An Optimization Method for Edge Computing Based on the Internet of Things

Abstract

This invention belongs to the technical field of computer resource allocation and relates to an edge computing optimization method based on the Internet of Things (IoT). The method includes: acquiring real-time working status information of edge nodes; determining the energy consumption type of nodes based on the real-time working status information and outputting a node type tag set; decomposing the computing task to be processed into independent execution unit blocks; generating a communication energy consumption compensation value list; establishing a dynamic matching link based on the node type tag set, independent execution unit blocks, and the communication energy consumption compensation value list, ensuring that the total communication energy consumption compensation value between all associated nodes is less than a predetermined limit; sending operation instructions carrying execution block data to the node group corresponding to the dynamic matching link to trigger distributed computing; triggering a reclassification of energy consumption types, generating a reorganized execution block set, and activating the establishment of a new dynamic matching link. This invention addresses the problem of unquantified communication energy consumption compensation effects caused by node collaboration, which leads to hidden energy consumption overhead during cross-node task allocation.

Description

Technical Field

[0001] This invention belongs to the technical field of computer resource allocation and relates to an edge computing optimization method based on the Internet of Things. Background Technology

[0002] In current edge computing environments, massive heterogeneous nodes exhibit significant differences in energy consumption behavior and dynamic resource fluctuations. Traditional static task allocation schemes struggle to adapt to real-time changes in node power, temperature, and load. Especially in mobile edge scenarios, the instability of communication links caused by node displacement further exacerbates the risk of uncontrollable system energy consumption. The mismatch between task urgency requirements and node energy consumption characteristics often leads to delays in critical tasks or overload of inefficient nodes.

[0003] Existing technologies primarily employ load balancing algorithms or fixed sharding strategies for optimization. Load balancing schemes implement round-robin scheduling based on fundamental metrics such as node CPU / memory utilization, while fixed sharding strategies divide computational tasks according to a preset ratio. Some improved solutions introduce simple power threshold mechanisms, stopping task allocation when a node's power falls below a critical value. While this can alleviate local resource constraints, it fails to establish a deep correlation between energy consumption behavior patterns and task characteristics.

[0004] Based on the above problems, the communication energy consumption compensation effect generated by node collaboration is not quantified, resulting in hidden energy consumption overhead when allocating tasks across nodes, causing a decrease in the system's effective computing power, uncontrolled energy consumption peaks, and an increase in the interruption rate of critical tasks. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides an edge computing optimization method based on the Internet of Things.

[0006] An edge computing optimization method based on the Internet of Things includes the following steps:

[0007] S1. Obtain real-time working status information of edge nodes, including chip power consumption rate, remaining power percentage and temperature data;

[0008] S2. Determine the energy consumption type of the node based on the real-time working status information, and output a set of node type tags, including direct-through, buffer, relay and floating types;

[0009] S3. The computational task to be processed is decomposed into independent execution unit blocks, which are divided into main execution blocks, secondary execution blocks and fragment execution blocks according to the urgency of the task.

[0010] S4. Predict the communication energy consumption compensation value of different node type tag combinations, set an additional energy consumption compensation value when the physical distance between nodes exceeds the physical distance threshold or the energy consumption types are mutually exclusive, and generate a list of communication energy consumption compensation values.

[0011] S5. Establish a dynamic matching link based on the node type tag set, independent execution unit block and communication energy consumption compensation value list, including assigning the main execution block to the direct type node, injecting the secondary execution block into the preload queue of the buffer type node, splitting the fragment execution block to multiple relay type nodes, and ensuring that the total communication energy consumption compensation value between all associated nodes is less than the predetermined limit.

[0012] S6. Send an operation command carrying execution block data to the node group corresponding to the dynamic matching link to trigger distributed computing;

[0013] S7. When any node reaches the overload critical point, trigger the re-division of energy consumption type, generate a set of reorganized execution blocks and activate the establishment of a new dynamic matching link.

[0014] A further aspect of the present invention involves obtaining real-time working status information of edge nodes, including the following steps:

[0015] The chip's energy consumption rate is obtained by periodically collecting the chip's power loss value through the preset status monitoring component; the remaining capacity ratio reading of the energy storage element provided by the battery management chip is the remaining power percentage; and the quantified value of the heat generation reported by the thermal sensing element distributed in the core area of ​​the chip is the temperature data.

[0016] Perform fluctuation verification on parameter values ​​within a continuous acquisition period: if the fluctuation of the same parameter in five consecutive acquisitions exceeds the preset fluctuation threshold, discard the last data and re-acquire; if the fluctuation of the last data in three consecutive acquisitions is less than the preset stability threshold, use the last data.

[0017] The verified chip power consumption rate, remaining power percentage, and temperature data are transmitted to the central scheduling system.

[0018] A further aspect of this invention involves determining the type of energy consumption at a node, including the following steps:

[0019] The logic compares the three elements of the acquired chip power consumption rate, remaining power percentage, and temperature data with the preset thresholds, and reads the classification threshold set in the configuration file. The classification threshold set includes the direct-type power consumption threshold, the floating power threshold, and the overheat protection threshold.

[0020] First, check whether the remaining power percentage triggers the floating condition. If the floating condition is not triggered, sequentially filter through the through nodes that simultaneously meet the following conditions: chip power consumption rate is greater than the through power consumption threshold, remaining power percentage is higher than the floating power threshold, and temperature data has not reached the overheat protection threshold.

[0021] The next screening selects buffer nodes that simultaneously meet the following criteria: chip power consumption rate is lower than the buffered type power consumption limit, remaining power percentage is higher than the floating type power threshold, and have data caching hardware.

[0022] The rest are classified as relay-type conditions by default, ensuring that each node has one and only one type of energy consumption.

[0023] A further aspect of this invention includes node energy consumption type categories, including:

[0024] Straight-through type refers to a node type that is continuously in a state of high computational intensity and is suitable for real-time computing tasks that cannot be interrupted.

[0025] Buffered type refers to a node type that has a data preloading hardware structure and whose chip power consumption rate is lower than the buffered type power consumption limit, reducing computation latency by preloading data;

[0026] Relay type, specifically refers to a node type suitable for supporting task sharding processing, which reduces the peak energy consumption of a single node by splitting tasks;

[0027] Floating type refers to the forced conversion type when the remaining power percentage is lower than the floating power threshold. Regardless of the original type of the node, it will switch to this type, and its behavior mode will automatically inherit the task processing method of the relay type node.

[0028] A further aspect of the present invention involves decomposing the computational task to be processed into independent execution unit blocks, including the following steps:

[0029] Based on the continuity of task functions, detect whether there are separable computational sub-units;

[0030] By traversing the data input / output table of the calculation instructions, it is detected whether there is a sequence of instructions that must be executed consecutively.

[0031] The calculation sub-units are classified into different timing levels according to the allowable response delay value, namely 00 = real-time, 01 = non-real-time, and 10 = background. The real-time level indicates millisecond-level response requirements, the non-real-time level represents the allowable delay of hundreds of milliseconds, and the background level indicates no time limit requirement.

[0032] After the division is completed, three types of independent execution unit blocks are output.

[0033] A further embodiment of the present invention includes three independent execution unit blocks, comprising the following steps:

[0034] The task sub-unit marked as the highest priority forms the main execution block. The determination is based on the task packet header response delay requirement field value being equal to the real-time level, or the functional continuity test showing that the interruption of this module will cause the entire task to fail.

[0035] The decision to allow task sub-units with limited delay to form secondary execution blocks is based on the task packet header response delay requirement field value being non-real-time level, and functional testing showing that delay processing does not affect the correctness of the final result.

[0036] Background tasks without a fixed completion time form fragmented execution blocks. The criteria for this are: the response delay requirement field in the task package header is set to the background level, or the function continuity test shows that it is a non-critical statistical operation and is automatically classified into this category.

[0037] A further aspect of the present invention involves generating a communication energy consumption compensation value list, comprising the following steps:

[0038] Obtain the node energy consumption type category and the coordinate location table of deployed nodes. The node coordinate location table is collected by a preset location sensing component.

[0039] When the physical distance between any two nodes in a node combination exceeds a preset distance threshold, a distance penalty mechanism is activated: the distance penalty component is generated by multiplying the excess distance value by a preset distance penalty coefficient.

[0040] Query the preset type mutual exclusion relationship table, and when there is mutual exclusion in the combination of node types, load a fixed penalty value to generate a mutual exclusion penalty component;

[0041] Accumulate the distance penalty component and the mutual exclusion penalty component to generate a communication energy consumption compensation value, and output a list of compensation values.

[0042] A further aspect of the present invention establishes a dynamic matching link, comprising the following steps:

[0043] The main execution block is assigned to the lightest load pass-through node, which is determined by comparing temperature data fluctuations in real-time operating status information;

[0044] The execution block is injected into the preset preload queue of the buffered node, and the preload queue adopts a first-in-first-out management mechanism;

[0045] The fragmented execution block is broken down into micro-task units according to the sharding algorithm and distributed to multiple relay nodes. The number of shards is determined by multiplying the number of available nodes by a preset load adjustment coefficient.

[0046] The total compensation value is obtained by accumulating the communication energy consumption compensation values ​​of the associated node pairs. When the total compensation value is exceeded, the node pair with the largest compensation value is replaced, and a dynamic matching link topology map is generated.

[0047] A further aspect of the present invention involves sending operation instructions carrying execution block data, comprising the following steps:

[0048] The structure consists of three parts: the header identifier segment stores the node's physical address code and the category of its independent execution block; the payload segment encapsulates the complete original data of the independent execution unit block; and the collaboration information appendix loads the communication topology relationship data, listing the addresses of the partner nodes that the current node needs to interact with and the corresponding communication energy consumption compensation values.

[0049] Transmit command packets in the order of direct-access nodes, buffer nodes, and relay nodes;

[0050] After all nodes return a confirmation signal, a distributed computing start command carrying a synchronization timestamp is issued.

[0051] A further aspect of this invention involves triggering a reclassification of energy consumption types, including the following steps:

[0052] When the chip's power consumption rate continuously exceeds the preset overload threshold, the node will be marked as floating and a type change notification will be broadcast.

[0053] Freeze the independent execution unit blocks processed by the node and roll back the operation instructions, and encapsulate the incomplete computation state to generate execution blocks to be reassembled;

[0054] The execution block to be reorganized is submitted to the task decomposition step for secondary decomposition, generating a set of reorganized execution blocks, and skipping floating nodes when establishing a new dynamic matching link.

[0055] Maintain the node ready status bitmap, set the corresponding status bit after receiving the confirmation signal, resend the operation command if the first confirmation timeout exceeds the preset timeout threshold, remove the node and activate the energy consumption type reclassification process if the second confirmation timeout occurs.

[0056] In summary, the present invention has the following beneficial technical effects:

[0057] 1. Achieve deep matching and optimization between hardware resources and task characteristics. By dynamically dividing nodes into four types with high energy consumption and establishing classification rules for task execution blocks, an adaptive scheduling framework is constructed at the node resource level. Cut-through nodes ensure deterministic response to real-time tasks, buffered nodes reduce latency of critical services using a preloading mechanism, relay nodes distribute peak energy consumption through sharding, and floating mode provides a safe backoff mechanism for overloaded nodes;

[0058] 2. Overcoming the energy consumption bottleneck of cross-node collaboration. An innovative dual energy consumption compensation model is established, based on the mutual exclusion of physical distance and energy consumption type, to predict the potential energy loss of node combination schemes in advance. A linearly increasing distance penalty mechanism is introduced for long-distance wireless transmission scenarios, simultaneously identifying energy consumption type conflicts to trigger fixed energy consumption compensation, and finally using composite compensation values ​​to constrain and select the optimal collaborative link;

[0059] 3. Establish a continuously evolving energy consumption management framework. Based on actual operational data, the energy consumption type determination thresholds are reverse-calibrated, key parameters are fine-tuned using a progressive strategy, and system stability is ensured through multi-dimensional simulation verification. The final established negative feedback optimization mechanism ensures that the matching degree between core thresholds and the real-time operating environment continuously improves over time. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings are used to provide a further understanding of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 A flowchart illustrating an embodiment of this application is disclosed.

[0062] Figure 2 Structural schematic diagrams of embodiments of this application are disclosed. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] The following is in conjunction with the appendix Figures 1-2 A preferred description of the present invention is provided below.

[0065] See attached document Figure 1 This invention proposes an edge computing optimization method based on the Internet of Things, comprising the following steps:

[0066] S1. Obtain real-time working status information of edge nodes, including chip power consumption rate, remaining power percentage and temperature data;

[0067] S2. Determine the energy consumption type of the node based on the real-time working status information, and output a set of node type tags, including direct-through, buffer, relay and floating types;

[0068] S3. The computational task to be processed is decomposed into independent execution unit blocks, which are divided into main execution blocks, secondary execution blocks and fragment execution blocks according to the urgency of the task.

[0069] S4. Predict the communication energy consumption compensation value of different node type tag combinations, set an additional energy consumption compensation value when the physical distance between nodes exceeds the physical distance threshold or the energy consumption types are mutually exclusive, and generate a list of communication energy consumption compensation values.

[0070] S5. Establish a dynamic matching link based on the node type tag set, independent execution unit block and communication energy consumption compensation value list, including assigning the main execution block to the direct type node, injecting the secondary execution block into the preload queue of the buffer type node, splitting the fragment execution block to multiple relay type nodes, and ensuring that the total communication energy consumption compensation value between all associated nodes is less than the predetermined limit.

[0071] S6. Send an operation command carrying execution block data to the node group corresponding to the dynamic matching link to trigger distributed computing;

[0072] S7. When any node reaches the overload critical point, trigger the re-division of energy consumption type, generate a set of reorganized execution blocks and activate the establishment of a new dynamic matching link.

[0073] In one embodiment of the present invention, step S1 includes the following steps:

[0074] Obtain real-time operating status information of edge nodes, including chip power consumption rate, remaining power percentage, and temperature data.

[0075] Specifically, the controller periodically collects operating parameters through pre-set status monitoring components on all edge nodes. These parameters include chip power loss values ​​measured synchronously by current and voltage metering components, residual capacity ratio readings of energy storage elements provided by the battery management chip, and quantified heat levels reported by the thermal sensors in the core area of ​​the distributed chip. The status monitoring components read these three types of raw data from the node hardware interface at set time intervals and transmit the verified raw data to the central scheduling system.

[0076] The minimum interval for periodically collecting operating parameters is no less than 1 second, and the maximum interval is no more than 60 seconds. The specific interval is automatically adjusted according to the urgency of the task being processed by the edge node: the minimum interval is used when the edge node is processing a task that requires real-time response, and the maximum interval is used when processing regular tasks.

[0077] The chip power consumption rate refers to the amount of electrical energy consumed by the main processing chip of the edge node per unit time. It is calculated by simultaneously measuring the chip's operating current and supply voltage, multiplying them to obtain the current power value, and then multiplying this current power value by the duration of the time interval to obtain the amount of electrical energy consumed per unit time, satisfying the following formula:

[0078]

[0079] Among them, E ΔT P(τ) represents the energy loss per unit time (chip energy consumption rate); P(τ) represents the instantaneous power function, obtained by multiplying the chip's operating current and supply voltage; ΔT represents the preset time interval, with a minimum interval of 1 second and a maximum interval of 60 seconds. The specific interval is automatically adjusted by the urgency of the current task being processed by the edge node: the minimum interval is used when the edge node is processing a task requiring real-time response, and the maximum interval is used when processing regular tasks; t represents the start time of data collection; τ is used in the formula to express the energy consumption accumulation process.

[0080] The remaining power percentage refers to the ratio of the total amount of electrical energy that the energy storage element provided by the battery management chip can currently release to the initial maximum amount of electrical energy that can be released. This value is obtained by calculating the operating voltage and discharge curve characteristics of the energy storage element provided by the edge node battery management chip. When the edge node is powered by an external DC power supply, the remaining power percentage is fixed at a constant value.

[0081] Temperature data specifically refers to the quantified value of the degree of heat generated by a thermistor on the surface of the core computing area of ​​the chip. The thermistor needs to be physically in contact with the chip's heat source. During measurement, the thermistor converts the physical temperature signal into a change in resistance value, which is then converted into a transmittable digital temperature reading through an analog-to-digital converter circuit.

[0082] All collected real-time operational status information must undergo verification before being transmitted to the central dispatch system:

[0083] 1. If the value of the same parameter fluctuates by more than 20% within five consecutive acquisition cycles (i.e., the range between the maximum and minimum values ​​exceeds 20%), discard the last acquired data for that node (the data from the fifth acquisition) and immediately start a new round of acquisition (restarting the five-cycle acquisition sequence). For example, in five consecutive acquisition cycles of chip power consumption rate, the chip power consumption rate values ​​are 100, 110, 95, 105, and 80 (unit: W). The maximum value is 110W and the minimum value is 80W. The range between the maximum and minimum values ​​is (110W - 80W) / 80W = 37.5%, exceeding 20%.

[0084] 2. If the fluctuation of the same parameter in three consecutive acquisitions is less than 5%, the latest acquisition value should be used for reporting. For example, if the chip power consumption rate is collected in three consecutive cycles and the values ​​are 100, 101, and 100.5 (unit: W), the maximum value is 101W and the minimum value is 100W. The fluctuation percentage is (101W-100W) / 100W = 1%, which is less than 5%. Therefore, the latest acquisition value of 100.5W should be used for reporting.

[0085] In one embodiment of the present invention, step S2 includes the following steps:

[0086] Based on real-time working status information, the power consumption type of edge nodes is determined, and edge nodes are dynamically divided into one of the following four categories: pass-through type, which is suitable for high-power computing tasks that cannot be interrupted; buffer type, which supports stable low-power tasks that support data preloading; relay type, which allows task sharding to reduce peak power consumption; and floating type, which automatically switches to relay type when the power is below the critical value.

[0087] Specifically, after receiving the real-time operating status information of the edge nodes, the controller determines the energy consumption type category of the edge nodes. The energy consumption type category refers to the classification result of the energy consumption behavior pattern exhibited by the edge nodes in the current operating state. The energy consumption behavior pattern classification result determines the way the edge nodes process tasks. The classification basis comes from the logical comparison of three elements—chip energy consumption rate, remaining power percentage, and temperature data—obtained in step S1 with preset thresholds. First, the classification threshold set in the configuration file is read. This classification threshold set contains four sets of judgment conditions:

[0088] 1. Pass-through type refers to a node type that is continuously in a high computing intensity state. It is suitable for real-time computing tasks that cannot be interrupted. The conditions for determining pass-through type must be met simultaneously: the chip power consumption rate is greater than the pass-through type power consumption threshold, the remaining power percentage is higher than the floating type power threshold, and the temperature data has not reached the overheat protection threshold.

[0089] The direct-access energy consumption threshold adopts the engineering safety margin principle, set based on the safety margin of the chip's peak workload, and is set at 70% of the chip's maximum theoretical energy consumption (with a 30% buffer to prevent instantaneous overload). The floating-type power threshold is a protection threshold determined by the battery's chemical characteristics, with a lower limit set according to the battery type (e.g., 20% of total power for lithium batteries). The overheat protection threshold follows the physical limits of semiconductors and the temperature safety limits specified by the chip manufacturer.

[0090] 2. Buffered type refers to a node type that has a data preloading hardware structure and whose chip power consumption rate is lower than the buffered type power consumption limit. The conditions for determining buffered type must be met simultaneously: the chip power consumption rate is lower than the buffered type power consumption limit, the remaining power percentage is higher than the floating type power threshold, and the node has built-in data caching hardware (such as an SSD module). This type of node reduces computation latency by preloading data.

[0091] Among them, the upper limit of buffered energy consumption is the average of the node's historical lowest and highest energy consumption.

[0092] 3. Relay type specifically refers to a node type suitable for supporting task fragmentation. The criteria for relay type must be met simultaneously: the remaining battery percentage is higher than the floating battery threshold, the chip has a multi-core parallel computing architecture, and it does not meet the criteria for pass-through or buffer type. This type of node reduces the peak power consumption of a single node by splitting tasks.

[0093] 4. Floating type refers to the forced conversion type when the remaining power percentage is lower than the floating power threshold. Regardless of the original type of the node, it will switch to this type, and the behavior mode will automatically inherit the task processing method of the relay type node.

[0094] The following operations are performed sequentially during the determination process: First, check whether the remaining battery percentage triggers the floating switching condition. If the remaining battery percentage does not trigger the floating switching condition, compare the chip power consumption rate with the pass-through power consumption threshold and then check the hardware function support characteristics. Finally, output a unique power consumption type category label. If there is a data verification anomaly in the real-time working status information (chip power consumption rate, remaining battery percentage, temperature data) required for a single determination, the verification operation performed in step S1 is automatically invoked to re-collect data. Floating conditions are judged first, then pass-through conditions and buffer conditions are filtered level by level, and the rest are classified as relay conditions by default, ensuring that each node has one and only one power consumption type category.

[0095] In one embodiment of the present invention, step S3 includes the following steps:

[0096] The computational tasks to be processed are decomposed into independent execution unit blocks. Each independent execution unit block refers to the smallest task fragment that can be scheduled and executed independently, containing complete computational logic and data encapsulation. The decomposition is based on parallelization comments in the task's source code or a runtime dependency graph, ensuring that the decomposed blocks can run independently. Based on task urgency, they are categorized as follows: main execution blocks requiring real-time output, secondary execution blocks allowing for millisecond-level delays, and fragmented execution blocks without explicit time constraints.

[0097] Specifically, after receiving the computation task to be processed, the gateway device parses the priority identifier in the task packet header and breaks the task down into several independent execution unit blocks according to the response speed requirements defined by the priority identifier. The breakdown process adopts a two-level partitioning rule:

[0098] First, based on the continuity of task functions, it is determined whether there are separable computational sub-units. The continuity of task functions refers to analyzing the data dependencies between each step within the task. By traversing the data input and output tables of computational instructions, it is determined whether there is a sequence of instructions that must be executed continuously. If two instruction steps share unpersisted intermediate variables, they are determined to be inseparable.

[0099] Then, the calculation sub-units are classified into different timing levels according to the response delay allowance value. The response delay allowance value is stored in the encoded value of the third bit of the task packet header. After decoding, it corresponds to three types of values ​​(i.e., 00 = real-time, 01 = non-real-time, 10 = background): the real-time level represents the millisecond-level response requirement, the non-real-time level represents the allowable delay of hundreds of milliseconds, and the background level represents no time limit requirement. This encoding is written by the business system that publishes the task when creating the task.

[0100] After partitioning, three types of independent execution unit blocks are output:

[0101] The task sub-unit marked as the highest priority forms the main execution block. The main execution block is the task unit that must be processed immediately and return the result in real time. It must be processed within one computing cycle. The judgment criteria are that the value of the response delay requirement field in the task packet header is equal to the real-time level, or the functional continuity test shows that the interruption of this module will cause the entire task to fail.

[0102] Task sub-units that allow limited delays form sub-execution blocks. Sub-execution blocks are task units that allow limited processing delays. The execution of such blocks can wait for data preloading to complete. The criteria for judgment are that the response delay requirement field value in the task packet header is not real-time but less than the threshold of hundreds of milliseconds, and the functional test shows that the delay processing does not affect the correctness of the final result.

[0103] Background tasks without a fixed completion time form fragmented execution blocks. These fragmented execution blocks are task units with no hard constraints on their processing cycle. They are automatically classified into this category when the response latency requirement field in the task header is set to the background level, or when functional continuity detection indicates that the task is a non-critical statistical operation. These blocks can be arbitrarily split into smaller computational fragments.

[0104] The decomposition process requires ensuring that each independent execution unit block has complete data input and output boundaries, and that a synchronization tag is automatically added if there is data correlation between blocks. The synchronization tag is metadata added to ensure data consistency. When two independent execution unit blocks have a sequential processing order, corresponding matching tags are added to the output of the preceding block and the input of the subsequent block. The synchronization tag consists of a triplet data structure composed of a 64-bit unique hash identifier of the preceding block, an MD5 fingerprint digest of the output data, and a timestamp validity window (start time to end time).

[0105] When an intermediate variable is detected to be shared by two execution unit blocks, the triple is automatically injected into the output interface metadata area of ​​the preceding block (marked with the _Send suffix) and the input interface metadata area of ​​the subsequent block (marked with the _Recv suffix). When the computational task to be processed contains multiple features, the final partitioning scheme is determined according to the principle of prioritizing the most urgent feature. The principle of prioritizing the most urgent feature is only activated when the priority identifier conflicts with the functional continuity detection result. A feature urgency value table is defined based on historical statistical data.

[0106] Security protection features (biometrics / anomaly detection) are assigned an urgency score of 90 points; real-time control features (closed-loop regulation / emergency braking) are assigned 80 points; critical business features (payment verification / authorization verification) are assigned 70 points; and data analysis features (statistical aggregation / log processing) are assigned 50 points. During task decomposition, feature scanning is performed, and the computational unit corresponding to the feature with the highest urgency is forcibly assigned to the main execution block. When multiple features have the same urgency score, the feature unit with the larger data volume is prioritized. The minimum computational load of the smallest independent execution unit block is a thousand times the computational load of a single chip clock cycle to avoid problems with overly fine-grained scheduling.

[0107] If the functional continuity check fails, the entire task is treated as the main execution block, and after disassembly, the integrity verification tool is run to check the closure of the data input and output interfaces of the independent execution unit blocks. The integrity verification tool performs four layers of verification: Data dimension verification checks the matching degree of the number of input and output parameters, which must be greater than or equal to 95%; Type dimension verification checks the strict consistency of the data structure (requiring complete byte-level matching); Temporal dimension verification checks the continuity of processing time slots, i.e., analyzing the continuity of the start and end timestamps of the five most recent processing tasks (allowing a maximum deviation of 5 milliseconds); Resource dimension scanning memory access boundaries (marking out-of-bounds address offsets), i.e., scanning whether memory pointers exceed the preset safe address space boundaries. Any anomaly detected in any dimension generates a dimension identifier code (a binary mask marking the location of the anomaly dimension). If all four dimensions pass, a closure verification flag is output.

[0108] The preset safe address space boundary is based on the historical memory usage peak of the execution block type, and the maximum value in the last 100 runs is taken as the reference baseline. For example, the extension boundary of the same type of main execution block is 0x3BFFFFFF (rounded to memory page alignment).

[0109] In one embodiment of the present invention, step S4 includes the following steps:

[0110] Predict the communication energy consumption compensation value of different energy consumption type node combinations. The energy consumption type node combination is the direct type, buffer type, relay type or floating type output in step S2. Set an additional energy consumption compensation value when the physical distance between nodes exceeds the physical distance threshold or the energy consumption types are mutually exclusive.

[0111] Specifically, based on the node energy consumption type determined in step S2 and the coordinate location table of deployed nodes, the scheduling controller executes a node location mapping process during the system initialization phase. First, it broadcasts a positioning initialization command to all edge nodes. Upon receiving the command, each node activates its built-in location sensing component. The location sensing component refers to the physical location detection device built into the node, including but not limited to GPS receivers, Bluetooth beacons, and UWB positioning chips. Mobile nodes must be equipped with a GPS module, and fixed nodes must support at least one wireless positioning technology. Mobile nodes with GPS modules (such as vehicle-mounted devices) use satellite positioning data, while fixed deployment nodes (such as street light gateways) use base station triangulation to calculate coordinates.

[0112] Base station triangulation is based on the principle that the propagation speed of radio frequency signals is constant. It measures the transmission delay difference between different base stations and establishes a hyperbolic equation system to solve for the spatial coordinates of the nodes. It requires the coordinated operation of three or more base stations and satisfies the following formula:

[0113]

[0114] Where (x, y) represents the coordinates of the mobile node; (x0, y0) represents the coordinates of the reference base station; (x1, y1) represents the coordinates of base station 1; (x2, y2) represents the coordinates of base station 2; and c represents the electromagnetic wave propagation speed, a fixed value of 3 × 10⁻⁶. 8 m / s; Δt1 represents the time difference between the signal from base station 1 and the reference base station; Δt2 represents the time difference between the signal from base station 2 and the reference base station. The coordinate data returned by all nodes is encrypted to form a table of the coordinate locations of the deployed nodes.

[0115] The deployed node coordinate location table update mechanism is divided into two categories: fixed nodes update their coordinates completely every 24 hours, while mobile nodes report new coordinates immediately when a displacement event is triggered. A displacement event is triggered when the accelerometer of a mobile node detects continuous movement for more than 5 seconds or a distance traveled exceeding 5 meters, automatically activating the position sensing component to collect new coordinates. Each entry in the deployed node coordinate location table stores the node identifier code, three-dimensional coordinate values ​​(longitude / latitude / altitude), positioning method marker (GPS / triangulation), and estimated positioning accuracy. The estimated positioning accuracy is automatically assigned based on the positioning method (GPS positioning = 2 meters; triangulation = 5 meters; indoor beacon positioning = 0.5 meters).

[0116] For each independent execution unit block generated in step S3, a cross-node collaborative energy consumption assessment is performed on the possible node combination schemes to be assigned. The node combination scheme refers to the set of collaborative nodes selected from the available edge nodes to complete the processing task of a specific independent execution unit block. The setting of the node combination scheme must simultaneously meet the following three elements: collaboration necessity means that it includes at least two different types of nodes with different energy consumption (such as pass-through type + buffer type) to achieve task collaboration; spatial proximity means that all nodes are in the same communication subnet to ensure low latency interaction; scheduling unit correlation means that the number of nodes in the combination must match the computing needs of the execution block (such as the main execution block must contain ≥1 pass-through type node, and the fragment execution block must be allocated ≥3 relay type nodes).

[0117] Implement cross-node collaborative energy consumption assessment:

[0118] 1. The system measures the straight-line physical distance between nodes in a power-consuming node combination. When the distance between any two nodes exceeds a set physical distance threshold, a distance penalty mechanism is activated. The physical distance threshold is a pre-set critical value, determined through on-site signal attenuation testing: the target scenario selects the furthest stable communication distance, deducts a safety redundancy margin to form the final threshold. The safety redundancy margin is typically 10%–30% of the furthest stable communication distance, with the specific percentage dynamically adjusted based on on-site risk conditions. After deployment, the physical distance threshold for wired connection nodes is set to a constant zero value (physical line connections are not affected by wireless distance attenuation).

[0119] The distance penalty mechanism includes: subtracting a preset physical distance threshold from the actual physical distance measurement between two nodes to obtain the excess distance value; multiplying the excess distance value by a distance penalty coefficient to obtain the distance penalty component. The distance penalty coefficient is a positive real number, and its actual value is obtained by averaging historical communication energy consumption test data, typically greater than 0. The coefficient is set based on statistical analysis of historical energy consumption test data, reflecting the actual impact cost of unit excess distance on system energy consumption.

[0120] 2. Verify the compatibility of node energy consumption types. For mutually exclusive combinations with conflicting energy consumption characteristics, activate the mutual exclusion compensation mechanism. Type mutual exclusion specifically refers to antagonistic combinations where certain energy consumption types inevitably incur additional overhead when nodes cooperate, including but not limited to: a direct-connection node and another direct-connection node forming a resource contention mutual exclusion; a direct-connection node and a buffer node combination having no mutual exclusion; and a floating node and a buffer node combination forming a compatibility mutual exclusion due to unstable power supply. Establish a type mutual exclusion relationship table, for example:

[0121]

[0122] The type-based mutual exclusion table supports laboratory conflict reproduction testing and regression analysis of field fault logs to verify novel mutual exclusion relationships. Incremental rule patches are released after review. Querying the type-based mutual exclusion table confirms whether two nodes belong to a preset mutually exclusive combination. If so, the corresponding fixed penalty value is directly loaded as the mutual exclusion penalty component, which is the mutual exclusion compensation mechanism. The fixed penalty value comes from the average measured data of typical mutually exclusive scenarios in the same environment, based on Standard Energy Equivalent Percentage (BECE) measurement. An example is:

[0123] Straight-through + Straight-through 12.7% BECE 200 laboratory channel contention tests + 17 field overload logs Floating type + buffer type 8.2% BECE 150 charge-discharge cycle tests + 23 high-altitude field failures Floating type + floating type 15.3% BECE 90 cascaded ripple tests + 9 offshore wind power failures Straight-through + Buffered 0% (No mutual exclusion) Compatibility verification test (without triggering compensation) Buffer type + buffer type 0% (No mutual exclusion) Asynchronous scheduling stress test (without triggering compensation)

[0124] The distance penalty component and the mutual exclusion penalty component are added together to generate the communication energy consumption compensation value. This value represents the estimated additional energy loss due to cross-node collaboration and is composed of two independent components: a mutual exclusion penalty component representing fixed losses due to incompatible node energy consumption types, and a distance penalty component representing losses that linearly increase with the physical distance exceeding a threshold. The final output is a list of communication energy consumption compensation values ​​for each candidate combination. This list clearly identifies node pairs exceeding the physical distance threshold and node pairs with mutually exclusive types. The output format of the communication energy consumption compensation value list includes: combination number, node pair identifier, excess distance value, mutual exclusion type name, and estimated additional energy loss.

[0125] In one embodiment of the present invention, step S5 includes the following steps:

[0126] Dynamically establish matching links between execution blocks and nodes, including injecting secondary execution blocks into the preload queue of buffered nodes, assigning main execution blocks to direct-access nodes, and splitting fragmented execution blocks to multiple relay nodes, ensuring that the total communication energy consumption compensation value between all associated nodes is less than a predetermined limit.

[0127] Specifically, based on the node energy consumption type category determined in step S2 and the communication energy consumption compensation value list generated in step S4, a dynamic matching link between independent execution unit blocks and nodes is established. The dynamic matching link refers to the set of execution paths that independent execution unit blocks are scheduled to a designated node group. The execution path carries both data flow and control command flow. The validity of the dynamic matching link must simultaneously meet the node type adaptability and communication energy consumption constraints.

[0128] First, the main execution block filters the set of available pass-through nodes and forces them to be assigned to the lightest-loaded pass-through node. Pass-through nodes strictly correspond to the type criteria defined in step S2. The filtering algorithm is as follows: iterate through the real-time working status information of all pass-through nodes and select the node with the lightest load. For example, if multiple nodes have the same resource level, prioritize the node with the smallest temperature data fluctuation in the last five minutes. The main execution block can only be assigned to pass-through nodes; if no pass-through node is currently available, a node type change must be triggered.

[0129] Next, an idle buffered node is matched for the secondary execution block, and its data is injected into the preload queue of the buffered node. The preload queue of the buffered node is a data buffer area specifically set up for the secondary execution block. The preload queue of the buffered node adopts a first-in, first-out management mechanism. The injection process requires two operations: first, the original data of the secondary execution block is stored at the end of the preload queue, and then a pointer to the data to be processed is sent to the node's computing core. If the buffered node fails, the secondary execution block can be degraded to a pass-through node.

[0130] The fragmented execution blocks are then broken down into microtask units according to a sharding algorithm. The splitting of fragmented execution blocks in relay nodes is implemented by a sharding controller and distributed to multiple relay nodes. The sharding rule implemented by the sharding controller is as follows: the input data of the fragmented execution block is evenly divided into several data segments, each data segment carrying an identical copy of the computation instructions. The number of data segments is limited by the real-time load of the nodes: actual number of shards = number of available relay nodes × load adjustment coefficient.

[0131] It can be seen that the load adjustment coefficient is set based on the average real-time load rate of the cluster (range: 0.1 to 0.9). A typical value of 0.75 corresponds to a steady-state scenario where the average real-time load rate of the cluster is ≤70%, satisfying the following formula:

[0132]

[0133] Among them, L avg L represents the average real-time load rate of the cluster. max The maximum safe load threshold set for the system (typically 85%), k adj This is the load adjustment coefficient.

[0134] Each operation requires verification of the communication relationships between associated nodes: Step S4 extracts the sum of communication energy consumption compensation values ​​for the current node combination. The sum of communication energy consumption compensation values ​​is the result of accumulating the compensation values ​​of all node pairs within the current combination. This accumulation operation includes the communication energy consumption compensation values ​​generated by any two nodes within the combination that have a cooperative relationship. When the sum of communication energy consumption compensation values ​​exceeds a predetermined limit, the node pair with the largest communication energy consumption compensation value is removed and replaced with a candidate node. The matching result forms a dynamic matching link topology diagram containing the complete flow of three types of independent execution unit blocks, along with a list of communication energy consumption compensation values ​​for all node pairs.

[0135] The predetermined limit is based on a constraint threshold set according to historical best energy consumption. The final limit is formed by multiplying the basic communication energy consumption record when the edge system is idle by a compensation tolerance coefficient. The limit is recalibrated every 24 hours during system operation. The compensation tolerance coefficient represents the system's preset dynamic scaling factor, with a value range of 1.15 to 1.35. A typical value of 1.25 corresponds to the upper limit of historical energy consumption fluctuation in the 95% confidence interval.

[0136] In one embodiment of the present invention, step S6 includes the following steps:

[0137] Send operation instructions carrying execution block data to the matching node group to start distributed computing.

[0138] Specifically, based on the dynamically matched link topology generated in step S5, the scheduling controller sends operation instruction data packets to the relevant node cluster. The operation instruction data packet refers to the smallest control unit for interaction between the scheduling controller and the edge nodes, containing three structured parts: a header identifier segment storing the node's physical address code specified by the matched link in step S5 and its category of independent execution block (main execution block / secondary execution block / fragment execution block); a data payload segment encapsulating the complete original data of the independent execution unit block (including input parameters and computational instruction sets), using a node authentication key for lightweight data encryption, employing the AES-128-CTR encryption algorithm, with the preset key changing every 24 hours; and a collaboration information appendix loading communication topology relationship data, listing the addresses of partner nodes that the current node needs to interact with and their corresponding communication energy consumption compensation values. The maximum data size of the collaboration information appendix is ​​128 bytes; if the data size exceeds this limit, paging is used for transmission.

[0139] Among them, the communication energy consumption compensation value list of the node combination generated in step S4 is inherited from the communication topology relationship data, and only the information of partner nodes that have a cooperative relationship with the target node is selected.

[0140] During transmission, the controller implements a type-based transmission strategy with three implementation methods: the direct transmission mode, which uses the highest priority communication channel for immediate transmission of direct node groups, occupies a dedicated real-time communication channel, and the sender marks an emergency transmission flag in the header of the instruction packet.

[0141] The buffered transmission mode, when facing a group of buffered nodes, includes a preloaded queue offset parameter (the offset of the buffer's starting address) to ensure that data is correctly injected into the specified position in the queue. The instruction packets sent to relay nodes embed the fragmentation calculation coordination code.

[0142] In the relay-type transmission mode, a global readiness check is triggered at the controller after all command packets have been transmitted. Only when all target nodes return a reception confirmation signal is the distributed computing start command synchronously issued, ensuring that all nodes start executing computing tasks within the same time reference window. The distributed computing start command is a unified control signal that enables edge nodes to execute operation instructions. The distributed computing start command carries a hardware-level synchronization timestamp. After receiving the distributed computing start command, all nodes deploy a high-precision time synchronization protocol (such as IEEE 1588v2) and calibrate their own clock counters according to the timestamps of the command packets.

[0143] Transmission timing control: Straight-through command packets are sent first, followed by buffered command packets, and finally relay command packets are transmitted.

[0144] Global Readiness Status Verification Process: The controller maintains a bitmap of node readiness status. Each time a receive confirmation signal is received from a node, the corresponding bit in the bitmap is set. When all target node status bits are set, the final verification pass event is triggered. The process of receiving a receive confirmation signal from a node sets a timeout threshold. The timeout threshold represents the round-trip time of the signal from the farthest node, which is analyzed in conjunction with historical statistical data.

[0145] In one embodiment of the present invention, step S7 includes the following steps:

[0146] When any node reaches the overload critical point, the energy consumption type is reclassified: the node is marked as floating, the execution block processed by the node is re-disassembled, and step S5 is activated to establish a new dynamic matching link.

[0147] Specifically, the system continuously monitors the real-time operating status of edge nodes. When the chip power consumption rate of any node exceeds the overload threshold for three consecutive acquisition cycles, the overload threshold refers to the maximum load limit for safe chip operation. The maximum load limit is set according to the rated operating parameters published by the chip manufacturer: for chips with intelligent frequency modulation function, the theoretical maximum power consumption value under full load operation at the base frequency is taken; for fixed frequency chips, 85% of the thermal design power rating is taken.

[0148] Trigger the adaptive rescheduling process:

[0149] 1. The controller marks the edge node as floating and performs three mandatory measures: clears the node's current energy consumption type category cache, forces a refresh of the node's registered type identifier, and broadcasts a type change notification to associated nodes. It fully inherits the definition and behavioral constraints of floating type from step S2, i.e., automatically switches to relay-type task processing mode.

[0150] 2. Freeze all independent execution unit blocks processed by the edge node and revert related operation instruction data packets. Operation instruction reversion must be sent according to the communication specifications of step S6. For direct-access nodes, the emergency channel is used for priority transmission; for buffered nodes, the corresponding entry in the preload queue is cleared; and for relay nodes, unfinished data fragmentation is terminated. The frozen execution blocks to be reassembled are resubmitted to step S3 for secondary decomposition, generating new sets of main execution blocks, secondary execution blocks, or fragmented execution blocks. Each reversion operation requires a confirmation receipt from the node.

[0151] Among them, the execution block to be reassembled specifically refers to the computation task segment that was interrupted during the execution of the original node, which includes the incomplete computation state and intermediate data; the execution block when frozen encapsulates the state snapshot data packet, which includes the current instruction pointer position, the temporary value of the register, and the summary of the processed data fragment.

[0152] Secondary decomposition requires inheriting the basic attributes of the original execution block: if the original is a primary execution block, at least one primary execution block must be retained after the new decomposition; the original secondary execution block can be split into multiple secondary execution blocks or fragmented execution blocks; the original fragmented execution block is only allowed to be split into more fragmented execution blocks, and other blocks can be downgraded but not upgraded.

[0153] 3. Activate the dynamic matching link establishment mechanism in step S5. When calling the interface in step S5, inject the reorganized execution block set. The matching process automatically skips faulty nodes marked as floating, ensuring that the reorganized execution block establishes a communication link that includes other available nodes. The rescheduling process ensures that the floating node stops receiving new tasks until its subsequent working state returns to normal.

[0154] Step S6 global ready state verification process, exception handling rules:

[0155] If the time taken to receive a receive confirmation signal from a node for the first time exceeds the set timeout threshold, the operation command data packet is retransmitted. If the time taken to receive a receive confirmation signal from a node continues to exceed the set timeout threshold, the dynamic matching link removes the edge node and triggers the energy consumption type reclassification in step S7.

[0156] See appendix Figure 2 The present invention also proposes an edge computing optimization system based on the Internet of Things, comprising the following modules:

[0157] The status monitoring module is used to acquire real-time working status information of edge nodes, including chip power consumption rate, remaining power percentage and temperature data;

[0158] The type classification module determines the energy consumption type of nodes based on real-time working status information and outputs a set of node type tags, including direct-through, buffer, relay, and floating types.

[0159] The task decomposition module is used to decompose the computing tasks to be processed into independent execution unit blocks, and divide them into main execution blocks, secondary execution blocks and fragment execution blocks according to the urgency of the tasks.

[0160] The energy consumption prediction module predicts the communication energy consumption compensation value of different node type tag combinations, sets the additional energy consumption compensation value when the physical distance between nodes exceeds the physical distance threshold, or when the energy consumption types are mutually exclusive, and generates a list of communication energy consumption compensation values.

[0161] The dynamic matching module establishes a dynamic matching link based on the node type tag set, independent execution unit blocks, and communication energy consumption compensation value list. This includes assigning the main execution block to a direct-access node, injecting the secondary execution block into the pre-loading queue of the buffer node, splitting the fragmented execution block into multiple relay nodes, and ensuring that the total communication energy consumption compensation value between all associated nodes is less than a predetermined limit.

[0162] The instruction scheduling module is used to send operation instructions carrying execution block data to the node group corresponding to the dynamically matched link to trigger distributed computing;

[0163] The exception handling module is used to detect when any node reaches the overload critical point, trigger the re-division of energy consumption type, generate a set of reorganized execution blocks, and activate the establishment of a new dynamic matching link.

[0164] It should be noted that the formulas described above, through the principle of dimensional consistency and mathematical standardization methods (such as normalization, dimensionless parameter conversion, or unit system unification), can translate physical quantities with different properties into unitless standard values ​​or superimposed parameters of the same dimension. This eliminates the interference of different dimensions on the computational logic, allowing the formulas to retain the original data distribution characteristics while possessing mathematical rationality and adaptability to objective laws. The descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the invention.

[0165] Each of the modules can be implemented in whole or in part through software, hardware, or a combination thereof. It supports hardware embedded in or independent of the processor in the computer device, and also supports software stored in the memory of the computer device, so that the processor can call and execute the operations corresponding to each of the above modules.

[0166] It should be noted that the human information (including but not limited to human device information and personal information) and data (including but not limited to data used for analysis, data stored and data displayed) involved in this invention are all information and data authorized by the human body or fully authorized by all parties. The collection, use and processing of related data require relevant legal standards.

[0167] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An edge computing optimization method based on the Internet of Things, characterized in that, Includes the following steps: S1. Obtain real-time working status information of edge nodes, including chip power consumption rate, remaining power percentage and temperature data; S2. Determine the energy consumption type of the node based on the real-time working status information, and output a set of node type tags, including direct-through, buffer, relay and floating types; S3. The computational task to be processed is decomposed into independent execution unit blocks, which are divided into main execution blocks, secondary execution blocks and fragment execution blocks according to the urgency of the task. S4. Predict the communication energy consumption compensation value of different node type tag combinations, set an additional energy consumption compensation value when the physical distance between nodes exceeds the physical distance threshold or the energy consumption types are mutually exclusive, and generate a list of communication energy consumption compensation values. S5. Establish a dynamic matching link based on the node type tag set, independent execution unit block and communication energy consumption compensation value list, including assigning the main execution block to the direct type node, injecting the secondary execution block into the preload queue of the buffer type node, splitting the fragment execution block to multiple relay type nodes, accumulating the communication energy consumption compensation value of the associated node pair to obtain the total compensation value, and replacing the node pair with the largest compensation value when the total compensation value exceeds the preset limit, generating a dynamic matching link topology map so that the total communication energy consumption compensation value between all associated nodes is less than the predetermined limit. S6. Send an operation command carrying execution block data to the node group corresponding to the dynamic matching link to trigger distributed computing; S7. When any node reaches the overload critical point, the energy consumption type is reclassified, the node is marked as floating, the execution blocks processed by the node are re-disassembled, a set of recombined execution blocks is generated, and a new dynamic matching link is activated.

2. The edge computing optimization method based on the Internet of Things according to claim 1, characterized in that, Obtaining real-time operational status information of edge nodes includes the following steps: The chip's energy consumption rate is obtained by periodically collecting the chip's power loss value through the preset status monitoring component; the remaining capacity ratio reading of the energy storage element provided by the battery management chip is the remaining power percentage; and the quantified value of the heat generation reported by the thermal sensing element distributed in the core area of ​​the chip is the temperature data. Perform fluctuation verification on parameter values ​​within a continuous acquisition period: if the fluctuation of the same parameter in five consecutive acquisitions exceeds the preset fluctuation threshold, discard the last data and re-acquire; if the fluctuation of the last data in three consecutive acquisitions is less than the preset stability threshold, use the last data. The verified chip power consumption rate, remaining power percentage, and temperature data are transmitted to the central scheduling system.

3. The edge computing optimization method based on the Internet of Things according to claim 1, characterized in that, Determining the type of node energy consumption includes the following steps: The logic compares the three elements of the acquired chip power consumption rate, remaining power percentage, and temperature data with the preset thresholds, and reads the classification threshold set in the configuration file. The classification threshold set includes the direct-type power consumption threshold, the floating power threshold, and the overheat protection threshold. First, check whether the remaining power percentage triggers the floating condition. If the floating condition is not triggered, sequentially filter through the through nodes that simultaneously meet the following conditions: chip power consumption rate is greater than the through power consumption threshold, remaining power percentage is higher than the floating power threshold, and temperature data has not reached the overheat protection threshold. The next screening selects buffer nodes that simultaneously meet the following criteria: chip power consumption rate is lower than the buffered type power consumption limit, remaining power percentage is higher than the floating type power threshold, and have data caching hardware. The rest are classified as relay-type conditions by default, ensuring that each node has one and only one type of energy consumption.

4. The edge computing optimization method based on the Internet of Things according to claim 3, characterized in that, Node energy consumption type categories include: Straight-through type refers to a node type that is continuously in a state of high computational intensity and is suitable for real-time computing tasks that cannot be interrupted. Buffered type refers to a node type that has a data preloading hardware structure and whose chip power consumption rate is lower than the buffered type power consumption limit, reducing computation latency by preloading data; Relay type, specifically refers to a node type suitable for supporting task sharding processing, which reduces the peak energy consumption of a single node by splitting tasks; Floating type refers to the forced conversion type when the remaining power percentage is lower than the floating power threshold. Regardless of the original type of the node, it will switch to this type, and its behavior mode will automatically inherit the task processing method of the relay type node.

5. The edge computing optimization method based on the Internet of Things according to claim 1, characterized in that, The computational task to be processed is decomposed into independent execution unit blocks, including the following steps: Based on the continuity of task functions, detect whether there are separable computational sub-units; By traversing the data input / output table of the calculation instructions, it is detected whether there is a sequence of instructions that must be executed consecutively. The calculation sub-units are classified into different timing levels according to the allowable response delay value, namely 00=real-time, 01=non-real-time, and 10=background; the real-time level indicates millisecond-level response requirements, the non-real-time level represents the allowable delay of hundreds of milliseconds, and the background level indicates no time limit requirement. After the division is completed, three types of independent execution unit blocks are output.

6. The edge computing optimization method based on the Internet of Things according to claim 5, characterized in that, Three independent execution unit blocks, including the following steps: The task sub-unit marked as the highest priority forms the main execution block. The determination is based on the task packet header response delay requirement field value being equal to the real-time level, or the functional continuity test showing that the interruption of the task sub-unit will cause the entire task to fail. The decision to allow task sub-units with limited delay to form secondary execution blocks is based on the task packet header response delay requirement field value being non-real-time level, and functional testing showing that delay processing does not affect the correctness of the final result. Background tasks without a fixed completion time form fragmented execution blocks. The criteria for this are: the response delay requirement field in the task package header is set to the background level, or the function continuity test shows that it is a non-critical statistical operation and is automatically classified into this category.

7. The edge computing optimization method based on the Internet of Things according to claim 1, characterized in that, Generate a list of communication power consumption compensation values, including the following steps: Obtain the node energy consumption type category and the coordinate location table of deployed nodes. The node coordinate location table is collected by a preset location sensing component. When the physical distance between any two nodes in a node combination exceeds a preset distance threshold, a distance penalty mechanism is activated: the distance penalty component is generated by multiplying the excess distance value by a preset distance penalty coefficient. Query the preset type mutual exclusion relationship table, and when there is mutual exclusion in the combination of node types, load a fixed penalty value to generate a mutual exclusion penalty component; Accumulate the distance penalty component and the mutual exclusion penalty component to generate a communication energy consumption compensation value, and output a list of compensation values.

8. The edge computing optimization method based on the Internet of Things according to claim 1, characterized in that, Establishing a dynamic matching link includes the following steps: The main execution block is assigned to the lightest load pass-through node, which is determined by comparing temperature data fluctuations in real-time operating status information; The execution block is injected into the preset preload queue of the buffered node, and the preload queue adopts a first-in-first-out management mechanism; The fragmented execution block is broken down into micro-task units according to the sharding algorithm and distributed to multiple relay nodes. The number of shards is determined by multiplying the number of available nodes by a preset load adjustment coefficient. The total compensation value is obtained by accumulating the communication energy consumption compensation values ​​of the associated node pairs. When the total compensation value is exceeded, the node pair with the largest compensation value is replaced, and a dynamic matching link topology map is generated.

9. The edge computing optimization method based on the Internet of Things according to claim 1, characterized in that, Sending operation instructions carrying execution block data includes the following steps: The structure consists of three parts: the header identifier segment stores the node's physical address code and the category of its independent execution block; the payload segment encapsulates the complete original data of the independent execution unit block; and the collaboration information appendix loads the communication topology relationship data, listing the addresses of the partner nodes that the current node needs to interact with and the corresponding communication energy consumption compensation values. Transmit command packets in the order of direct-access nodes, buffer nodes, and relay nodes; After all nodes return a confirmation signal, a distributed computing start command carrying a synchronization timestamp is issued.

10. The edge computing optimization method based on the Internet of Things according to claim 1, characterized in that, Triggering a reclassification of energy consumption types includes the following steps: When the chip's power consumption rate continuously exceeds the preset overload threshold, the node will be marked as floating and a type change notification will be broadcast. Freeze the independent execution unit blocks processed by the node and roll back the operation instructions, and encapsulate the incomplete computation state to generate execution blocks to be reassembled; The execution block to be reorganized is submitted to the task decomposition step for secondary decomposition, generating a set of reorganized execution blocks, and skipping floating nodes when establishing a new dynamic matching link. Maintain the node ready status bitmap, set the corresponding status bit after receiving the confirmation signal, resend the operation command if the first confirmation timeout exceeds the preset timeout threshold, remove the node and activate the energy consumption type reclassification process if the second confirmation timeout occurs.

Images