Multi-source fusion and meta-learning driven intelligent diagnosis method and system for power equipment

Abstract

The invention relates to the field of artificial intelligence, and discloses a multi-source fusion and meta-learning driven intelligent diagnosis method and system for power equipment. The method comprises the following steps: collecting multi-source operation characteristic data such as vibration, current and temperature of equipment; performing space-time alignment and normalization processing on the acquired data, and extracting a standardized feature vector; multi-source features are subjected to fusion analysis through a dynamic weight fusion model, feature matching is performed in combination with a fault knowledge base, and intelligent diagnosis output of equipment fault types, predicted downtime and risk levels is realized; when the diagnosis confidence exceeds a threshold value or an actual fault is confirmed, triggering a meta-learning module to optimize and update the fusion weight; a diagnosis result is displayed through the edge terminal and fed back to the cloud platform, and the knowledge base and model performance is further improved. The method has adaptivity and intelligence, can improve the accuracy and response efficiency of power equipment fault diagnosis, and is suitable for various scenes such as a transformer substation and a power distribution network.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and specifically, to an intelligent diagnosis method and system for power equipment driven by multi-source fusion and meta-learning. Background Art

[0002] With the improvement of the digital and intelligent level of power systems, the fault diagnosis technology of power equipment (such as transformers, circuit breakers, generators, etc.) has increasingly become the key to ensuring the safe and stable operation of the power grid. Traditional power equipment fault diagnosis methods mainly rely on single-sensor data (such as vibration, current, or temperature) and static analysis models based on rules or thresholds, lacking the collaborative fusion of multi-source data. For example, the high-frequency vibration signals (4 - 8 kHz) of early bearing wear are easily masked by noise, and it is difficult to accurately identify such weak fault features only relying on current or temperature data, resulting in a relatively high fault omission rate.

[0003] In recent years, multi-source data fusion and meta-learning technologies have gradually received attention in the field of industrial fault diagnosis. Multi-source fusion methods can improve the completeness of fault features by integrating heterogeneous data such as vibration, current, and temperature; meta-learning can quickly adapt to new tasks using a small number of samples and reduce the dependence on labeled data. However, the spatio-temporal alignment method for multi-source data is not perfect, and the normalization process is prone to losing key feature information such as high-frequency vibration; the application of meta-learning in power equipment diagnosis is still limited to the simulation environment and has not been combined with the edge computing architecture, making it difficult to meet the real-time requirements.

[0004] Based on the above background, the present invention provides an intelligent diagnosis method and system for power equipment driven by multi-source fusion and meta-learning. The present invention retains the physical meaning of multi-source features through spatio-temporal alignment, realizes high-precision diagnosis by combining dynamic weight fusion and fault knowledge base matching, and uses the meta-learning mechanism to optimize the model weights online. The present invention significantly improves the early fault detection rate and diagnosis timeliness while reducing the dependence on labeled data, providing an innovative solution for the intelligent and safe operation and maintenance of power equipment. Summary of the Invention

[0005] The purpose of the embodiments of this application is to provide an intelligent diagnosis method and system for power equipment driven by multi-source fusion and meta-learning. Through the improved dynamic weight fusion mechanism and meta-learning online optimization, the model can be quickly adapted with only a small number of samples, significantly reducing the demand for labeled data; providing an innovative solution for power equipment intelligent diagnosis with high precision, strong adaptability, and low implementation cost.

[0006] To achieve the above purpose, this application provides the following technical solutions:

[0007] In a first aspect, the present invention discloses an intelligent diagnosis method for power equipment driven by multi-source fusion and meta-learning, comprising the following steps: S1. Collect multi-source operation characteristic data such as vibration, current, and temperature of the equipment; S2. Preprocess and perform spatio-temporal alignment on the collected multi-source data to extract standardized feature vectors; S3. Perform fusion analysis on the multi-source features through a dynamic weight fusion model, and perform feature matching in combination with a fault knowledge base to achieve intelligent diagnosis output of the equipment fault type, predicted downtime, and risk level; S4. When the diagnosis confidence exceeds the threshold or an actual fault is confirmed, trigger the meta-learning module to optimize and update the fusion weights; S5. Display the diagnosis result through an edge terminal and feedback it to the cloud platform to further improve the knowledge base and model performance.

[0008] More specifically, for the above-mentioned intelligent diagnosis method for power equipment driven by multi-source fusion and meta-learning, in step S1, it includes: installing vibration sensors at key parts of the equipment to obtain three-dimensional ( axis, axis, axis) vibration signals during operation; collecting the real-time current signal of the equipment through a current sensor; extracting the surface temperature distribution of the equipment through a thermal imaging camera and obtaining the temperature data of key internal points of the equipment in combination with internal temperature sensors.

[0009] More specifically, for the above-mentioned intelligent diagnosis method for power equipment driven by multi-source fusion and meta-learning, the preprocessing and spatio-temporal alignment of the multi-source data in step S2 include:

[0010] Adopting the wavelet packet energy method for the vibration signal to extract the energy parameters in the frequency band of 4 - 8 kHz;

[0011] Adopting d-q axis harmonic analysis for the current data to extract the total harmonic distortion (THD) index;

[0012] Performing image processing on the thermal imaging image to extract the highest surface temperature.

[0013] Performing window alignment through a unified timestamp to obtain a vibration feature vector , a current feature vector , and a temperature feature vector ;

[0014] Each type of feature has independent diagnostic significance in the physical system, and no normalization processing is performed during the spatio-temporal alignment process, and the following information is respectively retained:

[0015] The vibration feature retains the RMS value and peak acceleration within the window; the current feature includes the instantaneous value and harmonic amplitude; the temperature feature includes the highest surface temperature, the temperature of the internal temperature measurement point, and their temperature difference.

[0016] More specifically, for the multi-source fusion and meta-learning-driven intelligent diagnosis method of power equipment described above, it is characterized in that in step S3, the dynamic weight fusion model for fusing and analyzing multi-source features is as follows:

[0017] ,

[0018] where , , are respectively the learnable dynamic weights of vibration features, current features, and temperature features, is the activation function, represents the diagnostic confidence after fusion.

[0019] The fused features are matched with the fault templates in the fault knowledge base by cosine similarity:

[0020] ,

[0021] where represents the template vector of the th fault mode. The fault type with the highest output similarity and its similarity are output. When the similarity exceeds the set similarity threshold, it is diagnosed as the corresponding fault type; the value range of the set threshold is 0.85 to 0.95, preferably 0.90.

[0022] According to the fusion output and the fault model matching result, the equipment status is divided into different risk levels,

[0023] The risk level division criteria are as follows:

[0024] The equipment risk level is divided into the following three categories:

[0025] (Low risk): Slight anomalies or initial fault signs are detected, which do not affect operation for the time being and can be continuously monitored;

[0026] (Medium risk): The fault features are obvious but not yet stable. It is recommended to handle them in the planned maintenance window, and there is a certain operation risk;

[0027] (High risk): The probability of failure is high, and it is expected to fail in a short time. Immediate shutdown for maintenance is required.

[0028] More specifically, for the intelligent diagnosis method of power equipment driven by multi-source fusion and meta-learning described above, it is characterized in that in step S3, the predicted downtime is calculated based on the matching result between the fusion features and the historical cases of the same type in the fault knowledge base, the current risk level, and the diagnostic confidence level:

[0029]

[0030] where is the average failure time of the historical cases of the th type of fault that is matched, is the risk level factor ( ), and are adjustment coefficients, is the confidence function (such as linear or exponential form).

[0031] More specifically, for the intelligent diagnosis method of power equipment driven by multi-source fusion and meta-learning described above, it is characterized in that in step S4, the meta-learning optimization module is triggered under the following conditions. When the diagnostic confidence level or a fault occurs in actual verification, the meta-learning module is enabled to optimize the fusion weights:

[0032] ,

[0033] where the loss function is defined using the mean square error:

[0034] ,

[0035] where represents the target confidence level corresponding to the fusion feature sample.

[0036] More specifically, for the intelligent diagnosis method of power equipment driven by multi-source fusion and meta-learning described above, it is characterized in that the diagnostic results in step S5 include: fault type alarm, display of equipment risk level, and output of predicted downtime. All data are fed back to the cloud platform through the edge terminal for continuous model optimization and historical data archiving.

[0037] In a second aspect, the present invention discloses a system for implementing an intelligent diagnosis method of power equipment driven by multi-source fusion and meta-learning, which is characterized in that the system includes: a multi-source data acquisition unit, an edge computing unit, a cloud platform management unit, and a human-machine interface unit.

[0038] The multi-source data acquisition unit is used to collect vibration signals, current signals, internal temperature, and thermal imaging image data during equipment operation, and extract temperature features from the thermal imaging images through an image processing module.

[0039] An edge computing unit, connected to the multi-source data acquisition unit, the edge computing unit includes:

[0040] A spatio-temporal alignment and feature extraction module, configured to preprocess the acquired data and extract corresponding features;

[0041] An intelligent fusion diagnosis module, configured to calculate a diagnosis output according to a fusion model and match it with a fault template stored in a fault knowledge base to implement fault type identification, predicted downtime prediction, and risk level classification;

[0042] A meta-learning weight update module, configured to online update dynamic weights according to actual diagnosis results and a preset loss function.

[0043] A cloud platform management unit, configured to store a fault knowledge base, implement data interaction and model distribution with the edge computing unit, and support global optimization of the model by a meta-learning engine.

[0044] A human-machine interface unit, configured to display and record diagnosis results on-site, provide functions of fault alarm, device status display, data recording, and historical data query, and perform data interaction with the cloud platform to implement end-to-end real-time monitoring and diagnostic feedback.

[0045] Compared with the prior art, the beneficial effects of the present invention are: through multi-source data fusion and in combination with spatio-temporal alignment technology, the present invention retains key features such as high-frequency vibration, current harmonics, and temperature gradient of physical dimensions, improving the accuracy of fault diagnosis; by introducing a self-learning dynamic weight mechanism, the diagnosis accuracy of the model under different working conditions is improved, and parameter adjustment by humans is not required. The present invention can significantly improve the accuracy, timeliness, and economy of power equipment fault diagnosis, providing an innovative solution for the safe operation and maintenance of smart grids. Description of the Drawings

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required to be used in the embodiments of the present application. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative efforts.

[0047] Figure 1 A flowchart of a multi-source fusion and meta-learning driven intelligent diagnosis method for power equipment provided by the application.

[0048] Figure 2 An architecture diagram of a system of a multi-source fusion and meta-learning driven intelligent diagnosis method for power equipment provided by the application. Detailed Embodiments

[0049] Next, the technical solutions in the embodiments of the present application will be described in conjunction with the accompanying drawings in the embodiments of the present application. It should be noted that: Similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0050] The term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the phrase "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0051] Terms such as "first", "second", etc. are only used to distinguish one entity or operation from another entity or operation, and cannot be construed as indicating or implying relative importance, nor can they be construed as requiring or implying any actual relationship or order between these entities or operations.

[0052] The basic idea of the present invention is to collect multi-source operation characteristic data such as vibration, current and temperature of the device; preprocess and spatio-temporally align the collected multi-source data to extract standardized feature vectors; perform fusion analysis on the multi-source features through a dynamic weight fusion model, and perform feature matching in combination with a fault knowledge base to achieve intelligent diagnosis output of the device fault type, predicted downtime and risk level; when the diagnosis confidence exceeds the threshold or an actual fault is confirmed, trigger the meta-learning module to optimize and update the fusion weights; display the diagnosis result through the edge terminal and feedback it to the cloud platform to further improve the knowledge base and model performance. Through multi-source data fusion, combined with spatio-temporal alignment technology, key features such as high-frequency vibration with physical dimension, current harmonics, and temperature gradient are retained to improve the accuracy of fault diagnosis; by introducing a self-learning dynamic weight mechanism, the diagnosis accuracy of the model under different working conditions is improved, and no manual parameter adjustment is required.

[0053] As Figure 1 shown, the present invention provides a multi-source fusion and meta-learning-driven intelligent diagnosis method for power equipment, including the following steps:

[0054] S1. Collect multi-source operation characteristic data such as vibration, current and temperature of the device;

[0055] Install vibration sensors at key parts of the device to obtain three-dimensional ( axis, axis, Vibration signals of the shaft); collect the real-time current signal of the device through a current sensor; extract the surface temperature distribution of the device through a thermal imaging camera, and obtain the temperature data of key points inside the device in combination with internal temperature sensors.

[0056] S2. Preprocess and align the multi-source data in space and time, and extract the standardized feature vectors;

[0057] Adopt the wavelet packet energy method for the vibration signal to extract the energy parameters in the frequency band of 4 - 8 kHz;

[0058] Adopt d-q axis harmonic analysis for the current data to extract the total harmonic distortion (THD) index;

[0059] Perform image processing on the thermal imaging image to extract the highest surface temperature.

[0060] Perform window alignment through a unified timestamp to obtain the vibration feature vector 、current feature vector 、temperature feature vector ;

[0061] Each type of feature has independent diagnostic significance in the physical system, and no normalization process is performed during the space-time alignment process. The following information is retained respectively:

[0062] The vibration feature retains the RMS value and peak acceleration within the window; the current feature includes the instantaneous value and harmonic amplitude; the temperature feature includes the highest surface temperature, the temperature of the internal temperature measurement point, and their temperature difference.

[0063] S3. Perform fusion analysis on the multi-source features through a dynamic weight fusion model, and perform feature matching in combination with the fault knowledge base to achieve intelligent diagnostic output of the device fault type, predicted downtime, and risk level;

[0064] 1) Perform dynamic weight fusion for the fusion analysis of multi-source features. The dynamic weight fusion model:

[0065] ,

[0066] where , , are the learnable dynamic weights of the vibration feature, current feature, and temperature feature respectively, is the activation function, represents the diagnostic confidence after fusion.

[0067] 2) Fuse the feature and perform cosine similarity matching with the fault templates in the fault knowledge base:

[0068] ,

[0069] Among them, represents the template vector of the th fault mode. Output the fault type with the highest similarity and its similarity. When the similarity exceeds the set similarity threshold, diagnose it as the corresponding fault type; the value range of the set threshold is 0.85 to 0.95, preferably 0.90.

[0070] 3) According to the fusion output and the fault model matching result, divide the device status into different risk levels.

[0071] The risk level division criteria:

[0072] The device risk level is divided into the following three categories:

[0073] (Low risk): Slight anomalies or initial fault signs are detected, which do not affect the operation for the time being and can be continuously monitored;

[0074] (Medium risk): The fault characteristics are obvious but not yet stable. It is recommended to handle it in the planned maintenance window, and there is a certain operation risk;

[0075] (High risk): The probability of failure is high, and it is expected to fail in a short time. Immediate shutdown for maintenance is required.

[0076] 4) Calculate the estimated downtime based on the matching result of the fusion features and the historical cases of the same type in the fault knowledge base, the current risk level, and the diagnostic confidence:

[0077]

[0078] Among them is the average historical failure time of the th type of fault for matching, is the risk level factor ( ), and are adjustment coefficients, is the confidence function (such as linear or exponential form).

[0079] S4. When the diagnostic confidence exceeds the threshold or the actual fault is confirmed, trigger the meta-learning module to optimize and update the fusion weights;

[0080] The meta-learning optimization module is triggered under the following conditions. When the diagnostic confidence or an actual verified fault occurs, enable the meta-learning module to optimize the fusion weights:

[0081] ,

[0082] Among them, the loss function is defined by the mean square error:

[0083] ,

[0084] where represents the target confidence corresponding to the fused feature sample.

[0085] S5. Display the diagnosis result through the edge terminal and feedback it to the cloud platform to further improve the knowledge base and model performance.

[0086] The diagnosis result includes: fault type alarm, display of equipment risk level, and output of the predicted downtime. All data are fed back to the cloud platform through the edge terminal for continuous model optimization and historical data archiving.

[0087] As Figure 2 shown, the present invention provides an architecture diagram of a system for a multi-source fusion and meta-learning-driven intelligent diagnosis method for power equipment. The system includes: a multi-source data acquisition unit, an edge computing unit, a cloud platform management unit, and a human-machine interface unit.

[0088] The multi-source data acquisition unit is used to collect vibration signals, current signals, internal temperature, and thermal imaging image data during equipment operation, and extract temperature features from the thermal imaging images through an image processing module.

[0089] The edge computing unit is connected to the multi-source data acquisition unit. The edge computing unit includes:

[0090] A spatio-temporal alignment and feature extraction module, which is used to preprocess the collected data and extract corresponding features;

[0091] An intelligent fusion diagnosis module, which is used to calculate the diagnosis output according to the fusion model and match it with the fault templates stored in the fault knowledge base to realize fault type identification, predicted downtime prediction, and risk level classification;

[0092] A meta-learning weight update module, which is used to online update the dynamic weight according to the actual diagnosis result and the preset loss function.

[0093] The cloud platform management unit is used to store the fault knowledge base, realize data interaction and model distribution with the edge computing unit, and support the global optimization of the model by the meta-learning engine.

[0094] The human-machine interface unit is used to display and record the diagnosis result on site, provide functions such as fault alarm, equipment status display, data recording, and historical data query, and perform data interaction with the cloud platform to realize end-to-end real-time monitoring and diagnosis feedback.

[0095] The above are only the embodiments of the present application and are not intended to limit the protection scope of the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. An intelligent diagnosis method for power equipment driven by multi-source fusion and meta-learning, characterized in that It includes the following steps: S1. Collect multi-source operation characteristic data such as the vibration, current, and temperature of the device; S2. Preprocess and perform spatio-temporal alignment on the collected multi-source data, and extract standardized feature vectors; S3. Conduct fusion analysis on multi-source features through a dynamic weight fusion model, perform feature matching in combination with a fault knowledge base, and achieve intelligent diagnostic output of the device fault type, predicted downtime, and risk level; S4. When the diagnostic confidence exceeds the threshold or an actual fault is confirmed, trigger the meta-learning module to optimize and update the fusion weights; S5. Display the diagnostic results through the edge terminal and feedback them to the cloud platform to further improve the knowledge base and model performance.

2. The intelligent diagnosis method for power equipment driven by multi-source fusion and meta-learning according to claim 1, wherein Step S1 includes: installing vibration sensors at key parts of the device to obtain three-dimensional ( axis, axis, axis) vibration signals during operation; collecting real-time current signals of the device through a current sensor; extracting the surface temperature distribution of the device through a thermal imaging camera, and obtaining the temperature data of key internal points of the device in combination with internal temperature sensors.

3. A multi-source fusion and meta-learning-driven intelligent diagnosis method for power equipment according to claim 1, wherein, The preprocessing and spatio-temporal alignment of the multi-source data in step S2 include: Adopt the wavelet packet energy method for the vibration signal to extract the energy parameters in the frequency band of 4 - 8 kHz; Adopt d-q axis harmonic analysis for the current data to extract the total harmonic distortion (THD) index; Perform image processing on the thermal imaging image to extract the highest surface temperature; Align the windows by unifying the timestamps to obtain the vibration feature vector , the current feature vector , and the temperature feature vector ; Each type of feature has independent diagnostic significance in the physical system, and no normalization processing is performed during the spatio-temporal alignment process. The following information is retained respectively: The vibration features retain the RMS value and peak acceleration within the window; the current features include the instantaneous value and harmonic amplitude; the temperature features include the highest surface temperature, the temperature of the internal temperature measurement point, and their temperature difference.

4. A multi-source fusion and meta-learning driven intelligent diagnosis method for power equipment according to claim 1, characterized in that, In step S3, a dynamic weight fusion model for fusing and analyzing multi-source features: ; where , , are respectively the learnable dynamic weights of vibration features, current features, and temperature features, is the activation function, represents the diagnostic confidence after fusion; Fusion feature Perform cosine similarity matching with the fault templates in the fault knowledge base: ; among them, represents the template vector of the th fault mode; output the fault type with the highest similarity and its similarity. When the similarity exceeds the set similarity threshold, it is diagnosed as the corresponding fault type; the value range of the set threshold is 0.85 to 0.95, preferably 0.90; According to the fusion output and the fault model matching results, the device status is divided into different risk levels The risk level classification criteria are as follows: ; The equipment risk level is divided into the following three categories: (Low risk): Slight anomalies or initial signs of failure are detected, which do not affect operation for the time being and continuous monitoring can be carried out; (Medium risk): The fault characteristics are obvious but not yet stable. It is recommended to handle it during the planned maintenance window, with a certain operation risk; (High risk): High probability of failure, expected to fail within a short time, immediate shutdown for maintenance is required.

5. A multi-source fusion and meta-learning driven intelligent diagnosis method for power equipment according to claim 1, characterized in that, The estimated downtime in step S3 is calculated based on the matching result of the fusion features with historical cases of the same type in the fault knowledge base, the current risk level, and the diagnostic confidence: ; where is the average failure time of the th type of fault history in the match, is the risk level factor ( ), and are adjustment coefficients, is the confidence function, and the confidence function is in a linear or exponential form.

6. A multi-source fusion and meta-learning driven intelligent diagnosis method for power equipment according to claim 1, characterized in that, In the step S4, the meta-learning optimization module is triggered under the following conditions. When the diagnostic confidence or an actual verification fails, the meta-learning module is enabled to optimize the fusion weights: ; where the loss function is defined by the mean square error: ; where represents the target confidence corresponding to the fusion feature sample.

7. A multi-source fusion and meta-learning driven intelligent diagnosis method for power equipment according to claim 1, characterized in that, The diagnostic results in step S5 include: fault type alarm, display of the device risk level, and output of the predicted downtime. All data are fed back to the cloud platform through the edge terminal for continuous model optimization and historical data archiving.

8. A system for implementing the multi-source fusion and meta-learning driven intelligent diagnosis method for power equipment described in claims 1-6, characterized in that, The system includes: a multi-source data acquisition unit, an edge computing unit, a cloud platform management unit, and a human-machine interface unit; The multi-source data acquisition unit is used to collect the vibration signal, current signal, internal temperature, and thermal imaging image data during the operation of the device, and extract temperature features from the thermal imaging image through an image processing module; The edge computing unit is connected to the multi-source data acquisition unit. The edge computing unit includes: A spatio-temporal alignment and feature extraction module, which is used to preprocess the collected data and extract corresponding features; An intelligent fusion diagnosis module, which is used to calculate diagnostic output according to the fusion model and match the fault templates stored in the fault knowledge base to achieve fault type identification, predicted downtime prediction, and risk level classification; A meta-learning weight update module, which is used to online update the dynamic weights according to the actual diagnostic results and a preset loss function; The cloud platform management unit is used to store the fault knowledge base, realize data interaction and model distribution with the edge computing unit, and at the same time support the global optimization of the model by the meta-learning engine; The human-machine interface unit is used to display and record the diagnostic results on-site, provide functions such as fault alarm, device status display, data recording, and historical data query, and perform data interaction with the cloud platform to achieve end-to-end real-time monitoring and diagnostic feedback.

Images