Substation equipment image identification and fault detection method, medium and system

Abstract

The invention discloses a substation equipment image recognition and fault detection method, a medium and a substation equipment image recognition and fault detection system. The method comprises the following steps: acquiring a real-time image of substation equipment and preprocessing the real-time image; an improved YOLOv8 algorithm model is adopted to extract image features in the preprocessed image; specifically, a deformation convolution module is introduced into a YOLOv8 backbone network, and a bidirectional feature pyramid network is adopted in a YOLOv8 neck network to form an improved YOLOv8 algorithm model; and carrying out classification and fault detection on the extracted image features by using a convolutional neural network. According to the invention, the accuracy and efficiency of fault detection are improved.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of power equipment detection, and specifically to a method, medium and system for image recognition and fault detection of substation equipment. Background Art

[0002] Substations are an important part of the power system, responsible for power transmission, transformation and distribution. The reliability and stability of their operation are crucial to the safe operation of the power system. Substation equipment includes transformers, circuit breakers, current transformers, voltage transformers, lightning arresters, busbars and their connecting conductors, etc. In the long-term operation of these devices, due to factors such as environmental conditions, load changes and equipment aging, various faults such as overheating, leakage, insulation aging, etc. are prone to occur. If these faults are not discovered and handled in time, the normal operation of the power system will be seriously affected, and even a large-scale power failure will be caused.

[0003] Traditional substation equipment monitoring and maintenance mainly rely on manual inspections, which have many problems. On the one hand, the frequency and coverage of manual inspections are limited, making it difficult to detect and handle equipment failures in a timely manner; on the other hand, the results of manual inspections are easily affected by subjective factors, and the accuracy and consistency of detection are difficult to guarantee. In addition, manual inspections are inefficient, consume a lot of manpower and time costs, and are difficult to meet the requirements of modern power systems for efficient and accurate monitoring of substation equipment.

[0004] With the rapid development of computer vision and deep learning technologies, the application of automated monitoring methods based on image recognition in substation equipment monitoring is gradually increasing. Computer vision technology can collect images of substation equipment through camera equipment, and use image processing and analysis technology to monitor and diagnose the equipment status, thereby realizing automated monitoring of equipment. Deep learning technology, especially convolutional neural network (CNN), performs well in the field of image recognition, can automatically learn feature information in images, perform target detection and classification, and has broad application prospects in substation equipment fault detection.

[0005] YOLO (You Only Look Once) is an advanced target detection algorithm that transforms the target detection problem into a regression problem and predicts the position and category of the target through a single forward pass. The YOLO algorithm has the advantages of fast detection speed and high accuracy, and is suitable for real-time target detection tasks. YOLOv5 is an important version of the YOLO series algorithm. It optimizes the model structure and training methods to improve the detection performance. However, in the specific application of substation equipment, YOLOv5 still has some shortcomings when detecting small targets and complex scenes, such as insufficient detection accuracy and insufficient recognition ability of small targets.

[0006] In order to further improve the accuracy and efficiency of substation equipment fault detection, researchers have made many improvements to YOLOv5 and proposed the YOLOv8 algorithm. YOLOv8 introduces more advanced feature extraction and fusion technologies based on YOLOv5, such as deformable convolution and bidirectional feature pyramid network (BiFPN), which enhances the model's detection ability for targets of different scales. However, in the actual application of substation equipment monitoring, YOLOv8 still needs to be further optimized and adjusted according to specific needs to adapt to complex and changing application scenarios. Summary of the invention

[0007] In view of the technical problems existing in the prior art, the present invention provides a method, medium and system for substation equipment image recognition and fault detection which improve the accuracy and efficiency of fault detection.

[0008] In order to solve the above technical problems, the technical solution proposed by the present invention is:

[0009] A method for image recognition and fault detection of substation equipment, comprising the steps of:

[0010] Acquire real-time images of substation equipment and perform preprocessing;

[0011] An improved YOLOv8 algorithm model is used to extract image features from preprocessed images; specifically, a deformable convolution module is introduced into the YOLOv8 backbone network, and a bidirectional feature pyramid network is used in the YOLOv8 neck network to form an improved YOLOv8 algorithm model;

[0012] Convolutional neural networks are used to classify and detect faults based on the extracted image features.

[0013] Preferably, the formula of the deformation convolution module is:

[0014]

[0015] Among them, Δp n is the offset, p0 is the center position of the current convolution operation, and p n is the offset of the corresponding position in the standard convolution window, R is the receptive field of the convolution kernel, w is the convolution kernel weight, x is the input feature map, and y is the output feature map; by learning the offset Δp n , the convolution kernel can be adaptively adjusted according to the shape of the target.

[0016] Preferably, the feature fusion formula of the bidirectional feature pyramid network is:

[0017]

[0018] Where P i outis the output feature map, P j is the input feature map, w j and w k is the weighting coefficient, and ε is a small value to prevent the denominator from being zero.

[0019] Preferably, in the process of image feature extraction, an improved anchor frame strategy is used to optimize target detection. The specific method is to adjust the size and proportion of the anchor frame to adapt to the diversity of substation equipment, and calculate the anchor frame dimension D by the following formula:

[0020]

[0021] where h i and w i are the height and width of the i-th true label respectively, and N is the total number of labels.

[0022] Preferably, the image preprocessing includes grayscale conversion, Gaussian filtering denoising and image enhancement.

[0023] Preferably, the fault detection process is:

[0024] Use multi-layer perceptrons for high-level abstraction and classification of features;

[0025] The failure probability is calculated using the Sigmoid activation function.

[0026] Preferably, during the fault detection process, whether a fault occurs is determined by a dynamic threshold, where the dynamic threshold T is:

[0027] T=μ+3×δ

[0028] where μ and δ are the mean and standard deviation of the detection probability, respectively.

[0029] Preferably, during the fault detection process, the weights of the convolutional neural network are adjusted according to the cross entropy loss function, and the formula is:

[0030]

[0031] Where M is the number of categories, y o,c is the true label of the cth class of the oth sample, p o,c is the predicted probability.

[0032] The present invention also discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described above are executed.

[0033] The present invention further discloses a substation equipment image recognition and fault detection system, comprising a memory and a processor connected to each other, wherein a computer program is stored in the memory, and when the computer program is run by the processor, the steps of the method described above are executed.

[0034] Compared with the prior art, the advantages of the present invention are:

[0035] The present invention is based on the substation equipment image recognition and fault detection method of improved YOLOv8 and convolutional neural network. The real-time image of the substation equipment is obtained and the image is preprocessed. The improved YOLOv8 algorithm is used to extract the equipment image features, and the bidirectional feature pyramid network (BiFPN) structure is combined to improve the detection ability of small targets and multi-scale targets. Finally, the convolutional neural network is used to detect faults on the features and generate fault reports. The method optimizes the feature extraction and fusion process by introducing deformable convolution and bidirectional feature pyramid network (BiFPN), thereby improving the accuracy and efficiency of fault detection, and is suitable for the high-demand environment of power grid automation and intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 The figure is a flow chart of an embodiment of the substation equipment image recognition and fault detection method of the present invention.

[0037] Figure 2 The improved YOLOv8 algorithm model structure diagram in the present invention. DETAILED DESCRIPTION

[0038] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0039] like Figure 1 As shown, the substation equipment image recognition and fault detection method based on improved YOLOv8 and convolutional neural network in the embodiment of the present invention includes the following steps:

[0040] S1. Image acquisition: Use high-definition cameras to obtain real-time images of substation equipment;

[0041] S2, image preprocessing: grayscale, denoising, enhancement and other preprocessing operations are performed on the collected images;

[0042] S3, feature extraction: use the improved YOLOv8 algorithm to extract image features;

[0043] The YOLOv8 algorithm combines the deformable convolution and bidirectional feature pyramid network (BiFPN) structures, where the deformable convolution module is used to enhance the model's adaptability to target morphology and scale; the bidirectional feature pyramid network is used to perform weighted fusion of multi-scale features to improve detection accuracy;

[0044] S4, Fault Detection: Use convolutional neural network to classify and detect faults on extracted image features; use multi-layer perceptron (MLP) to perform high-level abstraction and classification on image features; apply sigmoid activation function to calculate fault probability;

[0045] S5. Result output: Display the test results through a visual interface and generate a test report.

[0046] Specifically, in the feature extraction step of step S3, an improved anchor frame strategy is used to optimize target detection. The specific method is to adjust the size and proportion of the anchor frame to adapt to the diversity of substation equipment, and calculate the anchor frame dimension D by the following formula:

[0047]

[0048] where h i and w i are the height and width of the i-th true label respectively, and N is the total number of labels.

[0049] The present invention is based on the substation equipment image recognition and fault detection method of improved YOLOv8 and convolutional neural network. The real-time image of the substation equipment is obtained and the image is preprocessed. The improved YOLOv8 algorithm is used to extract the equipment image features, and the bidirectional feature pyramid network (BiFPN) structure is combined to improve the detection ability of small targets and multi-scale targets. Finally, the convolutional neural network is used to detect faults on the features and generate fault reports. The method optimizes the feature extraction and fusion process by introducing deformable convolution and bidirectional feature pyramid network (BiFPN), thereby improving the accuracy and efficiency of fault detection, and is suitable for the high-demand environment of power grid automation and intelligence.

[0050] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0051] The substation equipment image recognition and fault detection method based on improved YOLOv8 and convolutional neural network of the present invention comprises the following specific steps:

[0052] S1. System configuration and image acquisition

[0053] System configuration: Install high-definition cameras at key locations in the substation to ensure coverage of all major equipment. The cameras should have night vision and high dynamic range to ensure clear images under all lighting conditions.

[0054] Configure edge computing devices and connect cameras through the network to realize the collection and processing of real-time image data. Edge computing devices should have high computing performance and low latency to ensure real-time performance.

[0055] Image acquisition: Through the timing control system, the image data of the substation equipment is automatically collected every 5 minutes. In special circumstances (such as detecting an abnormal situation), image acquisition can be triggered immediately.

[0056] All collected image data are transmitted to the central control system via the network for storage and analysis.

[0057] S2. Image preprocessing:

[0058] Grayscale: Convert a color image to a grayscale image to simplify subsequent processing steps.

[0059] Gaussian filter denoising: Use a Gaussian filter to denoise the image to reduce noise interference in the image; the core size of the filter is adjusted according to the following formula:

[0060]

[0061] Where K is the adjustment factor of the filter kernel size and σ is the estimated standard deviation of the image noise.

[0062] Image Enhancement: Applies contrast and brightness adjustment techniques to enhance the visual quality of images and make key features of devices stand out more.

[0063] S3. Feature extraction:

[0064] The improved YOLOv8 algorithm is used to extract features from the preprocessed image. The improved YOLOv8 algorithm combines deformable convolution and bidirectional feature pyramid network (BiFPN) structure to improve the detection ability of objects of different scales. The specific improvements of the improved YOLOv8 algorithm model can be found in Figure 2 :

[0065] (1) Traditional convolutional neural networks perform well in processing fixed-shape features, but they are unable to cope with the problem of different shapes of substation equipment. Therefore, the deformable convolution module is introduced into the backbone network of YOLOv8. Deformable convolution introduces a learnable offset so that the convolution kernel can flexibly adapt to the shape changes and scales of different targets, thereby achieving adaptive learning of the target shape and scale. The formula for deformable convolution is:

[0066]

[0067] Among them, Δp n is the offset, p0 is the center position of the current convolution operation, and p nis the offset of the corresponding position in the standard convolution window, R is the receptive field of the convolution kernel, w is the convolution kernel weight, x is the input feature map, and y is the output feature map. n , the convolution kernel can be adaptively adjusted according to the shape of the target.

[0068] (2) The traditional feature pyramid network has the problem of insufficient information fusion when dealing with multi-scale targets. BiFPN is used to replace the traditional feature pyramid network (FPN), and a bidirectional feature pyramid network (BiFPN) is used in the neck network to perform weighted fusion of multi-scale features, which improves the fusion effect of multi-scale features and enhances the model's detection ability for small targets and multi-scale targets. The feature fusion formula of BiFPN is:

[0069]

[0070] Where P i out is the output feature map; P j is the input feature map; w j and w k is a weighting coefficient used to allocate weights to feature maps at different levels; ε is a small value that prevents the denominator from being zero.

[0071] The characteristics of BiFPN include:

[0072] 1. A weighting mechanism is introduced during feature fusion so that features of different scales can be adaptively fused according to their importance.

[0073] 2. Through two-way transmission, high-level semantic features and low-level detail features can complement each other and improve detection accuracy.

[0074] (3) In addition, in substation equipment, small targets are difficult to detect and are easily missed by traditional detection algorithms. To solve this problem, an additional small target detection layer is added to improve the detection capability of small targets by adding more fine-grained feature maps in the network.

[0075] The design of the small target detection layer includes:

[0076] 1. Introduce high-resolution feature maps to ensure that the features of small objects are not lost in the downsampling process.

[0077] 2. Add a dedicated detection branch for small targets in the detection head to ensure that the features of small targets can be fully utilized.

[0078] (4) Combine real-time image processing technology to optimize the algorithm's running speed and response time:

[0079] Real-time image processing technology is crucial in substation equipment monitoring. By optimizing the structure and parameter settings of YOLOv8, the algorithm's running speed and response time are improved.

[0080] Specific optimization measures include:

[0081] 1. Reduce the number of network layers and parameters, and reduce computational complexity through pruning and quantization techniques.

[0082] 2. Use efficient activation functions and normalization layers, such as Swish activation function and Batch Normalization, to improve the convergence speed and stability of the network.

[0083] 3. Through edge computing devices, data processing is performed close to the data source to reduce data transmission delays.

[0084] S4. Fault detection:

[0085] The extracted image features are analyzed using convolutional neural networks to identify possible equipment failures.

[0086] The specific steps are as follows:

[0087] (1) Multilayer Perceptron (MLP): Use MLP to perform high-level abstraction and classification of features;

[0088] (2) Sigmoid activation function: Calculate the failure probability. The formula is:

[0089]

[0090] (3) Use the loss function optimization algorithm in the deep learning model, which adjusts the weights according to the cross entropy loss function. The formula is:

[0091]

[0092] Where M is the number of categories, y o,c is the true label of the cth class of the oth sample, p o,c is the predicted probability.

[0093] (4) Dynamic threshold processing technology is used to determine whether a fault has occurred, where the threshold T is dynamically calculated using the following formula:

[0094] T=μ+3×δ

[0095] where μ and δ are the mean and standard deviation of the detection probability, respectively.

[0096] S5. Result output:

[0097] The detection results are displayed through a visual interface, including information such as equipment status, fault type and fault location.

[0098] The specific steps are as follows:

[0099] (1) Graphical interface: Displays the detected device image and marks the fault location and type on the image.

[0100] (2) Test report: Generate a detailed test report including equipment name, fault type, test time, fault location and other information.

[0101] Notifications and Archives:

[0102] (1) Notify the operation and maintenance personnel via email or text message and provide the download link of the test report and troubleshooting suggestions.

[0103] (2) The test report shall be archived for subsequent analysis and reference.

[0104] The present invention also discloses a computer-readable storage medium, on which a computer program is stored, and the computer program executes the steps of the above method when it is executed by a processor. The present invention further discloses a substation equipment image recognition and fault detection system, including a memory and a processor connected to each other, and the memory stores a computer program, and the computer program executes the steps of the above method when it is executed by the processor. The medium and system of the present invention correspond to the above method and also have the advantages described in the above method.

[0105] The present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by hardware related to computer program instructions. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned method embodiment when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. Computer-readable storage media include: any entity or device that can carry computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. The memory is used to store computer programs and / or modules, and the processor implements various functions by running or executing computer programs and / or modules stored in the memory, and calling data stored in the memory. The memory may include high-speed random access memory and may also include non-volatile memory, such as a hard disk, an internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0106] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A method for image recognition and fault detection of substation equipment, characterized in that: Includes steps: Acquire real-time images of substation equipment and perform preprocessing; An improved YOLOv8 algorithm model is used to extract image features from preprocessed images; specifically, a deformable convolution module is introduced into the YOLOv8 backbone network, and a bidirectional feature pyramid network is used in the YOLOv8 neck network to form an improved YOLOv8 algorithm model; Convolutional neural networks are used to classify and detect faults based on the extracted image features.

2. The substation equipment image recognition and fault detection method according to claim 1, characterized in that: The formula of the deformable convolution module is: Among them, Δp n is the offset, p0 is the center position of the current convolution operation, and p n is the offset of the corresponding position in the standard convolution window, R is the receptive field of the convolution kernel, w is the convolution kernel weight, x is the input feature map, and y is the output feature map; by learning the offset Δp n , the convolution kernel can be adaptively adjusted according to the shape of the target.

3. The substation equipment image recognition and fault detection method according to claim 1, characterized in that: The feature fusion formula of the bidirectional feature pyramid network is: Where P i out is the output feature map, P j is the input feature map, w j and w k is the weighting coefficient, and ε is a small value to prevent the denominator from being zero.

4. The substation equipment image recognition and fault detection method according to claim 1, 2 or 3, characterized in that: In the process of image feature extraction, an improved anchor frame strategy is used to optimize target detection. The specific method is to adjust the size and proportion of the anchor frame to adapt to the diversity of substation equipment. The anchor frame dimension D is calculated by the following formula: where h i and w i are the height and width of the i-th true label respectively, and N is the total number of labels.

5. The substation equipment image recognition and fault detection method according to claim 1, 2 or 3, characterized in that: Image preprocessing includes grayscale, Gaussian filtering denoising and image enhancement.

6. The substation equipment image recognition and fault detection method according to claim 1, 2 or 3, characterized in that: The process of fault detection is as follows: Use multi-layer perceptrons for high-level abstraction and classification of features; The failure probability is calculated using the Sigmoid activation function.

7. The substation equipment image recognition and fault detection method according to claim 1, 2 or 3, characterized in that: In the fault detection process, the dynamic threshold is used to determine whether a fault occurs, where the dynamic threshold T is: T=μ+3×δ where μ and δ are the mean and standard deviation of the detection probability, respectively.

8. The substation equipment image recognition and fault detection method according to claim 1, 2 or 3, characterized in that: In the fault detection process, the weights of the convolutional neural network are adjusted according to the cross entropy loss function, as follows: Where M is the number of categories, y o，c is the true label of the cth class of the oth sample, p o，c is the predicted probability.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the computer program performs the steps of the method according to any one of claims 1 to 8.

10. A substation equipment image recognition and fault detection system, comprising a memory and a processor connected to each other, wherein a computer program is stored in the memory, characterized in that: When the computer program is executed by a processor, the computer program performs the steps of the method according to any one of claims 1 to 8.

Images