Internet of Things card sampling priority scheduling system driven by behavior analysis

Abstract

The invention relates to the technical field of Internet of Things, in particular to a behavior analysis-driven Internet of Things card sampling priority scheduling system, which comprises a behavior analysis unit, a sampling priority evaluation unit, a wireless local area network resource sensing unit, a scheduling execution unit and an upper computer interaction unit. According to the method, rapid adaptive modeling, accurate anomaly detection and dynamic risk assessment of the behavior of the Internet of Things card are realized through a meta-learning and variational auto-encoder hybrid architecture, a reliable basis is provided for priority scheduling, and a multi-objective optimization function is dynamically balanced based on reinforcement learning, so that the reliability of the behavior of the Internet of Things card is improved. The timeliness of the high-priority equipment is guaranteed while the fairness of the low-priority equipment is maintained, and elastic resources are reserved in combination with behavior prediction, so that the intelligence, precision and reliability of sampling and scheduling of the Internet of Things card are improved, and diversified requirements in a complex network environment are effectively met.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Things, and in particular to an Internet of Things card sampling priority scheduling system driven by behavior analysis. Background Art

[0002] An IoT card is an embedded SIM card or data card designed specifically for IoT devices. It's primarily used to enable wireless data communication between IoT devices and between devices and cloud platforms, serving as the core identity identifier and communication carrier for IoT terminals accessing the network. With the rapid development of IoT technology, IoT cards have been widely used in fields such as industrial control, smart homes, and environmental monitoring. The number of connected devices has exploded, and the service requirements (such as real-time performance and data criticality) and communication characteristics (such as sampling frequency and data volume) of different devices vary significantly. Conventional scheduling systems are unable to comprehensively consider device behavior characteristics, business rules, and the real-time status of the network. This leads to irrational resource allocation, insecurity in the timeliness of high-priority services, and difficulty balancing fairness for low-priority services. Furthermore, there is a lack of effective detection and analysis methods for abnormal IoT card behavior.

[0003] Based on this, the present invention provides a behavior analysis-driven IoT card sampling priority scheduling system to solve the above-mentioned technical problems. Summary of the Invention

[0004] The purpose of the present invention is to provide an Internet of Things card sampling priority scheduling system driven by behavior analysis. The meta-learning and variational autoencoder hybrid architecture of the present invention realizes rapid adaptive modeling, accurate anomaly detection and dynamic risk assessment of Internet of Things card behavior, provides a reliable basis for priority scheduling, and dynamically balances multi-objective optimization functions based on reinforcement learning. While ensuring the timeliness of high-priority devices, it maintains the fairness of low-priority devices, and reserves elastic resources in combination with behavior prediction, thereby improving the intelligence, precision and reliability of Internet of Things card sampling scheduling, and effectively responding to diverse needs in complex network environments.

[0005] To achieve the above object, the present invention provides the following technical solutions: The present invention provides a behavior analysis-driven IoT card sampling priority scheduling system, comprising a behavior analysis unit, a sampling priority evaluation unit, a wireless local area network resource perception unit, a scheduling execution unit, and a host computer interaction unit, wherein: The behavior analysis unit: through a hybrid architecture of meta-learning and variational autoencoders, combined with hierarchical feature extraction, lightweight online updates and interpretable analysis, it performs rapid adaptive modeling, anomaly detection and dynamic evaluation of IoT card behavior; The sampling priority evaluation unit is used to combine the behavior analysis results with the preset business rules and use a weighted scoring algorithm to perform multi-dimensional priority quantitative sorting on the sampling requirements of each IoT card; The wireless local area network resource sensing unit is used to monitor the signal strength, channel quality, bandwidth occupancy and network delay status of the wireless local area network in real time; The scheduling execution unit is used to adaptively schedule the timeliness of high-priority devices and the fairness of low-priority devices based on priority sorting and network resource status, through reinforcement learning and dynamic balancing of multi-objective optimization functions, and to reserve elastic resources to cope with sudden demands based on behavior prediction results; The host computer interaction unit is used to receive host computer configuration instructions through a standardized interface protocol and provide real-time feedback on system operation status and scheduling execution logs.

[0006] The behavior analysis unit includes a hybrid architecture modeling module, a hierarchical feature extraction module, a lightweight update module, an interpretability analysis module, and an anomaly detection and evaluation module, wherein: The hybrid architecture modeling module: based on the hybrid architecture of meta-learning and variational autoencoder, performs fast adaptive basic modeling of IoT card behavior; The hierarchical feature extraction module is used to extract multi-dimensional features of IoT card behavior data through hierarchical processing; The lightweight update module is used to dynamically update the behavior model online using a lightweight algorithm; The explainability analysis module is used to perform explainability analysis on the output results of the behavior model to clarify the causes of abnormal behavior; The anomaly detection and assessment module identifies abnormal behavior of IoT cards based on modeling results and dynamically assesses the risk level of the behavior.

[0007] The hybrid architecture modeling module is based on a hybrid architecture of meta-learning and variational autoencoders to perform fast adaptive basic modeling of IoT card behavior. The specific operations are as follows: A1: Construct multiple meta-training tasks, each of which samples support sets and query sets from historical behavior data of different categories of IoT cards; A2: Build a basic encoding-decoding structure based on the variational autoencoder. Its loss function includes the reconstruction error term and the KL divergence regularization term: , Where x is the input behavior sequence, To reconstruct the output, is the reconstruction error term, is a hidden variable, is the KL divergence, β is the regularization weight; A3: A model-independent meta-learning algorithm is used to optimize the initial parameters. After a small number of gradient updates, the model can achieve good performance on new tasks. The meta-update goal is: , in, , which means minimizing the task loss on the query set after updating the parameters once on the support set; A4: Deploy the trained general prior model to the edge.

[0008] The interpretability analysis module performs interpretability analysis on the output results of the behavior model to clarify the causes of abnormal behavior. The specific operations are as follows: B1: receives the abnormality judgment result output by the behavior model and its corresponding original input behavior feature vector; B2: Use the feature attribution algorithm to calculate the contribution score of each input feature to the anomaly judgment result; B3: Generate a visual explanation report based on the contribution score, marking the key behavioral characteristics that lead to abnormal judgments and their impact direction; B4: Associate the interpretation results with the device meta-information and output a readability diagnosis statement.

[0009] The sampling priority evaluation unit includes an analysis result adaptation module, a business rule management module, a weighted score calculation module, and a priority sorting module, wherein: The analysis result adaptation module is used to convert the output results of the behavior analysis unit into basic parameters for priority evaluation; The business rule management module is used to store, call and dynamically update preset business rules; The weighted scoring calculation module is used to quantitatively calculate the multi-dimensional indicators of behavioral risks and business needs using a weighted scoring algorithm; The priority sorting module is used to prioritize the sampling requirements of each IoT card according to the scoring results and output a sorted list.

[0010] The wireless local area network resource sensing unit includes a signal strength monitoring module, a channel quality analysis module, a bandwidth occupancy statistics module, a network delay measurement module, and a resource status integration module, wherein: The signal strength monitoring module is used to collect the signal strength data of each IoT card in the wireless local area network in real time and evaluate the connection stability; The channel quality analysis module is used to monitor channel interference and signal-to-noise ratio parameters and analyze the transmission quality of the wireless local area network channel; The bandwidth occupancy statistics module is used to count the bandwidth usage of the current wireless local area network and calculate the remaining available bandwidth; The network delay measurement module is used to measure the communication delay between each device and the network node in real time and evaluate the transmission timeliness; The resource status integration module is used to summarize the above monitoring data and generate a comprehensive report on the wireless local area network resource status.

[0011] The scheduling execution unit includes a multi-objective optimization module, a reinforcement learning decision module, a resource allocation execution module, a fairness adjustment module, and a flexible resource reservation module, wherein: The multi-objective optimization module is used to construct a multi-objective optimization function including priority satisfaction, fairness, and resource utilization; The reinforcement learning decision module is used to dynamically solve the optimization function through the reinforcement learning algorithm and generate an adaptive scheduling strategy; The resource allocation execution module is used to allocate sampling time slots and bandwidth to devices of different priorities according to the scheduling strategy; The fairness adjustment module is used to dynamically balance resource allocation and ensure the minimum service quota of low-priority devices; The elastic resource reservation module is used to reserve elastic resources in combination with the prediction results of the behavior analysis unit.

[0012] The reinforcement learning decision module dynamically solves the optimization function through the reinforcement learning algorithm to generate an adaptive scheduling strategy. The specific operations are as follows: C1: Constructs a state space, including the current sampling priority of each IoT card, wireless LAN channel quality level, bandwidth utilization, and waiting delay of high-priority devices; C2: Defines the action space as a combination of time slot allocations for devices of different priorities, where each action represents a set of scheduling resource configuration instructions; C3: Design a multi-objective reward function R to evaluate the comprehensive performance of the scheduling strategy: , in, The sampling time rate of high priority devices, is the fairness score for low priority devices, is the resource over-allocation penalty item, is the preset weight coefficient; C4: Use deep reinforcement learning algorithms to train the scheduling strategy model, select the optimal action based on the real-time status, and generate an adaptive scheduling strategy.

[0013] The elastic resource reservation module reserves elastic resources in combination with the prediction results of the behavior analysis unit. The specific operations are as follows: D1: Receives the IoT card behavior risk level and communication activity prediction value in the future time window output by the behavior analysis unit; D2: Determine the target device set for which sampling resources need to be reserved based on risk level and predicted activity; D3: Based on risk level Calculate the resource reservation ratio of the i-th device, satisfying: , in, Score device behavior risk, The proportion of pre-allocated bandwidth or time slots, γ is the control coefficient of the total amount of reserved resources in the system; D4: Allocate calculated resources to high-risk devices before the scheduling period begins and release unused resources after the period ends.

[0014] The host computer interaction unit includes an instruction parsing module, a status reporting module, and an interface protocol adaptation module, wherein: The instruction parsing module is used to receive control instructions for configuration changes and policy updates issued by the host computer; The status reporting module is used to upload system operation indicators, device status and scheduling logs periodically or in an event-triggered manner; The interface protocol adaptation module is used to perform data interaction through the standardized REST API protocol.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention realizes rapid adaptive modeling, accurate anomaly detection and dynamic risk assessment of IoT card behavior through a hybrid architecture of meta-learning and variational autoencoder, providing a reliable basis for priority scheduling. Based on the dynamic balance multi-objective optimization function of reinforcement learning, it maintains the fairness of low-priority devices while ensuring the timeliness of high-priority devices, and reserves elastic resources in combination with behavior prediction, thereby improving the intelligence, precision and reliability of IoT card sampling scheduling, and effectively responding to diverse needs in complex network environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a system diagram of a behavior analysis-driven IoT card sampling priority scheduling system of the present invention.

[0017] Figure 2 This is a workflow diagram of a behavior analysis unit in a behavior analysis-driven IoT card sampling priority scheduling system of the present invention.

[0018] Figure 3 This is a priority evaluation flow chart in a behavior analysis-driven IoT card sampling priority scheduling system of the present invention.

[0019] Description of Figure Numbers: 100, Behavior Analysis Unit; 101, Hybrid Architecture Modeling Module; 102, Hierarchical Feature Extraction Module; 103, Lightweight Update Module; 104, Explainability Analysis Module; 105, Anomaly Detection and Evaluation Module; 200, Sampling Priority Evaluation Unit; 201, Analysis Result Adaptation Module; 202, Business Rule Management Module; 203, Weighted Score Calculation Module; 204, Priority Sorting Module; 300, Wireless LAN Resource Perception Unit; 301, Signal Strength Monitoring Module; 3 02. Channel quality analysis module; 303. Bandwidth occupancy statistics module; 304. Network delay measurement module; 305. Resource status integration module; 400. Scheduling execution unit; 401. Multi-objective optimization module; 402. Reinforcement learning decision module; 403. Resource allocation execution module; 404. Fairness adjustment module; 405. Elastic resource reservation module; 500. Host computer interaction unit; 501. Instruction parsing module; 502. Status reporting module; 503. Interface protocol adaptation module. DETAILED DESCRIPTION

[0020] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0021] Example: like Figure 1-Figure 3As shown, this embodiment provides a behavior analysis driven IoT card sampling priority scheduling system, including a behavior analysis unit 100, a sampling priority evaluation unit 200, a wireless local area network resource perception unit 300, a scheduling execution unit 400, and a host computer interaction unit 500, wherein: the behavior analysis unit 100: through a hybrid architecture of meta-learning and variational autoencoder, combined with hierarchical feature extraction, lightweight online update and interpretability analysis, to perform rapid adaptive modeling, anomaly detection and dynamic evaluation of IoT card behavior; the sampling priority evaluation unit 200: for combining the behavior analysis results with preset business rules, using a weighted scoring algorithm to evaluate the behavior of each IoT card; The sampling requirements of the network card are quantitatively prioritized in multiple dimensions; the wireless LAN resource perception unit 300 is used to monitor the signal strength, channel quality, bandwidth occupancy and network delay status of the wireless LAN in real time; the scheduling execution unit 400 is used to adaptively schedule the timeliness guarantee of high-priority devices and the fairness maintenance of low-priority devices based on priority sorting and network resource status through reinforcement learning dynamic balancing multi-objective optimization function, and at the same time reserve elastic resources to cope with sudden demands based on behavior prediction results; the host computer interaction unit 500 is used to receive host computer configuration instructions through a standardized interface protocol, and provide real-time feedback on the system operation status and scheduling execution log.

[0022] Among them, it should be noted that the behavior analysis unit 100 models and detects anomalies in the behavior of the Internet of Things card, and its output is passed to the sampling priority evaluation unit 200 for multi-dimensional quantitative sorting. At the same time, the wireless local area network resource perception unit 300 monitors the network status in real time. The data of the two are jointly input into the scheduling execution unit 400, and the reinforcement learning algorithm dynamically optimizes resource allocation and reserves elastic resources, and finally realizes human-computer collaborative control through the host computer interaction unit 500.

[0023] In this embodiment, it should also be noted that the behavior analysis unit 100 includes a hybrid architecture modeling module 101, a hierarchical feature extraction module 102, a lightweight update module 103, an interpretability analysis module 104, and an anomaly detection and evaluation module 105, wherein: the hybrid architecture modeling module 101: based on a hybrid architecture of meta-learning and variational autoencoder, performs fast adaptive basic modeling of IoT card behavior; the specific operations are as follows: A1: constructs multiple meta-training tasks, each task samples support sets and query sets from historical behavior data of IoT cards of different categories; A2: constructs a basic encoding-decoding structure based on the variational autoencoder, and its loss function includes a reconstruction error term and a KL divergence regularization term: , Where x is the input behavior sequence, To reconstruct the output, is the reconstruction error term, is a hidden variable, is the KL divergence, and β is the regularization weight. A3: A model-independent meta-learning algorithm is used to optimize the initial parameters. After a small amount of gradient updates, the model can achieve good performance on the new task. The meta-update goal is: , in, , which represents minimizing the task loss on the query set after updating the parameters once on the support set. A4: Deploys the trained general prior model to the edge. The hierarchical feature extraction module 102 extracts multidimensional features from IoT card behavior data through hierarchical processing. The lightweight update module 103 dynamically updates the behavior model online using a lightweight algorithm. The interpretability analysis module 104 performs interpretability analysis on the behavior model output to identify the causes of abnormal behavior. The specific operations are as follows: B1: Receives the anomaly determination result output by the behavior model and its corresponding original input behavior feature vector. B2: Uses a feature attribution algorithm to calculate the contribution score of each input feature to the anomaly determination result. B3: Generates a visual explanation report based on the contribution score, annotating the key behavioral features that led to the anomaly determination and their impact direction. B4: Associates the explanation result with device metadata and outputs a readable diagnostic statement. The anomaly detection and assessment module 105 identifies abnormal IoT card behavior based on the modeling results and dynamically assesses the behavioral risk level.

[0024] Among them, it should be noted that the hybrid architecture modeling module 101 constructs a hybrid model of meta-learning and variational autoencoder, the hierarchical feature extraction module 102 extracts multi-dimensional behavioral features, the lightweight update module 103 realizes online model optimization, the interpretability analysis module 104 parses the model decision logic and outputs a visual diagnostic report, and finally the anomaly detection and evaluation module 105 completes the dynamic assessment of risk level.

[0025] Furthermore, it should be noted that the different categories of IoT cards in A1 specifically refer to types categorized by application scenarios, including industrial control (e.g., production line sensors), smart home (e.g., temperature and humidity sensors), and environmental monitoring (e.g., PM2.5 monitors). Each card type must contain at least 1,000 pieces of historical behavioral data (each piece of data includes 10+ dimensional features such as timestamp, transmission rate, and connection status). The "support set and query set" are divided in a 7:3 ratio, with the support set used for model initialization training and the query set used for meta-learning adaptability verification. The hierarchical processing approach in hierarchical feature extraction module 102 is as follows: at the physical layer, signal strength, transmission frequency, and signal-to-noise ratio are extracted; at the data link layer, packet size, retransmission rate, and frame error rate are extracted; and at the application layer, sampling interval, data type, and service identifier are extracted. Lightweight update module 103 uses the elastic weight integration algorithm from incremental learning to update only the parameters of the model's output layer and the last fully connected layer. Update trigger conditions include: ① 50 newly collected behavioral data items have accumulated; ② the model's reconstruction error for the new data exceeds 15% for three consecutive times; and ③ the host computer proactively issues an update command. The specific criteria for identifying abnormal behavior include "Euclidean distance from the historical behavior baseline > threshold σ, or reconstruction error > 95th percentile," as well as a quantitative standard for "dynamically assessing behavioral risk level": Low Risk (Level 1): Abnormal behavior lasts <5 minutes, affecting only a single device with no service interruption; Medium Risk (Level 2): ​​Abnormal behavior lasts 5-30 minutes, affecting 1-5 devices and causing non-core service delays; High Risk (Level 3): Abnormal behavior lasts 30 minutes to 2 hours, affecting 5-20 devices and impacting core services; Critical Risk (Level 4): Abnormal behavior lasts >2 hours, affecting >20 devices and potentially causing service interruption.

[0026] In this embodiment, it should also be noted that the sampling priority evaluation unit 200 includes an analysis result adaptation module 201, a business rule management module 202, a weighted score calculation module 203, and a priority sorting module 204, wherein: the analysis result adaptation module 201 is used to convert the output results of the behavior analysis unit 100 into basic parameters for priority evaluation; the business rule management module 202 is used to store, call and dynamically update preset business rules; the weighted score calculation module 203 is used to use a weighted scoring algorithm to quantitatively calculate multi-dimensional indicators of behavioral risks and business needs; the priority sorting module 204 is used to prioritize the sampling requirements of each IoT card according to the scoring results and output a sorted list.

[0027] Among them, it should be noted that the analysis result adaptation module 201 converts the behavior analysis data into evaluation parameters, combines the dynamic business rules called by the business rule management module 202, and the weighted score calculation module 203 performs quantitative weighted calculation on the multi-dimensional indicators, and finally forms a device sampling priority ranking list through the priority sorting module 204.

[0028] Furthermore, it should be noted that the parameter conversion rules of the analysis result adaptation module 201 are as follows: the core data output by the received behavior analysis unit 100 include: device behavior risk level (level 1-4, corresponding to low to emergency risk), anomaly confidence (0-100%), and behavior fluctuation coefficient (0-1, measuring behavior stability); conversion logic: map the risk level to 10-40 points (level 1 is 10 points, and each level up adds 10 points), the anomaly confidence is converted to 0-20 points according to "confidence × 0.2", and the behavior fluctuation coefficient is converted to 0-10 points according to "(1-fluctuation coefficient) × 10", and finally merged into a "behavioral feature parameter" of 0-70 points as the basic input for priority evaluation. The evaluation dimensions in the weighted score calculation module 203 include: behavioral characteristic parameters (40%), business rule parameters (30%), and historical scheduling performance (30%, such as the sampling completion rate over the past 24 hours). The weighted scoring formula is: Final score = 0.4 × behavioral characteristic parameters + 0.3 × business rule score + 0.3 × historical performance score. The business rule score is calculated as "business rule weight × corresponding dimension score" (for example, the real-time score for industrial control = actual response time / standard response time × 100). The score ranges from 0 to 100 and is divided into five priority levels at 20-point intervals (100-80 is the highest, and so on). The output "priority ranking list" includes: device ID, priority level, final score, and key influencing factors (such as "behavioral risk 40 points + business importance 25 points"). It is synchronized to the scheduling execution unit 400 in JSON format and archived through the host computer interaction unit 500.

[0029] In this embodiment, it should also be noted that the wireless local area network resource perception unit 300 includes a signal strength monitoring module 301, a channel quality analysis module 302, a bandwidth occupancy statistics module 303, a network delay measurement module 304, and a resource status integration module 305, wherein: the signal strength monitoring module 301 is used to collect signal strength data of each Internet of Things card in the wireless local area network in real time and evaluate the connection stability; the channel quality analysis module 302 is used to monitor channel interference and signal-to-noise ratio parameters and analyze the transmission quality of the wireless local area network channel; the bandwidth occupancy statistics module 303 is used to count the bandwidth usage of the current wireless local area network and calculate the remaining available bandwidth; the network delay measurement module 304 is used to measure the communication delay between each device and the network node in real time and evaluate the transmission timeliness; the resource status integration module 305 is used to summarize the above monitoring data and generate a comprehensive report on the resource status of the wireless local area network.

[0030] Among them, it should be noted that the signal strength monitoring module 301 evaluates the stability of the device connection, the channel quality analysis module 302 detects the channel transmission quality, the bandwidth occupancy statistics module 303 calculates the available bandwidth resources, and the network delay measurement module 304 analyzes the transmission timeliness. Finally, the resource status integration module 305 generates a multi-dimensional resource status comprehensive report including signal strength, channel quality, bandwidth occupancy and network delay.

[0031] Furthermore, it should be noted that the connection stability assessment rules in the signal strength monitoring module 301 are as follows: Stable: signal strength ≥ -65dBm, with fluctuation ≤ 5dBm within 1 minute; Fair: -75dBm ≤ signal strength < -65dBm, or fluctuation 5-10dBm; Unstable: signal strength < -75dBm, or fluctuation > 10dBm (triggering an alert and marking "Connection Optimization Required"). Exception handling: If no signal is acquired for three consecutive times (signal strength = -120dBm), the connection is considered "offline" and the result is pushed to the host computer in real time. The transmission timeliness grading in the network delay measurement module 304 is as follows: Excellent: RTT < 50ms (meets the needs of real-time control services); Good: 50-100ms (meets the needs of video streaming and high-frequency sampling); Acceptable: 100-300ms (meets the needs of periodic data reporting); Poor: > 300ms (impacts the service experience and triggers delay cause analysis, such as channel congestion or equipment failure).

[0032] In this embodiment, it should also be noted that the scheduling execution unit 400 includes a multi-objective optimization module 401, a reinforcement learning decision module 402, a resource allocation execution module 403, a fairness adjustment module 404, and a flexible resource reservation module 405, wherein: the multi-objective optimization module 401 is used to construct a multi-objective optimization function including priority satisfaction, fairness, and resource utilization; the reinforcement learning decision module 402 is used to dynamically solve the optimization function through a reinforcement learning algorithm to generate an adaptive scheduling strategy; the specific operations are as follows: C1: Construct a state space, including the current sampling priority of each IoT card, the wireless LAN channel quality level, the bandwidth occupancy rate, and the waiting delay of high-priority devices; C2: Define the action space as a time slot allocation combination of devices of different priorities, and each action represents a set of scheduling resource configuration instructions; C3: Design a multi-objective reward function R to evaluate the comprehensive performance of the scheduling strategy: , in, The sampling time rate of high priority devices, is the fairness score for low priority devices, is the resource over-allocation penalty item, is a preset weight coefficient; C4: uses a deep reinforcement learning algorithm to train the scheduling strategy model, and selects the optimal action according to the real-time status to generate an adaptive scheduling strategy. Resource allocation execution module 403: is used to allocate sampling time slots and bandwidth to devices of different priorities according to the scheduling strategy; fairness adjustment module 404: is used to dynamically balance resource allocation and ensure the minimum service quota of low-priority devices; elastic resource reservation module 405: is used to reserve elastic resources in combination with the prediction results of the behavior analysis unit 100. The specific operations are as follows: D1: Receive the behavior risk level of the Internet of Things card output by the behavior analysis unit 100 and the predicted value of communication activity in the future time window; D2: Determine the target device set for which sampling resources need to be reserved based on the risk level and predicted activity; D3: Based on the risk level Calculate the resource reservation ratio of the i-th device, satisfying: , in, Score device behavior risk, The proportion of pre-allocated bandwidth or time slots, γ is the control coefficient of the total amount of reserved resources in the system; D4: Before the start of the scheduling cycle, the calculated resources are allocated to high-risk devices, and unused resources are released after the cycle ends.

[0033] Among them, it should be noted that the multi-objective optimization module 401 constructs the optimization function, the reinforcement learning decision module 402 dynamically generates the scheduling strategy, the resource allocation execution module 403 implements differentiated resource allocation, the fairness adjustment module 404 ensures service fairness, and the elastic resource reservation module 405 reserves resources based on prediction.

[0034] Furthermore, it should be noted that the minimum service quota: low-priority devices obtain at least 10% of the total time slots and 10% of the total bandwidth (configurable), and if the quota is not reached for three consecutive cycles, "fairness compensation" is automatically triggered (the quota for the next cycle is temporarily increased to 15%); dynamic balancing strategy: when the demand for high-priority devices decreases (such as sampling is completed), the released resources are allocated according to "low-priority device waiting time × historical quota deficit" to avoid "starvation"; quota monitoring: after the end of each cycle, the "actual quota / theoretical quota" ratio of low-priority devices is calculated. If it is less than 0.8 for five consecutive times, a "fairness alarm" is pushed to the upper computer.

[0035] In this embodiment, it should also be noted that the host computer interaction unit 500 includes an instruction parsing module 501, a status reporting module 502, and an interface protocol adaptation module 503, wherein: the instruction parsing module 501 is used to receive control instructions for configuration changes and policy updates issued by the host computer; the status reporting module 502 is used to periodically or event-triggeredly upload system operation indicators, device status and scheduling logs; the interface protocol adaptation module 503 is used to interact with data through the standardized REST API protocol.

[0036] It should be noted that the instruction parsing module 501 receives and parses control instructions, the status reporting module 502 actively feeds back the system operation status, and the interface protocol adaptation module 503 standardizes the data interaction protocol.

[0037] Furthermore, it should be noted that periodic reporting (default 5 minutes) includes: system operation indicators such as CPU utilization, memory usage, module load, and scheduling success rate; global network status such as wireless LAN health score, average latency of high-priority devices, and bandwidth utilization trend (over the past 10 minutes); event-triggered reporting (real-time ≤ 1s): abnormal events such as device offline (three consecutive no-signal events), scheduling failure (resource allocation conflict), and module failure (such as reinforcement learning model loading failure); key operations such as successful policy update, triggering of elastic resource reservation, and fairness alerts (consecutive failure to meet low-priority quotas).

[0038] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0039] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A behavior analysis driven IoT card sampling priority scheduling system, characterized in that: The system comprises a behavior analysis unit (100), a sampling priority evaluation unit (200), a wireless local area network resource perception unit (300), a scheduling execution unit (400), and a host computer interaction unit (500), wherein: The behavior analysis unit (100) is configured to perform rapid adaptive modeling, anomaly detection, and dynamic evaluation of IoT card behaviors through a hybrid architecture of meta-learning and variational autoencoders, combined with hierarchical feature extraction, lightweight online updating, and interpretability analysis. The sampling priority evaluation unit (200) is used to combine the behavior analysis results with the preset business rules and use a weighted scoring algorithm to perform multi-dimensional priority quantitative sorting on the sampling requirements of each IoT card; The wireless local area network resource sensing unit (300) is used to monitor the signal strength, channel quality, bandwidth occupancy and network delay status of the wireless local area network in real time; The scheduling execution unit (400) is used to perform adaptive scheduling for timeliness assurance of high-priority devices and fairness maintenance of low-priority devices based on priority sorting and network resource status through reinforcement learning dynamic balancing multi-objective optimization function, and at the same time reserve elastic resources to cope with sudden demands in combination with behavior prediction results; The host computer interaction unit (500) is used to receive host computer configuration instructions through a standardized interface protocol and to provide real-time feedback on system operation status and scheduling execution logs.

2. The behavior analysis driven IoT card sampling priority scheduling system according to claim 1, characterized in that: The behavior analysis unit (100) includes a hybrid architecture modeling module (101), a hierarchical feature extraction module (102), a lightweight update module (103), an interpretability analysis module (104), and an anomaly detection and evaluation module (105), wherein: The hybrid architecture modeling module (101) is based on a hybrid architecture of meta-learning and variational autoencoder to perform fast adaptive basic modeling of IoT card behavior; The hierarchical feature extraction module (102) is used to extract multi-dimensional features of the Internet of Things card behavior data through a hierarchical processing method; The lightweight update module (103) is used to dynamically update the behavior model online using a lightweight algorithm; The explainability analysis module (104) is used to perform explainability analysis on the output results of the behavior model to clarify the causes of abnormal behavior; The anomaly detection and assessment module (105) identifies abnormal behaviors of the IoT card based on the modeling results and dynamically assesses the risk level of the behavior.

3. The behavior analysis driven IoT card sampling priority scheduling system according to claim 2, characterized in that: The hybrid architecture modeling module (101) is based on a hybrid architecture of meta-learning and variational autoencoder to perform fast adaptive basic modeling of the behavior of the IoT card. The specific operations are as follows: A1: Construct multiple meta-training tasks, each of which samples support sets and query sets from historical behavior data of different categories of IoT cards; A2: Build a basic encoding-decoding structure based on the variational autoencoder. Its loss function includes the reconstruction error term and the KL divergence regularization term: ， in, is the input behavior sequence, To reconstruct the output, is the reconstruction error term, is a hidden variable, is the KL divergence, β is the regularization weight; A3: A model-independent meta-learning algorithm is used to optimize the initial parameters. After a small number of gradient updates, the model can achieve good performance on new tasks. The meta-update goal is: ， in, , which means minimizing the task loss on the query set after updating the parameters once on the support set; A4: Deploy the trained general prior model to the edge.

4. The behavior analysis driven IoT card sampling priority scheduling system according to claim 2, characterized in that: The interpretability analysis module (104) performs interpretability analysis on the output results of the behavior model to clarify the causes of abnormal behavior. The specific operations are as follows: B1: receives the abnormality judgment result output by the behavior model and its corresponding original input behavior feature vector; B2: Use the feature attribution algorithm to calculate the contribution score of each input feature to the anomaly judgment result; B3: Generate a visual explanation report based on the contribution score, marking the key behavioral characteristics that lead to abnormal judgments and their impact direction; B4: Associate the interpretation results with the device meta-information and output a readability diagnosis statement.

5. The behavior analysis driven IoT card sampling priority scheduling system according to claim 1, characterized in that: The sampling priority evaluation unit (200) comprises an analysis result adaptation module (201), a business rule management module (202), a weighted score calculation module (203), and a priority sorting module (204), wherein: The analysis result adaptation module (201) is used to convert the output result of the behavior analysis unit (100) into a basic parameter for priority evaluation; The business rule management module (202) is used to store, call and dynamically update preset business rules; The weighted scoring calculation module (203) is used to quantitatively calculate the multi-dimensional indicators of behavioral risks and business needs using a weighted scoring algorithm; The priority sorting module (204) is used to prioritize the sampling requirements of each IoT card according to the scoring results and output a sorting list.

6. The behavior analysis driven IoT card sampling priority scheduling system according to claim 1, characterized in that: The wireless local area network resource sensing unit (300) comprises a signal strength monitoring module (301), a channel quality analysis module (302), a bandwidth occupancy statistics module (303), a network delay measurement module (304), and a resource status integration module (305), wherein: The signal strength monitoring module (301) is used to collect signal strength data of each IoT card in the wireless local area network in real time and evaluate the connection stability; The channel quality analysis module (302) is used to monitor channel interference and signal-to-noise ratio parameters and analyze the transmission quality of the wireless local area network channel; The bandwidth occupancy statistics module (303) is used to count the bandwidth usage of the current wireless local area network and calculate the remaining available bandwidth; The network delay measurement module (304) is used to measure the communication delay between each device and the network node in real time and evaluate the transmission timeliness; The resource status integration module (305) is used to aggregate the above monitoring data and generate a comprehensive report on the resource status of the wireless local area network.

7. The behavior analysis driven IoT card sampling priority scheduling system according to claim 1, characterized in that: The scheduling execution unit (400) includes a multi-objective optimization module (401), a reinforcement learning decision module (402), a resource allocation execution module (403), a fairness adjustment module (404), and a flexible resource reservation module (405), wherein: The multi-objective optimization module (401) is used to construct a multi-objective optimization function including priority satisfaction, fairness, and resource utilization; The reinforcement learning decision module (402) is used to dynamically solve the optimization function through the reinforcement learning algorithm to generate an adaptive scheduling strategy; The resource allocation execution module (403) is used to allocate sampling time slots and bandwidths to devices of different priorities according to a scheduling policy; The fairness adjustment module (404) is used to dynamically balance resource allocation and ensure the minimum service quota of low-priority devices; The elastic resource reservation module (405) is used to reserve elastic resources in combination with the prediction result of the behavior analysis unit (100).

8. The behavior analysis driven IoT card sampling priority scheduling system according to claim 7, characterized in that: The reinforcement learning decision module (402) dynamically solves the optimization function through the reinforcement learning algorithm to generate an adaptive scheduling strategy. The specific operations are as follows: C1: Constructs a state space, including the current sampling priority of each IoT card, wireless LAN channel quality level, bandwidth utilization, and waiting delay of high-priority devices; C2: Defines the action space as a combination of time slot allocations for devices of different priorities, where each action represents a set of scheduling resource configuration instructions; C3: Design a multi-objective reward function R to evaluate the comprehensive performance of the scheduling strategy: ， in, The sampling time rate of high priority devices, is the fairness score for low priority devices, is the resource over-allocation penalty item, is the preset weight coefficient; C4: Use deep reinforcement learning algorithms to train the scheduling strategy model, select the optimal action based on the real-time status, and generate an adaptive scheduling strategy.

9. The behavior analysis driven IoT card sampling priority scheduling system according to claim 7, characterized in that: The elastic resource reservation module (405) reserves elastic resources in combination with the prediction result of the behavior analysis unit (100). The specific operations are as follows: D1: receiving the behavior risk level of the IoT card and the predicted value of communication activity in the future time window output by the behavior analysis unit (100); D2: Determine the target device set for which sampling resources need to be reserved based on risk level and predicted activity; D3: Based on risk level Calculate the resource reservation ratio of the i-th device, satisfying: ， in, Score device behavior risk, The proportion of pre-allocated bandwidth or time slots, γ is the control coefficient of the total amount of reserved resources in the system; D4: Allocate calculated resources to high-risk devices before the scheduling period begins and release unused resources after the period ends.

10. The behavior analysis driven IoT card sampling priority scheduling system according to claim 1, characterized in that: The host computer interaction unit (500) includes an instruction parsing module (501), a status reporting module (502), and an interface protocol adaptation module (503), wherein: The instruction parsing module (501) is used to receive control instructions for configuration changes and policy updates issued by the host computer; The status reporting module (502) is used to upload system operation indicators, device status and scheduling logs periodically or in an event-triggered manner; The interface protocol adaptation module (503) is used for data interaction via a standardized REST API protocol.

Images