Multi-level Internet of Things equipment security management and access control method

Abstract

The invention belongs to the technical field of Internet of Things security, and particularly relates to a multi-level Internet of Things equipment security management and access control method, which comprises the following steps of: performing bidirectional verification on a digital certificate and a hardware fingerprint through an edge node when equipment is accessed; calculating credibility in real time and dynamically dividing security levels by using a hybrid evaluation algorithm based on equipment operation context data; a fine-grained access strategy is generated in combination with equipment attributes and environmental risks, and cross-level collaboration is realized through a lightweight protocol; abnormal behaviors are detected in real time in the access process, and high-risk operation is blocked within milliseconds; and strategy conflict resolution and adaptive optimization are realized through formalized verification and deep reinforcement learning. According to the technical scheme, the identity authentication reliability, the authority dynamic adaptability, the strategy consistency and the threat response speed of the Internet of Things equipment are remarkably improved, and a systematic security guarantee is provided for large-scale Internet of Things applications with high security requirements.

Description

Technical Field

[0001] This invention belongs to the field of Internet of Things (IoT) security technology, specifically a multi-level IoT device security management and access control method. Background Technology

[0002] With the deep application of IoT technology in fields such as smart grids and industrial internet, a large number of heterogeneous devices are connected to the network through a multi-level architecture of edge-gateway-cloud, which poses core requirements for device security management and access control: while ensuring lightweight deployment on the edge side, it is necessary to achieve trusted authentication, dynamic permission adaptation and real-time risk blocking for cross-level access, so as to avoid security vulnerabilities caused by permission generalization or authentication lag.

[0003] Existing technologies mostly rely on static access control lists or centralized authentication protocols, which are difficult to adapt to IoT scenarios with frequent device online / offline cycles and dynamic role changes, and are prone to policy update delays and permission redundancy. Some solutions attempt to introduce attribute-based access control or blockchain-assisted identity management, but they do not fully consider the limitations of edge node computing resources and do not effectively integrate device runtime context (such as geographical location, energy consumption status, and communication frequency), resulting in problems such as excessive resource consumption or insufficient security control granularity.

[0004] The inherent limitations of existing technologies constitute key bottlenecks: First, the lack of a real-time assessment mechanism for device security status makes it impossible to dynamically adjust permissions based on abnormal behavior or environmental risks; second, the fragmented policies across different layers in a multi-level architecture, lacking a unified coordination and conflict resolution mechanism, easily lead to permission overreach or false denial of legitimate access; and third, the difficulty in balancing the need for lightweight edge devices with the need for precise control. These problems restrict the reliable application of large-scale IoT systems in high-security scenarios.

[0005] Therefore, the present invention provides a multi-level Internet of Things (IoT) device security management and access control method. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0007] The technical solution adopted by this invention to solve its technical problem is as follows: A multi-level IoT device security management and access control method, comprising the following specific steps: Step 1: Trusted device identity verification. When a device accesses the network, a lightweight authentication module deployed at the edge node performs bidirectional verification of the device's digital certificate and hardware fingerprint. The hardware fingerprint is jointly generated by the device firmware hash value and the Physically Unclonable Function (PUF), forming a unique identifier with anti-cloning and anti-tampering properties. The verification process adopts a digital signature algorithm based on Elliptic Curve Cryptography (ECC), with a signature length of 256 bits, and introduces a zero-knowledge proof mechanism to complete identity confirmation without revealing the private key, ensuring the device's identity is verified. Step 1: Verify the authenticity, non-repudiation, and privacy protection of the device's identity; Step 2: Dynamic trust assessment and security status awareness. Continuously collect contextual data during device operation, including communication behavior patterns, resource consumption rates, and geographical location information. Calculate the device's real-time trust level using a hybrid assessment algorithm integrating Long Short-Term Memory (LSTM) and Hidden Markov Model (HMM). LSTM is used to model long-term behavioral dependencies, while HMM captures short-term state transitions. The outputs of both are weighted and fused to generate a continuous trust value between 0 and 1, with an update cycle of no more than 5 seconds. This trust level is not only used to classify device security levels but also serves as a core input parameter for subsequent policy generation and session management; Step 3: Cross-level permissions. Policy generation and coordination: Based on the trust level, network layer of the device, and sensitivity of the target resources accessed, as output in step 2, the policy decision point dynamically extracts more than 20 attributes from the distributed attribute library, including device type, environmental risk index, and historical compliance records. It then uses an improved attribute-based access control (ABAC) model to generate fine-grained permission policies. The policy decision point communicates with the policy execution points deployed on the regional gateway and cloud platform via a lightweight message queue telemetry transport protocol (MQTT) for low-latency communication. The protocol uses binary encoding and a "at least once" delivery mechanism based on the Quality of Service (QoS) level to ensure that the policy is synchronized to key nodes across the network within 100 milliseconds and supports policy version rollback and consistency verification. Step 4: Real-time Access control and abnormal behavior blocking: When a device initiates an access request, the policy enforcement point makes an access decision based on the latest permission policy. At the same time, it uses traffic mirroring technology to import the device's communication flow into the behavior analysis engine deployed on the regional gateway. This engine adopts a streaming computing architecture, combining a density-based clustering algorithm (DBSCAN) and an online learning mechanism to dynamically adjust the neighborhood radius and minimum sample number parameters, and detect abnormal traffic that deviates from the historical behavior baseline in real time. For identified high-risk access or abnormal behavior, the policy enforcement point can issue a blocking command to the corresponding network boundary device within 10 milliseconds and simultaneously revoke the current session token. The token validity period is dynamically set according to the device's real-time trust level, ranging from 5 minutes to 24 hours.Step 5: Policy Conflict Resolution and Adaptive Optimization. A unified policy library covering the edge to the cloud is established. Formal verification tools automatically convert access control policies into temporal logic formulas, and model detection algorithms are used to verify the existence of logical conflicts between policies. When a conflict is detected, a multi-dimensional conflict resolution algorithm based on priority, security level, and contextual semantics is used to correct the policy, prioritizing the enforcement of policies in high-security areas. The policy library synchronously records all access control logs, and a deep reinforcement learning model (using the Asynchronous Advantage Actor Critics algorithm, A3C) is trained weekly using over 1 million historical records to maximize system security benefits and minimize the false rejection rate of legitimate access, achieving continuous adaptive optimization of policy generation parameters.

[0008] Preferably, in step 1, the lightweight authentication module is compatible with multiple IoT protocols such as Narrowband Internet of Things (NB-IoT) and Long-Range Wireless Communication (LoRa), with an authentication data packet payload of no more than 200 bytes, a memory footprint of less than 50 kilobytes on a typical microcontroller unit (MCU), an authentication response time of less than 2 seconds, and support for offline emergency authentication mode. Preferably, in step 2, the context data is collected by lightweight agent software on the device side, with an agent size of less than 30 kilobytes. The collected data is processed locally for differential privacy desensitization before being uploaded. The communication behavior pattern analysis covers the number of data packets per second, connection duration, protocol compliance, and port usage distribution. Preferably, the LSTM network contains two hidden layers, each with 128 neurons; the number of states in the HMM is dynamically configured to 3 to 5 based on the device type, and the state transition probability matrix is ​​updated online as the device behavior evolves. Preferably, the attribute library is stored in a horizontally scalable distributed database that can hold hundreds of millions of device attribute records. The policy description language uses a simplified subset of Extensible Access Control Markup Language (XACML), with a parsing latency of less than 1 millisecond. Preferably, the regional gateway policy execution point caches high-frequency access policies, reducing cloud reliance; the MQTT message header compression rate reaches over 60%, significantly reducing WAN bandwidth consumption. Preferably, the behavior analysis engine automatically generates a structured security event report upon detecting anomalies, including behavioral feature vectors, confidence scores, and associated device graphs, and pushes it to the Security Information and Event Management System (SIEM) in real time. Preferably, the session token adopts a dynamic decay mechanism based on trust, automatically shortening its validity period when trust decreases, and supports geofencing-based access constraints. Preferably, the formal verification tool supports policy change impact analysis, and conflict resolution operations are fully traceable, meeting the Level 3 audit requirements of the Information Security Protection Standard 2.0. Preferably, an adversarial example enhancement mechanism is introduced during the training of the deep reinforcement learning model to improve the generalized defense capability against new attack patterns. Preferably, the system integrates a visual management interface, supporting device security heatmaps, policy topology views, abnormal behavior tracing and drill-down, and custom security report generation, assisting security administrators in achieving proactive defense and intelligent operation and maintenance.

[0009] Compared with existing technologies, this invention has the following beneficial effects: This invention is not a simple aggregation of known technologies, but rather a synergistic technological whole formed by introducing zero-knowledge proofs into lightweight device authentication, constructing an LSTM-HMM hybrid trust evaluation model, designing an ABAC policy generation mechanism based on multi-dimensional context, achieving millisecond-level anomaly blocking and dynamic token management, and establishing a formal verification-driven policy self-optimization closed loop. Specifically, this invention is the first to combine hardware fingerprints generated jointly by PUF and firmware hashes with zero-knowledge proofs for IoT device identity authentication. In dynamic trust evaluation, it integrates LSTM and HMM to model long-term behavioral dependencies and short-term state transitions respectively. Simultaneously, at the policy execution level, it achieves sub-hundred-millisecond policy synchronization based on MQTT and 10-millisecond blocking responses based on DBSCAN streaming anomaly detection. Furthermore, it combines formal verification and temporal logic model detection for policy conflict resolution and utilizes the A3C deep reinforcement learning model for continuous optimization of policy parameters. This organic combination of the above technical features is not disclosed in existing technologies and is not easily conceived by those skilled in the art, thereby significantly improving the identity credibility, policy consistency, threat response speed, and adaptive defense capabilities of IoT systems. Especially in large-scale heterogeneous IoT environments, it can effectively prevent identity forgery, permission abuse, lateral movement and covert attacks, solve the core pain point that traditional static access control models cannot cope with dynamic threats, and produce comprehensive technical effects that exceed the sum of the individual effects of each component. It has outstanding substantive features and significant progress. Attached Figure Description

[0010] The invention will now be further described with reference to the accompanying drawings.

[0011] Figure 1 This is a flowchart illustrating the multi-level IoT device security management and access control method proposed in this invention.

[0012] Figure 2 This is a schematic diagram of the core principle framework of dynamic trust assessment and cross-level strategy collaboration in this invention;

[0013] Figure 3 This is a logical flow diagram of real-time access control and abnormal behavior blocking in this invention. Detailed Implementation

[0014] Example 1

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0016] Currently, with the widespread application of IoT technology, a massive number of heterogeneous devices are accessing systems via wireless networks, making device security management and access control a core aspect of ensuring overall network security. The diverse types of devices, limited resources, and complex deployment scenarios in the IoT environment pose significant challenges to lightweight authentication mechanisms, dynamic adaptability of permission policies, and real-time blocking capabilities for unauthorized access. Especially in multi-level network architectures, devices with different security levels need to achieve fine-grained access control within a unified framework, which puts significant pressure on the scalability and policy execution efficiency of existing security systems. To address these technical issues, this invention proposes an end-to-end closed-loop security control system by innovatively integrating lightweight authentication, dynamic trust assessment, cross-level policy collaboration, real-time anomaly blocking, and adaptive policy optimization mechanisms. This system effectively solves the problems of insufficient security protection capabilities caused by static permission models, isolated security policies, and delayed response in existing technologies. It achieves high-efficiency, high-consistency, and highly adaptable security management in resource-constrained environments and is applied to a multi-level IoT device security management and access control method.

[0017] Reference Appendix Figure 1 The overall technical architecture of this invention includes a lightweight authentication module deployed on edge nodes, a behavior analysis engine and policy execution point on regional gateways, a policy decision point and unified policy library on the cloud platform, and low-latency communication and policy synchronization at each level through the Lightweight Message Queuing Telemetry Transport Protocol (MQTT). (Appendix) Figure 2 This demonstrates the core framework of dynamic trust assessment and cross-level policy collaboration. The device runtime context data is processed by an LSTM-HMM hybrid model to generate real-time trust levels, which serve as key inputs for policy generation. (Appendix) Figure 3 It details the logical process of real-time access control and abnormal behavior blocking, including key steps such as traffic mirroring, streaming anomaly detection, blocking command issuance, and dynamic management of session tokens.

[0018] In the aforementioned multi-level IoT device security management and access control method, step 1, device identity verification, involves bidirectional verification of the device's digital certificate and hardware fingerprint when the device accesses the network via a lightweight authentication module deployed at the edge node. Specifically, the hardware fingerprint is jointly generated by the device firmware hash value and the Physically Unclonable Function (PUF), forming a unique identifier with anti-cloning and anti-tampering properties. The firmware hash value is obtained by performing a one-way hash calculation on the device firmware image using the SHA-256 algorithm, while the PUF response is extracted based on the randomness of the SRAM's initial power-on state. The two are fused using the HMAC-SHA256 function to generate a 256-bit hardware fingerprint digest. This digest is bound to the public key certificate pre-installed at the device's factory and stored in the device's secure storage area. The verification process employs a digital signature algorithm based on Elliptic Curve Cryptography (ECC), with a signature length of 256 bits, and introduces a zero-knowledge proof mechanism to complete identity confirmation without revealing the private key. The zero-knowledge proof protocol uses a variant of the Sigma protocol. The challenge value is randomly generated by the edge node. The device uses its private key to sign the challenge value and construct a proof. The edge node only needs to verify the validity of the proof to confirm that the device holds the corresponding private key, without needing to obtain the private key itself. This process ensures the authenticity, non-repudiation, and privacy protection of the device's identity. The lightweight authentication module is compatible with multiple IoT protocols, including Narrowband IoT (NB-IoT) and LoRa. The authentication data packet payload does not exceed 200 bytes, and its memory usage on a typical microcontroller unit (MCU) is less than 50 kilobytes. The authentication response time is controlled within 2 seconds, and it supports an offline emergency authentication mode. In offline mode, the edge node can verify based on a locally cached list of recently valid certificates, ensuring basic access security during network outages.

[0019] In the aforementioned multi-level IoT device security management and access control method, step 2, dynamic trust assessment and security status awareness, continuously collects context data during device operation, including communication behavior patterns, resource consumption rates, and geographic location information. Specifically, the context data is collected by lightweight agent software on the device side, with an agent size of less than 30 kilobytes. The collected data is processed locally using differential privacy desensitization before being uploaded. The differential privacy mechanism is achieved by adding noise that satisfies a Laplace distribution to the original data, with a privacy budget ε set to 0.5 to balance data utility and privacy protection. Communication behavior pattern analysis covers the number of packets per second, connection duration, protocol compliance, and port usage distribution; these indicators are encoded into fixed-dimensional feature vectors. Resource consumption rates include CPU utilization, memory usage, and battery discharge rate, sampled once per second. Geographic location information is provided by the device's built-in GNSS module, with an accuracy better than 5 meters. The collected multi-source context data is input into a hybrid evaluation algorithm integrating Long Short-Term Memory (LSTM) networks and Hidden Markov Models (HMMs). The LSTM network comprises two hidden layers, each with 128 neurons, used to model long-term behavioral dependencies. Its input is a sequence of historical feature vectors within a sliding time window, and its output is a long-term trend score of device behavior. The number of states in the Hidden Model (HMM) is dynamically configured from 3 to 5 based on the device type. For example, sensor devices are configured with 3 states (normal, low activity, abnormal), while actuator devices are configured with 5 states (standby, execution, high load, maintenance, fault). The state transition probability matrix is ​​updated online as device behavior evolves to capture short-term state transitions. The outputs of both are weighted and fused to generate a continuous trust value between 0 and 1, with an update cycle of no more than 5 seconds. The weight coefficients are preset according to the device type and deployment scenario; for example, in industrial control scenarios, the HMM is given higher weights to quickly respond to state changes. This trust value is not only used to classify device security levels (e.g., a trust value greater than or equal to 0.8 indicates high security, 0.5 to 0.8 indicates medium security, and less than 0.5 indicates low security) but also serves as a core input parameter for subsequent policy generation and session management.

[0020] In the aforementioned multi-level IoT device security management and access control method, step 3, cross-level permission policy generation and collaboration, is based on the trust level, network level of the device, and sensitivity of the target resources accessed, as output in step 2. The policy decision point dynamically extracts more than 20 attributes from a distributed attribute library, including device type, environmental risk index, and historical compliance records. Specifically, the attribute library is stored in a horizontally scalable distributed database that can hold hundreds of millions of device attribute records. The policy description language uses a simplified subset of Extensible Access Control Markup Language (XACML), with a parsing latency of less than 1 millisecond. The environmental risk index is dynamically calculated by the regional gateway based on local weather, physical security alarms, and network threat intelligence. Historical compliance records include the device's access success rate, number of policy violations, and number of abnormal behavior reports over the past 30 days. The policy decision point adopts an improved attribute-based access control (ABAC) model, with policy rules formalized as: <subject attribute, resource attribute, environmental attribute, operation> → {allow, deny}. The policy generation engine matches predefined policy templates and instantiates specific policies based on the real-time input attribute set. Policy decision points and policy enforcement points deployed on regional gateways and cloud platforms communicate with each other using the Lightweight Message Queuing Telemetry Transport Protocol (MQTT) for low-latency communication. The protocol employs binary encoding and a Quality of Service (QoS) level of "at least once" delivery, ensuring that policies are synchronized to critical nodes across the network within 100 milliseconds. MQTT message header compression reaches over 60%, significantly reducing WAN bandwidth consumption. Regional gateway policy enforcement points cache frequently accessed policies, reducing reliance on the cloud; they also support policy version rollback and consistency verification, ensuring policy version consistency across nodes by comparing policy hash values.

[0021] In the aforementioned multi-level IoT device security management and access control method, step 4, real-time access control and abnormal behavior blocking, involves the policy enforcement point making an access decision based on the latest permission policy when a device initiates an access request. Specifically, the access request includes the device identifier, target resource identifier, operation type, and current session token. The policy enforcement point first verifies the token's validity. The token's validity period is dynamically set based on the device's real-time trust level, ranging from 5 minutes to 24 hours, and employs a dynamic decay mechanism based on trust level, automatically shortening the validity period as trust level decreases. Simultaneously, the policy enforcement point supports geofence-based access constraints, allowing access to specific resources only when the device's geographical location is within a preset geofence. Simultaneously with the access decision, traffic mirroring technology is used to import the device's communication flow into a behavior analysis engine deployed on the regional gateway. This engine employs a streaming computing architecture, combining a density-based clustering algorithm (DBSCAN) with an online learning mechanism to detect abnormal traffic deviating from historical behavior baselines in real time. The neighborhood radius (eps) and minimum sample size (minPts) parameters of the DBSCAN algorithm are dynamically adjusted based on device type and historical traffic characteristics. For example, for sensors with stable data reporting frequency, the initial value of eps is set to 1.5 times the historical average packet interval, and minPts is set to 5; for actuators with frequent interactions, eps is set to 2 times, and minPts is set to 10. The online learning mechanism continuously fine-tunes the parameters based on newly arrived data points to adapt to the slow drift of behavioral patterns. For identified high-risk access or abnormal behavior, the policy enforcement point can issue a blocking command to the corresponding network boundary device (such as firewall or switch ACL) within 10 milliseconds and simultaneously revoke the current session token. When anomalies are detected, the behavior analysis engine automatically generates a structured security event report, including behavioral feature vectors, confidence scores, and associated device graphs, and pushes it to the Security Information and Event Management System (SIEM) in real time.

[0022] In the aforementioned multi-level IoT device security management and access control method, step 5, policy conflict resolution and adaptive optimization, establishes a unified policy library covering the edge to the cloud. Specifically, a formal verification tool automatically converts access control policies into temporal logic formulas, and a model detection algorithm is used to verify whether logical conflicts exist between policies. The formal verification tool supports policy change impact analysis, and the conflict resolution operation is fully traceable, meeting the Level 3 audit requirements of the Information Security Protection Standard 2.0. When a conflict is detected, for example, one policy allows low-trust devices to access highly sensitive resources, while another policy prohibits such access, the system uses a multi-dimensional conflict resolution algorithm based on priority, security level, and contextual semantics to correct the policy. Priority is set by the policy creator, the security level is determined based on a comprehensive assessment of resource sensitivity and device trust, and the contextual semantics analyze the specific scenarios in which the policy applies. The algorithm prioritizes the enforcement of policies in high-security areas; for example, policies in industrial control networks have higher priority than those in ordinary office networks. The policy library synchronously records a full range of access control logs, including each access request, decision result, policy version, and contextual snapshot. The system trains a deep reinforcement learning model weekly using over 1 million historical records (employing the Asynchronous Advantage Actor Critics algorithm, A3C) to maximize system security gains (defined as the number of successfully blocked attacks minus the number of false positives for legitimate access) and minimize the false positive rate for legitimate access, achieving continuous adaptive optimization of policy generation parameters. An adversarial example enhancement mechanism is introduced during the deep reinforcement learning model training process, generating synthetic data simulating novel attack patterns to improve the generalized defense capability against unknown threats. The system integrates a visual management interface, supporting device security heatmaps, policy topology views, anomaly behavior tracing and drill-down, and custom security report generation, assisting security administrators in achieving proactive defense and intelligent operation and maintenance.

[0023] To illustrate the implementation effects of this invention more specifically, an application example in a smart grid scenario is constructed below. In this scenario, tens of thousands of smart meters (edge ​​nodes), hundreds of regional data concentrators (regional gateways), and a provincial power dispatch center (cloud platform) are deployed. When a new smart meter powers on and attempts to access the network, it first initiates an authentication request to the nearest regional data concentrator. The meter's built-in lightweight agent collects its firmware hash and SRAMPUF response, generates a hardware fingerprint, and signs the authentication challenge using an ECC private key to construct a zero-knowledge proof. After the regional data concentrator verifies the validity of the proof, it allows access and assigns an initial trust level of 0.7. Subsequently, the meter begins reporting electricity consumption data every 15 minutes. Its agent continuously collects context data: the number of data packets per second is stable at 0.1, the connection duration is approximately 2 seconds, only a preset port is used, the CPU utilization is less than 5%, and the geographical location is fixed. This data is uploaded to the regional gateway, evaluated by the LSTM-HMM model, and the trust level is stable at 0.85, classifying it as a high-security level. When the meter attempts to access the provincial dispatch center's electricity price adjustment interface (a highly sensitive resource), the policy decision point extracts its device type (smart meter), trust level (0.85), resource sensitivity (high), and environmental risk index (currently no anomalies) from the attribute database, generating the policy "Allow high-security-level smart meters to access the electricity price interface during working hours." The policy is synchronized to all relevant nodes via MQTT within 80 milliseconds. Access is granted, and the session token validity period is set to 2 hours. One day, the meter is infected with malware due to a firmware vulnerability and begins sending data to external IPs at a frequency of 10 packets per second. The regional gateway's behavior analysis engine detects that its communication behavior deviates significantly from the baseline (eps=0.5 seconds, minPts=5) via DBSCAN, with a confidence score of 0.95. It immediately generates a security incident report and pushes it to SIEM. The policy enforcement point issues a blocking command within 8 milliseconds, cutting off its external connection and revoking the token. Simultaneously, its trust level is rapidly reduced to 0.3, and all subsequent access requests are rejected. The unified policy library recorded this incident, and the A3C model learned the characteristics of such high-frequency outbound behaviors in subsequent training, optimized the initial parameter configuration of DBSCAN, and improved the detection speed of similar attacks.

[0024] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-tiered IoT device security management and access control method, characterized in that, The method comprises the following steps: Step 1, device identity trusted verification: when the device accesses the network, the lightweight authentication module deployed in the edge node performs bidirectional verification on the device digital certificate and the hardware fingerprint; the hardware fingerprint is generated by the device firmware hash value and the physically unclonable function, and the verification adopts the elliptic curve cryptography digital signature algorithm with a signature length of 256 bits; Step 2, dynamic trust evaluation and security state perception: continuously collect runtime context data such as device communication behavior mode, resource consumption rate and geographic location, and calculate real-time trust degree through a hybrid evaluation algorithm of long short-term memory network and hidden Markov model; the trust degree is a continuous value between 0 and 1, the update period is not more than 5 seconds, and the device security level is dynamically divided according to the preset threshold; Step 3, cross-level permission policy generation and cooperation: based on the device security level and the network level to which the device belongs, the policy decision point calls the device type, access resource sensitivity, environment risk index and other attributes in the attribute library, and generates a fine-grained permission policy by using an attribute-based access control model; The policy decision point communicates with the policy execution point of the regional gateway and the cloud platform through the lightweight message queue telemetry transmission protocol, and ensures that the policy is synchronized to all network key nodes within 100 milliseconds; Step 4, real-time access control and abnormal behavior blocking: when the device initiates an access request, the policy execution point makes a decision according to the latest permission policy; at the same time, the communication flow is introduced into the behavior analysis engine through the flow mirroring technology, and the engine detects abnormal flow deviating from the historical behavior mode by using a density-based clustering algorithm; when high-risk behavior is identified, the policy execution point sends a blocking instruction to the network boundary device within 10 milliseconds; Step 5, policy conflict resolution and adaptive optimization: a unified policy library from the edge to the cloud is established, the logical conflicts of different level policies are detected by using a formal verification tool, and the conflicts are corrected by using an algorithm based on priority and context; the policy library records historical access control logs, and optimizes the policy generation parameters by periodically training a deep reinforcement learning model offline.

2. The hierarchical IoT device security management and access control method of claim 1, wherein, The lightweight authentication module in step 1 supports narrowband Internet of Things and long-distance wireless communication protocols, the data packet load of the authentication process is not more than 200 bytes, the running memory occupation of the module on a typical microcontroller unit is less than 50 kilobytes, and the authentication response time is within 2 seconds.

3. The hierarchical IoT device security management and access control method of claim 1, wherein, The context data in step 2 is collected by a device-side proxy software, the code size of the proxy software is less than 30 kilobytes, and the collected data is transmitted to the trust evaluation module after light data desensitization processing; the communication behavior mode analysis includes per-second packet quantity, connection duration and protocol compliance check; the long short-term memory network contains 2 hidden layers with 128 neurons in each layer; the number of states of the hidden Markov model is dynamically adjusted between 3 and 5 according to the device type.

4. The hierarchical IoT device security management and access control method of claim 1, wherein, The attribute library in step 3 is stored in a distributed database, supports horizontal expansion to accommodate hundreds of millions of device attribute records, considers more than 20 dynamic attributes when generating a policy, and uses a simplified subset of the extensible access control markup language as a policy description language; the lightweight message queue telemetry transport protocol uses a quality of service level of at least once delivery, and the message format is binary encoding; the regional gateway policy enforcement point caches common policies to reduce cloud query delay.

5. The hierarchical IoT device security management and access control method of claim 1, wherein, The behavior analysis engine in step 4 is deployed in the regional gateway and uses a streaming computing architecture to process real-time traffic; the neighborhood radius and minimum sample size parameters in the density-based clustering algorithm are adaptively adjusted according to network traffic characteristics; for detected abnormal behavior, the engine generates a security event report and uploads it to a security information and event management system.

6. The hierarchical IoT device security management and access control method of claim 5, wherein, Step 4 also includes monitoring of device session state, with session token validity period dynamically set according to device security level, ranging from 5 minutes to 24 hours, and revoked immediately upon detection of suspicious activity.

7. The hierarchical IoT device security management and access control method of claim 1, wherein, In step 5, the formal verification tool converts the access control policy into a timed logic formula and uses a model detection algorithm to verify the consistency of the policy; the conflict resolution algorithm prioritizes the enforcement of policies in high-security level areas and records all resolution operations for auditing.

8. The hierarchical IoT device security management and access control method of claim 7, wherein, The deep reinforcement learning model in step 5 uses the asynchronous advantage actor critic algorithm for training, with a training cycle of once a week, and uses over 1 million access records in historical logs to optimize policy generation, aiming to maximize system security benefits while minimizing the probability of false rejection of legitimate access.

9. The hierarchical IoT device security management and access control method of claim 3, wherein, The geographic location information collected by the device-side proxy software is provided by the device's built-in GNSS module, with an accuracy of better than 5 meters; resource consumption rate includes CPU occupancy, memory usage, and battery discharge rate, with a sampling frequency of 1 second; differential privacy mechanism is achieved by adding noise satisfying Laplace distribution to the original data, with privacy budget ε set to 0.

5.

10. The hierarchical IoT device security management and access control method of claim 4, wherein, The environmental risk indicators are dynamically calculated by the regional gateway based on local weather, physical security alerts, and network threat intelligence; historical compliance records include the device's access success rate, policy violation times, and abnormal behavior report quantity in the past 30 days; The policy enforcement point supports access constraints based on geographic fences, allowing access to specific resources only when the device's geographic location is within the preset geographic fence.

Images