Traction substation equipment operation state intelligent image recognition method and system and medium

Abstract

The invention relates to the technical field of operation state detection of traction substation equipment, and discloses an intelligent image recognition method and system for the operation state of the traction substation equipment and a medium, and the method comprises the steps: obtaining initial image data of the operation state of the substation equipment; preprocessing the initial image data to obtain intermediate image data; obtaining a sample actual label from the intermediate image data, and taking the intermediate image data and the corresponding sample actual label as a group of training samples so as to form a training sample set; constructing a state recognition model based on a deep learning neural network, optimizing weights and thresholds of a feature extraction unit and a recognition unit based on a particle swarm algorithm in combination with a training sample set, and finally completing training of the state recognition model; and detecting the intermediate image data of the operation state of the substation equipment through the trained state recognition model. Through the method provided by the invention, the workload of operation and maintenance personnel is reduced, the inspection efficiency is improved, and the risk occurrence probability is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of traction substation equipment operating status detection, and in particular to a traction substation equipment operating status intelligent image recognition method, system and medium. Background Art

[0002] With the comprehensive construction and in-depth application of smart substations, the intelligence and automation levels of primary and secondary equipment are getting higher and higher. The accuracy and rationality of the operating status of various equipment in traction substations directly affect the stability, safety, and efficiency of substations. At present, the inspection of equipment status in traction substations mainly relies on manual inspections and machine inspections. With the increasing number of equipment types and inspection points in smart substations, relying solely on human visual inspections is laborious and time-consuming, and it is very easy to miss inspections and misdetections due to manual misjudgment, which in turn affects the stable operation of the power grid, damages power equipment, and even causes power safety accidents. Therefore, it is increasingly important to combine the latest technologies in various industries to improve the level of intelligent and information-based inspections.

[0003] Traditional traction substation inspection and analysis mainly relies on manual visual judgment, which is labor-intensive and increasingly unable to meet the massive image recognition needs generated by modern traction substation inspections. In recent years, the new generation of artificial intelligence technology represented by deep learning has achieved remarkable results in the fields of transportation, medical care, finance, etc., especially in the field of computer vision. The recognition and analysis accuracy of deep learning algorithms for some specific targets has reached or exceeded the level of human intelligence. Therefore, the development of a system suitable for intelligent monitoring of the operating status of traction substation equipment and intelligent image analysis, and supporting the solution to the problem of automated processing of traction substation inspection images is a development trend of power station inspection, which will help reduce the workload of operation and maintenance personnel, improve inspection efficiency, and reduce the probability of risk. Summary of the invention

[0004] The present invention provides a substation equipment inspection operation image recognition detection method, system and medium to solve the above problems.

[0005] The present invention is achieved through the following technical solutions:

[0006] An intelligent image recognition method for the operating status of traction substation equipment, comprising:

[0007] Acquire initial image data of the operation status of substation equipment;

[0008] Preprocessing the initial image data to obtain intermediate image data;

[0009] According to prior knowledge, actual labels of samples are obtained from the intermediate image data, and the intermediate image data and the corresponding actual labels of samples are used as a group of training samples to form a training sample set;

[0010] Building a state recognition model based on a deep learning neural network, the state recognition model includes a feature extraction unit and a recognition unit, and optimizing the weights and thresholds of the feature extraction unit and the recognition unit based on a particle swarm algorithm combined with the training sample set, and finally completing the training of the state recognition model;

[0011] The intermediate image data of the operating status of the substation equipment is detected by the trained status recognition model.

[0012] As an optimization, the preprocessing includes supervised sample augmentation based on noise addition, rotation flipping, color jittering, and PCA analysis, as well as an unsupervised sample augmentation method based on a generative adversarial network, and image enhancement technology based on spatial filtering, histogram equalization, gamma correction, and spatial transformation, so as to achieve the expansion of the initial image data and improve the quality of the initial image data.

[0013] As an optimization, the feature extraction unit includes a SIFT extraction model, a HOG extraction model and a LBP extraction model.

[0014] As an optimization, the specific process of optimizing the weights and thresholds of the feature extraction unit and the recognition unit based on the particle swarm algorithm combined with the training sample set is as follows:

[0015] A1. Initialize the particle swarm parameters, including the particle population size G, the maximum speed v max , maximum position x max , dimension W, maximum number of iterations T max , learning factors c1 and c2, the expected minimum error value ε, and the initial speed and position of particles are randomly generated within the constraint range, thereby obtaining several initialized state recognition models;

[0016] A2, inputting the training sample set into a plurality of state recognition models, and calculating the fitness value based on the position of each particle in the particle swarm;

[0017] A3. Determine the individual optimal extreme value and the global optimal extreme value of each particle according to the fitness value;

[0018] A4. updating each particle in the particle swarm according to an improved speed and position updating strategy, thereby updating the state recognition model;

[0019] A5. Determine whether the termination condition is met. If so, output the currently updated particle swarm to obtain the optimal feature extraction unit, optimal weight and optimal threshold of the state recognition model. Otherwise, return to A2.

[0020] As an optimization, the fitness value is the correct ratio of the sample prediction label and the sample actual label output by the state recognition model when the intermediate image data in the training sample is input into the state recognition model.

[0021] As an optimization, in A4, the specific formula for updating each particle in the particle swarm according to the improved speed position update strategy is:

[0022] The speed is updated as follows:

[0023]

[0024] The update formula of position is as follows:

[0025]

[0026] in, is the velocity vector of the w-th dimension of the ith particle in the t-th iteration, w∈[1,W], is the velocity vector of the w-th dimension of the ith particle in the t-1th iteration, is the w-th dimension component of the best position of the ith particle in the t-1th iteration, is the w-th dimension component of the optimal extreme value of the i-th particle in the t-1-th iteration process, is the w-th dimension component of the global optimal extreme value in the t-1th iteration process, r1 and r2 are random numbers ranging from 0 to 1, ω is the inertia weight, k is the shrinkage factor,

[0027] As an optimization, the termination condition includes reaching a maximum number of iterations or a deviation between two maximum fitness values ​​obtained in two adjacent iterations is within a set range.

[0028] The present invention also discloses an intelligent image recognition system for the operating status of traction substation equipment, which is used to execute the aforementioned intelligent image recognition method for the operating status of traction substation equipment, comprising:

[0029] The acquisition module is used to obtain the initial image data of the operation status of the substation equipment;

[0030] An image preprocessing module, used for preprocessing the initial image data to obtain intermediate image data;

[0031] A sample set generation module, used to obtain sample actual labels from the intermediate image data according to prior knowledge, and use the intermediate image data and the corresponding sample actual labels as a group of training samples to form a training sample set;

[0032] A training module, for building a state recognition model based on a deep learning neural network, wherein the state recognition model includes a feature extraction unit and a recognition unit, and optimizing the weights and thresholds of the feature extraction unit and the recognition unit based on a particle swarm algorithm combined with the training sample set, and finally completing the training of the state recognition model;

[0033] The execution module is used to detect the intermediate image data of the operating status of the substation equipment through the trained state recognition model.

[0034] The present invention also discloses an electronic device, comprising at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute an intelligent image recognition method for the operating status of a traction substation equipment as described above.

[0035] The present invention also discloses a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the aforementioned intelligent image recognition method for the operating status of traction substation equipment.

[0036] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0037] 1. Through the status recognition model of the present invention, the equipment operating status of the traction substation can be quickly identified without manual detection, which is more automated and intelligent, helping to reduce the workload of operation and maintenance personnel, improve inspection efficiency, and reduce the probability of risk.

[0038] 2. The present invention uses a particle swarm algorithm to find the optimal feature extraction unit and recognition unit, thereby obtaining a recognition model of image data that is most suitable for the operating status of the traction substation equipment of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0040] Figure 1 The present invention provides a flow chart of an intelligent image recognition method for the operating status of traction substation equipment. DETAILED DESCRIPTION

[0041] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0042] This embodiment 1 provides an intelligent image recognition method for the operating status of traction substation equipment, such as Figure 1 As shown, including:

[0043] S1, obtaining initial image data of the operation status of substation equipment;

[0044] The operating status of substation equipment includes the status of the pressure plate, various warning lights, operating status lights, and the status of various switches, such as closed and open status.

[0045] S2, preprocessing the initial image data to obtain intermediate image data;

[0046] The preprocessing includes supervised sample augmentation based on noise addition, rotation flipping, color jitter, and PCA analysis, as well as an unsupervised sample augmentation method based on a generative adversarial network, and image enhancement technology based on spatial filtering, histogram equalization, gamma correction, and spatial transformation, so as to achieve the expansion of the initial image data and improve the quality of the initial image data.

[0047] Supervised sample augmentation techniques:

[0048] The main feature of image transformation augmentation is to perform specific image transformation operations (such as geometric transformation, optical transformation, statistical transformation, etc.) on the image samples of the training data set, so that the label information of the generated new image samples is consistent with the label information of the original image samples, thereby achieving an increase in the number of samples.

[0049] Specifically, they include flip augmentation, noise augmentation, blur augmentation, scaling augmentation, and affine transformation augmentation. These are existing technologies and will not be elaborated here.

[0050] Unsupervised sample augmentation techniques:

[0051] Generative Adversarial Networks (GANs) is an effective unsupervised sample augmentation method. GANs-based image augmentation uses GANs and its derivative models as tools to generate richer image samples on existing data sets, so as to achieve the goal of enriching training data samples and improving the performance of application models in test sets.

[0052] The generative adversarial network is structurally divided into two models, including the generative network and the adversarial network. The generative network continuously generates new samples based on existing samples, and the adversarial network judges the generated samples and achieves sample expansion through mutual game. Currently, the commonly used generative adversarial networks for sample augmentation include naive generative adversarial networks, conditional generative adversarial networks, and auxiliary classification conditional generative adversarial networks.

[0053] S3. According to prior knowledge, obtain actual labels of samples from the intermediate image data, and use the intermediate image data and the corresponding actual labels of samples as a group of training samples to form a training sample set;

[0054] There are actually only two cases for the actual label of the sample, one is fault and the other is normal. Therefore, when calculating the subsequent fitness value, if the sample prediction label predicts a fault and the actual label of the sample is normal, then the actual label and the sample prediction label are different.

[0055] S4. Building a state recognition model based on a deep learning neural network, the state recognition model includes a feature extraction unit and a recognition unit, and optimizing the weights and thresholds of the feature extraction unit and the recognition unit based on a particle swarm algorithm combined with the training sample set, and finally completing the training of the state recognition model;

[0056] The features of the intermediate image data are extracted by the feature extraction unit and then input into the recognition unit. Since the feature extraction unit includes a SIFT extraction model, a HOG extraction model and a LBP extraction model, it is unknown which extraction model is suitable for the image data and the recognition unit of the present invention. Therefore, the present invention optimizes the weights and thresholds of the feature extraction unit and the recognition unit based on the particle swarm algorithm combined with the training sample set, so as to obtain a state recognition model that is most suitable for the operating state image data of the traction substation equipment of the present invention.

[0057] 4. According to claim 1, a method for intelligent image recognition of the operating status of traction substation equipment is characterized in that the specific process of optimizing the weights and thresholds of the feature extraction unit and the recognition unit based on the particle swarm algorithm combined with the training sample set is:

[0058] A1. Initialize the particle swarm parameters, including the particle population size G, the maximum speed v max , maximum position x max , dimension W, maximum number of iterations T max , learning factors c1 and c2, the expected minimum error value ε, and the initial speed and position of particles are randomly generated within the constraint range, thereby obtaining several initialized state recognition models;

[0059] A2, inputting the training sample set into a plurality of state recognition models, and calculating the fitness value based on the position of each particle in the particle swarm;

[0060] In some embodiments, the fitness value is the correct ratio of the sample prediction label and the sample actual label output by the state recognition model when the intermediate image data in the training sample is input into the state recognition model.

[0061] For example, there are 500 samples, and the 500 samples are input into several state recognition models respectively. The sample prediction labels output by the several state recognition models are compared with the corresponding sample actual labels respectively. The correct proportion = the number of sample prediction labels that are the same as the sample actual labels / the total number of samples.

[0062] A3. Determine the individual optimal extreme value and the global optimal extreme value of each particle according to the fitness value;

[0063] The selection of individual optimal extreme values ​​and global optimal extreme values ​​is an existing technology and will not be described in detail here.

[0064] A4. updating each particle in the particle swarm according to an improved speed and position updating strategy, thereby updating the state recognition model;

[0065] The specific formula for updating each particle in the particle swarm according to the improved speed position update strategy is:

[0066] The speed is updated as follows:

[0067]

[0068] The update formula of position is as follows:

[0069]

[0070] in, is the velocity vector of the w-th dimension of the ith particle in the t-th iteration, w∈[1,W], is the velocity vector of the w-th dimension of the ith particle in the t-1th iteration, is the w-th dimension component of the best position of the ith particle in the t-1th iteration, is the w-th dimension component of the optimal extreme value of the i-th particle in the t-1-th iteration process, is the w-th dimension component of the global optimal extreme value in the t-1th iteration process, r1 and r2 are random numbers ranging from 0 to 1, ω is the inertia weight, k is the shrinkage factor,

[0071] The half of particles with the highest fitness values ​​perform normal speed updates, and the half of particles with the lowest fitness values ​​perform updates using an improved speed update formula. By setting the contraction factor, different areas can be effectively searched to obtain higher quality solutions.

[0072] A5. Determine whether the termination condition is met. If so, output the currently updated particle swarm to obtain the optimal feature extraction unit, optimal weight and optimal threshold of the state recognition model, that is, the optimal state recognition model. Otherwise, return to A2.

[0073] S5. Detecting the intermediate image data of the operating status of the substation equipment by using the trained status recognition model.

[0074] The termination condition includes reaching the maximum number of iterations or the deviation between two maximum fitness values ​​obtained in two adjacent iterations is within a set range.

[0075] Embodiment 2 further discloses an intelligent image recognition system for the operation status of traction substation equipment, which is used to execute an intelligent image recognition method for the operation status of traction substation equipment in embodiment 1, including:

[0076] The acquisition module is used to obtain the initial image data of the operation status of the substation equipment;

[0077] The acquisition module may be a camera installed near each device, and is used to acquire image data of the device's operating status.

[0078] An image preprocessing module, used for preprocessing the initial image data to obtain intermediate image data;

[0079] A sample set generation module, used to obtain sample actual labels from the intermediate image data according to prior knowledge, and use the intermediate image data and the corresponding sample actual labels as a group of training samples to form a training sample set;

[0080] A training module, for building a state recognition model based on a deep learning neural network, wherein the state recognition model includes a feature extraction unit and a recognition unit, and optimizing the weights and thresholds of the feature extraction unit and the recognition unit based on a particle swarm algorithm combined with the training sample set, and finally completing the training of the state recognition model;

[0081] The execution module is used to detect the intermediate image data of the operating status of the substation equipment through the trained state recognition model.

[0082] Example 3 also discloses an electronic device, comprising at least one processor, and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute an intelligent image recognition method for the operating status of traction substation equipment as described in Example 3.

[0083] Embodiment 4 further discloses a storage medium storing a computer program, which, when executed by a processor, implements the method for intelligent image recognition of the operating status of traction substation equipment described in Embodiment 1.

[0084] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for intelligent image recognition of the operating status of traction substation equipment, characterized in that: include: Acquire initial image data of the operation status of substation equipment; Preprocessing the initial image data to obtain intermediate image data; According to prior knowledge, actual labels of samples are obtained from the intermediate image data, and the intermediate image data and the corresponding actual labels of samples are used as a group of training samples to form a training sample set; Building a state recognition model based on a deep learning neural network, the state recognition model includes a feature extraction unit and a recognition unit, and optimizing the weights and thresholds of the feature extraction unit and the recognition unit based on a particle swarm algorithm combined with the training sample set, and finally completing the training of the state recognition model; The intermediate image data of the operating status of the substation equipment is detected by the trained status recognition model.

2. The intelligent image recognition method for the operating status of traction substation equipment according to claim 1 is characterized in that: The preprocessing includes supervised sample augmentation based on noise addition, rotation flipping, color jitter, and PCA analysis, as well as an unsupervised sample augmentation method based on a generative adversarial network, and image enhancement technology based on spatial filtering, histogram equalization, gamma correction, and spatial transformation, so as to achieve the expansion of the initial image data and improve the quality of the initial image data.

3. The intelligent image recognition method for the operating status of traction substation equipment according to claim 1 is characterized in that: The feature extraction unit includes a SIFT extraction model, a HOG extraction model and a LBP extraction model.

4. The method for intelligent image recognition of the operating status of traction substation equipment according to claim 1, characterized in that: The specific process of optimizing the weights and thresholds of the feature extraction unit and the recognition unit based on the particle swarm algorithm combined with the training sample set is as follows: A1. Initialize the particle swarm parameters, including the particle population size G, the maximum speed v max , maximum position x max , dimension W, maximum number of iterations T max , learning factors c1 and c2, the expected minimum error value ε, and the initial speed and position of particles are randomly generated within the constraint range, thereby obtaining several initialized state recognition models; A2, inputting the training sample set into a plurality of state recognition models, and calculating the fitness value based on the position of each particle in the particle swarm; A3. Determine the individual optimal extreme value and the global optimal extreme value of each particle according to the fitness value; A4. updating each particle in the particle swarm according to an improved speed and position updating strategy, thereby updating the state recognition model; A5. Determine whether the termination condition is met. If so, output the currently updated particle swarm to obtain the optimal feature extraction unit, optimal weight and optimal threshold of the state recognition model. Otherwise, return to A2.

5. The method for intelligent image recognition of the operating status of traction substation equipment according to claim 4 is characterized in that: The fitness value is the correct ratio of the sample prediction label and the sample actual label output by the state recognition model when the intermediate image data in the training sample is input into the state recognition model.

6. The method for intelligent image recognition of the operating status of traction substation equipment according to claim 4 is characterized in that: In A4, the specific formula for updating each particle in the particle swarm according to the improved speed position update strategy is: The speed is updated as follows: The update formula of position is as follows: in, is the velocity vector of the w-th dimension of the ith particle in the t-th iteration, w∈[1,W], is the velocity vector of the w-th dimension of the ith particle in the t-1th iteration, is the w-th dimension component of the best position of the ith particle in the t-1th iteration, is the w-th dimension component of the optimal extreme value of the i-th particle in the t-1-th iteration process, is the w-th dimension component of the global optimal extreme value in the t-1th iteration process, r1 and r2 are random numbers ranging from 0 to 1, ω is the inertia weight, k is the shrinkage factor, 7. The method for intelligent image recognition of the operating status of traction substation equipment according to claim 4 is characterized in that: The termination condition includes reaching the maximum number of iterations or the deviation between two maximum fitness values ​​obtained in two adjacent iterations is within a set range.

8. An intelligent image recognition system for the operating status of traction substation equipment, used to execute the intelligent image recognition method for the operating status of traction substation equipment according to any one of claims 1 to 7, characterized in that: include: The acquisition module is used to obtain the initial image data of the operation status of the substation equipment; An image preprocessing module, used for preprocessing the initial image data to obtain intermediate image data; A sample set generation module, used to obtain sample actual labels from the intermediate image data according to prior knowledge, and use the intermediate image data and the corresponding sample actual labels as a group of training samples to form a training sample set; A training module, for building a state recognition model based on a deep learning neural network, wherein the state recognition model includes a feature extraction unit and a recognition unit, and optimizing the weights and thresholds of the feature extraction unit and the recognition unit based on a particle swarm algorithm combined with the training sample set, and finally completing the training of the state recognition model; The execution module is used to detect the intermediate image data of the operating status of the substation equipment through the trained state recognition model.

9. An electronic device, characterized in that: It includes at least one processor and a memory that is communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a method for intelligent image recognition of the operating status of traction substation equipment as described in any one of claims 1 to 7.

10. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, an intelligent image recognition method for the operating status of traction substation equipment according to any one of claims 1 to 7 is implemented.

Images