An intelligent management and control system based on device identification dynamic detection

Abstract

The application relates to the technical field of intelligent management and control, and discloses an intelligent management and control system based on dynamic detection of equipment identification, which comprises a perception module: used for continuously collecting multiple-dimension physical fingerprint data of unmanned equipment, and dynamically reconstructing physical fingerprint collection logic according to a risk level; a modeling module: used for modeling the physical fingerprint data into a dynamic entropy matrix and a dynamic physical fingerprint spectrum; and an analysis module: used for performing real-time analysis on the dynamic entropy matrix and the dynamic physical fingerprint spectrum based on a pre-trained baseline model. Through deep perception and active enhancement mechanism of multi-modal physical fingerprints, and through a programmable hardware security unit for periodically injecting a perturbation to generate a dynamic response spectrum, and through dynamic adjustment or reconstruction of the physical fingerprint collection logic in a high-risk state, the problems that a traditional single and static identification is easy to be forged or bypassed are effectively solved, and the identification precision and attack resistance of the system to physical tampering and illegal replacement are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, specifically to an intelligent control system based on dynamic detection of equipment identification. Background Technology

[0002] With the widespread application of unmanned equipment in military, industrial, and civilian fields, the security threats they face are becoming increasingly complex. Traditional cybersecurity measures mainly focus on the software layer, such as identity authentication, encrypted communication, access control, and intrusion detection systems. These measures have shown some effectiveness in dealing with attacks targeting software vulnerabilities, protocol flaws, or logical errors. However, as physical entities, the physical integrity and authenticity of the hardware platform and internal components of unmanned equipment are also the cornerstone of ensuring its secure operation. Once attackers illegally replace, tamper with, or implant malicious hardware into the hardware components of unmanned equipment through supply chain hijacking, physical contact, or insider attacks, traditional software-layer security protections will face the risk of failure.

[0003] In existing technologies, hardware-level security protection often relies on static hardware identifiers (such as serial numbers and MAC addresses), Physically Unclonable Functions (PUFs), or simple hardware encryption modules. However, these static identifiers are easily copied, simulated, or bypassed, especially when attackers have certain expertise and physical access capabilities. For example, by copying the PUF response or simulating the static identifier of a legitimate component, attackers may deceive the system, allowing the tampered device to impersonate a legitimate device to access the network or perform tasks. In addition, some hardware integrity verification methods may not have fine enough detection granularity, making it difficult to capture subtle changes in physical state at the micrometer or sub-millisecond level. They are often unable to effectively identify carefully designed low interception rate (LPI) or low detectability (LPI) hardware tampering behaviors.

[0004] Existing technologies also have significant limitations when facing changes in the physical relationships between internal components of unmanned equipment and the dynamic evolution of their physical states. Most solutions focus on the static identification of individual components, failing to effectively establish and analyze the collaborative fingerprints of multiple internal components in terms of power consumption, timing, thermal field, vibration, etc. This means that even if an attacker can circumvent the static identification detection of a single component, if their tampering behavior disrupts the inherent physical relationship patterns or dynamic collaborative characteristics between internal components, traditional systems may not be able to detect it. In addition, during long-term operation, the physical characteristics of the equipment will naturally evolve due to aging, wear, or environmental changes. Systems lacking adaptive capabilities may therefore generate a high false alarm rate or lose the ability to detect real threats due to the inability to update the baseline in a timely manner. This static detection and lack of insight into the dynamic evolution of internal physical relationships necessitate a new type of intelligent security system that can dynamically perceive the physical state of the equipment from multiple dimensions, deeply model internal physical relationships, and possess adaptive learning and hierarchical control capabilities. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent control system based on dynamic detection of equipment identifiers. This system solves the problem that existing unmanned equipment control systems are unable to accurately identify and effectively respond to hardware-level tampering when faced with illegal equipment replacement or tampering. This is because the traditional, single, static identifier detection mechanism is easily bypassed, and the system lacks the ability to accurately perceive and adaptively analyze the dynamic evolution and correlation changes of the internal physical state of the equipment.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent control system based on dynamic detection of device identification, comprising:

[0007] Perception module: Used to continuously collect multi-dimensional physical fingerprint data of unmanned equipment and dynamically reconstruct the physical fingerprint collection logic according to the risk level;

[0008] Modeling module: used to model the physical fingerprint data into a dynamic entropy matrix and a dynamic physical fingerprint spectrum;

[0009] Analysis module: used to perform real-time analysis on the dynamic entropy matrix and dynamic physical fingerprint spectrum based on a pre-trained baseline model to generate a physical tampering index;

[0010] And a control module: used to classify risk levels based on the physical tampering index and execute corresponding hierarchical control strategies.

[0011] Preferably, the sensing module is specifically used to: deploy a sensor array within the unmanned equipment to collect hardware-inherent noise fingerprints, micro-environment dynamic fingerprints, system-level behavioral fingerprints, and module-level communication features; and periodically inject micro-perturbations into selected physical fingerprint sources through a built-in programmable hardware security unit to generate dynamic response feature spectra, and dynamically adjust or reconstruct the physical fingerprint acquisition logic when there is a high risk.

[0012] Preferably, the modeling module is specifically used for: performing feature generation and entropy transformation on the collected physical fingerprint data to generate multiple physical fingerprint entropy value sequences; constructing the physical fingerprint entropy value sequences into a dynamic entropy matrix P(t), where the elements of P(t) are the entropy values ​​E of the j-th physical fingerprint feature at a specific time step kΔt. j (kΔt), and this matrix is ​​continuously updated over time; multiple sensor nodes or modules within the unmanned equipment are defined as nodes V of a graph, and static topological edges E are constructed based on the static topological relationships between the nodes. s And construct dynamic association edges E based on the real-time association strength between physical fingerprints of nodes. D (t) to form a dynamic physical fingerprint spectrum

[0013] Preferably, the analysis module includes a training unit for constructing a baseline model. Specifically, the training unit is used to: train the historical dynamic entropy matrix using nonnegative matrix factorization (NMF) or robust principal component analysis (RPCA) algorithms to extract the baseline pattern matrix W of the device; and train the historical dynamic physical fingerprint sequence using graph neural networks (GNNs) to learn the normal evolution pattern of node fingerprints and the baseline features of the graph structure.

[0014] Preferably, the analysis module includes an evaluation unit for real-time detection, specifically used to: employ a sparse coding algorithm to attempt to optimize the current dynamic entropy matrix P. current (t) represents a linear combination of the baseline pattern matrix W, and its reconstruction error ∈(t) is calculated, where the sparse coding algorithm is: The current dynamic physical fingerprint G(t) is input into the pre-trained GNN model to calculate the anomaly score δ. G (t); and combining the reconstruction error ∈(t) and the map anomaly score δ G Indicators such as (t) are obtained through the fusion function F. fuse (·) Physical tampering index R(t) of computing device.

[0015] Preferably, the analysis module is further configured to fine-tune the baseline pattern matrix W and / or the parameters of the GNN model using an incremental learning algorithm after confirming that the unmanned equipment is in a normal state.

[0016] Preferably, the control module is specifically used to: map the physical tampering index R(t) to a risk level that includes at least four levels: security, low-risk warning, medium-risk isolation, and high-risk blocking; and automatically trigger and execute corresponding hierarchical control strategies based on the risk level. The strategies include, but are not limited to, recording abnormal logs, sending early warning notifications, adjusting data transmission frequency, initiating fingerprint collection logic reconstruction, restricting resource access, forcing device self-tests, cutting off communication, locking functions, or triggering data self-destruction procedures.

[0017] Preferably, the inherent hardware noise fingerprint includes the power consumption ripple spectrum of the CPU / GPU, the clock jitter characteristics of the internal crystal oscillator, or the transient response characteristics of the RF module; the microenvironment dynamic fingerprint includes the internal micron-level temperature distribution pattern or vibration resonance response spectrum; the system-level behavioral fingerprint includes the transient voltage / current response characteristic curve of the power rail; and the module-level communication characteristics include the data transmission timing characteristics between internal modules or the differences in protocol stack implementation.

[0018] Preferably, the featureization and entropy transformation of the physical fingerprint data includes calculating the sample entropy or approximate entropy of the time-series signal to serve as the entropy value E of the physical fingerprint feature. j (t).

[0019] Preferably, the high-risk blocking strategy includes immediately cutting off all communication connections between the unmanned equipment and the external network, locking the key functions of the unmanned equipment, or triggering an encrypted data erasure procedure.

[0020] This invention provides an intelligent control system based on dynamic detection of device identifiers. It has the following advantages:

[0021] 1. This invention utilizes a multimodal physical fingerprinting deep perception and active enhancement mechanism. Specifically, it deploys a sensor array within unmanned equipment to collect hardware-inherent noise fingerprints, micro-environment dynamic fingerprints, system-level behavioral fingerprints, and module-level communication features. Furthermore, it periodically injects micro-perturbations through a programmable hardware security unit to generate a dynamic response spectrum. Under high-risk conditions, it dynamically adjusts or reconstructs the physical fingerprint acquisition logic. This effectively solves the problem of traditional single, static identifiers being easily forged or bypassed. This multi-dimensional, highly dynamic, and actively immune fingerprint acquisition method fundamentally enhances the uniqueness and unforgeability of device identification, making it difficult for attackers to accurately copy device fingerprints. Consequently, it greatly improves the system's accuracy in identifying physical tampering and illegal replacement, as well as its resistance to attacks.

[0022] 2. This invention models the collected physical fingerprint data into a dynamic entropy matrix and a spatiotemporal topology map. Specifically, it characterizes and transforms the fingerprint data to generate a physical fingerprint entropy sequence, and constructs a continuously updated dynamic entropy matrix in the time dimension based on this. At the same time, the device modules are constructed as nodes in the graph. Combining static topology and real-time correlation strength, a dynamic physical fingerprint map is formed, which realizes a more comprehensive and in-depth characterization of the physical state of the device. The dynamic entropy matrix captures the changing patterns of multi-dimensional physical features in the time series, enabling the system to discover correlation anomalies or distributed tampering behaviors that are difficult to reveal with a single fingerprint, significantly enhancing the detection capability of complex tampering patterns.

[0023] 3. This invention utilizes intelligent analysis and risk quantification based on matrix factorization and graph neural networks. Specifically, during the learning phase, it extracts the baseline pattern matrix using nonnegative matrix factorization (NMF) or robust principal component analysis (RPCA) algorithms and learns the normal topological behavior of the graph using graph neural networks (GNNs). During the real-time detection phase, it uses a sparse coding algorithm to calculate the reconstruction error of the dynamic entropy matrix and calculates the graph anomaly score using GNNs. Finally, it fuses these results to generate a physical tampering index, significantly improving the intelligence and accuracy of tampering detection.

[0024] 4. This invention employs an adaptive feedback and multi-level hierarchical control strategy. After confirming the normal state of the equipment, it fine-tunes the baseline model parameters through an incremental learning algorithm and classifies risks into multiple levels, such as safe, low-risk warning, medium-risk isolation, and high-risk blocking, based on the physical tampering index. Corresponding hierarchical control strategies are then implemented, significantly improving the robustness and decision-making efficiency of the system. The adaptive learning mechanism enables the system to cope with factors such as normal equipment aging and environmental changes, reducing false alarms, maintaining long-term operational accuracy, and ensuring the flexibility and effectiveness of the system response.

[0025] 5. This invention utilizes the proactive immunity and reconfiguration capabilities of the programmable hardware security unit to enable the system to dynamically adjust or reconfigure the physical fingerprint acquisition logic when high-risk tampering is detected. This generates a new fingerprint pattern for secondary verification. Combined with proactive defense and emergency response measures such as targeted restrictions on access to sensitive resources, forced device self-testing, and even communication cutoff, function locking, and triggering data self-destruction, a highly robust and adaptive adversarial intelligent management and control framework is constructed, effectively preventing serious losses caused by the leakage of sensitive information or malicious control of devices. Attached Figure Description

[0026] Figure 1 This is the main system flowchart of the present invention;

[0027] Figure 2 This is a flowchart of the sensing module processing of the present invention;

[0028] Figure 3 This is a flowchart of the analysis module processing of the present invention;

[0029] Figure 4 This is a flowchart of the control module processing of the present invention. Detailed Implementation

[0030] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Please see the appendix Figure 1 -Appendix Figure 4 This invention provides an intelligent management and control system based on dynamic detection of device identifiers, comprising:

[0032] The perception module is used to continuously collect multi-dimensional physical fingerprint data of the unmanned equipment and dynamically reconstruct the physical fingerprint acquisition logic according to the risk level. Specifically, the perception module is used to: deploy a sensor array within the unmanned equipment to collect hardware inherent noise fingerprints, micro-environment dynamic fingerprints, system-level behavioral fingerprints, and module-level communication characteristics; and periodically inject micro-perturbations into selected physical fingerprint sources through a built-in programmable hardware security unit to generate dynamic response characteristic spectra, and dynamically adjust or reconstruct the physical fingerprint acquisition logic when there is a high risk. The hardware inherent noise fingerprint includes the power consumption ripple spectrum of the CPU / GPU, the clock jitter characteristics of the internal crystal oscillator, or the transient response characteristics of the RF module; the micro-environment dynamic fingerprint includes the internal micron-level temperature distribution pattern or vibration resonance response spectrum; the system-level behavioral fingerprint includes the transient voltage / current response characteristic curve of the power rail; and the module-level communication characteristics include the data transmission timing characteristics between internal modules or differences in protocol stack implementation.

[0033] Specifically, a multi-dimensional array of micro-sensors is deployed within the unmanned equipment. These sensors continuously collect the device's inherent hardware noise fingerprints (such as CPU / GPU power consumption ripple spectrum, clock jitter, and RF module transient response), micro-environment dynamic fingerprints (such as micrometer-level temperature distribution and vibration resonance spectrum), system-level behavioral fingerprints (such as the transient response curve of the electro-filter track), and module-level communication characteristics (such as the timing characteristics of data transmission between internal modules). These fingerprints originate from the manufacturing differences and physical characteristics of the device's underlying hardware, making them difficult to accurately copy or forge. Furthermore, this system achieves proactive enhancement through a built-in programmable hardware security unit. This security unit periodically injects micro-perturbations (such as fine-tuning voltage or injecting high-frequency signals) into selected physical fingerprint sources and measures the device's dynamic response characteristic spectrum under the perturbation. This proactively generated response spectrum contains the unique interaction characteristics between the device and the perturbation, significantly improving the complexity and anti-forgery properties of the fingerprint. When a potential threat is detected, the security unit can also dynamically adjust or reconstruct the physical fingerprint acquisition logic (e.g., switch the sampling frequency, change the preprocessing algorithm, or selectively activate the sensor) to generate new fingerprint patterns that have not been mastered by attackers, thereby achieving fingerprint "immunity" and "adaptive evolution".

[0034] Modeling module: This module is used to model the physical fingerprint data into a dynamic entropy matrix and a dynamic physical fingerprint spectrum. Specifically, the modeling module is used to: perform feature extraction and entropy transformation on the collected physical fingerprint data to generate multiple physical fingerprint entropy value sequences; and construct the physical fingerprint entropy value sequences into a dynamic entropy matrix P(t), where the elements of P(t) are the entropy values ​​E of the j-th physical fingerprint feature at a specific time step kΔt. j (kΔt), and this matrix is ​​continuously updated over time; multiple sensor nodes or modules within the unmanned equipment are defined as nodes V of a graph, and static topological edges E are constructed based on the static topological relationships between the nodes. s And construct dynamic association edges E based on the real-time association strength between physical fingerprints of nodes. D (t) to form a dynamic physical fingerprint spectrum The process of characterizing and transforming the physical fingerprint data includes calculating the sample entropy or approximate entropy of the time-series signal to serve as the entropy value E of the physical fingerprint feature. j (t);

[0035] Specifically, this system performs feature extraction and entropy transformation on the collected raw and enhanced physical fingerprint data. Through nonlinear mapping and information theory methods, various fingerprints are converted into quantified entropy values, effectively characterizing their complexity, randomness, or uncertainty. Based on this, the system constructs two key mathematical models:

[0036] Dynamic entropy matrix P(t): This matrix aggregates the entropy values ​​of multi-dimensional physical fingerprints within a predefined time window and continuously updates them in the form of a sliding window. Each row of the matrix represents a historical time step, and each column represents the entropy value E of a physical fingerprint feature. j (kΔt) is used to capture the dynamic evolution of the physical state of the equipment.

[0037] Dynamic Physical Fingerprint Map G(t): This map treats key sensor nodes or modules inside unmanned equipment as nodes in a graph, with each node carrying its corresponding physical fingerprint feature vector. The edges of the map include not only pre-defined static topological edges (such as physical connections and power supply relationships) but also dynamically constructed time-varying relational edges, whose weights are determined by the real-time correlation strength (e.g., correlation coefficient) between the physical fingerprints of the nodes. This map can reveal the spatial correlation of physical features within the equipment and their dynamic evolution over time, providing structured information for identifying local or distributed tampering.

[0038] Analysis Module: This module performs real-time analysis of the dynamic entropy matrix and dynamic physical fingerprint map based on a pre-trained baseline model to generate a physical tampering index. The analysis module includes a training unit for building the baseline model. Specifically, this training unit is used to: train the historical dynamic entropy matrix using non-negative matrix factorization (NMF) or robust principal component analysis (RPCA) algorithms to extract the device's baseline pattern matrix W; and train the historical dynamic physical fingerprint map sequence using graph neural networks (GNNs) to learn the normal evolution pattern of node fingerprints and the baseline features of the map structure. The analysis module also includes an evaluation unit for real-time detection. Specifically, this evaluation unit is used to: employ a sparse coding algorithm to attempt to modify the current dynamic entropy matrix P... current (t) represents a linear combination of the baseline pattern matrix W, and its reconstruction error ∈(t) is calculated, where the sparse coding algorithm is: The current dynamic physical fingerprint G(t) is input into the pre-trained GNN model to calculate the anomaly score δ. G (t); and combining the reconstruction error ∈(t) and the map anomaly score δ G Indicators such as (t) are obtained through the fusion function F. fuse (·)Calculate the physical tampering index R(t) of the computing device. The analysis module is also used to fine-tune the baseline pattern matrix W and / or the parameters of the GNN model through an incremental learning algorithm after confirming that the unmanned equipment is in a normal state.

[0039] Specifically, this system performs intelligent analysis by combining a pre-trained baseline model with real-time detection. In the baseline model construction phase, the system uses Non-negative Matrix Factorization (NMF) or Robust Principal Component Analysis (RPCA) algorithms to train a large number of historical dynamic entropy matrices under normal conditions, extracting a baseline pattern matrix W representing the core and stable physical characteristics of the device. Simultaneously, it uses Graph Neural Networks (GNNs) to train historical dynamic physical fingerprint sequences, learning the normal evolution patterns of node fingerprints and the baseline features of the overall graph structure. In the real-time detection phase, the system uses the current dynamic entropy matrix P... current (t) is input into a sparse coding algorithm, attempting to represent it as a linear combination of the baseline pattern matrix W, and the reconstruction error is calculated. Simultaneously, the current dynamic physical fingerprint map G(t) is input into the pre-trained GNN model to calculate the map anomaly score δ. G (t), and finally the fusion function combines ∈(t) and δ G Multiple indicators, such as (t), are integrated into a continuous physical tampering index R(t), which enables precise quantification of the risk of physical tampering of equipment.

[0040] The control module is used to classify risk levels based on the physical tampering index and execute corresponding hierarchical control strategies. Specifically, the control module is used to: map the physical tampering index R(t) to a risk level that includes at least four levels: security, low-risk warning, medium-risk isolation, and high-risk blocking; and automatically trigger and execute corresponding hierarchical control strategies based on the risk level. These strategies include, but are not limited to, recording abnormal logs, sending early warning notifications, adjusting data transmission frequency, initiating fingerprint collection logic reconstruction, restricting resource access, forcing device self-tests, cutting off communication, locking functions, or triggering data self-destruction programs. The high-risk blocking strategy includes immediately cutting off all communication connections between the unmanned equipment and the external network, locking the key functions of the unmanned equipment, or triggering encrypted data erasure programs.

[0041] Specifically, after manual confirmation that the equipment is in normal condition or has undergone legitimate maintenance, the system fine-tunes the baseline model parameters through incremental learning algorithms. This allows the system to adapt to subtle changes in physical characteristics caused by natural wear and tear, aging, or environmental changes, thereby reducing the false alarm rate. Based on a quantified physical tampering index R(t), the system maps it to preset multi-level risk levels (such as security, low-risk warning, medium-risk isolation, and high-risk blocking / self-destruction). For different levels, the system automatically triggers and executes refined hierarchical control strategies. For example, at low risk, only logs and warnings are recorded; at medium risk, fingerprint collection logic reconstruction, resource access restrictions, or mandatory self-checks are initiated; and at high risk, communication is immediately cut off, functions are locked, or even data self-destruction procedures are triggered. This layered response mechanism ensures that the most appropriate and effective measures are taken at different risk levels.

[0042] In summary, this invention provides an intelligent control system based on dynamic detection of device identifiers. Through deep perception and active enhancement mechanisms using multimodal physical fingerprints—specifically, deploying sensor arrays within unmanned equipment to collect inherent hardware noise fingerprints, micro-environment dynamic fingerprints, system-level behavioral fingerprints, and module-level communication features—and periodically injecting micro-perturbations through programmable hardware security units to generate dynamic response spectra, dynamically adjusting or reconstructing the physical fingerprint acquisition logic under high-risk conditions, this effectively solves the problem of traditional single, static identifiers being easily forged or bypassed. Furthermore, through intelligent analysis and risk quantification based on matrix factorization and graph neural networks—specifically, using nonnegative matrix factorization (NMF) or robust principal component analysis (RPCA) algorithms to extract baseline pattern matrices during the learning phase and utilizing graph neural networks (GNNs) to learn the normal topological behavior of the graph—the system can significantly improve the intelligence and accuracy of tamper detection.

[0043] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent control system based on dynamic detection of equipment identification, characterized in that, include: Perception module: Used to continuously collect multi-dimensional physical fingerprint data of unmanned equipment and dynamically reconstruct the physical fingerprint collection logic according to the risk level; Modeling module: used to model the physical fingerprint data into a dynamic entropy matrix and a dynamic physical fingerprint spectrum; Analysis module: used to perform real-time analysis on the dynamic entropy matrix and dynamic physical fingerprint spectrum based on a pre-trained baseline model to generate a physical tampering index; Control module: used to classify risk levels based on the physical tampering index and execute corresponding hierarchical control strategies; The perception module is specifically used to: deploy a sensor array within the unmanned equipment to collect hardware inherent noise fingerprints, micro-environment dynamic fingerprints, system-level behavioral fingerprints, and module-level communication characteristics; And through the built-in programmable hardware security unit, micro-perturbations are periodically injected into selected physical fingerprint sources to generate dynamic response feature spectra, and the physical fingerprint acquisition logic is dynamically adjusted or reconstructed when there is a high risk. The modeling module is specifically used for: characterizing and transforming the collected physical fingerprint data to generate multiple physical fingerprint entropy value sequences; and constructing the physical fingerprint entropy value sequences into a dynamic entropy matrix P( t) , where P( t) The element is at a specific time step. kΔt The j Entropy value of physical fingerprint features E j ( kΔt Furthermore, this matrix is ​​continuously updated over time; multiple sensor nodes or modules within the unmanned equipment are defined as nodes in a graph. V Construct static topological edges based on the static topological relationships between nodes. E S Dynamic association edges are constructed based on the real-time association strength of the physical fingerprints between nodes. E D ( t To form a dynamic physical fingerprint spectrum. ; The analysis module includes a training unit for building a baseline model. Specifically, the training unit is used to: train the historical dynamic entropy matrix using nonnegative matrix factorization or robust principal component analysis algorithms to extract the baseline pattern matrix W of the device; and train the historical dynamic physical fingerprint sequence using a graph neural network (GNN) to learn the normal evolution pattern of node fingerprints and the baseline features of the graph structure. The analysis module includes an evaluation unit for real-time detection. Specifically, the evaluation unit is used to: employ a sparse coding algorithm to attempt to optimize the current dynamic entropy matrix P. current (t) represents a linear combination of the baseline pattern matrix W, and its reconstruction error is calculated. The sparse coding algorithm is as follows: The current dynamic physical fingerprint G(t) is input into the pre-trained GNN model to calculate the anomaly score. And combined with the reconstruction error and spectral anomaly score Indicators, through the fusion function F fuse (•) Physical tampering index of computing device R(t).

2. The intelligent control system based on dynamic detection of equipment identification as described in claim 1, characterized in that, The analysis module is also used to fine-tune the baseline pattern matrix W and / or the parameters of the GNN model through an incremental learning algorithm after confirming that the unmanned equipment is in a normal state.

3. The intelligent control system based on dynamic detection of equipment identification as described in claim 1, characterized in that, The control module is specifically used to: map the physical tampering index R(t) to a risk level that includes at least four levels: security, low-risk warning, medium-risk isolation, and high-risk blocking; Based on the risk level, the system automatically triggers and executes corresponding hierarchical control strategies, including recording abnormal logs, sending early warning notifications, adjusting data transmission frequency, initiating fingerprint collection logic reconstruction, restricting resource access, forcing device self-tests, cutting off communication, locking functions, or triggering data self-destruction procedures.

4. The intelligent control system based on dynamic detection of equipment identification as described in claim 1, characterized in that, The inherent hardware noise fingerprint includes the power consumption ripple spectrum of the CPU / GPU, the clock jitter characteristics of the internal crystal oscillator, or the transient response characteristics of the RF module; the microenvironment dynamic fingerprint includes the internal micron-level temperature distribution pattern or vibration resonance response spectrum; the system-level behavioral fingerprint includes the transient voltage / current response characteristic curve of the power rail; the module-level communication characteristics include the data transmission timing characteristics between internal modules or the differences in protocol stack implementation.

5. The intelligent control system based on dynamic detection of equipment identification as described in claim 1, characterized in that, The process of characterizing and transforming the acquired physical fingerprint data includes calculating the sample entropy or approximate entropy of the time-series signal to serve as the entropy value E of the physical fingerprint feature. j (t).

6. The intelligent control system based on dynamic detection of equipment identification according to claim 3, characterized in that, The high-risk blocking strategy includes immediately cutting off all communication connections between the unmanned equipment and external networks, locking the key functions of the unmanned equipment, or triggering an encrypted data erasure procedure.

Images