Intelligent analysis method and system for vibration characteristics of main transformer

Abstract

The invention relates to the technical field of power equipment state monitoring, and discloses a main transformer vibration characteristic intelligent analysis method and system, and the method comprises the steps: collecting vibration and current signals of a multi-load level in a health state, carrying out the decoupling of a vibration source, calculating a transfer function, and constructing a time-varying transfer function tensor representing the health state; during online monitoring, real-time signals are obtained, and a real-time transfer function matrix is calculated; interpolating the tensor according to a real-time load to obtain a health reference transfer function matrix matched with the current working condition; calculating a difference value between the real-time matrix and the health reference matrix, and obtaining a differential response field capable of stripping the influence of the working condition; and finally, on the basis of the differential response field, establishing and solving a physical inverse problem to realize reverse inversion and positioning of the abnormal state in the main transformer. According to the method, the interference of working condition change on vibration analysis is effectively solved, the decoupling of a fault source and the accurate positioning of internal abnormity are realized, and the accuracy and reliability of diagnosis are improved.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring technology, specifically to an intelligent analysis method and system for main transformer vibration characteristics. Background Technology

[0002] The main transformer (MT) is the core equipment in a power system for voltage transformation and power transmission, and its operational stability and reliability directly affect the safety of the entire power grid. During long-term operation, the MMT is subjected to a combination of electrical, thermal, and mechanical factors, which can cause changes in the mechanical structure of key components such as the core, windings, and supports. These changes can include winding deformation, core loosening, and clamp loosening. These mechanical faults are major hidden dangers leading to unexpected outages and even serious accidents. Therefore, effective online condition monitoring and fault diagnosis of the MMT are of paramount importance.

[0003] Vibration analysis, as a non-invasive online monitoring method, can reflect the mechanical state of internal components of a main transformer in real time and dynamically, and has become an important research direction in the field of condition monitoring. Currently, main transformer vibration analysis technology usually collects vibration signals by placing sensors on the tank wall, and compares the characteristics of the collected signals with a preset reference signal under healthy conditions. When the difference between the real-time signal and the reference signal exceeds a predetermined threshold, it is determined that there may be an internal structural abnormality.

[0004] However, in engineering practice, it has been found that this traditional analysis method has several inherent technical limitations. First, the vibration measured on the transformer tank wall is a complex superposition result of the combined effects of multiple physical sources within it. This includes vibrations caused by the magnetostrictive effect of the core, as well as electromagnetic force vibrations generated on the windings by the interaction of load current and leakage magnetic field. These two vibration sources have different physical mechanisms and different vibration characteristics. Existing technologies often directly analyze this mixed vibration signal as a whole, making it difficult to effectively distinguish the source of vibration anomalies. Consequently, when vibration changes, it is impossible to clearly determine whether the problem lies in the core or the windings, resulting in ambiguous diagnostic results.

[0005] More importantly, the intensity of the aforementioned vibration sources, especially the electromagnetic force vibration of the windings, is closely related to the square of the load current. Therefore, the overall vibration state of the main transformer is extremely sensitive to changes in operating conditions (especially load levels). This poses a severe challenge to diagnostic methods based on benchmark comparison. Traditional methods typically use vibration data collected under a specific operating condition (such as rated load) as a fixed health benchmark. When the main transformer deviates from this calibrated operating condition in actual operation, even if its internal structure is completely normal, its vibration signal will naturally deviate from this fixed benchmark due to load changes, thus triggering a large number of false alarms. Conversely, some weak vibration changes caused by early structural anomalies may also be masked by drastic changes in operating conditions, resulting in missed fault detection. This static and fixed benchmark lacks the ability to adapt to changes in operating conditions, severely limiting the accuracy and reliability of diagnosis.

[0006] Furthermore, even if existing technologies can detect vibration anomalies, they mostly remain at the macroscopic level of "abnormal condition." They cannot provide effective location information for the specific spatial location of the anomaly source inside the transformer, such as which phase winding is deformed or which area of ​​the core clamps is loose, thus failing to provide precise guidance for subsequent maintenance work.

[0007] Therefore, how to effectively separate the coupling effects of multi-source vibration, eliminate the interference of changes in operating conditions on diagnosis, and achieve accurate location of internal abnormal sources are the technical challenges that urgently need to be solved in the field of main transformer vibration analysis. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides an intelligent analysis method and system for the vibration characteristics of main transformers. This solves the problem that existing main transformer vibration analysis methods are unable to effectively separate the combined effects of multi-source vibration coupling and operating condition changes, resulting in insufficient accuracy in diagnosing internal structural anomalies and in terms of spatial positioning capabilities.

[0009] To achieve the above objectives, the present invention provides the following technical solution: an intelligent analysis method and system for the vibration characteristics of a main transformer, comprising: To address the aforementioned technical problems, the first aspect of this invention provides an intelligent analysis method for the vibration characteristics of a main transformer.

[0010] This method first calibrates the health benchmark when the main transformer is in a preset healthy state. During this stage, vibration signals from the transformer's tank wall vibration sensor array and current signals from the load side are simultaneously acquired when the transformer operates under multiple different load levels. The acquired vibration signals are further decoupled to separate the components contributed by different physical sources, specifically: the core vibration component related to the core magnetostriction effect, and the winding vibration component related to the electromagnetic force generated by the load current. For the two types of decoupled vibration components, the transfer function between any two vibration sensors is calculated at each load level, forming the transfer function matrix for that operating condition. Finally, the transfer function matrices obtained from all load levels are combined to construct a time-varying transfer function tensor that characterizes the structural response characteristics of the main transformer under healthy conditions and their dynamic changes with load.

[0011] During online monitoring of the main transformer, this method acquires vibration and load current signals in real time and uses the same technique as in the calibration phase to calculate the real-time transfer function matrix. Simultaneously, based on the real-time load information, interpolation is performed along the load level dimension in the constructed time-varying transfer function tensor to obtain a health baseline transfer function matrix that precisely corresponds to the current real-time operating conditions.

[0012] Subsequently, a differential response field is obtained by calculating the difference between the real-time transfer function matrix and the health baseline transfer function matrix. This differential response field eliminates the influence of changes in normal operating conditions, and its changes can more purely reflect the changes in the internal structural state of the main transformer.

[0013] Finally, this method transforms the fault diagnosis problem into an inverse physical problem. It is based on a forward physical model describing the mapping relationship between the internal structural disturbances of the main transformer and the differential response field generated at the tank wall. Using the actually calculated differential response field as observation data, the location, properties, or intensity of the internal structural disturbance source causing the change in the response field is solved through an inverse disturbance source inversion algorithm, thereby realizing the determination of the abnormal state inside the main transformer.

[0014] In one specific implementation, the step of decoupling the vibration signal into core vibration components and winding vibration components is achieved through the following operations: using a comb filter or narrowband filter bank with a center frequency twice the power grid frequency and its harmonics, a signal with a clear periodicity is extracted from the original vibration signal as the core vibration component; simultaneously, using the square of the load current signal as a reference input signal, an adaptive filtering algorithm is used to extract a signal portion from the original vibration signal that is highly correlated with the reference input signal as the winding vibration component.

[0015] In one specific implementation, the differential response field specifically includes a core path differential response field and a winding path differential response field. The core path differential response field is calculated from the transfer function matrix related to the core vibration component; the winding path differential response field is calculated from the transfer function matrix related to the winding vibration component. During the inverse disturbance source inversion, the contribution of these two differential response fields in the solution process can be used to further determine whether the abnormal state is related to the core or the winding.

[0016] In one specific implementation, any two vibration sensors and Transfer function between Calculated using the following formula: ; in, For frequency; For sensors and sensors Between in frequency The transfer function at the location; For sensors The vibration components at frequency The self-power spectral density at the location; For sensors and sensors The vibration components at frequency The cross-power spectral density at that location.

[0017] In one specific implementation, the solution process for the inverse perturbation source inversion employs the Tikhonov regularization method, the goal of which is to find a solution that minimizes the objective function. Optimal internal structure perturbation vector : ; in, Let the internal structure perturbation vector be the optimization variable; The observed data vector is composed of the differential response field; For the operators of the aforementioned positive physical model; For regularization parameters; For regularization operators; It is the square of the L2 norm of the vector.

[0018] A second aspect of the present invention provides an intelligent analysis system for the vibration characteristics of a main transformer, comprising: The data acquisition module is designed to synchronously acquire the vibration signal and load current signal of the main transformer.

[0019] A health benchmark construction module, which is connected to the data acquisition module, is used to construct a time-varying transfer function tensor characterizing the health state of the main transformer based on the vibration signal and load current signal covering multiple load levels under a preset health state.

[0020] The online diagnostic module, connected to the data acquisition module and the health benchmark construction module, is configured to: acquire real-time vibration signals and load current signals, and calculate the real-time transfer function matrix; calculate the health benchmark transfer function matrix corresponding to the real-time transfer function matrix based on the time-varying transfer function tensor and the real-time load; and calculate the difference between the real-time transfer function matrix and the health benchmark transfer function matrix to obtain the differential response field.

[0021] The inverse inversion module, which is connected to the online diagnostic module, is used to determine the abnormal state inside the main transformer by inverting the inverse disturbance source based on the differential response field.

[0022] This invention provides an intelligent analysis method and system for the vibration characteristics of a main transformer. It has the following beneficial effects: 1. This invention decouples the mixed vibration signal collected from the main transformer box wall into independent core vibration components and winding vibration components, and constructs transfer function models based on these components for comparative analysis, thereby achieving independent monitoring of different physical source paths. This method allows diagnostic results to be initially attributed to structural paths related to the core or windings, significantly improving the accuracy of fault diagnosis and the clarity of fault attribute judgment compared to directly analyzing mixed signals.

[0023] 2. This invention effectively overcomes the false alarm or missed alarm problems caused by changes in operating conditions in traditional fixed-reference methods by constructing a time-varying transfer function tensor covering multiple typical load levels of the main transformer and generating a dynamic health benchmark based on real-time load interpolation during online monitoring. This method can accurately isolate the influence of normal operating factors such as load fluctuations on vibration characteristics, thereby enhancing the sensitivity to early, subtle fault characteristics caused by structural anomalies and improving the overall reliability of diagnosis.

[0024] 3. This invention transforms the diagnostic problem into an inverse problem based on a physical model. Using the differential response field as observation data, it performs inverse inversion to solve for the structural disturbance sources within the main transformer, achieving source tracing from external vibration response to internal anomaly sources. This method not only determines the existence of faults but also spatially locates the anomaly sources and presents the results in a visual form, providing more intuitive and effective information support for operation and maintenance personnel to conduct condition assessments and make accurate maintenance decisions. Attached Figure Description

[0025] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system architecture diagram of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Example: Please see the appendix Figure 1 This invention provides an intelligent analysis method and system for the vibration characteristics of a main transformer, comprising the following steps: S1: Under the preset health state, construct a time-varying transfer function tensor characterizing the health state of the main transformer based on vibration signals and load current signals covering multiple load levels; This embodiment details the process of constructing a time-varying transfer function tensor representing the health state of the main transformer when it is in a preset health state. This process aims to establish a high-dimensional, dynamic health benchmark capable of adapting to changes in operating conditions, providing an accurate reference system for subsequent online diagnostics.

[0028] To comprehensively capture the structural response characteristics of the main transformer under different operating conditions, cross-condition data acquisition is first required. Preferably, multiple representative stable load levels covering the main transformer's range from no-load to rated load are selected, denoted as the set. At each load level Under these conditions, vibration signals and load current signals are collected synchronously and strictly through a vibration sensor array deployed on the surface of the main transformer box and current transformers installed on the line. The vibration sensor array is preferably arranged in an M×N grid, with a total of P sensors, which are monitored at specific times. The collected signal is denoted as ,in The synchronously acquired load current signal is denoted as... Strict synchronization here is a prerequisite for calculating the phase relationship between the signals, and is crucial for ensuring the accuracy of subsequent transfer function calculations.

[0029] Before proceeding with further analysis, the acquired raw signals are preprocessed. This preprocessing step includes processing all vibration signals. and current signal The purpose of applying a bandpass filter is to filter out DC bias and high-frequency random noise that are irrelevant to the analysis, while retaining the effective information bandwidth.

[0030] Considering that the vibration response of the main transformer tank wall is the result of the combined action of multiple internal physical sources, this embodiment introduces a multi-source vibration signal decoupling step to achieve accurate fault attribution. The core of this step lies in decoupling the vibration signals collected by each sensor. Decomposed into core vibration components related to the core. and winding vibration components associated with the winding .

[0031] Specifically, the extraction of the core vibration components is based on their well-defined physical causes. Since the magnetostrictive effect of the core is mainly caused by the alternating magnetic field of the power grid, its vibration frequency is constant at the power grid frequency. twice (i.e.) ) and its higher harmonics. Therefore, a comb filter or a series of center frequencies can be constructed to detect them. ( A narrowband filter bank (where the integers are positive) is used to represent the original vibration signal in the frequency domain. These specific frequency components are extracted to obtain the frequency domain representation of the core vibration components. The time-domain signal is obtained after inverse Fourier transform. .

[0032] The extraction of winding vibration components utilizes their strong correlation with the load current. Winding vibration originates from the electromagnetic force generated by the interaction between the load current and the leakage magnetic field; the magnitude of this force is proportional to the square of the load current. Based on this principle, this embodiment employs adaptive filtering technology. The square of the real-time acquired load current signal is then used... As a reference input, the original vibration signal As the desired output, the transmission path from the electromagnetic force source to the measuring point on the box wall is identified through an adaptive filter. The output of the filter is the estimated winding vibration component. This effectively separates the vibrational components that are strongly correlated with the load from the complex mixed signal.

[0033] After successfully separating the vibration components from different sources, the next step is to quantify the propagation characteristics of the vibration in the structure, i.e., to calculate the transfer function. For any two vibration sensors... and Collected homogeneous vibration components (e.g., core vibration components) and ), the transfer function between them Defined as the ratio of the cross-power spectral density of two signals to the self-power spectral density of the input signal, its calculation formula is: ; in, For frequency; For sensors and sensors Between in frequency The transfer function at that point, in its complex numerical form, contains gain and phase information; For sensors The self-power spectral density of the vibrational components; For sensors and sensors The cross power spectral density of the vibrational components.

[0034] The self-power spectral density and cross-power spectral density are estimated by the following formula: ; ; in, It is the Fourier transform of the corresponding time-domain signal. It is its complex conjugate. This represents the expectation operation. In engineering practice, the piecewise averaging method is usually used to obtain a robust estimate.

[0035] The core of this embodiment lies in constructing not a static health baseline, but a dynamic tensor that reflects changes in operating conditions. To achieve this, it is necessary to perform tensor calculations at each of the aforementioned selected load levels. The following steps are repeated for signal decoupling and transfer function calculation. Specifically, for each load level... Calculate the values ​​of all sensor pairs respectively Core vibration transfer function Winding vibration transfer function These constitute the core vibration transfer function matrix and the winding vibration transfer function matrix under this operating condition, respectively.

[0036] Finally, all load levels All the calculated transfer function matrices are aggregated to form a four-dimensional mathematical object, namely the time-varying transfer function tensor of this invention, denoted as . Its mathematical expression is: ; This tensor As a complete health benchmark, it internally stores all the response patterns of the structural system to excitations from two different physical sources under different frequencies, different propagation paths, and different load levels when the main transformer is in a healthy state.

[0037] Through the steps described above, the time-varying transfer function tensor constructed in this embodiment possesses higher dimensionality and stronger scenario adaptability compared to traditional single-condition, single-source health fingerprints. It integrates load as an inherent dimension into the health model, thus providing a solid foundation for accurately separating the influence of operating conditions and effectively highlighting subtle changes caused by structural anomalies in subsequent online diagnostics.

[0038] S2: During online monitoring, real-time vibration signals and load current signals are acquired, and the real-time transfer function matrix is ​​calculated. This embodiment details how, after the main transformer enters the online monitoring phase, it acquires real-time operating data and calculates the real-time transfer function matrix characterizing its current structural response. This process is the prerequisite and foundation for effectively comparing with a health benchmark and subsequently detecting abnormal states.

[0039] During normal operation of the main transformer, the system of this invention continuously performs data acquisition operations. Specifically, the same set of sensors and acquisition equipment used when establishing the health baseline—namely, the vibration sensor array deployed on the main transformer tank wall and the current transformers on the lines—are used to acquire data in real time and synchronously. At any given monitoring moment, the system acquires a set of instantaneous vibration signals, denoted as... (in (For the sensor index), and the load current signal strictly time-aligned with it, denoted as The continuous synchronous acquisition here ensures that the acquired vibration response corresponds precisely to its excitation source (load current) in time, providing distortion-free data input for subsequent signal decoupling and dynamic characteristic analysis.

[0040] To ensure consistency and accuracy in subsequent calculations, the acquired real-time raw signals... and The process must undergo the exact same preprocessing steps as the health benchmark construction phase. This process preferably includes applying a bandpass filter to remove DC bias and high-frequency random noise that may introduce errors. Maintaining consistency in the preprocessing methods is a key technical guarantee for ensuring the comparability of subsequent real-time calculation results with the health benchmark; its purpose is to eliminate systematic biases introduced by differences in signal processing methods.

[0041] Next, we move on to one of the core aspects of online monitoring: real-time decoupling of multi-source vibration signals. This is because the vibration signals are acquired in real-time... Since the response is a mixture of various physical effects, including core vibration and winding vibration, the same decoupling algorithm used when constructing the health benchmark must be employed to decompose it into independent components that reflect specific physical sources. This is necessary because only by decomposing the real-time signal to the same dimensions and physical meaning as the health benchmark can subsequent differential comparisons have clear physical significance.

[0042] Specifically, for the real-time core vibration components The extraction of frequency characteristics still utilizes the deterministic nature of these characteristics in this embodiment. This is achieved through the analysis of real-time vibration signals. Its spectrum is obtained by performing a Fourier transform. And apply a center frequency set at twice the power grid frequency ( Comb filters at the positions of the iron core and its harmonics are used to separate the frequency components related to the magnetostriction effect of the iron core from the mixed spectrum, and then the real-time iron core vibration components are obtained by inverse Fourier transform. .

[0043] For real-time winding vibration components The extraction of the load current signal utilizes its causal relationship with the real-time load current. This embodiment extracts the square of the real-time load current signal. As a reference input, the original real-time vibration signal is used. The desired output is processed by an online adaptive filter. This filter continuously adjusts its internal parameters so that its output best reproduces the portion of the vibration signal related to the square of the load current; this output is the estimated real-time winding vibration component. .

[0044] After obtaining the two sets of decoupled real-time vibration components, the real-time transfer function matrix characterizing the current system state can be calculated. This calculation is divided into two parallel parts, corresponding to the core vibration propagation path and the winding vibration propagation path, respectively.

[0045] For the propagation path of core vibration, any two vibration sensors and Real-time transfer function between It is based on the real-time core vibration components and Calculated. For the winding vibration propagation path, any two vibration sensors and Real-time transfer function between This is based on the real-time winding vibration components. and Calculated.

[0046] The calculation of both types of transfer functions above follows the same formula definition: ; in, For frequency; Represents the real-time transfer function; The self-power spectral density is calculated based on real-time vibration components; The cross-power spectral density is calculated based on real-time vibration components. These power spectral densities are estimated by performing a Fourier transform on the real-time signal and then performing expectation operations, for example, by using the Welch method to perform piecewise averaging on the continuous data stream to obtain stable calculation results.

[0047] Ultimately, all sensors... Real-time transfer function Combined, they form the real-time core vibration transfer function matrix. Similarly, These components combine to form the real-time winding vibration transfer function matrix. These two matrices together form a comprehensive, quantitative snapshot that accurately describes the dynamic response characteristics of the main transformer's internal structure to the two main excitation sources at the current monitoring moment. These two matrices will serve as direct input for the next step of comparison with a health benchmark and calculation of the differential response field.

[0048] S3: Calculate the health baseline transfer function matrix corresponding to the real-time transfer function matrix based on the time-varying transfer function tensor and the real-time load; This embodiment details how, after obtaining the real-time transfer function matrix, the corresponding health baseline transfer function matrix is ​​accurately calculated based on the constructed time-varying transfer function tensor and the current real-time load. This step is crucial for the invention to achieve adaptive diagnosis and effectively distinguish between changes in operating conditions and structural anomalies. Its purpose is to provide a fair and accurate health reference for subsequent differential comparisons.

[0049] Before proceeding to this step, the system has obtained real-time load levels characterizing its current operating status through online monitoring. This load level Preferably, by real-time acquisition of load current signal The process will be performed to determine this. Meanwhile, the system has internally stored the time-varying transfer function tensor calibrated under healthy conditions of the main transformer. .

[0050] Due to the actual operating load of the main transformer It is a continuously changing quantity, and its value is very likely not to fall precisely on the set of discrete load level points selected during pre-calibration. Above. If the health data of the closest discrete point is used directly as the benchmark, systematic comparison errors will be introduced due to differences in load levels. These errors may be misjudged as abnormalities or mask the true characteristics of early failures.

[0051] To address this technical problem, this embodiment employs a dynamic benchmark generation method based on interpolation. The core idea of ​​this method is to utilize the time-varying transfer function tensor... The health data at discrete load points is used to calculate the current real-time load level using an interpolation algorithm. Below, the transfer function matrix that a theoretically healthy principal variable should exhibit.

[0052] Specifically, this process first requires a set of discrete load level points. In the process, two values ​​that will change the current real-time load level were found. The two adjacent load points that are surrounded. Assume the two load points found are... and ,satisfy Subsequently, from the time-varying transfer function tensor... In the process, the complete health transfer function matrix corresponding to these two load points is extracted respectively, i.e. and .

[0053] After obtaining these two adjacent health benchmarks, the load level is calculated by linear interpolating each element in the transfer function matrix. Health baseline transfer function Preferably, the linear interpolation process can be represented by the following formula: ; in, It is pending, under the current load The health baseline transfer function is as follows; and It is at two adjacent calibration load points respectively. and The healthy transfer function is stored in the time-varying transfer function tensor. This calculation is performed on the transfer function itself, a complex number, while ensuring a smooth transition in its magnitude and phase.

[0054] It is important to emphasize that this interpolation calculation is performed independently on two components of the time-varying transfer function tensor. Specifically: on one hand, the current load is calculated using the transfer function component related to core vibration. The health baseline iron core transfer function matrix is ​​denoted as follows: On the other hand, the current load is calculated using the transfer function components related to winding vibration. The healthy reference winding transfer function matrix is ​​denoted as follows: .

[0055] This calculation process requires iterating through all sensor pairs. and frequency points of all analyses This ultimately generates two matrices related to the real-time transfer function. and The health benchmark transfer function matrix that is completely corresponding in both dimensional and physical sense and .

[0056] Furthermore, in some edge cases, such as the current real-time load... Exceeding the calibration range (i.e.) or In this embodiment, extrapolation or the use of the closest boundary value (i.e.) is preferably employed. or This serves as a health benchmark to ensure the robustness of the algorithm. Although linear interpolation is preferred in this embodiment, those skilled in the art can use other interpolation methods, such as cubic spline interpolation or polynomial interpolation, as needed to obtain potentially smoother transition characteristics.

[0057] Through the steps of this embodiment, the system no longer uses a static, fixed health template, but can dynamically generate a health benchmark that perfectly matches it based on real-time operating conditions. The generated health benchmark transfer function matrix accurately represents the structural response state that a healthy main transformer should have under the current load level, thus providing a high-precision, unbiased benchmark for the next step of calculating the differential response field that can truly reflect changes in structural state.

[0058] S4: Calculate the difference between the real-time transfer function matrix and the health baseline transfer function matrix to obtain the differential response field; This embodiment details how, based on obtaining the real-time transfer function matrix representing the current operating state and its precisely corresponding health baseline transfer function matrix, the difference between the two is calculated to obtain a differential response field that accurately represents changes in the system state. This step is the core of the diagnostic logic of this invention, and its purpose is to quantify the deviation between the current system response and the ideal health state, and to provide clean and reliable input data for subsequent fault tracing.

[0059] In the aforementioned steps, the system has obtained two sets of key matrix data in parallel: one is the real-time transfer function matrix obtained through real-time monitoring and signal processing, including the real-time core transfer function matrix. and real-time winding transfer function matrix Secondly, it is a health baseline transfer function matrix that perfectly matches the current real-time operating conditions, obtained by interpolating the health time-varying transfer function tensor, including the health baseline core transfer function matrix. and health reference winding transfer function matrix .

[0060] The core task of this step is to accurately subtract the two sets of data to eliminate response variations caused by changes in normal operating conditions, thereby highlighting the true deviation caused by abnormal internal structural conditions of the main transformer. This calculation is performed independently for each source to ensure the clarity of subsequent fault attribution.

[0061] Specifically, for the vibration propagation path originating from the magnetostrictive effect of the iron core, the differential response field of the iron core path is obtained by subtracting each element in the real-time iron core transfer function matrix from the corresponding element in the healthy baseline iron core transfer function matrix. The calculation of each element follows the formula: ; in, For frequency; To ensure that the sensor is at the current monitoring time, The real-time transfer function calculated from the core vibration components; To adjust according to the current real-time load The health transfer function obtained by interpolation from the health baseline at the same sensor pair and frequency; This refers to the response deviation at that specific path and frequency.

[0062] Similarly, for the vibration propagation path originating from the electromagnetic force of the winding, the differential response field of the winding path can be obtained by subtracting the real-time winding transfer function matrix from the healthy reference winding transfer function matrix element by element. The calculation of each element follows the formula: ; in, This is the real-time transfer function calculated from the winding vibration components; The corresponding health baseline transfer function; This represents the response deviation along that path.

[0063] It should be noted that, since the transfer function is a complex number containing both amplitude and phase, the subtraction operation here is a subtraction in the complex field. This means that the obtained differential response field and It contains not only information on the change in gain during vibration propagation, but also information on the change in phase. This complete preservation of amplitude and phase information is crucial for accurately describing changes in the dynamic characteristics of the system.

[0064] The physical significance of the differential response field obtained in this way lies in its quantification of the degree to which the main transformer's structural response deviates from its expected healthy state under current operating conditions. Since the health benchmark itself is dynamically adjusted according to operating conditions, this differential calculation fundamentally eliminates the interference of normal operating factors such as load fluctuations on the diagnostic results. When the main transformer structure is healthy, the real-time transfer function matrix should be highly consistent with the dynamically generated health benchmark; at this time, the amplitudes of each element in the calculated differential response field will approach zero. Conversely, when changes occur in the internal structure, such as winding deformation, core loosening, or abnormal support components, the vibration transmission path characteristics will change, causing the real-time transfer function to deviate from the health benchmark, thus producing significant non-zero values ​​in the differential response field.

[0065] Ultimately, the output of this step is two independent difference response field matrices. and These two matrices together constitute a high-dimensional set of anomalous features. The numerical distribution, amplitude, and distribution patterns in the frequency and spatial domains within this set contain rich information about internal structural anomalies. This differential response field will serve as the direct observation input for the subsequent inverse disturbance source inversion step, providing the final decision-making basis for accurate fault location and attribute determination.

[0066] S5: Based on the differential response field, the abnormal state inside the main transformer is determined by inversion of the reverse disturbance source.

[0067] This embodiment details how, after obtaining the differential response field that accurately reflects changes in the structural state, the abnormal state inside the main transformer is ultimately determined through the technique of inverse disturbance source inversion. This step constitutes the core of the intelligent analysis method of this invention for decision-making and localization. Its purpose is to trace the changes in external response observed from the sensor array back to their physical causes inside the main transformer, thereby achieving a leap from judging an abnormal state to fault localization.

[0068] In the preceding steps, the system has calculated and obtained two independent differential response field matrices: the iron core path differential response field. Differential response field of winding path These two matrices are the sole and complete input of the observation data for subsequent inversion calculations. In this embodiment, these two complex matrices are vectorized and expanded in the frequency dimension and the sensor pair dimension, and combined into a long one-dimensional complex vector, denoted as the observation data vector. .

[0069] One of the core ideas of this invention is to construct the fault diagnosis problem as an inverse problem supported by a physical model. To solve this inverse problem, it is first necessary to establish a forward physical model that can describe the relationship between cause and effect. Here, the cause refers to the small perturbation of structural parameters occurring at any location inside the main transformer, and the effect is the differential response field caused by this perturbation at the sensor on the tank wall.

[0070] Specifically, the internal structure of the main transformer is first discretized spatially into several units. The potential changes in the structural parameters of each unit are defined as the model parameter vector. One of the elements in the model parameter vector. It comprehensively describes the location and attributes of all possible anomalies within the main transformer.

[0071] Forward physical model, also known as forward operator Its function is to predict when a known disturbance occurs in the internal structure. What kind of differential response field will be generated at the box wall? This relationship can be linearized under the assumption of small perturbations as follows: ; in, It is a large Jacobian matrix or sensitivity matrix. Each element of this matrix... Indicates the first The unit change of the first internal structural parameter affects the first... The degree of influence of each observation data. This positive operator. It is not real-time, but pre-calculated and stored before system deployment. Preferably, this matrix can be constructed using high-precision numerical simulation methods.

[0072] After establishing the forward physical model, the diagnostic task is transformed into solving the inverse problem of the aforementioned system of linear equations: given the observation data... and positive operators Solve for the internal disturbance source that caused this observation. .

[0073] However, such inverse physical problems are often ill-conditioned, meaning that even small noises in the observation data can cause violent oscillations in the solution, and the solution may not be unique. To address this technical challenge, this embodiment employs the Tikhonov regularization method to seek a stable and physically reasonable solution. This method transforms the ill-conditioned problem into a well-conditioned one by imposing a constraint on the solution itself while minimizing the data fitting error.

[0074] Specifically, the objective of this embodiment is to solve for a solution that minimizes the following objective function. Optimal internal structure perturbation vector : ; In this objective function, the first term It is a data fitting term, and its physical meaning is to make the model predict the difference response field. The difference response field compared with the actual observation Minimize the squared L2 norm between the two sides to ensure that the solution can fully explain the observation data.

[0075] Second item This is a regularization term used to introduce prior information about the solution to stabilize the solution process. Among them, It is a regularization operator, preferably an identity matrix or a discrete Laplace operator. It is a regularization parameter, whose function is to balance the weights between the data fitting term and the regularization term. The value of is crucial to the quality of the solution, and can preferably be automatically optimized and determined by methods such as the L-curve method.

[0076] The optimal solution obtained by solving the above optimization problem is... It is a vector, where the value of each element represents the change in the structural parameters of the corresponding discrete unit inside the main transformer.

[0077] Finally, to facilitate understanding and decision-making by engineering technicians, this embodiment will present the obtained perturbation vector. The results are then remapped back into the three-dimensional physical space of the main transformer and presented in a visual manner. Preferably, a probability distribution cloud map or intensity distribution map of internal structural anomalies can be generated, with the highlighted areas in the map indicating the most likely locations of internal anomalies.

[0078] Furthermore, due to the observation data vector It is composed of the core path differential response field Differential response field of winding path Together, these two parts of data can also be used to drive the final solution in this embodiment. The contribution of each component in the formation process can be used to help identify anomalous attributes. For example, if the final solution is mainly composed of... If the significant change is caused by [a specific factor], then the anomaly is more likely related to the core or its fasteners; conversely, if it is caused by [another factor], then [the anomaly is more likely related to] the core or its fasteners. If the dominant factor is the winding or its supporting structure, then the anomaly is more likely to be related to the winding.

[0079] Please see the appendix Figure 2 The intelligent analysis system for main transformer vibration characteristics includes the following steps: The data acquisition module is used to synchronously acquire the vibration signal and load current signal of the main transformer; The health benchmark construction module is used to construct a time-varying transfer function tensor characterizing the health status of the main transformer based on vibration signals and load current signals covering multiple load levels under a preset health state. The online diagnostic module is configured as follows: The real-time vibration signal and load current signal are acquired, and the real-time transfer function matrix is ​​calculated. Based on the time-varying transfer function tensor and the real-time load, the health baseline transfer function matrix corresponding to the real-time transfer function matrix is ​​calculated. The difference between the real-time transfer function matrix and the health baseline transfer function matrix is ​​calculated to obtain the differential response field; The inverse inversion module is used to determine the abnormal state inside the main transformer by inverting the inverse disturbance source based on the differential response field.

[0080] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent analysis method for the vibration characteristics of a main transformer, characterized in that, Includes the following steps: S1: Under a preset health state, a time-varying transfer function tensor characterizing the health state of the main transformer is constructed based on vibration signals and load current signals covering multiple load levels. S2: During online monitoring, real-time vibration signals and load current signals are acquired, and the real-time transfer function matrix is ​​calculated. S3: Calculate the health baseline transfer function matrix corresponding to the real-time transfer function matrix based on the time-varying transfer function tensor and the real-time load; S4: Calculate the difference between the real-time transfer function matrix and the health baseline transfer function matrix to obtain the differential response field; S5: Based on the differential response field, the abnormal state inside the main transformer is determined by inversion of the reverse disturbance source.

2. The method according to claim 1, characterized in that, The steps for constructing the time-varying transfer function tensor include: The vibration signals and load current signals of the main transformer under multiple load levels are collected simultaneously from the vibration sensor array. The vibration signal is decoupled into core vibration component and winding vibration component; For the core vibration component and the winding vibration component, the transfer function between any two vibration sensors is calculated under each load level, and a transfer function matrix is ​​formed. The transfer function matrices under all load levels are combined to construct the time-varying transfer function tensor.

3. The method according to claim 2, characterized in that, The specific steps for decoupling the vibration signal into core vibration components and winding vibration components are as follows: The vibration signal with a frequency twice the power grid frequency and its harmonics is extracted from the vibration signal by using a comb filter or narrowband filter bank, and is used as the core vibration component. Using the square of the load current signal as a reference input, an adaptive filter is used to extract the signal related to the reference input from the vibration signal, which is then used as the winding vibration component.

4. The method according to claim 1, characterized in that, The specific steps for calculating the health baseline transfer function matrix are as follows: The current load level is determined based on the real-time load current signal; In the time-varying transfer function tensor, interpolation is performed along the load level dimension to obtain the health baseline transfer function matrix corresponding to the current load level.

5. The method according to claim 2, characterized in that, The differential response field includes the core path differential response field and the winding path differential response field; The core path differential response field is calculated from the transfer function matrix related to the core vibration components; The winding path differential response field is calculated from the transfer function matrix associated with the winding vibration components.

6. The method according to claim 1, characterized in that, The step of determining the abnormal state inside the main transformer through inversion of the reverse disturbance source includes: Construct a forward physical model from the internal structural disturbance of the main transformer to the differential response field of the box wall; Based on the aforementioned forward physical model, an inverse problem is established and solved, using the differential response field as the observed data and the internal structural disturbance as the parameter to be determined, in order to determine the abnormal location or properties inside the main transformer.

7. The method according to claim 6, characterized in that, The inverse problem is solved using the Tikhonov regularization method, with the goal of finding a solution that minimizes the objective function. Optimal internal structure perturbation vector : ； in, Let the internal structure perturbation vector be the optimization variable; The observed data vector is composed of the differential response field; For the operators of the aforementioned positive physical model; For regularization parameters; For regularization operators; It is the squared L2 norm of the vector.

8. The method according to claim 2, characterized in that, Any two vibration sensors and Transfer function between Calculated using the following formula: ； in, For frequency; For sensors and sensors Between in frequency The transfer function at the location; For sensors The vibration components at frequency The self-power spectral density at the location; For sensors and sensors The vibration components at frequency The cross-power spectral density at that location.

9. The method according to claim 5, characterized in that, The step of determining the abnormal state inside the main transformer also includes: Based on the contribution of the core path differential response field and the winding path differential response field to the solution of the inverse problem, it is determined whether the abnormal state is related to the core or the winding.

10. A main transformer vibration characteristic intelligent analysis system, and a main transformer vibration characteristic intelligent analysis method according to any one of claims 1-9, characterized in that, Includes the following steps: The data acquisition module is used to synchronously acquire the vibration signal and load current signal of the main transformer; The health benchmark construction module is used to construct a time-varying transfer function tensor characterizing the health state of the main transformer based on the vibration signal and load current signal covering multiple load levels under a preset health state. The online diagnostic module is configured as follows: The real-time vibration signal and load current signal are acquired, and the real-time transfer function matrix is ​​calculated. Based on the time-varying transfer function tensor and the real-time load, the health baseline transfer function matrix corresponding to the real-time transfer function matrix is ​​calculated. The difference between the real-time transfer function matrix and the health baseline transfer function matrix is ​​calculated to obtain the differential response field; The inverse inversion module is used to determine the abnormal state inside the main transformer by inverting the inverse disturbance source based on the differential response field.

Images