construct node energy consumption profiles; The method of mapping the task offloading decision to an IoT node and generating dynamic power control commands includes: The task offloading decision is mapped to the IoT node to match the edge gateway, and multiple offloading tasks are extracted. The multiple offloading tasks contain multiple source nodes, and there is a corresponding relationship between the multiple source nodes and the multiple offloading tasks. Based on the multiple source nodes and the real-time status data, calculations are performed to determine the wireless link quality parameters, conduct communication analysis, and generate a first type of power control command. The task unloading decision is mapped to the IoT node for local matching, and multiple local tasks are extracted. The multiple local tasks contain multiple local computing cycles, and there is a corresponding relationship between the multiple local computing cycles and the multiple local tasks. Based on the complexity of the task offloading decision calculated from the multiple local tasks, a second type of power control instruction is generated. The first type of power control instruction is integrated with the second type of power control instruction to generate the dynamic power control instruction.

2. The energy consumption optimization method for IoT nodes based on edge computing as described in claim 1, characterized in that, The process and methods for setting the data acquisition cycle include: Based on the node energy consumption profile, the energy consumption of IoT nodes is classified and multiple energy levels are set. By introducing the type parameter of the Internet of Things (IoT) node and combining it with the multiple energy levels, the IoT nodes are traversed to perform periodic data analysis, and multiple initial collection cycles are set. Set an energy critical threshold, and divide multiple energy levels according to the energy critical threshold; Energy levels below the energy critical threshold are identified as low-energy, and the first IoT node is determined. Energy levels greater than or equal to the energy critical threshold are identified as having sufficient energy to determine the second Internet of Things node; Arrange the multiple initial acquisition cycles in ascending order to determine the acquisition cycle sequence; Based on the acquisition cycle sequence, the first IoT node is traversed in reverse order, and the first matching cycle is associated with the first IoT node as the long data acquisition cycle. Based on the acquisition cycle sequence, the second IoT node is traversed in a forward manner, and the second matching cycle is associated with the second IoT node as a short data acquisition cycle.

3. The energy consumption optimization method for IoT nodes based on edge computing as described in claim 2, characterized in that, The method involves performing real-time sensing on IoT nodes according to the aforementioned data acquisition cycle to obtain real-time status data and data for sensing tasks to be performed on the IoT nodes. Based on the edge gateway, a first sensing request instruction is sent to the first IoT node according to the long data collection cycle; Based on the IoT node's response to the first sensing request instruction, the edge gateway is triggered to report the first real-time status data through the first sensing request event. Based on the edge gateway sending a second sensing request instruction to the second IoT node according to the short data collection cycle; Based on the IoT node's response to the second sensing request instruction, the edge gateway is triggered to report the second real-time status data through the second sensing request event. The first IoT node identifies the first real-time status data and generates the first sensing task data to be executed. The second IoT node identifies the second real-time status data and generates the second sensing task data to be executed.

4. The energy consumption optimization method for IoT nodes based on edge computing as described in claim 1, characterized in that, Based on the node energy consumption profile, the real-time status data, and the data of the sensing tasks to be executed, node energy consumption balancing optimization is performed to generate an optimized dataset. The method includes: Based on the real-time status data, energy analysis is performed on the IoT nodes to extract the remaining energy value of the nodes; The remaining energy value of the node is mapped to the node energy consumption profile for calculation to obtain the average energy consumption rate; The energy sufficiency index of the IoT node is set based on the node's remaining energy value and the average energy consumption rate. The data of the sensing task to be executed is analyzed according to the energy sufficiency index to generate task execution capability parameters; Energy consumption is calculated based on the task execution capability parameters, low-energy nodes are identified according to the first calculation result, and a network lifetime sub-objective is defined. Based on the remaining energy value, the IoT nodes are balanced, and the failed nodes are identified according to the second calculation result. The node energy consumption sub-target is defined. The network lifetime sub-objective and the node energy consumption sub-objective are used as joint optimization objectives to optimize node energy consumption balance, thereby generating the optimized dataset.

5. The energy consumption optimization method for IoT nodes based on edge computing as described in claim 4, characterized in that, The optimization dataset is generated by using the network lifetime sub-objective and the node energy consumption sub-objective as a joint optimization objective to optimize node energy consumption balance, and the method includes: Based on the data of the sensing tasks to be executed, task enumeration is performed to determine the execution positions of multiple tasks. The task allocation scheme is constructed by traversing the multiple task execution locations and allocating them. Based on the task allocation scheme, data transmission power analysis is performed to generate multiple candidate node power adjustment strategies. The multiple candidate node power adjustment strategies have a corresponding relationship with the task allocation scheme. The task allocation scheme is combined with the power adjustment strategies of the multiple candidate nodes to construct a decision solution space; Multiple constraints are preset, and the decision solution space is filtered according to the multiple constraints to obtain decision solutions that satisfy the multiple constraints. The decision solutions are used as feasible solutions for optimization evaluation to obtain the contribution value of the joint optimization objective. Feasible solutions are selected based on the contribution values ​​of the joint optimization objective to construct the optimization dataset.

6. The method for optimizing the energy consumption of IoT nodes based on edge computing as described in claim 1, characterized in that, Based on the optimized dataset, an edge gateway learning decision engine is constructed and deep reinforcement learning is performed to generate task offloading decisions. The method includes: The optimized dataset is transmitted to the edge gateway for initialization, and an edge gateway learning decision-maker is constructed. Traverse the key node information of IoT nodes, aggregate the key node information, and generate network status information; Energy analysis is performed based on the node energy consumption profile to extract energy sufficiency. Energy analysis is performed based on the real-time status data to extract the percentage of remaining energy. Based on the data of the sensing tasks to be executed, perform task analysis and extract the number of tasks; Based on the network status information, the energy sufficiency, the remaining energy percentage, and the number of tasks are normalized to generate normalized information; The normalized information is concatenated sequentially according to the IoT nodes to construct a network state vector. The network state vector is then synchronized to the edge gateway learning decision-maker to explore actions and generate network action parameters. The network action parameters are decoded to construct the task unloading decision.

7. The energy consumption optimization method for IoT nodes based on edge computing as described in claim 6, characterized in that, Decoding the network action parameters to construct the task unloading decision includes the following methods: The network action parameters are decoded to generate a task unloading instruction, which is then used as a preliminary decision. The energy consumption is monitored by simulating the execution of the preliminary decision, and energy consumption simulation effect data is generated. A reward signal is generated based on the energy consumption simulation effect data. The network state vector, the network action parameters, and the reward signal are combined to construct data pairs; Based on the data, iterative training is performed to make preliminary decisions, and the task unloading decision is constructed.

8. An IoT node energy consumption optimization system based on edge computing, characterized in that, The system is used to implement the edge computing-based IoT node energy consumption optimization method according to any one of claims 1-7, the system comprising: Dynamic connection module: Locates the edge gateway to dynamically connect with IoT nodes and builds a node energy consumption profile; Real-time sensing module: Sets a data acquisition cycle, performs real-time sensing of IoT nodes according to the data acquisition cycle, and obtains real-time status data and sensing task data to be executed of IoT nodes; Node energy consumption balancing optimization module: Based on the node energy consumption profile, the real-time status data, and the data of the sensing tasks to be executed, the module optimizes node energy consumption balancing and generates an optimized dataset. Deep reinforcement learning module: Based on the optimized dataset, an edge gateway learning decision engine is constructed to perform deep reinforcement learning and generate task unloading decisions; Energy consumption optimization module: Maps the task offloading decision to the IoT node, generates dynamic power control instructions, and sends the task offloading decision in conjunction with the dynamic power control instructions to the IoT node for execution, thereby optimizing the energy consumption of the IoT node.

Images