Rapid explosion suppression and flame prevention method for transformer

Abstract

The invention discloses a rapid explosion suppression and flame prevention method for a transformer, and belongs to the technical field of power equipment safety, and the method specifically comprises the steps: S1, collecting multi-physical field data in real time through a sensor network disposed in the transformer; s2, based on the multi-physical field data, determining an electrothermal instability index, a fractal dimension of fire nucleus growth, a maximum characteristic value of an irradiance Hessian matrix and a spectral coherence length attenuation rate; s3, combining the electrothermal instability index, the fractal dimension of fire core growth, the maximum characteristic value and the spectral coherence length attenuation rate to construct a characteristic vector; and S4, identifying the initial fault state based on the feature vector, correcting the initial fault state by using a preset physical model to obtain a final fault state, and triggering a grading response strategy according to the final fault state. And a solid data basis is provided for subsequent accurate diagnosis.

Description

Technical Field

[0001] This invention relates to the field of power equipment safety, specifically to a rapid explosion suppression and fire prevention method for transformers. Background Technology

[0002] The insulating oil medium inside power transformers deteriorates under long-term electrical and thermal stress. If localized overheating or discharge occurring under fault conditions is not detected in time, it can escalate into a serious accident, leading to combustion or even explosion. Existing detection technologies have limitations in dealing with the complex internal environment of transformers. The propagation of optical signals in the oil medium is affected by the coupling of the spectral, medium, and temperature fields, resulting in nonlinear distortion. In the critical stage before combustion, the weak photon radiation signal is severely masked by environmental noise. Multiple potential fault hotspots may coexist internally, evolving competitively through heat migration and electromagnetic field coupling, leading to uncertainty in predicting the actual ignition point. Therefore, a method is urgently needed that can penetrate complex environmental noise, accurately identify early fault characteristics, and predict their evolution trend.

[0003] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] The purpose of this invention is to provide a rapid explosion suppression and fire prevention method for transformers, so as to solve the problems mentioned in the background art.

[0005] The technical solution of the present invention includes the following specific steps: S1. Real-time acquisition of multi-physics data through a sensor network deployed inside the transformer; S2. Based on multiphysics data, determine the electrothermal instability index, the fractal dimension of fire nucleus growth, the maximum eigenvalue of the Hessian matrix of irradiance, and the spectral coherence length attenuation rate. S3. Construct an eigenvector by combining the electrothermal instability index, the fractal dimension of fire nucleus growth, the maximum eigenvalue, and the spectral coherence length attenuation rate; S4. Identify the initial fault state based on feature vectors, correct the initial fault state using a preset physical model to obtain the final fault state, and trigger a graded response strategy based on the final fault state.

[0006] Preferably, the multiphysics data includes at least multi-band optical spectral signals, electric field intensity signals, temperature distribution signals, and ultrasonic signals.

[0007] Preferably, in step S2, the determination of the electrothermal instability index, the fractal dimension of the fire nucleus growth, the maximum eigenvalue of the Hessian matrix of irradiance, and the spectral coherence length attenuation rate includes: The ratio of the relative rate of change of conductivity over time to the relative rate of change of temperature over time is calculated to obtain the electrothermal instability index. Based on temperature and light intensity data, the fractal dimension of fire nucleus growth is calculated using the box counting method. Calculate the Hessian matrix of the photon flux density scalar field and extract the largest eigenvalue of the Hessian matrix; The time series of spectral coherence lengths is numerically differentiated to calculate the spectral coherence length attenuation rate.

[0008] Preferably, the calculation of the fractal dimension of the fire core growth includes: Potential hotspots were identified using an adaptive thresholding method based on temperature and light intensity data. Define a series of observation scales and count the number of hotspots within each observation scale; In a logarithmic coordinate system, linear regression analysis is performed on data points with varying observation scales and hotspot numbers, and the slope of the fitted line from the linear regression analysis is used as the fractal dimension.

[0009] Preferably, the identification of the preliminary fault state includes: When the maximum eigenvalue exceeds the first preset threshold, the initial fault state is determined to be either a hotspot fixed state or a hotspot wandering state based on the time series stability of the peak position of the maximum eigenvalue; otherwise, it is determined to be normal. When the electrothermal instability index exceeds the second preset threshold, the fractal dimension is within the preset range, and the spectral coherence length attenuation rate exceeds the third preset threshold, the initial fault state is determined to be the pre-combustion glass state. A dynamic hotspot network is constructed and the degree distribution of the network is calculated. When the power-law fitting exponent of the degree distribution is within the preset range of 2 to 3, the initial fault state is determined to be a self-organized critical state.

[0010] Preferably, the correction of the initial fault state includes: Perform hysteresis loss asymmetry correction and use the hysteresis loop area space gradient map provided by the pre-stored transformer core finite element model to calibrate the weights on the optical detection path; Perform dissolved gas spectral absorption correction by subtracting faulty gas absorption peaks from the preset chemical database from the real-time spectral data; Plasma resonance pseudo-signal filtering is performed by identifying and filtering out signal peaks in the spectrum that conform to the model characteristics based on a preset surface plasmonic resonance frequency model. The demodulation of the acousto-optic coupling modulation is performed by identifying the sidebands of a specific frequency shift through a fast Fourier transform of the optical signal and applying a digital inverse filter.

[0011] Preferably, the hierarchical response strategy includes: When the final fault state is determined to be entering the pre-combustion glassy state or the self-organized critical state, it is determined to be a level three alarm state, and the rapid explosion suppression and fire prevention device that executes power cut-off or releases high-pressure inert gas is immediately triggered. When the final fault state does not meet the conditions for a level 3 alarm state, but meets the preset level 2 warning conditions, it is determined to be a level 2 warning state, and auxiliary cooling is automatically started or a command to reduce the load is issued. When the final fault state does not meet the conditions for a Level 3 alarm state or a Level 2 early warning state, it is determined to be a Level 1 early warning state, and a maintenance recommendation is issued.

[0012] This invention provides an improved method for rapid explosion suppression and fire prevention of transformers, which has the following improvements and advantages compared with the prior art: 1. This invention uses a sensor network deployed inside a transformer to collect multi-physics field data in real time, including at least multi-band optical spectral signals, electric field intensity signals, temperature distribution signals, and ultrasonic signals. This organic integration of optical, electrical, thermal, and acoustic signals ensures the richness and complementarity of the input information, solves the problem that a single signal source is easily interfered with and fails in complex electromagnetic and thermal environments, and provides a solid data foundation for subsequent accurate diagnosis. 2. This invention abandons the simple threshold judgment of the original signal; it achieves in-depth quantification of the fault state by determining four core parameters: electrothermal instability index, fractal dimension of fire nucleus growth, maximum eigenvalue of the Hessian matrix of irradiance, and spectral coherence length attenuation rate, and constructing them as feature vectors; these parameters, from four mutually orthogonal physical dimensions of electrothermal coupling stability, spatial morphological complexity, energy concentration degree, and signal transmission quality, profoundly reveal the intrinsic mechanism of fault evolution, can penetrate strong noise background, and accurately capture early weak fault signs that cannot be identified by traditional methods; 3. This invention establishes a cascaded diagnostic logic; it can not only identify early anomalies such as hot spot fixation and hot spot wandering, but also accurately determine the high-risk state of pre-combustion glass state through composite judgment, and identify the self-organized critical state that foreshadows cascading failure by constructing a dynamic hot spot network; this fine division of fault stages transcends the traditional binary theory of normal / fault, providing accurate decision input for graded disposal; more importantly, this method introduces a preset physical model to correct the preliminary diagnostic results. By performing hysteresis loss asymmetry correction, dissolved gas spectral absorption correction, plasma resonance pseudo-signal filtering, and acoustic-optical coupling modulation demodulation, it systematically removes interference from multiple physical processes such as magnetic, chemical, plasma, and acoustic processes, ensuring that the final fault state judgment is based on a true reflection of the fault itself, greatly reducing the risk of false alarms and missed alarms; 4. This invention triggers a corresponding graded response strategy based on the severity level of the final fault state. For extreme dangers such as pre-combustion glassy state or self-organized critical state, the system immediately performs forced interventions such as power cut-off or release of high-pressure inert gas. For developing hidden dangers, it automatically starts auxiliary cooling or issues load reduction commands for active suppression. For early deviations, it issues maintenance recommendations for early warning. This automated response system, which is precisely matched with risk assessment, achieves the best balance between safety and economic benefits, enabling instantaneous and powerful intervention before a disaster occurs while minimizing unnecessary downtime losses. Attached Figure Description

[0013] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of a rapid explosion suppression and fire prevention method for transformers according to the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0015] Example 1 Please see Figure 1 This invention provides a rapid explosion suppression and fire prevention method for transformers, the specific steps of which include: S1. Real-time acquisition of multi-physics data through a sensor network deployed inside the transformer; S2. Based on multiphysics data, determine the electrothermal instability index, the fractal dimension of fire nucleus growth, the maximum eigenvalue of the Hessian matrix of irradiance, and the spectral coherence length attenuation rate. S3. Construct an eigenvector by combining the electrothermal instability index, the fractal dimension of fire nucleus growth, the maximum eigenvalue, and the spectral coherence length attenuation rate; S4. Identify the initial fault state based on feature vectors, correct the initial fault state using a preset physical model to obtain the final fault state, and trigger a graded response strategy based on the final fault state. This embodiment provides a method for rapid explosion suppression and fire prevention of transformers; the overall technical process of the method is designed as a closed-loop sensing, analysis and response system, which aims to achieve accurate early warning and proactive intervention for potential internal faults of transformers; The specific steps of the method are implemented; data is acquired through a sensor network deployed inside the transformer; the sensor network refers to a spatially distributed data acquisition system composed of various types of sensors, with the aim of capturing multi-dimensional physical information of the transformer's internal operating status in real time without blind spots; in this embodiment, the network is configured at key locations inside the transformer tank to ensure continuous monitoring of the core area; For a three-phase oil-immersed power transformer, the sensor network can be deployed as follows: Three monitoring points (upper, middle, and lower) are set along the axial direction near the geometric center of the high, medium, and low voltage coils of each of the three phases A, B, and C, for a total of nine core monitoring points. Each monitoring point integrates a composite sensor probe, which includes a fiber Bragg grating sensor for temperature measurement, a miniature electric field probe for electric field strength measurement, and an ultraviolet-sensitive photodiode for detecting ultraviolet photon flux. In addition, a broadband piezoelectric ultrasonic sensor is installed at the center of each of the four inner surfaces of the tank wall to receive global acoustic signals. This three-dimensional layout of nine core points plus four auxiliary points enables comprehensive monitoring of the winding heating and discharge core areas and the overall state of the tank. Based on the collected multiphysics field data, a set of key characteristic parameters were determined. The aim of this step is to transform the raw, high-dimensional sensor data stream into low-dimensional, high-information-density indicators that can profoundly characterize the physical nature of the fault. To achieve this goal, four core parameters were determined: the electrothermal instability index, the fractal dimension of the fire nucleus growth, the maximum eigenvalue of the Hessian matrix of irradiance, and the spectral coherence length attenuation rate. These parameters, from four mutually orthogonal dimensions of electrothermal stability, spatial morphological complexity, energy concentration, and signal transmission quality, jointly depict the evolution of the fault. The determined electrothermal instability index, fractal dimension of fire nucleus growth, maximum eigenvalue, and spectral coherence length attenuation rate are used to construct a feature vector. The feature vector is an ordered set containing all four core parameters calculated in the previous steps. Its function is to provide a standardized, multi-dimensional input for subsequent pattern recognition algorithms. By integrating parameters with different physical meanings, the feature vector can reflect the overall state of the system more comprehensively and robustly than any single parameter. A cascaded fault diagnosis and response process is executed based on feature vectors. The internal logic of the process is as follows: the feature vectors are analyzed through a set of preset logical rules to identify the initial fault state; in order to eliminate the influence of known physical interference factors on the diagnosis results, the initial fault state is corrected using a preset physical model to obtain a more accurate and reliable final fault state; and based on the final fault state, a graded response strategy is triggered to execute corresponding safety measures according to the diagnosed risk level. This embodiment solves the problems of weak signals, strong noise interference, and uncertain fault evolution under complex operating conditions that are difficult to handle by traditional methods by constructing a complete technical link from multi-dimensional data acquisition, multi-angle feature extraction, intelligent status recognition to automated hierarchical response. The technical effect is that it can penetrate complex environmental noise, accurately identify early fault characteristics and predict their evolution trend, so as to effectively intervene before the fault evolves into a serious accident, which greatly improves the operational safety of power transformers.

[0016] Multiphysics data includes at least multi-band optical spectral signals, electric field intensity signals, temperature distribution signals, and ultrasonic signals; This embodiment defines the specific composition of multiphysics data; the purpose of multiphysics data is to capture early warning information of faults from multiple physical dimensions in order to construct a comprehensive state view; in this embodiment, the data is required to include at least multi-band optical spectral signals, electric field intensity signals, temperature distribution signals, and ultrasonic signals. The multi-band optical spectral signal is used to monitor photon radiation in the insulating medium caused by partial discharge or overheating, which serves as a precursor to combustion; the electric field intensity signal is used to directly monitor abnormal fluctuations in the electric field inside the transformer, indicating potential insulation breakdown risks; the temperature distribution signal is used to depict the spatial distribution of the internal thermal field and identify local hot spots; and the ultrasonic signal is used to detect sound waves generated by partial discharge or mechanical vibration, serving as further evidence of the fault. The technical advantage of this embodiment is that by explicitly specifying these four key data sources, the comprehensiveness and complementarity of the input information are ensured; the fusion of optical, electrical, thermal and acoustic signals enables subsequent feature extraction and state recognition to be based on a richer and more reliable information foundation, effectively avoiding the problem that a single signal source may fail due to specific interference, thereby significantly improving the diagnostic robustness and accuracy of the entire method under complex working conditions.

[0017] Example 2 In S2, the determination of the electrothermal instability index, the fractal dimension of fire nucleus growth, the maximum eigenvalue of the irradiance Hessian matrix, and the spectral coherence length attenuation rate includes: The ratio of the relative rate of change of conductivity over time to the relative rate of change of temperature over time is calculated to obtain the electrothermal instability index. Based on temperature and light intensity data, the fractal dimension of fire nucleus growth is calculated using the box counting method. Calculate the Hessian matrix of the photon flux density scalar field and extract the largest eigenvalue of the Hessian matrix; The time series of spectral coherence lengths is numerically differentiated to calculate the spectral coherence length attenuation rate. The calculation of the fractal dimension of fire core growth includes: Potential hotspots were identified using an adaptive thresholding method based on temperature and light intensity data. Define a series of observation scales and count the number of hotspots within each observation scale; In a logarithmic coordinate system, linear regression analysis was performed on the data points of the observation scale and the number of hotspots, and the slope of the fitted line of the linear regression analysis was used as the fractal dimension. Based on Example 1, this embodiment elaborates on the method for determining the four core parameters in step S2, ensuring the feasibility of the technical solution; The purpose of determining the electrothermal instability index is to quantify the risk of thermally induced insulation failure in insulating oil from the perspective of electrothermal coupling; the calculation model is defined as follows: in, The electrothermal instability index; Real-time conductivity measured by a conductivity sensor; The real-time absolute temperature is measured by a temperature sensor. and These are the rates of change of electrical conductivity and temperature over time, respectively. By numerically differentiating their respective time series data, their rate of change over time can be calculated, and thus the dimensionless exponent can be obtained. A rapid increase in the index is considered a key precursor to thermal runaway. The determination of the fractal dimension of fire core growth aims to distinguish between a truly continuously developing fire core and a randomly isolated, interfering hotspot from the perspective of spatial complexity. The calculation process follows a standard procedure, during which potential hotspots are identified. Potential hotspots refer to high-signal regions marked in temperature and light intensity data using an adaptive thresholding method. The adaptive thresholding method is an algorithm where the threshold is dynamically adjusted based on the statistical characteristics of the local background signal, effectively suppressing global noise and accurately segmenting meaningful outliers. For example, the method can employ a bimodal method or the Ozin method based on the local signal mean and standard deviation to ensure robustness of segmentation under different background noise levels. A series of observation scales covering millimeters to decimeters are defined. And count the number of hotspots included in each scale. In a logarithmic coordinate system, for data points Perform linear regression analysis; the slope of the fitted line obtained from the analysis is defined as the fractal dimension. : in, It is the fractal dimension; Number of hotspots; The observation scale; Δ: the difference symbol, representing the amount of change; : Logarithmic symbol; Stable within a specific range The value indicates that the fire core is expanding in a way that has a complex structure and statistical self-similarity, rather than random noise; The purpose of determining the largest eigenvalue of the Hessian matrix is ​​to quantitatively describe the spatial concentration of photon radiation energy around the fault hotspot; and to construct an irradiance scalar field. Hessian matrix ,in Real-time measurement is performed using an ultraviolet-sensitive photodiode array; the Hessian matrix is ​​a second-order tensor, with components derived from irradiance. Spatial coordinates The second-order partial derivatives constitute the basis for accurately characterizing... The local curvature of the field in space; to compare the tensor with the scalar threshold, the largest eigenvalue of the matrix needs to be extracted. ; Physically, it represents the curvature of the direction in which irradiance changes most drastically in space, and is the most sensitive indicator of hotspot intensity and energy concentration. The purpose of determining the relative attenuation rate of spectral coherence length is to evaluate the transmission quality of optical signals in oil media in real time. Deterioration is an important indicator of media degradation and the formation of fault environments; its calculation model is defined as: in, The attenuation rate; It is the spectral coherence length; For time; Differential symbol; Spectral coherence length It is a physical quantity calculated by performing Michelson interferometry on the acquired spectral signal; by... The decay rate can be obtained by numerically differentiating the time series data. ; The technical advantage of this embodiment is that it provides four parameters with clear physical meaning and calculation methods for the construction of feature vectors. The parameters quantify the fault state from four dimensions: thermoelectric stability, spatial morphology, energy accumulation and channel quality. This makes the subsequent fault identification no longer based on a simple threshold judgment of the original signal, but based on a precise characterization of the deep physical mechanism of the fault, thereby greatly improving the accuracy and reliability of diagnosis. To increase the robustness of the model, a protective threshold is set during computation. Only when the relative rate of change of temperature over time Only then is it calculated. Index; if If the system is in a thermally stable state, then it is considered to be in a thermally stable state. Set it to a safe default value, such as 0; this can effectively avoid meaningless divergence of the exponent due to stable temperature and an excessively small denominator, and prevent false alarms from the system.

[0018] The initial identification of fault conditions includes: When the maximum eigenvalue exceeds the first preset threshold, the initial fault state is determined to be either a hotspot fixed state or a hotspot wandering state based on the time series stability of the peak position of the maximum eigenvalue; otherwise, it is determined to be normal. When the electrothermal instability index exceeds the second preset threshold, the fractal dimension is within the preset range, and the spectral coherence length attenuation rate exceeds the third preset threshold, the initial fault state is determined to be the pre-combustion glass state. A dynamic hotspot network is constructed and the degree distribution of the network is calculated. When the power-law fitting exponent of the degree distribution is within the preset range of 2 to 3, the preliminary fault state is determined to be a self-organized critical state. In this embodiment, the specific logical rules for identifying the initial fault state are described in detail; the identification process is designed as a set of cascaded judgment logic to distinguish different stages of fault evolution. The identification of hotspot bistable phenomena is performed; the first preset threshold is set. The largest eigenvalue used to determine irradiance Whether it is abnormal; the determination of the preset threshold is based on the statistical analysis of a large amount of historical normal and fault data, such as the receiver operation characteristic curve analysis, to set the optimal decision boundary that can accept the false alarm rate while ensuring a high detection rate. For example, for an SFZ11-50000 / 110 type oil-immersed power transformer, based on a large amount of simulation and measured data statistics, the first preset threshold can be determined. Set as When the largest eigenvalue of the Hessian matrix of irradiance When the value is exceeded, it is considered that there is an abnormal energy accumulation that requires attention; when Exceed At that time, the system will conduct further analysis. The stability of the peak position over time; if the peak position remains stable in space, the initial fault state is determined to be a fixed hotspot state; if the peak position drifts irregularly in space, it is determined to be a wandering hotspot state; otherwise, the system is determined to be normal. The determination of the pre-combustion glassy state is performed; the state identification adopts a multi-condition composite judgment rule; when the electrothermal instability index Exceeding the second preset threshold Simultaneous fractal dimension Within the preset range Internal, and spectral coherence length attenuation rate Exceeding the third preset threshold At that time, the initial fault state is judged as the pre-burning glass state; the working principle of the preset threshold and preset interval is to determine the optimal decision boundary by statistical methods such as receiver operation characteristic curve analysis of multi-physics field data of historical failed transformers, which aims to maximize the identification of high-risk states while minimizing the false alarm rate. Taking the SFZ11-50000 / 110 transformer as an example, by analyzing historical failure data, the second preset threshold can be determined. Set it to 0.85, and set the third preset threshold. Set as fractal dimension The preset interval can then be determined as That is, when the electrothermal instability index fractal dimension And the spectral coherence length attenuation rate When all three conditions are met, it is preliminarily determined to be in a pre-combustion glassy state; State identification indicates that the system has entered a dangerous region with strong nonlinear correlation; The identification of self-organized critical states is performed; the identification process is based on network theory; a dynamic hotspot network is constructed, in which multiple identified hotspots are regarded as network nodes, and when the spatial distance between two hotspots is less than a preset distance threshold... Furthermore, their average temperature exceeds the preset temperature threshold. When the coupling threshold is reached, a connection is established between the two nodes. The method for determining the coupling threshold is to find a critical distance and temperature combination that best characterizes the significant energy transfer effect between hotspots by statistical analysis of historical fault data; the method for determining the preset coupling threshold is to define a critical strength value that characterizes the significant energy or information exchange between two hotspots by simulation and statistical analysis of the heat conduction or electromagnetic coupling effect between hotspots in historical fault data. For example, simulation can be used to define when the heat flux density between two hot spots exceeds... Or the intensity of its electric field interaction exceeds When significant coupling is considered to exist between the two, a connection is established for the corresponding nodes in the dynamic hotspot network. Calculate the degree distribution of the network Power-law fitting of the degree distribution ,in, : Degree of a network node; γ: Power exponent; It indicates proportionality to; when the fitted power exponent is obtained When the system is within the preset range of 2 to 3, the initial fault state is determined to be a self-organized critical state. The working principle of the preset range is based on the theory of complex systems, where the power exponent of 2 to 3 is a typical feature of scale-free networks, indicating that the system has entered an extremely unstable critical state, and any slight disturbance may trigger an avalanche-like chain reaction. The technical advantage of this embodiment is that, through this hierarchical logical rule, a refined classification of fault states is achieved; it goes beyond the traditional binary judgment of normal / fault, and can identify intermediate states with different physical connotations and risk levels, providing a more accurate basis for decision-making on the intensity of subsequent response measures.

[0019] Example 3 Corrections to the initial fault condition include: Perform hysteresis loss asymmetry correction and use the hysteresis loop area space gradient map provided by the pre-stored transformer core finite element model to calibrate the weights on the optical detection path; Perform dissolved gas spectral absorption correction by subtracting faulty gas absorption peaks from the preset chemical database from the real-time spectral data; Plasma resonance pseudo-signal filtering is performed by identifying and filtering out signal peaks in the spectrum that conform to the model characteristics based on a preset surface plasmonic resonance frequency model. Demodulation of acousto-optic coupling modulation is performed by identifying sidebands of specific frequency shifts through fast Fourier transform of the optical signal and applying a digital inverse filter. In this embodiment, four specific technical means for correcting the initial fault state to obtain the final fault state are described in detail. The overall purpose of the correction steps is to improve the final accuracy of the diagnosis by introducing prior physical knowledge to identify and filter out signal distortions caused by non-fault factors. Hysteresis loss asymmetry correction is performed; the correction utilizes a pre-stored finite element model of the transformer core; the working principle of the pre-stored model is to pre-calculate the spatial distribution of hysteresis loss of the transformer core under different operating conditions through offline, high-precision electromagnetic field simulation, and generate a spatial gradient map of the hysteresis loop area; in real-time diagnostics, the map is used as a lookup table to calibrate the weights on the optical detection path caused by non-uniform heating of the core, thereby eliminating the temperature field measurement deviation introduced by hysteresis loss; During application, the system retrieves the corresponding three-dimensional hysteresis loss thermal map from the pre-stored finite element model results based on the current transformer load rate and ambient temperature. As a first-order approximation, the temperature correction at a point on the optical sensor path. Modeled as the hysteresis loss value at a point Proportional, that is: in It is an equivalent fitting coefficient obtained by statistical analysis after performing multiple offline heat transfer simulations on the entire transformer. Temperature correction amount; Spatial coordinates; Three-dimensional hysteresis loss heatmap; The correction amount will be subtracted from the original reading of the optical temperature sensor at the corresponding location to obtain a true temperature value that better reflects the fault increment, eliminating the influence of steady-state hysteresis heating. Dissolved gas spectral absorption correction is performed; the correction relies on a preset chemical database; the preset database works by pre-entering the standard absorption spectra of common fault indicator gases in a transformer oil environment; in real-time diagnosis, by matching and subtracting the real-time acquired spectra with the characteristic absorption peaks in the database, the optical signals indicating the precursor of combustion that are masked by the selective absorption of fault gases can be recovered. For example, chemical databases prominently include acetylene ( ), ethylene ) and methane ( The system obtains characteristic absorption spectral data in the near-infrared band of 1500-1700 nm. When the real-time spectrum shows absorption in the band that highly matches the characteristic peak of acetylene in the database, such as the P-branch absorption peak at 1530.37 nm, the system will use the least squares method to fit the intensity and shape of the characteristic peak and subtract it from the original spectral data to more accurately analyze the blackbody radiation continuum spectrum generated by overheating or weak discharge in the background. Plasma resonance pseudo-signal filtering is performed; the surface plasma resonance frequency model based on the preset model is corrected; the working principle of the preset model is to calculate the specific frequency that may cause surface plasma resonance and generate a flame-like spectrum based on the material and geometry of the metal components inside the transformer through electrodynamic theory; in real-time diagnosis, signal peaks in the spectrum that conform to the model characteristics are identified as pseudo-signals and filtered out to avoid misjudging non-flame signals as fire precursors. Demodulation of the acousto-optic coupling modulation is performed; correction is used to eliminate the interference of ultrasound on the optical signal through the acousto-optic effect; by performing a fast Fourier transform on the optical signal, the modulation sideband with a specific frequency shift caused by the Brillouin scattering effect can be identified; a digital inverse filter is applied to demodulate the modulation; the working principle of the digital inverse filter is that its transfer function is designed to be conjugate and opposite to the transfer function of the acousto-optic modulation process, so as to recover the original optical signal that is not affected by ultrasound interference; The technical effect of this embodiment is that, through these four directional corrections targeting different physical interference sources, the signal-to-noise ratio and reliability of fault diagnosis are greatly improved; the contamination of key signals by various non-core factors such as magnetism, chemistry, plasma and acoustics is systematically removed, ensuring that the fault state used for final decision-making is based on a true reflection of the physical process of the fault itself, thereby significantly reducing the false alarm rate and the missed alarm rate.

[0020] Tiered response strategies include: When the final fault state is determined to be entering the pre-combustion glassy state or the self-organized critical state, it is determined to be a level three alarm state, and the rapid explosion suppression and fire prevention device that executes power cut-off or releases high-pressure inert gas is immediately triggered. When the final fault state does not meet the conditions for a level 3 alarm state, but meets the preset level 2 warning conditions, it is determined to be a level 2 warning state, and auxiliary cooling is automatically started or a command to reduce the load is issued. When the final fault state does not meet the conditions for a Level 3 alarm state or a Level 2 warning state, it is determined to be a Level 1 warning state, and a maintenance recommendation is issued. In this embodiment, the specific levels and triggering conditions of the graded response strategy are described in detail. The strategy aims to match the appropriate level of intervention measures according to the severity of the final fault state in order to achieve a balance between safety and economic benefits. Level 3 alarm status is defined as the highest risk level; the trigger condition is when the final fault state is judged to have entered the pre-combustion glassy state or the self-organized critical state; once triggered, the system will immediately activate the rapid explosion suppression and fire prevention device and execute the highest priority safety measures, such as forcibly cutting off the transformer power supply or releasing high-pressure inert gas into the oil tank. A Level 2 warning status is defined as a medium-risk level. The trigger condition is that the final fault state does not meet the conditions for a Level 3 alarm, but meets other preset Level 2 warning conditions. The logic behind these preset conditions lies in defining fault states with a lower level of risk; for example, when the duration of a fixed hotspot exceeds a preset duration. , The critical duration or electrothermal instability index is obtained based on historical data statistics. Although it did not reach the level three alarm threshold, its growth rate within the time window exceeded the preset slope. , A level-two warning can be triggered whenever the critical slope characterizes the rate of deterioration. Taking transformers as an example again, the critical time can be... Setting it to 30 minutes, the critical slope characterizing the rate of deterioration will be... The value is set to 0.02 / hour; that is, the duration of the hotspot in a fixed state exceeds 30 minutes, or the electrothermal instability index... Any growth rate exceeding 0.02 within one hour triggers a Level 2 warning. Once triggered, the system will automatically activate the preset suppression program, such as automatically activating the auxiliary cooling system or sending an instruction to the power grid dispatching system to reduce the transformer load. Level 1 warning status is defined as the lowest risk level; the trigger condition is when the final failure state does not meet the conditions of Level 3 or Level 2, but the system has deviated from the ideal healthy state; once triggered, the system will issue maintenance recommendations, providing maintenance personnel with detailed diagnostic information and potential risk points, and recommending further manual inspection or arranging future maintenance plans; The technical advantage of this embodiment lies in establishing an automated response system that precisely matches the risk assessment results; avoiding a one-size-fits-all approach, and achieving appropriateness in handling measures by finely classifying risks: providing instantaneous and forceful intervention for imminent dangers; proactively and gently suppressing developing hidden dangers; and providing timely information warnings for early anomalies; this strategy minimizes unnecessary downtime and economic losses while ensuring absolute safety. This invention significantly improves the safety and reliability of power transformer operation by constructing a technical chain of multi-physics field data acquisition, deep feature extraction, high-precision intelligent diagnosis and automated hierarchical response. It can accurately and effectively intervene before a fault evolves into a serious accident, which has significant technological progress and practical value.

[0021] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A rapid explosion suppression and fire prevention method for transformers, characterized in that, The specific steps include: S1. Real-time acquisition of multi-physics data through a sensor network deployed inside the transformer; S2. Based on multiphysics data, determine the electrothermal instability index, the fractal dimension of fire nucleus growth, the maximum eigenvalue of the Hessian matrix of irradiance, and the spectral coherence length attenuation rate. S3. Construct an eigenvector by combining the electrothermal instability index, the fractal dimension of fire nucleus growth, the maximum eigenvalue, and the spectral coherence length attenuation rate; S4. Identify the initial fault state based on feature vectors, correct the initial fault state using a preset physical model to obtain the final fault state, and trigger a graded response strategy based on the final fault state.

2. The rapid explosion suppression and fire prevention method for transformers according to claim 1, characterized in that, The multiphysics data includes at least multi-band optical spectral signals, electric field intensity signals, temperature distribution signals, and ultrasonic signals.

3. The rapid explosion suppression and fire prevention method for transformers according to claim 1, characterized in that, In S2, the determination of the electrothermal instability index, the fractal dimension of fire nucleus growth, the maximum eigenvalue of the irradiance Hessian matrix, and the spectral coherence length attenuation rate includes: The ratio of the relative rate of change of conductivity over time to the relative rate of change of temperature over time is calculated to obtain the electrothermal instability index. Based on temperature and light intensity data, the fractal dimension of fire nucleus growth is calculated using the box counting method. Calculate the Hessian matrix of the photon flux density scalar field and extract the largest eigenvalue of the Hessian matrix; The time series of spectral coherence lengths is numerically differentiated to calculate the spectral coherence length attenuation rate.

4. The rapid explosion suppression and fire prevention method for transformers according to claim 3, characterized in that, The calculation of the fractal dimension of the fire core growth includes: Potential hotspots were identified using an adaptive thresholding method based on temperature and light intensity data. Define a series of observation scales and count the number of hotspots within each observation scale; In a logarithmic coordinate system, linear regression analysis is performed on data points with varying observation scales and hotspot numbers, and the slope of the fitted line from the linear regression analysis is used as the fractal dimension.

5. The rapid explosion suppression and fire prevention method for transformers according to claim 1, characterized in that, The identification of the preliminary fault state includes: When the maximum eigenvalue exceeds the first preset threshold, the initial fault state is determined to be either a hotspot fixed state or a hotspot wandering state based on the time series stability of the peak position of the maximum eigenvalue; otherwise, it is determined to be normal. When the electrothermal instability index exceeds the second preset threshold, the fractal dimension is within the preset range, and the spectral coherence length attenuation rate exceeds the third preset threshold, the initial fault state is determined to be the pre-combustion glass state. A dynamic hotspot network is constructed and the degree distribution of the network is calculated. When the power-law fitting exponent of the degree distribution is within the preset range of 2 to 3, the initial fault state is determined to be a self-organized critical state.

6. The rapid explosion suppression and fire prevention method for transformers according to claim 1, characterized in that, The correction of the initial fault state includes: Perform hysteresis loss asymmetry correction and use the hysteresis loop area space gradient map provided by the pre-stored transformer core finite element model to calibrate the weights on the optical detection path; Perform dissolved gas spectral absorption correction by subtracting faulty gas absorption peaks from the preset chemical database from the real-time spectral data; Plasma resonance pseudo-signal filtering is performed by identifying and filtering out signal peaks in the spectrum that conform to the model characteristics based on a preset surface plasmonic resonance frequency model. Demodulation of the acousto-optic coupling modulation is performed by identifying the sidebands of the frequency shift caused by the Brillouin scattering effect through a fast Fourier transform of the optical signal, and then applying a digital inverse filter.

7. The rapid explosion suppression and fire prevention method for transformers according to claim 1, characterized in that, The hierarchical response strategy includes: When the final fault state is determined to be entering the pre-combustion glassy state or the self-organized critical state, it is determined to be a level three alarm state, and the rapid explosion suppression and fire prevention device that executes power cut-off or releases high-pressure inert gas is immediately triggered. When the final fault state does not meet the conditions for a level 3 alarm state, but meets the preset level 2 warning conditions, it is determined to be a level 2 warning state, and auxiliary cooling is automatically started or a command to reduce the load is issued. When the final fault state does not meet the conditions for a Level 3 alarm state or a Level 2 early warning state, it is determined to be a Level 1 early warning state, and a maintenance recommendation is issued.

Images