Electronic evidence system for non-site environmental supervision of pollution sources combined with hash operation

Abstract

The application relates to the technical field of electronic data digital processing, and discloses a pollution source non-site environmental supervision electronic evidence system combined with hash operation, which comprises the following steps: generating a pseudo-random cryptographic sequence by using a true random number generator, extracting a cross-modal consistency tensor, and generating a dynamic semantic hash through hash operation; adopting an interactive multi-model and an unscented Kalman filter to decouple physical link intrinsic noise, extract adaptive physical micro-entropy; deriving a hardware-level device private key in a trusted execution environment based on a physically unclonable function; performing strong binding signature on multi-dimensional trust features and a time stamp, generating a living body electronic evidence certificate, and completing synchronism and anti-replay verification through a cloud digital twin model. The application extends a trust chain to a bottom layer physical sensing link, realizes identification of AI injected fake data and physical layer tampering behavior, constructs a full-link trust closed loop from a physical source to a cloud judicial solidification, and guarantees the authenticity and non-repudiation of non-site environmental supervision data.

Description

Technical Field

[0001] This invention relates to the field of electronic data digital processing technology, specifically to an electronic evidence collection system for off-site environmental monitoring of pollution sources that combines hash operations. Background Technology

[0002] As ecological and environmental supervision evolves towards intelligence, off-site environmental monitoring of pollution sources has become an important business model. This model relies on IoT sensor terminals deployed at the front end of polluting enterprises to collect and remotely transmit multimodal environmental monitoring data in real time, including exhaust gas values, operating current, and on-site images. Electronic evidence collection technology, as a key support for ensuring the judicial validity of off-site monitoring data, is defined by using cryptographic methods such as hash operations, digital signatures, and hardware-level encryption to extract, solidify, and encapsulate collected digital evidence, aiming to ensure the authenticity, integrity, and non-repudiation of environmental monitoring data throughout its entire lifecycle.

[0003] Current technologies in this field primarily rely on standard cryptographic protocols based on Public Key Infrastructure (PKI) and purely software-based data storage methods. The mainstream solutions typically receive digital signals from sensors at the edge computing gateway, calculate a business data digest using a general one-way hash function, and digitally sign it using a software private key stored in the general operating system's memory. Finally, the encrypted payload is uploaded to the cloud or written to the blockchain for solidification. Representative related work focuses on optimizing the backend data storage structure or building transmission channels to enhance the concurrent processing capabilities of cloud-based data on-chain. These technical approaches emphasize ensuring the tamper-proof nature of data during network transmission and cloud storage.

[0004] However, existing electronic forensics solutions are not yet perfect and their ability to handle high-tech forgery methods remains insufficient. The main problem is that existing systems are unable to resist high-tech injection tampering and deepfake attacks targeting the measurement and sensing side. This is because the trust starting point of existing forensics mechanisms is usually established after the data has completed analog-to-digital conversion and entered the digital logic layer, failing to extend the trust chain down to the source of the underlying physical sensing link. This means that when some polluting enterprises use data interception and simulation transmission equipment to directly inject forged data synthesized by generative artificial intelligence (such as GAN networks) into the front-end probes, because such forged data can maintain high semantic consistency in exhaust gas values, operating current, and monitoring video footage, and reproduce logically consistent operating states, existing systems will treat it as legitimate underlying input and execute the subsequent standard signature solidification process. This causes the anti-counterfeiting trust chain to break at the very front-end physical sensing source, resulting in the system-solidified electronic evidence having a formally legitimate signature, but its represented physical state is false, ultimately weakening the enforcement effectiveness and judicial rigor of off-site environmental supervision. Therefore, there is an urgent need for an electronic evidence collection system for off-site environmental monitoring of pollution sources that incorporates hash operations to solve the above problems. Summary of the Invention

[0005] To address the problems in related technologies, this invention provides an electronic evidence collection system for off-site environmental monitoring of pollution sources that incorporates hash operations, thereby overcoming the aforementioned technical problems in existing related technologies.

[0006] To solve the aforementioned technical problem, the present invention is achieved through the following technical solution: This invention provides an electronic evidence collection system for off-site environmental monitoring of pollution sources that combines hash operations. Specifically, it includes: a multimodal dynamic semantic hash generation module, used to generate pseudo-random cryptographic sequences using a true random number generator, convert the pseudo-random cryptographic sequences into modulation instructions, inject them into the front-end device, demodulate and calculate the cross-consistency tensor, and perform dimensionality reduction and quantization using a symbolic random projection algorithm combined with adaptive dead-zone quantization to generate dynamic semantic hashes; an adaptive physical micro-entropy extraction module, used to collect the intrinsic physical feature vectors of the front-end sensor circuits, fuse interactive multi-model and unscented Kalman filtering to calculate the measurement residual vector, and decouple malicious features to generate adaptive hybrid physical micro-entropy random numbers; and a device privacy... The key derivation module is used to obtain the static physical response output by the physically unclonable function module, and calls the two-stage key derivation function through the security platform abstraction layer within the trusted execution environment to derive the asymmetric device private key; the evidence storage encapsulation and hardware signature module is used to concatenate the dynamic semantic hash, the adaptive hybrid physical micro-entropy random number, the system timestamp, and the terminal unique identifier into a payload to be signed, and call the asymmetric device private key to perform digital signature to generate a living electronic evidence certificate; the cloud independent verification module is used to perform independent verification of the living electronic evidence certificate by using a pre-calibrated device digital twin trust model and performing online updates through a parameter correction matrix to generate a legally valid result.

[0007] As a preferred embodiment of the electronic evidence collection system for off-site environmental monitoring of pollution sources combined with hash operations as described in this invention, the step of the multimodal dynamic semantic hash generation module generating dynamic semantic hashes includes: The cross-consistency tensor is flattened into a one-dimensional correlation feature vector, and adaptive dead-zone quantization is performed on the one-dimensional correlation feature vector using a preset dead-zone threshold to shield noise fluctuations. Using a static pseudo-random projection matrix generated based on hardware identifier codes, a symbolic random projection operation is performed on the quantized feature vector to generate a binary feature vector. The binary feature vector is concatenated with the original bit stream, and a one-way hash operation is performed to output the dynamic semantic hash.

[0008] As a preferred embodiment of the electronic evidence collection system for off-site environmental monitoring of pollution sources combined with hash operations as described in this invention, the demodulation calculation of the cross-consistency tensor includes: A physical information neural network model is constructed, which is pre-trained offline using positive samples generated by hardware-in-the-loop simulation and negative samples generated by a generative adversarial network. During the system initialization phase, the initial default values ​​of the system's equivalent time constant and the link's static gain are calibrated using a step test signal. During the forward inference phase, the physical dynamics transfer equation is constrained by the initial default values, and a self-supervised solution is performed in conjunction with an online-generated pseudo-random cryptographic sequence to decouple high-frequency feature components and calculate the cross-consistency tensor.

[0009] As a preferred embodiment of the electronic evidence collection system for off-site environmental monitoring of pollution sources combined with hash operations as described in this invention, the adaptive physical micro-entropy extraction module generates adaptive hybrid physical micro-entropy random numbers including: The measurement residual vector is transformed into a sparse optimization problem and solved to obtain the coefficient subset of the corresponding malicious sub-dictionary and calculate the projected energy, as well as the residual energy of the corresponding unmodeled signal. A decision weight function is constructed based on the projected energy, the residual energy, and the residual energy penalty coefficient. Calculate the overall tampering confidence score. When the overall tampering confidence score exceeds the preset lower limit, determine the adaptive hybrid physical micro-entropy random number.

[0010] As a preferred embodiment of the electronic evidence collection system for off-site environmental monitoring of pollution sources combined with hash operations as described in this invention, the step of acquiring the intrinsic physical feature vector includes: Inject a high-frequency composite physical detection signal containing wideband linear frequency modulated pulses and dual-frequency equal-amplitude sine waves into the sensor physical cable; The response signal is captured through the first channel, the frequency domain complex reflection coefficient is calculated, and the third-order intermodulation distortion power ratio based on dual-frequency equal-amplitude sine wave excitation is extracted. By suppressing the fundamental frequency interference through the second channel, the power spectral density of the sensor's intrinsic low-frequency electrical noise is calculated and the flicker noise slope constant is extracted. The intrinsic physical feature vector is determined by concatenating the third-order intermodulation distortion power ratio, the frequency domain complex reflection coefficient, and the flicker noise slope constant.

[0011] As a preferred embodiment of the electronic evidence collection system for off-site environmental monitoring of pollution sources that combines hash operations as described in this invention, the device private key derivation module includes: The static physical response is obtained through the unified interface of the security platform abstraction layer, and then concatenated with the factory security salt value to construct the initial key material. Within a trusted execution environment, a hash-based message authentication code key derivation function is executed. A pseudo-random key is generated through the extraction phase, and iterative computation is performed in the expansion phase in combination with specific application information to determine the output key material. The output key material is converted into an asymmetric device private key and loaded into a dedicated volatile secure register, followed by a forced erasure of the intermediate computation stack.

[0012] As a preferred embodiment of the electronic evidence collection system for off-site environmental monitoring of pollution sources combined with hash operations as described in this invention, the evidence storage and encapsulation and hardware signature module includes: The trusted real-time clock and hardware monotonic counter within the trusted execution environment are invoked and cross-validated with the cloud-based reference clock to generate a high-precision trusted system timestamp. The type length value encoding standard is used to serialize and encode each field of the payload to be signed, and dynamic field separators are injected at the boundaries of the encoding blocks; The payload is assembled within the encrypted isolated memory and read / write permissions are frozen in one direction. Then, the national cryptographic algorithm is called to execute the digital signature.

[0013] As a preferred embodiment of the electronic evidence collection system for off-site environmental monitoring of pollution sources combined with hash operations as described in this invention, after generating the live electronic evidence certificate, the evidence encapsulation and hardware signature module further performs the following: The payload to be signed, the digital signature, the firmware version hash, and the secure boot state machine vector are encapsulated into an independent electronic evidence compartment. The electronic evidence compartment is encrypted using a symmetric encryption algorithm and then pushed into a persistent asynchronous transmission queue based on a local encrypted file system.

[0014] As a preferred embodiment of the electronic evidence collection system for off-site environmental monitoring of pollution sources combined with hash operation as described in this invention, the cloud-based independent verification module introduces a cuckoo filter to perform concurrent existence verification on adaptive hybrid physical micro-entropy random numbers during verification, in order to intercept replay attacks.

[0015] As a preferred embodiment of the electronic evidence collection system for off-site environmental monitoring of pollution sources that combines hash operations as described in this invention, the device digital twin trust model includes: During the cold start phase, initial values ​​of temperature sensitivity coefficient, humidity correlation coefficient, and aging drift slope are obtained through laboratory pre-calibration parameters or big data prior libraries. During the verification phase, the parameter state vector of the digital twin trust model is corrected online using the recursive least squares method or the extended Kalman filter matrix based on the measured residuals.

[0016] The present invention has the following beneficial effects: 1. This invention injects cryptographic micro-modulation watermarks into the sensor's simulated reference voltage and the stroboscopic light of the monitoring video through a multimodal dynamic semantic hash generation module, and extracts the response features of the monitoring data to the out-of-band watermark. Since it is difficult to predict and reproduce such hidden physical layer fluctuations in real time, this invention can effectively identify injected forgery behavior from the source of physical perception, ensure the liveness attribute and authenticity of the evidence data, and thus solve the problem that existing technologies cannot identify high-tech injection tampering and AI forged data.

[0017] 2. This invention employs an adaptive physical micro-entropy extraction module to monitor the intrinsic electrical noise and impedance characteristics of the sensor-simulated link in real time, extending the trust chain to the physical medium layer. It can keenly detect malicious operations such as interception, bypassing, or replacement of physical cables, making up for the shortcomings of pure software encryption schemes in defending against physical layer tampering. This improves the robustness of the electronic forensics system and solves the problem that existing schemes have blind spots in supervision on the physical transmission link and cannot detect hardware tampering.

[0018] 3. This invention utilizes a physically unclonable function module built into the acquisition terminal to dynamically derive a unique device private key based on static manufacturing deviations of the chip. This private key resides within a trusted execution environment and possesses physical uniqueness and volatility. This not only eliminates the risk of malicious reading or firmware extraction of the private key at the hardware level but also establishes an immutable hardware root of trust, thereby ensuring the uniqueness and non-repudiation of the source of electronic evidence. This solves the problem of easily extracted or cloned device private keys leading to identity forgery.

[0019] 4. This invention uses a storage and encapsulation module and a hardware signature module to strongly bind semantic hash, physical micro-entropy, and high-precision timestamps in a secure and isolated environment. Combined with the digital twin model and anti-replay retrieval mechanism in the cloud-based independent verification module, it constructs a full-link trust closed loop from front-end physical perception to cloud-based judicial solidification, ensuring multiple consistency of electronic evidence in time, space, and generation environment, thereby strengthening the judicial validity of off-site environmental monitoring data.

[0020] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0022] Figure 1 This invention provides a schematic diagram of an electronic evidence collection system module for off-site environmental monitoring of pollution sources that incorporates hash operations.

[0023] Figure 2 This is a flowchart illustrating the electronic evidence collection method for off-site environmental monitoring of pollution sources that incorporates hash operations, as provided by the present invention.

[0024] Figure 3 This is a schematic diagram of the S5 process provided by the present invention. Detailed Implementation

[0025] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0026] Example 1 As high-tech fraud methods evolve, some polluting companies may employ data interception and simulation transmission equipment to directly inject compliant data synthesized by generative AI (such as GAN networks) into front-end probes. This data maintains high semantic consistency across exhaust gas values, operating current, and monitoring video footage; for example, it can replay a past compliant production process or fabricate a logically sound virtual operating status. Existing electronic forensics solutions are not yet perfect and lack the capacity to handle such high-tech fraud methods.

[0027] To solve the above technical problems, such as Figure 1 As shown, Embodiment 1 of the present invention provides an electronic evidence collection system for off-site environmental monitoring of pollution sources that combines hash operations, specifically including: The multimodal dynamic semantic hash generation module uses a true random number generator to generate pseudo-random cryptographic sequences, converts them into modulation instructions, injects them into the front-end device, demodulates and calculates the cross-consistency tensor, and generates dynamic semantic hashes through hashing operations. The adaptive physical micro-entropy extraction module collects the intrinsic physical feature vectors of the front-end sensor circuits, integrates interactive multi-model and unscented Kalman filtering to calculate the measurement residual vector, and decouples malicious features to generate adaptive hybrid physical micro-entropy random numbers. The device private key derivation module activates the physically unclonable function module to obtain the static physical response, and derives the asymmetric device private key in a trusted execution environment by combining the factory security salt value. The evidence storage and hardware signature module concatenates dynamic semantic hash, adaptive hybrid physical micro-entropy random number, system timestamp and terminal unique identifier into a payload to be signed, and calls the asymmetric device private key to execute digital signature to generate a live electronic evidence storage certificate. The cloud-based independent verification module performs independent verification on the received evidence storage credentials through hardware public key signature verification, timestamp determination, digital twin model verification, and anti-replay retrieval.

[0028] To better illustrate the technical solution of Embodiment 1 of the present invention, such as Figure 2As shown, this paper describes in detail an electronic evidence collection system for environmental supervision, combining a hash-based method for off-site environmental monitoring of pollution sources. Taking a monitoring point of a pollution discharge enterprise equipped with an edge computing terminal featuring an NPU (Neural Processing Unit) and a security chip (SE / TEE) as an example, a hardware-software physical fusion anti-counterfeiting architecture integrating non-intrusive measurement-side cryptographic watermarking and adaptive drift compensation micro-entropy is constructed to form a non-repudiable evidence collection closed loop while ensuring safe production. The method includes the following steps: S1. Use a true random number generator to generate a pseudo-random cryptographic sequence, convert the pseudo-random cryptographic sequence into a level fluctuation command and a high-frequency pulse width modulation command, and inject them into the analog reference voltage terminal of the front-end sensor and the fill light control terminal of the on-site monitoring camera, respectively. Simultaneously collect multidimensional datasets, merge and demodulate to extract micro-amplitude fluctuation sequences, calculate the cross-consistency tensor between demodulated sequences, and perform hash operation on the cross-consistency tensor after dimensionality reduction and splicing to obtain dynamic semantic hash. S2. Collect the intrinsic physical feature vector of the front-end sensor line, use the tracking architecture that combines interactive multi-model and unscented Kalman filtering to calculate the weighted fusion output of the physical state baseline and the measurement residual vector, and introduce a signal sparse decomposition mechanism to decouple malicious features and benign noise from the morphological dimension to generate adaptive hybrid physical micro-entropy random numbers. S3. During the initialization phase of the acquisition terminal, activate the built-in physical non-cloning function module and apply a challenge signal to obtain the static physical response output by the physical non-cloning function module. Combine the static physical response with the factory-set security salt value and send it to the key derivation function to recover the corresponding asymmetric device private key in the volatile register of the trusted execution environment. S4. The dynamic semantic hash, adaptive hybrid physical micro-entropy random number, system timestamp and terminal unique identifier are serialized and spliced ​​to form the payload to be signed. Within the security boundary of the trusted execution environment, the asymmetric device private key is called to execute the digital signature instruction on the payload to be signed, generate a live electronic evidence certificate and transmit it asynchronously to the cloud supervision center. S5 uses hardware public keys in the cloud to perform identity and integrity verification, extracts system timestamps to determine network freshness, verifies the synchronization of measurement-side watermarks through a cloud-based digital twin trust model, and performs existence-based anti-replay retrieval of adaptive hybrid physical micro-entropy random numbers in the historical buffer database to establish the judicial validity of electronic evidence.

[0029] In this invention, the dynamic semantic hash generated in step S1 is the core data payload for verifying the liveness attribute of the front-end monitoring data; the adaptive hybrid physical micro-entropy random number extracted in step S2 is a security trust benchmark for sensing and resisting physical layer hardware tampering; the asymmetric device private key derived in step S3 is a hardware trust root that ensures the uniqueness and unforgeability of the terminal identity; the encapsulation and signing operation performed in step S4 is a spatiotemporal snapshot of cryptographically binding multi-dimensional trust features and solidifying evidence; the independent verification performed in the cloud in step S5 is the final judgment link to confirm the authenticity of the data and complete the judicial solidification; the above steps work together to extend the anti-counterfeiting trust chain down to the source of the underlying physical sensing link, which can realize a closed-loop trust chain from the source of physical sensing to the cloud judicial solidification, thereby ensuring the authenticity and non-repudiation of off-site environmental monitoring data.

[0030] Furthermore, this method for electronic evidence collection in off-site environmental monitoring of pollution sources, which combines hash operations, is explained in detail, including the following: First, S1 constructs a dynamic semantic hash using a non-invasive measurement-side watermark, including the following sub-steps: S11. Utilize the True Random Number Generator (TRNG) built into the edge terminal to generate a set of extremely short pseudo-random cryptographic sequences. .

[0031] S12, will This is converted into level fluctuation commands and high-frequency pulse width modulation (PWM) commands, which are then injected into the analog reference voltage terminal of the analog-to-digital converter (ADC) of the front-end chemical / physical sensor. ) and the infrared fill light control terminal of the on-site monitoring camera.

[0032] S13, Edge terminal synchronous acquisition time window Multidimensional data sets within The algorithm executes a physical information neural network demodulation algorithm based on offline pre-training and online self-supervised mechanisms to extract hidden information from multimodal data sequences. Synchronize the small-amplitude fluctuation sequence and compute the cross-consistency tensor between the demodulated sequences. .

[0033] S14, will Dimensionality reduction and concatenation with the original bitstream are performed, followed by a one-way hash operation to obtain a dynamic semantic hash with liveness watermark proof. .

[0034] As an optional embodiment, considering the complex actual working conditions such as strong electromagnetic interference and wide temperature variation in industrial environments that can easily cause bias or periodic oscillations in the underlying physical entropy source of hardware, and the inability of out-of-band cryptographic watermarks to cause static drift in the sensor measurement baseline, the specific implementation steps of step S11 include: S111. Extract the original broadband thermal noise or ring oscillator phase jitter from the built-in true random number generator (TRNG) of the edge terminal as the initial entropy source. Introduce a dynamic debiasing algorithm and a conditioner of a cryptographically secure pseudo-random number generator (CSPRNG) to perform whitening processing on the initial entropy source to eliminate local entropy loss caused by environmental electromagnetic interference and temperature drift, and generate a high-entropy random seed with strictly uniform distribution characteristics.

[0035] S112. Generate a pseudo-random raw sequence of a set bit length (e.g., 16 to 64 bits) using a high-entropy random seed. Perform DC-balanced encoding on the pseudo-random raw sequence, such as Manchester encoding or equivalent zero-mean encoding, to force alignment of the energy integrals of the positive and negative levels in the sequence, generating an extremely short pseudo-random cryptographic sequence with a strictly zero mathematical mean. .

[0036] S113. Extract the high-precision hardware clock register parameters from the edge terminal's underlying layer, which are then converted into an extremely short pseudo-random cryptographic sequence. Bind the injected timestamp anchor. Push the sequence carrying the injected timestamp anchor into the secure isolation ring buffer of the direct memory access (DMA) controller so that the subsequent digital-to-analog converter front end or pulse width modulation (PWM) controller can independently execute the reading and modulation operations strictly according to the hardware clock tick without CPU interrupt intervention, ensuring rigid synchronization of micro-modulation instructions within an extremely short physical time window.

[0037] As an optional embodiment, considering the complex actual working conditions such as the susceptibility of the analog link of the front-end sensor in the industrial field to impedance mismatch and common-mode interference, the optical artifacts easily caused by the rolling shutter effect of the on-site camera, and the need to ensure that the out-of-band modulation signal does not compromise the compliance and validity of environmental monitoring data, the specific implementation steps of step S12 include: S121. To address the risk that signals may pollute the DC baseline of environmental data, extract extremely short pseudo-random cryptographic sequences. A direct sequence spread spectrum (DSSS) algorithm is introduced, using a megahertz-level high-frequency pseudo-random code (PN code) to perform spread spectrum modulation, uniformly broadening the signal energy and reducing it below the quantization noise floor of the ADC system. A high-resolution digital-to-analog converter (DAC) generates an extremely low-power spread spectrum baseband signal, which is then used to construct an independent differential injection bypass via a precision instrumentation amplifier with high common-mode rejection ratio (CMRR). At the physical dielectric layer, a high-frequency parasitic capacitance array at the Faraday (fF) level is employed to couple the differential spread spectrum signal in common-mode to the differential signal transmission line of the front-end sensor.

[0038] S122. To address image stripe artifacts and visual persistence effects caused by frame rate mismatch in on-site monitoring cameras, the vertical synchronization timing signal (VSYNC) and exposure trigger pulse of the on-site monitoring camera are extracted; a phase-locked loop (PLL) circuit is introduced to perform phase locking and frequency multiplication based on the vertical synchronization timing signal; based on an extremely short pseudo-random cryptographic sequence... The output of the phase-locked loop circuit generates a high-frequency pulse width modulation (PWM) command with a modulation frequency much higher than the reciprocal of the exposure time of a single frame of the surveillance camera and the critical fusion frequency of human vision (for example, raising it to a frequency band above 10kHz); the high-frequency pulse width modulation command is injected into the infrared fill light control terminal to achieve invisible embedding of optical stroboscopic watermark.

[0039] S123. For extreme operating conditions where sudden surges or control logic anomalies cause the modulation signal to exceed limits, an independent hard-wired clamping circuit with fail-safe characteristics is deployed at the injection front end. The system disconnects the injection path by default. A high-speed optocoupler solid-state relay with ultra-low parasitic capacitance is used as a physical isolation gate, which remains closed only when it receives a continuous high-frequency safety heartbeat pulse from the NPU hardware watchdog. A high-speed window comparator is deployed in the injection bypass to monitor the physical envelope amplitude of the differential spread spectrum signal in real time. Set strict physical safety red line thresholds. When the comparator detects When this occurs, a pure hardware-level asynchronous reset logic is triggered, directly discharging and grounding the gate drive capacitor of the solid-state relay.

[0040] As an optional embodiment, before the physical information neural network model is first used, a system hardware initialization calibration is performed to establish the initial parameter default values ​​of the physical mechanism equations, specifically including: During the cold start phase of the device, the edge computing terminal cuts off the actual physical input from the front-end sensors. A step test signal is then injected into the analog signal conditioning circuit via a digital-to-analog converter. Acquire the zero-state response curve at the sensor output. Based on the first-order dynamic response principle, the output curve is extracted to reach a steady-state value. The corresponding time interval is set as the initial default value of the system's equivalent time constant. Calculate the ratio of the steady-state output value to the amplitude of the step test signal to obtain the initial default value of the static gain of the injected link. Extract the factory calibration parameters of the front-end sensors to obtain the initial default value of the static gain of the sensor's service link. .

[0041] A single, extremely narrow pulse sequence is generated using a true random number generator and injected into the physical cable. A high-frequency analog-to-digital converter is activated to collect reflected and transmitted waves. The physical medium group delay for each mode of data is determined by calculating the time difference between the leading edges of the transmitted and received pulses. The group delay parameter is then assigned a value to the physical medium transmission phase compensation constant. initial default value .

[0042] Furthermore, in step S13, a physical information neural network demodulation algorithm based on offline pre-training and online self-supervised mechanisms is executed to extract the hidden micro-amplitude fluctuation sequence in the multimodal data sequence: S131. Before system deployment, construct an offline training dataset containing positive and negative samples, including: Positive Sample Acquisition and Labeling: Constructing a Hardware-in-the-Loop (HIL) simulation platform for target monitoring hardware. A pseudo-random code generator is used to generate orthogonal cryptographic sequence sets covering different Hamming weights and run lengths. This is injected into the physical hardware as a modulation command, and the output level sequence with physical layer distortion is acquired synchronously. .Will The corresponding injection sequence is labeled as a positive sample. Labeled as real.

[0043] Negative Sample Acquisition and Labeling: Low-frequency forged business data conforming to normal waste discharge logic is generated using a Generative Adversarial Network (GAN). This data is converted into an analog voltage injection acquisition channel without superimposing any high-frequency cryptographic sequences. The output sequence is acquired and labeled as a negative sample, while the true label is set to an all-zero sequence.

[0044] The aforementioned dataset was used to perform supervised pre-training on the feature extraction operators within the PINN model, enabling it to lock in the prior mapping of high-frequency micro-amplitude fluctuations in broadband background noise.

[0045] S132. Instantiate the pre-trained PINN model in the NPU built into the edge terminal. Construct the analog front-end (AFE) at time step S132. First-order physical dynamics transfer equations: ; In the formula, The system's equivalent time constant; The sensor outputs a response; For the static differential mode gain of the sensor business link; This represents the actual physical quantities of low-frequency environmentally friendly wastewater discharge. Gain of the isolation bias network for the high-frequency injection channel; This refers to the injected high-frequency spread spectrum microvolt sequence.

[0046] Furthermore, the initial calibration data obtained... , , Load the equations as the bias baseline. Extract the observations from the multimodal acquisition data within the current time window. Unlike offline training, the PINN model performs self-supervised solving during the online runtime phase. It utilizes a true random number generator to generate currently known sequences. and its spread spectrum results As a deterministic known term in the equation. The total loss function of the PINN model. Configured as follows: ; In the formula, The number of sampling points for the observation data; The number of points to allocate for the physical equations; Predict the output for the neural network; It is a dimensionless physical constraint penalty coefficient used to balance the weight of data-driven and physical constraints; Here is the Huber loss function.

[0047] The loss function is solved by gradient descent optimization, and the low-frequency physical components are decoupled using the network's internal automatic differentiation mechanism (Auto-Diff). The high-frequency feature components are then subjected to local despreading operations, outputting a demodulated feature sequence. This self-monitoring mechanism utilizes dynamically generated... Directly constraining the solution space helps eliminate the model's generalization blind spot for unseen random sequences.

[0048] S133. For the physical time delay difference between signal transmission via cable and optical space, a sliding window cross-correlation analysis with time offset is performed on the generated demodulated feature sequence to automatically search for and compensate for the maximum phase peak deviation between different modes. After phase alignment, the normalized cross-correlation (NCC) function is used to calculate the demodulated sequence of each mode and the original injected sequence. The correlation coefficients are used to aggregate and generate a cross-consistency tensor. : ; In the formula, For the first The physical medium transmission phase compensation constant corresponding to the mode; For the small signal demodulation operator executed by the NPU, This is a correlation analysis function. When a man-in-the-middle attack injects unmodulated, steady-state forged data directly through a generative adversarial network (GAN), it cannot reproduce the data carrying specific... Delay characteristics and conforms to Hidden fluctuations in the encoding will lead to The coefficients approach zero or diverge negatively, thus... Tensor collapse occurs and triggers the tamper circuit breaker mechanism.

[0049] Among them, the physical medium transmission phase compensation constant in step S133 The specific steps for obtaining and dynamically updating include: S1331. During the self-test cycle of system first power-on, hardware reset, or detection of sensor probe replacement, a training sequence (such as a Barker code sequence or pseudo-random code sequence) with sharp autocorrelation characteristics is generated using a true random number generator as a reference injection signal; the edge terminal acquires multimodal demodulation feature sequences. The discrete cross-correlation function between the signal and the reference injected signal is calculated more quickly using Fast Fourier Transform (FFT); the time lag index corresponding to the maximum peak value of the global absolute value of the cross-correlation function is extracted, converted, and locked to the first... Modal physical medium transmission phase compensation reference constant . It rigorously maps the inherent chemical reaction hysteresis, analog-to-digital conversion cycle and impedance network group delay of this specific sensor link at the physical level.

[0050] S1332. In the continuous operation of the system, to overcome the slow phase shift caused by environmental temperature drift, a digital delay-locked loop (DLL) algorithm model based on a sliding window is constructed in the built-in NPU; the transient micro-offset of the cross-correlation peak between the demodulated sequence and the reference sequence within the current short-time window is continuously calculated. The exponential moving average (EMA) function is introduced to perform low-pass filtering on transient micro-offsets to remove high-frequency random jitter noise. The dynamic physical medium transmission phase compensation constant is then iterated and output in real time. : .

[0051] S1333. Extract the theoretical phase drift limit parameters of this specific type of sensor and cable transmission medium under full temperature range operating conditions, and use this to construct a phase offset tolerance defense line. During the aforementioned dynamic delay phase-locked tracking process, boundary condition determination is continuously performed: In extreme cases where a man-in-the-middle attacker maliciously connects a digital delay buffer in the sensor physical link to gain time to compute forged data for a generative adversarial network (GAN), this will trigger... This triggers a low-level hardware interrupt, forcibly clearing the cross-consistency tensor. It also reports physical link delay anomalies to the cloud to ensure the spatiotemporal locking of the source of the evidence chain.

[0052] As an optional embodiment, for the working condition where heterogeneous chips in industrial field edge computing nodes have a hash avalanche effect caused by precision rounding errors when processing continuous floating-point tensors, and for the actual complex working condition where the length distribution of multi-source heterogeneous original data is extremely uneven and vulnerable to length expansion attacks, the specific implementation steps of step S14 include: S141. For underlying heterogeneous neural network processors or central processing units, in calculating the cross-consistency tensor In cases where differences in floating-point arithmetic precision can easily lead to slight biases in the underlying bitstream, a Locality Sensitive Hash (LSH) dimensionality reduction algorithm combining adaptive dead-time quantization and symbolic random projection is introduced. The specific steps include: Extracting the cross-consistency tensor Flatten it into a one-dimensional correlation feature vector Among them, feature dimension , The number of channels for multimodal data streams. This represents the delay compensation window constant in sliding window cross-correlation analysis. Addressing the physical phenomenon that broadband noise in industrial environments can easily induce spurious peaks in cross-correlation analysis, this is applied to the one-dimensional correlation feature vector. Perform adaptive dead-time quantization: ; In the formula, The first feature vector after filtering One component; The first of the original one-dimensional correlation feature vectors Each component represents the normalized cross-correlation coefficient; This is a dead-zone threshold constant used to shield against minor correlation fluctuations caused by stray electromagnetic interference from the environment. An offline static pseudo-random projection matrix is ​​constructed. ,in The bit length of the target dimension-reduction hash is constant. The unique hardware identifier of the device within the trusted execution environment is extracted as the initial seed for the pseudo-random number generator, generating a pseudo-random projection matrix. Each element is determined to strictly follow a standard normal distribution. The feature vector after dead zone quantization Mapping to a low-dimensional Hamming space, a symbolic random projection operation is performed to generate discretized features: ; In the formula, For projection feature sequence The One component; For pseudo-random projection matrix The Middle Line 1 Column elements; This is the symbol extraction function.

[0053] In extreme tampering scenarios where the front-end sensor is completely physically severed or replaced with a steady-state DC signal generator, the demodulated sequence and the original cryptographic sequence become completely mismatched, resulting in a loss of quality after dead-zone quantization. It degenerates into a vector of all zeros. To avoid abnormal algorithm output or stagnation, a mandatory boundary condition is set for the symbol extraction function: when the input value is... Constant output ,Right now This setting ensures that all-zero feature vectors are fixedly mapped to a specific configuration of incorrectly projected sequences, thereby triggering a hash avalanche effect in subsequent one-way hash operations and ensuring a rigid block on the synchronization verification of cloud-based digital twins. The projected feature sequence... The value in Convert to logical bits numerical values Convert to logical bits The polymerization generates a fixed length of The binary eigenvector of a bit Extract the binary feature vector. As a constant replacement for the cross-consistency tensor of the liveness watermark feature after dimensionality reduction This eliminates hash preload jitter caused by differences in the precision of underlying floating-point calculations.

[0054] S142. For situations where the data volume of high-frequency video frames is much larger than that of low-frequency sensor values, causing direct splicing to easily lead to data overflow or blurred delimitation, the multi-source raw bitstream is encoded according to a preset type-length-value encoding standard. Perform serialization encapsulation; inject strictly defined modal domain separators and length placeholders into the header of the serialized data block to prevent the byte boundaries of different modal data from being maliciously tampered with; fill the serialized encapsulated multimodal original bit stream with fixed-length bytes and concatenate it with physical blocks with the live watermark feature constant generated in step S141 to generate a structured hash preload.

[0055] S143. To address the replay scenario where static, isolated hash blocks are easily intercepted by attackers and sliced ​​for replacement, extract the preceding time sliding window. Internally generated historical hash node record aggregation item As a forward timing constraint constant, a one-way hash algorithm with high collision resistance (such as the national cryptographic standard SM3) is adopted. The structured hash preload and the forward timing constraint constant are used as the input sequence of the algorithm to perform hash operation; the output is a fixed-length dynamic semantic hash. .

[0056] Furthermore, S2 extracts adaptive micro-entropy that is robust to environmental conditions and defends against physical layer tampering, including the following sub-steps: S21. The high-frequency analog-to-digital conversion channel of the edge computing terminal continuously acquires the intrinsic physical feature vectors of the front-end sensor lines. It includes 1 / f flicker noise spectral density and broadband micro-reflection impedance.

[0057] S22. To address the failure condition caused by parameter convergence hysteresis of a single Kalman filter when faced with sudden environmental changes (such as a drastic change in temperature and humidity due to the sudden opening of a factory gate), a tracking architecture that integrates interactive multiple model (IMM) and unscented Kalman filter (UKF) is adopted to achieve smooth switching between different physical evolution rates. A system dynamic model set covering various physical environment evolution characteristics is constructed. .in, To characterize the extremely low-speed, steady-state drift model of natural aging, To characterize the conventional linear drift model of diurnal temperature range, This is a nonlinear accelerated evolution model characterizing sudden changes in the local environment. The system is defined at time [time value missing]. The joint physical state vector is Its internal elements encompass intrinsic impedance and noise power spectral density for a specific frequency band. Corresponding model The state transition equation is expressed as: In the formula, For the corresponding model The nonlinear physical evolution function; This is the process noise vector.

[0058] Observation equation: ,in The observed values ​​input in step S21, For the observation matrix, To observe noise.

[0059] Introducing the Markov transition probability matrix ,in The system represents the model Transfer to model The prior probability. Within each sampling period, based on the residual distribution between the current observation and the predicted values ​​of each model, the model probability weights at the current time are dynamically updated. And calculate the weighted fusion output of the global physical state baseline. : ; Calculation time Measurement residual vector This serves as a measure of how far a physical link deviates from the normal baseline.

[0060] Under extreme conditions, such as encountering strong electromagnetic pulse interference exceeding all preset physical evolution limits, the prediction residuals of all models tend to diverge, and the model probability weights... The data will degenerate to a uniform distribution. At this point, the system triggers a covariance reset interrupt, suspending baseline updates and locking them to the previous steady-state value to prevent anomalous data from contaminating long-term baseline parameters.

[0061] S23. For industrial operations such as starting and stopping large motors and relay activation, where electromagnetic interference steps can easily be confused with physical tampering, a signal sparse decomposition mechanism based on an overcomplete dictionary is introduced to decouple malicious features from benign noise from a morphological perspective. An offline physical prior waveform dictionary matrix is ​​constructed. Among them, benign sub-dictionary Includes deterministic transient pulse basis functions collected from the field during the start-up and shutdown of various legitimate electromechanical equipment; malicious sub-dictionary It includes the basis functions of the step response caused by physical tampering behaviors such as simulating sensor disconnection and bypass capacitor connection.

[0062] Extract the measurement residual sequence output in step S22 Transform it into a solution Norm-regularized sparse optimization problem: ; In the formula, Let be the sparse coefficient vector to be solved; To control sparsity, a regularization penalty parameter is used. The optimal sparse coefficient vector is solved using Orthogonal Matching Pursuit (OMP) or Fast Iterative Shrinking Threshold Algorithm (FISTA). .Will Split into coefficient subsets of the corresponding benign sub-dictionary coefficient subset of the corresponding malicious sub-dictionary Calculate the projected energy of the signal in the maliciously modified subspace. : ; Extracting the reconstruction error vector Calculate the residual energy characterizing the unknown perturbation. Extracting the measurement residual sequence Mean variance of floor noise within the unaltered historical steady-state time window Set constant false alarm rate multiplier Calculate the adaptive dynamic decision threshold. : Construct a tampering decision weight function based on energy distribution. To quantify classification confidence and suppress false alarms caused by unknown physical interference outside the dictionary: ; In the formula, This is the residual energy penalty coefficient, used to adjust the suppression strength of unmodeled features on the judgment result; To prevent the denominator from approaching zero, the output value of this weighting function is limited to a minimal constant. Interval. Calculate the final comprehensive tampering confidence score. : ; In the formula, the confidence score For interval Dimensionless parameters within.

[0063] Execute threshold determination logic: Set the lower limit of system confidence determination. When judged At that time, extract High-frequency effective feature inputs are used in the HKDF algorithm to generate adaptive hybrid physical micro-entropy random numbers. If determined and The system determines that an undefined physical anomaly has occurred, blocks the generation of micro-entropy random numbers, and sends an asynchronous warning to the cloud.

[0064] As an optional embodiment, considering the potentially strong 50Hz / 60Hz power frequency electromagnetic interference and harmonic noise in industrial environments, which can easily drown out weak 1 / f flicker noise, and the fact that directly superimposing wideband impedance detection pulses on physical cables can easily cause crosstalk to the low-frequency (such as 4-20mA or 0-5V) standard analog business signals output by the front-end sensors, the specific implementation steps of step S21 include: S211. For the collinear crosstalk conditions of impedance detection and service signal sampling, and the equal impedance passive dummy load replacement attack conditions, a dual-frequency division multiplexer or a high-frequency isolation bias network is connected to the analog front end of the edge terminal; the low-frequency service signal of the front-end sensor and the high-frequency physical detection channel are frequency-domain isolated in the physical circuit; a high-frequency composite physical detection signal with a center frequency much higher than the sensor response bandwidth is generated by a digital-to-analog converter. The high-frequency composite physical detection signal includes a wideband linear frequency modulated pulse and a dual-frequency equal-amplitude sine wave, which is injected into the sensor physical cable through the high-frequency physical detection channel.

[0065] S212. Deploy a heterogeneous dual-channel analog-to-digital converter architecture at the analog signal receiver; capture the transient micro-reflection wave and nonlinear response spectrum generated by the injected high-frequency composite physical detection signal through the first channel; suppress the power frequency fundamental wave and its higher harmonic interference through the pre-amplified hardware comb notch filter of the second channel, and collect the sensor intrinsic low-frequency electrical noise sequence after power frequency interference is removed.

[0066] S213. The raw bitstream acquired from the dual channels is transmitted to the built-in neural network processor via a direct memory access channel; the power spectral density is calculated for the low-frequency electrical noise sequence of the second channel, and the scintillation noise slope constant is extracted; for the transient micro-reflection wave and nonlinear response spectrum captured by the first channel, a discrete Fourier transform is performed to calculate the amplitude and phase parameters of the frequency domain complex reflection coefficient, and the third-order intermodulation distortion power ratio based on dual-frequency equal-amplitude sine wave excitation is extracted; the scintillation noise slope constant, the amplitude and phase parameters of the complex reflection coefficient, and the third-order intermodulation distortion power ratio are tensor-concatenated to determine the intrinsic physical feature vector. .

[0067] The formula for calculating the third-order intermodulation distortion power ratio in step S213 is as follows: ; In the formula, The third-order intermodulation distortion power ratio; The fundamental frequency of the injected dual-frequency equal-amplitude sine wave; In response to the signal at the fundamental frequency and Power value at; These represent the response signals at the third-order intermodulation frequency points. and The power value at that location.

[0068] Furthermore, S3 constructs a static hardware root of trust based on PUF, including the following sub-steps: S31. During the power-on initialization and security boot phase of the acquisition terminal, activate the Physically Unclonable Function (PUF) IP core built into the chip and apply a preset challenge signal.

[0069] The S32 and PUF modules output unique static physics responses. The key is combined with the factory-set security salt value and fed into the key derivation function to recover the corresponding asymmetric device private key from the volatile register of the Trusted Execution Environment (TEE). The private key is immediately lost when the device experiences an abnormal power outage or is physically tampered with.

[0070] As an optional embodiment, for industrial environments where extreme temperature differences and power fluctuations can easily cause the threshold voltage of the underlying semiconductor transistors with physically unclonable functions (PUFs) to drift, thereby increasing the physical response bit error rate, and for complex actual operating conditions where equipment is susceptible to side-channel attacks such as differential power analysis (DPA) during cold start-up to steal hardware physical attributes, the specific implementation steps of step S31 include: S311. During the edge terminal cold start or system hard reset phase, the on-chip read-only memory verifies the digital signature of the initial boot program to establish the basic hardware trust root. Before activating the PUF core, the hardware monitoring unit data of the underlying motherboard is extracted, and physical boundary interval verification is performed on the core power supply voltage, transient current and bare die temperature of the current chip. When extreme environmental characteristics that may induce physical fault injection attacks, such as external voltage spikes or temperature over-limits, are detected, the PUF activation sequence is forcibly suspended until the physical environment converges to the preset safe operation envelope.

[0071] S312. To address the vulnerability of static challenge signals to electromagnetic interference during internal bus transmission, the conventional path of transmitting challenge signals via hard-coded or plaintext methods is blocked. A one-time physical mask is generated using an independent micro-generated true random number generator within the security chip. A bitwise XOR operation is performed between the preset basic challenge seed and the one-time physical mask to generate a dynamic scrambling challenge instruction. The dynamic scrambling challenge instruction is then sent to the PUF physical array via an independent internal security isolation bus, eliminating the statistical correlation between power consumption and challenge data at the physical execution level and resisting side-channel analysis attacks.

[0072] S313. Based on the dynamic scrambling challenge instruction, apply physical excitation to the initial power-on flip state of the static random access memory (SRAM) array inside the PUF module or the delay differential characteristics of the ring oscillator array to extract the original physical response sequence containing the initial error; read the auxiliary error correction data securely burned during device factory registration from the off-chip secure non-volatile memory; within the Trusted Execution Environment (TEE), use the fuzzy extractor algorithm core, combined with the auxiliary error correction data, to perform error correction coding (such as BCH code or Reed-Muller code) on the original physical response sequence to eliminate physical response errors caused by wide industrial temperature differences and electromagnetic interference, so as to ensure that the output has mathematical uniqueness and stability of static physical response.

[0073] As an optional embodiment, for situations where the factory-installed security salt value stored in ordinary non-volatile memory is vulnerable to firmware extraction attacks, where the key derivation process easily leaves sensitive snapshots in general-purpose memory, and where volatile memory is susceptible to cold-start attacks due to data residue effects after a conventional physical power failure, the specific implementation steps of step S32 include: S321. To address the sniffing risk of reading key materials from the external bus, extract the factory-installed security salt value physically embedded in the one-time programmable (OTP) memory or electronic fuse (eFuse) physical array within the security chip; within the hardware isolation boundary of the Trusted Execution Environment (TEE), apply the mathematically unique static physical response obtained in step S31. By sequentially splicing the product with the factory-safe salt value, a derived base seed material that is immune to external bus detection is constructed.

[0074] S322. To address the issue of register state leakage due to the ease with which general-purpose processors can compute keys, an independent cryptographic coprocessor embedded within the trusted execution environment is invoked. Using a hash-based message authentication code key derivation function (HKDF), cryptographic expansion and compression operations are performed on the derived seed material to dynamically generate a standard-length asymmetric device private key. The device's private key is directly loaded into a dedicated volatile security register, which has addressing permissions that are only open to the cryptographic coprocessor, through a dedicated hardware encryption bus, thereby achieving operating system-level memory access isolation.

[0075] S323. To address extreme conditions where passive power-off cannot prevent data retention and physical probe intrusion, a miniature backup storage capacitor and a hardware-level active clear logic gate network are deployed on the outer periphery of the dedicated volatile security register. The impedance integrity of the physical anti-tamper mesh covering the chip surface and the voltage status of the low-voltage detector are monitored in real time. When a physical intrusion circuit breaker signal is triggered or the supply voltage drops to a critical threshold, the residual charge of the miniature backup storage capacitor is used to force the active clear logic gate network to overwrite a full zero level or high-entropy physical noise to the dedicated volatile security register within a nanosecond clock cycle, thereby achieving forced erasure of the device's private key at the physical circuit level.

[0076] In this embodiment, the device private key derivation based on HKDF specifically includes: S322-a. Within a Trusted Execution Environment (TEE), the asymmetric device private key The derivation process follows a two-stage logic: extraction followed by expansion. Its input parameter specifications are as follows: Initial Key Material (IKM): Consists of static physical response With factory safety salt value Physically assembled, that is .

[0077] Salt value selection criteria: The length is set to 256 bits, consistent with the output length of the underlying hash function (such as SM3 or SHA-256). This choice is based on ensuring effective elimination during the extraction phase. The statistical bias that may exist in the sequence can increase the entropy strength of the pseudo-random key (PRK).

[0078] Application-specific information (Info): Defined as containing the device's unique identifier. A fixed-length string containing the current algorithm version number and its purpose identifier (such as "ENV_MONITOR_ECC_KEY").

[0079] Invoke the cryptographic coprocessor to perform the operation according to the following formula: During the extraction phase, a uniformly distributed pseudo-random key is generated using Hash Message Authentication Code (HMAC). ,Right now ; In the expansion phase, output key material of a specified length is generated. : ; ; ; in, Because the derived target is a 256-bit SM2 private key or a 256-bit ECDSA private key ( Furthermore, the hash function used (SM3 / SHA-256) also outputs 256 bits, hence the number of iteration rounds. The value is 1.

[0080] S322-b: To address the non-executability issues caused by interface differences between different hardware platforms (such as ARM TrustZone and Intel SGX), the system introduces a security platform abstraction layer to implement hardware-level calls through a unified function interface. Specifically, this includes: Define a standard API in trusted applications: TEE_Result DeriveHardwareSecret(constuint8_t puf_resp,size_t puf_len,uint8_t out_priv_key).

[0081] In ARM TrustZone (such as OP-TEE): using the TEE_DeriveKey API, ... As the parent key, it is used in conjunction with the TEE_ALG_HMAC_SM3_DERIVE_KEY algorithm identifier, and encapsulated in encrypted and isolated memory. The structure directly outputs the private key handle in secure memory, ensuring that the private key does not leave the secure domain.

[0082] In Intel SGX: The `sgx_get_key` instruction is called internally within the Enclave, combined with the `sgx_key_request_t` structure. Mapped to key_request.key_policy. Since SGX's EGETKEY instruction natively supports derivation based on physical hardware characteristics (such as the sealing key), at this time... The key_id field is injected as an additional entropy source, and the extended calculation is completed through internal hardware circuitry.

[0083] S322-c. To address the vulnerability of the derivation process to side-channel attacks, the following constraints are added: Memory cleanup: In After converting to asymmetric private key format and loading it into the volatile secure register, SecureMemset is called to force erasure. , and The temporary stack space it resides in.

[0084] Constant-time execution: The internal iterative process of HMAC operations must employ a constant-time algorithm implemented using hardware instruction sets (such as ARMv8 CryptoExtension or Intel AES-NI / SHA Extensions) to prevent reverse derivation through power consumption analysis. .

[0085] Furthermore, S4 performs signature solidification of the liveness evidence data packet, including the following sub-steps: S41. Dynamic semantic hashing Adaptive physical micro-entropy random numbers System timestamp and the terminal's unique identifier Serialization and concatenation form the payload to be signed. .

[0086] S42. Within the TEE security boundary, access the device private key. right Execute the digital signature instruction to generate a live electronic evidence certificate. And asynchronously transmit to the cloud monitoring center.

[0087] As an optional embodiment, considering the complex real-world conditions such as the vulnerability of local clocks at industrial edge nodes to malicious tampering to forge evidence collection times, the ease with which simple concatenation of heterogeneous data can lead to parsing ambiguity and hash length expansion attacks, and the vulnerability of data to malicious processes intercepting and replacing it through memory during the assembly stage, the specific implementation steps of step S41 include: S411. Addressing the issue that industrial edge nodes are often located in weak network environments, where their local system clocks are highly susceptible to external interference, network time protocol (NTP) hijacking, and malicious clock rollback, the system abandons the software time of general-purpose operating systems and instead uses a trusted real-time clock powered by an independent battery within the Trusted Execution Environment (TEE). It extracts the current value of the trusted real-time clock and the accumulated count value of the underlying hardware monotonic counter, and performs cross-verification with a signed reference clock periodically distributed by a cloud-based secure time server to generate a high-precision trusted system timestamp with physical tamper-proof and anti-rollback attributes. .

[0088] S412. Given the varying lengths of data payloads generated by different environmental monitoring points, simple linear concatenation can easily lead to malicious byte boundary shifting by attackers, resulting in cryptographic ambiguity or length expansion attacks. Therefore, a strict type-length-value encoding standard is adopted for dynamic semantic hashing. Adaptive physical micro-entropy random numbers Trusted system timestamp and the unique identifier embedded in the hardware Each block is independently serialized and encoded; at the physical connection boundary of each encoded block, a preset dynamic field separator generated by the security chip is forcibly injected to block any cross-field byte tampering, shifting, and illegal splicing.

[0089] S413. Addressing the extreme scenario where the payload to be signed is assembled in the shared memory of a general-purpose operating system, making it vulnerable to privilege escalation and theft by malicious software or interception by direct memory access (DMA) attacks, potentially leading to data replacement, the serialization encoding and physical assembly of all fields are strictly restricted to be completed independently within encrypted, isolated memory within a Trusted Execution Environment (TEE). Upon completion of payload assembly, the Memory Management Unit (MMU) immediately performs hardware-level page table locking and one-way freeze of read / write permissions on this isolated memory region, outputting an immutable and completely physically isolated payload for signature verification. This ensures the purity of the entire data link and memory view before triggering the device's private key to perform a signing operation.

[0090] As an optional embodiment, considering the strict domestic compliance requirements for the underlying cryptographic algorithms of the government ecological environment supervision platform, the vulnerability of the underlying signature operation stage to execution time side-channel attacks, and the weak network conditions often faced by remote pollution source monitoring points, such as wireless network signal attenuation and frequent disconnections, the specific implementation steps of step S42 include: S421. In response to compliance requirements for government data encryption and the leakage of chip computing side channels, within the Trusted Execution Environment (TEE), the national cryptographic SM3 hash algorithm conforming to the standards of the State Cryptography Administration is forcibly invoked for the payload to be signed. Perform message digest calculation; then invoke the asymmetric device private key residing in the secure register. It executes the national standard SM2 elliptic curve digital signature algorithm. During the underlying scalar multiplication and modular inverse operations, it enforces constant-time large number arithmetic logic to ensure that the CPU clock cycle consumption and microarchitecture power consumption radiation of the signature instruction remain constant regardless of changes in the Hamming weight of the input payload. This physically blocks the side-channel attack path by attackers who can reverse-engineer the device's private key by measuring computation time, thus outputting a national standard compliant digital signature. .

[0091] S422. For situations where scattered data in complex systems is prone to structural damage due to memory overflow or pointer exceptions, after signing, the payload to be signed should still be stored within the secure physical boundaries of the TEE. Digital signature The firmware version hash of the current trusted execution environment and the state machine vector of the secure boot process are merged across layers. A pre-defined binary abstract syntax tree specification is used to serialize and encapsulate this data into an independent electronic evidence container with an tamper-proof shell. This evidence container enables a strong binding between business data, physical micro-entropy, and the underlying hardware and software execution environment context, ensuring that once the electronic certificate is generated, not only is its content immutable, but the security environment snapshot at the time of its generation is also fixed.

[0092] S423. In response to the complex working conditions of extremely unstable network connections in industrial sites, the synchronous transmission mode that easily leads to main thread blocking is abandoned; the electronic evidence compartment is encrypted twice using the national cryptographic SM4 symmetric encryption algorithm and session negotiation key, and then pushed into the persistent asynchronous transmission queue based on the local encrypted file system (EFS); the transmission daemon process detects the network connectivity health of the cloud monitoring center in real time, and automatically suspends the transmission and locks the local persistent pointer when a network interruption or handshake timeout is detected; after the network heartbeat is restored, the encrypted electronic evidence compartments in the queue are pushed to the cloud monitoring center in batches using a breakpoint resume transmission protocol based on block verification, ensuring "zero loss" of the evidence chain in extremely poor network environments.

[0093] Furthermore, S5 performs non-repudiation independent verification in the cloud, such as... Figure 3 As shown, it includes the following sub-steps: S51, cloud-based use of hardware public keys Perform the first layer of identity and integrity verification to confirm the data source and that it has not been tampered with.

[0094] S52, Extract Timestamp Perform a second layer of network freshness assessment.

[0095] S53, Extracting the load and The synchronization of the measurement-side watermark is verified through a cloud-based device digital twin trust model, and the data is then processed in the historical buffer database. Perform an existence-based replay protection search. Because... Invalid alarms caused by gradual environmental changes have been filtered out at the edge. The cloud only needs to verify the spatiotemporal uniqueness and non-abruptness to establish the final electronic evidence solidification and judicial validity.

[0096] As an optional embodiment, the specific implementation steps of step S53 include: S531, addressing the scenario where plaintext transmission of out-of-band watermark sequences is easily intercepted by physical link eavesdroppers, leading to watermark forgery, by blocking the direct transmission of any original watermark sequence; extracting the trusted system timestamp from the payload using a cloud-based digital twin trust model. Combined with the pre-shared master key, securely negotiated during device network access and residing only within the TEE, a hash-based pseudo-random number generation algorithm (Hash-DRBG) is used to synchronously deduce and reconstruct the edge device's... Expected cryptographic sequence injected at time The cloud utilizes twin computing resources to perform high-precision offline demodulation on the uploaded raw multimodal data bitstream, consistent with the edge endpoint, and calculates the twin consistency tensor. ;Will The hash operation is performed by concatenating the original bitstream with the reconstructed result and verifying it against the dynamic semantic hash in the payload. Whether they are strictly consistent is used to determine the synchronization of the watermark on the measurement side and whether the data has been tampered with.

[0097] S532. For distributed cloud monitoring platforms, where replay attacks may escape within a short time window due to underlying database read / write synchronization latency; a high-speed distributed anti-replay array based on a cuckoo filter is introduced in memory at the cloud level; physical micro-entropy random numbers in the payload are... With the device's unique identifier Serialization is combined into a joint anti-replay feature key; in the system timestamp Within a limited effective lifespan (TTL, such as a network tolerance window of 5 seconds), concurrent existence verification with underlying atomic operation semantics is performed; once the same feature key value is hit in the cache array, it is determined that a replay attack has occurred and the evidence collection circuit breaker mechanism is triggered instantly to block the entry of malicious dirty data into the database.

[0098] S533, targeting extreme cases where advanced attackers intercept the operational logic of the edge-end Kalman filter and use algorithms to brute-force generate fake high-frequency residual noise in an attempt to deceive the system; extracting and decrypting the payload in the cloud... Strongly bound intrinsic electrical noise statistical characteristic parameters, such as the amplitude envelope of the broadband complex reflection coefficient and the variance of the noise spectral density, are used. These statistical characteristic parameters are input into the cloud-based digital twin baseline model of the specific sensor node. The Gaussian mixture model (GMM) is used to calculate the relative information entropy distance (e.g., Kullback-Leibler divergence) between the currently uploaded micro-entropy distribution characteristics and the physical aging trajectory of the hardware throughout its entire life cycle (e.g., the impedance gradient curve caused by the aging of the cable insulation layer). It is then determined whether the information entropy distance is strictly constrained within a safe confidence interval representing non-jumping behavior. If and only if the identity signature, twin watermark, atomic anti-replay, and physical non-jumping behavior quadruple verification all pass, the system automatically adds a national cryptographic timestamp and a judicial electronic signature, completing the blockchain solidification of electronic evidence.

[0099] In this embodiment, the cloud-based digital twin baseline model is formed by coupling a symmetric synchronous cryptographic state machine with a physical evolutionary stochastic process model, specifically including: The cloud for each Maintain a dynamic parameter matrix Specifically, it includes: cryptographic parameter sequence : Includes pre-shared master key Current clock cycle counter and based on Clock offset correction amount .

[0100] Physical state characteristic benchmark : Initial intrinsic impedance of the sensor link Ambient temperature sensitivity coefficient Humidity correlation coefficient and the power-law exponential eigenvalues ​​of 1 / f noise. .

[0101] Lifetime decay factor Includes the drift slope of the electrolytic capacitor's equivalent series resistance (ESR) calculated based on equipment uptime. And the probability distribution parameters of cable insulation degradation.

[0102] To enable verification without transmitting the original watermark sequence, the cloud model performs the following calculations: Seed Reconstruction: Extracting Payload from the Cloud Combined with the device's master key stored locally Using a random number generation function based on the national cryptographic algorithm Calculation time Deterministic seeds: .

[0103] Watermark deduction: using As the initial state, the same watermark sequence injected at the edge is reconstructed using a linear feedback shift register (LFSR) or a hash derivation function. .

[0104] Cross-correlation verification: cloud utilization Digital down-conversion and sliding correlation operations are performed on the raw sewage discharge data bitstream. The resulting correlation peak value... If the watermark synchronization on the measurement side is successful, then the measurement side is considered to have been successfully synchronized. This is the preset confidence threshold.

[0105] To differentiate between gradual environmental changes and advanced tampering, the model employs a multivariate Gaussian distribution model with parameter drift. Baseline prediction: The cloud-based model calculates the physical feature vector for the current moment based on historically collected data trajectories and current meteorological data (temperature, humidity). Theoretical expected value : ; Distribution Consistency Verification: Extracting Adaptive Micro-Entropy from the Load This micro-entropy is considered a mapping of the observation residuals. The current observation residuals are calculated in the cloud. Gaussian distribution The distance from Maharanobis : ; in, This is the covariance matrix of the intrinsic noise of the device.

[0106] Decision logic: If If the physical characteristics change, it is determined that the change is due to natural environmental drift, and the evidence is valid; if If a non-physical evolutionary leap has occurred in the physical link, such as an attacker accessing a signal generator whose impedance characteristics no longer match the device's long-term physical evolution trajectory, the evidence is immediately deemed invalid. This is the significance level threshold under the chi-square distribution.

[0107] Closed-loop update mechanism: Whenever a piece of legitimate electronic evidence is successfully solidified, the cloud model uses the update gain of the Kalman filter to adjust the... In Dynamic correction is performed using equal coefficients. The specific implementation steps include: In the initial stage of system deployment (i.e., the cold start phase), in order to enable the cloud-based digital twin model to have initial predictive capabilities, the temperature sensitivity coefficient... Humidity correlation coefficient and aging drift slope Pre-calibration in the laboratory was used to obtain the following: Before the equipment left the factory, samples of sensors from the same batch were placed in a high and low temperature humidity alternating test chamber. Temperature gradients (e.g., -20°C to 60°C, in 10°C steps) and humidity gradients (e.g., 10%RH to 90%RH, in 20%RH steps) were set, and the changes in the sensor's intrinsic impedance and noise power spectral density were recorded. The partial derivatives of the physical characteristics with respect to temperature and humidity were calculated using a linear regression algorithm to obtain the initial coefficients of this sensor model. and The cloud database stores historical operating parameters for sensors from different industries and models. Based on the device's unique identifier... Search for the sensor type. If mature operating trajectory data for that model is already available in the database, extract the average aging attenuation coefficient after more than 5000 hours of operation, and use this as the initial aging drift slope for the current device. .

[0108] To ensure that the digital twin model can evolve in sync with the natural aging of the physical entity and environmental changes, the system uses a parametric state-space model in the cloud to perform real-time corrections: defining the time... The parameter state vector is The expected value prediction model for physical eigenvectors is expressed as: ;in, A vector representing environmental and temporal changes; These are the measured values ​​of the physical characteristics at the previous moment.

[0109] In each evidence verification cycle, the measured residuals are extracted in the cloud. The parameter state vector is corrected using a recursive matrix. The gain matrix is ​​then calculated. : ; in, The covariance matrix for parameter estimation (initial values) (Set as a diagonal matrix); To observe the noise variance scalar.

Claims

1. A non-site environmental monitoring electronic evidence collection system for pollution sources combining hash operations, characterized in that, include: The multimodal dynamic semantic hash generation module is used to generate pseudo-random cryptographic sequences using a true random number generator, convert the pseudo-random cryptographic sequences into modulation instructions, inject them into the front-end device, demodulate and calculate the cross-consistency tensor, and perform dimensionality reduction and quantization through a symbolic random projection algorithm combined with adaptive dead-zone quantization to generate dynamic semantic hashes. The adaptive physical micro-entropy extraction module is used to collect the intrinsic physical feature vectors of the front-end sensor circuit, fuse interactive multi-model and unscented Kalman filtering to calculate the measurement residual vector, and decouple malicious features to generate adaptive hybrid physical micro-entropy random numbers. The device private key derivation module is used to obtain the static physical response output by the physically unclonable function module. It calls the two-stage key derivation function through the security platform abstraction layer within the trusted execution environment to derive the asymmetric device private key. The evidence storage encapsulation and hardware signature module is used to concatenate the dynamic semantic hash, the adaptive hybrid physical micro-entropy random number, the system timestamp and the terminal unique identifier into a payload to be signed, and call the asymmetric device private key to perform digital signature to generate a live electronic evidence storage certificate. The cloud-based independent verification module utilizes a pre-calibrated device digital twin trust model to perform online updates via a parameter correction matrix, thereby independently verifying the live electronic evidence certificate to generate a legally valid result.

2. The electronic evidence collection system for off-site environmental monitoring of pollution sources combined with hash operation as described in claim 1, characterized in that, The steps of the multimodal dynamic semantic hash generation module to generate dynamic semantic hashes include: The cross-consistency tensor is flattened into a one-dimensional correlation feature vector, and adaptive dead-zone quantization is performed on the one-dimensional correlation feature vector using a preset dead-zone threshold to shield noise fluctuations. Using a static pseudo-random projection matrix generated based on hardware identifier codes, a symbolic random projection operation is performed on the quantized feature vector to generate a binary feature vector. The binary feature vector is concatenated with the original bit stream, and a one-way hash operation is performed to output the dynamic semantic hash.

3. The electronic evidence collection system for off-site environmental monitoring of pollution sources combined with hash operation as described in claim 1, characterized in that, The demodulation calculation of the cross-consistency tensor includes: A physical information neural network model is constructed, which is pre-trained offline using positive samples generated by hardware-in-the-loop simulation and negative samples generated by a generative adversarial network. During the system initialization phase, the initial default values ​​of the system's equivalent time constant and the link's static gain are calibrated using a step test signal. During the forward inference phase, the physical dynamics transfer equation is constrained by the initial default values, and a self-supervised solution is performed in conjunction with an online-generated pseudo-random cryptographic sequence to decouple high-frequency feature components and calculate the cross-consistency tensor.

4. The electronic evidence collection system for off-site environmental monitoring of pollution sources combined with hash operation as described in claim 1, characterized in that, The adaptive physical micro-entropy extraction module generates adaptive hybrid physical micro-entropy random numbers, including: The measurement residual vector is transformed into a sparse optimization problem and solved to obtain the coefficient subset of the corresponding malicious sub-dictionary and calculate the projected energy, as well as the residual energy of the corresponding unmodeled signal. A decision weight function is constructed based on the projected energy, the residual energy, and the residual energy penalty coefficient. Calculate the overall tampering confidence score. When the overall tampering confidence score exceeds the preset lower limit, determine the adaptive hybrid physical micro-entropy random number.

5. The electronic evidence collection system for off-site environmental monitoring of pollution sources combined with hash operation as described in claim 1, characterized in that, The steps for acquiring the intrinsic physical feature vectors include: Inject a high-frequency composite physical detection signal containing wideband linear frequency modulated pulses and dual-frequency equal-amplitude sine waves into the sensor physical cable; The response signal is captured through the first channel, the frequency domain complex reflection coefficient is calculated, and the third-order intermodulation distortion power ratio based on dual-frequency equal-amplitude sine wave excitation is extracted. By suppressing the fundamental frequency interference through the second channel, the power spectral density of the sensor's intrinsic low-frequency electrical noise is calculated and the flicker noise slope constant is extracted. The intrinsic physical feature vector is determined by concatenating the third-order intermodulation distortion power ratio, the frequency domain complex reflection coefficient, and the flicker noise slope constant.

6. The electronic evidence collection system for off-site environmental monitoring of pollution sources combined with hash operation as described in claim 1, characterized in that, The device private key derivation module includes: The static physical response is obtained through the unified interface of the security platform abstraction layer, and then concatenated with the factory security salt value to construct the initial key material. Within a trusted execution environment, a hash-based message authentication code key derivation function is executed. A pseudo-random key is generated through the extraction phase, and iterative computation is performed in the expansion phase in combination with specific application information to determine the output key material. The output key material is converted into an asymmetric device private key and loaded into a dedicated volatile secure register, followed by a forced erasure of the intermediate computation stack.

7. The electronic evidence collection system for off-site environmental monitoring of pollution sources combined with hash operation as described in claim 1, characterized in that, The evidence storage encapsulation and hardware signature module includes: The trusted real-time clock and hardware monotonic counter within the trusted execution environment are invoked and cross-validated with the cloud-based reference clock to generate a high-precision trusted system timestamp. The type length value encoding standard is used to serialize and encode each field of the payload to be signed, and dynamic field separators are injected at the boundaries of the encoding blocks; The payload is assembled within the encrypted isolated memory and read / write permissions are frozen in one direction. Then, the national cryptographic algorithm is called to execute the digital signature.

8. The electronic evidence collection system for off-site environmental monitoring of pollution sources combined with hash operation as described in claim 7, characterized in that, After generating the live electronic evidence certificate, the evidence encapsulation and hardware signature module also performs the following: The payload to be signed, the digital signature, the firmware version hash, and the secure boot state machine vector are encapsulated into an independent electronic evidence compartment. The electronic evidence compartment is encrypted using a symmetric encryption algorithm and then pushed into a persistent asynchronous transmission queue based on a local encrypted file system.

9. The electronic evidence collection system for off-site environmental monitoring of pollution sources combined with hash operation as described in claim 1, characterized in that: When performing verification, the cloud-based independent verification module introduces a cuckoo filter to perform concurrent existence verification on adaptive hybrid physical micro-entropy random numbers in order to intercept replay attacks.

10. The electronic evidence collection system for off-site environmental monitoring of pollution sources combined with hash operation as described in claim 1, characterized in that, The device digital twin trust model includes: During the cold start phase, initial values ​​of temperature sensitivity coefficient, humidity correlation coefficient, and aging drift slope are obtained through laboratory pre-calibration parameters or big data prior libraries. During the verification phase, the parameter state vector of the digital twin trust model is corrected online using the recursive least squares method or the extended Kalman filter matrix based on the measured residuals.

Images