Pole-mounted transformer state evaluation and fault early warning method fusing multi-source data

Abstract

The invention discloses a pole-mounted transformer state evaluation and fault early warning method fusing multi-source data, and relates to the technical field of intelligent power grid monitoring, and the method comprises the steps: executing dual state judgment according to a multi-source fusion feature set, extracting a statistical deviation degree of operation fluctuation abnormity, and calculating a physical deviation degree of multi-physical field coupling imbalance; the statistical deviation degree and the physical deviation degree are fused in a double-domain cooperation mode, and a real-time health index is obtained; the degradation rate and multi-scale energy characteristics of the real-time health index are extracted, a potential degradation aggregation mode is identified by referring to a historical slow-varying baseline, and a hidden danger trend factor is generated; performing composite attribution matching on the hidden danger trend factor and the multi-source fusion feature set, executing grading risk criteria according to a multi-physical field anomaly mechanism chain, and outputting an early warning grade and an early warning reason; according to the method, diagnosis is carried out by embedding a multi-physical field anomaly mechanism chain, and the physical root and evolution stage of the fault can be accurately positioned.

Description

Technical Field

[0001] This invention relates to the field of smart grid monitoring technology, and in particular to a method for assessing the condition of pole-mounted transformers and providing early warning of faults by integrating multi-source data. Background Technology

[0002] In the process of power grid intelligence, accurate assessment and early warning of the status of critical equipment are core to ensuring the reliable operation of the system. Traditional pole-mounted transformer condition assessment technology mainly relies on single or limited monitoring data such as electrical quantities, temperature, and insulation, judging the equipment status through threshold comparison or simple statistical models. With the rapid development of sensor technology and the popularization of the Internet of Things, modern condition assessment systems can now integrate multi-source, heterogeneous monitoring data and combine data-driven methods such as machine learning to extract more complex feature sets for generating equipment health indices or predicting failure probabilities. This development has significantly improved the perception capability and prediction accuracy of equipment operating status.

[0003] In the field of condition early warning for pole-mounted transformers, existing technologies often fail to pinpoint the root cause and severity of abnormal parameters such as current or temperature, leading to a lack of targeted maintenance decisions. For example, traditional methods typically only detect single phenomena like excessive total current and issue general alarms such as "overload," but cannot effectively distinguish whether the root cause of the overload is the natural increase in downstream user load, intermittent short circuits, or a surge in losses due to internal insulation degradation. This warning information is too generalized, often only concluding "abnormal" or "high risk," lacking in-depth analysis of the root cause and evolution path of the fault. Because the specific cause cannot be located, maintenance personnel struggle to develop precise countermeasures, often resorting to relatively conservative comprehensive strategies such as load limiting or planned power outages. This not only affects power supply reliability but also increases the blind spots and costs of maintenance. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for assessing the condition of pole-mounted transformers and providing early warning of faults by integrating multi-source data, which solves the problems of difficulty in locating the root cause of faults and the inability to effectively track the evolution path.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for condition assessment and fault early warning of pole-mounted transformers that integrates multi-source data, comprising:

[0008] Based on multi-source monitoring data of pole-mounted transformers, multi-physical field state characteristics of transformers are extracted; and a multi-domain collaborative evolution characteristic system is constructed to form a multi-source fusion characteristic set.

[0009] Based on the multi-source fusion feature set, a dual state determination is performed to extract the statistical deviation of abnormal operation fluctuations and calculate the physical deviation of multi-physics coupling imbalance. The statistical deviation and physical deviation are fused in a dual-domain collaborative manner to obtain the real-time health index.

[0010] Extract the degradation rate and multi-scale energy characteristics of the real-time health index, identify potential degradation clustering patterns by referring to historical slow-varying baselines, and generate potential risk trend factors;

[0011] The potential risk trend factors are combined with the multi-source fusion feature set for composite attribution matching, and graded risk judgment is performed based on the multi-physics anomaly mechanism chain to output the warning level and warning reason;

[0012] Based on the warning level and the cause of the warning, an online safety response plan is generated through calculation of the coordinated control strategy of the distribution network topology.

[0013] As a preferred embodiment of the method for assessing the condition of pole-mounted transformers and providing early warning of faults by fusing multi-source data as described in this invention, the method for forming a multi-source fusion feature set includes:

[0014] Based on multi-source monitoring data of pole-mounted transformers, we obtain the operation change patterns under multiple time scales and extract multi-physics field state features that characterize the cross-physics field state evolution.

[0015] By combining operational change patterns with multi-physics field state characteristics, a multi-domain collaborative evolution characteristic system is constructed;

[0016] A fusion operation is performed on the multi-physics field state characteristics and the multi-domain co-evolution characteristic system, and a cross-domain evolution consistency constraint is introduced to construct a co-evolution weight. Based on the co-evolution weight, a weighted screening and consistency dimensionality reduction are performed to form a multi-source fusion feature set that reflects the overall operation.

[0017] As a preferred embodiment of the method for assessing the condition and providing early warning of faults in pole-mounted transformers that integrates multi-source data as described in this invention, the method for performing dual-state determination includes:

[0018] Based on the multi-source fusion feature set, the temporal fluctuation characteristics of continuous variation intervals are extracted at multiple time scales, and the degree of abnormality of the operation fluctuation is quantified by the cross-scale consistency difference, and the statistical deviation is output.

[0019] Based on the multi-source fusion feature set, the coupling evolution deviation index of multiple physical field quantities at the same time scale is calculated, and the degree of coupling imbalance is quantified by the degree of deviation from the normal cooperative evolution relationship, and the physical deviation is output.

[0020] As a preferred embodiment of the method for assessing the condition of pole-mounted transformers and providing early warning of faults by integrating multi-source data as described in this invention, the method for obtaining the real-time health index includes:

[0021] The statistical deviation and the physical deviation are transformed to a unified dimension so that they are aligned in a common evaluation space, forming a two-domain deviation feature vector.

[0022] Based on the dual-domain deviation feature vector, the dual-domain contribution weights are dynamically allocated according to the uncertainty characteristics of statistical deviation and physical deviation. Through fusion operation, the fusion evaluation value and its confidence level under the synergistic effect of statistical domain and physical domain are generated.

[0023] The fusion assessment value and its confidence level are mapped to a health measurement range, and the confidence level information is integrated for scalar correction calculation to output a real-time health index.

[0024] As a preferred embodiment of the method for assessing the condition of pole-mounted transformers and providing early warning of faults by integrating multi-source data as described in this invention, the method for generating the hidden danger trend factor includes:

[0025] Based on the point-by-point changes of real-time health indices in continuous time series, the dynamic trend between adjacent health indices is continuously tracked; the difference in short time intervals is the instantaneous rate of change, and the cumulative change in long time intervals is the overall direction of deterioration, thus obtaining the deterioration rate characteristics.

[0026] Based on the fluctuation structure of the real-time health index at different time scales, the entire time series is segmented into short, medium and long windows, and key scale features at each scale are measured. The changes between scales are used to reflect the energy accumulation pattern of the health index at different time scales, forming a multi-scale energy feature.

[0027] Based on degradation rate characteristics and multi-scale energy characteristics, multi-dimensional pattern recognition and comparative analysis are performed with reference to historical slow-changing baselines. When there is a continuous deviation from the normal slow change range, abnormal coordinated changes between different time scales, or an accelerated clustering trend in the time series, potential degradation clustering patterns are identified and comprehensively quantified according to risk quantification elements to generate hidden danger trend factors.

[0028] As a preferred embodiment of the method for assessing the condition of pole-mounted transformers and providing early warning of faults by fusing multi-source data as described in this invention, the step of performing composite attribution matching includes:

[0029] Time window alignment is performed on the hidden danger trend factors and the multi-source fusion feature set. Features that evolve synchronously with the hidden danger trend factors in terms of time change and whose deviation exceeds the normal fluctuation range of the historical slow-change baseline are selected from the multi-source fusion feature set. A candidate attribution set is generated based on the time series consistency index system.

[0030] Based on the candidate attribution set, the attribution efficacy index of each candidate feature and potential risk trend factor is quantified; and scores and confidence levels are assigned to form an attribution scoring table.

[0031] Based on the attribution scoring table, the attribution results are integrated by weighting the scores and correcting for confidence, and the causal priority is ranked to obtain the attribution matching results.

[0032] As a preferred embodiment of the method for assessing the condition of a pole-mounted transformer and providing early warning of faults by integrating multi-source data as described in this invention, the step of executing a graded risk criterion based on a multi-physics anomaly mechanism chain includes:

[0033] The attribution features in the attribution matching results are embedded into the multiphysics anomaly mechanism chain, and associated with their corresponding physical action paths through causal relationship mapping, thereby verifying and establishing the starting node that triggers the potential anomaly mechanism.

[0034] Referring to the anomaly mechanism in the multiphysics anomaly mechanism chain, the fusion evaluation value of the cumulative damage degree of the physical action path and the multi-field synergistic correlation deviation index is used with the fault evolution critical threshold to perform a graded risk judgment to determine the risk level of the current operating status;

[0035] Based on the warning level and its corresponding anomaly mechanism, the key related factors leading to the risk level are extracted to form a structured warning cause, and the warning level and warning cause are output.

[0036] As a preferred embodiment of the method for assessing the condition of pole-mounted transformers and providing early warning of faults by integrating multi-source data as described in this invention, the method for generating an online safety response plan includes:

[0037] Based on the warning level and the cause of the warning, the affected regional units and related nodes are located; and by combining the physical action path corresponding to the cause of the warning, the key operational links that trigger the chain reaction are identified.

[0038] Based on the critical operation links and distribution network operation constraints, feasible coordinated control strategies are selected, and their risk reduction effects and costs are evaluated to generate strategy evaluation results. Then, strategy screening and time-series integration are performed to generate online safety handling solutions.

[0039] As a preferred embodiment of the pole-mounted transformer state assessment and fault early warning method integrating multi-source data described in this invention, the method for constructing the cross-domain evolution consistency constraint includes:

[0040] Based on the temporal variation relationship of the multi-physics field state characteristics and the multi-domain cooperative evolution characteristics system, we extract the evolutionary characteristic quantities that reflect the evolutionary direction and change rhythm of each domain, and establish constraints to regulate cross-domain cooperative relationships accordingly.

[0041] Based on the constraints, the degree of synergy between domains is quantified to form a synergistic weight system that can unify the cross-domain evolution law, thus obtaining cross-domain evolution consistency constraints.

[0042] As a preferred embodiment of the method for condition assessment and fault early warning of pole-mounted transformers that integrates multi-source data as described in this invention, wherein:

[0043] The method for constructing the multiphysics anomaly mechanism chain includes:

[0044] Based on the multi-physics operation characteristics of pole-mounted transformers, key state parameters that can characterize the mutual influence of multiple physical field quantities are extracted, and the causal action sequence of each physical quantity under abnormal conditions is analyzed to establish a logical link framework.

[0045] Based on the logical link framework, the action paths, coupling relationships, evolution bottlenecks, and corresponding critical constraints of each cross-physics field are integrated to construct a multi-physics anomaly mechanism chain.

[0046] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the method for assessing the condition of a pole-mounted transformer and providing early warning of faults by integrating multi-source data as described in the first aspect of the present invention.

[0047] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the pole-mounted transformer condition assessment and fault early warning method that integrates multi-source data as described in the first aspect of the present invention.

[0048] The beneficial effects of this invention are as follows: It deeply integrates data-driven and mechanism-driven approaches to generate a real-time health index that can accurately capture the early "sub-healthy" state of equipment. Through the complementary information of the two domains, it improves the robustness of state assessment and the sensitivity of early perception, overcoming the misjudgment caused by data noise or model bias due to single methods. By performing composite attribution matching between macroscopic hidden danger trends and microscopic multi-source characteristics, and embedding multi-physics anomaly mechanism chains for diagnosis, it achieves a leap from knowing what happened to knowing why. It can not only accurately locate the physical root cause and evolution stage of the fault, but also output a structured risk level and warning cause with clear physical meaning. It forms a complete closed loop from accurate perception to intelligent diagnosis to decision support, and transforms the operation and maintenance mode from passive post-event maintenance to proactive and precise intervention based on interpretable insights, thereby greatly improving the safety and economy of power grid operation. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart of the pole-mounted transformer condition assessment and fault early warning method that integrates multi-source data in this invention.

[0051] Figure 2 This is a flowchart for obtaining the real-time health index in this invention.

[0052] Figure 3 This is a flowchart for generating the hidden danger trend factor in this invention.

[0053] Figure 4 This is a flowchart illustrating the output of warning levels and warning reasons in this invention. Detailed Implementation

[0054] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0055] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0056] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0057] Reference Figure 1 , Figure 2 , Figure 3 and Figure 4 This is one embodiment of the present invention, which provides a method for assessing the condition and providing early warning of faults in pole-mounted transformers by fusing multi-source data, including the following steps:

[0058] Methods for forming multi-source fusion feature sets include:

[0059] Based on multi-source monitoring data of pole-mounted transformers, we can obtain the operation change patterns at multiple time scales and extract multi-physics state features that characterize the evolution of cross-physics state.

[0060] It should be noted that the multi-source monitoring data of pole-mounted transformers includes: electrical quantities, thermal quantities, mechanical quantities, environmental quantities, partial discharge quantities, and operation event records, which together constitute the basic information source reflecting the coupling characteristics of multiple physical fields of the transformer, including electrical, thermal, mechanical, and electromagnetic fields.

[0061] By analyzing transient changes (such as temperature rise slope fluctuations caused by sudden load changes) on short timescales (e.g., seconds to minutes), identifying intraday cyclical patterns (such as load cycles) on medium timescales (e.g., hours), and characterizing slow evolution trends (e.g., decreased heat dissipation efficiency) on long timescales (e.g., days to weeks), the system obtains cross-timescale operational change patterns and establishes multiphysics evolution trajectories accordingly. Subsequently, the system calculates the synchronicity, hysteresis response, and coupling strength between different physical quantities to extract multiphysics state characteristics.

[0062] By combining operational change patterns with multi-physics field state characteristics, a multi-domain collaborative evolution characteristic system is constructed.

[0063] Specifically, the operational change pattern is used as the main evolutionary line in the time dimension, and the evolutionary segments formed by the operational change pattern in short-term, medium-term, and long-term scales are registered with the multi-physics field state characteristics on a scale-by-scale basis. After registration, the electrical quantity change curves, thermal quantity change curves, mechanical quantity change curves, environmental quantity change curves, and partial discharge quantity change curves of each time segment are jointly analyzed to extract the synchronicity, response lag, and related change trends of each physical quantity in the same time segment. The synchronicity is used to characterize the cross-frequency domain energy coupling relationship, the response lag is used to characterize the time domain steady-state disturbance evolution path, and the related change trends are used to characterize the cross-source collaborative correlation law, and corresponding feature subsets are formed accordingly. Finally, the feature subsets are integrated in the order of short-term, medium-term, and long-term scales to form a multi-domain collaborative evolution feature system.

[0064] A fusion operation is performed on the multi-physics field state characteristics and the multi-domain co-evolution characteristic system, and a cross-domain evolution consistency constraint is introduced to construct a co-evolution weight. Based on the co-evolution weight, a weighted screening and consistency dimensionality reduction are performed to form a multi-source fusion feature set that reflects the overall operation.

[0065] It should be noted that the matching analysis is performed on the corresponding elements of the multi-physics field state features and the multi-domain co-evolution feature system at short, medium, and long time scales, and the consistency of each corresponding element in terms of change direction, change magnitude, and evolution trend is calculated. Then, cross-domain evolution consistency constraints are constructed based on the consistency degree, and the co-evolution weights are obtained through normalization. The co-evolution weights are then used to perform weighted screening on the multi-physics field state features and the multi-domain co-evolution feature system, so that feature components with low consistency (example < 0.3) are weakened. Based on the co-evolution weights, the fused features are subjected to consistency dimensionality reduction according to the time scale, and features with high consistency (example ≥ 0.8) at multiple time scales are retained to form a multi-source fusion feature set.

[0066] Methods for performing dual-state determination include:

[0067] Based on the multi-source fusion feature set, the temporal fluctuation characteristics of continuous change intervals are extracted at multiple time scales, and the degree of abnormality of the operation fluctuation is quantified by cross-scale consistency difference, and the statistical deviation is output.

[0068] It should be noted that, within short timescales, a sliding window is used to identify high-frequency continuously changing intervals and calculate transient fluctuation characteristics (such as change slope, local variance, and abrupt gradient); within medium timescales, a periodic window is used to extract stable repeating segments and calculate periodic fluctuation characteristics (such as peak-to-valley difference); within long timescales, a trend window is used to identify slowly changing segments and calculate long-term fluctuation characteristics (such as slow drift amplitude and trend curvature); the continuous changing interval characteristics of each scale are arranged in time sequence to form a set of time-series fluctuation characteristics at short-medium-long timescales;

[0069] Cross-scale consistency is jointly determined by the consistency of fluctuation direction, amplitude ratio, and trend continuity across short-term, medium-term, and long-term scales. The cross-scale consistency score expression for these three types of consistency indicators is as follows:

[0070] ;

[0071] in, It is the cross-scale consistency score. It is an indicator of the consistency of fluctuation direction. It is an amplitude ratio consistency index. It is an indicator of trend consistency. , and These are the weighting coefficients for the three types of consistency indicators, set based on their relative impact on overall cross-scale stability. The weights can be obtained by normalizing the variance contribution of the three indicators in normal operating data, giving higher weights to indicators with greater contributions (as exemplified). =0.3、 =0.4、 =0.3);

[0072] When the fluctuation characteristics at one scale do not match those at other scales, the cross-scale consistency score decreases, and the degree of deviation is as follows: ;in, It is the statistical deviation;

[0073] The volatility direction consistency index is used to quantify the consistency of volatility direction at different scales. Its value can be obtained by calculating the proportion of volatility sequences at each scale having the same sign (positive or negative) within the same time interval.

[0074] The amplitude ratio consistency index is used to quantify the relative stability of fluctuation amplitudes at different scales. Its value can be obtained by calculating the ratio of short-term to medium- and long-term fluctuation amplitudes (such as standard deviation or peak-to-peak value) and the degree of agreement between this ratio and the typical value of this ratio under a healthy baseline.

[0075] The trend consistency index is used to quantify the coordination of trend changes at different scales. Its value can be obtained by calculating the degree of consistency between the trend slope of medium and long time scales and the dominant direction of fluctuations in short time scales.

[0076] Based on the multi-source fusion feature set, the coupling evolution deviation index of multiple physical field quantities at the same time scale is calculated, and the degree of coupling imbalance is quantified by the degree of deviation from the normal cooperative evolution relationship, and the physical deviation is output.

[0077] It should be noted that, within the same time scale, the evolution sequences of multi-source monitoring data for pole-mounted transformers are registered, and the deviations in their direction of change consistency, amplitude ratio, and correlation trend are calculated; this can be achieved by constructing a coupling deviation index.

[0078] ;

[0079] in, It is the first Reference evolution characteristics of a physical quantity For the first Reference evolution characteristic values ​​of a physical quantity under healthy operating conditions. For the first Normalized weights for each physical quantity; Coupling evolution deviation index It is an index of physical quantities (electrical quantities, thermal quantities, mechanical quantities, environmental quantities, and partial discharge quantities);

[0080] The normal co-evolution relationship is derived from the historical multi-source monitoring data of the pole-mounted transformer during the healthy operation phase. By statistically analyzing the change direction, change magnitude ratio, and trend correlation of the multi-source monitoring data of the pole-mounted transformer at the same time scale, the reference evolution characteristics of the stable correspondence between various physical quantities under normal conditions are obtained.

[0081] During operation, when the difference between the coupling evolution deviation index of physical quantities and the reference evolution characteristics exceeds the typical deviation level corresponding to the normal co-evolution relationship (for example, 1.2 times the typical deviation level), it indicates that the coupling relationship between physical fields such as electro-thermal-mechanical-partial discharge is abnormal. The greater the difference, the higher the degree of coupling imbalance, and the output physical deviation.

[0082] Methods for obtaining real-time health indicators include:

[0083] The statistical deviation and physical deviation are transformed into a unified dimension to align them in a common evaluation space, forming a dual-domain deviation feature vector.

[0084] It should be noted that statistical deviation reflects the degree of stability disruption of cross-scale fluctuations, while physical deviation reflects the degree of shift in the coupling relationship of multi-physics fields. The two types of deviations are transformed in terms of dimensions by methods such as interval normalization, standard deviation scaling or robust scaling, so that the two-domain deviations are comparable in the same evaluation space. The relative change trends of the two are not distorted, forming a two-domain deviation feature vector that can simultaneously reflect the anomalies of temporal fluctuations and coupling shifts.

[0085] Based on the dual-domain deviation feature vector, the dual-domain contribution weights are dynamically allocated according to the uncertainty characteristics of statistical deviation and physical deviation. Through fusion operation, the fusion evaluation value and its confidence level under the synergistic effect of the statistical domain and physical domain are generated.

[0086] It should be noted that the uncertainty measures (such as deviation volatility, short-term stability, and confidence interval width) can be calculated based on the real-time characteristics of statistical and physical deviations during operation. These uncertainty measures are then normalized to construct a dual-domain weighting factor. The dual-domain contribution weights are calculated based on the principle that lower uncertainty equates to higher confidence, thus assigning higher weights to the deviation domain with lower uncertainty. After weighting, the dual-domain deviation feature vectors are weighted and fused using the dual-domain contribution weights to generate a fusion evaluation value and its confidence level under the synergistic effect of the statistical and physical domains.

[0087] The fusion assessment value and its confidence level are mapped to a health measurement range, and the confidence level information is integrated for scalar correction calculation to output a real-time health index.

[0088] It should be noted that the fusion evaluation value is normalized and mapped to a health measurement range (e.g., 0–1). A health index correction factor is then constructed based on the corresponding confidence level. The health index correction factor can be obtained by performing a linear or non-linear transformation on the confidence level. When the confidence level is high (e.g., ≥0.9), the correction factor is close to 1 (e.g., 0.95–1.00), and the health value remains basically unchanged. When the confidence level is medium (e.g., about 0.5), the correction factor is reduced to about 0.7–0.8 to moderately lower the health value. When the confidence level is low (e.g., ≤0.2), the correction factor can be compressed to ≤0.5, causing the health value to be significantly lowered, thereby avoiding misjudgment caused by unreliable biased data. The normalized health value is then scalarized using the correction factor to output the real-time health index.

[0089] Methods for generating potential hazard trend factors include:

[0090] Based on the point-by-point changes of real-time health indices in a continuous time series, the dynamic trend between adjacent health indices is continuously tracked; the difference in short time intervals is the instantaneous rate of change, and the cumulative change in long time intervals is the overall direction of deterioration, thus obtaining the deterioration rate characteristics.

[0091] It should be noted that a continuous sequence of real-time health indices can be constructed in chronological order, and the difference between adjacent sampling points can be calculated with a fixed step size to characterize the instantaneous rate of change on a short timescale. Simultaneously, the cumulative change at the beginning and end of a longer time window is calculated to describe the overall degradation direction on a long timescale. After completing the short-window differencing and long-window accumulation, the two are normalized to make them comparable within a unified metric space. The short-window rate captures sudden degradation, while the long-window accumulation identifies slow degradation. A degradation rate characteristic is formed by integrating these two types of trend information.

[0092] Based on the fluctuation structure of the real-time health index at different time scales, the entire time series is segmented into short, medium, and long windows, and key scale features at each scale are measured. The changes between scales are used to reflect the energy accumulation pattern of the health index at different time scales, forming a multi-scale energy feature.

[0093] It should be noted that the time series of the real-time health index is used, and multiple time windows such as short-term, medium-term, and long-term are set. Within each window, key feature quantities are calculated. For example, the local energy density (such as sum of squares or local variance) is calculated in the short-term window, the periodic energy accumulation (such as periodic envelope energy or kurtosis) is calculated in the medium-term window, and the trend energy (such as cumulative energy or trend slope integral) is calculated in the long-term window, so as to obtain the energy distribution of the health index at different time scales.

[0094] The energy characteristics at each scale are normalized, and by analyzing the correlation of energy changes between short, medium and long scales (such as the proportion of energy transfer, the rate of change of energy concentration with scale), a multi-scale energy characteristic reflecting the energy accumulation pattern of the health index at different time scales is constructed.

[0095] Based on degradation rate characteristics and multi-scale energy characteristics, multi-dimensional pattern recognition and comparative analysis are performed with reference to historical slow-changing baselines. When there is a continuous deviation from the normal slow change range, abnormal coordinated changes between different time scales, or an accelerated clustering trend in the time series, potential degradation clustering patterns are identified and comprehensively quantified according to risk quantification elements to generate hidden danger trend factors.

[0096] Specifically, based on the historical health index sequence of the pole-mounted transformer during the long-term healthy operation phase, the difference component, cumulative change and energy accumulation during the stable period are calculated in short-term, medium-term and long-term windows respectively; and steady-state clustering or robust statistical analysis is performed on the characteristics of each window to extract the stable distribution center, typical fluctuation boundary and slowly changing trend section as the historical slowly changing baseline.

[0097] When the short-term time difference component consistently falls outside the normal slow-change range of the historical slow-change baseline within a continuous time slice, the cumulative change over medium time shows a stable negative shift with the deviation exceeding the typical slow-change boundary of the baseline, and the concentration of long-term energy in the downward direction remains high, all three together indicate that the health index has continuously deviated from the normal slow-change range. If, on this basis, the deviation structures of short, medium, and long scales appear synchronously over time (e.g., a rapid decline in short time, a periodic decline in medium time, and a concentrated trend of energy concentration in long time are simultaneously manifested), it can be identified as an abnormal coordinated change. If the short-, medium-, and long-scales all show an accelerated concentration of energy in the downward direction (e.g., the energy concentration increases significantly with increasing scale), it is considered an accelerated concentration trend. By identifying consistent deviation structures across scales, they can be classified into deterioration concentration patterns such as sudden, gradual, or accelerated types, which can be used to describe the formation mechanism of potential hidden dangers.

[0098] The degradation rate characteristics and multi-scale energy characteristics are used to extract risk quantification elements. Specifically, short-term degradation intensity is obtained based on short-window differencing, collaborative deviation degree is obtained based on medium-window accumulation, and long-term aggregation trend degree is obtained based on long-window energy aggregation. The cross-scale consistency index is calculated from the energy relationship between different scales, and the historical deviation significance is obtained by comparing with the historical slow-changing baseline. Subsequently, each element is normalized and weighted according to risk contribution to form the hidden danger trend factor. By statistically analyzing the sensitivity, lead time, and stability of different elements before known degradation events, the relative weight of each element's influence on risk assessment can be obtained, and this weight is used as the risk contribution.

[0099] Performing composite attribution matching includes:

[0100] Time window alignment is performed on the hidden danger trend factors and the multi-source fusion feature set. Features that evolve synchronously with the hidden danger trend factors in terms of time change and whose deviation exceeds the normal fluctuation range of the historical slow-change baseline are selected from the multi-source fusion feature set. Based on the time series consistency index system, a candidate attribution set is generated.

[0101] It should be noted that by uniformly mapping the hidden danger trend factors and the multi-source fusion feature set to the same time coordinate axis, and slicing the data based on the same short, medium and long time windows, the two are accurately registered in time series.

[0102] From the registered multi-source fusion feature set, identify those features that are highly synchronized with the hazard trend factor in terms of change direction and consistent in response at key turning points. Among them, high synchronization and consistent response are quantitatively determined by a time-series consistency index system, which includes, for example, requiring that the Pearson correlation coefficient of the change slope of the candidate feature and the hazard trend factor in a short time scale be no less than 0.7, and requiring that the overlap rate between the two is no less than 80% within the time tolerance window of the occurrence of key local extreme points (peaks / troughs).

[0103] Furthermore, the specific values ​​of the selected features within the current time window are compared with the normal fluctuation range of the historical slow-varying baseline, and finally candidate features that both exhibit synchronous evolution and whose deviation exceeds the normal fluctuation range are selected.

[0104] The time-series consistency index system consists of quantitative indicators across multiple dimensions, including synchronicity of change direction, similarity of fluctuation patterns, and lag in response to key events. By calculating the comprehensive matching score between candidate features and potential risk trend factors within this time-series consistency index system, features scoring above the exemplary 0.75 are identified as strongly correlated features, generating a candidate attribution set.

[0105] Based on the candidate attribution set, the attribution efficacy index of each candidate feature and potential risk trend factor is quantified; and scores and confidence levels are assigned to form an attribution scoring table.

[0106] It should be noted that attribution efficacy metrics are quantitative standards used to measure the explanatory power of each feature in the candidate attribution set for the hazard trend factor. Examples include: temporal causal strength: quantifying the contribution of the candidate feature's historical values ​​to the current value of the hazard trend factor using methods such as transitive entropy; a larger contribution indicates higher causal strength. Evolutionary trend coupling: calculating the similarity between the candidate feature and the hazard trend factor's dynamic evolution trajectory within the same time window, for example, by measuring dynamic time regularization distance or trend correlation coefficient; higher similarity indicates stronger coupling. Anomaly deviation: calculating the degree of deviation of the candidate feature's current value from its normal fluctuation range within the historical slow-varying baseline, for example, quantified by standard deviation multiples or quantile differences; a larger deviation indicates a more significant anomaly.

[0107] Each attribution efficacy index is mapped to a score range of 0-1, and weights are assigned to each index according to their importance in determining the root cause. The comprehensive attribution score of each candidate feature is calculated by weighted fusion. Based on the quality of the candidate feature itself (such as signal-to-noise ratio and missing rate) and its stability over time, an attribution confidence score is assigned to the candidate feature. By establishing a one-to-one mapping relationship between each candidate feature and its comprehensive attribution score and attribution confidence score, and organizing them in the form of a list or matrix, a structured attribution scoring table is constructed.

[0108] Based on the attribution scoring table, the attribution results are integrated by weighting the scores and correcting for confidence, and the causal priority is ranked to obtain the attribution matching results.

[0109] It should be noted that the score weighting and confidence correction is to multiply the comprehensive attribution score of each candidate feature in the attribution scoring table with its attribution confidence to obtain an attribution power value that has been corrected for reliability.

[0110] All candidate features will be ranked in descending order of attribution power. The candidate feature with the highest attribution power will be considered the primary cause of the potential trend, and subsequent candidate features will be listed as secondary causes in descending order of their contribution. This ranked list is the final attribution matching result.

[0111] The risk classification criteria based on the multiphysics anomaly mechanism chain include:

[0112] The attribution features in the attribution matching results are embedded into the multiphysics anomaly mechanism chain, and associated with their corresponding physical action paths through causal relationship mapping, thereby verifying and establishing the starting node that triggers the potential anomaly mechanism.

[0113] It should be noted that the top-ranked attribution features in the attribution matching results are compared and matched one by one with the anomaly triggering nodes defined in the pre-constructed multiphysics anomaly mechanism chain.

[0114] When a certain attribution feature physically matches an anomaly triggering node, it is mapped to the subsequent physical action path corresponding to that node. Then, the state characteristics of subsequent stages on that physical action path are verified: if the feature already exists in the multi-source fusion feature set, and its value has deviated from the normal fluctuation range determined by the historical slow-varying baseline, while its direction of change conforms to the causal logic defined by the multiphysics anomaly mechanism chain, then the anomaly evidence on that path is determined to be valid, thereby ultimately establishing the starting node of the potential anomaly mechanism that triggered the current hidden danger trend.

[0115] Referring to the anomaly mechanism in the multiphysics anomaly mechanism chain, the fusion evaluation value of the cumulative damage degree of the physical action path and the multi-field synergistic correlation deviation index is used with the fault evolution critical threshold to perform a graded risk judgment to determine the risk level of the current operating state.

[0116] Specifically, the cumulative damage degree is calculated through the physical action path. This cumulative damage degree is quantified by integrating the abnormal amplitude and duration of key state parameters (such as insulation aging factor and mechanical fatigue degree) along the established physical action path to characterize the degree of accumulation of irreversible degradation. For example, for the physical action path of insulation aging caused by local overheating, its cumulative damage degree can be characterized by accumulating the product of the amplitude of the winding hot spot temperature exceeding the reference threshold and the time of exceeding the threshold. The reference threshold can be set to an example of 105°C (corresponding to the long-term allowable operating temperature of Class A insulation material). When the hot spot temperature exceeds 105°C, damage begins to accumulate. When the temperature reaches 130°C (the critical point where the thermal degradation of the insulation material is significantly accelerated), its damage accumulation rate per unit time will be given a higher weight.

[0117] The multi-field co-correlation deviation index is calculated by analyzing the actual coupling relationship between multiple physical fields (such as electrical, thermal, and vibration) related to the physical action path under the current operating state, comparing it with the normal co-correlation pattern shown in the historical baseline under the healthy state, and quantifying the degree of deviation between them. For example, the correlation between the current load current, winding temperature, and characteristic gas content in the oil can be calculated and compared with the typical correlation range obtained from the historical health data. The degree of deviation is the multi-field co-correlation deviation index.

[0118] Based on the definition of specific anomaly mechanisms in the multiphysics anomaly mechanism chain, appropriate weights are assigned to the cumulative damage degree of the physical action path and the multiphysics co-correlation deviation index, and a comprehensive and quantitative fusion evaluation value is generated through a weighted fusion algorithm. For example, for the slowly developing insulation aging problem, a weight of 0.6 can be assigned to the cumulative damage degree and a weight of 0.4 can be assigned to the co-correlation deviation index.

[0119] The fusion evaluation value is compared step-by-step with multiple fault evolution thresholds (e.g., threshold values ​​corresponding to attention, anomaly, and danger levels) preset for the anomaly mechanism in the multiphysics anomaly mechanism chain. Based on the threshold range in which the fusion evaluation value falls, the risk level of the current operating state is determined and output. For example, a fusion evaluation value below 0.3 can be set as the attention level, between 0.3 and 0.7 as the anomaly level, and above 0.7 as the danger level.

[0120] Based on the warning level and its corresponding anomaly mechanism, the key related factors leading to the risk level are extracted to form a structured warning cause, and the warning level and warning cause are output.

[0121] Specifically, from the established physical action paths, fusion assessment values ​​and their constituent components, the core elements that contribute the most to the risk level are identified. For example, for the risk of insulation aging caused by local overheating at the abnormal level, key related factors include: the characteristic of the A-phase winding hot spot temperature continuously exceeding the standard as the primary cause, the cumulative damage degree of 0.45 index indicating that the cumulative damage has reached the middle stage, and the load-temperature-characteristic gas correlation deviation index of 0.5 reflecting multi-field imbalance.

[0122] Based on the fault evolution logic described by the multiphysics anomaly mechanism chain, these key related factors are integrated into a structured early warning cause; the structure includes, for example: 1. Dominant anomaly mechanism: local overheating causes thermal aging of insulation; 2. Key precipitating characteristics: A-phase winding hot spot temperature exceeds the standard; 3. Risk quantification basis: cumulative damage degree 0.45, synergistic correlation deviation index 0.5; 4. Evolution stage location: overheating has caused initial deterioration of insulation material.

[0123] By combining the warning level with the above-mentioned structured warning reasons, the final warning information such as "[Warning Level] Abnormal: Due to the continuous exceedance of the hot spot temperature of phase A winding, insulation thermal aging has been triggered, and the cumulative damage has reached the mid-term level. Characteristic gas has begun to be abnormally produced" is output.

[0124] Methods for generating online security response plans include:

[0125] Based on the warning level and the cause of the warning, the affected regional units and related nodes are located; and by combining the physical action path corresponding to the cause of the warning, the key operational links that trigger the chain reaction are identified.

[0126] It should be noted that, based on the clearly defined abnormal mechanisms and key triggering characteristics in the warning reasons, specific faulty equipment and its electrical connection points are located in the distribution network topology. Furthermore, by combining the fault evolution path revealed by the multiphysics anomaly mechanism chain (such as thermal runaway potentially leading to insulation breakdown and short circuit), the chain reactions that may be triggered by the power flow shift and voltage over-limit after the faulty equipment and its electrical connection points fail, through the grid connection relationship, are deduced, thereby identifying the most vulnerable and highest-risk critical operating links.

[0127] Based on the critical operation links and distribution network operation constraints, feasible coordinated control strategies are selected, and their risk reduction effects and costs are evaluated to generate strategy evaluation results. Then, strategy screening and time-series integration are performed to generate online safety handling solutions.

[0128] It should be noted that the distribution network operation constraint system refers to a set of physical and operational rules that must be followed to ensure the safe and stable operation of the power grid. Examples include: capacity thermal stability limits for lines and transformers, upper and lower limits of node voltages, the coordination of protection device settings, and power supply reliability requirements. Any control strategy must be selected based on meeting these constraints.

[0129] Based on the identified critical operational links and early warning causes, executable coordinated control strategies are selected from the early warning strategy library (for example, for overload risk, strategies may include adjusting the state of tie switches to transfer load, switching capacitor banks to adjust voltage, or initiating demand-side response to reduce load). For each strategy, the percentage by which the strategy reduces the rate of increase in cumulative damage and the extent to which it restores multi-field coordinated deviation indicators to the normal range are calculated. At the same time, its operational costs are quantified, including the power supply reliability costs, equipment operation costs, and system operation costs caused by the strategy, thereby generating a comprehensive strategy evaluation result that includes both effects and costs for each strategy.

[0130] Based on the comprehensive strategy evaluation results, a multi-objective optimization algorithm is used to complete strategy selection and time-series integration: First, an optimization model is established with the objectives of maximizing risk reduction effect and minimizing operation cost, and the distribution network operation constraints are incorporated as constraints; by solving the optimization model with this objective, the Pareto optimal strategy set is selected from the candidate strategies; then, according to the operational logic dependency relationship and timeliness requirements between strategies, the execution order of strategies is time-seriesd to form a complete operation sequence and output an online safety handling plan.

[0131] It should also be noted that, based on the multiphysics anomaly mechanism chain analysis, the evolution path of various faults is determined, and effective intervention principles are identified; combined with the distribution network topology and operating procedures, the intervention principles are instantiated into a set of specific executable control operation instructions, forming an early warning strategy library.

[0132] Methods for constructing cross-domain evolutionary consistency constraints include:

[0133] Based on the temporal variation relationship between the multi-physics field state characteristics and the multi-domain cooperative evolution characteristics system, we extract evolutionary characteristic quantities that reflect the evolutionary direction and rhythm of each domain, and establish constraints to regulate cross-domain cooperative relationships.

[0134] It should be noted that, based on the inherent correlation between the multi-physics field state characteristics and the multi-domain co-evolutionary characteristics system in the time dimension, by extracting feature parameters that can characterize the consistency of evolutionary direction, the matching degree of change amplitude, and the coherence of trends in each domain (such as the frequency domain, time domain, and source domain), a set of mathematical constraints is established to regulate the synchronous and proportional or specific lag relationship between the evolution of different domains. These mathematical constraints together constitute a criterion system for judging whether cross-domain evolution is coordinated and consistent.

[0135] Based on the constraints, the degree of synergy between domains is quantified to form a synergistic weight system that can unify the cross-domain evolution law, thus obtaining cross-domain evolution consistency constraints.

[0136] It should be noted that, based on the established cross-domain collaborative relationship constraints, the actual degree of conformity between the multi-physics field state characteristics and the multi-domain collaborative evolution characteristic system in each constraint dimension is quantitatively calculated. These conformity indices are then normalized and comprehensively weighted to form a collaborative weighting system that dynamically reflects the importance of each domain's characteristics in the overall evaluation. This collaborative weighting system ensures that, in subsequent feature fusion, domains with consistent evolutionary patterns receive higher weights, resulting in a cross-domain evolution consistency constraint that unifies cross-domain evolutionary patterns.

[0137] Methods for constructing multiphysics anomaly mechanism chains include:

[0138] Based on the multi-physics operation characteristics of pole-mounted transformers, key state parameters that can characterize the mutual influence of multiple physical field quantities are extracted, and the causal action sequence of each physical quantity under abnormal conditions is analyzed to establish a logical link framework.

[0139] It should be noted that key state parameters that characterize the interaction between fields are extracted from multi-source monitoring data (such as the "loss-temperature rise" relationship parameter reflecting electro-thermal coupling, and the "thermal expansion stress" parameter reflecting thermo-mechanical coupling). By comprehensively applying the statistics of fault cases, the causal sequence and feedback mechanism of these key parameters under abnormal conditions are analyzed, thereby establishing a complete qualitative logical link framework that can clearly describe the anomaly from the triggering source (such as partial discharge and insulation moisture), through intermediate transmission and amplification (such as heat accumulation, material degradation and increased vibration), to the final functional failure (such as short circuit and breakdown).

[0140] Based on the logical link framework, the action paths, coupling relationships, evolution bottlenecks, and corresponding critical constraints of each cross-physics field are integrated to construct a multi-physics anomaly mechanism chain.

[0141] It should be noted that, based on the qualitative logical link framework, this approach integrates the specific action paths across various physical fields (e.g., "winding overheating → decrease in insulation paper polymerization degree → decrease in insulation mechanical strength → increased deformation under electrodynamic action → reduction in inter-turn insulation distance → initiation of partial discharge"), and statistically quantifies the coupling relationships between adjacent links in the path (e.g., establishing an Arrhenius model between temperature rise and insulation aging rate). Simultaneously, it identifies key bottlenecks and observable surrogate features in the evolution process (e.g., using measurable features such as "furfural content in oil" as a surrogate for the difficult-to-measure quantity "insulation paper polymerization degree"), and assigns corresponding quantifiable critical constraints (e.g., temperature threshold, vibration acceleration threshold, and characteristic gas concentration threshold) to each evolution link. Ultimately, this constructs a multi-physics anomaly mechanism chain integrating anomaly triggering nodes, multi-link action paths, quantified coupling relationships, and critical constraints.

[0142] This embodiment also provides a computer device applicable to the method for assessing the condition of a pole-mounted transformer and providing early warning of faults by integrating multi-source data, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the method for assessing the condition of a pole-mounted transformer and providing early warning of faults by integrating multi-source data as proposed in the above embodiment.

[0143] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0144] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the method for assessing the condition of a pole-mounted transformer and providing fault early warning by fusing multi-source data as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0145] In summary, this invention achieves this by: deeply integrating data-driven and mechanism-driven approaches to generate a real-time health index that accurately captures the early "sub-healthy" state of equipment; enhancing the robustness of condition assessment and the sensitivity of early perception through dual-domain information complementarity, overcoming misjudgments caused by data noise or model bias in single methods; and achieving a leap from knowing "what" to knowing "why" by performing composite attribution matching of macroscopic hidden danger trends and microscopic multi-source characteristics, and embedding a multi-physics anomaly mechanism chain for diagnosis. This not only accurately locates the physical root cause and evolution stage of the fault, but also outputs a structured risk level and warning cause with clear physical meaning. It forms a complete closed loop from accurate perception to intelligent diagnosis to decision support, transforming the operation and maintenance mode from passive post-event maintenance to proactive and precise intervention based on interpretable insights, thereby greatly improving the safety and economy of power grid operation.

[0146] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for fusion of multi-source data for pole-mounted transformer condition assessment and fault warning, characterized in that, The method comprises the following steps: S1, based on the multi-source monitoring data of the on-column transformer, the multi-physical field state characteristics of the transformer are extracted; and through the construction of a multi-domain collaborative evolution feature system, a multi-source fusion feature set is formed; S2, according to the multi-source fusion feature set, double state judgment is performed, the statistical deviation degree of operation fluctuation anomaly is extracted, and the physical deviation degree of multi-physical field coupling imbalance is calculated; the statistical deviation degree and the physical deviation degree are fused in a double-domain collaborative manner to obtain a real-time health index; S3, the degradation rate of the real-time health index and the multi-scale energy characteristics are extracted, the potential degradation aggregation mode is identified by referring to the historical slow-changing baseline, and a hidden danger trend factor is generated; S4, the hidden danger trend factor is matched with the multi-source fusion feature set, and a hierarchical risk criterion is executed according to the multi-physical field abnormal mechanism chain, and an early warning level and an early warning reason are output; S5, based on the early warning level and the early warning reason, an online safety disposal scheme is generated through the calculation of the distribution network topology structure collaborative control strategy.

2. The pole-mounted transformer condition assessment and failure warning method of fusing multi-source data according to claim 1, wherein, The method for forming the multi-source fusion feature set comprises: Based on the multi-source monitoring data of the on-column transformer, the operation change mode under the multi-time scale is obtained, and the multi-physical field state characteristics representing the state evolution across the physical field are extracted; combined with the operation change mode and the multi-physical field state characteristics, a multi-domain collaborative evolution feature system is constructed; the multi-physical field state characteristics and the multi-domain collaborative evolution feature system are subjected to fusion operation, and a collaborative weight is constructed by introducing a cross-domain evolution consistency constraint; according to the collaborative weight, weighted screening and consistency dimension reduction are performed to form a multi-source fusion feature set reflecting the overall operation.

3. The pole-mounted transformer condition assessment and failure warning method that fuses multiple sources of data of claim 2, wherein, The method for performing double state judgment comprises: Based on the multi-source fusion feature set, the time sequence fluctuation characteristics of the continuous change interval are extracted on the multi-time scale respectively, and the abnormal degree of operation fluctuation is quantified by the cross-scale consistency difference, and the statistical deviation degree is output; based on the multi-source fusion feature set, the coupling evolution deviation index of the multi-physical field quantity under the same time scale is calculated, and the coupling imbalance degree is quantified by the degree of deviation from the normal collaborative evolution relationship, and the physical deviation degree is output.

4. The pole-mounted transformer condition assessment and failure warning method that fuses multiple sources of data of claim 3, wherein, The method for obtaining the real-time health index comprises: the statistical deviation degree and the physical deviation degree are subjected to unified dimension conversion, so that they are aligned in the common evaluation space to form a double-domain deviation feature vector; based on the double-domain deviation feature vector, the double-domain contribution weight is dynamically allocated according to the uncertainty characteristics of the statistical deviation degree and the physical deviation degree, and through fusion operation, a fusion evaluation value and its confidence under the collaborative action of the statistical domain and the physical domain are generated; the fusion evaluation value and its confidence are mapped to the health measurement interval, the confidence information is integrated for scalarization correction calculation, and the real-time health index is output.

5. The pole-mounted transformer condition assessment and failure warning method that fuses multiple sources of data of claim 4, wherein, The method for generating the hidden danger trend factor comprises: based on the point-by-point change of the real-time health index in the continuous time sequence, the dynamic trend between adjacent health indexes is continuously tracked; the difference value of the short time interval is taken as the instantaneous change rate, and the cumulative change amount of the long time interval is taken as the overall degradation direction, and the degradation rate characteristics are obtained; Based on the fluctuation structure of the real-time health index at different time scales, the entire time series is segmented into short, medium and long multi-class windows, and the key scale feature quantity at each scale is measured. The energy aggregation law of the health index at different time scales is reflected by the change correlation between scales, and the multi-scale energy features are formed. Based on the deterioration rate characteristics and multi-scale energy features, multi-dimensional pattern recognition and comparison analysis are performed with reference to the historical slow baseline. When there is a sustained deviation from the normal slow change range, an abnormal coordinated change between different time scales, or an accelerating aggregation trend on the time series, a potential deterioration aggregation pattern is determined, and the risk quantification elements are integrated to generate a hidden danger trend factor.

6. The pole-mounted transformer condition assessment and failure warning method that fuses multiple sources of data of claim 5, wherein, The performing composite attribution matching includes: The time window alignment is performed on the hidden danger trend factor and the multi-source fusion feature set. The features that are synchronized with the hidden danger trend factor in time variation and deviate from the normal fluctuation interval of the historical slow baseline are selected from the multi-source fusion feature set. According to the time sequence consistency index system, a candidate attribution set is generated. Based on the candidate attribution set, the attribution efficiency index of each candidate feature and the hidden danger trend factor is quantified. The score is divided and the confidence is assigned to form an attribution score table. Based on the attribution score table, the attribution result is integrated by score weighting and confidence correction, and the causal priority is sorted to obtain the attribution matching result.

7. The pole-mounted transformer condition assessment and failure warning method that fuses multiple sources of data of claim 6, wherein, The hierarchical risk criterion based on the multi-physical field abnormal mechanism chain includes: The attribution features in the attribution matching result are embedded in the multi-physical field abnormal mechanism chain, and are associated with the corresponding physical action path through causal relationship mapping, to verify and determine the starting node of triggering the potential abnormal mechanism. Referring to the abnormal mechanism in the multi-physical field abnormal mechanism chain, the fusion evaluation value of the physical action path cumulative damage degree and the multi-field synergistic correlation deviation index is compared with the fault evolution critical threshold to determine the risk level of the current operating state. According to the warning level and the corresponding abnormal mechanism, the key associated factors leading to the risk level are extracted to form a structured warning cause, and the warning level and the warning cause are output.

8. The pole-mounted transformer condition assessment and failure warning method that fuses multiple sources of data of claim 7, wherein, The method for generating an online security disposal scheme includes: Based on the warning level and the warning cause, the affected area unit and the associated node are located, and the key operation link causing the chain effect is identified in combination with the physical action path corresponding to the warning cause. Based on the key operation link and the power distribution network operation constraint system, an executable synergistic control strategy is selected, and its risk reduction effect and cost are evaluated to generate a strategy evaluation result. The strategy is screened and time sequenced, and an online security disposal scheme is generated.

9. The pole-mounted transformer condition assessment and failure warning method that fuses multiple sources of data of claim 2, wherein, The construction method of the cross-domain evolution consistency constraint includes: Based on the time variation relationship between the multi-physical field state characteristics and the multi-domain synergistic evolution feature system, the evolution feature quantity reflecting the evolution direction and change rhythm of each domain is extracted, and the constraint condition for regulating the cross-domain synergistic relationship is established accordingly. According to the constraint condition, the synergistic degree between each domain is quantified to form a synergistic weight system that can unify the cross-domain evolution law, and a cross-domain evolution consistency constraint is obtained.

10. The pole-mounted transformer condition assessment and failure warning method that fuses multiple sources of data of claim 7, wherein, The construction method of the multi-physical field abnormal mechanism chain includes: Based on the multi-physical field operating characteristics of on-column transformer, the key state parameters that can represent the mutual influence relationship of multi-physical field quantities are extracted, and the causal action sequence of each physical quantity in abnormal state is analyzed to establish a logical link framework. According to the logical link framework, the action path, coupling relationship, evolution bottleneck and corresponding critical constraint conditions of each cross-physical field are integrated to construct a multi-physical field abnormal mechanism chain.

Images