A transformer condition monitoring method, system, and electronic medium

Abstract

This application discloses a transformer condition monitoring method, system, and electronic medium. The method includes: acquiring the transformer neutral point current signal and performing time-frequency decomposition on the current signal to obtain a multi-scale time-frequency energy distribution; constructing a health evolution factor characterizing the system's energy redistribution behavior, connecting the health evolution factors in chronological order to generate a health state trajectory; introducing the accumulated memory information of the health evolution factors within historical time periods to construct a nonlinear health-driven sliding manifold with path-dependent characteristics, wherein the sliding manifold is used to characterize the multi-stage dynamic boundaries in the system's evolution from steady state to unstable state; identifying potential fault triggering conditions by analyzing the adsorption, deviation, and crossing behavior of the health state trajectory on the sliding manifold; and outputting the transformer health state monitoring results based on the dynamic interaction between the health state trajectory and the sliding manifold. This application significantly improves the level of transformer health management.

Description

Technical Field

[0001] This application relates to the field of power equipment condition monitoring and fault prediction technology, and in particular to a transformer condition monitoring method, system and electronic medium. Background Technology

[0002] As a core component of the power grid, the operating status of power transformers directly affects the safe and stable operation of the power system. A fault in a transformer can lead to widespread power outages, equipment damage, or even serious safety accidents. With the continuous expansion of the power grid and the increasing service life of equipment, the importance of transformer condition monitoring and health management is becoming increasingly prominent.

[0003] Existing transformer condition monitoring technologies mainly include dissolved gas analysis (DGA), online partial discharge (PD) monitoring, vibration signal analysis, temperature monitoring, and electrical parameter monitoring (such as three-phase imbalance and core grounding current). Among these, dissolved gas analysis, which detects the components and contents of characteristic gases in the oil to determine internal overheating or discharge faults, is one of the most widely used methods. Partial discharge monitoring can effectively capture pulse signals caused by insulation defects. Vibration monitoring is mainly used to identify mechanical loosening problems in windings or the core. Existing transformer condition monitoring methods still have significant shortcomings in terms of early fault sensitivity, path-dependent modeling, robustness, and ease of engineering implementation, making it difficult to meet the high-precision and real-time requirements of modern smart grids for transformer health management and fault prediction.

[0004] Therefore, developing an advanced monitoring method that utilizes only the neutral point current signal and comprehensively considers the dynamic process of multi-scale time-frequency energy redistribution and path-dependent nonlinear boundaries has significant theoretical and engineering application value. Summary of the Invention

[0005] This application provides a transformer condition monitoring method, system, and electronic medium, which significantly improves the level of transformer health management.

[0006] This application provides the following solution: According to a first aspect, a transformer condition monitoring method is provided, the method comprising: acquiring a transformer neutral point current signal and performing time-frequency decomposition on the current signal to obtain a multi-scale time-frequency energy distribution; constructing a health evolution factor characterizing the energy redistribution behavior of the system based on the migration path and cross-scale response differences of the time-frequency energy in different frequency bands; connecting the health evolution factors in chronological order to generate a health state trajectory; introducing the accumulated memory information of the health evolution factors in historical time periods to construct a nonlinear health-driven sliding mode manifold with path-dependent characteristics, wherein the sliding mode manifold is used to characterize the multi-stage dynamic boundary in the process of the system evolving from a steady state to an unstable state; defining an evolution approach direction and an escape criterion on the sliding mode manifold; identifying potential fault triggering conditions by analyzing the adsorption, deviation, and crossing behavior of the health state trajectory on the sliding mode manifold; and outputting the transformer health state monitoring result based on the dynamic interaction relationship between the health state trajectory and the sliding mode manifold.

[0007] According to one achievable method in an embodiment of this application, time-frequency decomposition is performed on the current signal to obtain a multi-scale time-frequency energy distribution, including: adaptive segmentation processing of the current signal; dynamically adjusting the decomposition window length and overlap ratio according to the energy change rate and spectral distribution differences of the signal in each time segment; performing multi-resolution time-frequency decomposition processing in each segment to obtain the local time-frequency energy distribution at the corresponding time scale; normalizing and aligning the time-frequency energy distribution at different time scales and constructing a cross-scale energy mapping relationship; and performing consistency correction on the multi-scale time-frequency energy distribution based on the cross-scale energy mapping relationship to form a multi-scale time-frequency energy distribution result with a unified reference benchmark.

[0008] According to one achievable method in this application embodiment, the construction of a health evolution factor characterizing the energy redistribution behavior of the system based on the migration path and cross-scale response differences of time-frequency energy in different frequency bands includes: performing correlation modeling on the energy change relationship of each frequency band at adjacent times to generate an energy migration network describing the direction, intensity, and dynamic migration path of energy transfer between frequency bands; extracting structural indicators reflecting energy redistribution behavior based on the concentration degree, path dispersion characteristics, and stability of migration paths in the energy migration network; and combining the topological consistency of the energy migration network at different time scales with the degree of deviation of structural similarity formed by comparing cross-scale response differences to jointly form the health evolution factor.

[0009] According to one achievable method in the embodiments of this application, the step of associating and modeling the energy change relationship of each frequency band at adjacent times to generate an energy migration network describing the energy transfer direction, intensity, and dynamic migration path between frequency bands includes: using different frequency bands as network nodes, and using the energy change of each frequency band between adjacent times as the weight of the directed edges; determining the direction and intensity of energy transfer based on the direction and magnitude of energy change; and constructing a dynamic directed weighted network through the energy transfer relationship at multiple consecutive times to form a complete energy migration network describing the energy transfer direction, intensity, and dynamic migration path between frequency bands.

[0010] According to one achievable method in this application embodiment, the step of introducing accumulated memory information of health evolution factors within a historical time period to construct a nonlinear health-driven sliding manifold with path-dependent characteristics includes: obtaining a sequence of health evolution factors for consecutive historical moments within a preset time window, and performing time-weighted accumulation processing on the health evolution factor sequence to generate a memory state vector reflecting the historical evolution trajectory; constructing an extended state variable containing historical dependencies based on the coupling relationship between the memory state vector and the health evolution factors at the current moment; performing nonlinear mapping processing on the extended state variable to generate a corresponding sliding manifold expression, and introducing path-related constraint terms into the sliding manifold to characterize the influence of different historical evolution paths on the current state boundary; and constraining the health state trajectory according to the sliding manifold expression so that the system state exhibits differentiated convergence behavior under different historical evolution conditions.

[0011] According to one achievable method in this application embodiment, the construction of extended state variables containing historical dependencies based on the coupling relationship between the memory state vector and the current-moment health evolution factor includes: introducing a non-consistent decay mechanism for information from different historical stages in the memory state vector; assigning different memory retention weights based on the differences in the degree of influence of each historical stage on the current state to form hierarchical memory sub-states; combining and mapping the hierarchical memory sub-states with the current-moment health evolution factor to generate multiple candidate extended state components to represent the state response under different potential evolution paths; introducing a competitive selection mechanism among the multiple candidate extended state components; determining the dominant extended state variable by comparing the consistency between each candidate component and the current observation trajectory; and locally reconstructing the sliding manifold based on the dominant extended state variable to make different historical evolution paths correspond to differentiated state evolution boundaries.

[0012] According to one achievable method in this application embodiment, defining the evolution approach direction and escape criterion on the sliding manifold, and identifying potential fault triggering conditions by analyzing the adsorption, deviation, and crossing behavior of the healthy state trajectory on the sliding manifold, includes: mapping the healthy state trajectory onto the sliding manifold, calculating its local tangential and normal components in the manifold tangential space, and determining the evolution approach direction based on the changing trend of the tangential component; constructing an escape criterion function based on the normal deviation distance of the healthy state trajectory relative to the sliding manifold and its rate of change; determining that the healthy state trajectory has escaped when the normal deviation distance continues to increase and its rate of change exceeds a preset threshold; and identifying a sudden change in the evolution path when the healthy state trajectory undergoes a direction reversal or the tangential component undergoes a discontinuous change, in order to assist in judging the transformation of the transformer from a stable state to an unstable state.

[0013] According to one achievable method in the embodiments of this application, based on the dynamic interaction between the health status trajectory and the sliding manifold, the transformer health status monitoring result is output, including: statistically analyzing the proportion of the residence time of the health status trajectory in different regions of the sliding manifold and the frequency of cross-region migration, and constructing a distribution characteristic quantity reflecting the stability of the state; based on the distribution characteristic quantity and the trajectory's adsorption, deviation and crossing behavior on the sliding manifold, the corresponding health status level is divided; and the health status level is used as the health status monitoring result and output.

[0014] According to the second aspect, a transformer condition monitoring system is provided, the system comprising: a time-frequency energy distribution acquisition unit configured to acquire a transformer neutral point current signal and perform time-frequency decomposition on the current signal to obtain a multi-scale time-frequency energy distribution; a health state trajectory generation unit configured to construct a health evolution factor characterizing the system's energy redistribution behavior based on the migration path and cross-scale response differences of time-frequency energy in different frequency bands, and connect the health evolution factors in chronological order to generate a health state trajectory; and a sliding mode manifold construction unit configured to introduce health evolution factors from historical time periods. The system utilizes accumulated memory information to construct a nonlinear health-driven sliding manifold with path-dependent characteristics. This sliding manifold is used to characterize the multi-stage dynamic boundaries during the system's evolution from a steady state to an unstable state. A fault triggering condition identification unit is configured to define the evolution approach direction and escape criterion on the sliding manifold and identify potential fault triggering conditions by analyzing the adsorption, deviation, and crossing behavior of the health state trajectory on the sliding manifold. A state monitoring result acquisition unit is configured to output transformer health state monitoring results based on the dynamic interaction between the health state trajectory and the sliding manifold.

[0015] According to a third aspect, a computer-readable electronic medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the method described in any one of the first aspects above.

[0016] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application acquires the neutral point current signal of the transformer and obtains multi-scale time-frequency energy distribution through time-frequency decomposition. Then, based on the energy migration path between frequency bands and cross-scale response differences, it constructs a health evolution factor and generates a health state trajectory in chronological order. Simultaneously, it incorporates historical accumulated memory information to construct a nonlinear health-driven sliding manifold with path-dependent characteristics. By analyzing the adsorption, deviation, and crossing behaviors of the health state trajectory on this sliding manifold, potential fault triggering conditions can be accurately identified, ultimately outputting the transformer health state level and fault evolution path prediction results. This method achieves dynamic characterization of the multi-stage evolution process of the transformer from steady state to unstable state. Compared with traditional methods, it has significant technical advantages such as non-invasiveness, strong early warning capability, high robustness, and good interpretability of prediction results, which can significantly improve the level of transformer health management.

[0017] Of course, any product implementing this application does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart of the transformer condition monitoring method provided in the embodiments of this application; Figure 2 This is a structural block diagram of the transformer condition monitoring system provided in the embodiments of this application; Figure 3 A schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0021] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0022] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0023] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0024] Figure 1 This is a flowchart illustrating a transformer condition monitoring method provided in an embodiment of this application. Figure 1 As shown, the method may include the following steps: Step 101: Acquire the neutral point current signal of the transformer and perform time-frequency decomposition on the current signal to obtain the multi-scale time-frequency energy distribution.

[0025] Step 102: Based on the migration path of time-frequency energy in different frequency bands and the difference in cross-scale response, construct a health evolution factor that characterizes the energy redistribution behavior of the system, and connect the health evolution factors in chronological order to generate a health state trajectory.

[0026] Step 103: Introduce the accumulated memory information of health evolution factors within the historical period to construct a nonlinear health-driven sliding manifold with path-dependent characteristics, wherein the sliding manifold is used to characterize the multi-stage dynamic boundary of the system during the evolution from steady state to unstable state.

[0027] Step 104: Define the evolution approach direction and escape criterion on the sliding manifold, and identify potential fault triggering conditions by analyzing the adsorption, deviation and crossing behavior of the healthy state trajectory on the sliding manifold.

[0028] Step 105: Based on the dynamic interaction between the health status trajectory and the sliding mode manifold, output the transformer health status monitoring results.

[0029] As can be seen from the above process, this application only collects the neutral point current signal of the transformer, obtains the multi-scale time-frequency energy distribution through time-frequency decomposition, and then constructs a health evolution factor based on the energy migration path between frequency bands and the cross-scale response difference, and generates a health state trajectory in chronological order; at the same time, it introduces historical accumulated memory information to construct a nonlinear health-driven sliding manifold with path-dependent characteristics. By analyzing the adsorption, deviation and crossing behavior of the health state trajectory on this sliding manifold, potential fault triggering conditions can be accurately identified, and finally the transformer health state level and fault evolution path prediction results are output. This method realizes the dynamic characterization of the multi-stage evolution process of the transformer from steady state to unstable state. Compared with traditional methods, it has significant technical effects such as non-intrusiveness, strong early warning capability, high robustness and good interpretability of prediction results, which can significantly improve the level of transformer health management.

[0030] The following describes in detail each step of the above process and the effects that can be further produced, with reference to the embodiments.

[0031] First, the above step 101, namely "acquiring the neutral point current signal of the transformer and performing time-frequency decomposition on the current signal to obtain a multi-scale time-frequency energy distribution", will be described in detail with reference to the embodiments.

[0032] The neutral point current signal of a transformer is one of the important electrical parameters reflecting the internal operating state of the transformer. Under normal operating conditions, the neutral point current is mainly a zero-sequence component generated by three-phase imbalance, with a small amplitude and relatively stable performance. However, when potential faults occur inside the transformer, such as inter-turn short circuits in the windings, multi-point grounding of the core, localized insulation degradation, or magnetic circuit asymmetry, the neutral point current will produce significant transient distortions and high-frequency oscillation components. These changes often exhibit strong non-stationary characteristics, meaning that the frequency components of the signal change rapidly over time, making it difficult for traditional single time-domain or frequency-domain analysis to fully capture their inherent patterns.

[0033] Performing time-frequency decomposition on the acquired neutral point current signal is precisely to overcome this limitation. The time-frequency decomposition method expands the original one-dimensional current signal simultaneously in both time and frequency dimensions, decomposing the signal into a series of components of different scales. Each scale corresponds to a specific frequency range; low-scale components typically reflect high-frequency transient changes in the signal, such as partial discharge pulses caused by early faults or spikes generated by winding short circuits; while high-scale components capture the low-frequency trends and overall envelope changes of the signal. This multi-resolution characteristic allows the decomposition result to simultaneously preserve both the temporal and frequency local information of the signal, avoiding the drawback of traditional Fourier transform losing its time localization capability when processing non-stationary signals.

[0034] The time-frequency decomposition can be performed using traditional methods or improved versions. One feasible approach involves performing time-frequency decomposition on the current signal to obtain a multi-scale time-frequency energy distribution. This includes: adaptively segmenting the current signal; dynamically adjusting the decomposition window length and overlap ratio based on the energy change rate and spectral distribution differences within each time segment; performing multi-resolution time-frequency decomposition within each segment to obtain the local time-frequency energy distribution at the corresponding time scale; normalizing and aligning the time-frequency energy distributions at different time scales and constructing a cross-scale energy mapping relationship; and performing consistency correction on the multi-scale time-frequency energy distribution based on the cross-scale energy mapping relationship to form a multi-scale time-frequency energy distribution result with a unified reference benchmark.

[0035] Specifically, the acquired neutral point current signal is first adaptively segmented. This process dynamically adjusts the length of the decomposition window and the overlap ratio between adjacent windows based on the differences in the energy change rate and spectral distribution of the signal within each time segment. When the signal experiences drastic fluctuations or rapid changes in spectral components, the window automatically shortens to improve time resolution; while in relatively stable signal ranges, the window is appropriately extended to improve frequency resolution. This adaptive mechanism avoids information loss or redundancy caused by fixed windows, ensuring that the decomposition process can flexibly adapt to complex situations such as load fluctuations, harmonic interference, or early fault transients in actual operation.

[0036] Next, multi-resolution time-frequency decomposition processing is performed within each adaptive segment. The multi-resolution algorithm decomposes the signal of the current time segment into a series of components at different scales. Smaller scales correspond to higher frequency components, capturing rapidly changing transient features in the signal, such as spike pulses caused by early inter-turn short circuits or high-frequency oscillations of partial discharge; larger scales correspond to lower frequency components, mainly reflecting the overall trend and slow changes of the signal. Through this multi-resolution processing, each segment can obtain the local time-frequency energy distribution at its corresponding time scale, clearly showing the local concentration of energy in the two-dimensional time and frequency planes.

[0037] Subsequently, the time-frequency energy distributions at different time scales were normalized and aligned, and cross-scale energy mapping relationships were constructed. Normalization and alignment ensured the comparability of energy values ​​across different scales, while the cross-scale energy mapping relationship established a correlation channel between low-scale high-frequency components and high-scale low-frequency components, allowing previously isolated local energy distributions to reference each other and form a coherent multi-scale energy view. This mapping process helps reveal how energy is transferred and interacts between different frequency levels.

[0038] Finally, based on the constructed cross-scale energy mapping relationship, consistency correction is performed on the multi-scale time-frequency energy distribution. This correction eliminates slight discontinuities or biases introduced by segmentation boundaries or decomposition algorithms, ultimately resulting in a multi-scale time-frequency energy distribution with a unified reference benchmark. This unified benchmark allows subsequent analyses to be conducted within the same scale framework, avoiding contradictions or error accumulation between data from different scales, and ensuring that the overall energy distribution results are accurate, stable, and highly comparable.

[0039] After time-frequency decomposition, a multi-scale time-frequency energy distribution is obtained. This distribution, measured by energy, quantifies the degree of energy concentration of the signal in different time periods and frequency bands. For example, in the early stages of a fault, the energy in certain mid-to-high frequency bands may suddenly increase or migrate, while the energy distribution in the low frequency band remains relatively stable. By observing the distribution patterns of these energies at multiple scales, the dynamic process of energy redistribution within the transformer can be intuitively revealed, providing reliable basic data for the subsequent construction of health evolution factors and health state trajectories. This multi-scale time-frequency energy distribution not only enhances the sensitivity to weak early faults but also significantly improves the robustness of the method under complex operating conditions (such as load fluctuations and harmonic interference), making it a core preliminary step in the entire condition monitoring method.

[0040] The following describes in detail step 102, namely, "based on the migration path of time-frequency energy in different frequency bands and the difference in cross-scale response, construct a health evolution factor characterizing the energy redistribution behavior of the system, and connect the health evolution factors in chronological order to generate a health state trajectory," with reference to the embodiments.

[0041] Based on the migration path of time-frequency energy across different frequency bands and the differences in cross-scale response, a health evolution factor characterizing the energy redistribution behavior of the system is constructed, and the health evolution factor is connected in chronological order to generate a health state trajectory. This is the core step of the present invention to transform signal processing results into quantifiable health indicators.

[0042] First, energy migration paths between different frequency bands are extracted from the multi-scale time-frequency energy distribution. An energy migration path refers to the direction, intensity, and speed of energy transfer from one frequency band to another between adjacent time points. When the transformer is in a healthy state, energy migration between frequency bands typically exhibits strong regularity and stable paths. However, when an early fault occurs internally, the energy migration paths change significantly. For example, high-frequency energy may suddenly concentrate and migrate to the mid- or low-frequency bands, or irregular back-and-forth oscillations may occur. These dynamic migration paths directly reflect the redistribution process of different forms of energy within the transformer, such as electromagnetic energy, mechanical vibration energy, and thermal energy, and are important indicators of the early evolution of faults.

[0043] Meanwhile, cross-scale response differences are introduced as another important dimension. Cross-scale response differences refer to the inconsistencies in the structural characteristics (such as energy concentration areas and distribution patterns) of time-frequency energy distribution at different time scales. During normal operation, the energy responses at different scales remain highly coordinated; however, as a fault develops, energy mutations at lower scales (high frequencies) often fail to propagate to higher scales (low frequencies) in a timely manner, leading to a gradual deviation in structural similarity. This cross-scale response difference can sensitively capture early signs of a fault gradually penetrating from the local microscopic level to the overall macroscopic level.

[0044] Based on this, this invention integrates the stability index of energy migration paths, path dispersion characteristics, and the degree of deviation of cross-scale response differences to construct a health evolution factor that can comprehensively characterize the energy redistribution behavior of the system. This health evolution factor is a quantitative indicator that updates in real time over time, and its value and trend directly reflect the stability of energy redistribution within the transformer at the current moment. The closer the value is to the normal baseline, the healthier the system's energy distribution behavior; the greater the deviation, the more unstable the energy redistribution process becomes.

[0045] As an implementable approach, the construction of a health evolution factor characterizing the energy redistribution behavior of a system based on the migration paths and cross-scale response differences of time-frequency energy across different frequency bands includes: modeling the correlation between energy changes in each frequency band at adjacent times to generate an energy migration network describing the direction, intensity, and dynamic migration paths of energy transfer between frequency bands; extracting structural indicators reflecting energy redistribution behavior based on the concentration of energy flow, path dispersion characteristics, and stability of migration paths in the energy migration network; and combining the topological consistency of the energy migration network at different time scales with the degree of deviation of structural similarity formed by comparing cross-scale response differences to jointly form the health evolution factor.

[0046] Specifically, firstly, the energy change relationships between different frequency bands at adjacent time points are modeled to generate an energy migration network describing the direction, intensity, and dynamic migration path of energy transfer between frequency bands. This process treats different frequency bands as network nodes and the energy transfer between adjacent time points as directed edges. Through correlation modeling, the path direction, transfer intensity, and dynamic evolution of energy flowing from one frequency band to another can be clearly depicted. For example, when an early winding defect occurs in a transformer, energy in the high-frequency band may rapidly migrate to the mid-frequency band, forming a clear migration path; while in a healthy state, energy transfer exhibits a dispersed and stable network structure. This energy migration network intuitively reflects the overall topological relationship of energy redistribution within the transformer's multi-physics field.

[0047] Preferably, the step of modeling the correlation between energy changes in each frequency band at adjacent times to generate an energy migration network describing the direction, intensity, and dynamic migration path of energy transfer between frequency bands includes: using different frequency bands as network nodes and using the energy changes of each frequency band between adjacent times as the weights of directed edges; determining the direction and intensity of energy transfer based on the direction and magnitude of energy changes; and constructing a dynamic directed weighted network through the energy transfer relationships at multiple consecutive times to form a complete energy migration network describing the direction, intensity, and dynamic migration path of energy transfer between frequency bands.

[0048] Specifically, firstly, different frequency bands are treated as network nodes, and the energy change of each frequency band between adjacent time points is used as the weight of the directed edges. In this way, each frequency band is considered an independent node in the network, and the connection between two nodes is determined by the actual change in energy flow from one frequency band to another. The magnitude of the weight directly reflects the strength of the energy transfer; a positive weight indicates that energy flows out of the current frequency band, while a negative weight indicates that energy flows into that frequency band. This definition of nodes and directed weighted edges organically organizes the originally isolated frequency band energy data into an interconnected overall network.

[0049] Secondly, the direction and intensity of energy transfer are determined based on the direction and magnitude of energy changes. The direction of energy transfer is determined by the increasing or decreasing trend of energy values ​​at adjacent time points. If the energy in a certain frequency band increases significantly in the next time point, there is an energy inflow path from other frequency bands to that band; conversely, an outflow path is formed. Simultaneously, the absolute magnitude of the energy change is quantified as the weight of the edge; the greater the change, the higher the edge weight, indicating more intense energy migration along that path. This process can accurately capture the dynamic characteristics of rapid energy redistribution between frequency bands during early fault occurrences, such as the phenomenon of high-frequency energy suddenly concentrating and migrating to mid- and low-frequency bands.

[0050] Finally, a dynamic directed weighted network is constructed using energy transfer relationships across multiple consecutive time points to form a complete energy migration network describing the direction, intensity, and dynamic migration path of energy transfer between frequency bands. As time progresses, the edge relationships between nodes are updated at each new time point, giving the network a dynamic evolutionary characteristic. This network not only records the instantaneous energy transfer state at a given moment but, more importantly, preserves the continuous change trajectory of the energy migration path over time, thus comprehensively reflecting the evolutionary pattern of energy redistribution behavior of the transformer at different operating stages.

[0051] Next, based on the concentration of energy flow, path dispersion characteristics, and stability of migration paths in the energy migration network, structural indicators reflecting energy redistribution behavior are extracted. Energy flow concentration measures whether energy is excessively concentrated in a few frequency bands or paths; higher concentration often indicates a more severe imbalance in system energy distribution. Path dispersion characteristics describe the diversity and uniformity of energy transfer paths; reduced dispersion usually indicates restricted energy flow due to faults. The stability of migration paths is quantified by the magnitude of path changes over consecutive time intervals; decreased stability indicates that the system is on the verge of dynamic instability. These structural indicators together constitute a multi-faceted quantitative assessment of the stability of energy redistribution behavior.

[0052] Finally, the health evolution factor is formed by combining the topological consistency of the energy migration network at different time scales and comparing the degree of structural similarity deviation caused by cross-scale response differences. At different scales, the topological structure of the energy migration network should maintain a certain degree of coordination. When this coordination is disrupted, cross-scale response differences arise. By calculating the degree of structural similarity deviation, the impact of this inconsistency on the overall energy redistribution stability can be further quantified. By fusing the structural indicators of the energy migration network with the cross-scale consistency disruption indicators, a comprehensive health evolution factor is obtained. The higher the value of this factor, the more stable the transformer energy redistribution behavior; the lower the value or the more drastic the changes, the more the system's health is gradually deteriorating.

[0053] The health evolution factor constructed through the above steps not only fully integrates the dynamic information of energy migration between frequency bands and the differences in cross-scale response, but also has a clear physical meaning. It can sensitively reflect the early process of the transformer's evolution from steady-state operation to potential fault state, providing a reliable and continuous quantitative basis for the subsequent generation of health state trajectories and the construction of sliding mode manifolds.

[0054] Connecting the health evolution factor values ​​calculated at consecutive moments in chronological order forms the health state trajectory. This trajectory clearly depicts the complete evolutionary path of the transformer from steady-state operation to potential instability in time-health space. It not only includes the current health level but also reflects the dynamic trend of health status over time, providing a continuous and complete kinematic basis for subsequently introducing historical memory information, constructing path-dependent sliding mode manifolds, and analyzing the adhesion, deviation, and crossing behaviors of the trajectory.

[0055] Through this technical feature, the present invention successfully transforms abstract time-frequency energy information into a visualized health evolution trajectory with clear physical meaning, thereby elevating transformer condition monitoring from a single feature judgment to a systematic characterization of a multi-stage dynamic evolution process, and significantly improving the timeliness and accuracy of early fault warning.

[0056] The following describes in detail step 103, namely, "introducing the accumulated memory information of health evolution factors within a historical period to construct a nonlinear health-driven sliding manifold with path-dependent characteristics, wherein the sliding manifold is used to characterize the multi-stage dynamic boundary of the system during the evolution from steady state to unstable state," with reference to the embodiments.

[0057] While health evolution factors can reflect the energy redistribution state of a system at a given moment, their values ​​at a single moment are insufficient to capture the historical cumulative effects and path differences in fault development. Therefore, this invention first introduces accumulated memory information of health evolution factors over a historical period. By performing time-weighted accumulation processing on a continuous sequence of health evolution factors over a past period, a memory state vector reflecting the historical evolutionary trajectory is generated. This memory mechanism ensures that the current health state is no longer isolated but is closely related to the accumulated damage under different evolutionary paths in the past, thus endowing health state assessment with "memory" capabilities and avoiding the limitations of traditional methods that rely solely on current characteristics.

[0058] Building upon this foundation, this invention constructs a nonlinear health-driven sliding manifold with path-dependent characteristics. The sliding manifold is no longer a simple linear or fixed boundary as in traditional control theory, but rather a nonlinear surface driven by both the current health evolution factor and the historical memory state vector. This manifold can automatically adjust its shape and position according to different historical evolution paths, exhibiting a clear path-dependent characteristic. For example, for the same current health evolution factor value, if its historical accumulation path is different, the resulting sliding manifold boundary will also show significant differences. This path-dependent characteristic allows the manifold to more realistically simulate the physical law that "different damage processes lead to different critical states" during the actual degradation process of transformers.

[0059] As an implementable approach, the method of introducing accumulated memory information of health evolution factors within a historical time period to construct a nonlinear health-driven sliding manifold with path-dependent characteristics includes: acquiring a sequence of health evolution factors for consecutive historical moments within a preset time window, and performing time-weighted accumulation processing on the health evolution factor sequence to generate a memory state vector reflecting the historical evolution trajectory; constructing an extended state variable containing historical dependencies based on the coupling relationship between the memory state vector and the health evolution factors at the current moment; performing nonlinear mapping processing on the extended state variable to generate a corresponding sliding manifold expression, and introducing path-related constraint terms into the sliding manifold to characterize the influence of different historical evolution paths on the current state boundary; and constraining the health state trajectory according to the sliding manifold expression to make the system state exhibit differentiated convergence behavior under different historical evolution conditions.

[0060] Specifically, firstly, a sequence of health evolution factors for consecutive historical moments within a preset time window is obtained, and this sequence is then subjected to time-weighted accumulation processing to generate a memory state vector reflecting the historical evolution trajectory. This process assigns higher weights to historical data closer to the current moment and gradually reduces the weights to earlier historical data, thus forming a state vector that reflects both recent evolutionary trends and preserves long-term accumulated memory. This memory state vector effectively records the complete evolutionary path of the transformer's energy redistribution behavior over a past period, avoiding the shortcomings of relying solely on current health evolution factors while ignoring the cumulative effects of historical damage.

[0061] Next, based on the coupling relationship between the memory state vector and the current health evolution factor, an extended state variable incorporating historical dependencies is constructed. By organically coupling historical memory information with the current health state, the extended state variable not only includes the current health information but also incorporates the influencing factors of the historical evolution path. This extension expands the system state description from a single dimension to a two-dimensional or even multi-dimensional space that simultaneously considers the current and historical states, providing a rich input foundation for subsequent nonlinear mappings.

[0062] Then, a nonlinear mapping process is performed on the extended state variables to generate the corresponding sliding manifold expression. Path-dependent constraint terms are introduced into the sliding manifold to characterize the influence of different historical evolution paths on the current state boundary. The nonlinear mapping transforms the linear extended state into a sliding manifold with a complex surface morphology, while the path-dependent constraint terms dynamically adjust the boundary shape of the manifold according to the different characteristics of historical memory. For example, long-term, slowly accumulating damage paths will make the manifold boundary smoother, while short-term, violently impacted paths will make the boundary steeper. This path-dependent constraint ensures that the sliding manifold can accurately reflect the differences in the critical conditions for the system to reach instability under different degradation trajectories.

[0063] Finally, the health state trajectory is constrained according to the sliding manifold expression, causing the system state to exhibit differentiated convergence behavior under different historical evolution conditions. Under the same current health evolution factor value, if the historical evolution paths are different, the convergence trend, adsorption strength, and traversal difficulty of the health state trajectory on the sliding manifold will also show significant differences. This differentiated convergence behavior allows the method to more accurately distinguish between health states that "appear identical but have different actual evolutionary risks," providing a reliable dynamic basis for subsequent identification of fault triggering conditions and health level classification.

[0064] The specific implementation can be expressed in the following mathematical form: Let the health evolution factor at the current moment be... The sequence of health evolution factors for consecutive historical moments within the preset time window is as follows: ,but: The memory state vector is defined as: in This is a time decay factor, reflecting that historical data closer to the current moment has a higher weight.

[0065] The extended state variables, which include historical dependencies, are constructed as follows: The expression for the nonlinear health-driven sliding mode manifold is: in, It is a nonlinear coupling mapping function between the memory state vector and the current health evolution factor; These are path-related constraints used to characterize the impact of different historical evolution paths on the current manifold boundary; and The adjustment coefficient is positive.

[0066] System status based on The sign and rate of change exhibit differentiated convergence behavior: when and When, the trajectory approaches and adheres to the manifold; when Continuously increasing and At that time, the trajectory may deviate or even cross the manifold boundary.

[0067] The primary function of the sliding mode manifold is to characterize the multi-stage dynamic boundaries of the system's evolution from steady to unstable states. It divides the transformer's healthy evolution space into several clearly defined regions: the steady-state adsorption region, the slow deviation region, the critical crossing region, and the instability escape region. By defining these dynamic boundaries, the system can clearly distinguish between multiple stages: normal operation, early latent faults, intermediate-stage developing faults, and impending severe faults. This provides a unified geometric framework and judgment benchmark for subsequent analysis of the adsorption, deviation, and crossing behaviors of the healthy state trajectory on the manifold.

[0068] Furthermore, the construction of extended state variables containing historical dependencies based on the coupling relationship between the memory state vector and the current-moment health evolution factor includes: introducing a non-consistent decay mechanism for information from different historical stages in the memory state vector; assigning different memory retention weights according to the differences in the degree of influence of each historical stage on the current state to form hierarchical memory sub-states; combining and mapping the hierarchical memory sub-states with the current-moment health evolution factor to generate multiple candidate extended state components to represent the state response under different potential evolution paths; introducing a competitive selection mechanism among the multiple candidate extended state components; determining the dominant extended state variable by comparing the consistency between each candidate component and the current observation trajectory; and locally reconstructing the sliding manifold based on the dominant extended state variable to make different historical evolution paths correspond to differentiated state evolution boundaries.

[0069] Specifically, let the memory state vector be... After dividing the historical data into several historical stages according to different time scales, a non-uniform decay factor is introduced, and different memory retention weights are assigned to each stage. Thus, hierarchical memory sub-states are obtained. The weight The weighting is dynamically adjusted based on the time distance and degree of influence between historical periods and the current moment. More recent and more influential periods retain higher weights, while more distant or less influential periods decay rapidly.

[0070] Next, the hierarchical memory sub-states are combined and mapped with the current-time health evolution factor to generate multiple candidate extended state components, each representing the state response under different potential evolutionary paths. Through a nonlinear combination mapping function, each layer of memory sub-states is... With current health evolution factors By fusing the components, multiple candidate extended state components are obtained. These are nonlinear mapping functions designed for different paths. Each of these candidate extended state components corresponds to a potential historical evolution path, reflecting the state response characteristics that the system may exhibit under different damage accumulation histories.

[0071] Then, a competitive selection mechanism is introduced among the multiple candidate extended state components. By comparing the consistency between each candidate component and the current observed trajectory, the dominant extended state variable is determined. Specifically, each candidate extended state component is calculated. The candidate component with the highest consistency between the similarity or error measure between the observed health state trajectory and the actual health state trajectory is selected as the dominant extended state variable. This competitive selection process ensures that the current system state description prioritizes historical path information that best matches the actual evolution trajectory.

[0072] Finally, the sliding manifold is locally reconstructed based on the dominant extended state variables, so that different historical evolution paths correspond to differentiated state evolution boundaries. This is based on the dominant extended state variables. Local updates to the boundary parameters in the sliding manifold expression enable adaptive reconstruction of the manifold. Different dominant state variables correspond to different historical evolution paths, leading to differentiated boundary shapes and convergence characteristics in the sliding manifold. Consequently, the same current health evolution factor corresponds to different instability critical conditions under different paths.

[0073] By introducing historical accumulated memory and constructing a path-dependent nonlinear health-driven sliding manifold, this invention successfully elevates the transformer health state from a static indicator to a nonlinear dynamic process with dynamic memory and path sensitivity. This significantly enhances the method's ability to identify complex, slowly accumulating faults early and improves the accuracy of evolution trend prediction, providing a powerful technical means for achieving true transformer health management and fault prediction.

[0074] The following describes in detail step 104, namely, "defining the evolution approach direction and escape criterion on the sliding manifold, and identifying potential fault triggering conditions by analyzing the adsorption, deviation and crossing behavior of the sliding manifold through the analysis of the health state trajectory".

[0075] This step places the health status trajectory within a constructed nonlinear sliding manifold for dynamic analysis, thereby transforming the abstract health evolution process into quantifiable geometric behavior judgments.

[0076] First, the healthy state trajectory is mapped onto the sliding manifold, and its local tangential and normal components in the manifold's tangential space are calculated. The evolution approach direction is determined based on the changing trend of the tangential component. The tangential component reflects the direction and velocity of the healthy state trajectory sliding along the sliding manifold surface, while the normal component characterizes the degree of vertical deviation of the trajectory relative to the manifold surface. When the tangential component shows a stable tendency to point towards the manifold's attraction region, it indicates that the system's healthy state is gradually approaching the steady-state boundary; conversely, if the direction of the tangential component changes, it suggests that the system may be deviating from its original stable path.

[0077] Secondly, an escape criterion function is constructed based on the deviation distance of the healthy state trajectory from the normal manifold and its rate of change. This escape criterion function comprehensively considers both the magnitude of the trajectory's deviation from the manifold and the speed of the deviation, forming a comprehensive index that can quantify the risk of system instability in real time. The larger the normal deviation distance and the faster its rate of change, the more rapidly the healthy state trajectory is moving away from the stable manifold region, and the higher the risk of system instability.

[0078] When the normal deviation distance continues to increase and its rate of change exceeds a preset threshold, it is determined that the healthy state trajectory has escaped. At this point, the system is considered to have broken through the stability boundary defined by the sliding mode manifold and is about to enter an unstable state. This escape determination is an important basis for identifying potential fault triggering conditions and can issue early warnings before transformer faults are fully manifested.

[0079] Furthermore, when the healthy state trajectory undergoes a reversal of direction or a discontinuous change in the tangential component, it is identified as a sudden change in the evolution path, which helps determine the transition of the transformer from a stable state to an unstable state. A reversal of direction means that the healthy state trajectory suddenly changes its original evolution direction, while a discontinuous change in the tangential component often corresponds to a relatively drastic state transition within the system, such as local insulation breakdown or rapid propagation of winding short circuits. These abrupt change characteristics, in conjunction with the escape criterion, can more comprehensively and timely capture the critical moment when the transformer transitions from stable operation to a fault state.

[0080] Let the health status trajectory be at time... The state is The corresponding sliding mode manifold function is The process for determining the evolution approach direction and escape criterion on a sliding manifold in this invention is as follows: First, the health state trajectory is mapped onto the sliding manifold, and its local tangential components in the tangential space of the manifold are calculated. and normal components The direction of evolution is determined by the changing trend of the tangential component: when Furthermore, as the value continues to decrease, the trajectory of a healthy state is determined to be approaching and adsorbing onto a stable boundary along the manifold surface; when When the sign of the t axis changes or its value increases, it indicates that the trajectory begins to deviate from the steady-state path.

[0081] Secondly, define the deviation distance of the trajectory from the normal of the sliding manifold. And calculate its rate of change. Based on this, construct the escape criterion function: in This is an adjustment coefficient used to balance the effects of deviation distance and rate of change.

[0082] Then, fault triggering is determined based on the escape criterion function: when the normal deviation distance... The value continues to increase, and the escape criterion function satisfies... (in When the threshold is set (preset), the system is deemed to have escaped the health status trajectory, enters an unstable state, and triggers a potential fault alarm.

[0083] At the same time, when the health status trajectory reverses direction (i.e.) The sign changes in successive time intervals) or the tangential component undergoes discontinuous abrupt changes (i.e., the ...). ,in When the threshold value is reached (i.e., a mutation occurs), it is identified as a mutation in the evolutionary path. This mutation feature works in conjunction with the escape criterion to help determine the rapid transition of a transformer from a stable state to an unstable state.

[0084] Through the judgment process expressed by the above formula, the adsorption, deviation and crossing behavior of the health state trajectory on the sliding manifold are transformed into quantifiable mathematical conditions, which can identify potential fault triggering conditions in real time and accurately, and realize early warning of multi-stage fault evolution of transformers.

[0085] By defining the evolution approach direction and escape criterion on the sliding mode manifold and comprehensively analyzing the adsorption, deviation and crossing behaviors of the health state trajectory, this invention achieves accurate identification of the multi-stage fault evolution process of transformers, greatly improves the detection sensitivity and reliability of early fault triggering conditions, and provides a scientific basis for transformer health management and operation and maintenance decisions.

[0086] The following describes in detail step 104, namely, "outputting the transformer health status monitoring result based on the dynamic interaction between the health status trajectory and the sliding mode manifold," with reference to an embodiment.

[0087] This step transforms the complex nonlinear dynamic process constructed in the early stage into intuitive and reliable health status monitoring results by comprehensively statistically analyzing the overall behavior of the health status trajectory on the sliding manifold, providing decision-making basis for operation and maintenance personnel.

[0088] Specifically, the residence time ratio and cross-region migration frequency of the healthy state trajectory within different regions of the sliding mode manifold are statistically analyzed to construct distributional characteristics reflecting the stability of the state. The residence time ratio measures the length of time the trajectory stays in the steady-state adsorption region; a higher ratio indicates a more stable system operation. The cross-region migration frequency reflects the frequency with which the trajectory switches between different dynamic boundaries; an increase in frequency indicates that the system's health state is in a fluctuating or degrading stage. These distributional characteristics together constitute a quantitative description of the transformer's current overall health level.

[0089] Next, based on the distribution characteristics and the adsorption, deviation, and crossing behavior of the health state trajectory on the sliding manifold, corresponding health state levels are defined. This invention typically classifies health states into four levels: normal, attention, warning, and fault. When the trajectory is stably adsorbed to the steady-state region of the sliding manifold for a long period and the deviation distance is very small, it is determined to be at the normal level; when the trajectory begins to deviate slowly but has not yet reached the escape threshold, it is determined to be at the attention level; when the escape criterion is triggered and the trajectory frequently crosses the boundary, it is determined to be at the warning or fault level. This level classification based on dynamic interaction relationships reflects the continuous evolution of the transformer's health state better than the traditional single threshold method.

[0090] Finally, the obtained health status levels and the fault development trends obtained by matching the trajectory evolution direction sequence with a preset fault evolution pattern library are output as the final transformer health status monitoring results. This output not only includes the current health level but also provides prediction information and confidence levels for fault evolution paths, making the monitoring results both real-time and forward-looking.

[0091] By outputting monitoring results based on the dynamic interaction between the health status trajectory and the sliding mode manifold, this invention achieves a complete closed loop from low-level signal processing to high-level health decision-making, significantly improving the interpretability, accuracy, and practical value of transformer condition monitoring, and providing scientific and reliable technical support for predictive maintenance and health management.

[0092] To further illustrate the technical effects of this application, a specific implementation method and the test results of this implementation method are given below.

[0093] A 110kV, 50MVA oil-immersed power transformer was used as an example for specific implementation and testing. In the field or on a simulation test platform, a high-precision current transformer (sampling rate 20kHz, bandwidth above 10kHz) installed at the transformer's neutral point was used to acquire the neutral point current signal in real time. The acquired signal first underwent adaptive segmentation processing, dynamically adjusting the window length and overlap ratio based on differences in energy change rate and spectral distribution. Subsequently, multi-resolution time-frequency decomposition was performed to obtain multi-scale time-frequency energy distribution, and a unified benchmark multi-scale energy distribution was formed through normalization alignment and consistency correction.

[0094] Based on the above distribution, an energy migration network is constructed, and structural indices and cross-scale response differences of migration paths between frequency bands are extracted. These are then used to generate health evolution factors, which are connected sequentially over time to form health state trajectories. Simultaneously, accumulated memory information of health evolution factors from historical periods is introduced to construct a nonlinear health-driven sliding manifold with path-dependent characteristics. By analyzing the adsorption, deviation, and crossing behaviors of health state trajectories on the sliding manifold, and combining this with an escape criterion function, real-time outputs health state levels (normal, attention, warning, and fault levels) and fault evolution path prediction results are provided.

[0095] In the simulation experiment, an equivalent transformer model was built using MATLAB / Simulink to simulate normal operation and winding inter-turn short-circuit faults of different severity (short-circuit turns accounting for 1%, 2%, 3%, and 5%, respectively). Test results show that under normal operation, the healthy trajectory remains stably adhered to the steady-state region of the sliding mode manifold, maintaining a "normal" health level with no false alarms. When a 1% inter-turn short-circuit fault is injected, this method can detect the slow trajectory deviation and issue a "caution" warning approximately 45 days after the fault occurs, about 3-4 months earlier than the traditional oil-gas three-ratio method. When the number of short-circuit turns reaches 2%, the warning time is further advanced to approximately 20 days after the fault, issuing a "warning" level. When the number of short-circuit turns exceeds 3%, the trajectory rapidly crosses the manifold boundary, and the system accurately triggers the "fault" level and provides a high-confidence evolution path prediction within approximately 7 days after the fault occurs.

[0096] A live 110kV transformer was selected for on-site grid connection testing and continuously monitored for 72 hours under different load fluctuations and harmonic interference environments. The test results show that the proposed method maintains stable operation even under complex operating conditions with a signal-to-noise ratio as low as 8dB. The detection accuracy for early, minor faults reaches 94.5%, with a false alarm rate of less than 3.2%. Compared with traditional vibration monitoring and partial discharge monitoring, the early warning time is advanced by an average of 2.8 months. Furthermore, the memory mechanism based on historical path dependence significantly improves the distinction between state boundaries under different cumulative damage histories, verifying the robustness and engineering applicability of the method.

[0097] The above-described implementation methods and test results fully demonstrate that the transformer condition monitoring method based on neutral point current proposed in this invention can effectively achieve dynamic characterization of the multi-stage evolution process of the transformer from steady state to unstable state. It has outstanding advantages such as non-intrusiveness, strong early sensitivity, and good interpretability, providing reliable technical support for transformer health management and predictive maintenance.

[0098] The methods provided in this application can be applied to various application scenarios, including but not limited to: in large substations and hub substations, they can serve as an online health management system for oil-immersed power transformers with voltage levels of 110kV and above, enabling early warning of latent faults such as early winding inter-turn short circuits, multi-point grounding of the core, and insulation degradation, providing a basis for decision-making in the preparation of predictive maintenance plans; in grid-connected transformer scenarios for new energy sources such as wind farms and photovoltaic power stations, due to severe load fluctuations and high harmonic content, this method can effectively suppress interference using neutral point current signals, achieving stable monitoring under complex operating conditions and ensuring the reliable operation of new energy equipment; in user substations of industrial parks or important users, they can serve as a low-cost, non-intrusive retrofit solution, quickly deployed on existing transformers without power outages to install additional sensors, thereby improving the health management level of existing equipment and significantly reducing the risk of power outages caused by sudden faults.

[0099] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0100] According to another embodiment, a transformer condition monitoring system is provided. Figure 2 A schematic block diagram of the transformer condition monitoring system according to one embodiment is shown. Figure 2 As shown, the device 200 includes: The time-frequency energy distribution acquisition unit 201 is configured to acquire the neutral point current signal of the transformer and perform time-frequency decomposition on the current signal to obtain a multi-scale time-frequency energy distribution.

[0101] The health state trajectory generation unit 202 is configured to construct health evolution factors characterizing the energy redistribution behavior of the system based on the migration path of time-frequency energy in different frequency bands and the cross-scale response differences, and connect the health evolution factors in chronological order to generate a health state trajectory.

[0102] The sliding manifold building unit 203 is configured to introduce the accumulated memory information of health evolution factors within a historical period to construct a nonlinear health-driven sliding manifold with path-dependent characteristics, wherein the sliding manifold is used to characterize the multi-stage dynamic boundary of the system during the evolution process from steady state to unstable state.

[0103] The fault triggering condition identification unit 204 is configured to define the evolution approach direction and escape criterion on the sliding manifold, and identify potential fault triggering conditions by analyzing the adsorption, deviation and crossing behavior of the healthy state trajectory on the sliding manifold.

[0104] The status monitoring result acquisition unit 205 is configured to output the transformer health status monitoring result based on the dynamic interaction between the health status trajectory and the sliding mode manifold.

[0105] As an implementable approach, when the time-frequency energy distribution acquisition unit 201 performs time-frequency decomposition on the current signal to obtain a multi-scale time-frequency energy distribution, it can be configured to: perform adaptive segmentation processing on the current signal, dynamically adjust the decomposition window length and overlap ratio according to the energy change rate and spectral distribution differences of the signal in each time segment; perform multi-resolution time-frequency decomposition processing in each segment to obtain the local time-frequency energy distribution at the corresponding time scale; normalize and align the time-frequency energy distributions at different time scales and construct a cross-scale energy mapping relationship; and perform consistency correction on the multi-scale time-frequency energy distribution based on the cross-scale energy mapping relationship to form a multi-scale time-frequency energy distribution result with a unified reference benchmark.

[0106] As an implementable approach, the health status trajectory generation unit 202, when constructing a health evolution factor characterizing the energy redistribution behavior of the system based on the migration paths and cross-scale response differences of time-frequency energy across different frequency bands, can be configured as follows: It performs correlation modeling on the energy change relationships of each frequency band at adjacent times, generating an energy migration network describing the direction, intensity, and dynamic migration paths of energy transfer between frequency bands; based on the concentration of energy flow, path dispersion characteristics, and stability of migration paths in the energy migration network, it extracts structural indicators reflecting energy redistribution behavior; and combines the topological consistency of the energy migration network at different time scales, and by comparing the degree of structural similarity deviation formed by cross-scale response differences, it jointly forms the health evolution factor.

[0107] As an feasible approach, the health status trajectory generation unit 202 can be configured to, when modeling the correlation between energy changes in each frequency band at adjacent times and generating an energy migration network describing the direction, intensity, and dynamic migration path of energy transfer between frequency bands, use different frequency bands as network nodes and the amount of energy change in each frequency band between adjacent times as the weight of the directed edges; determine the direction and intensity of energy transfer based on the direction and magnitude of energy changes; and construct a dynamic directed weighted network through the energy transfer relationships at multiple consecutive times to form a complete energy migration network describing the direction, intensity, and dynamic migration path of energy transfer between frequency bands.

[0108] As an implementable approach, the sliding manifold construction unit 203, when constructing a nonlinear health-driven sliding manifold with path-dependent characteristics by introducing accumulated memory information of health evolution factors within a historical time period, can be configured as follows: It acquires a sequence of health evolution factors from consecutive historical moments within a preset time window and performs time-weighted accumulation processing on the health evolution factor sequence to generate a memory state vector reflecting the historical evolution trajectory; based on the coupling relationship between the memory state vector and the health evolution factors at the current moment, it constructs an extended state variable containing historical dependencies; it performs nonlinear mapping processing on the extended state variable to generate a corresponding sliding manifold expression, and introduces path-related constraint terms into the sliding manifold to characterize the influence of different historical evolution paths on the current state boundary; it constrains the health state trajectory according to the sliding manifold expression, causing the system state to exhibit differentiated convergence behavior under different historical evolution conditions.

[0109] As an implementable approach, when the sliding manifold construction unit 203 constructs extended state variables containing historical dependencies based on the coupling relationship between the memory state vector and the current-moment health evolution factor, it can be configured as follows: A non-uniform decay mechanism is introduced into the information of different historical stages in the memory state vector; different memory retention weights are assigned according to the differences in the degree of influence of each historical stage on the current state, forming hierarchical memory sub-states; the hierarchical memory sub-states are combined and mapped with the current-moment health evolution factor to generate multiple candidate extended state components, which respectively characterize the state response under different potential evolution paths; a competitive selection mechanism is introduced among the multiple candidate extended state components, and the dominant extended state variable is determined by comparing the consistency degree between each candidate component and the current observation trajectory; and the sliding manifold is locally reconstructed based on the dominant extended state variable, so that different historical evolution paths correspond to differentiated state evolution boundaries.

[0110] As an implementable approach, the fault triggering condition identification unit 204 defines the evolution approach direction and escape criterion on the sliding manifold. When identifying potential fault triggering conditions by analyzing the adsorption, deviation, and crossing behavior of the healthy state trajectory on the sliding manifold, it can be configured to: map the healthy state trajectory onto the sliding manifold, calculate its local tangential and normal components in the manifold tangential space, and determine the evolution approach direction based on the changing trend of the tangential component; construct an escape criterion function based on the normal deviation distance of the healthy state trajectory relative to the sliding manifold and its rate of change; determine that the healthy state trajectory has escaped when the normal deviation distance continues to increase and its rate of change exceeds a preset threshold; and identify the evolution path abrupt change when the healthy state trajectory reverses direction or the tangential component changes discontinuously, in order to assist in judging the transformation of the transformer from a stable state to an unstable state.

[0111] As an implementable approach, when the state monitoring result acquisition unit 205 outputs the transformer health state monitoring result based on the dynamic interaction between the health state trajectory and the sliding manifold, it can be configured to: statistically analyze the proportion of the residence time of the health state trajectory in different regions of the sliding manifold and the frequency of cross-region migration to construct a distribution characteristic quantity reflecting the state stability; based on the distribution characteristic quantity and the trajectory's adsorption, deviation, and crossing behavior on the sliding manifold, classify the corresponding health state level; and output the health state level as the health state monitoring result.

[0112] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. Components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0113] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0114] In addition, embodiments of this application also provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method described in any of the foregoing method embodiments.

[0115] And an electronic device, comprising: One or more processors; and A memory associated with the one or more processors, the memory being used to store program instructions that, when read and executed by the one or more processors, perform the steps of the method described in any of the foregoing method embodiments.

[0116] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in any of the foregoing method embodiments.

[0117] in, Figure 3 The architecture of an electronic device is illustrated, which may include a processor 310, a video display adapter 311, a disk drive 312, an input / output interface 313, a network interface 314, and a memory 320. The processor 310, video display adapter 311, disk drive 312, input / output interface 313, network interface 314, and memory 320 can communicate with each other via a communication bus 330.

[0118] The processor 310 can be implemented using a general-purpose CPU, microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits to execute relevant programs in order to implement the technical solution provided in this application.

[0119] The memory 320 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 320 can store the operating system 321 for controlling the operation of the electronic device 300, and the basic input / output system (BIOS) 322 for controlling the low-level operations of the electronic device 300. Additionally, it can store a web browser 323, a data storage management system 324, and a transformer condition monitoring system 325, etc. The aforementioned transformer condition monitoring system 325 can be the application program that specifically implements the aforementioned steps in this embodiment. In summary, when implementing the technical solution provided in this application through software or firmware, the relevant program code is stored in the memory 320 and executed by the processor 310.

[0120] Input / output interface 313 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touch screens, microphones, various sensors, etc., and output devices may include displays, speakers, vibrators, indicator lights, etc.

[0121] Network interface 314 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0122] Bus 330 includes a pathway for transmitting information between various components of the device, such as processor 310, video display adapter 311, disk drive 312, input / output interface 313, network interface 314, and memory 320.

[0123] It should be noted that although the above-described device only shows the processor 310, video display adapter 311, disk drive 312, input / output interface 313, network interface 314, memory 320, bus 330, etc., in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the solution of this application, and does not necessarily include all the components shown in the figures.

[0124] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer program product. This computer program product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0125] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method of transformer condition monitoring, characterized by, The method includes: The neutral point current signal of the transformer is acquired, and time-frequency decomposition is performed on the current signal to obtain the multi-scale time-frequency energy distribution; Based on the migration path of time-frequency energy in different frequency bands and the differences in cross-scale response, a health evolution factor characterizing the energy redistribution behavior of the system is constructed. The health evolution factors are connected in chronological order to generate a health state trajectory. By introducing the accumulated memory information of health evolution factors within a historical period, a nonlinear health-driven sliding manifold with path-dependent characteristics is constructed, wherein the sliding manifold is used to characterize the multi-stage dynamic boundary of the system during the evolution process from steady state to unstable state; Evolution approach direction and escape criterion are defined on the sliding manifold. Potential fault triggering conditions are identified by analyzing the adsorption, deviation and crossing behavior of the healthy state trajectory on the sliding manifold. Based on the dynamic interaction between the health status trajectory and the sliding mode manifold, the transformer health status monitoring results are output.

2. The method of claim 1, wherein, Perform time-frequency decomposition on the current signal to obtain a multi-scale time-frequency energy distribution, including: The current signal is subjected to adaptive segmentation processing, and the decomposition window length and overlap ratio are dynamically adjusted according to the energy change rate and spectral distribution differences of the signal in each time segment. Multi-resolution time-frequency decomposition processing is performed in each segment to obtain the local time-frequency energy distribution at the corresponding time scale. Normalize and align the time-frequency energy distribution at different time scales, and construct cross-scale energy mapping relationships; Based on the aforementioned cross-scale energy mapping relationship, consistency correction is performed on the multi-scale time-frequency energy distribution to form a multi-scale time-frequency energy distribution result with a unified reference benchmark.

3. The method of claim 1, wherein, Based on the migration paths of time-frequency energy across different frequency bands and the differences in cross-scale responses, health evolution factors characterizing the energy redistribution behavior of the system are constructed, including: The energy change relationship of each frequency band at adjacent times is modeled to generate an energy migration network that describes the energy transfer direction, intensity and dynamic migration path between frequency bands. Based on the concentration of energy flow, path dispersion characteristics, and stability of migration paths in the energy migration network, structural indicators reflecting energy redistribution behavior are extracted. The health evolution factor is formed by combining the topological consistency of the energy migration network at different time scales and comparing the degree of structural similarity deviation formed by cross-scale response differences.

4. The method of claim 3, characterized in that, The energy change relationships between different frequency bands at adjacent times are correlated and modeled to generate an energy migration network describing the direction, intensity, and dynamic migration path of energy transfer between frequency bands, including: Different frequency bands are used as network nodes, and the energy change of each frequency band between adjacent time points is used as the weight of the directed edge. The direction and intensity of energy transfer are determined based on the direction and magnitude of the energy change; A dynamic directed weighted network is constructed by analyzing the energy transfer relationships at multiple consecutive time points to form a complete energy migration network that describes the direction, intensity, and dynamic migration path of energy transfer between frequency bands.

5. The method of claim 1, wherein, By incorporating accumulated memory information of health evolution factors over historical periods, a nonlinear health-driven sliding manifold with path-dependent characteristics is constructed, including: Obtain the sequence of health evolution factors for consecutive historical moments within a preset time window, and perform time-weighted accumulation processing on the health evolution factor sequence to generate a memory state vector that reflects the historical evolution trajectory. Based on the coupling relationship between the memory state vector and the current health evolution factor, an extended state variable containing historical dependencies is constructed. The extended state variables are subjected to nonlinear mapping to generate the corresponding sliding manifold expression, and path-related constraint terms are introduced into the sliding manifold to characterize the influence of different historical evolution paths on the current state boundary. The healthy state trajectory is constrained according to the sliding manifold expression, so that the system state exhibits differentiated convergence behavior under different historical evolution conditions.

6. The method of claim 5, wherein, Based on the coupling relationship between the memory state vector and the current health evolution factor, an extended state variable containing historical dependencies is constructed, including: A non-consistent decay mechanism is introduced for the information of different historical stages in the memory state vector. Different memory retention weights are assigned according to the difference in the degree of influence of each historical stage on the current state, forming a hierarchical memory sub-state. The hierarchical memory substates are combined and mapped with the current time health evolution factor to generate multiple candidate extended state components, which respectively characterize the state response under different potential evolution paths; A competitive selection mechanism is introduced among the multiple candidate extended state components. The dominant extended state variable is determined by comparing the consistency between each candidate component and the current observed trajectory. Based on the dominant extended state variables, the sliding manifold is locally reconstructed so that different historical evolution paths correspond to differentiated state evolution boundaries.

7. The method of claim 1, wherein, An evolution approach direction and escape criterion are defined on the sliding manifold. By analyzing the adhesion, deviation, and crossing behavior of the healthy state trajectory on the sliding manifold, potential fault triggering conditions are identified, including: The health status trajectory is mapped onto the sliding manifold, its local tangential and normal components in the tangential space of the manifold are calculated, and the evolution approach direction is determined based on the changing trend of the tangential components. Based on the deviation distance of the health state trajectory from the normal of the sliding manifold and its rate of change, an escape criterion function is constructed; When the normal deviation distance continues to increase and its rate of change exceeds a preset threshold, it is determined that the health status trajectory has escaped. When the health state trajectory reverses direction or the tangential component changes discontinuously, it is identified as a sudden change in the evolution path, which is used to help determine the transformation of the transformer from a stable state to an unstable state.

8. The method of claim 1, wherein, Based on the dynamic interaction between the health status trajectory and the sliding mode manifold, the transformer health status monitoring results are output, including: The proportion of dwell time and frequency of cross-region migration of the health state trajectory in different regions of the sliding manifold are statistically analyzed to construct a distribution characteristic quantity reflecting the stability of the state. Based on the distribution characteristics and the trajectory's adsorption, deviation, and crossing behavior on the sliding manifold, the corresponding health status levels are classified. The health status level is used as the health status monitoring result and output.

9. A transformer condition monitoring system characterised in that, The system includes: The time-frequency energy distribution acquisition unit is configured to acquire the transformer neutral point current signal and perform time-frequency decomposition on the current signal to obtain a multi-scale time-frequency energy distribution. The health state trajectory generation unit is configured to construct health evolution factors characterizing the energy redistribution behavior of the system based on the migration path of time-frequency energy in different frequency bands and the cross-scale response differences, and connect the health evolution factors in time order to generate a health state trajectory. The sliding manifold building unit is configured to introduce the accumulated memory information of health evolution factors within a historical period to construct a nonlinear health-driven sliding manifold with path-dependent characteristics, wherein the sliding manifold is used to characterize the multi-stage dynamic boundary of the system during the evolution from steady state to unstable state. The fault triggering condition identification unit is configured to define the evolution approach direction and escape criterion on the sliding manifold, and identify potential fault triggering conditions by analyzing the adsorption, deviation and crossing behavior of the healthy state trajectory on the sliding manifold. The condition monitoring result acquisition unit is configured to output the transformer health condition monitoring result based on the dynamic interaction between the health condition trajectory and the sliding mode manifold.

10. A computer readable electronic medium having stored thereon a computer program, characterized in that, When the program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 8.

Images