Double-domain BBO transformer coupling fault diagnosis method based on habitat transformation mechanism

Abstract

The invention discloses a double-domain BBO transformer coupling fault diagnosis method based on a habitat transformation mechanism, and the method comprises the steps: constructing a chemical domain population and an electromagnetic domain population based on the operation feature data of a transformer; constructing a fault coupling matrix; fitness values of habitats in the current chemical domain population and the electromagnetic domain population are calculated, and migration operation of the standard BBO is executed based on the fitness values; in each iteration of executing the basic migration operation, when an elitist solution meeting a preset condition exists in the electromagnetic domain population, performing habitat transformation on the chemical domain population according to the elitist solution and the fault coupling matrix, including search space distortion and feature dimension dynamic reconstruction; and calculating fitness values of the current electromagnetic domain population and the chemical domain population, and determining the fault type of the transformer through weighted fusion. According to the method, the transformer to be detected does not need to be subjected to power failure hanging cover detection, high-precision state evaluation can be realized only by using online monitoring data, the operation and maintenance cost can be saved, and the method has extremely high engineering practical value.

Description

Technical Field

[0001] This invention belongs to the field of power equipment condition monitoring and artificial intelligence diagnosis technology, specifically involving a dual-domain BBO transformer coupling fault diagnosis method based on habitat modification mechanism. Background Technology

[0002] As the core hub of energy transmission and conversion in the power grid, the operational reliability of power transformers directly affects the safety and stability of the entire power system. With the rapid development of ultra-high voltage transmission technology, transformer capacity is constantly increasing, voltage levels are continuously rising, and their internal insulation structures are becoming increasingly complex. During long-term operation, transformers are subjected to the combined effects of electrical, thermal, and mechanical stresses, which can easily induce various types of latent faults. These faults typically exhibit significant multi-physics coupling characteristics, meaning that a fault is often accompanied by changes in chemical products (such as the decomposition of gases in oil) and electrical phenomena (such as partial discharge pulses and electromagnetic radiation). Therefore, how to accurately identify fault types from complex coupling characteristics has become a focus of attention in the power transmission and transformation industry.

[0003] Currently, transformer condition monitoring mainly relies on two independent technologies: dissolved gas analysis (DGA) and partial discharge (PD) detection. However, both technologies have their limitations in practical applications. On the one hand, DGA technology focuses on detecting the decomposition products (such as hydrogen and acetylene) of insulating oil and insulating paper under thermal or electrical faults. Although sensitive to cumulative thermal faults, it often shows a significant lag in reflecting sudden electrical faults due to the time required for oil and gas diffusion, and it is difficult to locate the fault source. On the other hand, PD detection technology based on ultra-high frequency (UHF) or ultrasound can capture instantaneous electrical pulses and is sensitive to insulation defects, but it is highly susceptible to interference from electromagnetic noise at the substation site, leading to a high false alarm rate. More importantly, existing diagnostic systems often treat chemical characteristics and electrical characteristics separately, forming "information silos." For example, when DGA data shows slight anomalies while the PD signal is strong, existing systems lack a mechanism to use the strong PD signal to "verify" or "correct" the interpretation of the weak DGA signal, resulting in frequent missed detections of complex composite faults such as "overheating accompanied by discharge."

[0004] To address these issues, researchers have introduced artificial intelligence algorithms such as Support Vector Machines (SVM) and Extreme Learning Machines (ELM) in recent years, and have utilized Genetic Algorithms (GA), Particle Swarm Optimization (PSO), or Biogeographical Optimization (BBO) algorithms to optimize classifier parameters. While these methods have improved diagnostic accuracy to some extent, the following technical problems remain to be solved: First, they lack prior guidance based on physical mechanisms. Traditional optimization algorithms (including standard BBO) are typically data-driven "black box" searches, with random migration and mutation operations, failing to utilize explicit physical laws such as "high-energy electric arcs inevitably produce acetylene" to guide the search direction. Second, they are prone to getting trapped in local optima. Transformer fault feature spaces are typically high-dimensional, nonlinear, and multi-peaked. With scarce samples, traditional algorithms are highly susceptible to converging to local extrema, resulting in poor generalization ability of the diagnostic model. Third, they cannot handle cross-domain coupling relationships. Existing algorithms cannot establish dynamic interactions between the chemical and electromagnetic domains during the optimization process, cannot simulate the ecological mechanisms of "successful species actively modifying their environment" in nature, and struggle to achieve true multimodal collaborative diagnosis. Summary of the Invention

[0005] To address the aforementioned shortcomings in existing technologies, the dual-domain BBO transformer coupling fault diagnosis method based on habitat modification mechanisms provided by this invention solves the problems of existing related methods, such as lack of prior guidance based on physical mechanisms, easy getting trapped in local optima, and inability to handle cross-domain coupling relationships, which in turn affect the fault diagnosis effect.

[0006] To achieve the aforementioned objectives, the technical solution adopted by this invention is: a dual-domain BBO transformer coupling fault diagnosis method based on habitat modification mechanisms, comprising the following steps:

[0007] S100. Based on the transformer's operating characteristic data, construct and initialize a dual-domain population, including a chemical domain population based on chemical characteristics and an electromagnetic domain population based on electrical characteristics.

[0008] S200. Based on the physical mechanism of transformer faults and expert experience database, a fault coupling matrix is ​​constructed to characterize the ecological impact coefficient of each fault type in the electromagnetic domain on each characteristic dimension of the chemical domain.

[0009] S300. Calculate the fitness values ​​of habitats in the current chemical domain population and electromagnetic domain population, and perform standard BBO migration operations based on the fitness values.

[0010] S400. In each iteration of the basic migration operation, determine whether there is an elite solution in the electromagnetic domain population that meets the preset conditions; if yes, activate the habitat modification mechanism and proceed to step S500; if no, return to step S300.

[0011] S500. Based on the elite solution and the fault coupling matrix, the habitat of the chemical domain population is modified, including search space distortion and dynamic reconstruction of feature dimensions.

[0012] S600: Calculate the fitness values ​​of the current electromagnetic domain population and chemical domain population, and determine the transformer fault type through weighted fusion.

[0013] Further, in S100, the feature vector of an individual in the chemical domain population is the main influencing factors extracted from transformer oil chromatographic data, including hydrogen concentration, methane concentration, ethane concentration, ethylene concentration, acetylene concentration, total hydrocarbon content and gas production rate, as well as the penalty factor and kernel parameters applied to the SVM classifier for diagnosing fault types.

[0014] The feature vectors of individuals in the electromagnetic field population are physical process statistical features extracted from the partial discharge UHF data, including maximum discharge quantity, average discharge quantity, pulse repetition rate, skewness, steepness, and phase window features, as well as penalty factors and kernel parameters for the SVM classifier used to diagnose fault types.

[0015] Furthermore, in S400, in each iteration, determining whether there is an elite solution in the electromagnetic domain population that satisfies a preset condition includes:

[0016] S401. Calculate the fitness value of all individuals in the electromagnetic field population;

[0017] S402. The maximum value among all fitness values ​​is taken as the optimal fitness value. And its corresponding fault type is determined by the SVM classifier;

[0018] S403. Determine the optimal fitness value. Is it greater than the preset fitness threshold? ;

[0019] If so, then there exists an elite solution in the electromagnetic domain population that meets the preset conditions, which activates the habitat modification mechanism, and thus S500.

[0020] If not, there is no elite solution in the electromagnetic domain population that meets the preset conditions, and no cross-domain transformation is performed in the current iteration round, maintaining the basic migration and mutation operations.

[0021] Furthermore, S500 includes the following sub-steps:

[0022] S501. Determine the fault type corresponding to the elite solution in the electromagnetic domain population;

[0023] S502. Based on the fault type, retrieve the corresponding row vector from the fault coupling matrix, which contains the ecological impact coefficient of the fault type on the characteristics of all chemical domains.

[0024] S503. Calculate the time decay factor for the current iteration number;

[0025] S504. Based on the time decay factor, the position of individuals in the chemical domain population is updated by calculating the nonlinear offset, thereby achieving search space distortion.

[0026] S505. Based on the retrieved ecological impact coefficient, dynamically adjust the variation rate of different characteristic dimensions of the chemical domain population to achieve dynamic reconstruction of characteristic dimensions.

[0027] Furthermore, in S504, the location update includes:

[0028] For the first in the chemical domain population For each individual instance, the standard BBO algorithm's migration operation is performed, and then the nonlinear offset is calculated. The nonlinear offset is superimposed onto the current position of the individual to be optimized in the chemical domain population to determine the updated position of the individual, which is expressed as:

[0029]

[0030]

[0031] In the formula, Indicates the first The first individual Dimensional features in Real-time location updates Indicates the first The first individual Dimensional features in Location at any given moment This represents the regular position update amount generated by standard BBO migration and mutation operations. Indicates the time decay factor. Indicates the first The first type of fault is related to the second type of fault. Ecological impact coefficient of dimensional characteristics, The ideal center to which the physical mechanism points. The Gaussian kernel width coefficient represents the range of influence of the gravitational field. This indicates the range of influence of the gravitational field.

[0032] Furthermore, the nonlinear offset is aligned with the direction of the ecological influence vector, and its strength decreases with the number of iterations.

[0033] Among them, the ecological impact vector Represented as:

[0034]

[0035] In the formula, This represents the confidence level of the elite solution. Represents the fault coupling matrix The row vector corresponding to the elite solution. This represents the index of the target fault type corresponding to the elite solution.

[0036] Furthermore, in S505, the dynamically adjusted mutation rate for:

[0037]

[0038] In the formula, Represents the basic rate of variation. Indicates the regulating factor. Indicates the first The first type of fault is related to the second type of fault. Ecological impact coefficient of dimensional characteristics.

[0039] Furthermore, S600 includes the following sub-steps:

[0040] S601. Calculate the fitness values ​​of elite solutions in the current electromagnetic domain population. Fitness values ​​of elite solutions in the chemical domain population ;

[0041] S602. Calculate the probability distribution diagrams of the current electromagnetic domain population and chemical domain population for each fault type prediction, respectively. and ;

[0042] S603, Based on fitness value and fitness value Dynamically calculate the corresponding weight coefficients and ;

[0043] S604. Based on the probability distribution diagram and and the corresponding weight calculation and Calculate the fault score for each fault type. The fault type corresponding to the highest fault score is taken as the transformer fault diagnosis result.

[0044] Furthermore, the S600 also includes:

[0045] According to the probability distribution diagram and Physical collision detection is performed using the Euclidean distance between them;

[0046] When Euclidean distance If so, the corresponding fault type is determined based on the fault score;

[0047] When Euclidean distance ,and and If all values ​​exceed the preset threshold, it is determined to be a heterogeneous fault, and a manual review alert is sent.

[0048] in, This indicates the safety threshold.

[0049] The beneficial effects of this invention are as follows:

[0050] (1) It breaks through the "information silo" of traditional single monitoring and realizes cross-domain collaborative diagnosis based on physical mechanism.

[0051] Existing technologies typically process oil chromatography (DGA) and partial discharge (PD) data in isolation, neglecting the physical-chemical coupling effects accompanying faults. This invention constructs a fault coupling matrix, transforming physical laws such as "arc discharge inevitably produces acetylene" and "partial discharge mainly breaks CH bonds" into mathematical "ecological impact coefficients." This makes the diagnostic process no longer a pure data fitting, but rather forces the algorithm to follow the physical mechanism of transformer faults. This mechanism utilizes the "strong real-time" nature of electromagnetic domain signals to actively guide the "cumulative" analysis of chemical domain signals, effectively solving the problems of delayed response of DGA to sudden electrical faults and susceptibility to false alarms in PD, significantly improving the diagnostic accuracy of complex coupled faults.

[0052] (3) The "habitat modification mechanism" overcomes the premature convergence of the algorithm and greatly improves the global optimization ability.

[0053] To address the problem of traditional BBO algorithms easily getting trapped in local optima due to the high-dimensionality, nonlinearity, and scarcity of samples in the transformer fault feature space, this invention innovatively introduces a search space distortion mechanism. When a high-confidence "elite solution" is found in the electromagnetic domain, a "gravitational funnel" is created in the search space of the chemical domain by calculating a nonlinear offset. This mechanism simulates the process of dominant species modifying their environment in nature, forcing the chemical domain population to escape inefficient random walks and rapidly shift towards the "ideal center" pointed to by the physical mechanism. This significantly avoids getting trapped in local optima while ensuring convergence speed, thus improving the model's generalization ability under conditions of few samples.

[0054] (3) Dynamic reconstruction of feature dimensions was achieved, effectively eliminating redundant interference and locking in key evidence.

[0055] To overcome the limitation of equal mutation rates across all feature dimensions in traditional algorithms, this invention introduces "terrain hardening" and "terrain loosening" strategies. Based on feedback from the fault coupling matrix, the mutation rate of each feature dimension is dynamically adjusted. For key features strongly correlated with the current suspected fault (such as acetylene under an arc fault), the algorithm automatically reduces its mutation rate (terrain hardening), locking in valid features and preventing random disturbances from destroying discovered evidence. For irrelevant features, a high mutation rate is maintained (terrain loosening), accelerating the escape from invalid search areas. This mechanism achieves a refined search that "discards false positives and retains true ones," significantly enhancing the robustness of the diagnostic system against on-site noise interference.

[0056] (4) A rigorous dual-domain physical conflict detection mechanism has been established, which reduces the false diagnosis rate and has the ability to warn of heterogeneous faults.

[0057] In the final decision-making stage, this invention does not simply output a mathematical probability average, but introduces physical conflict detection based on Euclidean distance. When the diagnostic results from the chemical and electromagnetic domains show a significant divergence (i.e., a physical-logical conflict, such as DGA indicating low-temperature overheating while PD indicates high-energy arcing) and both have high confidence levels, the system will not force an incorrect conclusion, but will instead classify it as a "heterogeneous fault" and trigger a manual review alarm. This "safety valve" design effectively avoids misjudgments caused by a single sensor failure or extreme composite faults, providing maintenance personnel with more objective and reliable decision support.

[0058] (5) It has extremely high engineering practical value and can significantly reduce operation and maintenance costs.

[0059] Based on the full-process diagnostic content of online monitoring data, this invention analyzes existing online oil chromatography monitoring devices and UHF partial discharge data for transformers, eliminating the need for power outages or cover removal inspections of the transformer under inspection. Through software-level algorithm upgrades, early detection and precise location of latent faults in transformers can be achieved, significantly reducing unplanned downtime and saving expensive manual inspections and offline testing costs, making it suitable for large-scale application in smart substations. Attached Figure Description

[0060] Figure 1 The flowchart of the dual-domain BBO transformer coupling fault diagnosis method based on habitat modification mechanism provided by the present invention is shown.

[0061] Figure 2 This is an overall architecture diagram of the dual-domain BBO transformer coupling fault diagnosis process based on habitat modification mechanism provided by the present invention.

[0062] Figure 3 This is a logical diagram of the dual-domain population interaction and fault coupling matrix provided by the present invention.

[0063] Figure 4 A schematic diagram illustrating the principle of search space distortion under the habitat modification mechanism provided by this invention.

[0064] Figure 5 This is a schematic diagram of dynamic dimension reconstruction based on ecological influence provided by the present invention.

[0065] Figure 6 The fault diagnosis result fusion decision strategy diagram provided by the present invention. Detailed Implementation

[0066] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0067] This invention provides a dual-domain BBO transformer coupled fault diagnosis method based on habitat modification mechanism. Through refined feature definition, physical mechanism-based matrix construction, visualized spatial distortion and dimension reconstruction mechanism, and rigorous fusion decision logic, it achieves deep perception of transformer faults.

[0068] refer to Figures 1-2 The dual-domain BBO transformer coupling fault diagnosis method based on habitat modification mechanisms includes the following steps:

[0069] S100. Based on the transformer's operating characteristic data, construct and initialize a dual-domain population, including a chemical domain population based on chemical characteristics and an electromagnetic domain population based on electrical characteristics.

[0070] S200. Based on the physical mechanism of transformer faults and expert experience database, a fault coupling matrix is ​​constructed to characterize the ecological impact coefficient of each fault type in the electromagnetic domain on each characteristic dimension of the chemical domain.

[0071] S300. Calculate the fitness values ​​of habitats in the current chemical domain population and electromagnetic domain population, and perform standard BBO migration operations based on the fitness values.

[0072] S400. In each iteration of the basic migration operation, determine whether there is an elite solution in the electromagnetic domain population that meets the preset conditions; if yes, activate the habitat modification mechanism and proceed to step S500; if no, return to step S300.

[0073] S500. Based on the elite solution and the fault coupling matrix, the habitat of the chemical domain population is modified, including search space distortion and dynamic reconstruction of feature dimensions.

[0074] S600: Calculate the fitness values ​​of the current electromagnetic domain population and chemical domain population, and determine the transformer fault type through weighted fusion.

[0075] In S100 of this embodiment, during the construction of the dual-domain population:

[0076] An online oil chromatography (DGA) monitoring device installed on the transformer, with a sampling frequency set to once every 4 hours, is used to separate and detect dissolved gases in the oil; and an ultra-high voltage (UHF) partial discharge sensor installed at the oil drain valve on the transformer tank wall, with an operating frequency band of 300MHz-1.5GHz and a sampling rate of not less than 20MS / s, is used to capture transient electromagnetic wave signals.

[0077] For the collected oil chromatography data, sliding window filtering was used to remove sampling noise and normalize it to the [0,1] interval; for the UHF signal, wavelet threshold denoising technology was used to remove white noise interference and convert it into a PRPD (phase-resolved partial discharge) spectrum.

[0078] For the chemical domain population, the individual feature vectors are the main influencing factors extracted from transformer oil chromatographic data, including hydrogen concentration, methane concentration, ethane concentration, ethylene concentration, acetylene concentration, total hydrocarbon content, and gas production rate, as well as the penalty factor and kernel parameters applied to the SVM classifier for diagnosing fault types. Specifically, the individual feature vectors in the chemical domain population are recorded as a sequence. In this embodiment, the normalized gas concentration value is used, where... Hydrogen ( ), methane ( ), Ethane ( ), ethylene ( ), Acetylene ( ), Total hydrocarbon content, This refers to the gas production rate. Additionally, it includes the optimization parameters for the SVM classifier. (Penalty Factor C) and (Kernel parameter ganmma), from which this sequence is obtained. This constitutes the basic optimization space of the chemical domain.

[0079] For the electromagnetic domain population, the feature vector of each individual is a statistical feature of the physical process extracted from the UHF partial discharge data, including maximum discharge quantity, average discharge quantity, pulse repetition rate, skewness, steepness, and phase window features, as well as the penalty factor and kernel parameters for the SVM classifier used to diagnose fault types. Specifically, the number of individual features in the electromagnetic domain population is recorded as a sequence. . Maximum discharge amount , Average discharge quantity , pulse repetition rate , Skewness , For steepness , Phase window features Similarly, the feature vector also contains the optimization parameters of the corresponding SVM classifier. (Penalty Factor) and (Kernel parameter ganmma).

[0080] In this embodiment, an “individual” in the electromagnetic and chemical domain populations can be understood as a virtual probe “carrying feature vectors and classifier parameters”. This probe wanders in the search space of the population to find the parameter combination that most accurately explains the current transformer state.

[0081] In this embodiment, the SVM classifier acts as the executor of fault diagnosis: it is responsible for mapping the complex input feature data (chemical or electromagnetic data) to specific transformer fault types (such as arc discharge, overheating, etc.). As the fitness evaluation function of the BBO algorithm, each "individual" (i.e., a combination of features and parameters) generated by the BBO algorithm needs to determine its fitness level. By substituting the feature vector parameters of the individual into the SVM classifier for training and testing, the accuracy or confidence of the SVM classification (i.e., the habitat fitness value) directly becomes the fitness value of that individual. If the SVM classification is effective, the individual will be retained and reproduce in the population; if the classification is ineffective, the individual will be eliminated. Since this invention employs a "dual-domain" approach, there are actually two independent and parallel SVM classifier models, processing data in the chemical and electromagnetic domains respectively. The individual feature vector contains the SVM penalty factor and kernel parameters; therefore, this method not only selects the gas / electric signal features that best represent the fault (feature selection) but also optimizes the model parameters.

[0082] In S200 of this embodiment, a fault coupling matrix is ​​constructed based on the physical mechanism of transformer faults and an expert experience base. Let K be a matrix, where This represents the identifiable fault types in the electromagnetic domain (such as high-energy arcs, floating potential discharges, surface discharges, etc.). The gaseous characteristic dimension representing the chemical domain.

[0083] Fault Coupling Matrix Matrix elements Defined the first The first type of electrical fault affects the first The "ecological impact" of chemical characteristics. Setting matrix elements. The value when the first Electrical faults are physically accompanied by the first When a gas is generated, set A positive value indicates ecological attraction; when the first... This type of electrical fault does not physically produce the first When using a gas, set A negative value indicates ecological exclusion.

[0084] For example, when Corresponding to "high-energy electric arc" and When dealing with "acetylene", set This indicates a strong positive attraction; when Corresponding to "partial discharge" and When dealing with "acetylene", set The symbol represents strong negative suppression. This setting ensures that the algorithm follows the physical law that "electric arcs always produce acetylene" during the search.

[0085] For example, a fault coupling matrix is ​​established by combining "expert scoring" with "historical big data correlation analysis". And the matrix dimension is ( Electrical faults (When a key gas is present);

[0086] Row vector 1 (corresponding to "High-energy arc arcing"): 0.95 corresponds to acetylene ( ), indicating extremely strong attraction; -0.5 corresponds to methane and ethane, indicating relative inhibition (because the main products of high-temperature cracking are not them).

[0087] Row vector 2 (corresponding to "partial discharge PD"): 0.85 corresponds to hydrogen ( The value of -0.9 indicates that PD primarily breaks the CH bond; -0.9 corresponds to acetylene, meaning that PD does not have enough energy to produce acetylene, and searching for acetylene would be an incorrect direction.

[0088] Row vector 3 (corresponding to "Overheating"): It primarily attracts hydrocarbon gases such as methane, ethane, and ethylene.

[0089] In S300 of this embodiment, the migration operation of the standard BBO algorithm based on the fitness value in the chemical domain population and the electromagnetic domain population is a routine operation of the BBO algorithm, and it is not limited in this embodiment.

[0090] In S400 of this embodiment, in each iteration, determining whether there is an elite solution in the electromagnetic domain population that satisfies a preset condition includes:

[0091] S401. Calculate the fitness value of all individuals in the electromagnetic field population;

[0092] S402. The maximum value among all fitness values ​​is taken as the optimal fitness value. And its corresponding fault type is determined by the SVM classifier;

[0093] Specifically, the optimal fitness value For the "elite solution" in the current iteration round, if the SVM classifier of this elite individual determines that the current situation is "high-energy arc", then "high-energy arc" is the fault type corresponding to the optimal fitness value. This invention selects the diagnostic model (individual) with the highest confidence (optimal fitness) and adopts its diagnostic conclusion (fault type) as a guiding signal.

[0094] S403. Determine the optimal fitness value. Is it greater than the preset fitness threshold? ;

[0095] If so, then there exists an elite solution in the electromagnetic domain population that meets the preset conditions, which activates the habitat modification mechanism, and thus S500.

[0096] If not, there is no elite solution in the electromagnetic domain population that meets the preset conditions, and no cross-domain transformation is performed in the current iteration round, maintaining the basic migration and mutation operations.

[0097] For example, in the above process, the fitness of elite solutions in the electromagnetic domain population Exceeding the fitness threshold When the confidence level is 0.85, it is determined that a high-confidence fault clue has been found, thereby activating the "habitat modification mechanism".

[0098] In this embodiment, S500, as... Figure 3 As shown, it includes the following steps:

[0099] S501. Determine the fault type corresponding to the elite solution in the electromagnetic domain population;

[0100] For example, the fault type pointed to by the elite solution in the electromagnetic domain population is determined. (For example: =High-energy electric arc).

[0101] S502. Based on the fault type, retrieve the corresponding row vector from the fault coupling matrix, which contains the ecological impact coefficient of the fault type on the characteristics of all chemical domains.

[0102] The retrieved row vectors contained faults. For all "Ecological impact coefficient" of each chemical characteristic For example, if If it is an electric arc, then for acetylene... A positive value (attraction) for methane It may be a negative value (rejection).

[0103] S503. Calculate the time decay factor for the current iteration number;

[0104] Specifically, to prevent oscillations in the later stages of the algorithm, the time decay factor at the current iteration number is calculated. The initial value is set to 0.5, which simulates the process of weathering in nature as the transformation intensity gradually weakens with iteration.

[0105] S504. Based on the time decay factor, the position of individuals in the chemical domain population is updated by calculating the nonlinear offset, thereby achieving search space distortion.

[0106] Specifically, location updates include:

[0107] For the first in the chemical domain population For each individual instance, the standard BBO algorithm's migration operation is performed, and then the nonlinear offset is calculated. The nonlinear offset is superimposed onto the current position of the individual to be optimized in the chemical domain population to determine the updated position of the individual, which is expressed as:

[0108]

[0109]

[0110] In the formula, Indicates the first The first individual Dimensional features in Real-time location updates Indicates the first The first individual Dimensional features in Location at any given moment This represents the regular position update amount generated by standard BBO migration and mutation operations. Indicates the time decay factor. Indicates the first The first type of fault is related to the second type of fault. Ecological impact coefficient of dimensional characteristics, The ideal center to which the physical mechanism points. The Gaussian kernel width coefficient represents the range of influence of the gravitational field. This indicates the range of influence of the gravitational field.

[0111] In this embodiment, the nonlinear offset is aligned with the direction of the ecological influence vector, and its strength decreases with the number of iterations.

[0112] Among them, the ecological impact vector is defined. It is A dimensional vector, each element of which Corresponding chemical domain number The forces acting on each feature dimension. First, a lookup table is performed on the fault coupling matrix, based on the target fault type index determined by the elite solution. From the fault coupling matrix Extract the corresponding row vector This row vector describes the physical impact pattern of that particular fault on all chemical features. It is then "weighted" using the confidence probabilities of the elite solutions. As a scalar, it is multiplied by the extracted row vector. This means that the more certain the elite solution is, the stronger its "ecological impact" will be.

[0113] Based on this, the ecological impact vector Represented as:

[0114]

[0115] Expand it as follows:

[0116]

[0117] In the formula, This represents the confidence level of the elite solution. Represents the fault coupling matrix The row vector corresponding to the elite solution. This represents the index of the target fault type corresponding to the elite solution. Indicates the first The target fault type affects the first [missing information] in the chemical domain. Ecological impact coefficient of dimensional characteristics.

[0118] The physical meaning of the above process can be visualized as follows: Figure 4 As shown, Figure 4 This vividly demonstrates how habitat modification mechanisms alter the topology of the chemical domain search space. Figure 4The search space consists of two states: State A (before modification) and State B (after modification). In State A, the terrain of the search space is flat and isotropic. At this time, the particles (solutions) in the chemical domain population search according to a random probability distribution, which easily leads to wandering in the local optimum region, resulting in a slow convergence speed.

[0119] Under the influence of the aforementioned ecological impact vector calculation formula, individuals in State A are subjected to a powerful "ecological gravitational force," forcing them to deviate from their original random trajectories and slide towards the center of the "gravitational funnel." This simulates the process by which successful species (electrical elites) modify the environment, guiding subsequent species (chemical solutions) to quickly find resources for survival.

[0120] Therefore, through the above-mentioned position update process, the present invention superimposes the nonlinear offset term onto the current position coordinates of the individual to be optimized in the chemical domain population, causing the individual to shift towards the target center pointed to by the physical mechanism in the search space, forming a "gravitational funnel" effect, thereby forcing the population to leave the local optimal region.

[0121] S505. Based on the retrieved ecological impact coefficient, dynamically adjust the variation rate of different characteristic dimensions of the chemical domain population to achieve dynamic reconstruction of characteristic dimensions.

[0122] Specifically, the variation rate of each feature dimension in the chemical domain population is dynamically adjusted according to the absolute value of the ecological influence coefficient. For key feature dimensions with large absolute values ​​of influence coefficients, their variation rate is reduced to achieve landform hardening and lock in effective features. For irrelevant feature dimensions with small absolute values ​​of influence coefficients, their variation rate is maintained or increased to achieve landform loosening and accelerate the population's escape from the ineffective search area.

[0123] In this embodiment, the dynamically adjusted variability rate for:

[0124]

[0125] In the formula, Represents the basic rate of variation. Indicates the regulating factor. Indicates the first The first type of fault is related to the second type of fault. Ecological impact coefficient of dimensional characteristics.

[0126] like Figure 5 As shown, two sets of histograms visually illustrate the dynamic adjustment strategy of the mutation rate. The horizontal axis represents the various characteristic dimensions of the chemical domain (corresponding to the sequence). The vertical axis represents the current mutation rate / search step size for that dimension.

[0127] Figure 5 The histogram in (a) illustrates the "ordinary state." At this point, the various feature dimensions (such as...) show the "ordinary state." The histograms of the mutation rates of the two algorithms are highly similar, which indicates that the algorithm is in a standard BBO search mode, and the exploration of all feature dimensions is balanced and random.

[0128] Figure 5 The histogram in (b) illustrates the "Controlled State." When the fault coupling matrix... Once activated, guided by the electromagnetic domain, key feature dimensions strongly correlated with the current fault undergo significant changes.

[0129] exist Figure 5 In this process, shorter pillars (reduced mutation rate) mean a significant reduction in the search step size for that dimension, and the algorithm shifts from "coarse search" to "fine development" in that dimension. This process, known as "terrain hardening," aims to lock in key evidence that has been discovered and prevent it from being destroyed by random perturbations in subsequent iterations.

[0130] Conversely, for features unrelated to the current fault (such as...) (For the corresponding methane), the column height remains constant or slightly increases. This means that this dimension remains "topographically loose," allowing particles to quickly jump out of the region with larger step sizes, avoiding invalid calculations.

[0131] like Figure 6 As shown, S600 in this embodiment includes the following sub-steps:

[0132] S601. Calculate the fitness values ​​of elite solutions in the current electromagnetic domain population. Fitness values ​​of elite solutions in the chemical domain population .

[0133] S602. Calculate the probability distribution diagrams of the current electromagnetic domain population and chemical domain population for each fault type prediction, respectively. and .

[0134] S603, Based on fitness value and fitness value Dynamically calculate the corresponding weight coefficients and ;

[0135] Specifically, in this embodiment, the fitness value is used as the basis for adaptively determining the weights, and the population with higher fitness has a larger weight coefficient.

[0136] S604. Based on the probability distribution diagram and and the corresponding weight calculation and Calculate the fault score for each fault type. The fault type corresponding to the highest fault score is taken as the transformer fault diagnosis result.

[0137] In this embodiment, S600 further includes:

[0138] According to the probability distribution diagram and Physical collision detection is performed using the Euclidean distance between them;

[0139] When Euclidean distance If so, the corresponding fault type is determined based on the fault score;

[0140] When Euclidean distance ,and and If all values ​​exceed the preset threshold, it is determined to be a heterogeneous fault, and a manual review alert is sent.

[0141] in, This indicates the safety threshold.

[0142] In this embodiment, the extreme case of inconsistent results between the two domains is handled through the aforementioned physical conflict detection process. When the chemical domain result points to "low-temperature overheating" (primarily characterized by ethane), while the electromagnetic domain result points to "high-energy arc" (primarily characterized by a broadband UHF signal), and both... and When all values ​​exceed a preset threshold, the system determines a physical logic conflict.

[0143] At this point, the system will not force the output of a weighted average, but will instead trigger a "Heterogeneous Fault Warning". Figure 6 The path indicates that the manual intervention port is being used, suggesting to maintenance personnel that the transformer may have a complex composite fault or that the sensor itself may be malfunctioning.

[0144] In this embodiment, to improve the robustness of fault diagnosis, time series analysis can also be introduced, that is, retaining the previous... The diagnostic results at any given time, if the current time's... In the event of a sudden change, it is necessary to smooth it out by combining it with historical trends to avoid misdiagnosis caused by instantaneous interference from sensors.

[0145] The dual-domain BBO transformer coupling fault diagnosis method based on the habitat modification mechanism provided by this invention achieves deep perception of transformer faults through refined feature sequence definition, physical mechanism-based matrix construction, visualized spatial distortion and dimensional reconstruction mechanism, and rigorous fusion decision logic. This method eliminates the need for power outage and cover removal inspection of the transformer under inspection; it only requires online monitoring data to achieve high-precision condition assessment, saving operation and maintenance costs and possessing extremely high engineering practical value.

[0146] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

[0147] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A dual-domain BBO transformer coupling fault diagnosis method based on habitat modification mechanisms, characterized in that, Includes the following steps: S100. Based on the transformer's operating characteristic data, construct and initialize a dual-domain population, including a chemical domain population based on chemical characteristics and an electromagnetic domain population based on electrical characteristics. S200. Based on the physical mechanism of transformer faults and expert experience database, a fault coupling matrix is ​​constructed to characterize the ecological impact coefficient of each fault type in the electromagnetic domain on each characteristic dimension of the chemical domain. S300. Calculate the fitness values ​​of habitats in the current chemical domain population and electromagnetic domain population, and perform standard BBO migration operations based on the fitness values. S400. In each iteration of the basic migration operation, determine whether there is an elite solution in the electromagnetic field population that satisfies the preset conditions. If yes, activate the habitat modification mechanism and proceed to step S500; otherwise, return to step S300. S500. Based on the elite solution and the fault coupling matrix, the habitat of the chemical domain population is modified, including search space distortion and dynamic reconstruction of feature dimensions. S600: Calculate the fitness values ​​of the current electromagnetic domain population and chemical domain population, and determine the transformer fault type through weighted fusion.

2. The dual-domain BBO transformer coupling fault diagnosis method based on habitat modification mechanism according to claim 1, characterized in that, In S100, the feature vector of an individual in the chemical domain population is the main influencing factors extracted from transformer oil chromatography data, including hydrogen concentration, methane concentration, ethane concentration, ethylene concentration, acetylene concentration, total hydrocarbon content and gas production rate, as well as the penalty factor and kernel parameters applied to the SVM classifier for diagnosing fault types. The feature vectors of individuals in the electromagnetic field population are physical process statistical features extracted from the partial discharge UHF data, including maximum discharge quantity, average discharge quantity, pulse repetition rate, skewness, steepness, and phase window features, as well as penalty factors and kernel parameters for the SVM classifier used to diagnose fault types.

3. The dual-domain BBO transformer coupling fault diagnosis method based on habitat modification mechanism according to claim 1, characterized in that, In step S400, in each iteration, determining whether there is an elite solution in the electromagnetic domain population that satisfies preset conditions includes: S401. Calculate the fitness value of all individuals in the electromagnetic field population; S402. The maximum value among all fitness values ​​is taken as the optimal fitness value. And its corresponding fault type is determined by the SVM classifier; S403. Determine the optimal fitness value. Is it greater than the preset fitness threshold? ; If so, then there exists an elite solution in the electromagnetic domain population that meets the preset conditions, which activates the habitat modification mechanism, and thus S500. If not, there is no elite solution in the electromagnetic domain population that meets the preset conditions, and no cross-domain transformation is performed in the current iteration round, maintaining the basic migration and mutation operations.

4. The dual-domain BBO transformer coupling fault diagnosis method based on habitat modification mechanism according to claim 3, characterized in that, The S500 includes the following steps: S501. Determine the fault type corresponding to the elite solution in the electromagnetic domain population; S502. Based on the fault type, retrieve the corresponding row vector from the fault coupling matrix, which contains the ecological impact coefficient of the fault type on the characteristics of all chemical domains. S503. Calculate the time decay factor for the current iteration number; S504. Based on the time decay factor, the position of individuals in the chemical domain population is updated by calculating the nonlinear offset, thereby achieving search space distortion. S505. Based on the retrieved ecological impact coefficient, dynamically adjust the variation rate of different characteristic dimensions of the chemical domain population to achieve dynamic reconstruction of characteristic dimensions.

5. The dual-domain BBO transformer coupling fault diagnosis method based on habitat modification mechanism according to claim 4, characterized in that, In step S504, the location update includes: For the first in the chemical domain population For each individual instance, the standard BBO algorithm's migration operation is performed, and then the nonlinear offset is calculated. The nonlinear offset is superimposed onto the current position of the individual to be optimized in the chemical domain population to determine the updated position of the individual, which is expressed as: In the formula, Indicates the first The first individual Dimensional features in Real-time location updates Indicates the first The first individual Dimensional features in Location at any given moment This represents the regular position update amount generated by standard BBO migration and mutation operations. Indicates the time decay factor. Indicates the first The first type of fault is related to the second type of fault. Ecological impact coefficient of dimensional characteristics, The ideal center to which the physical mechanism points. The Gaussian kernel width coefficient represents the range of influence of the gravitational field. This indicates the range of influence of the gravitational field.

6. The dual-domain BBO transformer coupling fault diagnosis method based on habitat modification mechanism according to claim 5, characterized in that, The nonlinear offset is aligned with the direction of the ecological influence vector, and its strength decreases with the number of iterations. Among them, the ecological impact vector Represented as: In the formula, This represents the confidence level of the elite solution. Represents the fault coupling matrix The row vector corresponding to the elite solution. This represents the index of the target fault type corresponding to the elite solution.

7. The dual-domain BBO transformer coupling fault diagnosis method based on habitat modification mechanism according to claim 4, characterized in that, In S505, the dynamically adjusted variation rate for: In the formula, Represents the basic rate of variation. Indicates the regulating factor. Indicates the first The first type of fault is related to the second type of fault. Ecological impact coefficient of dimensional characteristics.

8. The dual-domain BBO transformer coupling fault diagnosis method based on habitat modification mechanism according to claim 1, characterized in that, The S600 includes the following steps: S601. Calculate the fitness values ​​of elite solutions in the current electromagnetic domain population. Fitness values ​​of elite solutions in the chemical domain population ; S602. Calculate the probability distribution diagrams of the current electromagnetic domain population and chemical domain population for each fault type prediction, respectively. and ; S603, Based on fitness value and fitness value Dynamically calculate the corresponding weight coefficients and ; S604. Based on the probability distribution diagram and and the corresponding weight calculation and Calculate the fault score for each fault type. The fault type corresponding to the highest fault score is taken as the transformer fault diagnosis result.

9. The dual-domain BBO transformer coupling fault diagnosis method based on habitat modification mechanism according to claim 8, characterized in that, The S600 also includes: According to the probability distribution diagram and Physical collision detection is performed using the Euclidean distance between them; When Euclidean distance If so, the corresponding fault type is determined based on the fault score; When Euclidean distance ,and and If all values ​​exceed the preset threshold, it is determined to be a heterogeneous fault, and a manual review alert is sent. in, This indicates the safety threshold.

Images