Multi-cabin system transformer substation equipment defect detection method and device and medium

Abstract

The invention discloses a multi-cabin system transformer substation equipment defect detection method which comprises the following steps: acquiring positive and negative samples of multi-cabin system transformer substation equipment defects, and marking different types of defects; performing feature extraction on the marked positive and negative samples of the multi-cabin system substation equipment defects to obtain an equipment defect feature map, and performing correlation elimination processing on the obtained equipment defect feature map; constructing a semantic perception module branch network according to the processed equipment defect feature map, and introducing a category-level feature refining module to position category defect features in the equipment defect feature map; and performing cross-scale fusion processing on the positioned category defect features to obtain a detection result of the multi-cabin system substation equipment defects. According to the multi-cabin system transformer substation equipment defect detection method, the problem that a traditional transformer substation equipment detection method in the prior art depends on a large amount of manpower to participate is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of defect target detection of multi-compartment system substation equipment, and in particular to a defect detection method of multi-compartment system substation equipment. Background Art

[0002] With the iterative development of primary and secondary equipment integration technology, and the continuous improvement of the quality and efficiency requirements of substation construction in current power grid construction, multi-cabin system substations are gradually entering people's field of vision and emerging as a solution that integrates multiple advantages such as low-carbon intelligence, intensive and efficient. This type of substation adopts the advanced concept of "modular design, factory assembly, and mechanized construction". By completing the integration and prefabrication of primary and secondary equipment in the cabin in the factory, it significantly shortens the construction period, reduces on-site construction work, and reduces environmental impact and construction safety risks.

[0003] Defect detection in multi-compartment substations is crucial to maintaining stable equipment operation. However, existing traditional substation equipment detection methods rely on a large amount of manpower and require a high level of expertise, which increases uncertainty and the challenges of equipment defect diagnosis. Summary of the invention

[0004] The embodiment of the present application provides a method for detecting defects in a multi-compartment system substation equipment, thereby solving the problem in the prior art that traditional substation equipment detection methods rely on a large amount of manpower participation and require a high level of professional knowledge, thereby achieving accurate prediction of appearance defects in multi-compartment system substation equipment.

[0005] The embodiment of the present application provides a method for detecting defects in equipment in a multi-compartment system substation, comprising the following steps: obtaining positive and negative samples of defects in equipment in a multi-compartment system substation, and marking different types of defects; performing feature extraction on the marked positive and negative samples of defects in equipment in a multi-compartment system substation to obtain an equipment defect feature map, and performing correlation elimination processing on the obtained equipment defect feature map; constructing a semantic perception module branch network according to the processed equipment defect feature map, and introducing a category-level feature refinement module to locate category defect features in the equipment defect feature map; performing cross-scale fusion processing on the located category defect features to obtain detection results of defects in equipment in a multi-compartment system substation.

[0006] Furthermore, the feature extraction is performed on the positive and negative samples of the marked multi-compartment system substation equipment defects, specifically: 3 layers of 5×5 convolution kernels are used to extract the positive and negative samples of the marked multi-compartment system substation equipment defects to obtain the basic visual features of the positive and negative samples; and 1 layer of 1×1 convolution kernel is used to map the basic visual features to a higher dimension to obtain an equipment defect feature map.

[0007] Furthermore, the obtained equipment defect feature map is subjected to correlation elimination processing, and the specific method is as follows: the mean of each feature in the obtained equipment defect feature map is adjusted to 0, and the variance is adjusted to 1, so as to obtain the equipment defect feature map after feature standardization; the covariance matrix of the equipment defect feature map after feature standardization is calculated; the objective function is constructed to make the covariance matrix close to the unit matrix, so as to obtain the equipment defect feature map with correlation eliminated.

[0008] Furthermore, the method for obtaining the semantic perception module is as follows: a rectangular area is divided according to the defect label, and all elements of the rectangular area are set to 0 as a category mask, the equipment defect feature map and the corresponding category mask are multiplied to obtain the corresponding category branch network, and the category branch networks of several categories constitute a semantic perception module.

[0009] Furthermore, the cross-scale fusion processing of the located category defect features is specifically as follows: The intra-scale feature interaction of category defect features is carried out through position encoding and self-attention mechanism; and based on the convolution operation and fusion module, the category defect features are fused across scales to obtain the detection results of multi-compartment system substation equipment defects.

[0010] Furthermore, the device of the data detection and analysis method based on power operation is characterized in that it includes a sample collection module, a feature extraction module, a feature processing module, and a feature fusion module; the sample collection module is used to obtain positive and negative samples of multi-cabin system substation equipment defects and mark different types of defects; the feature extraction module is used to extract features from the marked positive and negative samples of multi-cabin system substation equipment defects to obtain an equipment defect feature map, and perform correlation elimination processing on the obtained equipment defect feature map; the feature processing module is used to construct a semantic perception module branch network according to the processed equipment defect feature map, and introduce a category-level feature refinement module to locate the category defect features in the equipment defect feature map; the feature fusion module is used to perform cross-scale fusion processing on the located category defect features to obtain the detection results of multi-cabin system substation equipment defects.

[0011] Furthermore, an electronic device is characterized in that the electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute a data detection and analysis method based on power operation as described in any one of claims 1 to 7.

[0012] Furthermore, a computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the method for data detection and analysis based on power operation as described in any one of claims 1 to 7 is implemented.

[0013] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. Through the multi-compartment system substation equipment defect detection method, the appearance defects of multi-compartment system substation equipment can be accurately predicted, which solves the problem that the traditional substation equipment detection method in the prior art relies on a large amount of manpower participation.

[0014] 2. Through a positive and negative sample driven method, the labeled defect positive and negative samples are used as the common input of the model, thereby eliminating the correlation between different channels and effectively avoiding the occurrence of misidentification.

[0015] 3. By designing a semantic perception module for multi-compartment substation equipment images, the category-level centers of features from different multi-compartment substation images are aligned, thereby enhancing the compactness of the distribution within the defect category and improving the classification performance of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flow chart of a method for detecting defects in a multi-compartment system substation equipment provided in an embodiment of the present application; Figure 2 A structural diagram of the semantic perception module provided in the embodiment of the present application; DETAILED DESCRIPTION The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or still includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "include..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0018] S1, obtain positive and negative samples of multi-cabin system substation equipment defects, and mark different types of defects in negative samples; Among the positive and negative samples of the multi-chamber system substation equipment defects, the positive samples are images when the equipment is free of defects, and the negative samples are images when the equipment has defects; Marking positive and negative samples of different defects can avoid the misidentification of defects in multi-compartment substation equipment.

[0019] S2, extracting features from positive and negative samples of marked multi-compartment system substation equipment defects to obtain an equipment defect feature map, and performing correlation elimination processing on the obtained equipment defect feature map; The feature extraction is performed on the positive and negative samples of the marked multi-compartment system substation equipment defects. The feature extraction process is as follows: Use 3 layers of 5×5 convolution kernels to extract positive and negative samples of marked multi-cabin system substation equipment defects and obtain the basic visual features of positive and negative samples; Use a layer of 1×1 convolution kernel to map the 3-channel features to a higher dimension.

[0020] The calculation process of feature extraction is expressed as:

[0021] Where: is the extracted equipment defect feature map, is a 1×1 convolution, is a 5×5 convolution, are the input positive and negative sample images.

[0022] It should be noted that The dimension is b*k*w*h, where b is the batch size of the model input, k is the number of channels of the device defect feature map, w is the width of the input positive and negative sample images, and h is the height of the input positive and negative sample images.

[0023] Batch size is a key hyperparameter in deep learning training. It determines the number of data samples processed by the model in each training iteration. The appropriate batch size can affect the training speed, stability, and memory usage, so it needs to be adjusted according to the specific situation.

[0024] The specific method of performing correlation elimination processing on the obtained equipment defect feature map is as follows: The mean of each feature in the obtained equipment defect feature map is adjusted to 0, and the variance is adjusted to 1, so as to obtain a standardized equipment defect feature map; Calculate the covariance matrix of the standardized device defect feature map; The objective function is constructed to make the covariance matrix close to the unit matrix, and the device defect feature map with correlation eliminated is obtained.

[0025] The covariance matrix is ​​a tool used in machine learning and statistics to measure the correlation between two variables. When processing the features of positive and negative samples, the covariance matrix can be used to eliminate the correlation of irrelevant features, thereby reducing the risk of misidentification of defects.

[0026] It should be noted that before analyzing the correlation between features, the features need to be standardized first. This is because the calculation of the covariance matrix is ​​very sensitive to the scale of the features, so standardization can ensure that the features are compared at the same scale.

[0027] The calculation formula for the feature standardization is as follows:

[0028] Where: X s is the standardized equipment defect feature map, X is the original feature value in the equipment defect feature map, is the mean, is the standard deviation.

[0029] Considering that covariance can represent the linear correlation between two random variables, and each element in the covariance matrix represents the covariance between two random variables, it can represent the linear correlation between any two variables in a set of random variables.

[0030] Therefore, calculating the covariance of the corresponding channels of the positive and negative sample features can represent the linear correlation between channels. Making the covariance matrix composed of the covariances calculated between different channels close to the unit matrix during training can effectively eliminate the correlation of irrelevant features, so that the model can effectively distinguish between positive and negative samples during training and avoid misidentification.

[0031] The process of obtaining the covariance matrix is ​​as follows:

[0032] Where: is the obtained covariance matrix, represents the kth channel of the positive sample, represents the kth channel of the negative sample, and Cov represents the calculated covariance.

[0033] The covariance calculation process of the two equipment defect feature maps is expressed as follows:

[0034] Where: is the mean of the kth channel of the positive sample, is the mean of the kth channel of the negative samples.

[0035] The objective function is constructed to make the covariance matrix close to the identity matrix, and the calculation formula is as follows:

[0036] Where: I is the identity matrix, is the de-correlated device defect signature map, is the 1-norm, is the covariance matrix.

[0037] S3, constructing a semantic perception module branch network based on the processed equipment defect feature map, and introducing a category-level feature refinement module to locate the category defect features in the equipment defect feature map; The semantic perception module includes several category branch networks. In each branch network, the feature map is multiplied by the corresponding category mask. The category mask is a rectangular area divided by the defect label. The elements of the rectangular area divided by the defect label are all set to 0. After the feature map is multiplied by the mask, the category area is highlighted, which helps the branch network learn the semantics of the corresponding defect category. The specific calculation formula of the mask feature map of the branch network is as follows:

[0038] Where: is the mask feature map of the c-th category branch network, is the category mask of the c-th category branch network, Feature map for the semantic perception module input.

[0039] Since the mask feature map obtained by the semantic perception module is too coarse to accurately locate the category features, a category-level feature refinement module is introduced to adaptively highlight the feature areas of the corresponding categories.

[0040] The calculation process of the category-level feature refinement module is as follows:

[0041] Where: is a 3×3 convolution kernel, is the mask feature map of the c-th category branch network, is the global maximum feature of the mask feature map of the c-th category branch network, is the global average feature of the mask feature map of the c-th category branch network, is the category mask of the c-th category branch network, and Sigm is the sigmoid function.

[0042] Semantic perception of features can effectively improve the distinctiveness of defects in multi-compartment substation equipment, enhance the degree of separation between defect categories, and improve the model's classification performance for different defects.

[0043] The designed region normalization layer further refines the category-level center alignment of defects. This layer only normalizes the features within the defect category area. After feature refinement, only the corresponding feature elements with high values ​​will be assigned to the category area.

[0044] In order to flexibly identify different feature elements, k-means clustering is applied to the defect space feature map obtained by average pooling along the channel axis, including: Divide the feature elements into k clusters; The first to tth clusters are considered as category regions, and the remaining clusters are ignored regions; Set t to 1 and find the optimal k through hyperparameter search.

[0045] S4, cross-scale fusion processing is performed on the located category defect features to obtain the detection results of multi-cabin system substation equipment defects.

[0046] The cross-scale fusion processing of the located category defect features is specifically as follows: Intra-scale feature interaction of category defect features through position encoding and self-attention mechanism; Based on the convolution operation and fusion module, the category defect features are fused across scales to obtain the detection results of multi-compartment system substation equipment defects.

[0047] It should be noted that an efficient hybrid encoder is designed in RT-DETR, which mainly includes two parts: intra-scale feature interaction and cross-scale feature fusion. The intra-scale feature interaction is mainly processed by position encoding and self-attention mechanism, while the cross-scale feature fusion relies on convolution operation and special fusion module to achieve.

[0048] It should be noted that RT-DETR is an improvement based on DETR, which is the first target detection framework that successfully integrates Transformer as a detection center building block. The Global Query it provides can eliminate the need for non-maximum suppression. However, the high computational cost of DETR limits its practicality in various fields and weakens its advantage in replacing NMS.

[0049] The process of position encoding is as follows:

[0050] Where: pos is the position, i is the dimension index, d model is the dimension of the model.

[0051] It should be noted that position encoding can help the model understand the relative or absolute relationship between different positions.

[0052] The cross-scale feature fusion is achieved through convolutions of different sizes and fusion modules. The fusion module represents the information interaction within the feature map channel and is achieved through cross-channel information integration.

[0053] RT-DETR uses uncertainty minimization query selection to determine a fixed number of encoder features as the initial object query for the decoder. The main task of uncertainty minimization query selection is to select a set of features with high information content from the features generated by the encoder. This process can be expressed as follows:

[0054] Where: is the initial query set selected, U is the uncertainty evaluation function, Q is the possible query candidate set, and F is the feature set output by the encoder.

[0055] It should be noted that the uncertainty evaluation function determines how to evaluate the uncertainty of each query candidate.

[0056] The uncertainty assessment function is defined as follows:

[0057] Where: represents the predicted probability distribution given the query candidate set and the feature set output by the encoder.

[0058] After selecting the initial queries, RT-DETR’s decoder refines these queries through an iterative optimization process. In each iteration, the decoder evaluates the current query using the auxiliary prediction head and updates the query to more accurately reflect the location and category of the object.

[0059] The iterative optimization process is expressed by the following formula:

[0060] Where: is the query for the t+1th iteration, The query for the tth iteration, is the learning rate, represents the vector differential operator, L is the loss function, D is the training data with true labels, Represents the gradient of the loss function with respect to the query.

[0061] It should be noted that the gradient provides the opposite direction of the steepest ascent direction of the loss function at the current query point, which is the key information for updating parameters in gradient descent.

[0062] Uncertainty minimization query selection is a strategy to improve model performance by optimizing epistemic uncertainty. In this approach, the model first evaluates the potential uncertainty of each possible query. These queries usually involve some part of the data that is not well learned or understood. In the application scenario of substation equipment defect detection, this means that the model will identify those areas that may have the greatest impact on the equipment status but have not yet been accurately detected or learned.

[0063] Uncertainty minimization query selection can construct and optimize epistemic uncertainty to model the joint latent variables of encoder features, with the goal of minimizing the uncertainty of the model during the learning process. Through uncertainty minimization query selection, the model can effectively identify the most critical areas for defect detection of multi-compartment system substation equipment, thereby achieving higher detection accuracy and efficiency under limited resources.

[0064] The device for the data detection and analysis method based on power operation is characterized by comprising a sample collection module, a feature extraction module, a feature processing module and a feature fusion module; Sample collection module: used to obtain positive and negative samples of defects in multi-compartment substation equipment and mark different types of defects; Feature extraction module: used to extract features from positive and negative samples of marked multi-compartment system substation equipment defects, obtain equipment defect feature maps, and perform correlation elimination processing on the obtained equipment defect feature maps; Feature processing module: used to construct a semantic perception module branch network based on the processed equipment defect feature map, and introduce a category-level feature refinement module to locate the category defect features in the equipment defect feature map; Feature fusion module: used to perform cross-scale fusion processing on the located category defect features to obtain the detection results of multi-cabin system substation equipment defects.

[0065] An electronic device, characterized in that the electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute a data detection and analysis method based on power operation as described in any one of claims 1 to 7.

[0066] A computer-readable storage medium storing a computer program, characterized in that when the computer program is executed by a processor, a data detection and analysis method based on power operation as described in any one of claims 1 to 7 is implemented.

[0067] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete 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 disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0068] 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 flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, 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 flowchart and / or block diagram. 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.

[0069] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate 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 A function specified in one or more boxes.

[0070] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions 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.

[0071] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0072] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for detecting defects in multi-compartment system substation equipment, characterized in that: The following steps are involved: Obtain positive and negative samples of defects in multi-compartment substation equipment and mark different types of defects; Feature extraction is performed on the positive and negative samples of the marked multi-compartment system substation equipment defects to obtain an equipment defect feature map, and correlation elimination processing is performed on the obtained equipment defect feature map; A semantic perception module branch network is constructed based on the processed equipment defect feature graph, and a category-level feature refinement module is introduced to locate the category defect features in the equipment defect feature graph; The located category defect features are fused across scales to obtain the detection results of multi-compartment system substation equipment defects.

2. The multi-compartment system substation equipment defect detection method according to claim 1, characterized in that: The feature extraction of positive and negative samples of marked multi-compartment system substation equipment defects is specifically as follows: Convolution kernels are used to extract positive and negative samples of marked multi-cabin system substation equipment defects to obtain the basic visual features of positive and negative samples; The basic visual features of positive and negative samples are processed to obtain the initial device defect feature map. The initial equipment defect characteristic diagram is calculated using the following formula: Where: is the extracted equipment defect feature map, is a 1×1 convolution, is a 5×5 convolution, are the input positive and negative sample images.

3. The multi-compartment system substation equipment defect detection method according to claim 2, characterized in that: The specific method of performing correlation elimination processing on the obtained equipment defect feature map is as follows: The mean and variance of each feature in the obtained equipment defect feature map are adjusted to obtain the equipment defect feature map after feature standardization, and the covariance matrix of the equipment defect feature map after feature standardization is calculated; According to the covariance matrix of the equipment defect feature map, an objective function is constructed to make the covariance matrix close to the unit matrix, and the equipment defect feature map with correlation eliminated is obtained.

4. A multi-compartment system substation equipment defect detection method as claimed in claim 3, characterized in that: The method for obtaining the semantic perception module is as follows: Divide a rectangular area according to the defect label, and set all elements of the rectangular area to 0 as a category mask; Multiply the device defect feature map and the corresponding category mask to obtain the corresponding category branch network; The category branch networks of several categories are combined to form a semantic perception module.

5. A method for detecting defects in multi-compartment system substation equipment as claimed in claim 4, characterized in that: The cross-scale fusion processing of the located category defect features is specifically as follows: Intra-scale feature interaction of category defect features through position encoding and self-attention mechanism; Based on the convolution operation and fusion module, the category defect features are fused across scales to obtain the detection results of multi-compartment system substation equipment defects.

6. A method for detecting defects in multi-compartment substation equipment as claimed in claim 3, characterized in that: The calculation formula for the feature standardization is as follows: Where: X s is the standardized equipment defect feature map, X is the original feature value in the equipment defect feature map, is the mean, is the standard deviation.

7. A multi-compartment system substation equipment defect detection method as claimed in claim 1, characterized in that: The feature refinement module introduced at the category level locates the category defect features in the device defect feature map, specifically: Where: is a 3×3 convolution kernel, is the mask feature map of the c-th category branch network, is the global maximum feature of the mask feature map of the c-th category branch network, is the global average feature of the mask feature map of the c-th category branch network, is the category mask of the c-th category branch network, and Sigm is the sigmoid function.

8. A device using the data detection and analysis method based on power operation according to any one of claims 1 to 7, characterized in that: It includes sample collection module, feature extraction module, feature processing module and feature fusion module; Sample collection module: used to obtain positive and negative samples of defects in multi-compartment substation equipment and mark different types of defects; Feature extraction module: used to extract features from positive and negative samples of marked multi-compartment system substation equipment defects, obtain equipment defect feature maps, and perform correlation elimination processing on the obtained equipment defect feature maps; Feature processing module: used to construct a semantic perception module branch network based on the processed equipment defect feature map, and introduce a category-level feature refinement module to locate the category defect features in the equipment defect feature map; Feature fusion module: used to perform cross-scale fusion processing on the located category defect features to obtain the detection results of multi-cabin system substation equipment defects.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute a data detection and analysis method based on power operation as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it implements a data detection and analysis method based on power operation as described in any one of claims 1 to 7.

Images