A robot data trusted collection method and system based on TPCM

Abstract

The application discloses a kind of robot data trusted collection method and system based on TPCM, belong to robot data security technical field.System includes TPCM trusted root module, multi-modal sensor uniform description module, static trusted measurement module, dynamic trusted measurement module, comprehensive trusted decision module, trusted data acquisition module, concept drift detection module and differential privacy publishing module.Method includes: TPCM is powered on to carry out firmware measurement in priority;Establish sensor uniform description model;Calculate static trusted degree and dynamic trusted degree, determine sensor trusted by particle swarm optimization fusion;Build trust chain, and measure equipment step by step on collection path;Concept drift is detected using ADWIN and EDDM double detector and adaptive update feature;Add adaptive Gaussian noise to realize differential privacy publishing.The application guarantees data source trusted by hardware trusted root, realizes robot data acquisition whole process of safe trusted and privacy protection by combining dynamic measurement and concept drift detection.

Description

Technical Field

[0001] This invention belongs to the field of robot data security technology, specifically relating to a reliable robot data acquisition method and system based on a Trusted Platform Control Module (TPCM). It is applicable to multimodal sensor data acquisition scenarios such as industrial robots, service robots, and special robots, ensuring the safety and reliability of the data acquisition process. Background Technology

[0002] With the rapid development of robotics technology, robot systems have become key equipment in fields such as intelligent manufacturing, home services, medical assistance, and autonomous driving. During operation, robots generate massive amounts of multimodal data, including images, point clouds (collected by LiDAR or depth cameras), force perception, and voice. This data is not only a core element of robot intelligence but also the foundation for realizing the value of data in the Industrial Internet. However, robots operate in open environments, and the data collection process faces severe security threats: sensors may be deceived or interfered with, communication links may be eavesdropped on or tampered with, and malicious nodes may inject false data. Existing data collection solutions mostly focus on privacy protection, trust management, and identity verification, but often neglect the trustworthiness measurement of the nodes themselves, failing to fundamentally guarantee the trustworthiness of the data source.

[0003] Scholars at home and abroad have proposed some data acquisition security schemes, such as data acquisition privacy protection schemes based on deep learning and acquisition node selection schemes based on trust degree calculation. However, these schemes have the following shortcomings: (1) They lack hardware-level root of trust support and cannot guarantee the trustworthiness of the nodes themselves; (2) They do not consider the dynamic changes and concept drift of sensor data and are difficult to adapt to complex environments; (3) They lack active measurement and dynamic monitoring of the entire data acquisition process and cannot detect and handle anomalies in real time.

[0004] In recent years, my country has made significant breakthroughs in the field of trusted computing. Academician Shen Changxiang proposed the Trusted Computing 3.0 framework, which uses TPCM (Trusted Platform Control Module) as the hardware root of trust and constructs a dual-system proactive immune architecture of "computing + protection." The release of the national standard GB / T40650-2021, "Trusted Platform Control Module," has laid the foundation for the industrial application of independent trusted computing technologies. Integrating TPCM root of trust technology into robot data acquisition systems is expected to fundamentally solve the trustworthiness problem of data sources. Summary of the Invention

[0005] The technical problems that this invention aims to solve are: (1) lack of hardware-level root of trust support, which makes it impossible to guarantee the trustworthiness of the node itself; (2) failure to consider the dynamic changes and concept drift of sensor data, which makes it difficult to adapt to complex environments; (3) lack of active measurement and dynamic monitoring of the entire data acquisition process, which makes it impossible to detect and process anomalies in real time.

[0006] The purpose of this invention is to provide a reliable robot data acquisition method and system based on TPCM. By introducing the TPCM root of trust, a three-stage reliable execution environment from hardware to application is constructed to realize the unified description and reliable measurement of multimodal sensors. Combined with identity authentication and dynamic reliable detection, the security and reliability of the entire data acquisition process are ensured.

[0007] This invention provides a reliable robot data acquisition method and system based on TPCM, belonging to the field of robot data security technology. The system includes a TPCM trusted root module, a multimodal sensor unified description module, a static trusted measurement module, a dynamic trusted measurement module, a comprehensive trusted judgment module, a trusted data acquisition module, a concept drift detection module, and a differential privacy publishing module. The method includes: prioritizing TPCM power-on for firmware measurement; establishing a sensor unified description model; calculating static and dynamic trustedness, and determining sensor trustworthiness through particle swarm optimization fusion; constructing a trust chain and measuring each device along the acquisition path step by step; using ADWIN and EDDM dual detectors to detect concept drift and adaptively update features; and adding adaptive Gaussian noise to achieve differential privacy publishing. This invention ensures the trustworthiness of the data source through a hardware trusted root, and combines dynamic measurement and concept drift detection to achieve secure, reliable, and privacy-protected data acquisition throughout the entire robot data acquisition process.

[0008] To achieve the above objectives, the present invention adopts the following technical solution.

[0009] A robot data trusted acquisition system based on TPCM includes: a TPCM trusted root module, which serves as the trust root of the system and is powered on before the robot's main CPU. It is connected to the firmware memory, hardware configuration acquisition unit, and trusted agent in the operating system through the SPI master control bus, low-speed communication bus, and high-speed communication bus, respectively, and is used to actively measure and control the startup code, firmware, operating system, and application.

[0010] The multimodal sensor unified description module connects to various sensors, as well as the static and dynamic trust measurement modules. It is used to establish a unified description model for various sensors mounted on the robot, including the sensor's inherent attributes, configuration attributes, operating status, and trust history.

[0011] The static reliability measurement module is connected to the multimodal sensor unified description module and the comprehensive reliability determination module, respectively, and is used to calculate static reliability based on the inherent properties of the sensor.

[0012] The dynamic credibility measurement module is connected to the multimodal sensor unified description module and the comprehensive credibility judgment module, respectively. It is used to update the dynamic credibility in real time based on the sensor's operating status and historical interaction records using Bayesian inference.

[0013] The comprehensive credibility determination module is connected to the static credibility measurement module, the dynamic credibility measurement module, and the credibility data acquisition module, respectively. It is used to perform weighted fusion of static credibility and dynamic credibility using the particle swarm optimization algorithm to obtain a comprehensive credibility value, and compare it with a preset threshold to determine whether the sensor is credible.

[0014] The trusted data acquisition module is connected to the integrated trusted judgment module, the TPCM trusted root module and the concept drift detection module respectively. It is used to build a trust chain based on the TPCM trusted root to measure the devices on the data acquisition path step by step, so as to ensure the complete and trusted transmission of data from the sensor to the processing unit.

[0015] The concept drift detection module is connected to the trusted data acquisition module and the differential privacy publishing module, respectively. It is used to detect concept drift in the sensor data stream in real time using ADWIN and EDDM dual detectors. When drift is detected, feature reselection and model update are triggered.

[0016] The differential privacy publishing module, connected to the concept drift detection module, is used to add adaptive Gaussian noise to the collected data to meet the Rényi differential privacy protection requirements and realize data publishing under privacy protection.

[0017] Furthermore, the unified description model of the multimodal sensor is represented as a quadruple:

[0018]

[0019] in These are inherent attributes, including sensor model, manufacturer, factory ID, hardware version, and security chip identifier; Configuration attributes include sampling rate, resolution, communication protocol, and calibration parameters; This indicates the operating status, including current output value, power consumption, temperature, communication latency, and error count. It is a trusted historical record, containing a historical credibility sequence and timestamps.

[0020] Furthermore, the static credibility calculation formula is as follows:

[0021]

[0022] in This is the factory default trust value. For security level, To enhance hardware tamper resistance, The weighting coefficients are determined by the analytic hierarchy process (AHP) and satisfy the following conditions: .

[0023] Furthermore, the dynamic credibility update employs Bayesian inference:

[0024]

[0025] Introducing a time decay factor:

[0026]

[0027] in To adjust the sliding window size, This represents the attenuation rate.

[0028] Furthermore, the comprehensive reliability determination employs a particle swarm optimization algorithm to adaptively learn weights. :

[0029]

[0030] The fitness function of the particle swarm optimization algorithm is the reciprocal of the false positive rate, and the optimal weights are obtained through iterative optimization. .

[0031] Furthermore, the TPCM trusted root module implements a three-stage trusted execution environment:

[0032] Standby phase: TPCM is powered on first, reads the BIOS / UEFI code in the firmware memory through the SPI master bus, calculates the SM3 hash value and compares it with the base value, and releases system control after confirming that the boot code is trustworthy;

[0033] Startup phase: Through a trusted proxy embedded in the startup code, information such as CPU configuration, memory configuration, and external devices is collected and sent to TPCM for verification via a low-speed communication bus;

[0034] During operation: It interacts with a trusted agent in the operating system via a high-speed communication bus to monitor memory code, system configuration, and process behavior in real time. When an anomaly is detected, it performs physical control according to policies (such as disconnecting USB ports or shutting down network devices).

[0035] Furthermore, the process of building a trust chain by the trusted data acquisition module is as follows: TPCM measures BootROM → BootROM measures Bootloader → Bootloader measures OS kernel → OS kernel measures application; before data acquisition, TPCM measures the sensors, CAN controller, gateway and other devices on the acquisition path in sequence. If all devices pass the measurement, data flow is allowed.

[0036] Furthermore, the concept drift detection module employs dual detectors ADWIN and EDDM. When both are triggered simultaneously, concept drift is considered to have occurred, triggering feature reselection. The feature selection method combines variance thresholding with SelectKBest to recalculate feature importance on the latest window data and update the feature set.

[0037] Furthermore, the differential privacy publishing module adopts the Rényi differential privacy mechanism, dynamically adjusting the privacy budget through a state-space model:

[0038]

[0039] in , The online convex optimization problem is solved using the DP-TOFW algorithm to minimize the regret value. This achieves a balance between privacy protection and data availability.

[0040] A reliable robot data acquisition method based on TPCM, applied to the above system, is characterized by comprising the following steps:

[0041] Step 1: System Initialization and TPCM Self-Test

[0042] The TPCM module powers on first and performs a self-test, including cryptographic algorithm module detection and internal code integrity verification. After the self-test passes, it reads the BIOS / UEFI code in the firmware memory via the SPI bus, calculates the SM3 hash value, and compares it with the baseline value stored in the TPCM secure flash memory. If they match, the main CPU is allowed to power on; otherwise, it prevents startup or enters untrusted mode.

[0043] Step 2: Unified Description and Reference Value Input of Multimodal Sensors

[0044] A unified description is provided for all types of sensors mounted on the robot, and a sensor information table is constructed, including inherent attributes, configuration attributes, and operating status fields. Static baseline values ​​(factory trust, security level) and initial dynamic parameters of the sensors are entered in a trusted environment.

[0045] Step 3: Static Trustworthiness Measurement

[0046] Based on the inherent properties of the sensor, the static reliability is calculated using the analytic hierarchy process (AHP). .

[0047] Step 4: Dynamic Trust Measurement

[0048] During robot operation, real-time data on sensor status (output values, power consumption, temperature, etc.) is collected, and Bayesian inference is used to update the dynamic reliability. A time decay factor is introduced to give higher weight to recent data.

[0049] Step 5: Comprehensive Reliability Assessment

[0050] By inputting static and dynamic confidence levels into the particle swarm optimization algorithm, the optimal fusion weights are learned. Calculate the overall credibility ;like ( If the threshold is set (preset), the sensor is deemed trustworthy; otherwise, it is marked as untrustworthy and an alarm is triggered.

[0051] Step 6: Trust Chain Construction and Data Collection

[0052] Using TPCM as the root of trust, a trusted chain is built from BootROM to application; during data acquisition, all devices on the acquisition path are measured sequentially, and data flow is allowed if all pass; data is transmitted from the sensor to the main processor via CAN bus and gateway, and is continuously monitored by TPCM during the process.

[0053] Step 7: Concept Drift Detection and Adaptive Feature Selection

[0054] The sensor data stream is monitored in real time using dual detectors ADWIN and EDDM. When concept drift is detected, the nearest window sample is obtained, feature selection is performed again (variance threshold + SelectKBest), and the feature set and online ensemble model are updated.

[0055] Step 8: Differential Privacy Release

[0056] Adaptive Gaussian noise is added to the collected raw data to meet Rényi differential privacy requirements; the noise scale is optimized using the DP-TOFW algorithm to maximize data availability while protecting privacy; and the published data is stored or transmitted to the cloud.

[0057] Step 9: Exception Handling and Feedback

[0058] If an anomaly is detected at any stage (such as a decrease in sensor confidence value, concept drift anomaly, hardware tampering, etc.), TPCM will handle it according to the preset policy, including logging, alarming, isolating ports, cutting off power, etc., and report the abnormal event to the Trust Management Center.

[0059] The present invention has the following beneficial effects:

[0060] 1. Hardware-level root of trust guarantee: The TPCM trusted chip, which conforms to national standards, is used as the root of trust. It is put into operation before the CPU and actively measures the firmware, operating system and application, which fundamentally solves the problem that the trustworthiness of the node itself cannot be guaranteed in traditional solutions.

[0061] 2. Unified Measurement of Multimodal Sensors: A unified description model for sensors is established, which integrates static attributes and dynamic behaviors. The particle swarm optimization algorithm is used to adaptively learn weights, which improves the accuracy and robustness of reliable judgment.

[0062] 3. Trusted execution environment throughout the entire lifecycle: Through a three-stage architecture, it achieves proactive protection throughout the entire process from standby, startup to operation, and performs step-by-step measurement on devices along the data acquisition path to ensure the trustworthiness of the data source.

[0063] 4. Concept drift adaptive detection: The system uses dual detectors to monitor changes in data distribution in real time and dynamically adjusts feature selection and model parameters, thereby improving the system's adaptability to complex environments.

[0064] 5. Differential Privacy Protection Release: Adopting the Rényi differential privacy mechanism, the privacy budget is dynamically adjusted, and the balance between privacy protection and data availability is optimized online to meet the privacy needs in data sharing.

[0065] 6. Proactive defense and intelligent response: TPCM has independent physical control capabilities. When an anomaly is detected, it can cut off the port or power in real time to prevent the spread of attacks. At the same time, it realizes situational awareness and intelligent decision-making through the Trusted Management Center. Attached Figure Description

[0066] Figure 1 This is the overall system architecture diagram.

[0067] Figure 2 This is a schematic diagram of the TPCM three-phase trusted execution environment.

[0068] Figure 3 This is a flowchart of the multimodal sensor reliability measurement process.

[0069] Figure 4 This is a diagram illustrating the construction of a trust chain for trusted data collection.

[0070] Figure 5 This is a flowchart of concept drift detection and adaptive feature selection.

[0071] Figure 6 This is a flowchart of the differential privacy publishing module.

[0072] Figure 7 This is the overall flowchart of the method of the present invention. Detailed Implementation

[0073] Figure 1This diagram illustrates the overall architecture of the system of the present invention, showcasing the overall composition and inter-module relationships of the TPCM-based robot data trusted acquisition system. The system consists of a TPCM trusted root module, a multimodal sensor unified description module, a static trusted measurement module, a dynamic trusted measurement module, a comprehensive trusted judgment module, a trusted data acquisition module, a concept drift detection module, and a differential privacy publishing module.

[0074] The TPCM Trusted Root Module includes an Active Measurement Engine, Trusted Storage (PCR), Trusted Reporting Module, Physical Control (GPIO), and three types of communication interfaces (SPI master controller, I2C low speed, USB / PCIe high speed).

[0075] TPCM implements three-stage measurement: BIOS / UEFI firmware is measured during standby, hardware configuration and OS are measured during startup, and memory and processes are measured during runtime.

[0076] The unified description module for multimodal sensors includes sensor inherent properties, configuration properties, operating status, and trusted historical records;

[0077] The sensor group includes cameras, lidar, force sensors, temperature sensors, etc.

[0078] The communication unit includes a CAN bus, Ethernet, a 5G module, and a trusted agent;

[0079] The system interacts with the cloud-based trusted management center via secure communication to enable policy distribution, log collection, and situational analysis.

[0080] Figure 2 The architecture and working principle of the TPCM three-phase trusted execution environment are demonstrated.

[0081] Three phases: Standby, Boot, and Runtime.

[0082] Three pathways: Master control measurement and control bus (SPI), low-speed communication bus (I2C), and high-speed communication bus (USB / PCIE).

[0083] Standby phase: TPCM is powered on first, reads firmware Flash via SPI bus, calculates SM3 hash value and compares it with base value, and enables main power supply after passing comparison.

[0084] During the boot phase: The trusted agent in the BootROM collects hardware configuration information via the I2C bus, and after verification by the TPCM, it decides whether to load the OS;

[0085] During operation: In the OS, the TSB monitors memory code hash, system configuration, and process behavior in real time via the USB / PCIE bus. The TPCM runtime behavior analysis model detects anomalies and handles them according to the strategy.

[0086] Each of the three buses performs its own function, enabling active measurement and control throughout the entire process from power-on to operation;

[0087] Figure 3 It demonstrates the complete process of trust measurement for multimodal sensors;

[0088] The process begins with building a unified description model for sensors, which includes inherent attributes, configuration attributes, operational status, and trusted history.

[0089] Static trust metrics are based on factory trust, security level, and hardware tamper resistance, and the weights are determined using the analytic hierarchy process (AHP).

[0090] Dynamic reliability metrics acquire sensor operating status in real time, are updated using Bayesian inference, and incorporate a time decay factor to give higher weight to recent data.

[0091] The particle swarm optimization algorithm learns the optimal fusion weight α and iteratively optimizes it to maximize fitness (minimize the misclassification rate).

[0092] Calculate overall credibility ;

[0093] With threshold Compare, If a data collection is deemed trustworthy, data collection is allowed; otherwise, if the data collection is deemed untrustworthy, an alarm is triggered or data collection is denied.

Claims

1. A reliable robot data acquisition system based on TPCM, characterized in that: include, The TPCM Trust Root Module, as the system's trust root, is powered on before the robot's main CPU. It connects to the firmware memory, hardware configuration acquisition unit, and trusted agent in the operating system via the SPI master control bus, low-speed communication bus, and high-speed communication bus, respectively, and is used to actively measure and control the startup code, firmware, operating system, and application. The multimodal sensor unified description module connects to various sensors, as well as the static and dynamic trust measurement modules. It is used to establish a unified description model for various sensors mounted on the robot, including the sensor's inherent attributes, configuration attributes, operating status, and trust history. The static reliability measurement module is connected to the multimodal sensor unified description module and the comprehensive reliability determination module, respectively, and is used to calculate static reliability based on the inherent properties of the sensor. The dynamic credibility measurement module is connected to the unified description model of multimodal sensors and the comprehensive credibility judgment module, respectively. It is used to update the dynamic credibility in real time based on the sensor's operating status and historical interaction records using Bayesian inference. The comprehensive credibility determination module is connected to the static credibility measurement module, the dynamic credibility measurement module, and the credibility data acquisition module, respectively. It is used to perform weighted fusion of static credibility and dynamic credibility using the particle swarm optimization algorithm to obtain a comprehensive credibility value, and compare it with a preset threshold to determine whether the sensor is credible. The trusted data acquisition module is connected to the integrated trusted judgment module, the TPCM trusted root module and the concept drift detection module respectively. It is used to build a trust chain based on the TPCM trusted root to measure the devices on the data acquisition path step by step, so as to ensure the complete and trusted transmission of data from the sensor to the processing unit. The concept drift detection module is connected to the trusted data acquisition module and the differential privacy publishing module, respectively. It is used to detect concept drift in the sensor data stream in real time using ADWIN and EDDM dual detectors. When drift is detected, feature reselection and model update are triggered. The differential privacy publishing module, connected to the concept drift detection module, is used to add adaptive Gaussian noise to the collected data to meet the Rényi differential privacy protection requirements and realize data publishing under privacy protection.

2. The reliable robot data acquisition system based on TPCM according to claim 1, characterized in that: The unified description model of the multimodal sensor is represented by a quadruple: ； in These are inherent attributes, including sensor model, manufacturer, factory ID, hardware version, and security chip identifier; Configuration attributes include sampling rate, resolution, communication protocol, and calibration parameters; This indicates the operating status, including current output value, power consumption, temperature, communication latency, and error count. It is a trusted historical record, containing a historical credibility sequence and timestamps.

3. The reliable robot data acquisition system based on TPCM according to claim 1, characterized in that: The static credibility calculation formula is as follows: ； in This is the factory default trust value. For security level, To enhance hardware tamper resistance, The weighting coefficients are determined by the analytic hierarchy process (AHP) and satisfy the following conditions: .

4. The reliable robot data acquisition system based on TPCM according to claim 1, characterized in that: The dynamic credibility update employs Bayesian inference: ； Introducing a time decay factor: ； in To adjust the sliding window size, This represents the attenuation rate.

5. The reliable robot data acquisition system based on TPCM according to claim 1, characterized in that: The comprehensive credibility determination employs a particle swarm optimization algorithm for adaptive weight learning. : ； The fitness function of the particle swarm optimization algorithm is the reciprocal of the false positive rate, and the optimal weights are obtained through iterative optimization. .

6. The reliable robot data acquisition system based on TPCM according to claim 1, characterized in that: The TPCM trusted root module implements a three-stage trusted execution environment: Standby phase: TPCM is powered on first, reads the BIOS / UEFI code in the firmware memory through the SPI master bus, calculates the SM3 hash value and compares it with the base value, and releases system control after confirming that the boot code is trustworthy; Startup phase: Through a trusted proxy embedded in the startup code, information such as CPU configuration, memory configuration, and external devices is collected and sent to TPCM for verification via a low-speed communication bus; During operation: It interacts with a trusted agent in the operating system via a high-speed communication bus to monitor memory code, system configuration, and process behavior in real time, and performs physical control according to policies when anomalies are detected.

7. The reliable robot data acquisition system based on TPCM according to claim 1, characterized in that: The process of building a trust chain by the trusted data acquisition module is as follows: TPCM measures BootROM, BootROM measures Bootloader, Bootloader measures OS kernel, and OS kernel measures application. Before data acquisition, TPCM measures the sensors, CAN controller, and gateway devices on the acquisition path in sequence. If all devices pass the measurement, data flow is allowed.

8. The reliable robot data acquisition system based on TPCM according to claim 1, characterized in that: The concept drift detection module uses ADWIN and EDDM dual detectors. When both are triggered simultaneously, it is considered that concept drift has occurred, and feature reselection is triggered. The feature selection method combines variance thresholding with SelectKBest to recalculate feature importance on the latest window data and update the feature set.

9. A reliable robot data acquisition system based on TPCM according to claim 1, characterized in that: The differential privacy publishing module adopts the Rényi differential privacy mechanism, which dynamically adjusts the privacy budget through a state-space model: ； in , The online convex optimization problem is solved using the DP-TOFW algorithm to minimize the regret value. This achieves a balance between privacy protection and data availability.

10. A reliable robot data acquisition method based on TPCM for implementing the system as described in any one of claims 1-9, applied to the above-mentioned system, characterized in that, Includes the following steps: Step 1: System initialization and TPCM self-test; The TPCM module powers on first and performs a self-test, including cryptographic algorithm module detection and internal code integrity verification. After the self-test passes, it reads the BIOS / UEFI code in the firmware memory via the SPI bus, calculates the SM3 hash value, and compares it with the baseline value stored in the TPCM secure flash memory. If they match, the main CPU is allowed to power on; otherwise, it prevents startup or enters untrusted mode. Step 2: Unified description and baseline value input for multimodal sensors; A unified description is provided for all types of sensors mounted on the robot, and a sensor information table is constructed, including inherent attributes, configuration attributes, and operating status fields; static baseline values ​​(factory trust, security level) and initial dynamic parameters of the sensors are entered in a trusted environment; Step 3: Static credibility measurement; Based on the inherent properties of the sensor, the static reliability is calculated using the analytic hierarchy process (AHP). ; Step 4: Dynamic Trust Measurement; During robot operation, real-time sensor status data is collected, and Bayesian inference is used to update the dynamic reliability. A time decay factor is introduced to give higher weight to recent data; Step 5: Comprehensive credibility assessment; By inputting static and dynamic confidence levels into the particle swarm optimization algorithm, the optimal fusion weights are learned. Calculate the overall credibility ;like ( If the threshold is set (preset), the sensor is deemed trustworthy; otherwise, it is marked as untrustworthy and an alarm is triggered. Step 6: Trust Chain Construction and Data Collection; Using TPCM as the root of trust, a trusted chain is built from BootROM to the application; During data acquisition, all devices along the acquisition path are measured sequentially, and data flow is allowed if all of them pass. Data is transmitted from the sensor to the main processor via the CAN bus and gateway, and is continuously monitored by the TPCM during the process. Step 7: Concept drift detection and adaptive feature selection; The sensor data stream is monitored in real time using dual detectors, ADWIN and EDDM; when concept drift is detected, the nearest window sample is acquired, feature selection is performed again, and the feature set and online ensemble model are updated. Step 8: Differential privacy release; Adaptive Gaussian noise is added to the collected raw data to meet Rényi differential privacy requirements; The DP-TOFW algorithm is used to optimize noise scale, maximizing data availability while protecting privacy; published data is stored or transmitted to the cloud. Step 9: Exception handling and feedback; If an anomaly is detected at any stage (such as a decrease in sensor confidence value, concept drift anomaly, hardware tampering, etc.), TPCM will handle it according to the preset policy, including logging, alarming, isolating ports, cutting off power, etc., and report the abnormal event to the Trust Management Center.

Images