Communication load adaptive scheduling method for edge nodes of intelligent lamp pole

Abstract

The embodiment of the invention provides a communication load adaptive scheduling method for edge nodes of an intelligent lamp post, which is applied to the technical field of edge computing and Internet of Things communication, and comprises the following steps: collecting electrical, network, queue and environment multi-dimensional data of nodes, and performing filtering, alignment and fusion processing to form a unified node operation state data set; calculating an instant messaging load through a multi-factor fusion model, and generating a load state mapping table by using a dynamic threshold value; the task of the high-load node is intelligently decomposed according to the mapping table, and a two-factor matching decision is carried out in combination with a real-time link state and a node resource, so that accurate migration from a task segment to a low-load node is realized; real-time monitoring and dynamic adjustment are carried out in the migration process; and based on the load report after each round of scheduling, self-adaptively feeding back and optimizing weight parameters of the load calculation model. According to the invention, accurate evaluation and prospective early warning of the node state are realized, efficient and reliable refined task scheduling is completed, and a closed-loop self-optimization intelligent system is formed.

Description

Technical Field

[0001] This application relates to the fields of edge computing and Internet of Things communication technology, and in particular to an adaptive scheduling method for communication load of edge nodes of smart light poles. Background Technology

[0002] In edge computing networks comprised of smart streetlights, node overload is a complex, multi-dimensionally coupled state that cannot be characterized by a single metric. Traditional scheduling methods typically rely on isolated metrics such as CPU utilization, memory usage, or network queue length for judgment. However, in harsh outdoor environments, node status is affected by multiple factors, including power supply fluctuations, ambient temperature and humidity, and wireless channel quality. For example, voltage dips may cause computing units to reduce their frequency, leading to processing queue backlogs, while high temperatures may exacerbate chip thermal noise, increasing the communication error rate.

[0003] Existing technologies lack the ability to synchronously collect, correlate, analyze, and model these cross-domain parameters, making it impossible to proactively capture performance bottlenecks or early signs of failures caused by coupling effects. This leads to distorted load assessments and creates hidden dangers for subsequent scheduling decisions.

[0004] Existing load scheduling systems often use statically configured or limitedly experience-based parameters such as load thresholds and evaluation weights, lacking self-learning and adaptive capabilities. Smart light pole networks face highly dynamic business loads, such as the ebb and flow of security video traffic during morning and evening peak hours and time-varying external environments. Static strategies may misjudge too many nodes when the overall load is high or react slowly during business downturns, failing to achieve optimal global resource allocation. As equipment ages and business models evolve, fixed strategies will quickly become ineffective. Summary of the Invention

[0005] The embodiments of this application provide an adaptive scheduling method for communication load of edge nodes of smart light poles, which fundamentally improves the accuracy of load balancing, maximizes resource utilization, and minimizes operation and maintenance costs in complex outdoor environments. To achieve the above objectives, this application adopts the following technical solution:

[0006] An adaptive scheduling method for communication load of edge nodes of smart light poles, the method comprising:

[0007] In the smart light pole network, the current, voltage, processing queue length, communication latency, packet loss rate, and ambient light, wind speed, and temperature of each edge node are acquired to form a node operation status dataset.

[0008] Based on the node running status dataset, the real-time communication load of each edge node is calculated, and the load value of each node is associated with an identifier indicating whether it exceeds a preset threshold and stored to form a load status mapping table.

[0009] Based on the load status mapping table, nodes marked as exceeding the threshold are selected to form a high-load node set, and the communication tasks of the high-load nodes are decomposed into transferable data fragments to generate a scheduling candidate task list.

[0010] Based on the scheduling candidate task list and combined with the inter-node communication link status, target low-load nodes are matched for each migrateable data segment in the list. According to the matching results, the migrateable data segments are forwarded to their respective target low-load nodes, and task migration execution logs are generated.

[0011] The communication latency, packet loss rate, and processing queue length of each node are collected during the migration process. The collected results are combined with the task migration execution log to form load adjustment data. The load adjustment data is used to adjust the transmission order and migration path of incomplete tasks.

[0012] Based on the load adjustment data, scheduling execution instructions are issued to high-load nodes and receiving nodes to perform task migration and load distribution for communication tasks, and generate a final node load status report.

[0013] Based on the final node load status report, the calculation parameters for subsequent scheduling processes are adaptively adjusted.

[0014] In some possible implementations, the acquisition of current, voltage, processing queue length, communication latency, packet loss rate, and ambient light, wind speed, and temperature of each edge node to form a node operating status dataset includes:

[0015] The raw electrical, network, and environmental data of the collected nodes are used to form a multi-source raw dataset, which includes raw electrical data, raw queue data, raw network data, and raw environmental data.

[0016] For the raw electrical data in the multi-source raw dataset, adaptive filtering and anomaly suppression processing are performed in combination with the node power fluctuation characteristics and sampling period characteristics to eliminate the influence of instantaneous interference and random noise, and obtain clean electrical parameters that can stably characterize the changing trend of node power supply status.

[0017] Using the cleanroom electrical parameters as a time reference, timestamp reconstruction and alignment processing are performed on the original queue data in the multi-source raw dataset. The load status of the control node and the power supply status of the node are established in a corresponding relationship under a unified time axis, generating time-synchronized node core operation data that reflects the core operation status of the node.

[0018] The core operating data of the time synchronization node is fused with the original network data and original environmental data in the multi-source raw dataset across domains based on the unique identifier of the node, and a node-level multi-dimensional state description structure is constructed to associate and encapsulate the internal operating state of the node with the external environmental state.

[0019] The node-level multidimensional state description structure, after being associated and encapsulated, is output as a node running state dataset for subsequent scheduling and analysis.

[0020] In some possible implementations, after the output node running state dataset, the following is also included:

[0021] Historical light intensity data and historical temperature data are extracted from the node operating status dataset, and real-time light intensity value and real-time temperature value are extracted at the current moment.

[0022] Statistical modeling is performed based on the temporal distribution characteristics of the historical light intensity data to calculate the light intensity benchmark value that characterizes the stability level of the node environment, and the corresponding temperature benchmark value is calculated based on the periodic fluctuation characteristics of the historical temperature data.

[0023] Using the aforementioned light intensity reference value and temperature reference value as references, the deviation of the real-time light intensity value and real-time temperature value is calculated respectively to obtain the light disturbance value and temperature disturbance value that characterize the degree of instantaneous change in the environment;

[0024] The illumination disturbance value and temperature disturbance value are jointly modeled, and a single environmental disturbance quantification parameter is generated through weighted fusion and scale unification processing.

[0025] The environmental disturbance quantification parameters are written into the node operating status dataset as environmental disturbance fields, and the fields are associated with the electrical and network parameters of the corresponding nodes to form an updated node operating status dataset.

[0026] In some possible implementations, calculating the real-time communication load of each edge node based on the node runtime status dataset includes:

[0027] A subset of operating parameters for load calculation is extracted concurrently based on node identifier. The subset of operating parameters includes at least a subset of electrical parameters characterizing the node's power supply status, a subset of network parameters characterizing the communication status, and an environmental disturbance field.

[0028] Based on the electrical parameter subset in the aforementioned operating parameter subset, the power consumption value of the node in the current scheduling cycle is calculated according to the node power supply model, and the network congestion value reflecting the degree of communication congestion is calculated based on the network parameter subset in the aforementioned operating parameter subset.

[0029] Normalization and scaling processing are performed on the power consumption value and network congestion value respectively, converting them into comparable standardized power consumption factor and standardized congestion factor;

[0030] Based on the preset weighting relationship, a weighted fusion calculation is performed on the standardized power consumption factor, the standardized congestion factor, and the environmental disturbance field. The calculation result is used as the real-time communication load of the corresponding edge node and written into the load field of the node running status dataset.

[0031] In some possible implementations, the step of calculating the real-time communication load of each edge node based on the node operating status dataset, and associating and storing the load value of each node with an identifier indicating whether it exceeds a preset threshold to form a load status mapping table includes:

[0032] The real-time communication load of each edge node is aggregated to form a full network load set covering the entire smart light pole network;

[0033] Statistical analysis is performed based on the entire network load set to calculate the average load value reflecting the overall load level and the load fluctuation range value characterizing the degree of load dispersion.

[0034] Based on the average load value of the entire network, and combined with the load fluctuation range value, a high load judgment threshold for identifying nodes with abnormal load is dynamically calculated and generated.

[0035] The instant messaging load value of each node is compared with the high load threshold one by one to filter out nodes that exceed the high load threshold.

[0036] Based on the range in which the instant communication load value of each node exceeds the high load determination threshold, the node load status is classified and processed accordingly.

[0037] The node identifier, instant messaging load value, and corresponding load classification results are structurally associated to generate a load status mapping table.

[0038] In some possible implementations, the step of filtering nodes identified as exceeding a threshold to form a high-load node set based on the load state mapping table, and decomposing the communication tasks of the high-load nodes into transferable data fragments to generate a scheduling candidate task list includes:

[0039] Based on the load classification results in the load status mapping table, the nodes in the high load classification range are determined, and the communication task information to be processed in the high load nodes is extracted.

[0040] The communication task information is parsed to identify the task type of each communication task, and the corresponding task decomposition strategy is matched according to the task type.

[0041] Based on the task decomposition strategy, the communication task is divided into multiple task units with associated identifiers that can be processed in parallel.

[0042] Each task unit is encapsulated with processing control information including sequence number, task identifier, data size, and expected processing time calculated based on data size and task type, forming a transferable data fragment;

[0043] Priority is calculated for the transferable data segments based on the expected processing time, and the transferable data segments carrying priority and processing control information are aggregated and sorted to generate a list of scheduling candidate tasks.

[0044] In some possible implementations, the method further includes:

[0045] The processing control information of each transferable data segment in the scheduling candidate task list is parsed to extract the corresponding task identifier;

[0046] Based on the task identifier, a preset task dependency database is queried to obtain the original dependency description describing the sequential constraints between tasks;

[0047] Based on the original dependency description, a task dependency graph representing the execution constraints between transferable data fragments is constructed;

[0048] Perform topological analysis on the task dependency graph to deduce the global execution order of each transferable data segment under the condition of satisfying dependency constraints;

[0049] The global execution order obtained from the deduction is written into the scheduling candidate task list, and the transferable data fragments are reordered to generate an updated scheduling candidate task list.

[0050] In some possible implementations, the step of matching each traversable data segment in the list of scheduling candidate tasks with a target low-load node based on the inter-node communication link status, and forwarding the traversable data segment to its corresponding target low-load node according to the matching result, includes:

[0051] Based on the current smart light pole network topology, link quality data characterizing the communication quality between nodes and resource status data of candidate low-load nodes are acquired concurrently.

[0052] Using the link quality data, a corresponding transmission efficiency score is calculated for the communication path from each high-load node to each candidate low-load node. Combining the resource status data and the features of transferable data segments extracted from the scheduling candidate task list, a matching score for the processing capability of the candidate low-load node for the data segment is calculated.

[0053] The transmission efficiency score and processing capability matching score corresponding to the same candidate low-load node are fused to generate a comprehensive forwarding evaluation result for that candidate low-load node.

[0054] Based on the comprehensive forwarding evaluation results, a corresponding target forwarding node is selected for each migrated data segment.

[0055] In some possible implementations, the step of issuing scheduling execution instructions to high-load nodes and receiving nodes based on the load adjustment data includes:

[0056] The load adjustment data is parsed to determine the new target node address and the updated sending order corresponding to the migrated data segment to be adjusted.

[0057] Based on the new target node address and the updated transmission order, scheduling control instructions are generated for the high-load node and the receiving node, respectively.

[0058] Send scheduling control instructions to high-load nodes to control them to forward migrateable data fragments to the corresponding new target node addresses according to the updated sending order;

[0059] Send a scheduling control instruction to the receiving node corresponding to the new target node address to trigger the reception and processing of the migrateable data segment;

[0060] The execution status information of the scheduling control instructions is collected and written into the task migration execution log.

[0061] In some possible implementations, the adaptive feedback adjustment of calculation parameters for subsequent scheduling processes based on the final node load status report includes:

[0062] Extract the real-time communication load values ​​of each edge node after scheduling is completed from the final node load status report;

[0063] A load balancing index reflecting the current scheduling result distribution characteristics is calculated based on the instant messaging load values.

[0064] The load balancing index is compared and analyzed with the pre-stored historical load balancing index to generate scheduling deviation analysis results.

[0065] Based on the scheduling deviation analysis results, the weight parameters used in the instant messaging load calculation process are updated;

[0066] The updated weight parameters are applied to the load calculation process in the next scheduling cycle.

[0067] As can be seen from the above technical solution, this application has the following beneficial effects:

[0068] 1. This method constructs a unified panoramic view of node operating status by collecting and fusing multi-dimensional physical parameters of nodes, including electrical, computational, network, and environmental parameters. Adaptive filtering and timestamp alignment techniques ensure data quality and spatiotemporal consistency. A multi-factor weighted fusion model transforms heterogeneous indicators into unified real-time communication load values, fundamentally overcoming the limitations of traditional single-dimensional monitoring. This method accurately reflects the real load pressure caused by cross-domain coupling effects. In particular, the introduction of environmental disturbance quantification parameters enables the system to proactively perceive external risks such as severe weather, thereby avoiding potential performance degradation before scheduling decisions are made and significantly improving the accuracy of the assessment.

[0069] 2. This method transforms the macroscopic problem of high node load into a series of microscopic task fragment migration decisions. Based on task type-aware intelligent decomposition and dependency analysis, it ensures the rationality and logical correctness of task splitting. During migration decisions, it combines link transmission efficiency and node processing capability matching as two factors for comprehensive evaluation, ensuring that the selected target node is not only idle but also easily reachable and capable of handling the task. This fine-grained scheduling mechanism avoids the additional network overhead and resource mismatch caused by coarse migration and ensures the real-time performance of critical services through priority ranking. While reducing the load on hot nodes, it maximizes the utilization of the entire network's computing and communication resources. Attached Figure Description

[0070] The invention will now be further described with reference to the accompanying drawings.

[0071] Figure 1 A first flowchart of a communication load adaptive scheduling method for edge nodes of smart light poles provided in this application embodiment;

[0072] Figure 2 A second flowchart of a communication load adaptive scheduling method for edge nodes of smart light poles provided in this application embodiment;

[0073] Figure 3 A third flowchart of a communication load adaptive scheduling method for edge nodes of smart light poles provided in this application embodiment;

[0074] Figure 4 The fourth flowchart of a communication load adaptive scheduling method for edge nodes of smart light poles provided in this application embodiment is shown. Detailed Implementation

[0075] The terms "first," "second," and "third," etc., used in this application specification, claims, and drawings are for distinguishing different objects, not for specifying a particular order.

[0076] In the embodiments of this application, the words "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the words "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0077] Research has revealed that existing technologies lack a communication load scheduling method that can deeply integrate the multi-dimensional states of node electrical, network, computing, and environment, and perform dynamic, precise, and closed-loop adaptive scheduling based on this. This makes it difficult to meet the requirements of smart light pole networks for high reliability, high resource utilization, and low task latency.

[0078] To address the aforementioned issues, this application provides an adaptive scheduling method for the communication load of edge nodes in smart light poles:

[0079] Example 1

[0080] To solve the above problems, such as Figure 1 As shown, the core idea of ​​this invention is to construct a complete scheduling system from multi-dimensional state perception to intelligent decision-making and execution, and then to closed-loop feedback optimization. By integrating electrical, network, queue, and environmental data across domains, it achieves refined and forward-looking assessment of node load status. Based on dynamic thresholds and multi-dimensional quantities, it achieves accurate identification of high-load nodes and intelligent task decomposition. By matching and deciding on the coupling of link status and task characteristics, it achieves efficient and reliable task migration. Furthermore, by utilizing feedback data from the migration process and the final state, it continuously optimizes the parameters of the scheduling strategy to form adaptive capabilities.

[0081] Step S1: Generate a node running status dataset.

[0082] Technical solution: In the smart light pole network, the current, voltage, processing queue length, communication delay, packet loss rate, and multi-dimensional physical parameters of the surrounding environment such as light intensity, wind speed, and temperature of each edge node are collected synchronously.

[0083] Overload or failure of smart light pole nodes is often the result of multiple coupled factors. For example, a sudden voltage drop may cause the computing unit to reduce its frequency, leading to processing queue backlog; high temperature may exacerbate chip thermal noise, resulting in increased communication error rate and latency. Traditional single-dimensional monitoring, such as looking only at CPU load, cannot capture such cross-domain coupled faults or performance bottleneck precursors. This invention provides a unique and reliable data foundation for establishing an evaluation model that can characterize the overall health or serviceability of nodes by synchronously collecting these parameters. By carefully selecting and synchronizing these parameters with inherent physical correlations, it creates the possibility of establishing an accurate load evaluation model that conforms to the physical reality of smart light poles.

[0084] Step S2: Calculate the instant messaging load and form a load status mapping table.

[0085] Based on the aforementioned dataset, the instant messaging load of each node is calculated. This load value is a quantitative indicator that integrates multiple dimensions of influence. Subsequently, the load value of each node is associated with an identifier indicating whether it exceeds a preset threshold and stored, forming a global load status mapping table.

[0086] This study addresses two key challenges: extracting an accurate and comparable load status indicator from multidimensional data and setting reasonable overload judgment criteria based on dynamic network conditions. The calculation of instant messaging load is essentially a multi-factor fusion modeling process. It integrates parameters with different dimensions and physical meanings, such as power consumption, latency, and environmental disturbances, into a single scalar through normalization and weighting. This scalar better reflects the comprehensive pressure a node experiences at a specific moment than a single indicator. Forming a mapping table means not only generating load values ​​but also classifying and identifying whether they exceed thresholds. The preset thresholds are not fixed values, allowing overload judgment to adapt to changes in the overall network load level. This avoids misjudging too many nodes during periods of high overall load or missing potential bottlenecks during periods of overall idleness. Therefore, it generates an accurate map that can dynamically and fairly identify bottlenecks or hotspots in the network, providing clear targets for subsequent scheduling interventions.

[0087] Step S3: Generate a list of scheduling candidate tasks.

[0088] Technical solution: Select a set of high-load nodes based on the mapping table, decompose the communication tasks to be processed on these nodes into smaller, independently movable data fragments, and generate a list of scheduling candidate tasks including these fragments and their priorities.

[0089] Directly migrating the entire large task may be impractical and detrimental to fine-grained load balancing. This step decomposes the task, refining the load scheduling granularity from the node level to the task fragment level. The decomposition considers task type, such as compute-intensive or I / O-intensive, and encapsulates information such as expected processing time. Generating a list and prioritizing it achieves two goals: firstly, it objectifies and serializes scheduling decisions, so the scheduler no longer faces ambiguous tasks but a series of well-defined units to be scheduled; secondly, it introduces a priority mechanism to ensure that critical or urgent tasks, such as security alarms, receive priority scheduling resources. This transforms the macroscopic problem of high node load into a series of microscopic, actionable task fragment migration decision problems, laying the foundation for fine-grained scheduling.

[0090] Step S4: Match the target node and execute forwarding, generating logs.

[0091] Technical solution: Based on the scheduling candidate task list and combined with the real-time inter-node communication link status, a target low-load node is intelligently matched for each migrateable data segment in the list. Then, forwarding is performed according to the matching result, and a detailed migration execution log is generated.

[0092] Technical Issues and Principle Derivation: This is the core decision-making and execution phase of scheduling. Its innovation lies in the two-factor coupling of the matching decision: it considers both network link status, such as RTT and packet loss rate, to ensure the migration process itself is efficient and reliable, avoiding migration on poor links that would increase network burden; and the matching degree between task fragment characteristics and target node resources, such as whether computing power and memory are sufficient. This overcomes the shortcomings of traditional methods that only consider node load margin while neglecting migration overhead and task adaptability. The generated logs record the complete trajectory of each migration, providing a data traceability basis for subsequent performance analysis, problem tracing, and, most importantly, dynamic adjustments. The beneficial effect of this step is to achieve intelligent migration, that is, reducing the load on the source node while minimizing migration side effects and maximizing task execution efficiency on the target node.

[0093] Step S5: Generate load adjustment data.

[0094] Technical solution: During the migration process, key performance indicators of each node are continuously collected, such as communication latency, packet loss rate, and queue length. These real-time monitoring data are then combined with the aforementioned task migration execution logs to form load adjustment data.

[0095] The effectiveness of scheduling decisions needs to be validated. Network conditions are dynamic, and this step establishes a real-time feedback loop. Migration logs describe the planned actions, while real-time data reflects what actually happened. Combining these two aspects reveals: Did the planned migration cause unexpected network congestion? Was the processing capacity of the target nodes overestimated? Did the performance of certain links deteriorate after scheduling? These insights are encapsulated in the load balancing data. This step enables the scheduling system to perceive and react, promptly capturing the side effects or new problems arising from scheduling actions, providing a basis for online adjustments.

[0096] Step S6: Issue the execution command and generate the final report.

[0097] Technical solution: Based on the load adjustment data, new scheduling execution instructions are issued to relevant nodes, high-load source nodes and receiving nodes to adjust unfinished or subsequent migration tasks, such as changing the transmission order or changing the migration path, and finally completing this round of scheduling and generating a final node load status report.

[0098] This is the execution end of the feedback loop. Based on the resulting adjustment data, the system can dynamically intervene in the ongoing scheduling process. For example, if a sudden increase in latency is detected on a certain path, subsequent task segments can be instructed to use alternative paths; if a receiving node's queue begins to grow, the distribution of new tasks to it can be paused. This achieves dynamic correction and elastic scaling during the scheduling process. The final report is a snapshot of the network's steady-state after this round of scheduling, serving as input for evaluating the overall scheduling effect and conducting long-term optimization. This step endows the scheduling method with strong resilience, enabling it to cope with instantaneous fluctuations in network and load.

[0099] Step S7: Adaptive feedback adjustment.

[0100] Technical solution: Based on the load status report of the final node, adaptive feedback adjustment is performed on the calculation parameters used in the subsequent scheduling process, such as the weights in the load calculation formula.

[0101] This is the system's long-term learning and optimization mechanism. By analyzing load balancing performance reports after multiple rounds of scheduling, such as load balancing metrics, the system can determine the effectiveness of the current load assessment model. If the load remains uneven after scheduling, it may mean that the weighting of certain factors is unreasonable, for example, underestimating the impact of ambient temperature on computational efficiency. The system can automatically or semi-automatically adjust these weighting parameters, thereby making better decisions in the next round of scheduling. This step evolves the entire method from static, experience-based scheduling to dynamic, self-optimizing intelligent scheduling, enabling continuous performance improvement to adapt to long-term business model changes and equipment aging.

[0102] Details of constructing the node running status dataset.

[0103] Multi-source raw data collection: The data is clearly divided into four categories:

[0104] Raw electrical data: collected through the power metering chip built into the node, including instantaneous and effective values ​​of current and voltage on both AC and DC sides. The sampling frequency must be higher than the mains power fluctuation frequency, such as several times 50Hz.

[0105] Raw queue data: obtained through APIs provided by the operating system or edge computing framework, including network send or receive queue lengths and computation task queues, such as thread pool wait queue lengths.

[0106] Raw network data: obtained through active probing or passive listening, for example, periodically sending probe packets to neighboring nodes to measure communication latency, round-trip time (RTT), and packet loss rate; listening to the number of MAC layer retransmissions as a reference for link quality.

[0107] Raw environmental data: acquired through the environmental sensor module integrated into the light pole, including light intensity, wind speed, and temperature.

[0108] All four types of data need to be stamped with high-precision timestamps during collection, such as by using network time protocols for synchronization.

[0109] Electrical data filtering: An adaptive filter is designed for the raw electrical data. The principle is as follows: First, the inherent fluctuation characteristics of the power supply are identified, such as the 50Hz mains frequency and its harmonics. Then, the filter cutoff frequency is dynamically adjusted according to the actual sampling period. A sliding window mean filtering combined with an outlier suppression algorithm is used: For a voltage sequence within a sampling window, its mean and standard deviation are calculated. Instantaneous values ​​deviating from the mean by more than three times the standard deviation are considered outliers, possibly caused by instantaneous load surges or measurement noise. These are replaced with interpolated or windowed mean values ​​from the preceding and following normal values, thus obtaining clean electrical parameters. This process stably characterizes the changing trend of the node's power supply status and filters out meaningless spikes.

[0110] Time synchronization processing: High-quality timestamps of filtered electrical parameters are used as a benchmark, as electrical sampling is typically the most frequent and stable method. The timestamps of the original queue data are reconstructed and aligned. For example, queue length data might be reported at 1-second intervals, reflecting the average state over the past second. Through interpolation algorithms, the queue data is aligned to the electrical parameter timestamp sequence, ensuring a one-to-one correspondence between the voltage value at a given moment and the estimated queue length at the same moment, generating core operational data for the time synchronization node.

[0111] Cross-domain fusion and encapsulation: This involves merging core operational data, raw network data, and raw environmental data after time synchronization, based on a globally unique identifier for each node, such as a MAC address or a pre-assigned ID. The fusion method involves constructing a node-level multi-dimensional state description structure, which can be understood as a data object or database record with fixed fields. This structure uses the node ID as the primary key and includes multiple nested or parallel field groups, corresponding to parameters and timestamps for dimensions such as electrical, queue, network, and environment. The encapsulation process involves filling data from different sources into the corresponding positions within this structure.

[0112] Feature Relationship: It will abstractly obtain... parameters specifically as technical actions with clear classification, processing flow, and output structure.

[0113] In smart light pole scenarios, electrical fluctuations, such as voltage dips caused by the start-up and shutdown of nearby large equipment and environmental interference, such as slight shaking of equipment due to strong winds affecting contact, are common. Adaptive filtering effectively removes these transient interferences, avoiding misjudgments of node power status due to data noise. For example, a single voltage sampling spike will not be interpreted as a node being out of power. Timestamp alignment ensures that the data used is strictly synchronized in time when analyzing cross-domain causal relationships such as whether high loads are caused by voltage drops, guaranteeing the accuracy of the analysis. Unified structured encapsulation allows all subsequent analysis modules to access data of any dimension from any node in a consistent and efficient manner, laying an engineering foundation for complex analyses.

[0114] Generation of environmental disturbance quantification parameters.

[0115] Data extraction: Extract historical light intensity data and historical temperature data of the target node from the dataset, such as data points at 5-minute intervals over the past 24 hours, as well as the real-time value at the current moment.

[0116] Benchmark value calculation: Statistical modeling of historical data.

[0117] For light intensity, analyze its temporal distribution characteristics. Since light intensity has a clear diurnal periodicity, we can use the moving average method or fit a diurnal variation curve to calculate the average light intensity at the current moment in the same historical period, such as the same moment in the past 7 days. This average light intensity can be used as a benchmark value, representing the light level that should be present at this point under normal weather conditions.

[0118] For temperature, we analyze its periodic fluctuation characteristics, diurnal variation and seasonal variation. Similarly, we calculate the historical average temperature for the same period at the current moment as the temperature benchmark.

[0119] Disturbance value calculation: Calculate the deviation between the real-time value and the corresponding reference value.

[0120] Illumination disturbance value = |Real-time illumination intensity - Illumination intensity baseline value| / Illumination intensity baseline value or a set normalization coefficient.

[0121] Temperature disturbance value = real-time temperature - temperature reference value.

[0122] This deviation value quantifies the degree of instantaneous change in the environment relative to its "normal" state. For example, a sudden drop in sunlight due to dark clouds blocking the sun, or a rapid drop in temperature due to the passage of cold air, will produce a large disturbance value.

[0123] Joint modeling generates a single parameter: weighted fusion of illumination disturbance values ​​and temperature disturbance values. The weights can be set based on experience or analysis of historical fault data. For example, temperature may have a greater impact on equipment stability, so it is given a higher weight. Then, scale unification processing is performed, such as using max-min normalization or Z-score standardization, to map the fused values ​​to [0,1] or a standard normal distribution interval, ultimately generating a single environmental disturbance quantification parameter.

[0124] Associated Writing: This parameter is written as a new environmental disturbance field into the state description structure of the node, and associated with electrical, network and other parameters.

[0125] In outdoor smart light pole scenarios, sudden environmental changes are commonplace. This solution transforms environmental factors from monitoring indicators into load impact factors that directly influence scheduling decisions. For example, when environmental disturbance parameters are high, the scheduling system can: Proactively adjust: Recognizing that high temperatures may lead to a decrease in the computational efficiency of all nodes, the system can pre-weight load calculations or trigger more conservative scheduling strategies, such as reserving more resource redundancy. Differentiated handling: For nodes primarily powered by solar energy, a sudden drop in sunlight directly means a decrease in their future power supply capacity. The scheduling system should prioritize offloading their tasks or avoid assigning new tasks to them. This enables proactive perception and preventative response to environmental risks, enhancing the system's resilience under severe weather conditions.

[0126] Calculation details of instant messaging load.

[0127] Parameter Extraction: From the updated node runtime status dataset, including the environmental perturbation field, a subset of runtime parameters is extracted for each node concurrently. This subset is a minimal complete set, including:

[0128] Subset of electrical parameters: such as RMS voltage, RMS current, or calculated instantaneous power.

[0129] Subset of network parameters: such as average communication latency and average packet loss rate over a recent period.

[0130] Environmental disturbance field: A single quantized value obtained from the calculation.

[0131] Intermediate value calculation:

[0132] Power consumption value: Calculated based on a subset of electrical parameters according to the node power supply model. A simplified model can be: Power consumption value = Voltage × Current × Power factor. More complex models can consider the individual power consumption of different hardware modules, reflecting the energy consumption rate of the node within the current scheduling cycle, which is an important indicator of the load.

[0133] Network congestion value: calculated based on a subset of network parameters. For example, a function can be designed as: Network congestion value = α × Normalized delay + β × Packet loss rate, where α and β are weighting coefficients. This value comprehensively reflects the busy and low-quality state of the node's communication link.

[0134] Normalization and scaling: Since power consumption and network congestion values ​​have different dimensions and ranges, they cannot be directly compared or merged. Therefore, they are normalized separately. For example, minimum-maximum normalization is used: Normalized power consumption factor = (current power value - historical minimum power) / (historical maximum power - historical minimum power). A similar process is performed on network congestion values ​​to obtain a normalized congestion factor. This maps them to the same numerical range (e.g., [0,1]), making them comparable factors.

[0135] Weighted fusion calculation: The standardized power consumption factor, standardized congestion factor, and environmental disturbance field are weighted and fused according to a preset weight relationship. For example: Instant messaging load = W1 × standardized power consumption factor + W2 × standardized congestion factor + W3 × environmental disturbance field, where W1 + W2 + W3 = 1. The initial weights can be set according to domain knowledge, such as W2 being slightly higher because communication load is the core. This weighted sum is the final, comprehensive instant messaging load value.

[0136] Storage: Write the load value to the dedicated load field in the node status dataset.

[0137] In a smart street light network, a node's high load can have multiple causes: it might be performing extensive video analysis, uploading large amounts of data, or experiencing performance degradation due to harsh environmental conditions. This solution uses a multi-factor weighted fusion model to quantify these heterogeneous causes into a single comprehensive load value. Its specific effects include: Fairness: A node experiencing reduced computing power due to high temperatures and a node busy due to queue congestion may calculate similar high load values, thus both receiving attention from the scheduling system. Foresight: The introduction of environmental disturbance fields gives load calculation a degree of predictability; for example, high temperatures indicate ongoing performance degradation, rather than simply a reactive consequence. This makes scheduling decisions more refined and predictive.

[0138] Details of generating the load status mapping table.

[0139] Aggregate network load: Collect the real-time communication load values ​​calculated by all edge nodes to form a network load aggregation.

[0140] Statistical analysis: Calculate the network-wide average load value (μ) and load fluctuation range for this set. The fluctuation range can be represented by the standard deviation (σ), or by (maximum value - minimum value). μ reflects the overall network busyness, and σ reflects the degree of load distribution balance.

[0141] Dynamic threshold calculation: The high load threshold (Th) is dynamically calculated based on μ and σ. A typical formula is: Th = μ + k × σ. Here, k is an adjustable parameter, such as k = 1 or 1.5. Its setting is based on the principle of normal distribution in statistics: when the load distribution is close to normal, μ + kσ can cover most nodes, and nodes exceeding this threshold can be considered abnormally high-load nodes. When the network is generally busy (high μ) or the load is very uneven (large σ), Th automatically increases to avoid misjudging too many nodes; conversely, when the network is generally idle, Th decreases, allowing even slightly high loads to be detected. This achieves adaptive overload detection.

[0142] Comparison and filtering: Compare the load value of each node with Th one by one, and filter out the nodes whose load value is greater than Th.

[0143] Tiered processing: Not all nodes exceeding the threshold are treated the same. Nodes are categorized based on the extent of the overload, such as ((load value - Th) / Th), for example, mild overload or severe overload. This provides more granular priority information for subsequent scheduling, prioritizing severely overloaded nodes.

[0144] Generate a mapping table: Structure the association between node ID, its load value, and the corresponding load level to generate a load status mapping table that can be quickly queried.

[0145] In scenarios with large-scale smart light pole networks and significant business fluctuations, fixed thresholds become completely ineffective. This solution's dynamic threshold mechanism offers key advantages: It avoids false positives during peak traffic periods: when the load on all nodes increases during the evening rush hour, the threshold automatically increases, only identifying truly bottleneck nodes. The scheduling system avoids futilely trying to balance a network already heavily loaded. It also improves sensitivity during quiet periods: during off-peak hours in the early morning, the threshold automatically decreases, allowing for the rapid detection of abnormal activity in individual nodes, such as malicious attacks or equipment malfunctions causing recurring tasks. This ensures that the scheduling system intelligently focuses on the nodes most in need of intervention under various overall network conditions, improving the efficiency of scheduling resource utilization.

[0146] Details of generating the candidate task list for scheduling.

[0147] Identify high-load nodes and extract tasks: Based on the load classification results in the generated load status mapping table, locate high-load level nodes and extract all pending communication task information from the task manager of these nodes, such as task descriptors, including task ID, source data location, estimated required computing resources, etc.

[0148] Task type analysis and matching strategy: The system analyzes task information to identify its task type. For example, by analyzing API call or resource request patterns in the task description, tasks can be categorized as video stream decoding, sensor data aggregation, model inference, etc. The system maintains a task decomposition strategy library, predefining decomposition rules for each task type. For example, video stream tasks can be decomposed by time slices, such as one frame per second, or by spatial blocks; data aggregation tasks can be decomposed by data source.

[0149] Task splitting: Based on the matched decomposition strategy, the original communication task is split into multiple smaller task units. Each unit is assigned a unique association identifier, such as parent task ID - child unit number, and is marked as capable of being processed in parallel.

[0150] Encapsulation into Portable Data Fragments: Processing control information is encapsulated for each task unit, forming a complete portable data fragment. This information includes:

[0151] Sequence number: The order within the parent task.

[0152] Task Identifier: Globally Unique ID.

[0153] Data size: The amount of data that needs to be processed in this segment.

[0154] Expected processing time: Estimated using a simple performance model based on data size and task type. For example, for a task with known computational density, expected processing time = data size × computation time per unit of data.

[0155] Priority Calculation and Sorting: Priority is calculated based on expected processing time. Generally, task segments with shorter expected processing times have higher priority, allowing for faster completion and resource release. Other strategies exist, such as prioritizing earliest deadlines, which involves aggregating all transferable data segments carrying priority and processing control information, sorting them by priority, and ultimately generating a list of scheduling candidate tasks.

[0156] Smart light poles handle a variety of tasks, some urgent, others massive. This solution breaks down large tasks into smaller, manageable parts through task type awareness, enabling refined load scheduling, estimation of expected processing time, and prioritization. This is particularly important in this scenario: it ensures that small, urgent analysis tasks, such as identifying abnormal personnel, are scheduled and executed before large, non-real-time tasks like generating monthly environmental data reports, thus meeting the differentiated needs of smart city applications for real-time performance and critical business assurance.

[0157] Task dependency handling.

[0158] Extract task identifier: Extract the task identifier from the segment processing control information of the scheduling candidate task list.

[0159] Query the dependency database: The system maintains a task dependency database, recording common sequential constraints between different task types. For example, the object detection task can only begin after the video decoding task is completed. Based on the type information associated with the task identifier, query this database to obtain the original dependency description.

[0160] Constructing a task dependency graph: Based on the query results, construct a directed acyclic graph (DAG) using transferable data fragments as nodes and dependencies as directed edges. For example, fragment A (decoding frame 1) points to fragment C (detecting frame 1).

[0161] Topology analysis: Topological sorting analysis is performed on the DAG, a graph theory algorithm that can deduce the global execution order of all task segments while satisfying all dependency constraints. For example, it guarantees that all decoding tasks in frame N are scheduled before detection tasks in frame N.

[0162] Update the task list: Use the new order result obtained from the topological sort as the highest priority sorting basis to update the scheduling candidate task list, overwriting the original order based only on expected processing time, and generate an updated scheduling candidate task list.

[0163] In the complex workflow of smart light pole applications, task dependencies are prevalent. For example, license plate recognition depends on vehicle detection, and data fusion relies on raw data reports from multiple sensors. Ignoring these dependencies and blindly migrating data fragments that depend on subsequent tasks can lead to target nodes waiting in vacancies or even deadlocks. This solution constructs a dependency graph and performs topological sorting to ensure that the logical order of tasks is strictly adhered to regardless of which node they are migrated to for execution. This is crucial in a distributed execution environment, guaranteeing the correctness of business logic and avoiding program errors or invalid waiting caused by scheduling.

[0164] Details of the matching decision for the target low-load node.

[0165] Acquiring decision-making data: Based on network topology, two types of data are acquired concurrently:

[0166] Link quality data: Real-time quality metrics of the links between all nodes are obtained through a centralized controller or distributed protocol, such as bandwidth, latency, and packet loss rate.

[0167] Candidate low-load node resource status data: From the load status mapping table and the node running status dataset, obtain the detailed resource status of nodes whose load has not exceeded the threshold and have idle resources, such as remaining CPU computing power, available memory, and storage IO capacity.

[0168] Calculate the transmission efficiency score: For each pair of high-load nodes to candidate low-load nodes, calculate the transmission efficiency score using the link quality data between them. The scoring function can be designed as: Efficiency Score = Available Bandwidth / (Latency × (1 + Packet Loss Rate)). This formula combines bandwidth, latency, and reliability; a higher value indicates that the path is more efficient for data migration.

[0169] Computational processing capacity matching score: This score combines the resource status data of candidate nodes with the requirements extracted from the characteristics of portable data segments, such as task type and data size. For example, for a computationally intensive segment, the score focuses on the remaining CPU power of the candidate nodes; for a memory-intensive segment, the score focuses on the available memory. The score can be a function of the matching degree between resource supply and demand.

[0170] The comprehensive evaluation result is generated by merging the two scores corresponding to the same candidate node. A weighted sum can be used: Comprehensive evaluation result = γ × transmission efficiency score + δ × processing capacity matching score. γ and δ are weights that can be dynamically adjusted according to network conditions; for example, γ is increased when the network is congested.

[0171] Selecting the target node: For a specific portable data segment, compare the comprehensive evaluation results of all candidate nodes and select the candidate node with the highest score as the target forwarding node.

[0172] Feature-based connections: It uses task lists and network and node status data to make specific migration decisions.

[0173] Smart light pole networks often employ wireless mesh connections, resulting in dynamically changing and asymmetrical link quality. This solution's two-factor matching decision addresses a key issue: avoiding the pitfalls of robbing Peter to pay Paul. It ensures that the selected target nodes are not only limited but also easily reachable and capable of handling the task. For example, it won't migrate a high-bandwidth video clip to a node with low load but a poor-quality connection, which would only shift network congestion from the source node to the link. This collaborative decision-making, considering both network transmission overhead and node processing adaptability, significantly improves the overall success rate and efficiency of task migration, and is crucial for achieving effective load balancing.

[0174] Details of the issuance of scheduling execution instructions.

[0175] Analysis and adjustment data: The load adjustment data includes dynamic adjustment strategies, such as changing the migration path of task segment T123 from node A to B to node A to C, or advancing the sending order of segment T456.

[0176] Generate control instructions: Based on the parsed new target node address and the updated transmission order, generate specific, executable scheduling control instructions. These instructions are of two types:

[0177] Instructions for high-load nodes include action forwarding, target fragment ID, new target address, and sequence number.

[0178] Instructions for the receiving node include: preparing to receive and process the action, source node ID, upcoming segment ID, and expected processing requirements.

[0179] Sending instructions: Through a reliable control channel, an independent, high-priority message queue sends instructions to both the high-load node and the receiving node.

[0180] Control execution: After receiving the instruction, the high-load node adjusts its local task scheduler to encapsulate the specified data fragments into network packets in a new order and send them to the new target address.

[0181] After receiving the instruction, the receiving node reserves the corresponding resources, such as memory buffers and computing threads, and enters a waiting state.

[0182] Record execution status: After the instruction is sent, the system collects its execution status information, such as whether the instruction has been delivered, whether the task fragment has started to be transmitted, and whether the target node has confirmed receipt. This status information is then appended to the task migration execution log to form a more complete audit trail.

[0183] In dynamic networks, static migration plans often require adjustments. This solution achieves fine-grained, real-time control of the migration process through a standardized, bidirectional control command mechanism. Specific scenario effects include: rapid response: when performance degradation is detected on a path, the migration flow on that path can be immediately interrupted, and a switch to an alternative path can be initiated via command, minimizing service interruption time. Collaborative preparation: receiving nodes are notified in advance, enabling them to prepare resources ahead of time, achieving zero-wait processing after data arrival and reducing the overall task completion time. This enhances the agility and controllability of the entire scheduling system.

[0184] Details of adaptive feedback adjustment.

[0185] Extract load values: Read the real-time communication load values ​​of all nodes after scheduling is completed from the final node load status report.

[0186] Calculate load balance metrics: Based on these load values, calculate a quantitative indicator reflecting the quality of scheduling results. Commonly used metrics include load variance or standard deviation. The smaller the variance, the more balanced the load across nodes. The Gini coefficient or the ratio of the heaviest to the lightest load node can also be used.

[0187] Comparative analysis generates deviation results: The calculated current load balancing index is compared with the pre-stored historical load balancing index, specifically the average or optimal value after the previous few rounds of scheduling. If the current index deteriorates, such as with increased variance, the scheduling deviation analysis indicates a decline in load balancing effectiveness.

[0188] Updating weight parameters: The system employs a gradient descent-based strategy for fine-tuning. It quantitatively assesses the magnitude of network congestion-related errors to improve load balancing, adjusting the weights accordingly: the weight of the network congestion factor is moderately increased, while the weight of the power consumption factor is correspondingly decreased. The adjustment magnitude is controlled by a small positive number called the learning rate to ensure the stability and gradualness of the update. To guarantee that the sum of all weight coefficients is 1, the system normalizes the updated weight set.

[0189] Apply updated parameters: Save the updated weight parameters and apply them to the real-time communication load calculation process of all nodes in the next scheduling cycle.

[0190] The business models and environment of smart light pole networks change slowly, such as the addition of new applications or the degradation of equipment performance due to aging. Fixed weight parameters may gradually become ineffective. The parameter adaptive mechanism of this solution gives the system long-term evolution capabilities. Its specific effect is that the system can automatically discover which factors in the current load assessment model are overestimated or underestimated, and automatically correct them. For example, with the arrival of summer, the weight of the impact of ambient temperature on node stability may be automatically increased by the system. This ensures that the scheduling strategy can keep pace with the times, maintain high efficiency in the long term, eliminate the need for frequent manual parameter readjustment, and significantly reduce operation and maintenance costs.

[0191] The foregoing has shown and described the basic principles, main features, and advantages of this application. Those skilled in the art should understand that this application is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this application. Various changes and modifications can be made to this application without departing from the spirit and scope thereof, and all such changes and modifications fall within the scope of this application as claimed. The scope of protection of this application is defined by the appended claims and their equivalents.

Claims

1. A method for adaptive scheduling of communication load at the edge nodes of a smart light pole, characterized in that, The method includes: In the smart light pole network, the current, voltage, processing queue length, communication latency, packet loss rate, and ambient light, wind speed, and temperature of each edge node are acquired to form a node operation status dataset. Based on the node running status dataset, the real-time communication load of each edge node is calculated, and the load value of each node is associated with an identifier indicating whether it exceeds a preset threshold and stored to form a load status mapping table. Based on the load status mapping table, nodes marked as exceeding the threshold are selected to form a high-load node set, and the communication tasks of the high-load nodes are decomposed into transferable data fragments to generate a scheduling candidate task list. Based on the scheduling candidate task list and combined with the inter-node communication link status, target low-load nodes are matched for each migrateable data segment in the list. According to the matching results, the migrateable data segments are forwarded to their respective target low-load nodes, and task migration execution logs are generated. The communication latency, packet loss rate, and processing queue length of each node are collected during the migration process. The collected results are combined with the task migration execution log to form load adjustment data. The load adjustment data is used to adjust the transmission order and migration path of incomplete tasks. Based on the load adjustment data, scheduling execution instructions are issued to high-load nodes and receiving nodes to perform task migration and load distribution for communication tasks, and generate a final node load status report. Based on the final node load status report, the calculation parameters for subsequent scheduling processes are adaptively adjusted.

2. The method according to claim 1, characterized in that, The process involves acquiring the current, voltage, processing queue length, communication latency, packet loss rate, and ambient light, wind speed, and temperature of each edge node to form a node operating status dataset, including: The raw electrical, network, and environmental data of the collected nodes are used to form a multi-source raw dataset, which includes raw electrical data, raw queue data, raw network data, and raw environmental data. For the raw electrical data in the multi-source raw dataset, adaptive filtering and anomaly suppression processing are performed in combination with the node power fluctuation characteristics and sampling period characteristics to eliminate the influence of instantaneous interference and random noise, and obtain clean electrical parameters that can stably characterize the changing trend of node power supply status. Using the cleanroom electrical parameters as a time reference, timestamp reconstruction and alignment processing are performed on the original queue data in the multi-source raw dataset. The load status of the control node and the power supply status of the node are established in a corresponding relationship under a unified time axis, generating time-synchronized node core operation data that reflects the core operation status of the node. The core operating data of the time synchronization node is fused with the original network data and original environmental data in the multi-source raw dataset across domains based on the unique identifier of the node, and a node-level multi-dimensional state description structure is constructed to associate and encapsulate the internal operating state of the node with the external environmental state. The node-level multidimensional state description structure, after being associated and encapsulated, is output as a node running state dataset for subsequent scheduling and analysis.

3. The method according to claim 2, characterized in that, After outputting the node running status dataset, the following is also included: Historical light intensity data and historical temperature data are extracted from the node operating status dataset, and real-time light intensity value and real-time temperature value are extracted at the current moment. Statistical modeling is performed based on the temporal distribution characteristics of the historical light intensity data to calculate the light intensity benchmark value that characterizes the stability level of the node environment, and the corresponding temperature benchmark value is calculated based on the periodic fluctuation characteristics of the historical temperature data. Using the aforementioned light intensity reference value and temperature reference value as references, the deviation of the real-time light intensity value and real-time temperature value is calculated respectively to obtain the light disturbance value and temperature disturbance value that characterize the degree of instantaneous change in the environment; The illumination disturbance value and temperature disturbance value are jointly modeled, and a single environmental disturbance quantification parameter is generated through weighted fusion and scale unification processing. The environmental disturbance quantification parameters are written into the node operating status dataset as environmental disturbance fields, and the fields are associated with the electrical and network parameters of the corresponding nodes to form an updated node operating status dataset.

4. The method according to claim 1, characterized in that, The calculation of the real-time communication load of each edge node based on the node operating status dataset includes: A subset of operating parameters for load calculation is extracted concurrently based on node identifier. The subset of operating parameters includes at least a subset of electrical parameters characterizing the node's power supply status, a subset of network parameters characterizing the communication status, and an environmental disturbance field. Based on the electrical parameter subset in the aforementioned operating parameter subset, the power consumption value of the node in the current scheduling cycle is calculated according to the node power supply model, and the network congestion value reflecting the degree of communication congestion is calculated based on the network parameter subset in the aforementioned operating parameter subset. Normalization and scaling processing are performed on the power consumption value and network congestion value respectively, converting them into comparable standardized power consumption factor and standardized congestion factor; Based on the preset weighting relationship, a weighted fusion calculation is performed on the standardized power consumption factor, the standardized congestion factor, and the environmental disturbance field. The calculation result is used as the real-time communication load of the corresponding edge node and written into the load field of the node running status dataset.

5. The method according to claim 4, characterized in that, Based on the node operating status dataset, the instant communication load of each edge node is calculated, and the load value of each node is associated with an identifier indicating whether it exceeds a preset threshold, forming a load status mapping table, including: The real-time communication load of each edge node is aggregated to form a full network load set covering the entire smart light pole network; Statistical analysis is performed based on the entire network load set to calculate the average load value reflecting the overall load level and the load fluctuation range value characterizing the degree of load dispersion. Based on the average load value of the entire network, and combined with the load fluctuation range value, a high load judgment threshold for identifying nodes with abnormal load is dynamically calculated and generated. The instant messaging load value of each node is compared with the high load threshold one by one to filter out nodes that exceed the high load threshold. Based on the range in which the instant communication load value of each node exceeds the high load determination threshold, the node load status is classified and processed accordingly. The node identifier, instant messaging load value, and corresponding load classification results are structurally associated to generate a load status mapping table.

6. The method according to claim 1, characterized in that, The process involves selecting nodes identified as exceeding a threshold based on the load status mapping table to form a high-load node set, and decomposing the communication tasks of the high-load nodes into transferable data fragments to generate a scheduling candidate task list, including: Based on the load classification results in the load status mapping table, the nodes in the high load classification range are determined, and the communication task information to be processed in the high load nodes is extracted. The communication task information is parsed to identify the task type of each communication task, and the corresponding task decomposition strategy is matched according to the task type. Based on the task decomposition strategy, the communication task is divided into multiple task units with associated identifiers that can be processed in parallel. Each task unit is encapsulated with processing control information including sequence number, task identifier, data size, and expected processing time calculated based on data size and task type, forming a transferable data fragment; Priority is calculated for the transferable data segments based on the expected processing time, and the transferable data segments carrying priority and processing control information are aggregated and sorted to generate a list of scheduling candidate tasks.

7. The method according to claim 6, characterized in that, The method further includes: The processing control information of each transferable data segment in the scheduling candidate task list is parsed to extract the corresponding task identifier; Based on the task identifier, a preset task dependency database is queried to obtain the original dependency description describing the sequential constraints between tasks; Based on the original dependency description, a task dependency graph representing the execution constraints between transferable data fragments is constructed; Perform topological analysis on the task dependency graph to deduce the global execution order of each transferable data segment under the condition of satisfying dependency constraints; The global execution order obtained from the deduction is written into the scheduling candidate task list, and the transferable data fragments are reordered to generate an updated scheduling candidate task list.

8. The method according to claim 7, characterized in that, The step of matching target low-load nodes for each migrateable data segment in the list based on the scheduling candidate task list and in conjunction with the inter-node communication link status, and forwarding the migrateable data segments to their respective target low-load nodes according to the matching results, includes: Based on the current smart light pole network topology, link quality data characterizing the communication quality between nodes and resource status data of candidate low-load nodes are acquired concurrently. Using the link quality data, a corresponding transmission efficiency score is calculated for the communication path from each high-load node to each candidate low-load node. Combining the resource status data and the features of transferable data segments extracted from the scheduling candidate task list, a matching score for the processing capability of the candidate low-load node for the data segment is calculated. The transmission efficiency score and processing capability matching score corresponding to the same candidate low-load node are fused to generate a comprehensive forwarding evaluation result for that candidate low-load node. Based on the comprehensive forwarding evaluation results, a corresponding target forwarding node is selected for each migrated data segment.

9. The method according to claim 1, characterized in that, The step of issuing scheduling execution instructions to high-load nodes and receiving nodes based on the load adjustment data includes: The load adjustment data is parsed to determine the new target node address and the updated sending order corresponding to the migrated data segment to be adjusted. Based on the new target node address and the updated transmission order, scheduling control instructions are generated for the high-load node and the receiving node, respectively. Send scheduling control instructions to high-load nodes to control them to forward migrateable data fragments to the corresponding new target node addresses according to the updated sending order; Send a scheduling control instruction to the receiving node corresponding to the new target node address to trigger the reception and processing of the migrateable data segment; The execution status information of the scheduling control instructions is collected and written into the task migration execution log.

10. The method according to claim 1, characterized in that, The adaptive feedback adjustment of calculation parameters for subsequent scheduling processes based on the final node load status report includes: Extract the real-time communication load values ​​of each edge node after scheduling is completed from the final node load status report; A load balancing index reflecting the current scheduling result distribution characteristics is calculated based on the instant messaging load values. The load balancing index is compared and analyzed with the pre-stored historical load balancing index to generate scheduling deviation analysis results. Based on the scheduling deviation analysis results, the weight parameters used in the instant messaging load calculation process are updated; The updated weight parameters are applied to the load calculation process in the next scheduling cycle.

Images