A transformer rapid explosion suppression and fire prevention method

Abstract

The application relates to a transformer rapid explosion suppression and fire prevention method, and belongs to the technical field of power equipment safety, which specifically comprises the following steps: S1, collecting multi-physical field data in real time through a sensor network arranged in the transformer; S2, determining an electric-thermal instability index, a fractal dimension of a fire core growth, a maximum eigenvalue of an irradiance Hessian matrix, and a spectral coherence length decay rate based on the multi-physical field data; S3, combining the electric-thermal instability index, the fractal dimension of the fire core growth, the maximum eigenvalue, and the spectral coherence length decay rate to construct a feature vector; and S4, identifying a preliminary fault state based on the feature vector, correcting the preliminary fault state by using a preset physical model to obtain a final fault state, and triggering a hierarchical response strategy according to the final fault state. The application solves the problem that a single signal source is prone to failure under the interference of a complex electromagnetic and thermal environment, and provides a solid data foundation for subsequent accurate diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power equipment safety, in particular to a transformer rapid explosion suppression and combustion prevention method. BACKGROUND

[0002] The internal insulation oil medium of a power transformer will deteriorate under long-term electrical stress and thermal stress. If the partial overheating or discharge generated under fault conditions cannot be discovered in time, it can evolve into a malignant accident, causing combustion or even explosion; the existing detection technology has limitations in dealing with complex internal environments of the transformer. The propagation of optical signals in the oil medium is affected by the coupling of spectrum-medium-temperature field, resulting in nonlinear distortion; during the critical stage before combustion, weak photon radiation signals are severely masked by environmental noise. Multiple potential fault hotspots that may exist simultaneously inside compete through heat migration and electromagnetic field coupling, leading to uncertainty in predicting the real ignition point. Therefore, there is an urgent need for a method that can penetrate complex environmental noise, accurately identify early fault characteristics and predict their evolution trend.

[0003] The above information disclosed in the above BACKGROUND section is only intended to strengthen the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0004] The purpose of the present application is to provide a transformer rapid explosion suppression and combustion prevention method to solve the problems raised in the above BACKGROUND.

[0005] The technical solution of the present application is as follows, the specific steps include:

[0006] S1, real-time collection of multi-physical field data through a sensor network deployed inside the transformer;

[0007] S2, determination of the electro-thermal instability index, fractal dimension of fire kernel growth, maximum eigenvalue of irradiance Hessian matrix, and spectral coherence length decay rate based on the multi-physical field data;

[0008] S3, construction of a feature vector combining the electro-thermal instability index, fractal dimension of fire kernel growth, maximum eigenvalue, and spectral coherence length decay rate;

[0009] S4, identification of a preliminary fault state based on the feature vector, correction of the preliminary fault state using a pre-set physical model to obtain a final fault state, and triggering of a hierarchical response strategy according to the final fault state.

[0010] Preferably, the multi-physical field data includes at least multi-band optical spectral signals, electric field intensity signals, temperature distribution signals, and ultrasonic signals.

[0011] Preferably, the determination of the electro-thermal instability index, the fractal dimension of the fire kernel growth, the maximum eigenvalue of the irradiance Hessian matrix, and the spectral coherence length decay rate in S2 comprises:

[0012] calculating the ratio of the relative time variation rate of the conductivity to the relative time variation rate of the temperature to obtain the electro-thermal instability index;

[0013] calculating the fractal dimension of the fire kernel growth based on the temperature and light intensity data by the principle of box counting method;

[0014] calculating the Hessian matrix of the photon flux density scalar field and extracting the maximum eigenvalue of the Hessian matrix;

[0015] numerically deriving the time series of the spectral coherence length to calculate the spectral coherence length decay rate.

[0016] Preferably, the calculation of the fractal dimension of the fire kernel growth comprises:

[0017] identifying potential hotspots based on the temperature and light intensity data by the adaptive threshold method;

[0018] defining a series of observation scales and counting the number of hotspots within each observation scale;

[0019] performing linear regression analysis on the data points of the observation scales and the number of hotspots in the logarithmic coordinate system, and taking the slope of the fitting straight line of the linear regression analysis as the fractal dimension.

[0020] Preferably, the identification of the preliminary fault state comprises:

[0021] when the maximum eigenvalue exceeds the first preset threshold, determining the preliminary fault state as the hotspot fixed state or the hotspot wandering state according to the time series stability of the maximum eigenvalue peak position, otherwise as normal;

[0022] when the electro-thermal instability index exceeds the second preset threshold, the fractal dimension is within the preset interval, and the spectral coherence length decay rate exceeds the third preset threshold, determining the preliminary fault state as the pre-combustion glass state;

[0023] constructing a dynamic hotspot network and calculating the degree distribution of the network, when the power-law fitting index of the degree distribution is within the preset range of 2 to 3, determining the preliminary fault state as the self-organized critical state.

[0024] Preferably, the modification of the preliminary fault state comprises:

[0025] performing hysteresis loss asymmetry correction, using the pre-stored hysteresis loop area spatial gradient map provided by the finite element model of the transformer core to calibrate the weight on the optical detection path;

[0026] Perform dissolved gas spectral absorption correction by subtracting faulty gas absorption peaks from the preset chemical database from the real-time spectral data;

[0027] 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.

[0028] 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.

[0029] Preferably, the hierarchical response strategy includes:

[0030] 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.

[0031] 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.

[0032] 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.

[0033] 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:

[0034] 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.

[0035] 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;

[0036] 3.The application establishes a set of cascading diagnostic logic; not only can it identify early abnormalities such as hotspot fixation and hotspot wandering, but also can accurately determine the pre-combustion glass state through composite judgment, and identify the self-organized critical state that indicates a chain failure through the construction of a dynamic hotspot network; this fine division of fault stages goes beyond the traditional binary normal / fault theory, providing accurate decision input for hierarchical disposal; more importantly, the method introduces a preset physical model to correct the preliminary diagnosis results, through the execution of hysteresis loss asymmetry correction, dissolved gas spectrum absorption correction, plasma resonance false signal filtering, and acousto-optic coupling demodulation, systematically stripping the interference from multiple physical processes such as magnetism, chemistry, plasma, and acoustics, ensuring that the final fault state judgment is based on the true reflection of the fault body, greatly reducing the risk of false positives and false negatives;

[0037] 4.According to the severity level of the final fault state, the application triggers the corresponding hierarchical response strategy; for extreme danger such as pre-combustion glass state or self-organized critical state, the system immediately executes forced intervention such as power-off or release of high-pressure inert gas; for developing hazards, it automatically starts auxiliary cooling or issues a load reduction instruction for active suppression; for early deviations, it issues maintenance recommendations for early warning; this automated response system that accurately matches risk assessment achieves the best balance between safety and economic benefits, enabling instantaneous and powerful intervention before disaster occurs, while minimizing unnecessary downtime losses. BRIEF DESCRIPTION OF DRAWINGS

[0038] The application will be further explained in conjunction with the accompanying drawings and examples:

[0039] Figure 1 is a flowchart of a transformer rapid explosion suppression and combustion prevention method of the application. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the application clearer, the application will be further described in detail below in conjunction with specific examples.

[0041] Example 1

[0042] Please refer to Figure 1 , the application provides a transformer rapid explosion suppression and combustion prevention method, the specific steps of which include:

[0043] S1, real-time acquisition of multi-physical field data through a sensor network deployed inside the transformer;

[0044] S2, based on the multi-physical field data, determine the electric heating instability index, the fractal dimension of the fire kernel growth, the maximum eigenvalue of the irradiance Hessian matrix, and the spectral coherence length decay rate;

[0045] S3, combine the electro-thermal instability index, the fractal dimension of fire kernel growth, the maximum eigenvalue, and the spectral coherence length decay rate to construct a feature vector;

[0046] S4, identify the preliminary fault state based on the feature vector, correct the preliminary fault state using a pre-set physical model to obtain the final fault state, and trigger a hierarchical response strategy according to the final fault state;

[0047] The embodiment provides a transformer rapid explosion suppression and combustion prevention method; the overall technical process of the method is designed as a closed-loop sensing, analyzing and responding system, aiming to realize accurate early warning and active intervention on potential faults inside the transformer;

[0048] The specific steps of the method are implemented; data collection is performed through a sensor network deployed inside the transformer; the sensor network refers to a spatially distributed data collection system composed of multiple types of sensors, aiming to capture multi-dimensional physical information of the running state inside the transformer in real time without dead angles; in the embodiment, the network is configured at key positions inside the transformer oil tank to ensure continuous monitoring of the core area;

[0049] For a three-phase oil-immersed power transformer, the sensor network can be deployed as follows: at each winding of the A, B and C three-phase, near the geometric center of the high-voltage, medium-voltage and low-voltage coils, three monitoring points are arranged in the axial direction, a total of nine core monitoring points; each monitoring point is integrated with a composite sensor probe, the probe includes a fiber Bragg grating sensor for measuring temperature, a micro electric field probe for measuring electric field intensity, and an ultraviolet sensitive photodiode for detecting ultraviolet photon flux; in addition, one wideband piezoelectric ultrasonic sensor is installed at the center of each of the four inner surfaces of the oil tank wall for receiving global acoustic signals; through this three-dimensional layout of nine points in three dimensions + four auxiliary surfaces, the winding heating, discharge core area and overall state of the oil tank can be monitored comprehensively;

[0050] Based on the collected multi-physical field data, a set of key feature parameters is determined; the step aims to convert the original, high-dimensional sensor data stream into low-dimensional, high-information-density indicators that can deeply represent the physical nature of the fault; to achieve the purpose, four core parameters are determined, which are the electro-thermal instability index, the fractal dimension of fire kernel growth, the maximum eigenvalue of the irradiance Hessian matrix, and the spectral coherence length decay rate; this set of parameters jointly describes the evolution state of the fault from four mutually orthogonal dimensions of electro-thermal stability, spatial form complexity, energy aggregation degree and signal transmission quality;

[0051] The determined electrothermal instability index, the fractal dimension of the fire core growth, the maximum eigenvalue and the spectral coherence length decay rate are constructed as a feature vector; the feature vector refers to an ordered set containing all four core parameters calculated in the previous steps, and functions to provide a standardized, multi-dimensional input for the subsequent pattern recognition algorithm; by integrating parameters with different physical meanings, the feature vector can more comprehensively and robustly reflect the overall state of the system than any single parameter;

[0052] A cascading fault diagnosis and response process is performed based on the feature vector; the internal logic of the process is: a set of pre-set logical rules are used to analyze the feature vector to identify the preliminary fault state; in order to eliminate the influence of known physical interference factors on the diagnosis result, the preliminary fault state is corrected using a pre-set physical model, so as to obtain a more accurate and reliable final fault state; and according to the final fault state, a hierarchical response strategy is triggered, and corresponding safety measures are executed according to the diagnosed risk level;

[0053] The embodiment solves the problems of weak signals, strong noise interference and uncertain fault evolution in complex working conditions that the traditional method cannot handle by constructing a complete technical link from multi-dimensional data acquisition, multi-angle feature extraction, intelligent state recognition to automated hierarchical response; the technical effect is that the early fault features can be accurately identified and the evolution trend can be predicted by penetrating through complex environmental noise, so that effective intervention can be performed before the fault evolves into a malignant accident, thereby greatly improving the operation safety of the power transformer.

[0054] The multi-physical field data at least includes multi-band optical spectrum signals, electric field intensity signals, temperature distribution signals and ultrasonic signals;

[0055] The embodiment limits the specific composition of the multi-physical field data; the purpose of the multi-physical field data is to capture the precursor information of the fault from multiple physical dimensions to construct a comprehensive state view; in the embodiment, the data is required to at least include multi-band optical spectrum signals, electric field intensity signals, temperature distribution signals and ultrasonic signals;

[0056] The multi-band optical spectrum signals function to monitor the photon radiation as a precursor of combustion caused by local discharge or overheating in the insulating medium; the electric field intensity signals are used to directly monitor the abnormal fluctuation of the electric field inside the transformer, indicating the potential risk of insulation breakdown; the temperature distribution signals are used to depict the spatial distribution of the internal thermal field and identify local hot spots; the ultrasonic signals are used to detect the sound waves generated by local discharge or mechanical vibration as another evidence of the fault;

[0057] The technical effect of the embodiment is that by explicitly specifying the four key data sources, the comprehensiveness and complementarity of the input information are ensured; the fusion of optical, electrical, thermal and acoustic signals enables the subsequent feature extraction and state recognition to be based on a more rich and 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.

[0058] Embodiment 2

[0059] In S2, the determination of the electro-thermal instability index, the fractal dimension of the fire kernel growth, the maximum eigenvalue of the irradiance Hessian matrix, and the spectral coherence length decay rate includes:

[0060] The ratio of the relative time variation rate of the electrical conductivity to the relative time variation rate of the temperature is calculated to obtain the electro-thermal instability index;

[0061] Based on the temperature and light intensity data, the fractal dimension of the fire kernel growth is calculated by the box counting method principle;

[0062] The Hessian matrix of the photon flux density scalar field is calculated, and the maximum eigenvalue of the Hessian matrix is extracted;

[0063] The time series of the spectral coherence length is numerically differentiated to calculate the spectral coherence length decay rate;

[0064] The calculation of the fractal dimension of the fire kernel growth includes:

[0065] Based on the temperature and light intensity data, the potential hot spots are identified by the adaptive threshold method;

[0066] A series of observation scales are defined, and the number of hot spots in each observation scale is counted;

[0067] In the logarithmic coordinate system, linear regression analysis is performed on the data points of the observation scale and the number of hot spots, and the slope of the fitting straight line of the linear regression analysis is taken as the fractal dimension;

[0068] The embodiment on the basis of embodiment 1 describes the determination method of the four core parameters in step S2, ensuring the implementability of the technical solution;

[0069] The purpose of determining the electro-thermal instability index is to quantify the risk of insulating oil tending to thermal insulation failure from the perspective of electro-thermal coupling; the calculation model is defined as:

[0070]

[0071] wherein, is the electro-thermal instability index; is the real-time electrical conductivity measured by the electrical conductivity sensor; the real-time absolute temperature measured by the temperature sensor; with respectively the time rate of change of conductivity and temperature;

[0072] By numerically differentiating the respective time series data, the time rate of change can be calculated, and then the dimensionless index is obtained; the rapid growth of the index is regarded as a key precursor of thermal runaway;

[0073] The determination of the fractal dimension of the kernel growth aims to distinguish the real sustained flame kernel from the random isolated interference hot spot from the perspective of spatial complexity; the calculation process follows a standard procedure; in the procedure, potential hot spots are identified; potential hot spots refer to high signal regions marked by an adaptive threshold method in temperature and light intensity data; the adaptive threshold method refers to an algorithm in which the threshold is dynamically adjusted according to the statistical characteristics of the local background of the signal, which effectively suppresses global noise and accurately segments meaningful abnormal points; for example, the method can be implemented by a bimodal method or an Otsu method based on the local signal mean and standard deviation to ensure the robustness of segmentation under different background noise levels; a series of observation scales covering from millimeter to decimeter are defined, and the number of hot spots contained in each scale is counted ; linear regression analysis is performed on the data points in the logarithmic coordinate system; the slope of the fitted straight line obtained by the analysis is defined as the fractal dimension :

[0074]

[0075] wherein, is the fractal dimension; is the number of hot spots; is the observation scale; Δ: difference symbol, indicating the change amount; : logarithmic symbol;

[0076] The value of stabilized in a certain interval represents that the kernel is expanding in a way with complex structure and statistical self-similarity, rather than random noise;

[0077] The determination of the largest eigenvalue of the Hessian matrix aims to quantitatively describe the degree of spatial concentration of photon radiation energy around the fault hot spot; the Hessian matrix of the irradiance scalar field is constructed , wherein is measured in real time by an array of ultraviolet-sensitive photodiodes; the Hessian matrix is a second-order tensor, and the components are composed of the second-order partial derivatives of irradiance with respect to spatial coordinates , which can accurately describe The local curvature of the field in space; in order to compare the tensor with the scalar threshold, the maximum eigenvalue of the matrix needs to be extracted ; It physically represents the curvature of the most intense direction of irradiance in space, and is the most sensitive indicator of hotspot intensity and energy concentration;

[0078] The determination of the relative decay rate of the spectral coherence length aims to evaluate the transmission quality of the optical signal in the oil medium in real time, and the deterioration is an important indicator of medium deterioration and fault environment formation; its calculation model is defined as:

[0079]

[0080] Among them, is the decay rate; is the spectral coherence length; is the time; : differential symbol;

[0081] Spectral coherence length is a physical quantity calculated by processing the collected spectral signal by Michelson interference; by numerically differentiating the time series of , the decay rate can be obtained;

[0082] The technical effect of this embodiment is to provide four parameters with clear physical meaning and clear calculation method for the construction of feature vectors; The parameters quantize the fault state from four dimensions of thermoelectric stability, spatial form, energy concentration and channel quality, so that the subsequent fault identification is no longer a simple threshold judgment based on the original signal, but an accurate characterization of the deep physical mechanism of the fault, thereby greatly improving the accuracy and reliability of the diagnosis;

[0083] In order to increase the robustness of the model, a protective threshold is set when calculating ; only when the relative time change rate of temperature is calculated ; if , it is considered that the system is in a thermally stable state, and is set to a safe default value such as 0; it can effectively avoid the meaningless divergence of the index due to temperature stability and too small denominator, and prevent system false positives.

[0084] The preliminary fault state identification includes:

[0085] When the maximum eigenvalue exceeds the first preset threshold, the preliminary fault state is determined to be a hotspot fixed state or a hotspot wandering state according to the time series stability of the peak position of the maximum eigenvalue, otherwise it is determined to be normal;

[0086] When the electric-thermal instability index exceeds the second preset threshold, the fractal dimension is within the preset interval, and the spectral coherence length decay rate exceeds the third preset threshold, the preliminary fault state is determined as pre-combustion glass state;

[0087] A dynamic hot spot network is constructed and the degree distribution of the network is calculated. When the power-law fitting index of the degree distribution is within the preset range of 2 to 3, the preliminary fault state is determined as a self-organized critical state;

[0088] In this embodiment, the specific logical rules for identifying the preliminary fault state are described in detail. The identification process is designed as a set of cascading judgment logic to distinguish different fault evolution stages.

[0089] The hot spot bistable state phenomenon is identified. The first preset threshold is used to determine the maximum eigenvalue of irradiance Abnormality; the determination of the preset threshold is based on statistical analysis of a large amount of historical normal and fault data, such as using receiver operating characteristic curve analysis to set the best decision boundary that can accept the false positive rate while ensuring high detection rate.

[0090] For example, for an SFZ11-50000 / 110 type oil-immersed power transformer, through a large number of simulation and measurement data statistics, the first preset threshold is set to ; when the maximum eigenvalue of the irradiance Hessian matrix exceeds the value, it is considered that there is abnormal energy aggregation that needs attention.

[0091] When exceeds , the system further analyzes the stability of the peak position in the time series; if the peak position remains stable in space, the preliminary fault state is determined as hot spot fixed state; if the peak position drifts irregularly in space, it is determined as hot spot wandering state; otherwise, the system is determined as normal.

[0092] The pre-combustion glass state is determined; the state recognition adopts a multi-condition composite judgment rule; when the electric-thermal instability index exceeds the second preset threshold , the fractal dimension is within the preset interval , and the spectral coherence length decay rate exceeds the third preset threshold ​When the three conditions are met simultaneously, the preliminary fault state is determined as pre-ignition glass state; the working principle of the preset threshold and the preset interval is the optimal decision boundary determined by statistical methods such as receiver operating characteristic curve analysis on the multi-physical field data of historical failed transformers, aiming to maximize the identification of high-risk states while minimizing the false positive rate;

[0093] Taking the SFZ11-50000 / 110 transformer as an example, through analysis of historical failure data, the second preset threshold may be set as 0.85, the third preset threshold may be set as ; and the preset interval of the fractal dimension may be determined as ; that is, when the electric thermal instability index , the fractal dimension , and the spectral coherence length decay rate all meet the conditions, the preliminary determination is pre-ignition glass state;

[0094] The identification of the state marks that the system enters a dangerous region of strong nonlinear correlation;

[0095] The self-organized critical state is identified; the identification process is based on network theory; a dynamic hot spot network is constructed, in which multiple identified hot spots are regarded as network nodes; when the spatial distance between two hot spots is less than a preset distance threshold , and the average temperature of the two hot spots exceeds a preset temperature threshold , a connection is established between the two nodes. The determination method of the coupling threshold is to find a critical distance and temperature combination that best represents the significant energy transfer effect between hot spots through statistical analysis of historical failure data; the determination method of the preset coupling threshold is to define a critical strength value that can represent the existence of significant energy or information exchange between two hot spots through simulation and statistical analysis of the heat conduction or electromagnetic coupling effect between hot spots in historical failure data;

[0096] For example, when the heat flux density between two hot spots exceeds or the electric field interaction strength exceeds , it is considered that there is significant coupling between the two, and a connection is established between the corresponding nodes in the dynamic hot spot network;

[0097] The degree distribution of the network is calculated ; the degree distribution is fitted by a power law , where : the degree of the network node; γ: the power index; represents the proportion; when the power index obtained by fitting is When in the preset range of 2 to 3, the preliminary fault state is determined to be a self-organizing critical state; the working principle of the preset range is based on complex system theory, in which the power index of 2 to 3 is a typical feature of a scale-free network, indicating that the system has entered a critically unstable state, and any slight disturbance may trigger an avalanche-like chain reaction;

[0098] The technical effect of the embodiment is that the hierarchical logical rules achieve fine classification of the fault state; the traditional normal / fault binary judgment is transcended, and the intermediate state with different physical connotations and risk levels can be identified, thereby providing more accurate decision basis for subsequent response measures.

[0099] Embodiment 3

[0100] The modification of the preliminary fault state includes:

[0101] The hysteresis loss asymmetry correction is performed, and the spatial gradient graph of the hysteresis loop area provided by the pre-stored finite element model of the transformer core is used to calibrate the weight on the optical detection path;

[0102] The dissolved gas spectrum absorption correction is performed, and the fault gas absorption peak in the preset chemical database is subtracted from the real-time spectrum data;

[0103] The plasma resonance pseudo-signal filtering is performed, and the signal peak meeting the model characteristics in the spectrum is identified and filtered according to the preset surface plasma resonance frequency model;

[0104] The demodulation of the acousto-optic coupling modulation is performed, the sideband with a specific frequency shift is identified by performing fast Fourier transform on the optical signal, and a digital inverse filter is applied;

[0105] In the embodiment, the four specific technical means for modifying the preliminary fault state to obtain the final fault state are described in detail; the overall purpose of the modification step is to identify and filter the signal distortion caused by non-fault factors by introducing prior physical knowledge, thereby improving the final accuracy of the diagnosis;

[0106] The hysteresis loss asymmetry correction is performed; the correction utilizes the pre-stored finite element model of the transformer core; the working principle of the preset model is that the spatial distribution of the hysteresis loss of the transformer core under different working conditions is pre-calculated through offline and high-precision electromagnetic field simulation, and a spatial gradient graph of the hysteresis loop area is generated; in real-time diagnosis, the graph is used as a lookup table to calibrate the weight on the optical detection path caused by the non-uniform heating of the core, thereby eliminating the temperature field measurement deviation introduced by the hysteresis loss;

[0107] In application, the system calls the corresponding three-dimensional hysteresis loss heat map from the pre-stored finite element model results according to the load rate and the ambient temperature of the current transformer ; as a first order approximation, the temperature correction at a certain point on the optical sensor path is modeled to be proportional to the hysteresis loss value at the point, i.e.

[0108]

[0109] wherein is an equivalent fitting coefficient statistically obtained through multiple off-line heat transfer simulations on the whole transformer;

[0110] temperature correction; spatial coordinates; three-dimensional hysteresis loss heat map;

[0111] The correction will be subtracted from the original reading of the optical temperature sensor at the corresponding position, so as to obtain a real temperature value that is free from the influence of steady-state hysteresis heating and better reflects the fault increment;

[0112] Dissolved gas spectrum absorption correction is performed; the correction relies on a preset chemical database; the working principle of the preset database is that the standard absorption spectrum of common fault indicating gases in the transformer oil environment is pre-recorded; in real-time diagnosis, by matching and subtracting the real-time collected spectrum from the characteristic absorption peaks in the database, the optical signal indicating the combustion precursor that is selectively absorbed by the fault gas and hidden can be recovered;

[0113] For example, the chemical database mainly contains the characteristic absorption spectrum data of acetylene ( ), ethylene ( ) and methane ( ) in the near-infrared waveband of 1500-1700 nm; when the real-time spectrum appears an absorption in the waveband that is highly consistent with the characteristic peak of acetylene in the database, such as the P-branch absorption peak at 1530.37 nm, the system will fit the intensity and shape of the characteristic peak using the least squares method, and subtract it from the original spectrum data to more accurately analyze the blackbody radiation continuum in the background produced by overheating or weak discharge;

[0114] Plasmon resonance pseudo-signal filtering is performed; the correction is based on a preset surface plasmon resonance frequency model; the working principle of the preset model is that according to the material and geometry of the metal components inside the transformer, the specific frequency that may occur surface plasmon resonance and produce a flame spectrum-like signal is calculated through electrodynamic theory; in real-time diagnosis, the signal peaks in the spectrum that meet the characteristics of the model are identified as pseudo-signals and filtered out to avoid misjudging non-flame signals as fire precursors;

[0115] ​​The demodulation of the acousto-optic coupling modulation is performed; the correction is used to eliminate the interference caused by the ultrasonic wave to the optical signal through the acousto-optic effect; by performing the 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 opposite to the conjugate of the transfer function of the acousto-optic modulation process, so that the original optical signal not disturbed by the ultrasonic wave can be recovered;

[0116] The technical effect of the embodiment is that, through the four directional corrections for different physical interference sources, the signal-to-noise ratio and reliability of fault diagnosis are greatly improved; various non-core factors such as magnetism, chemistry, plasma and acoustics pollute the key signal, ensuring that the final fault state for decision-making is based on the true reflection of the physical process of the fault body, thereby significantly reducing the false positive rate and the false negative rate.

[0117] The hierarchical response strategy includes:

[0118] When the final fault state is judged to enter the pre-combustion glass state or the self-organized critical state, a level three alarm state is determined, and a quick explosion suppression and fire prevention device for executing power-off or releasing high-pressure inert gas is immediately triggered;

[0119] When the final fault state does not meet the conditions of the level three alarm state, but meets the preset level two warning conditions, a level two warning state is determined, and auxiliary cooling or a command to reduce the load is automatically started;

[0120] When the final fault state does not meet the conditions of the level three alarm state and the level two warning state, a level one warning state is determined, and a maintenance suggestion is issued;

[0121] In the embodiment, the specific levels and triggering conditions of the hierarchical response strategy are described in detail; the strategy aims to match the intervention measures of corresponding intensity according to the severity of the final fault state, to achieve the balance between safety and economic benefits;

[0122] The level three alarm state is defined as the highest risk level; the triggering condition is that when the final fault state is judged to enter the pre-combustion glass state or the self-organized critical state; once triggered, the system will immediately activate the quick 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;

[0123] The level two warning state is defined as the medium risk level; the triggering condition is that when the final fault state does not meet the conditions of the level three alarm, but meets the preset level two warning conditions; the logic of the preset condition is to define the fault state of the next level of risk degree, for example, when the duration of the hot spot fixed state exceeds the preset time , is the critical time length obtained based on historical data statistics, or the thermal instability index Although the third-level alarm threshold is not reached, but its growth rate in the time window exceeds the preset slope , When the critical slope representing the deterioration speed is reached, the second-level early warning can be triggered.

[0124] Still taking the transformer as an example, the critical time length is set to 30 minutes, and the critical slope representing the deterioration speed is set to 0.02 / hour; that is, the duration of the hotspot fixed state exceeds 30 minutes, or the growth rate of the thermal instability index in one hour exceeds 0.02, both of which trigger the second-level early warning.

[0125] Once triggered, the system will automatically start the preset suppression program, such as automatically starting the auxiliary cooling system or issuing instructions to the power grid dispatching system to reduce the load of the transformer.

[0126] The first-level early warning state is defined as the lowest risk level; the triggering condition is when the final failure state does not meet the conditions of the third or second levels, but the system has deviated from the ideal healthy state; once triggered, the system will issue maintenance recommendations, provide detailed diagnostic information and potential risk points to the operation and maintenance personnel, and suggest further manual inspection or arrange future maintenance plans.

[0127] The technical effect of the embodiment is that an automated response system that accurately matches the risk assessment results is established; a one-size-fits-all approach is avoided, and the appropriateness of the disposal measures is achieved by fine-grained classification of risks: instantaneous and strong intervention for imminent danger; active and moderate suppression for developing hidden dangers; timely information early warning for early abnormalities; this strategy maximizes the reduction of unnecessary downtime and economic losses under the premise of ensuring absolute safety.

[0128] The present application significantly improves the safety and reliability of power transformer operation by constructing a technical link of multi-physical field data acquisition, deep feature extraction, high-precision intelligent diagnosis and automated grading response, and can accurately and effectively intervene before the fault evolves into a malignant accident, which has great technical progress and practical value.

[0129] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A transformer rapid explosion suppression and fire prevention method, characterized in that, The specific steps comprise: S1, collecting multi-physical field data in real time through a sensor network deployed inside the transformer; S2, determining an electro-thermal instability index, a fractal dimension of fire kernel growth, a maximum eigenvalue of an irradiance Hessian matrix, and a spectral coherence length decay rate based on the multi-physical field data; S3, constructing a feature vector in combination with the electro-thermal instability index, the fractal dimension of fire kernel growth, the maximum eigenvalue, and the spectral coherence length decay rate; S4, identifying a preliminary fault state based on the feature vector, correcting the preliminary fault state by using a preset physical model to obtain a final fault state, and triggering a hierarchical response strategy according to the final fault state.

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

3. The transformer fast explosion suppression and fire prevention method according to claim 1, characterized in that, In S2, the determination of the electro-thermal instability index, the fractal dimension of fire kernel growth, the maximum eigenvalue of the irradiance Hessian matrix, and the spectral coherence length decay rate comprises: calculating the ratio of the relative time variation rate of the conductivity to the relative time variation rate of the temperature to obtain the electro-thermal instability index; calculating the fractal dimension of fire kernel growth based on the temperature and light intensity data through the principle of box counting method; calculating the Hessian matrix of the photon flux density scalar field and extracting the maximum eigenvalue of the Hessian matrix; numerically deriving the time series of the spectral coherence length to calculate the spectral coherence length decay rate.

4. The transformer fast explosion suppression and fire prevention method according to claim 3, characterized in that, The calculation of the fractal dimension of fire kernel growth comprises: identifying potential hot spots through an adaptive threshold method based on the temperature and light intensity data; defining a series of observation scales and counting the number of hot spots inside each observation scale; in a logarithmic coordinate system, performing linear regression analysis on the data points of the observation scales and the number of hot spots, and taking the slope of the fitting straight line of the linear regression analysis as the fractal dimension.

5. The transformer fast explosion suppression and fire prevention method according to claim 1, characterized in that, The identification of the preliminary fault state comprises: when the maximum eigenvalue exceeds a first preset threshold, determining the preliminary fault state as a hot spot fixed state or a hot spot wandering state according to the time series stability of the peak position of the maximum eigenvalue, or as normal otherwise; when the electro-thermal instability index exceeds a second preset threshold, the fractal dimension is within a preset interval, and the spectral coherence length decay rate exceeds a third preset threshold, determining the preliminary fault state as a pre-combustion glass state; constructing a dynamic hot spot network and calculating the degree distribution of the network, when the power law fitting index of the degree distribution is within a preset range of 2 to 3, determining the preliminary fault state as a self-organized critical state.

6. The transformer fast explosion suppression and fire prevention method according to claim 1, characterized in that, The correction of the preliminary fault state comprises: performing hysteresis loss asymmetry correction, using a pre-stored magnetic hysteresis loop area spatial gradient graph provided by a transformer core finite element model to calibrate the weight on the optical detection path; performing dissolved gas spectral absorption correction, subtracting a fault gas absorption peak in a preset chemical database from real-time spectral data; performing plasma resonance pseudo-signal filtering, identifying and filtering out signal peaks in the spectrum that meet the characteristics of a preset surface plasmon resonance frequency model; performing demodulation of acousto-optic coupling modulation, identifying the frequency shift sideband caused by the Brillouin scattering effect by performing fast Fourier transform on the optical signal, and applying a digital inverse filter.

7. The transformer fast explosion suppression and fire prevention method according to claim 1, characterized in that, The hierarchical response strategy comprises: When the final failure state is judged to enter the pre-combustion glass state or the self-organized critical state, a third-level alarm state is determined, and a quick explosion suppression and fire prevention device for executing power-off or releasing high-pressure inert gas is immediately triggered; When the final failure state does not meet the conditions of the third-level alarm state, but meets the preset second-level warning conditions, a second-level warning state is determined, and auxiliary cooling is automatically started or an instruction to reduce the load is issued; When the final failure state does not meet the conditions of the third-level alarm state and the second-level warning state, a first-level warning state is determined, and a maintenance suggestion is issued.

Images