Station degradation trajectory prediction method fusing physical prior and spatio-temporal knowledge embedding

Abstract

The application belongs to the technical field of power distribution station operation state monitoring, and discloses a station building degradation trajectory prediction method fusing physical prior and space-time knowledge embedding; comprising: collecting multi-source heterogeneous data; constructing a physical constraint equation of key physical quantities in station building equipment; constructing a space-time knowledge graph, adopting a graph attention network as a graph embedding model, embedding the attention weight of a neighbor node of a node, and obtaining a node embedding vector set; performing feature extraction on the multi-source heterogeneous data, and obtaining mixed features; obtaining a degradation trajectory prediction result based on physical information neural network analysis, performing health degree evaluation on the degradation trajectory prediction result, obtaining a health degree score, and performing corresponding early warning according to the health degree score; the application realizes high-precision prediction of the degradation trajectory of station building equipment in a fault-free data scene, and guarantees the physical rationality and operation and maintenance guidance of the prediction result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power distribution station operation state monitoring, and in particular to a station building degradation trajectory prediction method fusing physical priori and spatiotemporal knowledge embedding. BACKGROUND

[0002] The station building equipment is the core infrastructure for the safe and stable operation of the power system, and its degradation trajectory prediction is crucial for early identification of equipment degradation trends and the development of accurate operation and maintenance strategies. However, existing degradation prediction techniques still have significant limitations and cannot meet actual needs. On the one hand, purely data-driven prediction methods can capture statistical patterns in data, but they are highly dependent on a large amount of historical fault data and complete degradation samples. However, actual station building equipment generally has long service life and low failure characteristics, and fault data, especially early degradation data, are scarce, leading to insufficient model training and poor generalization ability. In the absence of fault data, it is difficult to achieve accurate extrapolation prediction, and it is easy to deviate from the true degradation trend due to overfitting of normal data. On the other hand, purely physically mechanism-driven prediction methods are based on equipment degradation mechanisms to construct equation models, without relying on fault data. However, due to the complex structure of station building equipment and the characteristics of multi-physical field coupling, it is difficult to establish a comprehensive and accurate mechanism equation, and the dynamic correlation information of multi-source heterogeneous monitoring data is ignored, resulting in limited prediction accuracy and inability to effectively depict the subtle degradation process of equipment. In addition, existing technologies generally do not address the effective fusion of multi-source heterogeneous data, resulting in data silos, and do not consider the spatiotemporal correlation between station building equipment, making feature representation incomplete, further restricting the accurate perception of the overall degradation state of the equipment by the prediction model, and ultimately leading to the inability of existing technologies to accurately identify early degradation and reliably predict degradation trajectories of equipment, failing to meet the needs of station building equipment life cycle health management and power system safe operation.

[0003] A power distribution station building health evaluation system based on multi-source information with the authorization publication number CN118801579B discloses: a sensor unit for collecting operation data of the power distribution station building, which is then transmitted to a data processing unit and a data analysis unit; the data processing unit is used for preprocessing the operation data collected by the sensor unit; and the preprocessing result is transmitted to the data analysis unit; the data analysis unit is used for scoring analysis of the operation data collected and processed by the sensor unit and the data processing unit, and the scoring analysis result is transmitted to a data display unit; the data display unit is used to display the results obtained by the data analysis unit to relevant personnel in an intuitive manner.

[0004] The above method cannot achieve accurate extrapolation prediction in the absence of historical fault data.

[0005] To solve the above problems, the present application proposes a station building degradation trajectory prediction method fusing physical priori and spatiotemporal knowledge embedding. SUMMARY

[0006] The present application aims to propose a station building degradation trajectory prediction method that fuses physical priori and spatiotemporal knowledge embedding to solve the problems raised in the background art:

[0007] The prior art cannot achieve accurate extrapolation prediction in the absence of historical failure data.

[0008] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0009] The station building degradation trajectory prediction method that fuses physical priori and spatiotemporal knowledge embedding comprises the following steps:

[0010] Collecting station building equipment operation data and preprocessing to obtain multi-source heterogeneous data;

[0011] Constructing the degradation process physical constraint equation of the key physical quantity in the station building equipment, and the corresponding boundary constraint equation, to obtain the physical constraint equation;

[0012] Constructing a spatiotemporal knowledge graph containing a node set, a relationship set and an attribute set, using a graph attention network as a graph embedding model to embed the attention weight of the neighbor nodes of the nodes, to obtain a node embedding vector set;

[0013] Extracting features from the multi-source heterogeneous data to obtain data-driven features, and analyzing the physical constraint equation to obtain physical derivative features, which are 3-dimensional feature vectors obtained by splicing the wear rate, stiffness decay rate and aging acceleration factor;

[0014] Z-score standardization processing the node embedding vector set, the data-driven features and the physical derivative features to obtain standardized nodes, splicing the standardized node embedding vector set, the data-driven features and the physical derivative features to obtain mixed features;

[0015] Taking the mixed features as the input of the physical information neural network to obtain the degradation trajectory prediction result, evaluating the health degree of the degradation trajectory prediction result to obtain a health degree score, and performing corresponding early warning according to the health degree score.

[0016] Preferably, the station building equipment operation data includes electrical data, mechanical data, environmental data, visual images and inspection data.

[0017] Preferably, the key physical quantity includes switch cabinet contacts, circuit breaker springs and transformer windings.

[0018] Preferably, the method for obtaining the physical constraint equation comprises:

[0019] The cumulative number of opening and closing cycles and contact resistance are obtained. Combined with the electrical wear coefficient, mechanical wear coefficient and the current at the corresponding moment, an equation for the amount of contact wear is constructed.

[0020] The initial stiffness and stress cycle number of the spring are obtained, and the spring stiffness decay equation is constructed by combining the attenuation coefficient and the fatigue life at the corresponding stress level.

[0021] Obtain the dielectric loss angle and winding temperature, and combine the temperature acceleration coefficient and reference temperature to construct the winding insulation aging index equation.

[0022] Construct boundary constraint equations, including: wear amount is non-negative; spring stiffness is not lower than the lower stiffness threshold; winding insulation aging index is not higher than the upper aging index threshold.

[0023] The physical constraint equation is obtained by combining the contact wear equation, the spring stiffness decay equation, the winding insulation aging index equation, and the boundary constraint equation.

[0024] Preferably, the method for constructing a spatiotemporal knowledge graph that includes a set of nodes, a set of relationships, and a set of attributes includes:

[0025] Obtain the station's coverage area, including spatial scope, equipment scope, and time scope;

[0026] Entities are categorized into equipment unit entities, component entities, sensor entities, spatial region entities, and event / physical quantity entities;

[0027] Define the relationships between entities, and analyze the domain, range, and type of the relationships;

[0028] Define the attributes of an entity, including static attributes and dynamic attributes;

[0029] A pre-defined hierarchical structure of station buildings, floors, bays, and installation sites is adopted to associate each spatial area entity with the superior area;

[0030] Bind all entities within the station building to absolute coordinates based on the station building's coordinate system and relative coordinates based on the coordinate system of their respective spatial regions;

[0031] Acquire station equipment testing data and automatically extract and convert it based on preset conversion tools; obtain data entities and map the data entities into a graph format according to the correspondence between entities and attributes to obtain a spatiotemporal knowledge graph.

[0032] Preferably, the method for obtaining the node embedding vector set includes:

[0033] Randomly assign an initial vector to each node;

[0034] For nodes neighboring nodes The attention weight is calculated by combining the attention parameter, the linear transformation matrix and the node embedding vector;

[0035] The embedding vector of the node is updated based on the attention weight;

[0036] The embedding vectors of the nodes are spliced to obtain a set of node embedding vectors.

[0037] Preferably, the data-driven features include data features and image features;

[0038] The method for obtaining the data features includes:

[0039] The time series data including electrical data, mechanical data, environmental data and inspection data are counted to obtain the mean and variance, the absolute mean is obtained by counting the time series data, and the peak factor is calculated based on the absolute mean;

[0040] The mechanical data is subjected to Fourier transform to obtain a Fourier transform signal, the Fourier transform signal is extracted to obtain the main frequency, spectral energy and spectral entropy, and the main frequency, spectral energy and spectral entropy are spliced to obtain the frequency domain features;

[0041] The mechanical data is subjected to wavelet transform to obtain the wavelet entropy, and the wavelet entropy is taken as the time-frequency domain feature;

[0042] The time domain features, frequency domain features and time-frequency domain features are spliced to obtain the data features.

[0043] Preferably, the image features include visual features and infrared features;

[0044] The method for obtaining the image features includes:

[0045] The equipment appearance image is taken as the input of the convolution network to obtain the convolution features;

[0046] The equipment appearance image is subjected to grayscale processing and Gaussian denoising processing to obtain a pretreated image;

[0047] The pretreated image is taken as the input of the positioning model to obtain the damage area bounding box coordinates; the positioning model is a YOLOv8n model;

[0048] The pretreated image is taken as the input of the U-Net model to obtain the wear area probability, and the image region with the wear area probability higher than the probability threshold is segmented as the ROI region;

[0049] The pixel number of the damage area and the pixel number of the ROI region contained in the damage area bounding box coordinates are counted, and the ratio of the pixel number of the damage area to the pixel number of the ROI region is calculated to obtain the wear area pixel area proportion;

[0050] ​The convolutional features and the area proportion of the wear region pixels are spliced to obtain visual features;

[0051] The infrared image is converted based on a radiation calibration model of an infrared thermal imager to obtain pixel point temperatures;

[0052] An area in which continuous pixel point temperatures are higher than a temperature threshold and the number of pixel points higher than the temperature threshold is greater than G is obtained, the number of pixel points with a temperature higher than the temperature threshold is counted, and the total number of pixel points of the infrared image is counted, so that a ratio of the number of pixel points to the total number of pixel points is calculated to obtain a hot area proportion;

[0053] The highest temperature value among the pixel points with a temperature higher than the temperature threshold is counted;

[0054] The mean temperature of the pixel points with a temperature lower than the temperature threshold is calculated, and the difference between the highest temperature value and the mean temperature is calculated to obtain a temperature gradient;

[0055] The hot area proportion and the temperature gradient are spliced to obtain infrared features.

[0056] Preferably, the method for obtaining the screening result comprises:

[0057] The mutual information entropy of the data features, the image features and the physical derived features and the degradation index is calculated respectively, and the features greater than a degradation threshold are selected as the screening result.

[0058] Preferably, the physical information neural network is obtained by building a 2-layer LSTM, a Transformer encoder containing 4 attention heads and a space-time attention layer;

[0059] A data fitting loss is established based on the error between the real degradation index and the predicted value, a physical constraint loss is established based on equation constraints and boundary constraints, and a total loss function is calculated based on the data fitting loss and the physical constraint loss; the physical information neural network is optimized based on the Adam optimizer, and the physical information neural network at the minimum total loss function is obtained as the training output physical information neural network;

[0060] The method for obtaining the health score comprises:

[0061] The degradation trajectory prediction result is quantitatively processed based on a segmented function constructed according to a normal threshold of the degradation index and a failure threshold of the degradation index to obtain the health score;

[0062] When the health score is lower than a score threshold, a health warning is triggered, and when the health score change rate is higher than a change rate threshold, a trend warning is triggered.

[0063] Compared with the prior art, the present application provides a station building degradation trajectory prediction method that fuses physical priori and space-time knowledge embedding, and has the following beneficial effects:

[0064] The present application fills the information gap when constructing a full-dimensional device state representation to supplement historical failure data by collecting multi-source heterogeneous data of station house equipment and preprocessing; the physical constraint equation and boundary constraint equation of switch cabinet contact wear, circuit breaker spring stiffness attenuation, transformer winding insulation aging index, etc. are constructed to inject the physical mechanism anchor point of equipment degradation prediction, avoiding prediction that violates physical laws without failure data; the spatiotemporal knowledge graph containing entities such as equipment units, components, sensors, and spatiotemporal, physical, and functional relationships is constructed, and the GAT model is used to generate node embedding vector sets to convert device-related information into structured features to enhance the model's generalization ability in small sample scenarios; by extracting data-driven features and physical-derived features, and selecting features strongly associated with degradation indicators based on mutual information entropy, the feature quality is improved by removing false information; by building a physical information neural network, the model is optimized by combining the total loss function of data fitting loss and physical constraint loss, which can accurately capture the subtle degradation trend of equipment without relying on historical failure data, ensuring the rationality of extrapolation prediction; finally, the health score is calculated by piecewise function and the health warning and trend warning are triggered, effectively solving the core problem that the existing technology cannot achieve accurate extrapolation prediction when lacking historical failure data, realizing high-precision prediction of station house equipment degradation trajectory in the absence of failure data, while ensuring the physical rationality and operation guidance of the prediction results, significantly improving the reliability and generalization ability of equipment degradation prediction, and providing strong technical support for station house equipment life cycle health management and operation decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0065] Fig. 1 The method flowchart mentioned in embodiment 1 of the present application;

[0066] Fig. 2 The data flow direction diagram mentioned in embodiment 1 of the present application. DETAILED DESCRIPTION

[0067] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all.

[0068] Embodiment 1:

[0069] Please refer to Figs. 1-2 The station house degradation trajectory prediction method of the present application fuses physical priori and spatiotemporal knowledge embedding, including the following steps:

[0070] The system collects and preprocesses station equipment operation data to obtain multi-source heterogeneous data. This data includes electrical, mechanical, environmental, visual images, and inspection data. Electrical data includes corresponding current, voltage, contact resistance, partial discharge, and the number of electrical data samples. Mechanical data includes vibration acceleration, opening and closing times, and operating torque. Environmental data includes temperature, humidity, and dust concentration. Visual images include equipment appearance images and infrared images. Inspection data includes bolt tightness status, etc.

[0071] By collecting electrical, mechanical, environmental, visual images, and inspection data from station equipment and preprocessing this multi-source heterogeneous data, a state characterization system covering all dimensions of equipment electrical, mechanical, environmental, appearance, and key components can be constructed. This system overcomes the limitations of existing technologies that rely on historical fault data to mine degradation patterns. Even in scenarios lacking fault samples, it can provide rich prior information on equipment operation and degradation through the subtle degradation signals, multi-dimensional state correlation patterns, and external influencing factors provided by complementary multi-source data. This information is then used to inject rich prior information on equipment operation and degradation into subsequent prediction models that integrate physical priors and spatiotemporal knowledge. This supports the model in accurately capturing the inherent logic of equipment degradation trends, thereby achieving accurate extrapolation prediction of the future degradation trajectory of the equipment. This effectively solves the technical problem of insufficient extrapolation prediction accuracy caused by the scarcity of fault data in existing technologies.

[0072] The degradation process physical constraint equations of key physical quantities in the station equipment are constructed, as well as the corresponding boundary constraint equations. The degradation process physical constraint equations and the boundary constraint equations are integrated to construct the physical constraint equations. The physical constraint equations mentioned in other steps refer to the comprehensive physical constraint equations constructed in this step. Key physical quantities include switch cabinet contacts, circuit breaker springs and transformer windings.

[0073] Methods for obtaining physical constraint equations include:

[0074] By obtaining the cumulative number of opening and closing cycles and the contact resistance, and combining the electrical wear coefficient, mechanical wear coefficient, and current at the corresponding time, an equation for the contact wear amount is constructed; for example, the contact wear amount at time t. ,in, The coefficient of electrical wear; The mechanical wear coefficient; This is the initial time. for Current at any moment; for Constantly in contact with the resistor; The cumulative number of opening and closing operations at time t;

[0075] By obtaining the initial stiffness and stress cycle count of the spring, and combining the attenuation coefficient and fatigue life at the corresponding stress level, a spring stiffness attenuation equation is constructed; for example, the spring stiffness attenuation equation at time t. wherein, is the initial stiffness; is the fatigue decay coefficient; is the stress cycle number of the nth operation; N fl is the fatigue life corresponding to the stress level;

[0076] Obtain the dielectric loss angle and winding temperature, combine the temperature acceleration coefficient and the reference temperature, and build a winding insulation aging index equation, such as the winding insulation aging index:

[0077] ,

[0078] wherein, is the dielectric loss angle at time t; is the temperature acceleration coefficient; is the winding temperature at time t; T ref is the reference temperature, generally 25℃;

[0079] Build a boundary constraint equation, including non-negative wear amount; the spring stiffness is not lower than the lower limit threshold of the stiffness; the winding insulation aging index is not higher than the upper limit threshold of the aging index;

[0080] Integrate the contact wear equation, the spring stiffness decay equation, the winding insulation aging index equation and the boundary constraint equation as an equation set to obtain a dual constraint physical constraint equation of the mechanism and the boundary of the deterioration of the key components of the equipment. The essence of splicing the physical constraint equation is the coordinated constraint of the deterioration mechanisms of multiple components: the contact, the spring and the winding are the core failure components of the station house equipment, and their degradation equations respectively depict the deterioration laws of the electrical, mechanical and insulation systems, and the boundary equation ensures that the deterioration of each component does not exceed the safe operation limit, and the equation set collectively constitutes the physical anchor point of the prediction model, which can avoid the physical misalignment of the model prediction when there is no fault data.

[0081] Build a spatio-temporal knowledge graph including a node set, a relationship set and an attribute set, use a graph attention network, namely GAT, as a graph embedding model to embed the attention weight of neighbor nodes of a node, and obtain a node embedding vector set;

[0082] The method for building a spatio-temporal knowledge graph including a node set, a relationship set and an attribute set comprises:

[0083] Obtain the station house coverage range, including the spatial range, the equipment range and the time range;

[0084] Divide the entity categories, including the device unit entity, the component entity, the sensor entity, the space region entity and the event / physical quantity entity; ​​​

[0085] Defining the relationship between entities, and analyzing the definition domain, value domain and relationship type of the relationship;

[0086] Defining the attributes of the entity, including static attributes and dynamic attributes;

[0087] Using the preset hierarchical structure of station building, floor, interval and installation site to associate each space area entity with the upper-level area;

[0088] Binding all entities in the station building with absolute coordinates based on the station building coordinate system and relative coordinates based on the coordinate system of the space area to which they belong;

[0089] Obtaining station equipment detection data and performing automatic extraction and conversion based on a preset conversion tool, such as an ETL tool; obtaining data entities, mapping the data entities into a graph format according to the corresponding relationship between the entities and the attributes, and obtaining a spatio-temporal knowledge graph.

[0090] The method for obtaining the set of node embedding vectors includes:

[0091] Randomly assigning an initial vector to each node;

[0092] For the neighbor nodes of the node , combining the attention parameter, the linear transformation matrix and the node embedding vector to calculate the attention weight; for example, the attention weight of the neighbor nodes of the node :

[0093]

[0094] wherein, is the attention parameter; is the linear transformation matrix; is the neighbor set of the node ; represents vector splicing; is the negative slope, with a value of 0.2; h v is the embedding vector of the node ; h u is the embedding vector of the node ; is the embedding vector of the node ;

[0095] Updating the embedding vector of the node based on the attention weight, for example, the updated embedding vector of the node is:

[0096] ​

[0097] wherein, is an activation function;

[0098] The updated embedding vectors of each node are spliced to obtain a set of node embedding vectors.

[0099] By constructing the degradation process physical constraint equation and boundary constraint equation of the key physical quantities of the station building equipment, combining the construction of the spatiotemporal knowledge graph and the acquisition of the graph embedding vector set, the core support is provided for solving the problem that the existing technology cannot realize accurate extrapolation prediction when lacking historical failure data: on the one hand, for the switch cabinet contact, based on the obtained cumulative on-off times, contact resistance and corresponding current, combined with the electrical wear coefficient and the mechanical wear coefficient, the contact wear equation is constructed, which can quantify the degradation process caused by electrical erosion and mechanical friction of the contact; for the circuit breaker spring, the initial stiffness, stress cycle number, combined with the attenuation coefficient and the corresponding stress level of the fatigue life, the spring stiffness attenuation equation is built, which can represent the stiffness change rule of the spring fatigue aging; for the transformer winding, according to the dielectric loss angle, winding temperature, combined with the temperature acceleration coefficient and the reference temperature, the winding insulation aging index equation is constructed, which can reflect the aging trend of the winding insulation with the change of temperature and dielectric loss, and at the same time, the boundary constraint equation of the non-negative wear amount, the spring stiffness not lower than the lower threshold, and the winding temperature not higher than the upper threshold is supplemented. These equations constructed based on physical mechanism can provide the underlying physical law constraints of equipment degradation for the prediction model without relying on historical failure data, avoiding the deviation of pure data-driven models from the true degradation trend due to overfitting of normal data when the failure sample is scarce; on the other hand, by defining the spatial, physical, functional, temporal relationships and static, dynamic attributes between entities such as device units, components, sensors, space regions, events / physical quantities, and using the hierarchical structure of the station, floor, interval and installation site to associate the space region entities and bind the absolute and relative coordinates of the entities, combined with the ETL tool to extract and transform the detection data to generate the spatiotemporal knowledge graph, and then taking the graph attention network as the embedding model, assigning an initial vector to each node, combining the attention parameters, linear transformation matrix and node embedding vector to calculate the attention weight of the neighbor nodes of each node, and updating the node embedding vector based on the weight to form a vector set, this process can convert multi-source heterogeneous data into structured features rich in device spatiotemporal association and semantic relationship, even when the failure data is lacking, the missing information gap can be supplemented through the association information between devices, and finally the mechanism guidance provided by the physical constraint equation and the association features provided by the spatiotemporal knowledge graph embedding vector set work together, making the prediction model not only follow the physical law to ensure the rationality of the extrapolation direction, but also capture subtle degradation signals through multi-dimensional association information, effectively breaking through the limitations of relying on historical failure data in the existing technology, and realizing accurate extrapolation prediction of the equipment degradation trajectory.

[0100] The multi-source heterogeneous data is subjected to feature extraction to obtain data-driven features, the physical constraint equation is analyzed to obtain physical-derived features, and the node embedding vector set, the data-driven features and the physical-derived features are subjected to standardization processing and splicing to obtain mixed features; for example, assuming that the node embedding vector set contains 10 nodes, and each node embedding dimension is 16, then after flattening, it is a 1x160 vector; the data-driven features contain 3-dimensional time domain, 3-dimensional frequency domain and 10-dimensional image features, a total of 1x16 vectors; the physical-derived features are 1x3 vectors; after standardization, splicing is 1x(160+16+3)=1x179-dimensional mixed features, ensuring that the feature dimensions of the input model are uniform.

[0101] The data-driven features include data features and image features;

[0102] The method for obtaining the data features includes:

[0103] The time series data containing electrical data, mechanical data, environmental data and inspection data are subjected to statistics to obtain mean and variance, the absolute mean is obtained by statistical time series data, and the peak factor is calculated according to the absolute mean; for example, the absolute mean , the peak factor The mean, variance and peak factor are spliced to obtain the time domain features; is the number of time series data; is the time series data;

[0104] The mechanical data is subjected to Fourier transform to obtain a Fourier transform signal, and the main frequency, the spectral energy and the spectral entropy are extracted from the Fourier transform signal; the main frequency, the spectral energy and the spectral entropy are spliced to obtain the frequency domain features; for example, the Fourier transform signal , the main frequency , the spectral energy , and the spectral entropy , wherein, j is an imaginary unit; f is a frequency, used to describe the change characteristics of the signal in the frequency domain; is a time, used to describe the change characteristics of the signal in the time domain; is a vibration acceleration; f max is a maximum frequency range point; is the energy proportion of the frequency band;

[0105] The mechanical data is subjected to wavelet transform to calculate and obtain wavelet entropy; the wavelet entropy is used as a time-frequency domain feature. For example, the wavelet entropy , wherein, J is the number of decomposition layers; qjl For the first jl Energy proportion of layer wavelet coefficient;

[0106] The time domain feature, the frequency domain feature and the time-frequency domain feature are spliced to obtain a data feature.

[0107] By multi-dimensional feature extraction on the multi-source time sequence data of the station building equipment to construct data-driven features, key data support can be provided to solve the problem that the existing technology is difficult to realize accurate extrapolation prediction when lacking historical fault data: for time sequence data containing electrical, mechanical, environmental and inspection data, the mean, variance and peak factor calculated based on the absolute mean are obtained through statistical analysis, and the time domain features formed by splicing can depict the basic state of the equipment during normal operation from the data statistical level, even without fault data, early potential degradation can also be identified through mean shift and variance increase; Fourier transform is performed on the mechanical data to obtain Fourier transform signals, and the main frequency, frequency spectrum energy and spectral entropy calculated based on the energy proportion of each frequency band are further extracted, and the frequency domain features formed by splicing can mine the frequency domain degradation signals hidden in the mechanical data, without relying on fault data, early abnormalities of the mechanical system can be captured; wavelet transform is performed on the mechanical data, and the wavelet entropy calculated based on the decomposition layer number and the energy proportion of each layer wavelet coefficient is used as the time-frequency domain feature, which can accurately capture the instantaneous changes of the signal in the time and frequency dimensions, and further mine the weak degradation information easily ignored in the normal operation data. The complete data features formed by splicing the time domain, frequency domain and time-frequency domain features cover the full-dimensional state description of the multi-source time sequence data from statistical law, frequency distribution to transient change, and a rich normal state benchmark feature library of the equipment is constructed, even if there is no historical fault data, the model can also identify the feature deviation in the equipment operation process based on the feature library, and then accurately capture the degradation trend, providing reliable data level support for extrapolation prediction, effectively breaking through the limitations of the existing technology relying on fault data to mine degradation rules, and ensuring the accuracy of extrapolation prediction.

[0108] The image features include visual features and infrared features;

[0109] The method for obtaining the image features includes:

[0110] The appearance image of the equipment is taken as the input of the convolution network to obtain convolution features;

[0111] The appearance image of the equipment is subjected to grayscale processing and Gaussian denoising processing to obtain a preprocessed image;

[0112] The preprocessed image is taken as the input of a positioning model to obtain the coordinates of the bounding box of the damaged area; the positioning model is a YOLOv8n model;

[0113] The preprocessed image is taken as an input of the U-Net model to obtain a wear area probability, and an image area with a wear area probability higher than a probability threshold is segmented as an ROI area;

[0114] The number of pixels of the damage area and the number of pixels of the ROI area contained in the bounding box coordinates of the damage area are counted, and a ratio of the number of pixels of the damage area to the number of pixels of the ROI area is calculated to obtain a pixel area proportion of the wear area;

[0115] The convolutional features and the pixel area proportion of the wear area are spliced to obtain visual features;

[0116] The infrared image is converted based on a radiation calibration model of an infrared thermal imager to obtain pixel point temperatures, such as the temperature of a pixel point , wherein, and are calibration coefficients of the infrared thermal imager; is a digital quantization value of the pixel;

[0117] An area with a continuous pixel point temperature higher than a temperature threshold and a number of pixel points higher than the temperature threshold greater than G is obtained, the number of pixel points with a temperature higher than the temperature threshold is counted, and the total number of pixel points of the infrared image is counted, and a ratio of the number of pixel points to the total number of pixel points is calculated to obtain a hot area proportion;

[0118] The highest temperature value in the pixel points with a temperature higher than the temperature threshold is counted;

[0119] The temperature mean value of the pixel points with a temperature lower than the temperature threshold is calculated, and the difference between the highest temperature value and the temperature mean value is calculated to obtain a temperature gradient;

[0120] The hot area proportion and the temperature gradient are spliced to obtain infrared features, and if there is more than one hot area, the maximum hot area proportion and the maximum temperature gradient are spliced.

[0121] By multi-dimensional feature extraction of visual images and infrared images of station building equipment, image features are constructed, which can provide key visual and thermal degradation signal support for solving the problem that existing technologies are difficult to achieve accurate extrapolation prediction when lacking historical failure data: for visual images, first, the appearance image of the equipment is input into the convolution network to obtain the convolution features containing the appearance details, then after grayscale and Gaussian noise preprocessing, the YOLOv8n positioning model is used to accurately obtain the bounding box coordinates of the damaged area, and the U-Net model is used to segment the ROI area whose wear area probability is higher than the threshold, and the ratio of the number of pixels in the damaged area to the number of pixels in the ROI area is obtained to obtain the quantitative pixel area ratio of the wear area, and the visual features formed by splicing the convolution features and the ratio can convert the qualitative degradation of the equipment appearance into quantitative and traceable feature indicators, even without historical failure data, early appearance degradation can be identified through the slight increase of the wear area ratio from 0.5% to 2%, avoiding relying on failure cases to judge the degradation degree; for infrared images, the infrared image is converted into temperature data based on the radiation calibration model of the infrared thermal imager, and the heat area with more than G continuous high-temperature pixels is further screened out, the ratio of the number of pixels in the heat area to the total number of pixels in the infrared image is obtained to obtain the heat area ratio, and the difference between the maximum temperature of the heat area and the average temperature of the non-heat area is calculated to obtain the temperature gradient, and the infrared features formed by splicing the two can capture the implicit thermal degradation of the equipment, which has existed temperature anomaly before failure, and early thermal degradation can be identified through signals such as heat area ratio expansion and temperature gradient increase without historical failure data. The visual features and infrared features cover the visible appearance degradation and implicit thermal degradation of the equipment respectively, and the visual information of the image is converted into quantitative and comparable feature indicators, and the image feature benchmark of the normal operation of the equipment is constructed, when early degradation occurs, the features will deviate from the benchmark, and the prediction model can extrapolate the subsequent degradation trajectory based on the feature deviation trend, without relying on historical failure data to learn failure patterns, effectively breaking through the limitations of existing technologies that cannot capture early degradation and are difficult to accurately extrapolate due to the lack of failure data.

[0122] The method for obtaining the physical derived features comprises:

[0123] By calculating the dynamic change rate of the degradation process physical constraint equation in the physical constraint equation, the dynamic features of degradation, i.e. wear rate, stiffness decay rate and aging acceleration factor, are extracted; the specific acquisition method comprises:

[0124] The wear rate is calculated according to the contact wear amount equation and the wear time:

[0125]

[0126] wherein, is the 24h wear amount change; is the wear time;

[0127] The stiffness attenuation rate is calculated according to the spring stiffness attenuation equation:

[0128]

[0129] The aging acceleration factor is calculated according to the winding insulation aging index equation:

[0130]

[0131] The wear rate, stiffness attenuation rate and aging acceleration factor are sequentially concatenated as a 3-dimensional feature vector to obtain a physically derived feature.

[0132] By deriving the physically derived feature based on the degradation process equation of the key physical quantity, the problem that the existing technology is difficult to accurately extrapolate and predict when lacking historical failure data can be solved to provide core physical mechanism level degradation dynamic support: for the switch cabinet contact, the wear rate is obtained by calculating the ratio of the wear amount change to the wear time within 24 hours according to the contact wear amount equation, which can directly quantify the degradation speed of the contact caused by electrical wear and mechanical wear, even without failure data, the early warning signal of the contact degradation acceleration can be accurately captured through the change of the wear rate, avoiding the limitation that the statistical features cannot distinguish between normal fluctuations and degradation trends; for the circuit breaker spring, the stiffness attenuation rate is calculated based on the spring stiffness attenuation equation, which can reflect the degree of stiffness loss caused by spring fatigue aging in real time, without referring to historical failure cases, the spring can be judged whether it enters the accelerated degradation stage through the change of the stiffness attenuation rate, ensuring that the extrapolation prediction conforms to the physical law of spring fatigue aging; for the transformer winding, the aging acceleration factor is derived by means of the winding insulation aging index equation, which can quantify the acceleration effect of temperature on winding insulation aging, even lacking insulation breakdown failure data, the future trend of insulation aging can be predicted through the change of the aging acceleration factor with temperature rise. The physically derived feature formed by concatenating the wear rate, stiffness attenuation rate and aging acceleration factor is essentially to convert the degradation mechanism of the core components of the equipment into quantifiable and traceable dynamic indicators, which can reflect the real process and speed of degradation without relying on historical failure data, providing a physical level degradation trend anchor for the prediction model, avoiding the deviation of the pure data-driven model from the real degradation direction due to overfitting of normal data when the failure sample is scarce, and at the same time making the model extrapolation prediction always follow the physical law of equipment degradation, effectively filling the gap of degradation trend judgment when lacking failure data, and providing indispensable physical mechanism support for accurately extrapolating the future degradation trajectory of the equipment.

[0133] The mixed feature is taken as the input of the physical information neural network to obtain the degradation trajectory prediction result, the health degree of the degradation trajectory prediction result is evaluated to obtain a health degree score, and corresponding warning is performed according to the health degree score.

[0134] The physical information neural network is constructed by a 2-layer LSTM, a Transformer encoder with 4 attention heads, and a spatiotemporal attention layer;

[0135] A data fitting loss is established based on the error between the actual degradation index and the predicted value. A physical constraint loss is established based on equation constraints and boundary constraints. The total loss function is calculated based on the data fitting loss and the physical constraint loss. The Adam optimizer is used to optimize the physical information neural network, and the one with the minimum total loss function is used as the training output physical information neural network. For example, the data fitting loss is as follows:

[0136] ,

[0137] The physical constraint loss is established based on the equation constraint loss and the boundary constraint loss, where the equation constraint loss is:

[0138]

[0139] The boundary constraint loss is:

[0140]

[0141] In summary, the physical constraint loss is:

[0142]

[0143] The total loss function is:

[0144]

[0145] in, To predict the step size; This represents the number of training samples; This is a true indicator of degradation. The results are for predicting the degradation trajectory; For the first training samples Predicted value of contact wear at any given moment; For the first training samples Predicted current value at any given time; For the first training samples Predicted value of contact resistance at any time; For the first Predicted cumulative number of opening and closing times for each training sample at time t'; The balance coefficient can be obtained through historical data statistics.

[0146] Methods to obtain a health score include:

[0147] quantifying the degradation trajectory prediction result based on a segmentation function constructed according to the degradation index normal threshold and the degradation index fault threshold, to obtain a health score; for example the health score at the time t wherein, is the degradation index normal threshold; is the degradation index fault threshold; is the degradation index fault threshold; the degradation trajectory prediction result at the time t;

[0148] When the health score is lower than a score threshold, a health warning is triggered, and when the health score change rate is higher than a change rate threshold, a trend warning is triggered.

[0149] By inputting the mixed features into a physical information neural network built by a 2-layer LSTM, a Transformer encoder containing 4 attention heads, and a space-time attention layer, optimizing the model by combining a specific loss function, and outputting the degradation trajectory prediction result, and then quantifying the health score by a segmentation function and triggering the corresponding warning, the problem that the prior art cannot accurately extrapolate the prediction when lacking historical failure data can be effectively solved: on the one hand, the neural network architecture does not need to rely on historical failure data to learn the failure feature pattern, but can accurately mine the potential rules related to degradation in the data by means of the time series processing capability of LSTM and Transformer and the focusing capability of space-time attention, only by using existing normal operation data or a small amount of slightly deteriorated data; on the other hand, in the loss function design, the data fitting loss ensures that the prediction result fits the existing monitoring data, while the physical constraint loss constructed based on equation constraints and boundary constraints can force the prediction result to follow the physical mechanism of device degradation, avoiding the extrapolation deviation that violates the physical law caused by overfitting of normal data when there is no failure data, and the introduction of the balance coefficient further optimizes the adaptability of data fitting and physical law through historical data statistics; at the same time, the health score is calculated by a segmentation function based on the degradation index normal threshold and the fault threshold, which can realize quantitative evaluation without relying on health data of historical failure cases, and the health warning triggered based on the health score being lower than the threshold and the trend warning triggered based on the score change rate being higher than the threshold can also capture the dynamic trend of device degradation in time when lacking failure history, ensuring that the extrapolation prediction is not only accurate but also can be converted into a health state evaluation and warning signal with actual operation and maintenance guiding significance, which overall breaks through the limitations of the prior art that cannot guarantee the reasonableness of extrapolation prediction and cannot effectively evaluate the health state due to the lack of historical failure data, and realizes accurate prediction and operation and maintenance warning of the degradation trajectory of the device in the no-failure-data scenario.

[0150] The above merely describes preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art, according to the technical solution and inventive concept of the present application, makes equivalent replacement or change within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for predicting station degradation trajectories by integrating prior physical knowledge with spatiotemporal knowledge, characterized in that, Includes the following steps: Collect and preprocess station equipment operation data to obtain multi-source heterogeneous data; Physical constraint equations for the degradation process of key physical quantities in station equipment are constructed, along with corresponding boundary constraint equations. The degradation process physical constraint equations and boundary constraint equations are then integrated to construct the physical constraint equations. A spatiotemporal knowledge graph containing a set of nodes, a set of relationships, and a set of attributes is constructed. A graph attention network is used as the graph embedding model to embed the attention weights of nodes to their neighbors, thereby obtaining a set of node embedding vectors. Feature extraction is performed on multi-source heterogeneous data to obtain data-driven features. Physical constraint equations are analyzed to obtain physical derived features, which are 3-dimensional feature vectors obtained by splicing wear rate, stiffness decay rate and aging acceleration factor. The nodes are standardized by Z-score by embedding the vector set, data-driven features, and physical derived features. The standardized nodes are then concatenated with the embedding vector set, data-driven features, and physical derived features to obtain hybrid features. The hybrid features are used as input to the physical information neural network to obtain the degradation trajectory prediction results. The health of the degradation trajectory prediction results is evaluated to obtain a health score, and corresponding warnings are issued based on the health score.

2. The method for predicting station degradation trajectories by integrating physical priors and spatiotemporal knowledge embedding as described in claim 1, characterized in that, The station equipment operation data includes electrical data, mechanical data, environmental data, visual images, and inspection data.

3. The method for predicting station degradation trajectories by integrating physical priors and spatiotemporal knowledge embedding as described in claim 1, characterized in that, The key physical quantities include switchgear contacts, circuit breaker springs, and transformer windings.

4. The method for predicting station degradation trajectories by integrating physical priors and spatiotemporal knowledge embedding as described in claim 3, is characterized in that... Methods for obtaining physical constraint equations include: By obtaining the cumulative number of opening and closing cycles and the contact resistance, and combining the electrical wear coefficient, mechanical wear coefficient, and current at the corresponding moment, an equation for contact wear is constructed: The wear of the contact at time t is: in, k w The coefficient of electrical wear; k m The mechanical wear coefficient; t 0 represents the initial time. for Current at any moment; for Constantly in contact with the resistor; The cumulative number of opening and closing operations at time t; By obtaining the initial stiffness and stress cycle number of the spring, and combining the attenuation coefficient and the fatigue life at the corresponding stress level, a spring stiffness attenuation equation is constructed: The equation for the decrease in spring stiffness at time t is: in, k sp0 This is the initial stiffness; This is the fatigue attenuation coefficient; n l For the first l Number of stress cycles per operation; N fl This refers to the fatigue life corresponding to the stress level. By obtaining the dielectric loss angle and winding temperature, and combining the temperature acceleration factor and reference temperature, an equation for the winding insulation aging index is constructed: The winding insulation aging index is: in, for Constant time interval loss angle; The acceleration factor is the temperature coefficient. for Constant winding temperature; Reference temperature; Construct boundary constraint equations, including: wear amount is non-negative; spring stiffness is not lower than the lower stiffness threshold; winding insulation aging index is not higher than the upper aging index threshold. The equations for contact wear, spring stiffness decay, winding insulation aging index, and boundary constraints are integrated into a set of equations to obtain the physical constraint equations.

5. The method for predicting station degradation trajectories by integrating physical priors and spatiotemporal knowledge embedding as described in claim 1, characterized in that, Methods for constructing spatiotemporal knowledge graphs that include sets of nodes, relationships, and attributes include: Obtain the station's coverage area, including spatial scope, equipment scope, and time scope; Entities are categorized into equipment unit entities, component entities, sensor entities, spatial region entities, and event / physical quantity entities; Define the relationships between entities, and analyze the domain, range, and type of the relationships; Define the attributes of an entity, including static attributes and dynamic attributes; A pre-defined hierarchical structure of station buildings, floors, bays, and installation sites is adopted to associate each spatial area entity with the superior area; Bind all entities within the station building to absolute coordinates based on the station building's coordinate system and relative coordinates based on the coordinate system of their respective spatial regions; Acquire station equipment testing data and automatically extract and convert it based on preset conversion tools; obtain data entities and map the data entities into a graph format according to the correspondence between entities and attributes to obtain a spatiotemporal knowledge graph.

6. The method for predicting station degradation trajectories by integrating physical priors and spatiotemporal knowledge embedding as described in claim 5, is characterized in that, Methods for obtaining node embedding vector sets include: Randomly assign an initial vector to each node; For nodes neighboring nodes By combining attention parameters, linear transformation matrices, and node embedding vectors, the node is calculated. neighboring nodes Attention weights: in, a For attention parameters; E It is a linear transformation matrix; N ( v ) is a node v The set of neighbors; This indicates vector concatenation; The slope is negative, with a value of 0.

2. h v For nodes The embedding vector; h u For nodes The embedding vector; For nodes The embedding vector; Nodes based on attention weights Update the embedding vector; obtain the node The updated embedding vector is: in, For activation functions; The updated embedding vectors of each node are concatenated to obtain the node embedding vector set.

7. The method for predicting station degradation trajectories by integrating physical priors and spatiotemporal knowledge embedding as described in claim 1, characterized in that, The data-driven features include data features and image features; Methods for obtaining data features include: Statistical analysis was performed on time-series data including electrical, mechanical, environmental, and inspection data to obtain the mean and variance. The absolute mean of the time-series data was then calculated, and the peak value was determined based on the absolute mean. Peak factor The mean, variance, and peak factor are concatenated to obtain the time-domain features; This represents the number of time-series data points. For the first One time series data; Perform a Fourier transform on the mechanical data to obtain the Fourier transform signal. Extract the dominant frequency, spectral energy, and spectral entropy from the Fourier transform signal. Concatenate the dominant frequency, spectral energy, and spectral entropy to obtain the frequency domain characteristics. The Fourier transform signal... main frequency Spectral energy Spectral entropy ,in, j The imaginary unit; f For frequency; t For time; It is the vibration acceleration; f max This represents the point with the maximum frequency range. For the first Energy percentage of each frequency band; Wavelet transform is performed on the mechanical data to calculate the wavelet entropy; the wavelet entropy is then used as a time-frequency domain feature; wavelet entropy ,in, The number of decomposition layers; q jl For the first jl Energy proportion of layer wavelet coefficients; Data features are obtained by concatenating time-domain features, frequency-domain features, and time-frequency-domain features.

8. The method for predicting station degradation trajectories by integrating physical priors and spatiotemporal knowledge embedding as described in claim 7, is characterized in that... The image features include visual features and infrared features; Methods for obtaining image features include: The device appearance image is used as input to the convolutional network to obtain convolutional features; The image of the equipment's appearance is converted to grayscale and subjected to Gaussian denoising to obtain a preprocessed image; The preprocessed image is used as input to the localization model to obtain the coordinates of the bounding box of the damaged area; the localization model is the YOLOv8n model. The preprocessed image is used as input to the U-Net model to obtain the probability of wear regions. Image regions with wear region probabilities higher than the probability threshold are segmented as ROI regions. The number of pixels in the damaged area and the number of pixels in the ROI area included in the bounding box coordinates of the damaged area are counted, and the ratio of the number of pixels in the damaged area to the number of pixels in the ROI area is calculated to obtain the pixel area ratio of the worn area. Visual features are obtained by concatenating convolutional features and the pixel area ratio of the worn area. The infrared image is converted based on the radiation calibration model of the infrared thermal imager to obtain the pixel temperature; The region is identified where the temperature of consecutive pixels is higher than a temperature threshold and the number of pixels with a temperature higher than the temperature threshold is greater than G. The number of pixels with a temperature higher than the temperature threshold and the total number of pixels in the infrared image are counted. The ratio of the number of pixels to the total number of pixels is calculated to obtain the proportion of hot areas. The highest temperature value among pixels whose temperature exceeds the temperature threshold is counted. Calculate the average temperature of pixels whose temperature is below the temperature threshold, and calculate the difference between the highest temperature value and the average temperature value to obtain the temperature gradient. Infrared features are obtained by stitching together the proportion of hot zones and the temperature gradient. If there is more than one hot zone, the largest proportion of hot zone and the largest temperature gradient are taken for stitching.

9. The method for predicting station degradation trajectories by integrating physical priors and spatiotemporal knowledge embedding as described in claim 1, characterized in that, The physical information neural network is constructed from a 2-layer LSTM, a Transformer encoder with 4 attention heads, and a spatiotemporal attention layer. A data fitting loss is established based on the error between the actual degradation index and the predicted value: The physical constraint loss is established based on the equation constraint loss and the boundary constraint loss, where the equation constraint loss is: The boundary constraint loss is: In summary, the physical constraint loss is: The total loss function is calculated based on the data fitting loss and the physical constraint loss: in, To predict the step size; This represents the number of training samples; This is a true indicator of degradation. The results are for predicting the degradation trajectory; For the first training samples Predicted value of contact wear at any given moment; For the first training samples Predicted current value at any given time; For the first training samples Predicted contact resistance value at any time; For the first Predicted cumulative number of opening and closing times for each training sample at time t'; The balancing coefficient is used; the physical information neural network with the minimum total loss function is obtained by optimizing based on the Adam optimizer and used as the training output physical information neural network. Methods to obtain a health score include: The degradation trajectory prediction results are quantified based on a piecewise function constructed according to the normal threshold and the failure threshold of the degradation index to obtain a health score: The health score at time t'' is: ， in, y norm The normal threshold for degradation indicators; y fault The failure threshold is the degradation index. for The predicted degradation trajectory at any given time; When the health score is below the score threshold, a health warning is triggered; when the rate of change of the health score is above the rate of change threshold, a trend warning is triggered.

Images