Early identification method and system for building electrical hidden danger based on weak signal monitoring

Abstract

This invention relates to the field of building electrical fire early warning technology, specifically disclosing a method and system for early identification of building electrical hazards based on weak signal monitoring. The method includes: using geomagnetic induced current as a passive excitation source to capture the structural phonon flow excited in the micro-region of the hazard, collecting near-field evanescent electromagnetic waves leaking from the insulation surface, and establishing a phonon-evanescent wave coupling characteristic flow; inverting the electrochemical migration concentration field and space charge accumulation density field based on the phonon group velocity dispersion and evanescent wave polarization characteristics, constructing a time-series causal graph model using the transfer entropy algorithm, and calculating the causal structure deviation from the dynamic causal baseline; matching the deviation with multidimensional hazard criteria to distinguish between electrochemical migration-dominated and space charge breakdown-dominated hazards, and outputting accurate early warnings. This invention, by revealing the causal coupling mechanism between electrochemical migration and space charge accumulation, achieves non-invasive ultra-early identification and etiology-level diagnosis of building electrical hazards in their nascent stage.

Description

Technical Field

[0001] This invention relates to the field of building electrical fire early warning technology, specifically to a method and system for early identification of building electrical hazards based on weak signal monitoring. Background Technology

[0002] Existing building electrical hazard monitoring technologies mainly rely on residual current threshold alarms, infrared thermal imaging temperature over-limit triggering, or harmonic amplitude detection. These methods simplify hazard identification to the amplitude judgment of a single electrical parameter, triggering a response only when the fault has developed to the point of generating a strong signal or significant temperature rise. Essentially, they are reactive detection rather than early prevention. With the large-scale integration of nonlinear loads such as variable frequency air conditioners, LED lighting, and elevator group control into building power distribution systems, background electromagnetic interference and load fluctuations are becoming increasingly complex. Traditional monitoring schemes based on fixed thresholds frequently produce false alarms or missed alarms, making it difficult to detect nascent hazards such as slight increases in contact resistance and charge accumulation on insulation surfaces.

[0003] From the perspective of physical failure mechanisms, the emergence of electrical hazards in buildings is essentially a competitive and coupled evolution of two microscopic processes: electrochemical migration at the metal contact interface and space charge accumulation on the surface of the insulating material. Electrochemical migration refers to the formation of conductive dendrites by metal ions crossing the contact interface under the drive of an electric field, while space charge accumulation refers to the accumulation of charge at insulation defects, resulting in local electric field distortion. Current technologies have failed to establish a causal relationship model between these two processes, making it impossible to identify which is the dominant driving factor, let alone reveal the temporal dynamics of ion migration driving charge accumulation or charge feedback accelerating migration. This leads to a lack of targeted early warning information and significant blindness in maintenance decisions.

[0004] Furthermore, existing monitoring methods mostly employ static baselines or fixed thresholds, making it difficult to adapt to dynamic changes in building loads and background drift caused by seasonal environmental fluctuations. They also lack the statistical learning capability for the range of electrochemical migration and natural fluctuations in space charge under normal operating conditions. For the natural excitation formed by geomagnetic induced currents in building steel structures, existing technologies treat it as an interference source and filter it out, failing to utilize this passive excitation to stimulate the characteristic responses of potential hazard micro-regions.

[0005] Therefore, there is an urgent need for an early identification method for building electrical hazards that can penetrate complex electromagnetic environment noise, reveal the dynamic causal relationship between electrochemical migration and space charge accumulation, and assess the degree of distortion of the current causal network based on historical statistical baselines. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for early identification of building electrical hazards based on weak signal monitoring, so as to solve the technical problems of existing building electrical monitoring technologies that rely solely on a single physical quantity threshold for judgment, thus failing to reveal the dynamic causal coupling mechanism of electrochemical migration and space charge accumulation, and making it difficult to accurately distinguish the dominant type of hazard in the early stage.

[0007] To solve the above-mentioned technical problems, the present invention specifically provides the following technical solution:

[0008] An early identification method for electrical hazards in buildings based on weak signal monitoring includes the following steps:

[0009] S1. Capture the structural phonon flow and near-field evanescent electromagnetic waves excited by potential electrical hazards in buildings, and output the phonon-evanescent wave coupled characteristic flow after time-frequency domain coupling and background removal;

[0010] S2. Based on the phonon-evanescent wave coupled characteristic flow inversion electrochemical migration concentration field and space charge accumulation density field, construct a time-series causal graph model between the two, and calculate the causal structure deviation between the current time-series causal graph model and the dynamic causal baseline graph;

[0011] S3. Match the causal structure deviation with the multidimensional hidden danger criteria, determine the hidden danger type based on the matching result, and output early warning information.

[0012] As a preferred embodiment of the present invention, S1 specifically includes:

[0013] S11. A piezoelectric surface acoustic wave sensor array is attached to the metal connection interface of the building electrical load-bearing structure to capture the structural phonon flow excited by the geomagnetic induced current through the hidden danger micro-region, the structural phonon flow carrying the scattering characteristics of lattice thermal vibration at the grain boundary;

[0014] S12. Simultaneously arrange near-field capacitive coupling probes on the surface of the insulating material, with a probe spacing of no more than 2 mm and a sampling rate of no less than 5 MHz, to collect near-field evanescent electromagnetic waves leaking from the surface of the insulating material due to the redistribution of electrical hazards.

[0015] S13. Perform short-time Fourier transform on the structured phonon flow and the near-field evanescent electromagnetic wave respectively, calculate the coherence function and mutual information of the two within the same time window, extract the coherence peak frequency band and mutual information delay characteristics, and establish a coupling mapping relationship in the time and frequency domain.

[0016] S14. Perform empirical mode decomposition on the time spectrum corresponding to the coupling mapping relationship, extract and remove the diurnal variation periodic component of the geomagnetic background and the baseline drift component of the building thermal cycle, and output the phonon-evanescent wave coupling characteristic flow that reflects the energy dissipation state of the hidden danger micro-area.

[0017] As a preferred embodiment of the present invention, S11 specifically includes:

[0018] S111. At the metal connection interface of the building's electrical load-bearing structure, a piezoelectric surface acoustic wave sensor array is arranged at equal intervals along the main current path direction, with the spacing between adjacent sensors not exceeding 50mm, and the array coverage area covers areas prone to electrical hazards; the center frequency of the piezoelectric surface acoustic wave sensor array is configured to be between 500kHz and 2MHz, and the sampling rate is configured to be not less than 10MHz, so that the frequency response of the sensor array covers the main frequency band of the thermal vibration of the metal lattice;

[0019] S112. Using geomagnetic induced current as a natural excitation source, when geomagnetic induced current flows through the hidden danger micro-region, the abnormal Joule heating effect excites lattice thermal vibration at the metal grain boundary, and the structure phonon flow propagating along the metal surface is captured by the piezoelectric surface acoustic wave sensor array.

[0020] S113. Extract scattering features generated at metal grain boundaries by lattice thermal vibrations from the captured structural phonon flow, the scattering features including phonon group velocity dispersion curves, grain boundary scattering attenuation coefficients, and time delay broadening of phonon wave packets.

[0021] As a preferred embodiment of the present invention, S14 specifically includes:

[0022] S141. Obtain the time spectrum formed by the time-frequency domain coupling mapping of the structural phonon flow and the near-field evanescent electromagnetic wave. The time spectrum includes the energy dissipation characteristics of the hidden danger micro-region, as well as the diurnal variation periodic component of the geomagnetic background and the baseline drift component of the building thermal cycle. Perform empirical mode decomposition on the time spectrum to decompose it into multiple intrinsic mode function components, each of which corresponds to different time scale characteristics.

[0023] S142. Identify the components with a period of 24 hours ± 2 hours among the multiple intrinsic mode function components, and use them as the diurnal variation periodic components of the geomagnetic background and remove them from the time spectrum;

[0024] S143. Identify the slow-changing trend terms with a variation period greater than 12 hours among the multiple intrinsic mode function components, and use them as the building thermal cycle baseline drift components and remove them from the time spectrum;

[0025] S144. The remaining intrinsic mode function components after removing the diurnal variation periodic component of the geomagnetic background and the baseline drift component of the building thermal cycle are reconstructed to obtain the net time spectrum reflecting the energy dissipation state of the hidden danger micro-area, which is used as the output of the phonon-evanescent wave coupling characteristic flow.

[0026] As a preferred embodiment of the present invention, S2 specifically includes:

[0027] S21. Extract the phonon group velocity dispersion curve and grain boundary scattering attenuation coefficient from the phonon-evanescent wave coupled characteristic flow, establish constitutive relations based on the anisotropic medium propagation theory, and inversely derive the metal ion enrichment concentration distribution at the grain boundary through least squares fitting, which serves as the electrochemical migration concentration field.

[0028] S22. Extract the polarization state rotation angle and phase delay accumulation from the near-field evanescent electromagnetic wave, establish the functional relationship between polarization state rotation angle, phase delay and space charge density using the propagation model of evanescent wave in dielectric constant layered medium, and substitute the measured values ​​to obtain the space charge accumulation density field.

[0029] S23. Using the spatiotemporal sequence of the electrochemical migration concentration field and the space charge accumulation density field as nodes, calculate the bidirectional transfer entropy value, take the transfer entropy value that is significantly greater than zero as the causal correlation strength of the directed edge and record the time delay characteristics, and construct a time-series causal graph model.

[0030] S24. Compare the current temporal causal graph model with the dynamic causal baseline graph, calculate the topological difference of the causal edge set, the magnitude deviation of the causal intensity, and the time delay deviation of the time delay feature, and use the weighted sum of the three as the causal structure deviation.

[0031] As a preferred embodiment of the present invention, S21 specifically includes:

[0032] S211. Extract the phonon group velocity dispersion curve and grain boundary scattering attenuation coefficient from the phonon-evanescent wave coupled characteristic flow. The phonon group velocity dispersion curve records the relationship between the propagation speed of phonon wave packets at different frequencies and the frequency. The grain boundary scattering attenuation coefficient characterizes the degree of energy attenuation of the phonon flow at the metal grain boundary due to scattering.

[0033] S212. Based on the theory of sound wave propagation in anisotropic media, establish the first constitutive relationship between phonon group velocity and lattice thermal conductivity and grain boundary density, and the second constitutive relationship between grain boundary scattering attenuation coefficient and metal ion enrichment concentration at grain boundaries.

[0034] S213. Substitute the phonon group velocity dispersion curve and grain boundary scattering attenuation coefficient into the first constitutive relation and the second constitutive relation, and use the least squares fitting algorithm to invert the spatial distribution of metal ion enrichment concentration at the grain boundary. Output the inverted spatial distribution of metal ion enrichment concentration as an electrochemical migration concentration field. The electrochemical migration concentration field records the metal ion concentration values ​​at each spatial location of the electrical contact interface in the form of a three-dimensional grid.

[0035] As a preferred embodiment of the present invention, S23 specifically includes:

[0036] S231. The electrochemical migration concentration field obtained by inversion is organized into a spatiotemporal sequence of electrochemical migration concentration field according to the time order. At the same time, the space charge accumulation density field obtained by inversion is organized into a spatiotemporal sequence of space charge accumulation density field according to the same time axis. The time resolution of the two spatiotemporal sequences is consistent.

[0037] S232. Using the spatiotemporal sequence of the electrochemical migration concentration field as the source node sequence and the spatiotemporal sequence of the space charge accumulation density field as the target node sequence, calculate the transfer entropy value from the electrochemical migration concentration field to the space charge accumulation density field. The transfer entropy value quantifies the contribution of the historical information of the source node sequence to the prediction of the future state of the target node sequence.

[0038] S233. Using the spatiotemporal sequence of the space charge accumulation density field as the source node sequence and the spatiotemporal sequence of the electrochemical migration concentration field as the target node sequence, calculate the reverse transfer entropy value from the space charge accumulation density field to the electrochemical migration concentration field.

[0039] S234. The propagation entropy value and the reverse propagation entropy value are compared with a preset significance threshold respectively. The propagation entropy value that is greater than the significance threshold is taken as the causal association strength of the directed edge, and the time delay offset when the propagation entropy value is reached is recorded as the time delay feature.

[0040] S235. Using the electrochemical migration concentration field and the space charge accumulation density field as nodes, the causal correlation strength of the directed edges as edge weights, and the time delay characteristics as edge attributes, construct a time-series causal graph model.

[0041] As a preferred embodiment of the present invention, S24 specifically includes:

[0042] S241. Retrieve the dynamic causal baseline map from the dynamic causal baseline model. The dynamic causal baseline map is statistically generated based on multiple time-series causal graph models under historical no-hazard conditions. It includes the set of directed edges of the baselines, the causal intensity distribution range of each directed edge of the baselines, and the time delay characteristic distribution range of each directed edge of the baselines.

[0043] S242. Calculate the topological difference between the current temporal causal graph model and the dynamic causal baseline graph, wherein the topological difference includes the number of new edges, the number of missing edges, and the number of reverse edges in the current directed edge set relative to the baseline directed edge set;

[0044] S243. Calculate the strength deviation between the current temporal causal graph model and the dynamic causal baseline graph. For the directed edges in the current temporal causal graph model that are shared with the set of directed edges of the baseline, calculate the deviation ratio of the current causal association strength relative to the distribution range of the baseline causal strength.

[0045] S244. Calculate the time delay deviation between the current temporal causal graph model and the dynamic causal baseline graph. For the directed edges in the current temporal causal graph model that are shared with the set of directed edges of the baseline, calculate the deviation ratio of the current time delay feature relative to the distribution range of the baseline time delay feature.

[0046] S245. The topological difference, intensity deviation and time delay deviation are weighted and summed according to preset weights to obtain the causal structure deviation.

[0047] As a preferred embodiment of the present invention, S3 specifically includes:

[0048] S31. Obtain the deviation of the causal structure and the time delay characteristics of the directed edges in the time-series causal graph model, and compare the deviation of the causal structure with a number of preset hidden danger criteria respectively;

[0049] S32. When the deviation of the causal structure exceeds the first threshold and there is a directed edge in the time-series causal graph model pointing from the electrochemical migration concentration field to the space charge accumulation density field, and the time delay feature of the directed edge falls into the first time delay interval, the hidden danger type is determined to be an electrochemical migration-dominated hidden danger.

[0050] S33. When the deviation of the causal structure exceeds the second threshold and there is a directed edge in the time-series causal graph model pointing from the space charge accumulation density field to the electrochemical migration concentration field, and the time delay feature of the directed edge falls into the second time delay interval, the hidden danger type is determined to be a space charge breakdown dominant hidden danger.

[0051] S34. Based on the determined hazard type, combined with the numerical value and time delay characteristics of the causal structure deviation, estimate the remaining safe time, and output early warning information including hazard type, causal structure deviation, and remaining safe time.

[0052] An early identification system for electrical hazards in buildings based on weak signal monitoring is used to implement a method for early identification of electrical hazards in buildings based on weak signal monitoring, including:

[0053] The weak signal collaborative acquisition module is used to deploy a high-sensitivity surface acoustic wave sensor array and a near-field capacitive coupling probe on the surface of the building's electrical load-bearing structure to simultaneously capture the structural phonon flow and near-field evanescent electromagnetic waves, establish a time-frequency domain coupling mapping relationship, and eliminate the diurnal variation period of the geomagnetic background and the baseline drift of the building's thermal cycle, and output the phonon-evanescent wave coupling characteristic flow.

[0054] The physical field inversion and temporal causality analysis module is used to invert the electrochemical migration concentration field based on the phonon group velocity dispersion characteristics and grain boundary scattering attenuation intensity in the phonon-evanescent wave coupled characteristic flow, and to invert the space charge accumulation density field based on the polarization state rotation angle and phase delay accumulation of the near-field evanescent electromagnetic wave. It constructs a temporal causal graph model with the electrochemical migration concentration field and the space charge accumulation density field as nodes and the directed causal correlation strength calculated based on the transfer entropy algorithm as directed edges. The module compares the current temporal causal graph model with the dynamic causal baseline graph and calculates the causal structure deviation.

[0055] The hazard identification and early warning module is used to match the causal structure deviation with preset multidimensional hazard criteria. When the causal structure deviation meets any hazard criterion, the hazard type is determined and early warning information is output.

[0056] The beneficial effects of this invention are:

[0057] 1. By using geomagnetic induced current as a passive excitation source, the coupling characteristics of structural phonon current and near-field evanescent wave are captured, and the geomagnetic diurnal variation and thermal cycling baseline are eliminated to achieve non-invasive ultra-early monitoring, capturing micro-Euclidean level contact degradation and charge accumulation in the nascent stage.

[0058] 2. Based on transfer entropy, a time-series causal graph of electrochemical migration and space charge accumulation is constructed. By calculating the deviation of the causal structure from the dynamic causal baseline, the topological distortion of the causal network is identified rather than just the physical quantity exceeding the limit, thus accurately distinguishing between normal fluctuations and hidden evolution mechanisms.

[0059] 3. By matching multi-dimensional criteria such as deviation threshold, directed edge direction and time delay interval, the system can accurately distinguish between electrochemical migration-dominated and space charge breakdown-dominated hidden dangers, clearly identify the failure-dominant mechanism, and achieve the leap from abnormal alarm to cause diagnosis, thus guiding targeted maintenance. Attached Figure Description

[0060] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0061] Figure 1 This is a flowchart illustrating the method described in Embodiment 1 of the present invention.

[0062] Figure 2 This is a framework diagram of the system described in Embodiment 2 of the present invention. Detailed Implementation

[0063] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0064] In the description of this invention, unless otherwise explicitly specified and limited, the terms "connected," "linked," and "fixed" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0065] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0066] In the description of this embodiment, the terms "upper," "lower," "left," and "right," etc., refer to the orientation or positional relationship shown in the accompanying drawings. They are used only for ease of description and simplification of operation, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention. In addition, the terms "first" and "second" are used only for distinction in description and have no special meaning.

[0067] Example 1

[0068] like Figure 1 As shown, this invention provides a method for early identification of electrical hazards in buildings based on weak signal monitoring, comprising the following steps:

[0069] S1. Capture the structural phonon flow and near-field evanescent electromagnetic waves excited by potential electrical hazards in buildings, and output the phonon-evanescent wave coupled characteristic flow after time-frequency domain coupling and background removal; specifically including:

[0070] S11. Array deployment and phonon stream capture, specifically:

[0071] S111. Array topology deployment and frequency response configuration, specifically:

[0072] At the metal connection interface of the building's electrical load-bearing structure, a piezoelectric surface acoustic wave sensor array is mounted at equal intervals along the main current path. The spacing between adjacent sensors is set to be no more than 50mm. The overall coverage of the array must completely cover areas prone to electrical hazards, including terminal blocks, copper busbar lap surfaces, and circuit breaker inlet and outlet terminals.

[0073] The center frequency of the piezoelectric surface acoustic wave sensor array is configured to be in the 500kHz to 2MHz band, and the sampling rate is configured to be no less than 10MHz, so that the frequency response of the sensor array can effectively cover the main frequency band of the thermal vibration of the metal lattice.

[0074] S112. Geomagnetic excitation and phonon excitation trapping, specifically:

[0075] Using geomagnetic induced current as a natural excitation source, generated by changes in the Earth's magnetic field within the circuitry of a building's steel structure, this current, when flowing through a micro-region of electrical hazard, induces lattice thermal vibrations at the metal grain boundaries due to an abnormal Joule heating effect. This generates a structural phonon flow that propagates along the metal surface. This structural phonon flow is captured in real-time using a piezoelectric surface acoustic wave sensor array. This flow carries information about the scattering characteristics of the lattice thermal vibrations at the metal grain boundaries within the hazard micro-region, reflecting the energy dissipation state of the hazard micro-region.

[0076] S113. Scattering feature extraction and parameter analysis, specifically:

[0077] Scattering characteristics of lattice thermal vibrations at metal grain boundaries were extracted from the structured phonon flow captured by a piezoelectric surface acoustic wave sensor array. These scattering characteristics specifically included phonon group velocity dispersion curves, grain boundary scattering attenuation coefficients, and time delay broadening characteristics of phonon wave packets. Broadband spectral analysis and group velocity calculations of the structured phonon flow were performed to obtain phonon group velocity dispersion curves for phonon components of different frequencies. The grain boundary scattering attenuation coefficient, characterizing the scattering intensity of impurity ions at grain boundaries, was obtained by analyzing the energy attenuation law during phonon propagation. The time delay broadening parameter, reflecting the phonon coherence length, was obtained by measuring the time broadening of the phonon wave packets.

[0078] S12. Near-field evanescent wave acquisition, specifically:

[0079] A near-field capacitive coupling probe array is arranged above the surface of the insulating material. The probe adopts a subwavelength scale electrode structure, the electrode spacing is set to no more than 2 mm, and the sampling rate is configured to no less than 5 MHz to ensure high-fidelity capture of high-frequency charge fluctuations.

[0080] The probe senses the electromagnetic field leaked from the surface of the insulating material due to charge redistribution caused by electrical hazards via capacitive coupling. This electromagnetic field exists as a near-field evanescent wave, decaying exponentially with distance, and carries information about the gradient of space charge accumulation density on the insulating surface. The probe array is strictly clock-synchronized with the piezoelectric surface acoustic wave sensor array to achieve synchronous acquisition of the structural phonon flux and the near-field evanescent electromagnetic wave, ensuring the time alignment accuracy when establishing the subsequent time-frequency domain coupling mapping.

[0081] S13. Establishment of time-frequency domain coupling mapping, specifically as follows:

[0082] S131. Input the structured phonon stream and the near-field evanescent electromagnetic wave into the signal processing unit for short-time Fourier transform. Set a sliding time window, the window length of which covers the phonon wave period, and the step interval to ensure time-frequency resolution balance. Map the time-domain signal to the time-frequency plane through fast Fourier transform to obtain the time-frequency spectrum representation of the structured phonon stream and the time-frequency spectrum representation of the near-field evanescent electromagnetic wave.

[0083] S132. Perform frequency domain coherence analysis on the time spectra of the two signals at the same time window. Calculate the ratio of the cross-power spectral density to the individual power spectral density of the two signals at each frequency point to obtain the coherence function value characterizing the phase coherence strength. The coherence function value ranges from 0 to 1; the closer the value is to 1, the stronger the phase coherence of the two signals in that frequency band.

[0084] The mutual information is calculated synchronously by estimating the probability density of the amplitude values ​​of the two signals within the same time window, thus obtaining the joint probability distribution and marginal probability distribution. The mutual information value is calculated by summing the log-likelihood ratio. The larger the mutual information value, the stronger the statistical dependence between the two signals, reflecting the intensity of information transmission.

[0085] S133. When extracting the coherent peak frequency band, scan the coherence function curve to identify the continuous frequency band range where the coherence function value exceeds a preset threshold, and determine this range as the coherent peak frequency band. This frequency band reflects the main frequency range of the coupling between the structure phonon current and the near-field evanescent electromagnetic wave energy, and characterizes the dominant frequency band of multi-physics coupling in the hidden danger micro-region.

[0086] When extracting mutual information delay features, cross-correlation analysis using a sliding time window is used to find the time offset corresponding to the maximum mutual information value. This offset is the mutual information delay feature. Positive values ​​indicate that information is transmitted from the structural phonon flow to the near-field evanescent electromagnetic wave, while negative values ​​indicate the reverse transmission. The magnitude of the value reflects the coupling delay time.

[0087] Spatiotemporal registration is performed based on coherent peak frequency band and mutual information delay characteristics to construct a coupling mapping relationship in the time-frequency domain. This mapping relationship, with time as the horizontal axis and frequency as the vertical axis, is weighted by coherence intensity and mutual information value to characterize the energy dissipation of electrical hazard micro-regions in multiple physical fields and the distribution of coupling strength.

[0088] S14. Background removal and feature stream output, specifically:

[0089] S141. Empirical Mode Decomposition and Eigenmode Extraction, specifically:

[0090] The time-frequency spectrum is obtained after time-frequency domain coupling mapping of the structural phonon flow and the near-field evanescent electromagnetic wave. This time-frequency spectrum simultaneously includes the rapid transient characteristics caused by energy dissipation in the micro-regions of electrical hazards, as well as the slow-varying trends caused by the diurnal variation periodic component of the geomagnetic background and the baseline drift component of the building thermal cycle. Empirical mode decomposition is performed on this time-frequency spectrum. First, all local maxima and local minima of the time-frequency spectrum sequence are identified. The upper and lower envelopes are fitted by cubic spline interpolation, and the mean of the upper and lower envelopes is calculated as the instantaneous mean envelope.

[0091] Subtracting the mean envelope from the original time-frequency spectrum yields the first residual sequence. The residual sequence is then checked to determine if it meets the intrinsic mode function (IMF) criteria: the difference between the number of extrema and the number of zero-crossings is no greater than 1, and the mean of the upper and lower envelopes is zero. If not, the screening process is repeated until the criteria are met, yielding the first IMF component. The remaining residuals are screened repeatedly, and subsequent IMF components are extracted sequentially until the residuals are monotonic or their energy is below a threshold. This process ultimately yields a set of multiple IMF components arranged from high to low frequency, each component corresponding to a different timescale characteristic.

[0092] S142. Identification and removal of geomagnetic periodic components, specifically:

[0093] The obtained intrinsic mode function components were periodically and accurately identified. The dominant frequency was obtained by calculating the period of the autocorrelation function of each component or by performing a fast Fourier transform. Intrinsic mode function components whose main period falls within the range of 24 hours ± 2 hours were selected. Further verification with cross-correlation data of geomagnetic induced current confirmed that this component is synchronized with the diurnal variation of the geomagnetic field and was identified as a diurnal variation periodic component of the geomagnetic background.

[0094] The component is removed from the valid signal by either component elimination or amplitude zeroing. At the same time, the amplitude envelope of the component is recorded as a reference for geomagnetic interference. This ensures that the daily fluctuation of geomagnetic induced current does not interfere with the extraction of energy dissipation characteristics of the hidden danger micro-area, and retains the non-periodic transient characteristics related to the hidden danger for subsequent analysis.

[0095] S143. Identification and removal of thermal cycling trend items, specifically:

[0096] In the remaining intrinsic modal function components, the instantaneous frequency trajectory of each component is obtained through Hilbert transform, or the average period of each component is calculated to identify low-frequency slow-varying trend terms with a variation period greater than 12 hours. Alternatively, low-frequency components characterizing long-term trends can be separated using energy weight accumulation curves. Further comparison of the correlation between this type of component and building structure temperature monitoring data or load switching records confirms that it reflects the baseline drift characteristics caused by building load thermal cycling, classifying it as a building thermal cycle baseline drift component. This slow-varying trend term is then removed from the intrinsic modal function component set to eliminate the influence of temperature drift and mechanical thermal expansion and contraction background vibrations on the hazard signal, ensuring that only transient characteristics reflecting energy dissipation in the hazard micro-area are retained.

[0097] S144. Signal reconstruction and feature stream output, specifically:

[0098] The remaining intrinsic mode function (IMF) components, after removing the diurnal variation periodic component of the geomagnetic background and the baseline drift component of the building thermal cycle, are linearly superimposed and reconstructed. During reconstruction, the IMF components characterizing the transient features of the hazard in the mid-to-high frequency bands are strictly preserved, while residual low-frequency trend interference is eliminated. The amplitudes of each component are linearly superimposed after energy normalization, or weighted and summed according to their matching degree with the standard hazard characteristic template, generating a high signal-to-noise ratio net-time spectrum. This net-time spectrum is output as a phonon-evanescent wave coupling characteristic stream. This characteristic stream has completely filtered out environmental background interference, retaining only the coupling characteristics between the structural phonon stream excited by the anomalous Joule heating effect in the electrical hazard micro-region and the near-field evanescent electromagnetic wave from charge redistribution leakage, used for subsequent inversion calculations of the electrochemical migration concentration field and the space charge accumulation density field.

[0099] S2. Based on the phonon-evanescent wave coupled characteristic flow inversion of the electrochemical migration concentration field and the space charge accumulation density field, a time-series causal graph model between the two is constructed, and the causal structure deviation between the current time-series causal graph model and the dynamic causal baseline graph is calculated; specifically including:

[0100] S21. Phonon feature extraction and electrochemical migration field inversion, specifically:

[0101] S211. Phonon flux spectral decomposition and scattering feature extraction, specifically:

[0102] The time series of the structured phonon flow was separated from the phonon-evanescent wave coupled characteristic flow. Broadband spectral decomposition was performed on this time series to extract phonon wave packet components at different center frequencies. For each frequency component of the phonon wave packet, the group velocity of the phonon wave packet propagating along the metal surface was calculated by dividing the distance by the time difference, using the arrival time difference of the signals received by each element in the piezoelectric surface acoustic wave sensor array and the array's geometric layout parameters. By traversing all frequency components, a mapping relationship between the phonon group velocity and frequency was established, and the phonon group velocity dispersion curve was plotted.

[0103] Simultaneously, by calculating the ratio of the incident phonon wave packet energy to the transmitted phonon wave packet energy after scattering by the metal grain boundary, or by analyzing the exponential decay law of the phonon flux along the propagation path, and combining the distance normalization with the sensor array spacing, the percentage of energy attenuation per unit propagation distance is obtained as the grain boundary scattering attenuation coefficient.

[0104] S212. Anisotropic constitutive relation modeling and parameter mapping, specifically:

[0105] Based on the theories of continuum mechanics and anisotropic acoustics, a quantitative constitutive correlation between phonon propagation characteristics and material microstructure is established for the polycrystalline aggregation characteristics of metal connection interfaces in building electrical load-bearing structures.

[0106] For the first constitutive relation, considering the constraint effect of grain boundaries on the mean free path of phonons during the thermal vibration transmission of the metal lattice, an effective medium approximation model is introduced to model grain boundaries as phonon scattering centers. A theoretical correlation is established between the phonon group velocity and the lattice thermal conductivity and grain boundary density. This correlation states that the phonon group velocity increases with the increase of lattice thermal conductivity and decreases with the increase of grain boundary density. Specifically, by constructing a correction function for grain boundary density and effective sound velocity, and combining it with the anisotropic elastic tensor to describe the differences in phonon propagation in different crystal orientations, an analytical expression reflecting the influence of lattice integrity on phonon transport is obtained.

[0107] For the second constitutive relation, based on the phonon scattering cross section theory in solid-state physics, the local lattice distortion and acoustic impedance mismatch effect caused by the enrichment of metal ions at the grain boundary are analyzed. A quantitative mapping between the grain boundary scattering attenuation coefficient and the concentration of metal ions enriched at the grain boundary is established. This mapping takes into account that the enrichment of metal ions at the grain boundary will increase the phonon scattering cross section in the grain boundary region. By establishing the response function of the scattering cross section and the ion enrichment concentration, a nonlinear or linear relationship model in which the grain boundary scattering attenuation coefficient increases monotonically with the increase of the metal ion enrichment concentration is obtained, thus completing the construction of a physical bridge from phonon propagation characteristics to microscopic ion distribution.

[0108] S213. Inversion optimization and concentration field generation, specifically:

[0109] The measured phonon group velocity dispersion curve and grain boundary scattering attenuation coefficient were used as the observation dataset and substituted into the forward modeling physical model composed of the first and second constitutive relations. An initial spatial distribution prediction of the metal ion enrichment concentration was set, and the theoretical phonon group velocity and theoretical grain boundary scattering attenuation coefficient at each frequency point under this predicted distribution were calculated using the forward modeling. A weighted least squares objective function was constructed, defined as the sum of squares of the residuals between the theoretical prediction and the measured values, with the weights determined based on the signal-to-noise ratio of each frequency component in the sensor's frequency response curve.

[0110] The Levenberg-Marquardt nonlinear least squares optimization algorithm or trust region reflection algorithm is employed to iteratively adjust the distribution parameters of metal ion enrichment concentration at the nodes of a 3D spatial grid, ensuring that the objective function converges to a global minimum or a local minimum. Nonnegativity constraints and spatial smoothness constraints are applied during the optimization process to ensure the physical validity of the solution. The calculation terminates when the residual norm is less than a preset convergence threshold or the number of iterations reaches its upper limit. The metal ion concentration values ​​at each grid node in the 3D space are output, and this 3D grid data field is used as an electrochemical migration concentration field for subsequent node construction of the time-series causal graph model.

[0111] S22. Evanescent wave analysis and space charge field inversion, specifically:

[0112] S221. The near-field evanescent electromagnetic wave acquired by the near-field capacitively coupled probe array is first input into the polarization analytical unit. The electromagnetic wave is orthogonally polarized and decomposed to extract its horizontal electric field component and vertical electric field component. By calculating the amplitude ratio and phase difference of the two components, the polarization ellipse parameters are constructed, and then the tilt angle of the principal axis of the polarization ellipse relative to the reference direction is solved to obtain the polarization state rotation angle.

[0113] Simultaneously, using the excitation phase of the geomagnetic induced current or the reference phase provided by the internal synchronization clock of the probe array as a reference, the phase lag of the near-field evanescent electromagnetic wave relative to the reference phase is measured. By tracking the cumulative effect of this phase lag along the propagation path or with the observation time, the cumulative phase delay is obtained.

[0114] S222. Establish the propagation mechanism of evanescent waves in dielectric constant layered media. Model the surface layer of the insulating material as a non-uniform medium in which the dielectric constant varies with depth. The accumulation of space charge forms a charge density gradient, which modulates the local effective dielectric constant through electrostriction effect or dielectric response mechanism, forming a dielectric constant layered structure.

[0115] Based on Maxwell's equations, the propagation equation of near-field evanescent waves in such non-uniform dielectric constant distributions is derived. The sensitivity of polarization state rotation and phase delay accumulation to dielectric constant gradient changes is analyzed. A quantitative functional relationship between polarization state rotation angle, phase delay accumulation, and space charge density gradient is established. This functional relationship describes how the dielectric tensor change caused by space charge accumulation modulates the polarization characteristics and propagation phase of electromagnetic waves.

[0116] S223. Substitute the measured polarization rotation angle and phase delay accumulation as observed values ​​into the above functional relationship to construct a set of observation equations with the space charge accumulation density field as the unknown. Solve the set of equations using a regularized least squares inversion algorithm or an iterative perturbation inversion method based on the Born approximation. During the inversion process, apply a non-negativity constraint on the space charge density and a spatial smoothing regularization term to ensure the physical rationality and spatial continuity of the solution. Minimize the residual between the theoretical prediction and the measured observation through iterative optimization, and finally obtain the space charge density values ​​at each spatial location on the surface of the insulating material. Output this spatial distribution data as the space charge accumulation density field.

[0117] S23. Transitive entropy causal discovery and temporal causal graph construction, specifically:

[0118] S231. Construction of spatiotemporal sequences of concentration and density fields, specifically:

[0119] The electrochemical migration concentration field obtained by inversion is arranged and organized according to the chronological order of data acquisition to form a spatiotemporal sequence of the electrochemical migration concentration field. Specifically, a unique timestamp is assigned to the three-dimensional grid data of the electrochemical migration concentration field obtained at each sampling moment, and the concentration field data at each moment are arranged sequentially according to the timestamp from early to late, forming a four-dimensional spatiotemporal sequence that combines the time and spatial dimensions.

[0120] Simultaneously, the inverted space charge accumulation density field is organized along the exact same time axis to form a spatiotemporal sequence of the space charge accumulation density field, ensuring that the two spatiotemporal sequences are strictly aligned at the time nodes. A uniform time resolution is set, that is, the time interval between two adjacent concentration field or density field data remains constant and consistent. This time resolution matches the original sampling rate of the phonon-evanescent wave coupled characteristic flow to ensure the comparability of the time scales of the two sequences in subsequent transfer entropy calculations.

[0121] S232. Forward transit entropy calculation and driving causal edge construction, specifically:

[0122] Using the spatiotemporal sequence of the electrochemical migration concentration field as the source node sequence and the spatiotemporal sequence of the space charge accumulation density field as the target node sequence, the transfer entropy value characterizing the causal direction of electrochemical migration driving space charge accumulation is calculated. In specific implementation, the phase space of the spatiotemporal sequence of the electrochemical migration concentration field is first reconstructed, and the embedding dimension and delay time parameters are set to construct a set of historical state vectors characterizing the dynamic state of the system.

[0123] The same operation is performed simultaneously on the spatiotemporal sequence of the space charge accumulation density field. The historical state vector of the electrochemical migration concentration field spatiotemporal sequence is used as a set of conditional information. The conditional entropy of this set of information for the future state of the space charge accumulation density field spatiotemporal sequence is calculated and compared with the unconditional marginal entropy, yielding the conditional entropy subtraction, i.e., the transfer entropy value. This transfer entropy value quantifies the degree of reduction in uncertainty of space charge accumulation after knowing the historical information of electrochemical migration, reflecting the amount of information transferred and the predictive contribution of the electrochemical migration process to the space charge accumulation process.

[0124] By sliding the calculation across multiple historical time windows and taking the statistical average, a stable forward transfer entropy estimate is obtained. This forward transfer entropy estimate will serve as a candidate value for the causal correlation strength of the directed edge from the electrochemical migration concentration field node to the space charge accumulation density field node in the time-series causal graph model, characterizing the causal driving strength of the electrochemical migration process driving the space charge accumulation process. Its time delay characteristics are determined by scanning the peak position of the transfer entropy at different time offsets.

[0125] S233. Calculation of reverse transitive entropy and construction of feedback causal edges, specifically:

[0126] Using the spatiotemporal sequence of the space charge accumulation density field as the source node sequence and the spatiotemporal sequence of the electrochemical migration concentration field as the target node sequence, the reverse propagation entropy value characterizing the causal direction of electrochemical migration driven by space charge accumulation feedback is calculated. Employing a computational flow symmetric to S232, the phase space of the space charge accumulation density field spatiotemporal sequence is first reconstructed. The same embedding dimension and delay time parameters are set to ensure the comparability of bidirectional calculations, thus constructing a set of historical state vectors for space charge accumulation.

[0127] By using the historical state vector of the spatiotemporal sequence of the space charge accumulation density field as a set of conditional information, the conditional entropy reduction of this set on the future state of the electrochemical migration concentration field is calculated, yielding an estimate of the back-transfer entropy. This estimate quantifies the contribution of historical space charge accumulation information to the prediction of the future state of electrochemical migration, reflecting the reverse causal driving force of the space charge accumulation process accelerating electrochemical migration ion transport through the feedback enhancement effect of local electric field distortion.

[0128] By sliding the calculation across multiple historical time windows and taking the statistical average, a stable estimate of the reverse transfer entropy is obtained. This estimate of the reverse transfer entropy will serve as a candidate value for the causal correlation strength of the directed edge from the space charge accumulation density field node to the electrochemical migration concentration field node in the time-series causal graph model. Its time delay characteristics are determined by scanning the peak position of the transfer entropy at different time offsets.

[0129] S234. Causal significance determination and edge attribute extraction, specifically:

[0130] The forward and backward propagation entropy values ​​are compared with a preset significance threshold, which is determined by a permutation test method. This method involves randomly shuffling the temporal order of the source node sequence to destroy temporal correlation, calculating the zero propagation entropy distribution, and taking the upper boundary of its confidence interval as the significance criterion.

[0131] If the forward transfer entropy value is greater than the significance threshold, a significant causal relationship is confirmed from the electrochemical migration concentration field to the space charge accumulation density field. The forward transfer entropy value is assigned to the corresponding directed edge in the time-series causal graph model as the causal association strength, and the time offset of the maximum information transmission efficiency when calculating the transfer entropy is recorded as the time delay feature of the directed edge. This time delay feature characterizes the time lag of the space charge accumulation response caused by the change in electrochemical migration state.

[0132] The same judgment process is applied to the reverse propagation entropy value. If it is greater than the significance threshold, a reverse directed edge is established from the space charge accumulation density field to the electrochemical migration concentration field, and the corresponding causal correlation strength and time delay characteristics are assigned. For propagation entropy values ​​that fail the significance test, no corresponding directed edge is established in the time-series causal graph model to ensure that the graph structure only retains statistically significant causal relationships.

[0133] S235. Temporal cause-effect graph network integration and topology solidification, specifically:

[0134] Using the physical field corresponding to the spatiotemporal sequence of the electrochemical migration concentration field as the first node, and the physical field corresponding to the spatiotemporal sequence of the space charge accumulation density field as the second node, a complete temporal causal graph model is constructed, with the directed edges retained after significance determination as the edge set, the causal correlation strength attached to each directed edge as the edge weight, and the time delay characteristics attached to each directed edge as the edge attribute. The topological structure of this temporal causal graph model is stored in the form of an adjacency matrix, where rows represent source nodes, columns represent target nodes, and matrix elements store causal correlation strength values, with zero values ​​indicating no significant causal relationship. Edge attributes are stored in the form of an independent time delay matrix, recording the time delay characteristics of the corresponding directed edges.

[0135] This time-series causal graph model explicitly includes two physical field nodes and zero, one, or two directed edges, corresponding to no causal relationship, unidirectional driving relationship, or bidirectional feedback relationship, respectively. This model not only records the causal driving relationship between electrochemical migration and space charge accumulation at the current moment, but also quantifies the driving intensity through edge weights and records the response delay through edge attributes, forming a snapshot of the causal network topology characterizing the dynamic mechanism of electrical hazard evolution. This model, as the final output of step S2, is input into subsequent steps to calculate topological differences, intensity deviations, and time delay deviations compared to the dynamic causal baseline graph, thereby assessing the degree of causal structure deviation caused by electrical hazards.

[0136] S24. Dynamic baseline comparison and causal structure deviation calculation are as follows:

[0137] S241. Baseline model retrieval and baseline graph structure analysis, specifically:

[0138] The dynamic causal baseline model is retrieved from the long-term historical data storage module deployed in the central processing unit. This model is generated iteratively using a statistical learning algorithm based on several sets of historical time-series causal graph models continuously collected and accumulated under manually confirmed no-hazard operating conditions. The internal data structure of this dynamic causal baseline model is stored in a graph database or matrix format, and its core comprises three main components: a set of directed edges on the baselines, statistical distribution parameters of the causal correlation strength corresponding to each directed edge, and statistical distribution parameters of the time delay characteristics corresponding to each directed edge. The set of directed edges on the baselines is recorded in the form of an adjacency matrix or edge list, clearly defining which causal relationship links between electrochemical migration concentration field nodes and space charge accumulation density field nodes are stable under no-hazard conditions, as well as the directionality of the edges. For each directed edge on the baseline, the statistical distribution parameters of the causal correlation strength store the statistical moments such as the mean, standard deviation, skewness, and kurtosis of its historical samples, or store the parameters of the fitted probability distribution function, to characterize the natural fluctuation range and confidence interval of the causal driving intensity under normal conditions. The statistical distribution parameters of the latency characteristics are also stored for each directed edge of the baseline, storing the mean, standard deviation and distribution type of historical latency samples, which characterize the stable range and degree of variation of information transmission response latency under normal conditions.

[0139] The dynamic causal baseline map used for comparison is extracted from the dynamic causal baseline model. The dynamic causal baseline map presents the set of directed edges of the baseline and its corresponding statistical distribution parameters in the form of a fixed graph topology, which serves as a quantitative reference benchmark for assessing whether the current electrical hazard status deviates from the normal operating conditions.

[0140] S242. Topological structure difference quantification and edge set comparison analysis, specifically:

[0141] The current temporal causal graph model generated in step S235 is compared with the dynamic causal baseline graph retrieved in step S241 using a deep topological structure comparison to accurately calculate the topological differences between the two. First, the current directed edge set of the current temporal causal graph model is extracted, which contains all significant causal links detected at the current time and their direction information.

[0142] An element-by-element comparison analysis is performed between the current set of directed edges and the baseline set of directed edges in the dynamic causal baseline graph. The number of directed edges that exist only in the current set but are missing from the baseline set is identified and counted; this is defined as the number of newly added edges. This indicator reflects the emergence of new causal driving paths or feedback loops induced by electrical hazards. The number of directed edges that exist only in the baseline set but are missing from the current set is identified and counted; this is defined as the number of missing edges. This indicator reflects the interruption of causal links or the failure of information transmission channels under normal operating conditions. The number of directed edges whose starting and ending nodes are completely opposite to those of the baseline edges is identified and counted; this is defined as the number of reverse edges. This indicator reflects the reversal of the dominant causal driving direction, such as a shift from electrochemical migration driving space charge accumulation to space charge accumulation dominating electrochemical migration.

[0143] The number of newly added edges, missing edges, and reverse edges are weighted according to their impact on system stability and then summed in a weighted manner. Alternatively, a set similarity index such as Jaccard distance can be used to obtain a quantified value of topological difference. The larger the value, the more severe the distortion of the current causal network topology relative to the normal state, indicating that the system dynamics mechanism caused by hidden dangers has undergone structural changes.

[0144] S243. Calculation and statistical distribution comparison of causal strength deviation, specifically:

[0145] For shared directed edges that coexist in both the current temporal causal graph model and the dynamic causal baseline graph—that is, causal links where the topology remains consistent but the causal driving strength may experience anomalous drift—a refined calculation of causal association strength deviation is performed. For each shared directed edge, the current causal association strength value of that edge is extracted from the current temporal causal graph model. This value is obtained by normalizing the transit entropy value calculated in step S232 or S233. The statistical distribution parameters of the causal association strength corresponding to that edge are extracted from the dynamic causal baseline model, including the mean and standard deviation of historical samples, or the probability density function parameters of complete Gaussian or gamma distributions.

[0146] Calculate the absolute deviation of the current causal correlation strength value relative to the baseline mean, and divide this absolute deviation by the baseline standard deviation to obtain the standardized deviation ratio, i.e., calculate the standard score. If the current value falls outside the 95% confidence interval of the baseline distribution, it is judged as a significant deviation and assigned a larger deviation weight. For extreme deviations exceeding the confidence interval, the tail probability of its cumulative distribution function can be further calculated. Traverse all common directed edges, and perform weighted statistical aggregation on the standardized deviation ratios of the strength of each edge. The weights can be set according to the importance or stability of the edge under normal operating conditions. The weighted average or maximum value is used as the strength deviation index. This index quantitatively reflects the degree of abnormal amplification or attenuation of the current causal driving strength relative to the normal fluctuation range, indicating the strengthening or weakening of the driving mechanism caused by hidden dangers.

[0147] S244. Calculation of time delay characteristic deviation and evaluation of response dynamics, specifically:

[0148] For the shared directed edges in the current temporal causal graph model and the dynamic causal baseline graph, the deviation of time delay features is calculated to assess the anomalous changes in system response dynamics and the variation in information transmission speed. For each shared directed edge, the current time delay feature value of the edge is extracted from the edge attributes of the current temporal causal graph model. This value records the time delay between the change in the dependent variable and the response of the effect variable. The statistical distribution parameters of the time delay feature corresponding to this edge are extracted from the dynamic causal baseline model, including the mean time delay, standard deviation of time delay, and confidence interval boundaries of the time delay distribution for historical samples.

[0149] The deviation of the current time delay characteristic value from the baseline time delay mean is calculated and normalized according to the baseline time delay standard deviation to obtain a standardized time delay deviation ratio. When the time delay characteristic value is significantly less than the baseline mean minus twice the standard deviation, it indicates an abnormally accelerated causal response, which may indicate that electrical hazards are entering a rapid deterioration stage or that a positive feedback mechanism has been activated. When the time delay characteristic value is significantly greater than the baseline mean plus twice the standard deviation, it indicates an abnormally sluggish response, which may indicate a coupling mechanism failure or a decrease in energy transfer efficiency. The time delay deviation ratios of each edge are weighted according to the causal importance of the edge, such as by weighted summation or taking the weighted root mean square, to obtain a time delay deviation index. This index reflects the degree of abnormal drift of the system response time characteristics relative to the normal state, providing a basis for judging the evolution speed of hazards.

[0150] S245. Multidimensional Deviation Fusion and Causal Structure Deviation Generation, specifically:

[0151] The deviation indicators of three dimensions—topological difference, intensity deviation, and time delay deviation—are weighted and fused to generate the final causal structure deviation, thereby comprehensively quantifying the level of causal network distortion caused by electrical hazards.

[0152] First, set the topology difference weight, intensity deviation weight, and time delay deviation weight. The sum of the three weights is normalized to one. The topology difference weight reflects the severity of network structure distortion, the intensity deviation weight reflects the degree of abnormality in the strength of causal driving, and the time delay deviation weight reflects the degree of variation in response speed.

[0153] The topology difference index, intensity deviation index, and time delay deviation index are each multiplied by their corresponding weights and then linearly summed, or nonlinearly fused using nonlinear fusion functions such as exponential weighting or logistic mapping, to obtain a quantified causal structure deviation value. This causal structure deviation comprehensively characterizes the overall deviation level of the current time-series causal graph model from the dynamic causal baseline graph in terms of topology, causal strength, and time delay characteristics. The larger the value, the more significant the distortion of the causal relationship network caused by electrical hazards, and the further the system dynamic mechanism deviates from the normal state. This causal structure deviation value, as the final output of step S2, is passed to step S3 for threshold comparison and pattern matching with preset multidimensional hazard criteria to determine the specific hazard type and risk level, and trigger the corresponding early warning mechanism.

[0154] S3. Match the causal structure deviation with the multidimensional hazard criteria, determine the hazard type based on the matching results, and output early warning information; specifically including:

[0155] S31. Hazard criterion matching and threshold comparison, specifically:

[0156] S311. Obtain the current causal structure deviation value from the causal structure deviation calculation result output in step S24. Simultaneously, traverse and extract all directed edge sets that pass the significance test and their attached time delay feature values ​​from the time-series causal graph model data structure solidified in step S235. Retrieve the pre-configured hazard criterion library from the system storage medium. This library contains a first criterion for electrochemical migration-dominated hazards and a second criterion for space charge breakdown-dominated hazards.

[0157] The first threshold for the first criterion is determined based on the statistical distribution of the deviation of the causal structure calculated from several sets of time-series causal graph models accumulated under historical no-hazard operation conditions. Specifically, it is the mean of the statistical distribution plus two to three times the standard deviation, or the upper boundary of the 95% confidence interval. The numerical range is usually set between 0.3 and 0.8, which characterizes the critical identification level of the causal relationship network distortion caused by electrochemical migration-dominated hazards.

[0158] The second threshold for the second criterion is independent of the first threshold. It is determined based on the causal structure deviation distribution of historical space charge breakdown-dominated hidden danger samples. It also adopts the method of the mean plus two standard deviations or the boundary of the 95% confidence interval. The numerical range is usually set between 0.5 and 0.9 and is generally higher than the first threshold to reflect the more severe network distortion characteristics of space charge breakdown-dominated hidden dangers.

[0159] S312. Compare the real-time acquired causal structure deviation value with the first threshold and the second threshold respectively. Simultaneously traverse the set of directed edges of the current time-series causal graph model to check whether the direction pattern of the directed edges matches the criteria requirements, and verify whether the time delay feature value falls within the time delay interval specified by the criteria. Perform joint logical matching operation of multi-dimensional conditions, mark the hidden danger criteria that meet all judgment dimensions as active and pass them to subsequent steps.

[0160] S32. Identification of potential hazards dominated by electrochemical migration, specifically:

[0161] When the comparison results in step S31 show that the value of the causal structure deviation exceeds the first threshold preset for electrochemical migration-dominated hidden dangers; at the same time, it is verified whether there is a directed edge in the current time-series causal graph model's set of directed edges pointing from the electrochemical migration concentration field node to the space charge accumulation density field node. The existence of this edge indicates that the electrochemical migration process, as an information source, has a statistically significant driving effect on the space charge accumulation process, rather than random fluctuations or spurious correlations. The directionality of the edge reflects the physical driving direction of charge redistribution caused by ion migration; further, it is verified whether the time delay characteristic value of this directed edge falls into the preset first time delay interval. This first time delay interval is usually set to the range of milliseconds to seconds, corresponding to the physical and dynamic delay of metal ions crossing the contact interface under the action of an electric field and causing the charge response of the insulating surface.

[0162] The system determines that the current building electrical hazard is an electrochemical migration-dominated hazard only when all three of the above conditions are met. This determination indicates that the electrochemical migration and dendrite growth of metal ions at the electrical contact interface under the drive of the electric field is the core failure mechanism leading to system deterioration. It also indicates that the contact resistance will enter a stage of rapid increase while the insulation has not yet broken down, providing maintenance personnel with a basis for targeted tightening or replacement of contact components.

[0163] S33. Determination of space charge breakdown-dominated hidden dangers, specifically:

[0164] When the comparison results in step S31 show that the value of the causal structure deviation exceeds the second threshold preset for the space charge breakdown-dominated hidden danger; at the same time, it is verified whether there is a directed edge in the current time-series causal graph model's directed edge set that points from the space charge accumulation density field node to the electrochemical migration concentration field node. The existence of this reverse directed edge indicates that the strong local electric field formed by space charge accumulation has a significant feedback acceleration driving effect on metal ion migration, forming a positive feedback coupling mechanism, reflecting the physical process of charge accumulation enhancing the electric field and thus accelerating ion migration; further, it is verified whether the time delay characteristic value of this reverse directed edge falls into the preset second time delay interval. This second time delay interval is usually set to the range of microseconds to milliseconds, shorter than the first time delay interval, corresponding to the rapid feedback delay characteristics from charge accumulation to electric field enhancement and then to ion migration acceleration, reflecting the rapid conduction of the electric field enhancement effect and the establishment speed of the positive feedback mechanism.

[0165] The system determines that the current building electrical hazard is a space charge breakdown-dominated hazard only when all three of the above conditions are met. This determination indicates that surface charge accumulation and electric field distortion on the insulation material have replaced ion migration as the dominant failure mechanism, which indicates that surface flashover or breakdown failure is about to occur on the insulation surface, providing maintenance personnel with a basis for replacing insulation materials or improving insulation coatings.

[0166] S34. Remaining safety time estimation and early warning information output, specifically:

[0167] Based on the specific hazard type identification determined in step S32 or S33, and combined with the specific numerical value of the causal structure deviation and the specific value of the time delay characteristic, the remaining safe time required for the building electrical system to evolve from the current hazard state to the critical failure state is calculated using a preset remaining safe time estimation model. This estimation model can employ a lookup table mapping relationship established based on historical statistical data. A larger deviation value indicates that the system is closer to the critical state and corresponds to a shorter remaining time, or a shorter time delay characteristic indicates a faster feedback response and corresponds to a shorter remaining time. The time unit can be hours or days.

[0168] The identified hazard types, such as electrochemical migration-dominated or space charge breakdown-dominated, along with quantified causal structure deviation values, estimated remaining safety time values, and hazard physical spatial coordinates determined by sensor array geometry and signal arrival time difference positioning algorithms, are integrated into a complete early warning information data package.

[0169] The system displays real-time visual warning details through a human-computer interaction interface, including textual descriptions of hazard types, deviation curves, countdown timers, and spatial location markers. It also simultaneously triggers tiered audible and visual alarm devices, such as buzzers and warning lights at different frequencies. Alternatively, it can encrypt and push structured warning data packets to the mobile terminals of operation and maintenance personnel and the central monitoring and management system via the building IoT communication network, enabling early detection, accurate classification, and closed-loop response to hazards.

[0170] Example 2

[0171] like Figure 2 As shown, the building electrical hazard early identification system based on weak signal monitoring is used to implement the building electrical hazard early identification method based on weak signal monitoring, including:

[0172] A. A weak signal collaborative acquisition module is used to deploy a high-sensitivity surface acoustic wave sensor array and a near-field capacitive coupling probe on the surface of the building's electrical load-bearing structure to simultaneously capture the structural phonon flow and near-field evanescent electromagnetic waves, establish a time-frequency domain coupling mapping relationship, and eliminate the diurnal variation period of the geomagnetic background and the baseline drift of the building's thermal cycle, outputting the phonon-evanescent wave coupling characteristic flow; specifically including:

[0173] The phonon flow capture unit is used to mount a piezoelectric surface acoustic wave sensor array at the metal connection interface of the building electrical load-bearing structure. The sensor center frequency is 500kHz-2MHz and the sampling rate is ≥10MHz. It captures the structural phonon flow excited by the geomagnetic induced current flowing through the micro-area of ​​the hidden danger. The structural phonon flow carries the scattering characteristics of lattice thermal vibration at the metal grain boundary.

[0174] The evanescent wave acquisition unit is used to arrange near-field capacitive coupling probes on the surface of insulating materials. The probe spacing is ≤2mm and the sampling rate is ≥5MHz to acquire near-field evanescent electromagnetic waves leaked due to the redistribution of electrical hazards.

[0175] The coupling mapping unit is used to perform short-time Fourier transform on the structured phonon flow and the near-field evanescent electromagnetic wave respectively, calculate the coherence function and mutual information, extract the coherence peak frequency band and mutual information delay characteristics, and establish a time-frequency domain coupling mapping relationship.

[0176] The background removal unit is used to perform empirical mode decomposition on the time spectrum corresponding to the coupling mapping relationship, remove the diurnal variation periodic component of the geomagnetic background and the baseline drift component of the building thermal cycle, and output the phonon-evanescent wave coupled characteristic flow.

[0177] B. The physical field inversion and temporal causality analysis module is used to invert the electrochemical migration concentration field based on the phonon group velocity dispersion characteristics and grain boundary scattering attenuation intensity in the phonon-evanescent wave coupled characteristic flow, and to invert the space charge accumulation density field based on the polarization state rotation angle and phase delay accumulation of near-field evanescent electromagnetic waves. It constructs a temporal causal graph model with the electrochemical migration concentration field and the space charge accumulation density field as nodes and the directed causal correlation strength calculated based on the transfer entropy algorithm as directed edges. The module compares the current temporal causal graph model with the dynamic causal baseline graph to calculate the causal structure deviation. Specifically, it includes:

[0178] The electrochemical migration concentration field inversion unit is used to extract the phonon group velocity dispersion curve and grain boundary scattering attenuation coefficient from the phonon-evanescent wave coupled characteristic flow. Based on the anisotropic medium propagation theory, the constitutive relation is established, and the concentration distribution of metal ions enriched at the grain boundary is inverted through least squares fitting as the electrochemical migration concentration field.

[0179] The space charge accumulation density field inversion unit is used to extract the polarization state rotation angle and phase delay accumulation from near-field evanescent electromagnetic waves. The functional relationship is established using the propagation model of evanescent waves in dielectric constant layered media, and the space charge accumulation density field is obtained by solving the problem.

[0180] The temporal causal graph construction unit is used to calculate the bidirectional transfer entropy value with the spatiotemporal sequence of the electrochemical migration concentration field and the space charge accumulation density field as nodes. The transfer entropy value that is significantly greater than zero is used as the causal association strength of the directed edge and the time delay characteristics are recorded to construct the temporal causal graph model.

[0181] The deviation calculation unit is used to compare the current temporal causal graph model with the dynamic causal baseline graph, calculate the topological difference of the causal edge set, the magnitude deviation of the causal intensity, and the time delay deviation of the time delay feature, and use the weighted sum of the three as the causal structure deviation.

[0182] C. Hazard identification and early warning module, used to match the causal structure deviation with preset multi-dimensional hazard criteria. When the causal structure deviation meets any hazard criterion, the hazard type is determined and early warning information is output; specifically including:

[0183] The criterion matching unit is used to obtain the deviation degree of causal structure and the time delay characteristics of directed edges in the temporal causal graph model, and compares the deviation degree of causal structure with multiple preset hidden danger criteria respectively.

[0184] The hazard type determination unit is used to determine the hazard as an electrochemical migration-dominated hazard when the deviation of the causal structure exceeds the first threshold and there is a directed edge from the electrochemical migration concentration field to the space charge accumulation density field and the time delay feature falls within the first time delay interval; and to determine the hazard as a space charge breakdown-dominated hazard when the deviation of the causal structure exceeds the second threshold and there is a reverse directed edge and the time delay feature falls within the second time delay interval.

[0185] The early warning output unit is used to estimate the remaining safe time based on the determined hazard type, combined with the causal structure deviation and time delay characteristics, and output early warning information including hazard type, causal structure deviation, and remaining safe time.

[0186] As can be seen from the above description, the embodiments of the present invention achieve the following technical effects:

[0187] This invention overcomes the limitations of existing technologies that rely on active electrical detection or strong signal threshold monitoring. It utilizes the geomagnetic induced current generated in the steel structure of a building due to changes in the Earth's magnetic field as a natural passive excitation source. By monitoring the structural phonon flow excited by this weak current flowing through the micro-region of potential hazards and the near-field evanescent electromagnetic waves leaking from the insulation surface, a phonon-evanescent wave coupling characteristic flow across the solid-gas interface is established. This mechanism eliminates the need to inject detection signals into the power distribution system or install electrical contact sensors, achieving truly non-invasive and zero-interference monitoring. Simultaneously, by eliminating the geomagnetic diurnal variation period and the building's thermal cycle baseline, it effectively penetrates complex electromagnetic environment noise, capturing hazard characteristics at the nascent stage when contact resistance is still at the micro-ohmic level and the insulation surface has not yet shown a detectable temperature rise. This advances the identification window to the early stages of physical evolution, which are imperceptible by traditional methods.

[0188] Unlike existing technologies that only monitor the absolute value of physical field strength or simple correlations, this invention introduces a transfer entropy algorithm to construct a time-series causal graph model between the electrochemical migration concentration field and the space charge accumulation density field, quantifying the asymmetric causal driving relationship and information transmission delay between the two major microscopic failure processes. By establishing a dynamic causal baseline model based on statistical learning of historical no-hazard states, and using the causal structure deviation index to evaluate the degree of distortion of the current causal relationship network relative to normal operating conditions in three dimensions: topology, driving strength, and response delay, this invention achieves a paradigm shift from detecting physical quantity threshold exceedances to detecting causal mechanism topological anomalies. It can identify early hazards where the physical field strength has not yet exceeded the limit but the causal feedback loop has undergone qualitative changes, significantly reducing the false alarm rate caused by load fluctuations and seasonal temperature differences.

[0189] By setting multi-dimensional hazard criteria including causal structure deviation thresholds, directed edge direction patterns, and time delay characteristic intervals, this invention can accurately distinguish between two fundamentally different hazard evolution paths: electrochemical migration-dominated and space charge breakdown-dominated. The former corresponds to a millisecond-delayed causal link where ion migration drives charge accumulation, indicating a risk of rapidly increasing contact resistance; the latter corresponds to a short-delayed positive feedback loop where charge accumulation feedback accelerates ion migration, indicating a risk of surface flashover in insulation. Based on this, the system outputs targeted remaining safety time estimates and physical coordinate locations, solving the maintenance blindness caused by traditional single alarm information, and achieving a leap from anomaly detection to diagnosing the cause and predicting remaining life, providing maintenance personnel with precise decision-making basis for tightening contacts or replacing insulation materials.

[0190] The embodiments and / or implementation methods described above are merely preferred embodiments and / or implementation methods for implementing the technology of the present invention, and are not intended to limit the implementation methods of the technology of the present invention in any way. Any person skilled in the art may make some modifications or alterations to other equivalent embodiments without departing from the scope of the technical means disclosed in the present invention, but these should still be regarded as the technology or embodiments that are substantially the same as the present invention.

[0191] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are only preferred embodiments of this application. It should be noted that due to the limitations of written expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A method for early identification of electrical hazards in buildings based on weak signal monitoring, characterized in that, include: The structural phonon flow and near-field evanescent electromagnetic waves excited by potential electrical hazards in buildings are captured, and after time-frequency domain coupling and background removal, the phonon-evanescent wave coupled characteristic flow is output. Based on the phonon-evanescent wave coupled characteristic flow inversion electrochemical migration concentration field and space charge accumulation density field, a time-series causal graph model between the two is constructed, and the causal structure deviation between the current time-series causal graph model and the dynamic causal baseline graph is calculated. The deviation of the causal structure is matched with the multidimensional hazard criterion, and the hazard type is determined based on the matching result, and early warning information is output.

2. The method for early identification of building electrical hazards based on weak signal monitoring according to claim 1, characterized in that, The structure phonon flow and near-field evanescent electromagnetic waves excited by the capture of building electrical hazards are coupled in the time and frequency domains and background removal to output a phonon-evanescent wave coupled characteristic flow, specifically including: A piezoelectric surface acoustic wave sensor array is attached to the metal connection interface of the building's electrical load-bearing structure to capture the structural phonon flow excited by the geomagnetic induced current through the micro-region of the hidden danger. Near-field capacitive coupling probes are simultaneously deployed on the surface of the insulating material to collect near-field evanescent electromagnetic waves leaking from the surface of the insulating material due to the redistribution of charges caused by electrical hazards. The structured phonon flow and the near-field evanescent electromagnetic wave are subjected to short-time Fourier transforms respectively. The coherence function and mutual information of the two within the same time window are calculated. The coherence peak frequency band and mutual information delay characteristics are extracted to establish a coupling mapping relationship in the time and frequency domain. Empirical mode decomposition is performed on the time spectrum corresponding to the coupling mapping relationship to extract and remove the diurnal variation periodic component of the geomagnetic background and the baseline drift component of the building thermal cycle, and output the phonon-evanescent wave coupled characteristic flow.

3. The method for early identification of building electrical hazards based on weak signal monitoring according to claim 2, characterized in that, The method of attaching a piezoelectric surface acoustic wave sensor array at the metal connection interface of the building's electrical load-bearing structure to capture the structural phonon flow excited by the geomagnetic induced current through the micro-region of the hidden danger specifically includes: At the metal connection interface of the building's electrical load-bearing structure, piezoelectric surface acoustic wave sensor arrays are arranged at equal intervals along the main current path, and the array coverage covers areas prone to electrical hazards. Using geomagnetic induced current as a natural excitation source, when the geomagnetic induced current flows through the micro-region of hidden danger, the abnormal Joule heating effect excites lattice thermal vibration at the metal grain boundary, and the structure phonon flow propagating along the metal surface is captured by the piezoelectric surface acoustic wave sensor array. Scattering features generated at metal grain boundaries by lattice thermal vibrations are extracted from the captured structural phonon stream. These scattering features include phonon group velocity dispersion curves, grain boundary scattering attenuation coefficients, and time delay broadening of phonon wave packets.

4. The method for early identification of building electrical hazards based on weak signal monitoring according to claim 3, characterized in that, Empirical mode decomposition is performed on the time spectrum corresponding to the coupling mapping relationship to extract and remove the diurnal variation periodic component of the geomagnetic background and the baseline drift component of the building thermal cycle, and the phonon-evanescent wave coupled characteristic flow is output, specifically including: The time spectrum formed by the time-frequency domain coupling mapping of the structured phonon flow and the near-field evanescent electromagnetic wave is obtained. Empirical mode decomposition is performed on the time spectrum to decompose it into multiple intrinsic mode function components, each of which corresponds to different time scale characteristics. The components with a period of 24 hours ± 2 hours among the multiple intrinsic mode function components are identified and treated as diurnal variation periodic components of the geomagnetic background, and then removed from the time spectrum. Slow-changing trend terms with a variation period greater than 12 hours are identified among the multiple intrinsic mode function components and are used as the building thermal cycle baseline drift component and removed from the time spectrum; The remaining intrinsic mode function components after removing the diurnal variation periodic component of the geomagnetic background and the baseline drift component of the building thermal cycle are reconstructed to obtain the net time spectrum reflecting the energy dissipation state of the micro-area with hidden dangers, which is used as the output of the phonon-evanescent wave coupling characteristic flow.

5. The method for early identification of building electrical hazards based on weak signal monitoring according to claim 4, characterized in that, Based on the phonon-evanescent wave coupled characteristic flow inversion electrochemical migration concentration field and space charge accumulation density field, a time-series causal graph model between the two is constructed. The causal structure deviation between the current time-series causal graph model and the dynamic causal baseline graph is calculated, specifically including: Phonon group velocity dispersion curves and grain boundary scattering attenuation coefficients are extracted from the phonon-evanescent wave coupled characteristic flow. Constitutive relations are established based on the propagation theory of anisotropic media. The concentration distribution of metal ions enriched at the grain boundary is inverted by least squares fitting and used as the electrochemical migration concentration field. The polarization state rotation angle and phase delay accumulation are extracted from the near-field evanescent electromagnetic wave. The propagation model of the evanescent wave in the dielectric constant layered medium is used to establish the functional relationship between the polarization state rotation angle, phase delay and space charge density. The space charge accumulation density field is obtained by substituting the measured values ​​into the model. Using the spatiotemporal sequence of the electrochemical migration concentration field and the space charge accumulation density field as nodes, the bidirectional transfer entropy value is calculated. The transfer entropy value that is significantly greater than zero is taken as the causal correlation strength of the directed edge and the time delay characteristics are recorded to construct a time-series causal graph model. The current temporal causal graph model is compared with the dynamic causal baseline graph. The topological difference of the causal edge set, the magnitude deviation of the causal intensity, and the time delay deviation of the time delay feature are calculated. The weighted sum of the three is taken as the causal structure deviation.

6. The method for early identification of building electrical hazards based on weak signal monitoring according to claim 5, characterized in that, Phonon group velocity dispersion curves and grain boundary scattering attenuation coefficients are extracted from the phonon-evanescent wave coupled characteristic flow. Constitutive relations are established based on the anisotropic medium propagation theory. The concentration distribution of metal ions enriched at the grain boundaries is inverted through least-squares fitting and used as the electrochemical migration concentration field. Specifically, this includes: Phonon group velocity dispersion curves and grain boundary scattering attenuation coefficients are extracted from the phonon-evanescent wave coupling characteristic flow. Based on the theory of sound wave propagation in anisotropic media, a first constitutive relationship is established between phonon group velocity and lattice thermal conductivity and grain boundary density, and a second constitutive relationship is established between grain boundary scattering attenuation coefficient and metal ion enrichment concentration at grain boundaries. Substituting the phonon group velocity dispersion curve and grain boundary scattering attenuation coefficient into the first and second constitutive relations, the spatial distribution of metal ion enrichment concentration at the grain boundary is inverted using the least squares fitting algorithm. The inverted spatial distribution of metal ion enrichment concentration is then used as the output of the electrochemical migration concentration field.

7. The method for early identification of building electrical hazards based on weak signal monitoring according to claim 6, characterized in that, Using the spatiotemporal sequences of the electrochemical migration concentration field and the space charge accumulation density field as nodes, the bidirectional transfer entropy value is calculated. The transfer entropy value significantly greater than zero is taken as the causal correlation strength of the directed edge, and the time delay characteristics are recorded to construct a time-series causal graph model, specifically including: The electrochemical migration concentration field obtained by inversion is organized into a spatiotemporal sequence of electrochemical migration concentration field according to time order. At the same time, the space charge accumulation density field obtained by inversion is organized into a spatiotemporal sequence of space charge accumulation density field according to the same time axis. The time resolution of the two spatiotemporal sequences is consistent. Using the spatiotemporal sequence of the electrochemical migration concentration field as the source node sequence and the spatiotemporal sequence of the space charge accumulation density field as the target node sequence, the transfer entropy value from the electrochemical migration concentration field to the space charge accumulation density field is calculated. Using the spatiotemporal sequence of the space charge accumulation density field as the source node sequence and the spatiotemporal sequence of the electrochemical migration concentration field as the target node sequence, calculate the reverse transfer entropy value from the space charge accumulation density field to the electrochemical migration concentration field; The propagation entropy value and the reverse propagation entropy value are compared with a preset significance threshold. The propagation entropy value that is greater than the significance threshold is taken as the causal association strength of the directed edge, and the time delay offset when the propagation entropy value is reached is recorded as the time delay feature. Using the electrochemical migration concentration field and the space charge accumulation density field as nodes, the causal correlation strength of the directed edges as edge weights, and the time delay characteristics as edge attributes, a time-series causal graph model is constructed.

8. The method for early identification of building electrical hazards based on weak signal monitoring according to claim 7, characterized in that, The current temporal causal graph model is compared with the dynamic causal baseline graph. The topological differences of the causal edge set, the magnitude deviation of causal strength, and the delay deviation of delay characteristics are calculated. The weighted sum of these three factors is taken as the causal structure deviation, specifically including: The dynamic causal baseline map is retrieved from the dynamic causal baseline model. The dynamic causal baseline map is statistically generated based on multiple sets of time-series causal graph models under historical no-hazard conditions. It includes the set of directed edges of the baseline, the causal intensity distribution range of each directed edge of the baseline, and the time delay characteristic distribution range of each directed edge of the baseline. Calculate the topological difference between the current temporal causal graph model and the dynamic causal baseline graph. The topological difference includes the number of new edges, missing edges, and reverse edges in the current directed edge set relative to the baseline directed edge set. Calculate the strength deviation between the current temporal causal graph model and the dynamic causal baseline graph. For the directed edges in the current temporal causal graph model that are shared with the set of directed edges of the baseline, calculate the deviation ratio of the current causal association strength relative to the distribution range of the baseline causal strength. Calculate the time delay deviation between the current temporal causal graph model and the dynamic causal baseline graph. For the directed edges in the current temporal causal graph model that are shared with the set of directed edges of the baseline, calculate the deviation ratio of the current time delay feature from the distribution range of the baseline time delay feature. The topological difference, intensity deviation, and time delay deviation are weighted and summed according to preset weights to obtain the causal structure deviation.

9. The method for early identification of building electrical hazards based on weak signal monitoring according to claim 8, characterized in that, The causal structure deviation is matched with the multidimensional hazard criterion. Based on the matching result, the hazard type is determined and early warning information is output, specifically including: Obtain the deviation of the causal structure and the time delay characteristics of the directed edges in the time-series causal graph model, and compare the deviation of the causal structure with a number of preset hidden danger criteria respectively; When the deviation of the causal structure exceeds the first threshold and there is a directed edge in the time-series causal graph model pointing from the electrochemical migration concentration field to the space charge accumulation density field, and the time delay feature of the directed edge falls into the first time delay interval, the hidden danger type is determined to be an electrochemical migration-dominated hidden danger. When the deviation of the causal structure exceeds the second threshold and there is a directed edge in the time-series causal graph model pointing from the space charge accumulation density field to the electrochemical migration concentration field, and the time delay feature of the directed edge falls into the second time delay interval, the hidden danger type is determined to be a space charge breakdown-dominated hidden danger. Based on the identified hazard type, combined with the numerical value and time delay characteristics of the causal structure deviation, the remaining safe time is estimated, and early warning information is output.

10. A building electrical hazard early identification system based on weak signal monitoring, characterized in that, The method for early identification of building electrical hazards based on weak signal monitoring as described in any one of claims 1-9 includes: The weak signal collaborative acquisition module is used to deploy a high-sensitivity surface acoustic wave sensor array and a near-field capacitive coupling probe on the surface of the building's electrical load-bearing structure to simultaneously capture the structural phonon flow and near-field evanescent electromagnetic waves, establish a time-frequency domain coupling mapping relationship, and eliminate the diurnal variation period of the geomagnetic background and the baseline drift of the building's thermal cycle, and output the phonon-evanescent wave coupling characteristic flow. The physical field inversion and temporal causality analysis module is used to invert the electrochemical migration concentration field based on the phonon group velocity dispersion characteristics and grain boundary scattering attenuation intensity in the phonon-evanescent wave coupled characteristic flow, and to invert the space charge accumulation density field based on the polarization state rotation angle and phase delay accumulation of the near-field evanescent electromagnetic wave. It constructs a temporal causal graph model with the electrochemical migration concentration field and the space charge accumulation density field as nodes and the directed causal correlation strength calculated based on the transfer entropy algorithm as directed edges. The module compares the current temporal causal graph model with the dynamic causal baseline graph and calculates the causal structure deviation. The hazard identification and early warning module is used to match the causal structure deviation with preset multidimensional hazard criteria. When the causal structure deviation meets any hazard criterion, the hazard type is determined and early warning information is output.

Images