A transformer fault monitoring method and system based on artificial intelligence

Abstract

The application provides a transformer fault monitoring method and system based on artificial intelligence, and belongs to the technical field of transformers. In the method, a plurality of to-be-detected components of a target transformer are identified, the corresponding multi-modal monitoring data of each to-be-detected component is obtained, the corresponding weighted feature data of each to-be-detected component is determined based on the multi-modal monitoring data and a weight matrix of each to-be-detected component, the corresponding fault monitoring result of each to-be-detected component is output by inputting the weighted feature data into a target fault monitoring model, and the fault monitoring information of the target transformer is generated. The different target fault monitoring models are used to identify the multi-modal monitoring data of different to-be-detected components, so that the comprehensiveness and accuracy of transformer fault monitoring can be effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of transformers, and particularly relates to a transformer fault monitoring method and system based on artificial intelligence. BACKGROUND

[0002] Transformers play a very key role in power systems. Power transformers have functions of voltage level conversion, electric energy transfer and distribution, and the safe and stable operation of the transformers is extremely important for the condition of the entire power grid. If a transformer fails, it may cause equipment damage and scrap, and even lead to large-area power outage and other major life and production accidents.

[0003] With the rapid development of artificial intelligence technology, more and more scientific research companies introduce artificial intelligence into transformer fault monitoring. That is, a fault monitoring model is used to analyze and process input transformer monitoring data to achieve rapid prediction of transformer faults.

[0004] However, due to the complex structure of the transformer, which is usually composed of multiple components, and the different fault manifestations of different components, the current fault monitoring model can only roughly analyze the transformer as a whole, and it is difficult to achieve accurate prediction of transformer faults. SUMMARY

[0005] The embodiment of the present application provides a transformer fault monitoring method and system based on artificial intelligence, and the present application adopts the following technical scheme:

[0006] In a first aspect, a transformer fault monitoring method based on artificial intelligence is provided, and the method comprises the following steps:

[0007] obtaining the operating condition and multi-dimensional image data of a target transformer;

[0008] based on the multi-dimensional image data of the target transformer, identifying a plurality of to-be-detected components of the target transformer;

[0009] obtaining multi-modal monitoring data corresponding to each to-be-detected component, and determining a weight matrix of the multi-modal monitoring data of each to-be-detected component based on the operating condition;

[0010] based on the multi-modal monitoring data and the weight matrix corresponding to each to-be-detected component, determining weighted feature data corresponding to each to-be-detected component;

[0011] send each corresponding weighting feature data of the to-be-detected components to a cloud server, so that the cloud server determines a target fault monitoring model of any to-be-detected component based on a component type of the to-be-detected component in a pre-constructed fault monitoring model library, inputs the weighting feature data of the to-be-detected component into the target fault monitoring model, and outputs a corresponding fault monitoring result of the to-be-detected component;

[0012] receive the corresponding fault monitoring result of each to-be-detected component fed back by the cloud server, and generate fault monitoring information of the target transformer based on the corresponding fault monitoring result of each to-be-detected component.

[0013] In an embodiment of the present application, the multi-dimensional image data includes visible light images and infrared images.

[0014] Based on the multi-dimensional image data of the target transformer, a plurality of to-be-detected components of the target transformer are identified, including:

[0015] The visible light images and the infrared images are preprocessed to obtain visible light feature data and infrared feature data of the target transformer, respectively.

[0016] The visible light feature data and the infrared feature data are fused to obtain fused feature data.

[0017] The fused feature data is input into a pre-set classification model to obtain a plurality of to-be-detected components of the target transformer output by the classification model.

[0018] In an embodiment of the present application, the visible light images and the infrared images are preprocessed to obtain visible light feature data and infrared feature data of the target transformer, respectively, including:

[0019] The visible light images and the infrared images are registered to obtain registered visible light images and registered infrared images.

[0020] The registered visible light images are subjected to a first preprocessing operation to obtain the visible light feature data; the first preprocessing operation includes a histogram equalization operation and a sharpening operation.

[0021] The registered infrared images are subjected to a second preprocessing operation to obtain the infrared feature data; the second preprocessing operation includes a denoising operation and a non-uniformity correction operation.

[0022] In an embodiment of the present application, the visible light feature data and the infrared feature data are fused to obtain fused feature data, including:

[0023] determine a first fusion weight of the visible light feature data and a second fusion weight of the infrared feature data;

[0024] perform weighted fusion on the visible light feature data and the infrared feature data based on the first fusion weight and the second fusion weight to obtain the fusion feature data.

[0025] In an embodiment of the present application, the method further comprises:

[0026] obtain illumination information of the target transformer;

[0027] determine a first fusion weight of the visible light feature data and a second fusion weight of the infrared feature data based on the illumination information.

[0028] In an embodiment of the present application, the plurality of components to be detected include a plurality or all of a winding, an iron core, a bushing, an oil tank and a tap changer; the method further comprises:

[0029] in a case where the component to be detected is the winding, the multi-modal monitoring data of the winding include an infrared image, a vibration signal and an electrical parameter of the winding;

[0030] in a case where the component to be detected is the iron core, the multi-modal monitoring data of the iron core include an infrared image and a ground current of the iron core;

[0031] in a case where the component to be detected is the bushing, the multi-modal monitoring data of the bushing include an infrared image, a visible light image and a leakage current of the bushing;

[0032] in a case where the component to be detected is the oil tank, the multi-modal monitoring data of the oil tank include an infrared image and a visible light image of the oil tank;

[0033] in a case where the component to be detected is the tap changer, the multi-modal monitoring data of the tap changer include an infrared image, a visible light image, a vibration signal and a sound signal of the tap changer.

[0034] In an embodiment of the present application, the fault monitoring model library includes a winding fault monitoring model, an iron core fault monitoring model, a bushing fault monitoring model, an oil tank fault monitoring model and a tap changer fault monitoring model;

[0035] based on the component type of the component to be detected, determine a target fault monitoring model of the component to be detected in a pre-constructed fault monitoring model library, including:

[0036] In a case where the component type of the component to be detected is a winding, the target fault monitoring model is determined as the winding fault monitoring model; the winding fault monitoring model is trained based on sample weighted feature data of a sample winding;

[0037] In a case where the component type of the component to be detected is an iron core, the target fault monitoring model is determined as the iron core fault monitoring model; the iron core fault monitoring model is trained based on sample weighted feature data of a sample iron core;

[0038] In a case where the component type of the component to be detected is a bushing, the target fault monitoring model is determined as the bushing fault monitoring model; the bushing fault monitoring model is trained based on sample weighted feature data of a sample bushing;

[0039] In a case where the component type of the component to be detected is an oil tank, the target fault monitoring model is determined as the oil tank fault monitoring model; the oil tank fault monitoring model is trained based on sample weighted feature data of a sample oil tank;

[0040] In a case where the component type of the component to be detected is a tap changer, the target fault monitoring model is determined as the tap changer fault monitoring model; the tap changer fault monitoring model is trained based on sample weighted feature data of a sample tap changer.

[0041] In an embodiment of the present application, based on the operating condition, a weight matrix of the multi-modal monitoring data of each of the components to be detected is determined, including:

[0042] For any of the components to be detected, a fault performance coefficient of the multi-modal monitoring data of the component to be detected under the operating condition is determined; the fault performance coefficient represents a performance degree of the multi-modal monitoring data under the operating condition on a fault to be detected of the component to be detected;

[0043] Based on the fault performance coefficient, a weight matrix of the multi-modal monitoring data of the component to be detected under the operating condition is determined.

[0044] In an embodiment of the present application, the fault monitoring result includes a fault type of the component to be detected and a fault position of the fault type in the component to be detected;

[0045] Based on the respective fault monitoring result of each of the components to be detected, fault monitoring information of the target transformer is generated, including:

[0046] For any to-be-detected component, based on the fault monitoring result of the to-be-detected component, fault identification information of the to-be-detected component is generated; the fault identification information is used to indicate the fault type at the fault position of the to-be-detected component.

[0047] The fault identification information of each to-be-detected component is displayed as the fault monitoring information of the target transformer on the three-dimensional simulation model of the target transformer.

[0048] In a second aspect, based on the same inventive concept, an artificial intelligence-based transformer fault monitoring system is provided, and the system comprises:

[0049] A data acquisition module is configured to acquire the operating condition and multi-dimensional image data of a target transformer.

[0050] A component identification module is configured to identify a plurality of to-be-detected components of the target transformer based on the multi-dimensional image data of the target transformer.

[0051] A weight determination module is configured to acquire multi-modal monitoring data corresponding to each to-be-detected component, and determine a weight matrix of the multi-modal monitoring data of each to-be-detected component based on the operating condition.

[0052] A feature determination module is configured to determine weighted feature data corresponding to each to-be-detected component based on the multi-modal monitoring data corresponding to each to-be-detected component and the weight matrix.

[0053] A fault monitoring module is configured to send the weighted feature data corresponding to each to-be-detected component to a cloud server, so that the cloud server determines, for any to-be-detected component, a target fault monitoring model of the to-be-detected component in a pre-constructed fault monitoring model library based on the component type of the to-be-detected component, and inputs the weighted feature data of the to-be-detected component into the target fault monitoring model to output a corresponding fault monitoring result of the to-be-detected component.

[0054] An information generation module is configured to receive the fault monitoring result corresponding to each to-be-detected component fed back by the cloud server, and generate fault monitoring information of the target transformer based on the fault monitoring result corresponding to each to-be-detected component.

[0055] In an embodiment of the present application, the multi-dimensional image data comprises visible light images and infrared images; and the component identification module comprises:

[0056] A preprocessing submodule is configured to pre-process the visible light images and the infrared images to obtain visible light feature data and infrared feature data of the target transformer, respectively.

[0057] a feature fusion submodule, configured to perform feature fusion on the visible light feature data and the infrared feature data to obtain fusion feature data;

[0058] a component identification submodule, configured to input the fusion feature data into a preset classification model to obtain a plurality of to-be-detected components of the target transformer output by the classification model.

[0059] In an embodiment of the present application, the preprocessing submodule comprises:

[0060] a registration unit, configured to perform a registration operation on the visible light image and the infrared image to obtain a visible light registration image and an infrared registration image;

[0061] a first preprocessing unit, configured to perform a first preprocessing operation on the visible light registration image to obtain the visible light feature data; the first preprocessing operation comprises a histogram equalization operation and a sharpening operation;

[0062] a second preprocessing unit, configured to perform a second preprocessing operation on the infrared registration image to obtain the infrared feature data; the second preprocessing operation comprises a denoising operation and a non-uniformity correction operation.

[0063] In an embodiment of the present application, the feature fusion submodule comprises:

[0064] a fusion weight determination unit, configured to determine a first fusion weight of the visible light feature data and a second fusion weight of the infrared feature data;

[0065] a weighted fusion unit, configured to perform weighted fusion on the visible light feature data and the infrared feature data based on the first fusion weight and the second fusion weight to obtain the fusion feature data.

[0066] In an embodiment of the present application, the fusion weight determination unit comprises:

[0067] an illumination acquisition subunit, configured to acquire illumination information of the target transformer;

[0068] a fusion weight determination subunit, configured to determine a first fusion weight of the visible light feature data and a second fusion weight of the infrared feature data based on the illumination information.

[0069] In an embodiment of the present application, the plurality of to-be-detected components comprises a plurality or all of a winding, an iron core, a bushing, an oil tank and a tap changer; and the transformer fault monitoring system based on artificial intelligence comprises:

[0070] The first data determination module is configured to determine, when the component to be detected is the winding, that the multi-modal monitoring data of the winding comprises an infrared image, a vibration signal and an electrical parameter of the winding.

[0071] The second data determination module is configured to determine, when the component to be detected is the iron core, that the multi-modal monitoring data of the iron core comprises an infrared image and a grounding current of the iron core.

[0072] The third data determination module is configured to determine, when the component to be detected is the bushing, that the multi-modal monitoring data of the bushing comprises an infrared image, a visible light image and a leakage current of the bushing.

[0073] The fourth data determination module is configured to determine, when the component to be detected is the oil tank, that the multi-modal monitoring data of the oil tank comprises an infrared image and a visible light image of the oil tank.

[0074] The fifth data determination module is configured to determine, when the component to be detected is the tap changer, that the multi-modal monitoring data of the tap changer comprises an infrared image, a visible light image, a vibration signal and a sound signal of the tap changer.

[0075] In an embodiment of the present application, the fault monitoring model library comprises a winding fault monitoring model, an iron core fault monitoring model, a bushing fault monitoring model, an oil tank fault monitoring model and a tap changer fault monitoring model.

[0076] The fault monitoring module comprises:

[0077] The first model determination submodule is configured to determine, when the component type of the component to be detected is a winding, that the target fault monitoring model is the winding fault monitoring model; the winding fault monitoring model is trained based on sample weighted feature data of a sample winding.

[0078] The second model determination submodule is configured to determine, when the component type of the component to be detected is an iron core, that the target fault monitoring model is the iron core fault monitoring model; the iron core fault monitoring model is trained based on sample weighted feature data of a sample iron core.

[0079] The third model determination submodule is configured to determine, when the component type of the component to be detected is a bushing, that the target fault monitoring model is the bushing fault monitoring model; the bushing fault monitoring model is trained based on sample weighted feature data of a sample bushing.

[0080] The fourth model determination sub-module is configured to determine that the target fault monitoring model is an oil tank fault monitoring model when the component type of the component to be detected is an oil tank, and the oil tank fault monitoring model is trained based on sample weighted feature data of a sample oil tank.

[0081] The fifth model determination sub-module is configured to determine that the target fault monitoring model is a tapping switch fault monitoring model when the component type of the component to be detected is a tapping switch, and the tapping switch fault monitoring model is trained based on sample weighted feature data of a sample tapping switch.

[0082] In an embodiment of the present application, the weight determination module comprises:

[0083] The coefficient determination sub-module is configured to determine, for any component to be detected, a fault performance coefficient of the multi-modal monitoring data of the component to be detected under the operating condition, and the fault performance coefficient represents a performance degree of the multi-modal monitoring data of the component to be detected under the operating condition on a fault to be detected of the component to be detected.

[0084] The weight matrix determination sub-module is configured to determine, based on the fault performance coefficient, a weight matrix of the multi-modal monitoring data of the component to be detected under the operating condition.

[0085] In an embodiment of the present application, the fault monitoring result comprises a fault type of the component to be detected and a fault position of the fault type in the component to be detected.

[0086] The information generation module comprises:

[0087] The information generation sub-module is configured to generate, for any component to be detected, fault identification information of the component to be detected based on the fault monitoring result of the component to be detected, and the fault identification information is used to indicate the fault position of the fault type in the component to be detected.

[0088] The information display sub-module is configured to display the fault identification information of each component to be detected as the fault monitoring information of the target transformer on the three-dimensional simulation model of the target transformer.

[0089] In summary, the transformer fault monitoring method and system based on artificial intelligence have the following technical effects:

[0090] The embodiment of the application can accurately identify a plurality of to-be-detected components of the target transformer by acquiring multi-dimensional image data of the target transformer, and then can acquire multi-modal monitoring data corresponding to each to-be-detected component respectively, and match an appropriate weight matrix for the multi-modal monitoring data of each to-be-detected component based on the operating condition. Then, based on the multi-modal monitoring data and the weight matrix corresponding to each to-be-detected component respectively, the weighted feature data corresponding to each to-be-detected component is determined, and the weighted feature data corresponding to each to-be-detected component is sent to the cloud server, so that the cloud server determines the target fault monitoring model of the to-be-detected component in the pre-constructed fault monitoring model library based on the component type of the to-be-detected component, and inputs the weighted feature data of the to-be-detected component into the target fault monitoring model to output the corresponding fault monitoring result of the to-be-detected component. Finally, the fault monitoring result corresponding to each to-be-detected component respectively fed back by the cloud server is received, and the fault monitoring information of the target transformer is generated based on the fault monitoring result corresponding to each to-be-detected component respectively. The embodiment of the application can realize the targeted monitoring of each to-be-detected component by acquiring the multi-modal monitoring data corresponding to each to-be-detected component respectively and matching the appropriate weight matrix for the multi-modal monitoring data according to the operating condition, thereby effectively improving the comprehensiveness and accuracy of the transformer fault monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0091] Figure 1 A step schematic diagram of a transformer fault monitoring method based on artificial intelligence provided by the embodiment of the application;

[0092] Figure 2 A function module schematic diagram of a transformer fault monitoring system based on artificial intelligence provided by the embodiment of the application. DETAILED DESCRIPTION

[0093] The technical solutions in the application will be described below with reference to the drawings.

[0094] With reference to Figure 1 The embodiment of the application provides a transformer fault monitoring method based on artificial intelligence, which can include the following steps:

[0095] S101: acquiring the operating condition and multi-dimensional image data of the target transformer.

[0096] It should be noted that the execution subject of the embodiment can be a computing service device with data processing, network communication and program running functions, or an electronic device with the above functions. The present embodiment will be described taking a monitoring terminal as an example.

[0097] In the embodiment, one end of the monitoring terminal is connected with the cloud server, and the other end of the monitoring terminal can be connected with a plurality of transformers respectively. That is, the monitoring terminal can quickly and accurately monitor the faults of the plurality of transformers with the aid of the computing power of the cloud server.

[0098] Specifically, each transformer is configured with a sensor cluster including a plurality of sensors, and the sensor cluster is used to collect various monitoring data of the transformer. The monitoring terminal can monitor the faults of each transformer by acquiring the various monitoring data collected by each sensor cluster.

[0099] In the embodiment, the operating condition of the target transformer represents the state of the target transformer in operation. Specifically, according to the load condition of the target transformer, the operating condition of the target transformer can specifically include no-load operation, light-load operation, normal operation, overload operation and the like.

[0100] In the embodiment, the multi-dimensional image data of the target transformer represents image data of multiple categories. Specifically, the multi-dimensional image data can include visible light images and infrared images.

[0101] S102: Based on the multi-dimensional image data of the target transformer, a plurality of to-be-detected components of the target transformer are identified.

[0102] It should be noted that the visible light image is an image formed by light intensity information in the visible light band (about 380-780 nanometers), which can provide rich texture and structural details, and thus based on the visible light image, the external components of the transformer can be effectively observed; the infrared image is formed by capturing the thermal radiation information of the object, which can reflect the temperature distribution of the object, and thus based on the infrared image, the internal components of the transformer can be effectively observed.

[0103] In the embodiment, by comprehensively considering the multi-dimensional image data of the target transformer, accurate identification of the plurality of to-be-detected components of the target transformer can be achieved.

[0104] S103: Acquire the multi-modal monitoring data corresponding to each to-be-detected component respectively, and determine the weight matrix of the multi-modal monitoring data of each to-be-detected component based on the operating condition.

[0105] In the embodiment, considering that different to-be-detected components usually have different key parameters and fault manifestations, in order to accurately monitor different to-be-detected components, different multi-modal monitoring data can be matched for different to-be-detected components.

[0106] It should be noted that the multi-modal monitoring data represents data collected through multiple perception channels (such as vision, hearing, touch, etc.), which come from different sensors or measuring devices and can provide more comprehensive and rich information.

[0107] Exemplarily, when the component to be detected is a winding located inside the transformer, the multi-modal monitoring data can include infrared images, vibration signals and electrical parameters of the winding; when the component to be detected is a bushing located outside the transformer, the multi-modal monitoring data can include infrared images, visible light images and leakage currents of the winding.

[0108] It should be noted that the multi-modal monitoring data of the component to be detected can correspond to different weight matrices under different operating conditions, and the weight matrix includes a plurality of weight factors, which are in one-to-one correspondence with the plurality of monitoring data in the multi-modal monitoring data. Exemplarily, if the multi-modal monitoring data of a certain component to be detected includes monitoring data of three modalities, the weight matrix corresponding to the multi-modal monitoring data includes three weight factors corresponding to the monitoring data of the three modalities. The weight factor represents the influence degree of the corresponding monitoring data on the fault of the component to be detected. The greater the weight factor corresponding to a certain monitoring data of the component to be detected, the greater the relevance of the monitoring data to the component fault.

[0109] In this embodiment, it is considered that different monitoring data in the multi-modal monitoring data have different performance degrees on the component fault under different operating conditions of the target transformer. For example, under the normal operating condition, the appearance features (such as leakage, bushing state, etc.) of the target transformer are important monitoring indexes, at this time, a higher weight can be given to the visible light image, and a lower weight can be given to the infrared image; while under the overload operating condition, the operating temperature of the target transformer is a key index, at this time, the weight of the infrared image can be significantly increased, and the weight of the visible light image can be reduced.

[0110] S104: Determine the weighted feature data corresponding to each component to be detected based on the multi-modal monitoring data corresponding to each component to be detected and the weight matrix.

[0111] In this embodiment, by weighting the multi-modal monitoring data corresponding to each component to be detected by using the weight matrix, different weights can be given to different monitoring data in the multi-modal monitoring data, the capture ability of the subsequent fault monitoring model to the key monitoring data is enhanced, and thus the prediction accuracy of the model is improved.

[0112] S105: Send the weighted feature data corresponding to each component to be detected to the cloud server, so that the cloud server determines, for any component to be detected, a target fault monitoring model of the component to be detected based on the component type of the component to be detected in the pre-constructed fault monitoring model library, and inputs the weighted feature data of the component to be detected into the target fault monitoring model to output the corresponding fault monitoring result of the component to be detected.

[0113] In the embodiment, considering that the terminal has limited computing power and needs to simultaneously monitor faults of multiple transformers, in order to improve the data processing speed and ensure the timeliness of fault monitoring, the terminal sends the weighted feature data of each to-be-detected component to the cloud server after determining the weighted feature data of each to-be-detected component, so that the cloud server can use the pre-stored multiple fault monitoring models to quickly monitor faults of each to-be-detected component of different transformers.

[0114] In a specific implementation, the cloud server stores a pre-constructed fault monitoring model library and a model mapping table, where the fault monitoring model library is used to store multiple pre-trained fault monitoring models, and the model mapping table represents a correspondence relationship between different component types and different fault monitoring models. Based on the model mapping table and the component type of the to-be-detected component, the cloud server can match a corresponding fault monitoring model for each to-be-detected component in the fault monitoring model library to perform targeted fault monitoring.

[0115] S106: receiving the fault monitoring result of each to-be-detected component corresponding to each to-be-detected component fed back by the cloud server, and generating fault monitoring information of the target transformer based on the fault monitoring result of each to-be-detected component corresponding to each to-be-detected component.

[0116] In the embodiment, after the cloud server feeds back the fault monitoring result of each to-be-detected component to the monitoring terminal, the monitoring terminal can comprehensively consider the fault monitoring result of each to-be-detected component corresponding to each to-be-detected component, generate fault monitoring information of the target transformer, and display the fault monitoring information of the target transformer by using a preset display terminal, so that the monitoring personnel can obtain the fault condition of the target transformer in real time, and then take corresponding maintenance measures in a timely manner.

[0117] The transformer fault monitoring method based on artificial intelligence provided by the embodiment can realize targeted monitoring of each to-be-detected component by acquiring the multi-modal monitoring data corresponding to each to-be-detected component and matching an appropriate weight matrix for the multi-modal monitoring data according to the operating condition, and thus the comprehensiveness and accuracy of transformer fault monitoring can be effectively improved.

[0118] In a feasible implementation, S102 can specifically include the following sub-steps:

[0119] S102-1: pre-processing the visible light image and the infrared image to obtain visible light feature data and infrared feature data of the target transformer, respectively.

[0120] In the embodiment, by pre-processing the visible light image, the key features such as texture and edge can be better extracted, and more accurate input is provided for subsequent fault detection; by pre-processing the infrared image, the temperature abnormal area in the infrared image can be highlighted, and the thermal fault inside the transformer can be detected.

[0121] In a specific implementation, S102-1 can specifically include the following sub-steps:

[0122] S102-1-1: Perform a registration operation on the visible light image and the infrared image to obtain a visible light registration image and an infrared registration image.

[0123] In the embodiment, considering that the imaging principles of the visible light image and the infrared image are different, they are inconsistent in space. For example, the infrared image mainly reflects temperature information, and the visible light image provides appearance details, and the imaging devices of the two may have different fields of view, resolutions, and distortion characteristics. Therefore, in order to make full use of the advantages of the two kinds of images, they are registered to the same coordinate system for fusion processing.

[0124] In a specific implementation, SIFT (Scale-Invariant Feature Transform, a scale-invariant feature transformation algorithm) or SURF (Speeded-Up Robust Features, an accelerated robust feature extraction algorithm) or the like can be used to extract feature points in the visible light image and the infrared image; by calculating the descriptor matching between the feature points, a feature point pair reflecting the corresponding relationship between the visible light image and the infrared image is obtained; the matched feature point pair is used to obtain a transformation model; the transformation model is applied to align the infrared image to the coordinate system of the visible light image to obtain the visible light registration image and the infrared registration image.

[0125] In the embodiment, the registered visible light registration image and infrared registration image can more effectively integrate the information of the visible light image and the infrared image, and further generate high-quality fusion feature data.

[0126] S102-1-2: Perform a first pre-processing operation on the visible light registration image to obtain visible light feature data.

[0127] In the embodiment, considering that the visible light image mainly provides appearance information of the transformer, the first pre-processing operation can specifically include an image enhancement operation to improve the definition and detail performance of the visible light image.

[0128] Specifically, the image enhancement operation can include a histogram equalization operation and a sharpening operation. The histogram equalization operation can enhance the contrast of the image by adjusting the gray scale distribution of the image; the sharpening operation can use a Laplacian filter or a Sobel operator to enhance the edge information of the image, facilitating subsequent feature extraction.

[0129] S102-1-3: Perform a second preprocessing operation on the infrared registration image to obtain infrared feature data.

[0130] In this embodiment, considering that the infrared image mainly reflects temperature information, the second preprocessing operation can specifically include a denoising operation and a non-uniformity correction operation to improve the accuracy and clarity of the temperature information.

[0131] In a specific implementation, a two-point correction method can be used to perform a non-uniformity correction operation on the infrared registration image. Specifically, a plurality of images are respectively captured at low temperature (e.g., 20℃) and high temperature (e.g., 35℃); the average response value of each pixel at the two temperatures is calculated; the gain and offset correction parameters of each pixel are calculated according to the response values at low temperature and high temperature; and the infrared registration image is corrected using the gain and offset correction parameters. By performing the non-uniformity correction operation, the corrected image presents a uniform brightness distribution as a whole, reducing the brightness difference caused by inconsistent pixel responses, and more accurately reflecting the temperature distribution of the target, improving the accuracy and sensitivity of imaging.

[0132] S102-2: Perform feature fusion on the visible light feature data and the infrared feature data to obtain fusion feature data.

[0133] In a specific implementation, the gray scale features, texture features, and pixel density features of the visible light image and the infrared image can be extracted, and these features can be compressed and fused through discrete transformation (e.g., K-L transformation) to finally obtain the fusion feature data.

[0134] In a specific implementation, high-frequency and low-frequency features of the image can also be extracted through wavelet transformation, and then these features can be fused to finally obtain the fusion feature data. This method can effectively combine the details of the visible light image and the thermal information of the infrared image.

[0135] In a specific implementation, the features of the visible light image and the infrared image can also be extracted through a convolutional neural network, and then fused according to feature similarity to obtain the fusion feature data. This method can automatically learn the feature representation of the image, avoiding the manual design of complex fusion rules.

[0136] In this embodiment, to improve the accuracy of the fusion feature data, S102-2 can specifically include the following sub-steps:

[0137] S102-2-1: Determine the first fusion weight of the visible light feature data and the second fusion weight of the infrared feature data.

[0138] In the embodiment, the visible light feature data and the infrared feature data can be fused according to the environmental information of the target transformer.

[0139] In a specific implementation, considering that the illumination condition has a greater impact on the quality of the visible light image, and the infrared image mainly reflects temperature information and is less affected by the illumination, the first fusion weight and the second fusion weight can be dynamically adjusted according to the illumination condition. Specifically, when the illumination is sufficient, the visible light image has rich detail and texture information, and at this time, the first fusion weight of the visible light feature data can be increased and the second fusion weight of the infrared feature data can be reduced; when the illumination is insufficient, the temperature information of the infrared image is more advantageous, and at this time, the second fusion weight of the infrared feature data can be increased and the first fusion weight of the visible light feature data can be reduced.

[0140] S102-2-2: The visible light feature data and the infrared feature data are weighted and fused based on the first fusion weight and the second fusion weight to obtain fused feature data.

[0141] In the embodiment, by dynamically adjusting the fusion weights of the visible light feature data and the infrared feature data according to the illumination information of the transformer, the accuracy and reliability of fault monitoring can be effectively improved.

[0142] S102-3: The fused feature data is input into a preset classification model to obtain a plurality of to-be-detected components of the target transformer output by the classification model.

[0143] In the embodiment, by fusing the visible light feature data and the infrared feature data, more comprehensive fused feature data can be obtained, and then by inputting the fused feature data into the preset classification model, accurate identification of the plurality of to-be-detected components can be realized.

[0144] In the embodiment, the classification model can be constructed based on a deep convolutional neural network and trained using sample fused feature data.

[0145] In a specific implementation, the classification model can be trained in the following manner: obtaining a sample visible light image and a sample infrared image of a sample transformer, sequentially performing preprocessing and feature fusion on the sample visible light image and the sample infrared image to obtain sample fusion feature data, the sample fusion feature data containing component type labels of multiple components to be detected in the sample transformer; training an initial deep convolutional neural network using the sample fusion feature data to obtain classification results of the multiple components to be detected output by the initial deep convolutional neural network; then calculating a loss function value according to the classification results of the multiple components to be detected and the component type labels, and iteratively updating network parameters of the initial deep convolutional neural network according to the loss function value until the initial deep convolutional neural network meets a training termination condition, to obtain the classification model. After the model training is completed, a test data set can also be used for verification to evaluate the accuracy and generalization ability of the classification model.

[0146] In one possible implementation, the multiple components to be detected include multiple or all of the winding, the core, the bushing, the oil tank, and the tap changer; the transformer fault monitoring method based on artificial intelligence can further include the following steps:

[0147] S201: In the case where the component to be detected is a winding, the multi-modal monitoring data of the winding includes an infrared image, a vibration signal, and an electrical parameter of the winding; wherein the electrical parameter includes a current, a voltage, and a resistance of the winding.

[0148] It should be noted that the transformer winding generates heat during operation, and the heat is transmitted to the surface of the shell through the transformer oil, and the heat distribution can be indirectly reflected by the temperature change of the surface of the shell.

[0149] In this embodiment, by obtaining the infrared image of the winding, the thermal distribution of the transformer winding can be quickly and intuitively displayed, and by analyzing the temperature change, the overheating area of the winding can be found, which may be caused by problems such as insulation aging, poor contact, or local short circuit.

[0150] In this embodiment, the transformer winding generates vibration during operation, and the change of the vibration signal can reflect the mechanical stability of the winding. Specifically, a high-precision vibration sensor can be installed at the bottom of the winding and the side of the tank wall to collect the vibration signal of the winding, and then by analyzing the vibration signal of the winding, mechanical problems such as winding deformation, loosening, or core fault can be identified.

[0151] In this embodiment, by monitoring the change of the electrical parameter, various electrical faults of the transformer winding can be effectively reflected, such as insulation fault, partial discharge fault, and overheating fault.

[0152] S202: In the case of the component to be detected being an iron core, the multi-modal monitoring data of the iron core includes an infrared image of the iron core and a grounding current.

[0153] It should be noted that the iron core generates heat during operation, and the temperature distribution on the surface of the iron core can be quickly and intuitively detected by infrared imaging technology. If the iron core has problems such as multi-point grounding or insulation aging, it will cause local overheating, and the infrared image can clearly show these abnormal areas; the transformer iron core must have a reliable grounding point during normal operation to prevent the generation of suspended voltage due to static induction. If the iron core has multi-point grounding, a closed loop will be formed between the grounding points, resulting in circulating current, local overheating, insulation aging, and even damage to the iron core.

[0154] In this embodiment, the infrared image and the grounding current monitoring reflect the operation of the iron core from two dimensions of thermal state and electrical state, respectively. Through comprehensive analysis, the health status of the iron core can be more comprehensively evaluated, and the accuracy of fault diagnosis can be improved.

[0155] S203: In the case of the component to be detected being a bushing, the multi-modal monitoring data of the bushing includes an infrared image, a visible light image, and a leakage current of the bushing.

[0156] It should be noted that the infrared image of the bushing can quickly and intuitively show the temperature distribution on the surface of the bushing. If the bushing has problems such as local overheating, oil shortage, or insulation aging, the infrared image can clearly show these abnormal areas; the visible light image can effectively detect the appearance defects of the bushing, such as cracks, dirt, damage, and other faults; the leakage current is an important electrical parameter for evaluating the insulation state of the bushing. By monitoring the leakage current, aging, dampness, or partial discharge of the bushing insulation can be detected in a timely manner.

[0157] In this embodiment, the infrared image, the visible light image, and the leakage current monitoring reflect the operation of the bushing from three dimensions of thermal state, appearance state, and electrical state, respectively. Through comprehensive monitoring, the health status of the bushing can be more comprehensively evaluated.

[0158] S204: In the case of the component to be detected being an oil tank, the multi-modal monitoring data of the oil tank includes an infrared image and a visible light image of the oil tank.

[0159] It should be noted that the infrared image of the oil tank can quickly and intuitively show the temperature distribution on the surface of the transformer oil tank. By analyzing the temperature changes, abnormal heating areas inside or outside the oil tank can be detected in a timely manner, such as cooling system failure, oil line blockage, or local overheating; by analyzing the visible light image of the oil tank, appearance defects of the oil tank can be effectively detected, such as cracks, corrosion, paint peeling, abnormal oil level, and other faults. These defects may affect the sealing and insulation performance of the oil tank.

[0160] In the embodiment, the infrared image and the visible light image reflect the state of the oil tank from two dimensions of temperature and appearance respectively, and the health condition of the oil tank can be more comprehensively evaluated through comprehensive monitoring.

[0161] S205: In the case that the component to be detected is a tap changer, the multi-modal monitoring data of the tap changer includes an infrared image, a visible light image, a vibration signal and a sound signal of the tap changer.

[0162] It should be noted that the infrared image of the tap changer can quickly detect the thermal state of the tap changer, and through analysis of the temperature distribution, overheating areas such as poor contact of the contact, arc discharge and the like can be found in time; through analysis of the visible light image of the tap changer, appearance defects of the tap changer such as contact wear, oil stains, corrosion and the like can be effectively detected; through acquisition of the vibration signal of the tap changer, the mechanical performance state of the tap changer such as contact loosening, jamming of the transmission mechanism and the like can be effectively reflected, and specifically, a high-precision vibration sensor such as a piezoelectric acceleration sensor can be installed at a key position of the tap changer to collect the vibration signal of the tap changer; through acquisition of the sound signal of the tap changer, the action process of the tap changer such as contact switching, arc discharge and the like can be effectively reflected, and different types of faults will produce different sound frequency characteristics.

[0163] In the embodiment, the infrared image, the visible light image, the vibration signal and the sound signal reflect the operation of the tap changer from four dimensions of thermal state, appearance state, mechanical state and sound characteristics respectively, and the health condition of the tap changer can be more comprehensively evaluated through comprehensive monitoring.

[0164] In the embodiment, by acquiring different multi-modal monitoring data for different components to be detected, targeted fault monitoring can be performed on key components such as the winding, the core, the bushing, the oil tank and the tap changer in the target transformer, so as to improve the accuracy and reliability of the transformer fault monitoring.

[0165] In a feasible embodiment, the fault monitoring model library includes a winding fault monitoring model, a core fault monitoring model, a bushing fault monitoring model, an oil tank fault monitoring model and a tap changer fault monitoring model; S105 can specifically include the following sub-steps:

[0166] S105-1: In the case that the component type of the component to be detected is a winding, the target fault monitoring model is determined as the winding fault monitoring model.

[0167] In the embodiment, the winding fault monitoring model is trained based on the sample weighted feature data of the sample winding.

[0168] Specifically, the winding fault monitoring model can be trained in the following manner: an original sample data set of a sample winding is obtained, the original sample data set containing sample multi-modal monitoring data and a sample weight matrix, each sample multi-modal monitoring data containing a corresponding fault label; sample weighted feature data is obtained based on the sample multi-modal monitoring data and the sample weight matrix of the sample winding; an initial winding fault monitoring model is trained using the sample weighted feature data to obtain a fault monitoring result output by the initial winding fault monitoring model; then, a loss function value is calculated according to the fault monitoring result and the corresponding fault label, and network parameters of the initial winding fault monitoring model are iteratively updated according to the loss function value until the initial winding fault monitoring model meets a training stop condition, to obtain a final winding fault monitoring model, which is used for fault prediction of the winding.

[0169] S105-2: In a case where the component type of the component to be detected is an iron core, it is determined that the target fault monitoring model is an iron core fault monitoring model.

[0170] In this embodiment, the iron core fault monitoring model is trained based on sample weighted feature data of a sample iron core. The model training process of the iron core fault monitoring model is referred to the model training process of the winding fault monitoring model, which is not described herein again.

[0171] S105-3: In a case where the component type of the component to be detected is a bushing, it is determined that the target fault monitoring model is a bushing fault monitoring model.

[0172] In this embodiment, the bushing fault monitoring model is trained based on sample weighted feature data of a sample bushing. The model training process of the bushing fault monitoring model is referred to the model training process of the winding fault monitoring model, which is not described herein again.

[0173] S105-4: In a case where the component type of the component to be detected is an oil tank, it is determined that the target fault monitoring model is an oil tank fault monitoring model.

[0174] In this embodiment, the oil tank fault monitoring model is trained based on sample weighted feature data of a sample oil tank. The model training process of the oil tank fault monitoring model is referred to the model training process of the winding fault monitoring model, which is not described herein again.

[0175] S105-5: In a case where the component type of the component to be detected is a tap changer, it is determined that the target fault monitoring model is a tap changer fault monitoring model.

[0176] In this embodiment, the tap changer fault monitoring model is trained based on sample weighted feature data of a sample tap changer. The model training process of the tap changer fault monitoring model is referred to the model training process of the winding fault monitoring model, which is not described herein again.

[0177] In the embodiment, considering that different to-be-detected components correspond to different multi-modal monitoring data, different fault monitoring models are constructed to enable each fault monitoring model to better learn the pattern between the multi-modal monitoring data of the different to-be-detected components and the component fault, thereby improving the prediction accuracy of the model and realizing accurate identification of the component fault of each to-be-detected component.

[0178] In a feasible embodiment, S103 can specifically include the following sub-steps:

[0179] S103-1: For any to-be-detected component, determine the fault performance coefficient of the multi-modal monitoring data of the to-be-detected component under the operating condition.

[0180] It should be noted that the fault performance coefficient represents the performance degree of the multi-modal monitoring data under the operating condition on the to-be-detected fault of the to-be-detected component.

[0181] In the embodiment, by analyzing the fault feature performance of the to-be-detected component under different operating conditions of the multi-modal data for different fault types of the to-be-detected component, reasonable allocation of the fault performance coefficient can be realized. For example, the winding overheating fault is obvious in the infrared image in the multi-modal data, and the weight of the infrared image should be increased.

[0182] S103-2: Determine the weight matrix of the multi-modal monitoring data of the to-be-detected component under the operating condition based on the fault performance coefficient.

[0183] In the embodiment, after determining the fault performance coefficient of the multi-modal monitoring data of the to-be-detected component under the operating condition, normalization operation can be performed on each fault performance coefficient, and then the weight matrix of the multi-modal monitoring data of the to-be-detected component under the operating condition is determined.

[0184] In the embodiment, the fault performance coefficient of the multi-modal monitoring data under each operating condition can be reasonably determined according to the operating condition and fault feature of the transformer, and the multi-modal monitoring data is matched with a suitable weight matrix according to the operating condition of the transformer, thereby improving the accuracy and reliability of fault monitoring.

[0185] In a feasible embodiment, the fault monitoring result includes the fault type of the to-be-detected component and the fault position of the fault type in the to-be-detected component; S106 can specifically include the following sub-steps:

[0186] S106-1: For any to-be-detected component, generate fault identification information of the to-be-detected component based on the fault monitoring result of the to-be-detected component.

[0187] In the embodiment, the fault identification information is used to indicate the fault type at the fault position of the component to be detected.

[0188] S106-2: Display the fault identification information of each component to be detected as the fault monitoring information of the target transformer on the three-dimensional simulation model of the target transformer.

[0189] In the embodiment, after obtaining the fault monitoring result of each component to be detected, the fault identification information indicating the fault type at the fault position of the component to be detected can be generated, and the display device is controlled to display the fault identification information on the three-dimensional simulation model of the target transformer. In this way, the fault monitoring information of each component to be detected of the target transformer is comprehensively and intuitively output, so that the monitor can timely understand the fault type and the fault position of the target transformer, and then timely repair the target transformer, thereby effectively ensuring the operation safety of the target transformer.

[0190] Based on the same inventive concept, referring to Figure 2 , an embodiment of the present application provides a transformer fault monitoring system based on artificial intelligence 200, which comprises:

[0191] a data acquisition module 201 configured to acquire the operating condition and multi-dimensional image data of a target transformer;

[0192] a component recognition module 202 configured to recognize a plurality of components to be detected of the target transformer based on the multi-dimensional image data of the target transformer;

[0193] a weight determination module 203 configured to acquire multi-modal monitoring data corresponding to each component to be detected, and determine a weight matrix of the multi-modal monitoring data of each component to be detected based on the operating condition;

[0194] a feature determination module 204 configured to determine weighted feature data corresponding to each component to be detected based on the multi-modal monitoring data corresponding to each component to be detected and the weight matrix;

[0195] a fault monitoring module 205 configured to send the weighted feature data corresponding to each component to be detected to a cloud server, so that the cloud server determines a target fault monitoring model of the component to be detected in a pre-constructed fault monitoring model library based on the component type of the component to be detected, and inputs the weighted feature data of the component to be detected into the target fault monitoring model to output the fault monitoring result corresponding to the component to be detected;

[0196] The information generation module 206 is configured to receive the respective fault monitoring result of each to-be-detected component fed back by the cloud server, and generate the fault monitoring information of the target transformer based on the respective fault monitoring result of each to-be-detected component.

[0197] In an embodiment of the present application, the multi-dimensional image data includes a visible light image and an infrared image; and the component identification module 202 includes:

[0198] The preprocessing sub-module is configured to pre-process the visible light image and the infrared image to obtain visible light feature data and infrared feature data of the target transformer, respectively.

[0199] The feature fusion sub-module is configured to perform feature fusion on the visible light feature data and the infrared feature data to obtain fusion feature data.

[0200] The component identification sub-module is configured to input the fusion feature data into a preset classification model to obtain a plurality of to-be-detected components of the target transformer output by the classification model.

[0201] In an embodiment of the present application, the preprocessing sub-module includes:

[0202] The registration unit is configured to perform a registration operation on the visible light image and the infrared image to obtain a visible light registered image and an infrared registered image.

[0203] The first preprocessing unit is configured to perform a first preprocessing operation on the visible light registered image to obtain the visible light feature data; and the first preprocessing operation includes a histogram equalization operation and a sharpening operation.

[0204] The second preprocessing unit is configured to perform a second preprocessing operation on the infrared registered image to obtain the infrared feature data; and the second preprocessing operation includes a denoising operation and a non-uniformity correction operation.

[0205] In an embodiment of the present application, the feature fusion sub-module includes:

[0206] The fusion weight determination unit is configured to determine a first fusion weight of the visible light feature data and a second fusion weight of the infrared feature data.

[0207] The weighted fusion unit is configured to perform weighted fusion on the visible light feature data and the infrared feature data based on the first fusion weight and the second fusion weight to obtain the fusion feature data.

[0208] In an embodiment of the present application, the fusion weight determination unit includes:

[0209] The light acquisition sub-unit is configured to acquire light information of the target transformer.

[0210] The fusion weight determination subunit is configured to determine a first fusion weight of the visible light feature data and a second fusion weight of the infrared feature data based on the illumination information.

[0211] In an embodiment of the present application, the plurality of components to be detected include a plurality or all of the winding, the core, the bushing, the oil tank, and the tap changer; the transformer fault monitoring system 200 based on artificial intelligence includes:

[0212] The first data determination module is configured to, in a case where the component to be detected is the winding, determine that the multi-modal monitoring data of the winding includes the infrared image, the vibration signal, and the electrical parameter of the winding.

[0213] The second data determination module is configured to, in a case where the component to be detected is the core, determine that the multi-modal monitoring data of the core includes the infrared image and the ground current of the core.

[0214] The third data determination module is configured to, in a case where the component to be detected is the bushing, determine that the multi-modal monitoring data of the bushing includes the infrared image, the visible light image, and the leakage current of the bushing.

[0215] The fourth data determination module is configured to, in a case where the component to be detected is the oil tank, determine that the multi-modal monitoring data of the oil tank includes the infrared image and the visible light image of the oil tank.

[0216] The fifth data determination module is configured to, in a case where the component to be detected is the tap changer, determine that the multi-modal monitoring data of the tap changer includes the infrared image, the visible light image, the vibration signal, and the sound signal of the tap changer.

[0217] In an embodiment of the present application, the fault monitoring model library includes a winding fault monitoring model, a core fault monitoring model, a bushing fault monitoring model, an oil tank fault monitoring model, and a tap changer fault monitoring model.

[0218] The fault monitoring module 205 includes:

[0219] The first model determination sub-module is configured to, in a case where the component type of the component to be detected is the winding, determine that the target fault monitoring model is the winding fault monitoring model; the winding fault monitoring model is trained based on the sample weighted feature data of the sample winding.

[0220] The second model determination sub-module is configured to, in a case where the component type of the component to be detected is the core, determine that the target fault monitoring model is the core fault monitoring model; the core fault monitoring model is trained based on the sample weighted feature data of the sample core.

[0221] The third model determination sub-module is configured to determine, in a case where the component type of the component to be detected is a bushing, that the target fault monitoring model is a bushing fault monitoring model, and the bushing fault monitoring model is trained based on sample weighted feature data of a sample bushing.

[0222] The fourth model determination sub-module is configured to determine, in a case where the component type of the component to be detected is an oil tank, that the target fault monitoring model is an oil tank fault monitoring model, and the oil tank fault monitoring model is trained based on sample weighted feature data of a sample oil tank.

[0223] The fifth model determination sub-module is configured to determine, in a case where the component type of the component to be detected is a tap switch, that the target fault monitoring model is a tap switch fault monitoring model, and the tap switch fault monitoring model is trained based on sample weighted feature data of a sample tap switch.

[0224] In an embodiment of the present application, the weight determination module 203 comprises:

[0225] The coefficient determination sub-module is configured to determine, for any component to be detected, a fault performance coefficient of the multi-modal monitoring data of the component to be detected under the operating condition, and the fault performance coefficient represents a performance degree of the multi-modal monitoring data under the operating condition on the detected fault of the component to be detected.

[0226] The weight matrix determination sub-module is configured to determine, based on the fault performance coefficient, a weight matrix of the multi-modal monitoring data of the component to be detected under the operating condition.

[0227] In an embodiment of the present application, the fault monitoring result comprises a fault type of the component to be detected and a fault position of the fault type in the component to be detected.

[0228] The information generation module 206 comprises:

[0229] The information generation sub-module is configured to generate, for any component to be detected, fault identification information of the component to be detected based on the fault monitoring result of the component to be detected, and the fault identification information is used to indicate the fault position of the fault type in the component to be detected.

[0230] The information display sub-module is configured to display the fault identification information of each component to be detected as the fault monitoring information of the target transformer on the three-dimensional simulation model of the target transformer.

[0231] It should be noted that the specific implementation of the transformer fault monitoring system 200 based on artificial intelligence in the embodiments of the present application is described with reference to the specific implementation of the transformer fault monitoring method based on artificial intelligence in the first aspect of the embodiments of the present application, which will not be described here.

[0232] The above-described embodiments can be implemented in whole or in part by software, hardware (such as a circuit), firmware, or any combination thereof. When implemented in software, the above-described embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions according to the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable systems. The computer program or instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another computer-readable storage medium, for example, the computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center through a wired (for example, infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. containing one or more available medium sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0233] It should be understood that the term "and / or" herein merely describes an association relationship of associated objects, which means that there can be three relationships, for example, A and / or B can represent three cases of A alone, A and B together, and B alone, where A and B can be singular or plural. In addition, the character " / " herein generally represents an "or" relationship between the associated objects before and after it, but it can also represent an "and / or" relationship. The specific meaning can be understood according to the context before and after it.

[0234] In this application, "at least one" means one or more, and "multiple" means two or more. "At least one of the following" or similar expressions means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can represent a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, and c can be single or multiple.

[0235] It should be understood that in various embodiments of the present application, the size of the sequence number of the above-described processes does not mean the order of execution, and the execution order of the processes should be determined according to their functions and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0236] Those skilled in the art can clearly understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0237] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the above-described system, system and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0238] In several embodiments provided in the present application, it should be understood that the disclosed system, system and method can be implemented in other ways. For example, the above-described system embodiments are only schematic, for example, the division of units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be through some interface, indirect coupling or communication connection between the systems or units can be electrical, mechanical or other forms.

[0239] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or they can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0240] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0241] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the parts of the prior art or the parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of each embodiment method of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0242] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An artificial intelligence based transformer fault monitoring method, characterized by, The method comprises: acquiring the operating condition and multi-dimensional image data of a target transformer; based on the multi-dimensional image data of the target transformer, identifying a plurality of components to be detected of the target transformer; acquiring multi-modal monitoring data corresponding to each of the components to be detected, and determining a weight matrix of the multi-modal monitoring data of each of the components to be detected based on the operating condition; based on the multi-modal monitoring data and the weight matrix corresponding to each of the components to be detected, determining weighted feature data corresponding to each of the components to be detected; sending the weighted feature data corresponding to each of the components to be detected to a cloud server, so that the cloud server, for any of the components to be detected, determines a target fault monitoring model of the component to be detected based on the component type of the component to be detected in a pre-constructed fault monitoring model library, and inputs the weighted feature data of the component to be detected into the target fault monitoring model to output a fault monitoring result corresponding to the component to be detected; receiving the fault monitoring result corresponding to each of the components to be detected fed back by the cloud server, and generating fault monitoring information of the target transformer based on the fault monitoring result corresponding to each of the components to be detected.

2. The method of claim 1, wherein the method is based on artificial intelligence. The multi-dimensional image data comprises visible light images and infrared images; based on the multi-dimensional image data of the target transformer, identifying a plurality of components to be detected of the target transformer, comprises: preprocessing the visible light images and the infrared images to obtain visible light feature data and infrared feature data of the target transformer respectively; performing feature fusion on the visible light feature data and the infrared feature data to obtain fusion feature data; inputting the fusion feature data into a pre-set classification model to obtain a plurality of components to be detected of the target transformer output by the classification model.

3. The method of claim 2, wherein the method further comprises: The preprocessing of the visible light images and the infrared images to obtain the visible light feature data and the infrared feature data of the target transformer respectively comprises: performing registration operation on the visible light images and the infrared images to obtain visible light registered images and infrared registered images; performing first preprocessing operation on the visible light registered images to obtain the visible light feature data; the first preprocessing operation comprises histogram equalization operation and sharpening operation; performing second preprocessing operation on the infrared registered images to obtain the infrared feature data; the second preprocessing operation comprises denoising operation and non-uniformity correction operation.

4. The method of claim 2, wherein the method further comprises: The feature fusion of the visible light feature data and the infrared feature data to obtain fusion feature data comprises: determining first fusion weight of the visible light feature data and second fusion weight of the infrared feature data; based on the first fusion weight and the second fusion weight, performing weighted fusion on the visible light feature data and the infrared feature data to obtain the fusion feature data.

5. The method of claim 4, wherein the method further comprises: Determining the first fusion weight of the visible light feature data and the second fusion weight of the infrared feature data comprises: acquiring illumination information of the target transformer; Determine a first fusion weight of the visible light feature data and a second fusion weight of the infrared feature data based on the illumination information.

6. The method of claim 1, wherein the method is based on artificial intelligence. The plurality of components to be detected include a plurality of components selected from a winding, an iron core, a bushing, an oil tank, and a tap changer; The method further includes: In a case where the component to be detected is the winding, the multi-modal monitoring data of the winding include an infrared image, a vibration signal, and an electrical parameter of the winding; In a case where the component to be detected is the iron core, the multi-modal monitoring data of the iron core include an infrared image and a ground current of the iron core; In a case where the component to be detected is the bushing, the multi-modal monitoring data of the bushing include an infrared image, a visible light image, and a leakage current of the bushing; In a case where the component to be detected is the oil tank, the multi-modal monitoring data of the oil tank include an infrared image and a visible light image of the oil tank; In a case where the component to be detected is the tap changer, the multi-modal monitoring data of the tap changer include an infrared image, a visible light image, a vibration signal, and a sound signal of the tap changer.

7. The method of claim 1, wherein the method is based on artificial intelligence. The fault monitoring model library includes a winding fault monitoring model, an iron core fault monitoring model, a bushing fault monitoring model, an oil tank fault monitoring model, and a tap changer fault monitoring model; Based on the component type of the component to be detected, determine a target fault monitoring model of the component to be detected in a pre-constructed fault monitoring model library, including: In a case where the component type of the component to be detected is the winding, determine that the target fault monitoring model is the winding fault monitoring model; The winding fault monitoring model is trained based on sample weighted feature data of a sample winding; In a case where the component type of the component to be detected is the iron core, determine that the target fault monitoring model is the iron core fault monitoring model; the iron core fault monitoring model is trained based on sample weighted feature data of a sample iron core; In a case where the component type of the component to be detected is the bushing, determine that the target fault monitoring model is the bushing fault monitoring model; the bushing fault monitoring model is trained based on sample weighted feature data of a sample bushing; In a case where the component type of the component to be detected is the oil tank, determine that the target fault monitoring model is the oil tank fault monitoring model; the oil tank fault monitoring model is trained based on sample weighted feature data of a sample oil tank; In a case where the component type of the component to be detected is the tap changer, determine that the target fault monitoring model is the tap changer fault monitoring model; the tap changer fault monitoring model is trained based on sample weighted feature data of a sample tap changer.

8. The method of claim 1, wherein the method is based on artificial intelligence. Based on the operating condition, determine a weight matrix of the multi-modal monitoring data of each of the components to be detected, including: For any of the components to be detected, determine a fault performance coefficient of the multi-modal monitoring data of the component to be detected under the operating condition; the fault performance coefficient represents a performance degree of the multi-modal monitoring data on a to-be-detected fault of the component to be detected under the operating condition; Determine a weight matrix of the multi-modal monitoring data of the component to be detected under the operation condition based on the failure performance coefficient.

9. The method of claim 1, wherein the method is based on artificial intelligence. The failure monitoring result includes a failure type of the component to be detected and a failure position of the failure type in the component to be detected. Generate failure monitoring information of the target transformer based on the failure monitoring result of each component to be detected, including: For any component to be detected, generate failure identification information of the component to be detected based on the failure monitoring result of the component to be detected; the failure identification information is used to indicate the failure position of the failure type in the component to be detected; Display the failure identification information of each component to be detected as the failure monitoring information of the target transformer on a three-dimensional simulation model of the target transformer.

10. An artificial intelligence based transformer fault monitoring system characterized in that, The system includes: A data acquisition module is configured to acquire an operation condition and multi-dimensional image data of a target transformer. A component identification module is configured to identify a plurality of components to be detected of the target transformer based on the multi-dimensional image data of the target transformer. A weight determination module is configured to acquire multi-modal monitoring data corresponding to each component to be detected, and determine a weight matrix of the multi-modal monitoring data of each component to be detected based on the operation condition. A feature determination module is configured to determine weighted feature data corresponding to each component to be detected based on the multi-modal monitoring data and the weight matrix corresponding to each component to be detected. A failure monitoring module is configured to send the weighted feature data corresponding to each component to be detected to a cloud server, so that the cloud server determines a target failure monitoring model of the component to be detected based on a component type of the component to be detected in a pre-constructed failure monitoring model library, and inputs the weighted feature data of the component to be detected into the target failure monitoring model to output a failure monitoring result corresponding to the component to be detected. A result generation module is configured to receive the failure monitoring result corresponding to each component to be detected fed back by the cloud server, and generate failure monitoring information of the target transformer based on the failure monitoring result corresponding to each component to be detected.

Images