Appearance abnormality diagnosis model construction method and appearance abnormality diagnosis method and system

Abstract

The invention provides a construction method of an appearance abnormity diagnosis model, and an appearance abnormity diagnosis method and system, and the construction method comprises the steps: carrying out the image collection of a plurality of pole-mounted transformers which have appearance abnormity and are in an operation state, and obtaining an image data set; acquiring an operation data set and an environment data set of the plurality of pole-mounted transformers at historical moments; wherein the historical moment is a time sequence before an identification moment, and the identification moment is a corresponding moment when the appearance abnormity of the plurality of pole-mounted transformers is identified; and training a convolutional neural network by adopting the image data set, the operation data set, the environment data set and the pole-mounted transformer appearance anomaly knowledge graph to obtain an appearance anomaly diagnosis model of the pole-mounted transformer. According to the invention, the accuracy and reliability of appearance abnormity diagnosis of the pole-mounted transformer can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment fault diagnosis, and in particular to a method for constructing an appearance abnormality diagnosis model, an appearance abnormality diagnosis method and a system. Background Art

[0002] Pole-mounted transformers are widely used in the distribution chain of power systems. Their stable and reliable operation has a significant impact on the overall power supply quality and safety. To ensure the normal operation of pole-mounted transformers, the industry generally uses traditional diagnostic algorithms or manual inspections to monitor and analyze transformer appearance data. However, such methods often only handle limited data types and lack sufficient attention to external factors such as the equipment's operating environment and load conditions. This makes it difficult to maintain accurate and timely diagnostic results in various complex situations.

[0003] Conventional technologies for identifying transformer appearance anomalies rely heavily on single-source data or basic feature extraction methods, lacking effective correlation with multiple environmental and operating factors. To improve detection efficiency, some existing solutions have attempted to introduce more real-time monitoring equipment. However, insufficient linkage between subsystems and imperfect information exchange make it difficult to timely account for both equipment structural information and dynamic operating conditions, resulting in the inability to continuously iterate and optimize the diagnostic system. Once rare or complex fault scenarios occur, traditional methods tend to be slow to respond to risk signs, making it difficult to identify potential root causes in a short period of time. Summary of the Invention

[0004] In order to solve the problems of the prior art, the present invention proposes a method for constructing an appearance abnormality diagnosis model, an appearance abnormality diagnosis method and a system, aiming to improve the accuracy and reliability of appearance abnormality diagnosis of pole-mounted transformers.

[0005] The purpose of the present invention is achieved by adopting the following technical solutions:

[0006] In one aspect, an embodiment of the present invention provides a method for constructing a model for diagnosing abnormalities in the appearance of a pole-mounted transformer, the method comprising:

[0007] Capture images of multiple pole-mounted transformers that have abnormal appearances and are in operation to obtain an image dataset.

[0008] Obtaining operation data sets and environmental data sets of the plurality of pole-mounted transformers at historical moments; wherein the historical moments are time series before the identification moment, and the identification moment is the moment corresponding to when the plurality of pole-mounted transformers are identified to have appearance abnormalities;

[0009] The image dataset, the operation dataset, the environment dataset and the knowledge graph of pole-mounted transformer appearance anomaly are used to train a convolutional neural network to obtain a pole-mounted transformer appearance anomaly diagnosis model.

[0010] Optionally, the use of the image dataset, the operation dataset, the environment dataset, and the knowledge graph of pole-mounted transformer appearance anomalies to train a convolutional neural network to obtain a pole-mounted transformer appearance anomaly diagnosis model includes:

[0011] Using a probabilistic decision model, based on the operation data set and the environmental data set, inducement data of the plurality of pole-mounted transformers with abnormal appearances are inferred to obtain inducement data sets of the plurality of pole-mounted transformers with abnormal appearances;

[0012] Using the inducement dataset and the image dataset, the convolutional neural network is trained to obtain an intermediate abnormality diagnosis model;

[0013] The pole-mounted transformer appearance anomaly knowledge graph and the image dataset are used to train the intermediate anomaly diagnosis model to obtain the appearance anomaly diagnosis model.

[0014] Optionally, the adopting of a probabilistic decision model to perform inducement data inference on the multiple pole-mounted transformers with abnormal appearances based on the operation data set and the environmental data set to obtain inducement data sets of the multiple pole-mounted transformers with abnormal appearances includes:

[0015] Based on each pole-mounted transformer with an abnormal appearance, the probabilistic decision model is used to infer the operating weight ratio of the operating data set of the pole-mounted transformer with the abnormal appearance at a historical moment, and the environmental weight ratio of the environmental data set of the pole-mounted transformer with the abnormal appearance at a historical moment;

[0016] Multiplying the operation weight ratio by the operation data set of the pole-mounted transformer with the appearance abnormality at the historical moment to obtain an intermediate operation data set, and multiplying the environment weight ratio by the environment data set of the pole-mounted transformer with the appearance abnormality at the historical moment to obtain an intermediate environment data set;

[0017] The intermediate operation data set and the intermediate environment data set are fused to obtain the inducement data set of the pole-mounted transformer with abnormal appearance.

[0018] Optionally, the use of the inducement dataset and the image dataset to train the convolutional neural network to obtain an intermediate abnormality diagnosis model includes:

[0019] Inputting the image dataset into the convolutional neural network for abnormality diagnosis to obtain an initial diagnostic dataset;

[0020] Using a cosine similarity loss function, determining an initial similarity loss between the inducement dataset and the initial diagnosis dataset;

[0021] Based on the initial similarity loss, the network parameters of the convolutional neural network are adjusted until the intermediate abnormality diagnosis model whose output meets preset conditions is obtained.

[0022] Optionally, the use of the pole-mounted transformer appearance anomaly knowledge graph and the image dataset to train the intermediate anomaly diagnosis model to obtain the appearance anomaly diagnosis model includes:

[0023] Inputting the image dataset into the pole-mounted transformer appearance anomaly knowledge graph to perform anomaly query to obtain an anomaly dataset;

[0024] Inputting the image data set into the intermediate abnormality diagnosis model to perform abnormality diagnosis to obtain an intermediate diagnosis data set;

[0025] A cosine similarity loss function is used to determine the intermediate similarity loss between the abnormal data set and the intermediate diagnostic data set, and based on the intermediate similarity loss, the network parameters of the intermediate abnormality diagnosis model are adjusted until the appearance abnormality diagnosis model whose output meets the preset conditions is obtained.

[0026] Optionally, the step of inputting the image dataset into the pole-mounted transformer appearance anomaly knowledge graph to perform an anomaly query to obtain an anomaly dataset includes:

[0027] Determining an initial query range in the pole-mounted transformer appearance anomaly knowledge graph based on the pole-mounted transformer model carried in the image dataset;

[0028] The abnormal data set is determined within the initial query range according to the type of the appearance abnormality carried in the image data set.

[0029] Correspondingly, an embodiment of the present invention further provides a system for constructing an appearance abnormality diagnosis model for a pole-mounted transformer, the system comprising:

[0030] The first image acquisition module is used to acquire images of a plurality of pole-mounted transformers that have abnormal appearances and are in operation, to obtain an image data set;

[0031] An acquisition module is configured to acquire an operation data set and an environmental data set of the plurality of pole-mounted transformers at historical moments; wherein the historical moments are time series before an identification moment, and the identification moment is a moment corresponding to when an abnormal appearance of the plurality of pole-mounted transformers is identified;

[0032] The training module is used to train a convolutional neural network using the image dataset, the operation dataset, the environmental dataset, and the knowledge graph of the appearance anomaly of the pole-mounted transformer to obtain an appearance anomaly diagnosis model for the pole-mounted transformer.

[0033] Optionally, the training module includes:

[0034] an inference unit, configured to use a probabilistic decision model to perform inducement data inference on the plurality of pole-mounted transformers with abnormal appearances based on the operation data set and the environmental data set, to obtain inducement data sets of the plurality of pole-mounted transformers with abnormal appearances;

[0035] A first training subunit is configured to train the convolutional neural network using the inducement dataset and the image dataset to obtain an intermediate abnormality diagnosis model;

[0036] The second training unit is used to train the intermediate abnormality diagnosis model using the pole-mounted transformer appearance abnormality knowledge graph and the image dataset to obtain the appearance abnormality diagnosis model.

[0037] Optionally, the inference unit is specifically used to infer, based on each pole-mounted transformer with an abnormal appearance, the operating weight ratio of the operating data set of the pole-mounted transformer with an abnormal appearance at a historical moment, and the environmental weight ratio of the environmental data set of the pole-mounted transformer with an abnormal appearance at a historical moment, using the probabilistic decision model; multiplying the operating weight ratio by the operating data set of the pole-mounted transformer with an abnormal appearance at a historical moment to obtain an intermediate operating data set, and multiplying the environmental weight ratio by the environmental data set of the pole-mounted transformer with an abnormal appearance at a historical moment to obtain an intermediate environmental data set; and fusing the intermediate operating data set and the intermediate environmental data set to obtain the inducement data set of the pole-mounted transformer with an abnormal appearance.

[0038] Optionally, the first training unit is specifically used to input the image dataset into the convolutional neural network for abnormality diagnosis to obtain an initial diagnostic dataset; use a cosine similarity loss function to determine the initial similarity loss between the cause dataset and the initial diagnostic dataset; and adjust the network parameters of the convolutional neural network based on the initial similarity loss until the intermediate abnormality diagnosis model whose output meets preset conditions is obtained.

[0039] Optionally, the second training unit is specifically used to input the image dataset into the pole-mounted transformer appearance anomaly knowledge graph for anomaly query to obtain an anomaly dataset; input the image dataset into the intermediate anomaly diagnosis model for anomaly diagnosis to obtain an intermediate diagnostic dataset; use the cosine similarity loss function to determine the intermediate similarity loss between the anomaly dataset and the intermediate diagnostic dataset, and adjust the network parameters of the intermediate anomaly diagnosis model based on the intermediate similarity loss until the appearance anomaly diagnosis model whose output meets preset conditions is obtained.

[0040] Optionally, the second training unit is specifically used to determine an initial query range in the pole-mounted transformer appearance anomaly knowledge graph based on the pole-mounted transformer model carried in the image dataset; and determine the anomaly dataset within the initial query range based on the type of appearance anomaly carried in the image dataset.

[0041] In another aspect, an embodiment of the present invention provides a method for diagnosing appearance abnormalities of a pole-mounted transformer, the method comprising:

[0042] Capturing images of the pole-mounted transformer to be tested in operation to obtain images to be tested;

[0043] When abnormal features are identified in the image to be tested as existing in the appearance of the pole-mounted transformer to be tested, the image to be tested is input into the appearance abnormality diagnosis model obtained by any of the above-mentioned construction methods to perform abnormality diagnosis, and obtain an abnormality diagnosis data set of the abnormal features.

[0044] Optionally, identifying abnormal features in the appearance of the pole-mounted transformer to be tested in the image to be tested includes:

[0045] Using an abnormal target detection algorithm to identify abnormalities in the image to be tested, and obtain features to be evaluated;

[0046] When the abnormality confidence score of the feature to be evaluated is greater than a preset threshold, it is determined that an abnormal feature exists in the appearance of the transformer on the pole to be tested identified in the image to be tested.

[0047] Correspondingly, an embodiment of the present invention further provides a system for diagnosing abnormal appearance of a pole-mounted transformer, the system comprising:

[0048] The second image acquisition module is used to acquire an image of the pole-mounted transformer to be tested in operation to obtain an image to be tested;

[0049] The abnormality diagnosis module is used to input the image to be tested into the appearance abnormality diagnosis model obtained by the above-mentioned construction system to perform abnormality diagnosis when abnormal features are identified in the image to be tested, so as to obtain an abnormality diagnosis data set of the abnormal features.

[0050] Optionally, the abnormality diagnosis module further includes:

[0051] an identification unit, configured to use an abnormal target detection algorithm to perform abnormality identification on the image to be tested and obtain features to be evaluated;

[0052] When the abnormality confidence score of the feature to be evaluated is greater than a preset threshold, it is determined that an abnormal feature exists in the appearance of the transformer on the pole to be tested identified in the image to be tested.

[0053] On the other hand, an embodiment of the present invention further provides an electronic device, comprising: at least one processor and a memory; the memory and the processor are connected via a bus;

[0054] The memory is used to store one or more programs;

[0055] When the one or more programs are executed by the at least one processor, the method for constructing a model for diagnosing anomalies of the appearance of a pole-mounted transformer as described above and the method for diagnosing anomalies of the appearance of a pole-mounted transformer as described above are implemented.

[0056] On the other hand, an embodiment of the present invention further provides a readable storage medium having an execution program stored thereon. When the execution program is executed, it implements a method for constructing a model for diagnosing the appearance abnormality of a pole-mounted transformer as described in any of the above-mentioned methods, and a method for diagnosing the appearance abnormality of a pole-mounted transformer as described in any of the above-mentioned methods.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] An embodiment of the present invention provides a method and system for constructing a pole-mounted transformer appearance anomaly diagnosis model. During the execution of this method, a convolutional neural network is trained using a knowledge graph of pole-mounted transformer appearance anomalies and multi-source data (image datasets, historical operation datasets, and environmental datasets) of multiple pole-mounted transformers with appearance anomalies and in operation, thereby obtaining an appearance anomaly diagnosis model with improved overall anomaly diagnosis effectiveness. In this way, by fully integrating the operational data and environmental information of the pole-mounted transformer, problems such as missed detection or misjudgment when appearance anomalies occur can be avoided to the greatest extent possible. Furthermore, combined with the knowledge graph of pole-mounted transformer appearance anomalies, the cause of the appearance anomaly can be more accurately located, thereby improving the generalization capability of the obtained appearance anomaly diagnosis model and enabling it to adapt to complex and changing environments. Furthermore, the appearance anomaly diagnosis model can be used to improve the accuracy and reliability of pole-mounted transformer appearance anomaly diagnosis.

[0059] Embodiments of the present invention provide a method and system for diagnosing appearance anomalies of pole-mounted transformers. During the execution of this method, when abnormal features of the appearance of the pole-mounted transformer under test are identified in an image, an appearance anomaly diagnosis model trained using multi-source data and related knowledge graph technology is used to perform an anomaly diagnosis on the image of the pole-mounted transformer under test, thereby generating an anomaly diagnosis dataset with high accuracy and reliability. This not only improves the accuracy and stability of pole-mounted transformer appearance anomaly diagnosis, thereby reducing misjudgments and missed detections, but also enhances the intelligent level of appearance anomaly diagnosis prediction and operation and maintenance management for pole-mounted transformers.

[0060] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions provided by the embodiments of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive work, among which:

[0062] Figure 1 A schematic flow chart of a method for constructing a model for diagnosing abnormalities in the appearance of a pole-mounted transformer provided by an embodiment of the present invention;

[0063] Figure 2 A schematic flow chart of a method for diagnosing abnormalities in the appearance of a pole-mounted transformer provided in an embodiment of the present invention;

[0064] Figure 3A schematic diagram of the structure of a system for identifying and diagnosing appearance anomalies of a pole-mounted transformer provided by an embodiment of the present invention;

[0065] Figure 4 A schematic diagram of the composition of a system for constructing an appearance abnormality diagnosis model for a pole-mounted transformer provided by an embodiment of the present invention;

[0066] Figure 5 A schematic diagram of the composition of a system for diagnosing abnormalities in the appearance of a pole-mounted transformer provided by an embodiment of the present invention;

[0067] Figure 6 A schematic diagram of the composition of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The following describes the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art will readily appreciate the other advantages and benefits of the present invention from the disclosure herein. The present invention may also be implemented or applied through various other specific embodiments, and the various details in this specification may be modified or altered based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are intended only to illustrate the present invention and are not intended to limit the scope of protection of the present invention.

[0069] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0070] In the following description, the terms "first\second\third" are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It is understandable that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present invention described herein can be implemented in an order other than that illustrated or described herein.

[0071] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art in the art of the embodiments of the present invention. The terms used herein are only for the purpose of describing the embodiments of the present invention and are not intended to limit the embodiments of the present invention.

[0072] Example 1

[0073] See also Figure 1 FIG. 1 is a flow chart of a method for constructing a model for diagnosing abnormalities in the appearance of a pole-mounted transformer according to an embodiment of the present invention, wherein the method includes the following steps:

[0074] Step 101: Capture images of a plurality of pole-mounted transformers that have abnormal appearances and are in operation to obtain an image data set.

[0075] In some embodiments of the present invention, a pole-mounted transformer generally refers to an outdoor distribution transformer mounted on a pole. Furthermore, multiple pole-mounted transformers in operation with abnormal appearances may each have different or partially identical abnormal appearances. Exemplarily, these abnormal appearances include, but are not limited to, surface oil stains, surface cracks, surface rust, and missing components.

[0076] It should be noted that multiple pole-mounted transformers with abnormal appearance and in operation can be in the same power area or in different power areas, and the present invention is not limited to this.

[0077] In some embodiments of the present invention, image or video data is collected from a plurality of pole-mounted transformers that have abnormal appearances and are in operation, thereby obtaining the image dataset. The image dataset can be stored in the form of images or in the form of a video frame sequence.

[0078] Specifically, a high-definition camera and / or infrared thermal imager installed around the external structure of the pole-mounted transformer can be used to collect (capture) the overall appearance of the pole-mounted transformer, thereby generating an image dataset presenting the appearance (information) of the pole-mounted transformer. The high-definition camera can have automatic zoom and nighttime infrared fill light functions to enable clear image acquisition under various lighting conditions.

[0079] It should be noted that if the collected image dataset is a video frame sequence, the video frame sequence can adopt a dynamic frame sampling method to reduce data storage and computing overhead, while ensuring that the key image information corresponding to the moment when the appearance anomaly is identified is not lost.

[0080] In some embodiments of the present invention, first, images of multiple pole-mounted transformers with abnormal appearance and in operation are collected to obtain an initial image set; then, data preprocessing is performed on the initial image set to obtain an image data set; wherein the data preprocessing includes but is not limited to: contrast stretching, gamma correction, grayscale correction, color adjustment, image enhancement, denoising and restoration, color normalization, edge enhancement and noise suppression, etc., so as to improve or enhance the subsequent identifiability of abnormal features on the surface or appearance of the pole-mounted transformer.

[0081] In some embodiments of the present invention, high-definition cameras and / or infrared thermal imagers can be used at fixed time intervals to capture images of multiple pole-mounted transformers that have abnormal appearances and are in operation, thereby obtaining an initial image set (including infrared images obtained using infrared thermal imaging). Frame sampling processing can then be performed on the captured initial image set at fixed time intervals. To ensure the quality of subsequently obtained images, the initial image set can also be standardized. For example, images of varying resolutions within the initial image set can be resized to a uniform size, and image quality can be enhanced using methods such as grayscale transformation and histogram equalization to obtain an image dataset that reduces the effects of illumination and noise. Furthermore, for image datasets stored as video data, keyframe extraction techniques can be used to select representative key images from consecutive frames for processing, thereby reducing computational overhead and improving subsequent recognition efficiency.

[0082] It should be noted that in this embodiment of the present invention, when infrared thermal imaging is used to capture images of multiple pole-mounted transformers in operation and exhibiting visual anomalies, the resulting infrared images can be converted into pseudo-color images using a temperature gradient mapping method to enhance visualization of areas with temperature anomalies. Furthermore, edge detection based on computer vision techniques can be used to highlight the structural contours of the pole-mounted transformers, providing clearer features for subsequent target detection.

[0083] Step 102: Obtain operation data sets and environmental data sets of the plurality of pole-mounted transformers at historical moments.

[0084] The historical moments are time series before the identification moment, and the identification moment is the moment corresponding to when it is identified that the multiple pole-mounted transformers have appearance abnormalities.

[0085] In some embodiments of the present invention, the times corresponding to the identification of appearance abnormalities in multiple pole-mounted transformers may be the same or different, and the present invention does not impose any limitation on this; at the same time, the historical moment is a time series before the identification moment, which may refer to a time series corresponding to one month before the identification moment, or a time series corresponding to one week before the identification moment.

[0086] For example, the time corresponding to the identification of an appearance abnormality (crack) of pole transformer 1 is 2025.4.10.12:00:00, the time corresponding to the identification of an appearance abnormality (device misalignment) of pole transformer 2 is 2025.4.08.11:23:45, and the time corresponding to the identification of an appearance abnormality (oil leakage) of pole transformer 3 is 2025.3.15.08:33:12.

[0087] In some embodiments of the present invention, a variety of different sensors can be used to collect and obtain operating data sets and environmental data sets of multiple pole-mounted transformers at historical moments; wherein, the collection and acquisition of the operating data sets mainly rely on sensors installed on the pole-mounted transformers, such as: voltage sensors, current sensors, temperature sensors, and load rate detection devices, etc., and the operating data sets can be stored in a time series manner and time-series aligned in subsequent processing to ensure the timeliness and consistency of the data.

[0088] Correspondingly, the collection and acquisition of environmental data sets primarily relies on a set of meteorological sensors installed on or around the pole-mounted transformer (e.g., at a meteorological monitoring point near the transformer), including but not limited to temperature and humidity sensors, wind speed and direction sensors, and atmospheric pressure sensors. Alternatively, real-time meteorological information for the area where the pole-mounted transformer is located can be obtained through remote access to a meteorological data platform. To improve data reliability, multiple meteorological data sources for the pole-mounted transformer within the same time period can be verified to remove potential outliers.

[0089] In some embodiments of the present invention, the operating data set includes but is not limited to parameters such as voltage data, current data, temperature, load rate, etc., and the environmental data set includes but is not limited to parameters such as temperature, humidity, and wind speed.

[0090] For example, if it is identified at moment 1 that the pole-mounted transformer A has an abnormal appearance (cracks on the surface), the voltage data, current data, temperature and load rate of the pole-mounted transformer A within one month before moment 1 can be obtained accordingly, as well as the environmental data of the pole-mounted transformer A within one month before moment 1, namely: temperature, humidity and wind speed, etc.

[0091] In some embodiments of the present invention, data preprocessing can be performed on the acquired operational and environmental datasets of multiple pole-mounted transformers at historical moments to improve data quality and enhance the accuracy of subsequent analysis. Here, the acquired operational and environmental datasets can first be denoised, using sliding window filtering, low-pass filtering, or Kalman filtering to eliminate abnormal data points within a short period of time, while data interpolation can be used to fill in possible missing values.

[0092] Furthermore, after acquiring the operational and environmental datasets of multiple pole-mounted transformers at historical moments, these datasets can be formatted and aligned. Operational and environmental meteorological data are typically stored in structured data formats, such as time series databases or the lightweight data exchange format JavaScript Object Notation (JSON). Using a unified indexing system, data in different formats can be standardized for storage, facilitating subsequent fusion analysis. During the data alignment process, timestamp correction methods can be used to ensure time synchronization across all data sources, preventing time deviations from impacting the effective association of multi-source data (operational and environmental datasets).

[0093] Step 103: Use the image dataset, the operation dataset, the environment dataset, and the knowledge graph of pole-mounted transformer appearance anomalies to train a convolutional neural network to obtain a pole-mounted transformer appearance anomaly diagnosis model.

[0094] In some embodiments of the present invention, Convolutional Neural Networks (CNN) is a type of feedforward neural network that includes convolution calculations and has a deep structure, and is one of the representative algorithms of deep learning.

[0095] It should be noted that CNNs typically employ a multi-layered architecture, consisting of an input layer, multiple convolutional layers, a pooling layer, a fully connected layer, and an output layer. The input layer receives data and normalizes it to maintain a consistent numerical scale. The convolutional layer utilizes multi-channel feature extraction techniques to analyze the input data. For example, one-dimensional convolution is used to extract trend features from time series data, while two-dimensional convolution is used to extract features of faulty areas in images. The pooling layer reduces data dimensionality, improving computational efficiency while reducing the risk of overfitting. The fully connected layer fuses the features extracted from each channel and uses activation functions such as Softmax or Sigmoid to generate a classification output, which determines the diagnosis of the appearance anomaly of the pole-mounted transformer.

[0096] In some embodiments of the present invention, a pole-mounted transformer appearance anomaly diagnosis model can be based on physical information: a convolutional neural network and an adaptive learning module included therein, enabling diagnosis of pole-mounted transformer appearance anomalies, namely, deep feature extraction and anomaly identification of pole-mounted transformer appearance anomalies. The pole-mounted transformer appearance anomaly diagnosis model can receive image / video data exhibiting abnormal features, perform feature analysis and anomaly identification using a deep learning algorithm, and continuously optimize the model's generalization capability and diagnostic accuracy through an adaptive learning mechanism to identify the cause of abnormalities in pole-mounted transformers exhibiting abnormal features (appearance anomalies).

[0097] In some embodiments of the present invention, the knowledge graph of pole-mounted transformer appearance anomalies can be generated in advance based on empirical values; the knowledge graph of pole-mounted transformer appearance anomalies can be described as: a knowledge base that graphically represents the logical relationship between the appearance anomalies of the pole-mounted transformer and the cause data of its appearance anomalies.

[0098] It should be noted that a knowledge graph is a knowledge base that graphically represents knowledge such as entities, relationships, and attributes. By representing knowledge in a structured manner, it enables computers to better understand and process human language. A knowledge graph is typically a large, semi-structured, subject-oriented, multimodal knowledge base that contains information such as various entities, relationships, and attributes. This information is processed and reasoned through a series of algorithms and models, allowing computers to automatically acquire, reason, and generate new knowledge from it.

[0099] In some embodiments of the present invention, a convolutional neural network can be batch-trained using an image dataset, an operational dataset, an environmental dataset, and a knowledge graph of pole-mounted transformer appearance anomalies to obtain a pole-mounted transformer appearance anomaly diagnosis model. For example, the convolutional neural network is initially trained using the image dataset, the operational dataset, and the environmental dataset to obtain an intermediate anomaly diagnosis model. The intermediate anomaly diagnosis model is then trained using the image dataset and the knowledge graph of pole-mounted transformer appearance anomalies to ultimately obtain the pole-mounted transformer appearance anomaly diagnosis model.

[0100] In some embodiments of the present invention, the above step 103 can be implemented by following the steps 1031 to 1033 ( Figure 1 Not shown):

[0101] Step 1031: Using a probabilistic decision model, based on the operation data set and the environmental data set, infer the inducement data of the multiple pole-mounted transformers with abnormal appearances, and obtain the inducement data sets of the multiple pole-mounted transformers with abnormal appearances.

[0102] In some embodiments of the present invention, the probabilistic decision model may be a Bayesian model (Bayesian reasoning) or a Markov decision process, which is not limited in the present invention.

[0103] Among them, the probabilistic decision model can be adopted, that is, the operation data set and the environmental data set of the pole-mounted transformer can be fused with multi-source information by using Bayesian reasoning or Markov decision process, that is, the possible cause data of the appearance abnormality of the pole-mounted transformer can be inferred from the operation data set and the environmental data set of the pole-mounted transformer.

[0104] In some embodiments of the present invention, the above step 1031 can be implemented by following steps A1 to A3:

[0105] Step A1: Based on each pole-mounted transformer with an abnormal appearance, the probability decision model is used to infer the operating weight ratio of the operating data set of the pole-mounted transformer with the abnormal appearance at a historical moment, and the environmental weight ratio of the environmental data set of the pole-mounted transformer with the abnormal appearance at a historical moment.

[0106] In some embodiments of the present invention, for each pole-mounted transformer with an abnormal appearance, the probabilistic decision model can be used to infer the operational weight ratio of the operational dataset of the pole-mounted transformer with the abnormal appearance, as well as the environmental weight ratio of the environmental dataset of the pole-mounted transformer with the abnormal appearance. The operational weight ratio and environmental weight ratio for each pole-mounted transformer with an abnormal appearance can be completely different or partially the same, and this is not limited in any way by the present invention.

[0107] It should be noted that the operation weight ratio and the environment weight ratio can be described by numerical values, such as: 0.2, 0.5, etc.

[0108] For example, the operating weight ratio of the operating data set of the pole-mounted transformer 1 with an abnormal appearance is 0.7, and the environmental weight ratio of the environmental data set of the pole-mounted transformer 1 with an abnormal appearance is 0.3; the operating weight ratio of the operating data set of the pole-mounted transformer 2 with an abnormal appearance is 0.5, and the environmental weight ratio of the environmental data set of the pole-mounted transformer 2 with an abnormal appearance is 0.5; the operating weight ratio of the operating data set of the pole-mounted transformer 3 with an abnormal appearance is 0.1, and the environmental weight ratio of the environmental data set of the pole-mounted transformer 1 with an abnormal appearance is 0.9.

[0109] Step A2: multiply the operating data set of the pole-mounted transformer with the appearance abnormality at the historical moment according to the operating weight ratio to obtain an intermediate operating data set, and multiply the environmental data set of the pole-mounted transformer with the appearance abnormality at the historical moment according to the environmental weight ratio to obtain an intermediate environmental data set.

[0110] In some embodiments of the present invention, for each pole-mounted transformer with an abnormal appearance, the corresponding data set is multiplied by its corresponding weight ratio to obtain an intermediate operation data set and an intermediate environment data set of the pole-mounted transformer with an abnormal appearance.

[0111] Step A3: Fusing the intermediate operation dataset and the intermediate environment dataset to obtain a cause dataset of the pole-mounted transformer with abnormal appearance.

[0112] In some embodiments of the present invention, for each pole-mounted transformer with abnormal appearance, the corresponding intermediate operation data set and intermediate environment data set are fused to obtain the inducement data set of the pole-mounted transformer with abnormal appearance.

[0113] It should be noted that the inducement datasets corresponding to different pole-mounted transformers with abnormal appearances are all composed of their operating datasets and environmental datasets at historical moments, and their internal composition ratios, that is, the weight ratios involved, may be different.

[0114] In some embodiments of the present invention, a data fusion unit can be used to fuse the operating data set and environmental data set of the pole-mounted transformer at historical moments. Furthermore, the data fusion unit can assign different weights to different data by calculating the feature correlation of different data sources based on the weighted feature fusion method. For example, when the current fluctuation of the pole-mounted transformer is large within a certain time period and the image data set of the pole-mounted transformer indicates that there are discharge marks on the surface of the bushing of the pole-mounted transformer, the weight of the current fluctuation feature (operating data set) is increased to enhance its influence in subsequent diagnostic analysis. If a crack is detected in the image (an abnormal appearance exists), and there is no obvious abnormality in the operating data set of the pole-mounted transformer in the historical time period, the weight ratio of the operating data set is reduced, and the wind speed, temperature and other parameters in the environmental data set of the pole-mounted transformer are combined to assess whether there is cumulative damage caused by long-term external factors. If it is caused by external environmental data, the weight ratio of the environmental data set is increased accordingly.

[0115] In this way, through the probabilistic decision model, the possible cause data of the appearance abnormality of the pole-mounted transformer is inferred from the operation data set and environmental data set of the pole-mounted transformer, so as to obtain the inducement data set of the pole-mounted transformer with appearance abnormality, thereby providing more accurate data for the subsequent training of the convolutional neural network, and thus improving the diagnostic accuracy of the subsequent appearance abnormality diagnosis model.

[0116] Step 1032: Use the inducement dataset and the image dataset to train the convolutional neural network to obtain an intermediate abnormality diagnosis model.

[0117] In some embodiments of the present invention, the inducement dataset and the image dataset are directly used as training datasets to perform (iterative) training on the convolutional neural network to obtain an intermediate abnormality diagnosis model whose output results meet preset conditions.

[0118] In some embodiments of the present invention, Bayesian inference or a Markov decision process is first used to fuse the operational and environmental datasets using multi-source information. Specifically, the possible causal factors of the appearance anomaly are inferred from the operational and environmental datasets closest to the anomaly occurrence, thereby obtaining a causal dataset. This causal dataset and images corresponding to the abnormal appearance (image dataset) are then used to initially train a CNN to obtain the intermediate anomaly diagnosis model. Here, if the intermediate anomaly diagnosis model is used to detect external deformation of the pole-mounted transformer and abnormal wind speed data (environmental dataset), it can be inferred that the external deformation is likely caused by strong winds rather than an internal fault in the pole-mounted transformer, thereby avoiding misjudgments.

[0119] In some embodiments of the present invention, the convolutional neural network can be trained using the image dataset as the input training set and the cause dataset as the labels corresponding to the training set, thereby obtaining a trained model, namely the intermediate anomaly diagnosis model. In other words, the CNN is initially trained using multiple transformer images with visual anomalies and the operational and environmental meteorological data closest to the anomaly occurrence time, thereby obtaining a preliminary diagnostic model, namely the intermediate anomaly diagnosis model.

[0120] In some embodiments of the present invention, the above step 1032 may be implemented by following steps B1 to B3:

[0121] Step B1: input the image dataset into the convolutional neural network for abnormality diagnosis to obtain an initial diagnostic dataset.

[0122] Step B2: using a cosine similarity loss function to determine the initial similarity loss between the inducement dataset and the initial diagnosis dataset.

[0123] Step B3: Based on the initial similarity loss, adjust the network parameters of the convolutional neural network until the intermediate abnormality diagnosis model whose output meets the preset conditions is obtained.

[0124] In some embodiments of the present invention, the cosine similarity loss function is a loss function suitable for calculating similarity. The cosine similarity loss function measures the similarity between two vectors by calculating the cosine similarity between them.

[0125] In some embodiments of the present invention, a cosine similarity function may be used to determine the initial similarity loss between the vector corresponding to the inducement dataset and the vector corresponding to the initial diagnosis dataset.

[0126] In some embodiments of the present invention, the initial similarity loss can be used to adjust the network parameters of the convolutional neural network, such as: convolution layer parameters, convolution kernel size, fully connected layer parameters, channel input of input data, etc. Here, the initial similarity loss between the inducement dataset (label data) and the initial diagnosis dataset (prediction data) can be used to iteratively train the network parameters of the convolutional neural network until an intermediate abnormality diagnosis model is obtained whose output meets preset conditions. Here, the preset condition can be that the initial similarity loss is less than or equal to a set threshold, etc.

[0127] Step 1033: Use the pole-mounted transformer appearance anomaly knowledge graph and the image dataset to train the intermediate anomaly diagnosis model to obtain the appearance anomaly diagnosis model.

[0128] In some embodiments of the present invention, a knowledge graph of pole-mounted transformer appearance anomalies and an image dataset (image data of multiple pole-mounted transformers with appearance anomalies) are used to further train the obtained intermediate anomaly diagnosis model to obtain an appearance anomaly diagnosis model of the pole-mounted transformer.

[0129] In some embodiments of the present invention, the above step 1033 may be implemented by following steps C1 to C3:

[0130] Step C1: input the image dataset into the pole-mounted transformer appearance anomaly knowledge graph to perform an anomaly query to obtain an anomaly dataset.

[0131] In some embodiments of the present invention, an image dataset may be used to perform anomaly retrieval in a knowledge graph of appearance anomalies of a pole-mounted transformer, thereby obtaining an anomaly dataset that matches each image data in the image dataset.

[0132] In some embodiments of the present invention, the above step C1 may be implemented by the following process:

[0133] The first step is to determine the initial query range in the pole-mounted transformer appearance anomaly knowledge graph based on the pole-mounted transformer model carried in the image dataset.

[0134] In the second step, the abnormal data set is determined within the initial query range according to the type of the appearance abnormality carried in the image data set.

[0135] In some embodiments of the present invention, a multi-level matching mechanism can be used to query within the knowledge graph of pole-mounted transformer appearance anomalies. Specifically, the query process employs a multi-level matching mechanism. First, an initial query range is determined based on the pole-mounted transformer model number contained in the image dataset. Then, based on the type of appearance anomaly contained in the image dataset, namely, the abnormal features in the abnormal image (e.g., cracks, deformation, oil leakage, rust, etc.), the search targets are narrowed down, and the corresponding anomaly cause data is determined, thereby obtaining an anomaly dataset. For example, if the image dataset indicates cracks on the pole-mounted transformer casing, the subgraph related to the casing corresponding to the pole-mounted transformer model number can be retrieved from the knowledge graph of pole-mounted transformer appearance anomalies. Based on the specific characteristics and location of the cracks, the corresponding anomaly cause data can be determined, thereby obtaining an anomaly dataset.

[0136] It's important to note that to improve query efficiency within the knowledge graph for pole-mounted transformer appearance anomalies, a graph traversal algorithm can be used to quickly identify the anomaly cause data most relevant to the anomaly information contained in the image dataset. Furthermore, the knowledge graph for pole-mounted transformer appearance anomalies can also return a 3D model or installation diagram of the pole-mounted transformer for subsequent analysis.

[0137] Step C2: inputting the image data set into the intermediate abnormality diagnosis model to perform abnormality diagnosis to obtain an intermediate diagnosis data set.

[0138] Step C3: Using the cosine similarity loss function, determine the intermediate similarity loss between the abnormal data set and the intermediate diagnostic data set, and based on the intermediate similarity loss, adjust the network parameters of the volume intermediate abnormality diagnosis model until the output of the appearance abnormality diagnosis model that meets the preset conditions is obtained.

[0139] In some embodiments of the present invention, the description of the above steps C2 and C3 may refer to the above description of obtaining the intermediate abnormality diagnosis model. The implementation methods of the two are similar, so the description of steps B2 and B3 may be referred to and will not be repeated here.

[0140] In some embodiments of the present invention, the knowledge graph of the appearance anomaly of the pole transformer uses the stored knowledge base of the appearance anomaly of the pole transformer to search and match the image data set. Therefore, based on the method of similarity calculation, the similarity between the abnormal data set found from the knowledge graph of the appearance anomaly of the pole transformer and the diagnosis result output by the current intermediate abnormality diagnosis model, that is, the intermediate diagnosis data set, can be calculated. For example, the cosine similarity calculation method (cosine similarity loss function) is adopted, that is, the following formula (1) is used to compare the feature vector F of the intermediate diagnosis data set d The feature vector F of the abnormal dataset found from the knowledge graph of abnormal appearance of the pole transformer kThe similarity between them is used to determine the degree of matching, that is, the similarity score S:

[0141]

[0142] Wherein, S represents a similarity score, and its value is between [0, 1]. The closer S is to 1, the more similar the abnormal dataset and the intermediate diagnostic dataset are. Otherwise, the difference between the abnormal dataset and the intermediate diagnostic dataset is large.

[0143] In some embodiments of the present invention, when the similarity score S is higher than a preset threshold T S (e.g., 0.85), the diagnostic result (intermediate diagnostic data set) output by the intermediate abnormality diagnosis model is considered to be consistent with the abnormal data set found in the knowledge graph of the appearance abnormality of the pole-mounted transformer, and the training of the intermediate abnormality diagnosis model can be stopped; if the similarity score S is lower than or equal to T S , it is believed that there may be a deviation between the diagnostic results (intermediate diagnostic data set) output by the intermediate abnormality diagnosis model and the abnormal data set found in the knowledge graph of pole-mounted transformer appearance abnormalities, and the intermediate abnormality diagnosis model needs to be further optimized.

[0144] Here, when there may be a deviation between the diagnostic result (intermediate diagnostic data set) output by the intermediate abnormality diagnosis model and the abnormal data set found in the knowledge graph of the appearance abnormality of the pole-mounted transformer, the parameter optimization unit in the intermediate abnormality diagnosis model can be used to trigger a local correction of the intermediate abnormality diagnosis model. This optimization process calculates the diagnostic error and dynamically adjusts the feature extraction weight of the intermediate abnormality diagnosis model based on a feedback optimization mechanism. The error E between the diagnostic result (intermediate diagnostic data set) output by the intermediate abnormality diagnosis model and the abnormal data set found in the knowledge graph of the appearance abnormality of the pole-mounted transformer exceeds the threshold E t , then perform the following local optimization as shown in formula (2):

[0145]

[0146] Among them, ω′ i,j is the weight of the updated intermediate anomaly diagnosis model, ω i,j is the current weight, η' is the learning rate, is the gradient of the error to the weight, λ' is the adjustment factor, so that when the similarity S is equal to the threshold T S When the gap is large, the weight update amplitude is increased accordingly to speed up the model correction.

[0147] Here, after the optimization of the intermediate abnormality diagnosis model is completed, the appearance abnormality diagnosis model is obtained to re-execute the abnormality diagnosis, and the obtained abnormality diagnosis result is matched again with the result (inference result) found in the knowledge graph of the appearance abnormality of the pole transformer to form a cyclic correction mechanism until the abnormality diagnosis result is consistent with the reasoning result of the knowledge graph, or the set maximum number of optimization times N is reached. max The update is terminated when , in order to ensure the convergence and computational efficiency of the optimization process.

[0148] In some embodiments of the present invention, the convolutional neural network may also include an adaptive learning module for continuously optimizing its own parameters in the two training links mentioned above (the initial training link, i.e., using the inducement data set and the image data set to train the convolutional neural network to obtain an intermediate abnormality diagnosis model, and the re-training link, i.e., using the pole-mounted transformer appearance abnormality knowledge graph and the image data set to train the intermediate abnormality diagnosis model to obtain an appearance abnormality diagnosis model) to improve the diagnostic accuracy and reliability of the final appearance abnormality diagnosis model.

[0149] In some embodiments of the present invention, the adaptive learning module may include functions such as an online learning mechanism, an environmental adaptability enhancement mechanism, and a feedback optimization mechanism. Specific descriptions are as follows:

[0150] 1. Online Learning Mechanism: This online learning mechanism continuously optimizes the parameters of the CNN or intermediate anomaly diagnosis model based on new data received in real time. This mechanism utilizes a gradient self-regulation module to dynamically adjust the network's learning rate based on feedback from the loss function, enabling the CNN or intermediate anomaly diagnosis model to converge quickly with new data input. Furthermore, during the online learning process, incremental learning methods can be used to fine-tune only the new data without affecting existing parameter weights, thus avoiding the forgetting problem in the CNN or intermediate anomaly diagnosis model.

[0151] Correspondingly, in order to improve the adaptability of the online learning mechanism, the embodiment of the present invention proposes a dynamic gradient adaptive adjustment function to ensure that the update step size of the CNN or intermediate abnormality diagnosis model can be automatically adjusted at different learning stages, thereby improving training efficiency and reducing the risk of overfitting. The definition of the dynamic gradient adaptive adjustment function can be seen in formula (3):

[0152]

[0153] In the formula (3), η t is the learning rate of the current iteration, η0 is the initial learning rate, λ is the learning rate attenuation factor, N is the total number of current trainable parameters, is the loss function L for the i-th weight parameter ω iThe gradient of , α is the gradient influence coefficient, which is used to adjust the sensitivity of the learning rate to the gradient change.

[0154] Here, when the accumulated gradient during the training process of CNN or intermediate anomaly diagnosis model is large, it means that the adjustment amplitude of the parameters needs to be reduced to prevent gradient explosion or unstable update; conversely, when the gradient is small, the learning rate is kept high to ensure that CNN or intermediate anomaly diagnosis model can still effectively learn the feature information of the new data.

[0155] Therefore, based on formula (3), the online learning mechanism provided by the embodiment of the present invention can adapt to different data update situations more robustly, while maintaining the stability of the output of the CNN or intermediate abnormality diagnosis model, and enhancing the ability to quickly adapt to sudden abnormal patterns.

[0156] 2. Environmental Adaptability Enhancement Mechanism: This mechanism dynamically adjusts the feature weights of the CNN or intermediate anomaly diagnosis model by combining historical environmental and operational datasets of the pole-mounted transformer. This allows the resulting appearance anomaly diagnosis model to adapt to different climate and load conditions. This mechanism introduces environmental adjustment factors to adjust the parameters of the convolution kernels within the CNN or intermediate anomaly diagnosis model based on current meteorological conditions (such as temperature, humidity, and wind speed) and the load of the pole-mounted transformer. For example, in high-temperature and high-load environments, the CNN or intermediate anomaly diagnosis model increases the feature weights for heat dissipation components, increasing sensitivity to overheating faults, while in low-temperature environments, it decreases these weights to reduce the likelihood of false alarms. Furthermore, this mechanism leverages historical data (historical environmental and operational datasets of the pole-mounted transformer) to establish an environmental impact factor matrix. This weighted calculation modifies the fault judgment threshold of the CNN or intermediate anomaly diagnosis model, improving its adaptability under different operating conditions.

[0157] Correspondingly, in order to more effectively adjust the feature weights of CNN, an environment-adaptive weight adjustment formula is proposed based on the environmental adaptability enhancement mechanism. This formula can reconstruct the weights of the convolution kernels of the feature layer of CNN or the intermediate abnormality diagnosis model based on the dynamic changes of environmental parameters, so that the CNN or the intermediate abnormality diagnosis model can optimize the fault feature extraction capability under different environmental conditions. The environment-adaptive weight adjustment formula can be shown as follows (4):

[0158]

[0159] In formula (4), W′ i,j is the adjusted convolution kernel weight, W i,jis the original convolution kernel weight, γ is the environment adaptive adjustment coefficient, which controls the adjustment amplitude of the weight, tanh(·) is the hyperbolic tangent function, which makes the weight adjustment smooth and prevents the weight from changing too fast and causing unstable training, M is the total number of environmental variables, β k is the influencing factor of the kth environmental factor, which is used to measure the contribution of different environmental variables to feature extraction. k is the real-time value of the kth environment variable, is the normal operating reference value of the kth environment variable, and are the historical maximum and minimum values ​​of the variable, respectively, used for normalization.

[0160] Therefore, based on formula (4), the feature extraction weights of CNN or intermediate abnormality diagnosis model can be dynamically adjusted using the current environmental parameters, so that CNN or intermediate abnormality diagnosis model can automatically enhance or weaken the influence of certain features under different environmental conditions. For example, when both temperature and humidity are higher than the normal operating range, the adjusted W′ i,j This makes the CNN or intermediate anomaly diagnosis model more sensitive to the characteristics of the heat dissipation component area, while reducing the impact of this area in low temperature and high humidity conditions to reduce false positives. In this way, the present invention can effectively improve the adaptability of the subsequent appearance anomaly diagnosis model in complex environmental conditions and enhance the intelligent adjustment capabilities of the appearance anomaly diagnosis model.

[0161] 3. Feedback Optimization Mechanism: This feedback optimization mechanism is used to improve the accuracy of the anomaly diagnosis results of the appearance anomaly diagnosis model. The feedback optimization mechanism can trigger local parameter updates of the appearance anomaly diagnosis model and adjust the feature extraction weights of the CNN or intermediate anomaly diagnosis model involved in the training process. This process can employ an adaptive optimization method based on backpropagation (BP) to calculate the error between the output of the CNN or intermediate anomaly diagnosis model and labeled data, such as the error between the cause dataset and the initial diagnosis dataset, or the error between the inference results (anomaly dataset) of the pole-mounted transformer appearance anomaly knowledge graph and the intermediate diagnosis dataset. Based on the obtained error, the weights of specific layers of the CNN or intermediate anomaly diagnosis model are fine-tuned. For example, during the training of the intermediate anomaly diagnosis model, if the pole-mounted transformer appearance anomaly knowledge graph indicates that a certain type of appearance anomaly is more likely to occur in a specific component, but the initial judgment of the intermediate anomaly diagnosis model is inconsistent, the feedback optimization mechanism will adjust the weights of the relevant features of that component, so that the intermediate anomaly diagnosis model is more likely to align with the inference results of the pole-mounted transformer appearance anomaly knowledge graph during subsequent training, thereby improving the reliability of appearance anomaly diagnosis.

[0162] Here, in order to enhance the adaptability of the feedback optimization mechanism, the present invention proposes a local gradient dynamic adjustment formula, which enables the CNN or intermediate anomaly diagnosis model to adjust the parameter optimization rate according to the confidence level corresponding to the label data during the update process, thereby enhancing the convergence stability and improving the accuracy of appearance anomaly diagnosis. The local gradient dynamic adjustment formula is shown in the following formula (5):

[0163]

[0164] In formula (5), Δω i is the update amount of the i-th weight parameter, η is the basic learning rate, is the loss function L for parameter ω i The gradient of S kg is the confidence corresponding to the label data, S cnn is the confidence of the initial diagnosis of CNN or intermediate abnormality diagnosis model, ρ is the feedback adjustment coefficient, and ε is a small positive number (e.g., 10 -6 ), used to prevent the denominator from approaching zero.

[0165] Here, we take the training of the intermediate abnormality diagnosis model as an example to illustrate that the formula (3) introduces the confidence difference between the reasoning result of the knowledge graph of the appearance abnormality of the pole transformer and the output of the intermediate abnormality diagnosis model, and uses the hyperbolic tangent function for nonlinear adjustment. When the reasoning result of the intermediate abnormality diagnosis model is highly consistent with the knowledge graph of the appearance abnormality of the pole transformer (i.e., S kg ≈S cnn ), the parameter update amplitude is small to maintain stability; when the confidence difference between the two is large (ie: S kg With S cnn If significant deviations are found, local weight adjustments are accelerated to quickly correct the abnormality diagnosis capability of the intermediate abnormality diagnosis model. This feedback optimization mechanism effectively improves the adaptability of the appearance abnormality diagnosis model in complex environments, enabling it to correct misjudgments more quickly and improve overall diagnostic accuracy.

[0166] Through the adaptive learning module described above, the final appearance anomaly diagnosis model can maintain high recognition accuracy and stability under complex working conditions, making the appearance anomaly diagnosis of pole-mounted transformers more intelligent and efficient.

[0167] Based on the above description, as is common knowledge in this field, when diagnosing the appearance abnormality of a pole-mounted transformer, if the operating data and environmental information of the pole-mounted transformer cannot be fully integrated, problems such as missed detection or misjudgment of the appearance abnormality may occur. This will not only cause duplication and waste of resources in on-site maintenance of the pole-mounted transformer, but will also bring certain potential hidden dangers to the safe and stable operation of the power system. For example, if a single point failure or abnormal signal caused by the appearance abnormality of the pole-mounted transformer is not identified and processed in a timely manner, in extreme cases it may evolve into a large-scale local power outage or damage to the power supply equipment in the area where the pole-mounted transformer is located, thereby adversely affecting the overall reliability of the distribution network.

[0168] Based on this, the embodiment of the present invention achieves accurate diagnosis of appearance anomalies of pole-mounted transformers through an intelligent optimization model, namely, a CNN-based appearance anomaly diagnosis model. It also accurately integrates and analyzes multi-source data of the pole-mounted transformer in the diagnostic results, thereby dynamically perceiving the corresponding state changes when the pole-mounted transformer's appearance is abnormal, effectively improving the accuracy and real-time performance of diagnosing appearance anomalies or faults of the pole-mounted transformer. Furthermore, the CNN-based appearance anomaly diagnosis model can possess online learning and environmental adaptability, significantly enhancing the generalization performance of the appearance anomaly diagnosis model to address the problems of poor model generalization and difficulty in coping with complex environmental changes in traditional methods, thereby improving the overall diagnostic effect. Furthermore, the appearance anomaly diagnosis model provided by the embodiment of the present invention utilizes a knowledge graph of pole-mounted transformer appearance anomalies, namely, intelligent retrieval technology that utilizes the structural information of the pole-mounted transformer to more accurately locate the cause of the appearance anomaly, reducing maintenance costs and improving the safety and stability of distribution network operation.

[0169] In other words, the method for constructing a pole-mounted transformer appearance abnormality diagnosis model provided by an embodiment of the present invention includes: collecting operating data, environmental data and image data of pole-mounted transformers with appearance abnormalities at historical moments; and fusing the collected operating data and environmental data to form a cause data set, and with the help of a knowledge graph of pole-mounted transformer appearance abnormalities, that is, obtaining a cause data set matching the appearance abnormalities based on intelligent retrieval technology, and finally training a CNN with an online learning and adaptability enhancement mechanism with the help of the inducement data set, the cause data set and the image data of the pole-mounted transformer to obtain an intelligent optimization model, that is, an appearance abnormality model of the pole-mounted transformer; that is, this embodiment can make full use of the pole-mounted transformer operating data, environmental meteorological data and image data to realize comprehensive analysis and diagnosis of existing pole-mounted transformers with appearance abnormalities. Among them, the CNN can dynamically adjust the weights according to different working conditions and environmental factors based on its internal adaptive learning ability, which can further enhance the generalization and self-update performance of the appearance anomaly diagnosis model. That is, it can significantly improve the generalization ability and real-time performance of the appearance anomaly diagnosis model of pole-mounted transformers, thereby improving the accuracy and reliability of the appearance anomaly diagnosis model, and thus providing important technical support for the stable operation of the distribution system.

[0170] The present invention provides a method for constructing a pole-mounted transformer appearance anomaly diagnosis model. During the execution of this method, a convolutional neural network is trained using a knowledge graph of pole-mounted transformer appearance anomalies and multi-source data (image datasets, historical operation datasets, and environmental datasets) of multiple pole-mounted transformers with appearance anomalies and in operation, thereby obtaining an appearance anomaly diagnosis model with improved overall anomaly diagnosis effectiveness. In this way, by fully integrating the operational data and environmental information of the pole-mounted transformer, problems such as missed detection or misjudgment when appearance anomalies occur can be avoided as much as possible. Furthermore, combined with the knowledge graph of pole-mounted transformer appearance anomalies, the cause of the appearance anomaly can be more accurately located, thereby improving the generalization capability of the obtained appearance anomaly diagnosis model and enabling it to adapt to complex and changing environments. Furthermore, the appearance anomaly diagnosis model can be used to improve the accuracy and reliability of pole-mounted transformer appearance anomaly diagnosis.

[0171] Example 2:

[0172] Based on the same inventive concept, an embodiment of the present invention provides a method for diagnosing abnormal appearance of a pole-mounted transformer, such as Figure 2 As shown, the method includes the following:

[0173] Step 201: Capture an image of the pole-mounted transformer to be tested in operation to obtain an image to be tested;

[0174] Step 202: When abnormal features are identified in the image to be tested as existing in the appearance of the pole-mounted transformer to be tested, the image to be tested is input into the appearance abnormality diagnosis model obtained by any of the above-mentioned construction methods to perform abnormality diagnosis, thereby obtaining an abnormality diagnosis data set of the abnormal features.

[0175] In some embodiments of the present invention, an image of the appearance or surface of the pole-mounted transformer to be tested in operation may be captured to obtain an image to be tested; wherein the image to be tested may be video frame data or an infrared image, etc., and the present invention does not impose any limitation on this.

[0176] It should be noted that the image to be tested can be one or multiple images.

[0177] In some embodiments of the present invention, a deep neural network-based image recognition method can be used to identify the image to be tested and determine whether the pole-mounted transformer to be tested has any abnormal features. Abnormal features include, but are not limited to, the presence of oil stains, cracks, rust, surface component misalignment, surface deformation, surface component deformation, and surface component detachment.

[0178] In some embodiments of the present invention, the abnormal diagnostic data set of abnormal characteristics may include but is not limited to: operating data of the pole transformer to be tested in a historical time period (including but not limited to: abnormal voltage, abnormal current, abnormal load, etc.) and environmental data (high temperature, low temperature, high humidity, etc.).

[0179] It should be noted that the present invention can also generate an abnormality diagnosis report in a structured manner using the abnormality diagnosis dataset obtained with abnormal characteristics. This abnormality diagnosis report can include the abnormality characteristics, location, severity, possible causes, treatment suggestions, and historical case comparison results. The report can be displayed through a human-computer interaction interface or sent to an operation and maintenance terminal via remote communication for reference and decision-making by maintenance personnel.

[0180] In some embodiments of the present invention, the above step 202 can be implemented by the following steps 2021 and 2022 ( Figure 2 Not shown):

[0181] Step 2021: Use an abnormal target detection algorithm to perform abnormality recognition on the image to be tested to obtain features to be evaluated.

[0182] Step 2022: When the abnormality confidence score of the feature to be evaluated is greater than a preset threshold, it is determined that an abnormal feature exists in the appearance of the pole-mounted transformer to be tested identified in the image to be tested.

[0183] In some embodiments of the present invention, the abnormal target detection algorithm may be a cluster analysis method, a domain affinity method, a reconstruction error method, etc., and the present invention does not impose any limitation on this.

[0184] In some embodiments of the present invention, the abnormal target detection algorithm can adopt a target detection algorithm based on a deep neural network, such as: (You Only Look Once, YOLO), or a deep learning model such as Region-based Convolutional Neural Networks (R-CNN). Here, the image features of the image to be tested can be first extracted by CNN, and combined with the bounding box regression method, the key parts of the appearance of the pole-mounted transformer, such as: bushings, heat sinks, oil pillows, porcelain bottles, etc. When abnormal features are detected in the image to be tested (such as: oil leakage, cracks, rust, component detachment, etc.), the abnormal target detection algorithm will mark the abnormal area on the image to be tested and assign the corresponding abnormal category label. Here, for scenes that require more precise area division, a semantic segmentation algorithm is used to perform pixel-level segmentation on the abnormal area on the surface of the pole-mounted transformer to more accurately outline the scope of the abnormal area.

[0185] It should be noted that outlier detection algorithms are used to identify abnormal points in data. These algorithms are widely used in various fields, such as network security, finance, and biomedicine. The main goal of anomaly detection is to identify points that are significantly different from other data points. These points may indicate potential problems, vulnerabilities, or diseases.

[0186] In some embodiments of the present invention, the obtained feature to be evaluated may be scored for abnormality confidence first. Here, a model based on probability output, such as a supervised / semi-supervised classification model, or a deep learning model may be used to score the confidence of the feature to be evaluated to obtain an abnormality confidence score for the feature to be evaluated. Then, a preset threshold set by an additional abnormality confidence threshold setting mechanism is used to compare the obtained abnormality confidence score to further determine whether abnormal features are identified in the appearance of the transformer to be tested in the test image.

[0187] Here, only when the anomaly confidence score of the feature to be evaluated exceeds a preset threshold can the identification of an abnormal feature in the pole-mounted transformer's appearance in the image under test be determined, and subsequent anomaly diagnosis can be performed. This can reduce the false alarm rate of anomaly diagnosis. Furthermore, if the image under test is video frame data, time series analysis can be used to eliminate false positives due to transient illumination changes or environmental interference before performing subsequent anomaly confidence assessments, thereby improving the stability and accuracy of pole-mounted transformer appearance anomaly diagnosis.

[0188] In some embodiments of the present invention, the image to be tested can be preprocessed first, and then the preprocessed image can be input into the appearance abnormality diagnosis model obtained by the construction method provided in the above embodiment to achieve automatic recognition, analysis and diagnosis of the appearance abnormality of the pole-mounted transformer.

[0189] The pole-mounted transformer abnormality diagnosis method provided by an embodiment of the present invention, during the execution of which, when abnormal features of the pole-mounted transformer's appearance are identified in the image to be tested, uses an appearance abnormality diagnosis model trained using multi-source data and related knowledge graph technology to perform abnormality diagnosis on the image to be tested, thereby generating a highly accurate and reliable abnormality diagnosis dataset. This not only improves the accuracy and stability of pole-mounted transformer appearance abnormality diagnosis, thereby reducing misjudgments and missed detections, but also enhances the intelligent level of appearance abnormality diagnosis prediction and operation and maintenance management of pole-mounted transformers.

[0190] Continuing from the above description, the embodiment of the present invention further provides a system 300 for identifying and diagnosing abnormalities in the appearance of a pole-mounted transformer, such as Figure 3 As shown, a method for constructing a pole-mounted transformer appearance abnormality diagnosis model and a pole-mounted transformer appearance abnormality diagnosis method provided in some embodiments of the present invention are simultaneously implemented in a system, wherein the pole-mounted transformer appearance abnormality recognition and diagnosis system 300 includes:

[0191] The data acquisition and preprocessing module 301 is used to acquire image data of the pole-mounted transformer (the image data includes: test data and training data, wherein the test data is used in the appearance abnormality diagnosis link, and the training data is used in the training link of the appearance abnormality diagnosis model of the pole-mounted transformer), operation data and environmental data (training link of the appearance abnormality diagnosis model of the pole-mounted transformer), etc., and simultaneously perform preprocessing operations on the acquired image data, operation data and environmental data.

[0192] The image recognition and appearance anomaly detection module 302 is used to perform anomaly recognition on the image data of the pole-mounted transformer acquired by the data acquisition and preprocessing module 301 to determine whether the appearance of the pole-mounted transformer has any abnormal appearance phenomenon or abnormal features.

[0193] The multi-source data fusion module 303 is used for the training of the appearance abnormality diagnosis model of the pole-mounted transformer. Specifically, it is used to adopt a probabilistic decision model to fuse and infer the environmental data and operation data obtained by the data acquisition and preprocessing module 301 to obtain the inducement data of the pole-mounted transformer with abnormal appearance, etc., so that the training and optimization module 306 can train the CNN based on the inducement data and image data to obtain the intermediate abnormality diagnosis model of the pole-mounted transformer.

[0194] The pole-mounted transformer query module 304 is used for the training phase of the pole-mounted transformer appearance anomaly diagnosis model. Specifically, it is used to input image data into the pole-mounted transformer appearance anomaly knowledge graph to perform anomaly query and obtain anomaly data, etc., so that the intermediate anomaly diagnosis model can be trained with the help of the anomaly data and image data to obtain the final pole-mounted transformer appearance anomaly diagnosis model.

[0195] The diagnostic module 305 includes a pole-mounted transformer appearance abnormality diagnostic model, which is used to identify abnormal features in the appearance of the pole-mounted transformer to be tested in the test data obtained by the data acquisition and preprocessing module 301. The training and optimization module 306 is used to train and optimize the pole-mounted transformer appearance abnormality diagnostic model to perform abnormal diagnosis on the test image of the pole-mounted transformer to be tested, thereby obtaining an abnormality diagnosis data set of the abnormal features.

[0196] Example 3

[0197] Based on the same inventive concept, the embodiment of the present invention also provides a system for constructing an appearance abnormality diagnosis model of a pole-mounted transformer, see Figure 4 As shown, the construction system 400 of the appearance abnormality diagnosis model of the pole-mounted transformer includes:

[0198] The first image acquisition module 401 is used to acquire images of a plurality of pole-mounted transformers that have abnormal appearances and are in operation, to obtain an image data set;

[0199] An acquisition module 402 is configured to acquire an operating data set and an environmental data set of the plurality of pole-mounted transformers at historical moments; wherein the historical moments are time series before an identification moment, and the identification moment is a moment corresponding to when an abnormal appearance of the plurality of pole-mounted transformers is identified;

[0200] The training module 403 is used to train a convolutional neural network using the image dataset, the operation dataset, the environmental dataset, and the knowledge graph of pole-mounted transformer appearance anomalies to obtain a pole-mounted transformer appearance anomaly diagnosis model.

[0201] Optionally, the training module 403 includes:

[0202] an inference unit, configured to use a probabilistic decision model to perform inducement data inference on the plurality of pole-mounted transformers with abnormal appearances based on the operation data set and the environmental data set, to obtain inducement data sets of the plurality of pole-mounted transformers with abnormal appearances;

[0203] A first training subunit is configured to train the convolutional neural network using the inducement dataset and the image dataset to obtain an intermediate abnormality diagnosis model;

[0204] The second training unit is used to train the intermediate abnormality diagnosis model using the pole-mounted transformer appearance abnormality knowledge graph and the image dataset to obtain the appearance abnormality diagnosis model.

[0205] Optionally, the inference unit is specifically used to infer, based on each pole-mounted transformer with an abnormal appearance, the operating weight ratio of the operating data set of the pole-mounted transformer with an abnormal appearance at a historical moment, and the environmental weight ratio of the environmental data set of the pole-mounted transformer with an abnormal appearance at a historical moment, using the probabilistic decision model; multiplying the operating weight ratio by the operating data set of the pole-mounted transformer with an abnormal appearance at a historical moment to obtain an intermediate operating data set, and multiplying the environmental weight ratio by the environmental data set of the pole-mounted transformer with an abnormal appearance at a historical moment to obtain an intermediate environmental data set; and fusing the intermediate operating data set and the intermediate environmental data set to obtain the inducement data set of the pole-mounted transformer with an abnormal appearance.

[0206] Optionally, the first training unit is specifically used to input the image dataset into the convolutional neural network for abnormality diagnosis to obtain an initial diagnostic dataset; use a cosine similarity loss function to determine the initial similarity loss between the cause dataset and the initial diagnostic dataset; and adjust the network parameters of the convolutional neural network based on the initial similarity loss until the intermediate abnormality diagnosis model whose output meets preset conditions is obtained.

[0207] Optionally, the second training unit is specifically used to input the image dataset into the pole-mounted transformer appearance anomaly knowledge graph for anomaly query to obtain an anomaly dataset; input the image dataset into the intermediate anomaly diagnosis model for anomaly diagnosis to obtain an intermediate diagnostic dataset; use the cosine similarity loss function to determine the intermediate similarity loss between the anomaly dataset and the intermediate diagnostic dataset, and adjust the network parameters of the intermediate anomaly diagnosis model based on the intermediate similarity loss until the appearance anomaly diagnosis model whose output meets preset conditions is obtained.

[0208] Optionally, the second training unit is specifically used to determine an initial query range in the pole-mounted transformer appearance anomaly knowledge graph based on the pole-mounted transformer model carried in the image dataset; and determine the anomaly dataset within the initial query range based on the type of appearance anomaly carried in the image dataset.

[0209] It should be noted that the description of the system for constructing a pole-mounted transformer appearance anomaly diagnosis model is similar to the description of the aforementioned method embodiment for constructing a pole-mounted transformer appearance anomaly diagnosis model, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the system embodiment of the present invention, please refer to the description of the method embodiment of the present invention for an understanding.

[0210] Example 4

[0211] Based on the same inventive concept, the embodiment of the present invention also provides a system for diagnosing abnormal appearance of a pole-mounted transformer, see Figure 5 As shown, the appearance abnormality diagnosis system 500 of the pole-mounted transformer includes:

[0212] The second image acquisition module 501 is used to acquire an image of the pole-mounted transformer to be tested in operation to obtain an image to be tested;

[0213] The abnormality diagnosis module 502 is used to input the image to be tested into the appearance abnormality diagnosis model obtained by the above-mentioned construction system to perform abnormality diagnosis when abnormal features are identified in the image to be tested, thereby obtaining an abnormality diagnosis data set of the abnormal features.

[0214] Optionally, the abnormality diagnosis module 502 further includes:

[0215] an identification unit, configured to use an abnormal target detection algorithm to perform abnormality identification on the image to be tested and obtain features to be evaluated;

[0216] When the abnormality confidence score of the feature to be evaluated is greater than a preset threshold, it is determined that an abnormal feature exists in the appearance of the transformer on the pole to be tested identified in the image to be tested.

[0217] It should be noted that the description of the pole-mounted transformer appearance abnormality diagnosis system is similar to the description of the pole-mounted transformer appearance abnormality diagnosis method embodiment described above, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the system embodiment of the present invention, please refer to the description of the method embodiment of the present invention for understanding.

[0218] Example 5:

[0219] Based on the same inventive concept, Figure 6As shown, the present invention also provides an electronic device, which may be a computer device, a single-chip microcomputer device, a smart mobile device, etc. The electronic device in this embodiment may include a processor 610, a memory 620, a transceiver component 630, etc. The processor 610, the memory 620, and the transceiver component 630 are connected via a bus 640. The memory 620 may be used to store an execution program, which may include instructions. The processor 610 is used to execute the instructions stored in the memory. The memory 620 may also be used to store data, which may be accessed and / or modified during the execution of the instructions.

[0220] The processor may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the storage medium to implement the corresponding method flow or corresponding function, so as to realize the method for constructing the appearance abnormality diagnosis model of the pole-mounted transformer in the above embodiment, and the appearance abnormality diagnosis method of the pole-mounted transformer in the above embodiment.

[0221] Example 6:

[0222] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory), which is a memory device in the electronic device for storing programs and data. It is understandable that the storage medium here can include both built-in storage media in the electronic device and, of course, extended storage media supported by the electronic device. The storage medium provides a storage space that stores the terminal's operating system. In addition, the storage space also stores one or more instructions suitable for being loaded and executed by the processor. These instructions can be one or more execution programs (including program code). It should be noted that the storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage. The processor loads and executes one or more instructions stored in the storage medium to implement the method for constructing the appearance abnormality diagnosis model of the pole-mounted transformer in the above-mentioned embodiment, and the appearance abnormality diagnosis method of the pole-mounted transformer in the above-mentioned embodiment.

[0223] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0224] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0225] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0226] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0227] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A method for constructing an appearance abnormality diagnosis model for a pole-mounted transformer, characterized in that: The method comprises: Capture images of multiple pole-mounted transformers that have abnormal appearances and are in operation to obtain an image dataset. Obtaining operation data sets and environmental data sets of the plurality of pole-mounted transformers at historical moments; wherein the historical moments are time series before the identification moment, and the identification moment is the moment corresponding to when the plurality of pole-mounted transformers are identified to have appearance abnormalities; The image dataset, the operation dataset, the environment dataset and the knowledge graph of pole-mounted transformer appearance anomaly are used to train a convolutional neural network to obtain a pole-mounted transformer appearance anomaly diagnosis model.

2. The method according to claim 1, characterized in that The image dataset, the operation dataset, the environment dataset, and the knowledge graph of pole-mounted transformer appearance anomalies are used to train a convolutional neural network to obtain a pole-mounted transformer appearance anomaly diagnosis model, including: Using a probabilistic decision model, based on the operation data set and the environmental data set, inducement data of the plurality of pole-mounted transformers with abnormal appearances are inferred to obtain inducement data sets of the plurality of pole-mounted transformers with abnormal appearances; Using the inducement dataset and the image dataset, the convolutional neural network is trained to obtain an intermediate abnormality diagnosis model; The pole-mounted transformer appearance anomaly knowledge graph and the image dataset are used to train the intermediate anomaly diagnosis model to obtain the appearance anomaly diagnosis model.

3. The method according to claim 2, characterized in that The method of using a probabilistic decision model to infer inducement data of the plurality of pole-mounted transformers with abnormal appearances based on the operation data set and the environmental data set to obtain inducement data sets of the plurality of pole-mounted transformers with abnormal appearances includes: Based on each pole-mounted transformer with an abnormal appearance, the probabilistic decision model is used to infer the operating weight ratio of the operating data set of the pole-mounted transformer with the abnormal appearance at a historical moment, and the environmental weight ratio of the environmental data set of the pole-mounted transformer with the abnormal appearance at a historical moment; Multiplying the operation weight ratio by the operation data set of the pole-mounted transformer with the appearance abnormality at the historical moment to obtain an intermediate operation data set, and multiplying the environment weight ratio by the environment data set of the pole-mounted transformer with the appearance abnormality at the historical moment to obtain an intermediate environment data set; The intermediate operation data set and the intermediate environment data set are fused to obtain the inducement data set of the pole-mounted transformer with abnormal appearance.

4. The method according to claim 2, characterized in that The method of using the inducement dataset and the image dataset to train the convolutional neural network to obtain an intermediate abnormality diagnosis model includes: Inputting the image dataset into the convolutional neural network for abnormality diagnosis to obtain an initial diagnostic dataset; Using a cosine similarity loss function, determining an initial similarity loss between the inducement dataset and the initial diagnosis dataset; Based on the initial similarity loss, the network parameters of the convolutional neural network are adjusted until the intermediate abnormality diagnosis model whose output meets preset conditions is obtained.

5. The method according to claim 2, characterized in that The method of using the pole-mounted transformer appearance anomaly knowledge graph and the image dataset to train the intermediate anomaly diagnosis model to obtain the appearance anomaly diagnosis model includes: Inputting the image dataset into the pole-mounted transformer appearance anomaly knowledge graph to perform anomaly query to obtain an anomaly dataset; Inputting the image data set into the intermediate abnormality diagnosis model to perform abnormality diagnosis to obtain an intermediate diagnosis data set; A cosine similarity loss function is used to determine the intermediate similarity loss between the abnormal data set and the intermediate diagnostic data set, and based on the intermediate similarity loss, the network parameters of the intermediate abnormality diagnosis model are adjusted until the appearance abnormality diagnosis model whose output meets the preset conditions is obtained.

6. The method according to claim 5, characterized in that The image dataset is input into the pole-mounted transformer appearance anomaly knowledge graph to perform an anomaly query, and an anomaly dataset is obtained, including: Determining an initial query range in the pole-mounted transformer appearance anomaly knowledge graph based on the pole-mounted transformer model carried in the image dataset; The abnormal data set is determined within the initial query range according to the type of the appearance abnormality carried in the image data set.

7. A system for constructing an appearance abnormality diagnosis model for a pole-mounted transformer, characterized in that: The system comprises: The first image acquisition module is used to acquire images of a plurality of pole-mounted transformers that have abnormal appearances and are in operation, to obtain an image data set; An acquisition module is configured to acquire an operation data set and an environmental data set of the plurality of pole-mounted transformers at historical moments; wherein the historical moments are time series before an identification moment, and the identification moment is a moment corresponding to when an abnormal appearance of the plurality of pole-mounted transformers is identified; The training module is used to train a convolutional neural network using the image dataset, the operation dataset, the environmental dataset, and the knowledge graph of the appearance anomaly of the pole-mounted transformer to obtain an appearance anomaly diagnosis model for the pole-mounted transformer.

8. A method for diagnosing abnormal appearance of a pole-mounted transformer, characterized in that: The method comprises: Capturing images of the pole-mounted transformer to be tested in operation to obtain images to be tested; When abnormal features are identified in the image to be tested as existing in the appearance of the transformer on the pole to be tested, the image to be tested is input into the appearance abnormality diagnosis model obtained by any construction method described in weights 1 to 6 for abnormality diagnosis, and an abnormality diagnosis data set of the abnormal features is obtained.

9. The method according to claim 8, characterized in that The image to be tested identifies abnormal features in the appearance of the pole-mounted transformer to be tested, including: Using an abnormal target detection algorithm to identify abnormalities in the image to be tested, and obtain features to be evaluated; When the abnormality confidence score of the feature to be evaluated is greater than a preset threshold, it is determined that an abnormal feature exists in the appearance of the transformer on the pole to be tested identified in the image to be tested.

10. A system for diagnosing abnormal appearance of a pole-mounted transformer, characterized in that: The system comprises: The second image acquisition module is used to acquire an image of the pole-mounted transformer to be tested in operation to obtain an image to be tested; The abnormality diagnosis module is used to input the image to be tested into the appearance abnormality diagnosis model obtained by the construction system described in right 7 to perform abnormal diagnosis when abnormal features are identified in the image to be tested, so as to obtain an abnormality diagnosis data set of the abnormal features.

Images