An intelligent patrol system for protection devices based on multi-sensing technology

Through an intelligent patrol system based on multi-perception technology, using infrared cameras, multi-perception devices and convolutional neural network models, automatic patrol and fault diagnosis of relay protection devices is realized, and the existing technology cannot meet high-quality patrol without dead angles. The problem has improved the efficiency and safety of inspections.

CN113991837BActive Publication Date: 2024-11-22STATE GRID JIANGSU ELECTRIC POWER CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111168767.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-09-30
Publication Date
2024-11-22
Estimated Expiration
2041-09-30

AI Technical Summary

Technical Problem

The existing inspection methods cannot meet the high-quality inspection needs without blind spots, making it difficult to ensure the safe and reliable operation of relay protection devices, especially when the safety operation requirements of the power grid are constantly increasing.

Method used

Design an intelligent inspection system for protection devices based on multi-perception technology, including device perception layer, network interconnection layer and analysis application layer, and use infrared cameras, multi-perception devices, convolutional neural network model and object detection model to realize the protection device Automatic inspection and fault diagnosis, collect and analyze data through self-travel devices, identify equipment status and faults, and assist operation and maintenance personnel in troubleshooting problems.

Benefits of technology

It realizes unmanned or less-manned inspections of the protection devices, improves the inspection work efficiency, can quickly respond to equipment abnormalities and accurately locate hidden dangers, and ensures the safe and reliable operation of the relay protection devices.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113991837B_ABST
    Figure CN113991837B_ABST
Patent Text Reader

Abstract

The present invention discloses a protection device intelligent patrol system based on multi-sensing technology, including: equipment perception layer, network interconnection layer, analysis application layer, platform sharing layer, the equipment perception layer includes: infrared camera, multi-sensing device; the network interconnection layer includes: station control layer central switch, hard disk recorder, intelligent on-site module; the analysis application layer includes: image recognition host, secondary intelligent patrol device; the platform sharing layer includes cloud platform, control system. The protection device intelligent patrol system based on multi-sensing technology provided by the present invention can integrate and use information of different dimensions to discover equipment and circuit anomalies, and issue early warnings, so as to achieve rapid response to protection device defects and accurate positioning of hidden dangers, improve the efficiency of secondary equipment patrol work, and truly realize unmanned or less-manned patrol.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to an intelligent patrol system for protection devices based on multi-sensing technology, belonging to the technical field of intelligent patrol systems. Background Art

[0002] With the strengthening of the power grid, the number of substations and relay protection devices has increased significantly, and the daily inspection workload of relay protection operation and maintenance personnel has reached saturation. However, in order to ensure the safe and reliable operation of the relay protection devices, regular inspections of the protection devices are essential. As the guarantee rate of safe operation of the power grid continues to increase, the inspection methods have also evolved from normal inspections to lights-out inspections, comprehensive inspections, special inspections, etc.

[0003] It can be seen that the existing patrol methods can no longer meet such high-quality, no-dead-angle patrol requirements. In order to better protect the existing protective devices, a new intelligent patrol system must be designed. Summary of the invention

[0004] Purpose: In order to overcome the problems of the existing patrol methods in the prior art, the present invention provides an intelligent patrol system for protection devices based on multi-sensing technology, which lays a foundation for realizing unmanned or less-manned patrol of relay protection.

[0005] Technical solution: To solve the above technical problems, the technical solution adopted by the present invention is:

[0006] An intelligent patrol system for protection devices based on multi-sensing technology, comprising: a device sensing layer, a network interconnection layer, and an analysis application layer;

[0007] The equipment sensing layer includes: an infrared camera and a multi-sensing device. The infrared camera and the multi-sensing device are installed on a self-propelled device. The self-propelled device patrols on the tracks inside and outside the secondary equipment cabinet to monitor the devices, circuits and environment outside the cabinet, and collect the data required by the intelligent patrol system.

[0008] The network interconnection layer includes: a central switch of the station control layer, a hard disk recorder, and an intelligent local module. After the central switch of the station control layer obtains the online monitoring data and alarm information of the protection device, it is sent to the secondary intelligent inspection device of the analysis application layer. The hard disk recorder is used to connect to the infrared camera, store and send images to the image recognition host. The intelligent local module sends the information collected by the multi-sensing device to the secondary intelligent inspection device.

[0009] The analysis application layer includes: an image recognition host and a secondary intelligent inspection device. The image recognition host uses a convolutional neural network model to detect the images collected and uploaded by the hard disk recorder. According to the different image contents of the inspection, the protection device status, switch handle status, pressure plate status, terminal block wiring, grounding copper plate, cable status, optical fiber status, spare optical fiber dust cap, cabinet blockage and cabinet foreign matter are identified. The secondary intelligent inspection device uses the online monitoring data of the protection device and the image recognition results output by the image recognition host, and uses homologous comparison and trend analysis to judge the operating status and change trend of the protection device, and automatically identify the explicit and hidden faults of the protection device. The possible fault location is determined by the alarm signal, and the self-propelled device moves to the abnormal equipment and possible fault location to collect images, so as to assist the operation and maintenance personnel in troubleshooting.

[0010] As a preferred solution, it also includes a platform sharing layer, which includes a cloud platform and a control system. The dispatching data network sends relevant information to the dispatching end to provide data support for advanced applications of the control system and the cloud platform.

[0011] As a preferred solution, the multi-sensing device includes: a temperature and humidity sensor, and a smoke detector.

[0012] As a preferred solution, the convolutional neural network model includes three convolution layers and three pooling layers.

[0013] As a preferred solution, the convolution layer adopts depth-wise separable convolution, and the depth-wise separable convolution formula is as follows:

[0014] K S ×K S ×C×Q

[0015] Among them, K S is the size of the convolution kernel, C is the number of channels, Q is the number of convolution kernels, the convolution kernels of the three depthwise separable convolutions are 2, 3, and 4, and the numbers of convolution kernels are 8, 16, and 16, respectively.

[0016] As a preferred solution, the downsampling ratio of the pooling layer is 2.

[0017] As a preferred solution, the image recognition host also includes a target detection model, which is used to annotate the image recognition results. The target detection model adopts a Faster-CNN structure.

[0018] Beneficial effects: The intelligent inspection system for protection devices based on multi-sensing technology provided by the present invention can integrate and utilize information of different dimensions to discover equipment and circuit abnormalities, and issue early warnings, thereby achieving rapid response to defects in protection devices and accurate positioning of hidden dangers, improving the efficiency of secondary equipment inspections, and truly realizing unmanned or less-manned inspections. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a structural diagram of the system of the present invention.

[0020] Figure 2 It is a specific data signal flow diagram of the system of the present invention.

[0021] Figure 3 It is an abstract flow chart of the system of the present invention.

[0022] Figure 4 It is a flow chart of image detection using image feature model. DETAILED DESCRIPTION

[0023] The present invention will be further described below in conjunction with specific embodiments.

[0024] An intelligent patrol system for protection devices based on multi-sensing technology includes: a device perception layer, a network interconnection layer, an analysis application layer, and a platform sharing layer, and applies image recognition technology and intelligent patrol and fault diagnosis technology to realize automatic patrol of protection devices.

[0025] The equipment sensing layer includes: infrared cameras and multi-sensing devices. The infrared cameras and multi-sensing devices are installed on self-propelled devices. The self-propelled devices patrol on the tracks inside and outside the secondary equipment cabinet to monitor the devices, circuits and environment outside the cabinet, and collect data required by the intelligent patrol system. Multi-sensing devices include: temperature and humidity sensors, and smoke detectors.

[0026] The network interconnection layer includes: the central switch of the station control layer, the hard disk recorder, and the intelligent local module. After the central switch of the station control layer obtains the online monitoring data and alarm information of the protection device, it is sent to the secondary intelligent inspection device of the analysis application layer. The hard disk recorder is used to connect the high-definition infrared camera, store and send images to the image recognition host. The intelligent local module sends the information collected by the multi-sensing device to the secondary intelligent inspection device.

[0027] The analysis application layer includes: image recognition host and secondary intelligent inspection device. The image recognition host uses a convolutional neural network model to detect the images collected and uploaded by the hard disk recorder. According to the different image contents of the inspection, the protection device status, switch handle status, pressure plate status, terminal block wiring, grounding copper plate, cable status, optical fiber status, spare optical fiber dust cap, cabinet blockage and cabinet foreign matter are identified, and the corresponding convolutional neural network models are trained respectively.

[0028] Under the condition of ensuring accuracy, the convolutional neural network model proposed in the present invention includes three convolutional layers and three pooling layers, wherein the convolutional layer adopts depth-separable convolution. As shown in Formula 1, it is the standard convolution formula:

[0029] K S ×K S ×C+C×Q (1)

[0030] K S is the size of the convolution kernel, C is the number of channels, and Q is the number of convolution kernels.

[0031] Compared with standard convolution, depthwise separable convolution has fewer parameters. As shown in Equation 2, depthwise separable convolution first considers the region and then the channel, thus achieving the separation of channel and region.

[0032] K S ×K s ×C×Q (2)

[0033] In the formula, K S is the size of the convolution kernel, C is the number of channels, Q is the number of convolution kernels, and the convolution layers in this model use K S The small convolution kernels are 2, 3, and 4, and the number of convolution kernels Q is 8, 16, and 16 respectively.

[0034] In addition, the downsampling ratio of the pooling layer is 2, so the image needs to go through K S = 2 depth convolution, normalization layer, activation layer, K S =1 point-by-point convolution, normalization layer, activation layer, and then pooling layer, stacked three times to get the final feature map. The reason for using a small convolution kernel is that it can reduce the amount of calculation, increase the model reasoning speed, and fully consider the different inspection project recognition tasks and the collected 800×600 resolution image samples. After many experiments, when the number of layers of the image feature extraction model is 6, the features extracted by the model have good detail information and semantic information. In addition, the number of samples input for each training is unified as 50, and the test accuracy of different inspection project models after training can be maintained above 96%.

[0035] The secondary intelligent inspection device uses the online monitoring data of the protection device and the image recognition results output by the image recognition host, and uses homology comparison and trend analysis to judge the operating status and change trend of the protection device, and automatically identify the explicit and implicit faults of the protection device. The possible fault location is determined by the alarm signal, and the self-propelled device moves to the abnormal equipment and possible fault location for image acquisition to assist the operation and maintenance personnel in troubleshooting.

[0036] Among them, the platform sharing layer includes the cloud platform and the control system. The dispatching data network sends relevant information to the dispatching end to provide data support for the advanced applications of the D5000 control system and the cloud platform.

[0037] Embodiment 1:

[0038] like Figure 2-4 As shown, an intelligent patrol system for protective devices based on multi-sensing technology includes a device perception layer, a network interconnection layer, an analysis application layer, and a platform sharing layer.

[0039] The data flow is mainly divided into three paths: the first path is the alarm information of the secondary equipment online monitoring and analysis system transmitted to the monitoring host and the secondary intelligent inspection device through the central switch of the station control layer; the second path is to collect images through high-definition and infrared cameras, and then store them in the hard disk recorder. The image recognition host obtains image data from the hard disk recorder for target detection, and the detection results will be sent to the secondary intelligent inspection device; the third path is that the intelligent local module containing temperature and humidity sensors and smoke detectors transmits the collected environmental information to the secondary intelligent inspection device. The secondary intelligent inspection device comprehensively utilizes multi-dimensional data, which is more conducive to detecting the device status and troubleshooting.

[0040] The convolutional neural network model in the image recognition host has fewer parameters and less computation, while ensuring accuracy, and the detection speed is greatly improved. In addition, it also includes a target detection model, which is based on Faster R-CNN. The detection algorithm involves an anchor mechanism. Usually, the number of anchor frames for the foreground is relatively small, and most of them are anchor frames for the background, which will lead to an imbalance in the foreground and background samples. Sample imbalance makes the model tend to favor types with a large number of samples, which will weaken the generalization ability of the model. For this reason, we introduced focal loss in the final classification stage, as shown in Formula 3. The principle is to assign weights to different categories of loss.

[0041] FL(p t )=-α t (1-p t ) γ log(p t ) (3)

[0042] As shown in the formula, α t It is used to adjust the category weight, γ is used to adjust the weights of samples with different detection difficulties, so that the model can pay attention to difficult samples faster, and t represents different categories.

[0043] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. An intelligent patrol system for protection devices based on multi-sensing technology, Features: include: Device perception layer, network interconnection layer, and analysis application layer; The equipment sensing layer includes: infrared cameras and multi-sensing devices. The infrared cameras and multi-sensing devices are installed on a self-propelled device. The self-propelled device patrols on the tracks inside and outside the secondary equipment cabinet to monitor the devices, circuits and environment outside the cabinet, and collect data required by the intelligent patrol system. The network interconnection layer includes: a central switch of the station control layer, a hard disk recorder, and an intelligent local module; after the central switch of the station control layer obtains the online monitoring data and alarm information of the protection device, it is sent to the secondary intelligent inspection device of the analysis application layer; the hard disk recorder is used to connect to the infrared camera, store and send images to the image recognition host; the intelligent local module sends the information collected by the multi-sensing device to the secondary intelligent inspection device; The analysis application layer includes: an image recognition host and a secondary intelligent inspection device; the image recognition host uses a convolutional neural network model to detect the images collected and uploaded by the hard disk recorder; for different image contents of the inspection, the protection device status, switch handle status, pressure plate status, terminal block wiring, grounding copper plate, cable status, optical fiber status, spare optical fiber dust cap, cabinet blockage and cabinet foreign matter are identified; the secondary intelligent inspection device uses the online monitoring data of the protection device and the image recognition results output by the image recognition host, and uses homology comparison and trend analysis to judge the operating status and change trend of the protection device, and automatically identifies the explicit and hidden faults of the protection device; the possible fault location determined by the alarm signal is used to move the self-propelled device to the abnormal equipment and the possible fault location for image acquisition, so as to assist the operation and maintenance personnel in troubleshooting; It also includes a platform sharing layer, which includes a cloud platform and a control system. The dispatching data network sends relevant information to the dispatching end to provide data support for advanced applications of the control system and the cloud platform; The multi-sensing device includes: temperature and humidity sensor, smoke detector; The convolutional neural network model includes three convolutional layers and three pooling layers; The convolution layer adopts depth-separable convolution, and the depth-separable convolution formula is as follows: K S ×K S ×C×Q Among them, K S is the size of the convolution kernel, C is the number of channels, Q is the number of convolution kernels, the convolution kernels of the three depthwise separable convolutions are 2, 3, and 4, and the number of convolution kernels are 8, 16, and 16, respectively; The downsampling ratio of the pooling layer is 2; The image is K S =2 depth convolution, normalization layer, activation layer, K S =1 point-by-point convolution, normalization layer, activation layer, pooling layer, K S =3 depth convolution, normalization layer, activation layer, K S =1 point-by-point convolution, normalization layer, activation layer, pooling layer, K S = 4 depth convolution, normalization layer, activation layer, K S =1 point-by-point convolution, normalization layer, activation layer, pooling layer to obtain the final feature map; The image recognition host also includes a target detection model, which is used to mark the image recognition results. The target detection model adopts a Faster r-cnn structure; The target detection model introduces focal loss in the classification stage, assigns weights to different categories of loss, and the focal loss calculation formula is as follows: ; in, is the probability of different categories t, Used to adjust the category weights, Used to adjust the weights of samples with different detection difficulty.