Intelligent identification model of substation equipment defects based on "RPA+AI" technology

Abstract

The present invention discloses an intelligent defect recognition model for substation equipment based on "RPA+AI" technology, including a high-definition video inspection system for substations, an RPA technology middle platform, an AI technology middle platform and a data processing and storage module. The high-definition video inspection system for substations uses a high-definition camera to obtain images of target equipment, the RPA technology middle platform is responsible for automatically classifying and processing images of target equipment, the AI ​​technology middle platform is responsible for processing and analyzing images classified by the RPA robot using artificial intelligence, the data processing and storage module processes the analysis results, and calls the target result disposal method from the material library to form an automatic output report. Under the traditional working mode, manual judgment needs to analyze the image status first and then analyze the equipment location and disposal method, which has the problems of slow speed and easy errors. The use of the present invention can quickly and accurately perform intelligent analysis of data, the defect analysis is accurate, the judgment speed is fast, the total inspection time can be effectively reduced, and the labor efficiency can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image intelligent recognition, and specifically to a substation equipment defect intelligent recognition model based on "RPA+AI" technology. Background Art

[0002] With the digital transformation and development of the power industry, the operation and maintenance of power grid equipment are gradually facing new challenges and opportunities. From the initial reliance on manual inspections of equipment defects to now being able to conduct inspections through high-definition video inspection systems, the current high-definition inspection system has a low level of intelligence. The pictures and videos taken by the high-definition inspection system of insulator cracks, discoloration of respirator silicone, opening and closing of switch cabinet pressure plates, and suspended foreign objects still need to be manually identified for equipment defects, and intelligent judgment is not possible. This requires extremely high work experience for the operation and maintenance personnel, and results in a large number of repetitive operations, with problems such as slow speed and many errors. In the context of promoting the "two substitutions" of substation operation and maintenance, namely, remote intelligent inspections replacing manual routine inspections and one-button sequential control replacing traditional switching operations, the intelligent recognition system in the inspection system still faces huge challenges.

[0003] To this end, how to use remote intelligent patrols to replace manual labor has become one of the technical problems that technicians in related technical fields need to solve. Summary of the invention

[0004] The technical problem to be solved by the present invention is that, in view of the deficiencies in the above-mentioned prior art, an intelligent recognition model for defects in substation equipment using the "RPA+AI" fusion technology is constructed. Based on the high-definition video inspection system, without changing the original system architecture, the "RPA+AI" fusion technology is adopted to use the RPA technology middle platform to classify the target images, and then the AI ​​technology middle platform uses artificial intelligence to intelligently identify defective equipment, thus solving the problem of defect identification of substation equipment for the front-line personnel of substation operation and maintenance, replacing manual work with remote intelligent patrols, and realizing efficient, precise and intelligent equipment defect identification.

[0005] To achieve the above purpose, the present invention provides the following solutions:

[0006] Substation HD video inspection system, which monitors the target equipment through HD cameras to obtain and display the target equipment images;

[0007] RPA robot module: The RPA robot module is responsible for automatically processing digital business data;

[0008] AI analysis module: The AI ​​analysis module is responsible for intelligent analysis of the data processed by the RPA robot module;

[0009] Data storage module: The data storage module is used to store digital business data and processing results.

[0010] Preferably, the visual operation interface can be accessed using a variety of devices such as browsers, APPs and technical platforms.

[0011] Preferably, the high-definition video inspection system converts the original image data generated by photography into a common image format of jpg or png for output, and the video data is output frame by frame into an image format of jpg or png for output.

[0012] Preferably, the RPA robot module can first divide the images transmitted by the RPA robot from the power station high-definition video inspection system into different time domains when simulating manual classification, so as to avoid idle system resources and improve the speed and accuracy of self-inspection.

[0013] Preferably, the artificial intelligence recognition system can set a target device according to the warning threshold of different defect facilities, such as color change. When the pictures obtained by the high-definition video inspection system are classified by the RPA robot and transmitted to the AI ​​technology center, the artificial intelligence recognition model will compare the pictures with the pictures in the sample library. When the set NBS (a color difference unit) value is exceeded, the system will issue a warning, so that the device can design different warning trigger conditions according to actual needs of different devices during use, making the recognition more sensitive and accurate.

[0014] Compared with the prior art, the present invention has the following beneficial effects:

[0015] (1) The present invention improves business processing efficiency: the RPA robot module can automatically complete a large amount of repetitive, low-value work, reducing the manual burden; the AI ​​analysis module can quickly and accurately perform intelligent analysis of data, improving analysis efficiency;

[0016] (2) The present invention can reduce the error rate: the RPA robot module can strictly perform tasks according to the rule engine and workflow, reducing human errors; the AI ​​analysis module can learn and predict based on a large amount of data, improving the accuracy and reliability of analysis;

[0017] (3) The present invention uses an artificial intelligence recognition system, and the database can be modified according to actual needs during use to increase the effectiveness and reliability of the artificial intelligence recognition model;

[0018] (4) The present invention can be modified using the original high-definition video inspection system of the substation and adapt to most operating systems and application platforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The RPA middle platform technology architecture;

[0020] Figure 2This is a diagram of equipment connection relationships for one implementation of a substation equipment defect intelligent identification model based on "RPA+AI" technology;

[0021] Figure 3 This is a workflow diagram of an implementation method of the intelligent identification model system for substation equipment defects based on "RPA+AI" technology. DETAILED DESCRIPTION

[0022] The technical solution of this patent is further described in detail below in conjunction with specific implementation methods.

[0023] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", and "set" should be understood in a broad sense, for example, they can be fixedly connected or set, or detachably connected or set, or integrally connected or set. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0024] The present invention consists of RPA technology center and AI technology center. The RPA technology center architecture complies with the application architecture management requirements of SG-ERP, strictly follows the principles of application integrity, horizontal integration, business drive and architecture flexibility, and is constructed using the "B / S+C / S" architecture. The design of a reasonable automation operation center application architecture ensures that the application fully covers the ability to reduce the burden on the automation operation center. The RPA center technology architecture is mainly divided into:

[0025] (1) Presentation layer: This layer is implemented using HTML, Element-UI, and VUE technologies to build a functional visualization operation interface that users access through a browser. It mainly hosts backend management applications and provides users with a simple, friendly, and easy-to-use operation interface.

[0026] (2) Control layer: responsible for system access control, input validation, and unified exception handling.

[0027] (3) Business layer: As the core layer of the platform, the business layer carries the core business of the entire system. The business layer is mainly divided into system services, basic services, and software robots.

[0028] (4) Data layer: MySQL is used to store structured data. The files generated by monitoring videos and execution results are uploaded to the management and dispatching center via HTTP and stored in the local file system of the management and control center.

[0029] (5) Infrastructure layer: provides basic guarantee for the normal operation of the system.

[0030] The AI ​​technology platform can provide company-level artificial intelligence application computing power, cloud-edge intelligent collaborative services, and business samples and intelligent models jointly built and shared by various professionals. By outputting AI capabilities such as intelligent video patrol, intelligent image analysis, intelligent behavior analysis, intelligent voice interaction, and intelligent text processing, it provides application support for various business scenarios. The AI ​​technology platform architecture includes the following 9 parts:

[0031] (1) Heterogeneous computing hardware layer: The underlying layer of the AI ​​platform is the heterogeneous computing hardware layer, including CPU and GPU, which can be deployed on different cloud platform environments.

[0032] (2) Scheduling layer: Cluster operating system and resource scheduling module, including K8S and Docker to solve the problems of environment inconsistency and resource pooling. The two-level resource management mechanism meets the hierarchical resource management needs of enterprise users.

[0033] (3) Framework layer: It is integrated with mainstream deep learning frameworks (Tensorflow, Paddlepaddle, Pytorch, Cafe, etc., or other frameworks supported by self-made Docker images) and machine learning frameworks (sklearn, XGBoost, Spark, MLlib algorithms). The administrator will regularly update and upgrade offline to keep the mainstream learning framework versions synchronized with the community.

[0034] (4) Storage layer: Integrates a private cloud storage solution that performs a lot of performance optimization for scenarios with massive amounts of small files; provides a standard S3 interface; also supports data synchronization, permission management, data lifecycle management, data encryption, etc.; and uses open source storage GlusterFs.

[0035] (5) Model production: Model production is carried out around data processing, model training, model management, and model prediction service processes; it supports five modeling modes, including notebook, visualization, automated modeling, production line modeling, and operational modeling, to meet the model development needs of users at different levels. The artificial intelligence platform provides parameter modification and custom code functions to help enterprises tailor their own models for rapid application.

[0036] (6) Prediction service: supports accessing the prediction service via URL, supports grayscale release, resource load balancing, etc.

[0037] (7) Model repository: supports the import of third-party models and model version management.

[0038] (8) User management: complete authentication and authorization schemes, and multi-user management.

[0039] (9) Operation and maintenance monitoring: comprehensive monitoring, anomaly detection, etc.

[0040] In summary, the intelligent identification model of substation equipment faults and defects based on the "RPA+AI" fusion technology makes full use of the advantages of the RPA technology middle platform and the AI ​​technology middle platform, combines the two technologies, and realizes the intelligent identification of station-side primary equipment faults and defects by substation operation and maintenance personnel, so that substation operation and maintenance personnel can efficiently and accurately remotely analyze and judge station-side equipment faults and defects.

[0041] In the traditional working mode, the manual judgment of equipment faults at the station end requires manual identification of data types, which has problems such as slow manual judgment and easy errors. At present, 1500 pieces of equipment data at the station end are collected remotely. According to the calculation of 20 seconds for a single material judgment (including defect model type and network delay, etc.), it takes 500 minutes to complete a fault judgment of equipment. By calculating the expression of the reduction of the total man-hours of annual inspections, substituting it into formula (1), it can be obtained that the total man-hours of inspection and maintenance can be reduced by about 3041 hours.

[0042]

[0043] Among them, represents the number of decreases in the total patrol and maintenance working hours, represents the type of patrol and maintenance. Here, patrol and maintenance are divided into four types according to camera network connectivity, no picture, blurred picture, and position movement. M represents the type of patrol, represents the number of annual patrol type i, represents the standard working hours for patrol type i operations, and represents the reduction ratio of patrol type operations after the use of "RPA+AI" fusion technology.

[0044] The intelligent identification application of fault defects in substation equipment based on the fusion technology of "RPA+AI" frees substation operation and maintenance personnel from the heavy work of image data analysis. The staff only needs to view the inspection results in the specified directory and selectively verify the intelligent identification results. More importantly, the application of this achievement facilitates the more timely and efficient discovery of hidden dangers and defects in the operation of station-side equipment, and can arrange more targeted power grid maintenance plans, which plays a very important role in supporting the safe operation of the power grid.

[0045] From the above, it can be concluded that the quality of the intelligent recognition model directly determines whether the power grid can operate safely and stably. The model, target equipment and database form a closely connected digital ecosystem. In this digital ecosystem, the high-definition video inspection system plays a core role in anomaly detection, and the target database stores the RPA technology platform and AI technology platform to be processed. It is worth noting that before the official operation, the three have completed strict cross-certification processing to ensure the security and reliability of data transmission.

[0046] When the target device reaches the self-check cycle, a patrol command is manually input, or a suspected abnormality is manually input, an abnormality detection request will be actively reported to the detection system. The abnormality detection request contains basic information such as the time, location, and scope of the abnormality, providing the detection system with preliminary abnormality clues.

[0047] After receiving the request, the system will immediately link to the high-definition video inspection system to obtain pictures from the target device database. The pictures are accompanied by interval numbers, and the numbers are accompanied by the location of the high-definition camera, shooting time, and shooting target information. By obtaining the target device pictures, they are transferred to the RPA technology center for classification and processing, and divided by device type. The classified pictures are grouped separately and saved in the sample library.

[0048] During the data acquisition process, the detection system fully utilizes the communication connection with the target database to ensure the efficiency and accuracy of data transmission. At the same time, since the cross-certification process has been completed before, the detection system can safely access the data in the target database without worrying about the risk of data leakage or tampering.

[0049] After obtaining the RPA business data to be processed, the system will input this data into the pre-designed artificial intelligence recognition model. The defect anomaly detection algorithm is one of the core components of the present invention. It is built based on the Paddlepaddle deep learning framework. The artificial intelligence recognition model is trained by Cafe and Pytorch to perform convolution operations on the device differences by sliding the convolution kernel on the image to extract different textures, edges and other features. The normal and defective device images can be encoded as supervision to enable it to identify large-scale target device image data to infer the complex device defect problems that exist. The XGBoost machine learning framework is used to enable the model to flexibly adapt to most application scenarios, and by being able to automatically analyze the defect anomalies in the target device images, give the corresponding defect labels, and output the device image sets with identified defects to the data processing and storage module.

[0050] The data processing and storage module receives a set of defective equipment images, determines the location of the defective equipment by recording the number on the image source, calls the target equipment result disposal method from the material library, forms an automatic output report, and transmits the automatic output report to the user end. After manual inspection, the list of defective equipment in the system is updated after another inspection.

[0051] 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 in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1The steps for the functions specified in one or more boxes.

[0052] The above-mentioned embodiments are only preferred embodiments of the present invention and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and substitutions made by technicians in this field on the basis of the present invention shall fall within the scope of protection required by the present invention.

Claims

1. The intelligent identification model of substation equipment defects based on "RPA+AI" technology is characterized by: include: The substation high-definition video inspection system includes a high-definition camera, a camera controller, a 5G communication module, a picture collection server and a server communication domain. The camera controller controls the high-definition camera to monitor the target device to obtain the target device image, and the 5G communication module is used to transmit the target device image to the picture collection server. The picture collection server can be set in different server communication domains; The RPA technology platform includes a presentation layer, a control layer, a business layer, a data layer, and an infrastructure layer. The presentation layer uses HTML, Element-UI, and VUE technologies to provide a visual operation interface, allowing users to access it through a browser; the control layer provides system access control, input verification, and unified exception handling; the business layer provides system services, basic services, and RPA robot modules. The RPA robot module simulates manual process operations, identifies the target device images captured by high-definition cameras, determines whether there is a target device in the image, annotates the image with the identified image information, classifies it into different device image groups according to the annotations, and transmits the image groups to the data layer; the data layer uses Mysql to store structured data, and the infrastructure layer provides basic guarantees for the normal operation of the system; The AI ​​technology middle platform includes heterogeneous computing hardware layer, scheduling layer, framework layer, storage layer, model production, prediction service, model warehouse, user management, and operation and maintenance monitoring. The heterogeneous computing hardware layer includes CPU and GPU, which can be deployed on different cloud platform environments. The scheduling layer includes K8S and Docker to solve the problems of inconsistent environments and resource pooling. The framework layer is equipped with an artificial intelligence recognition model integrated with the mainstream deep learning framework and machine learning framework, which is responsible for intelligent analysis of target device images; the storage layer is equipped with an integrated private cloud storage solution; the model production, including data processing, model training, model management, and model prediction, provides a platform for parameter modification and custom code, and can adjust the model according to different needs; the model warehouse supports the import of third-party models; the user management can complete different user authentication, authentication solutions and multi-user management; the operation and maintenance monitoring regularly performs system self-inspection and anomaly detection processing; The data processing and storage module is used to store the processing results of the AI ​​analysis module, summarize the artificial intelligence processing results, and call the target device result processing method from the material library to form an automatic output report. The information transmission between the RPA technology center and the AI ​​technology center is as follows: S1: The substation high-definition video inspection system takes pictures of target equipment according to business processing requirements and transmits them to the TPA technology center; S2: Based on the received target device images, the robot module in the TPA technology center simulates the manual process operation, recognizes the target device images obtained by the high-definition camera, annotates the images based on the recognized image information, and organizes them into different device image groups according to the annotated classification; S3: The RPA robot module passes the processed image groups of different devices to the AI ​​technology middle platform data layer; S4: The AI ​​technology center retrieves the data in the data layer and compares it with the standard control group stored in the sample library to obtain the target device defect image group, and associates and annotates the target device defect image group with the defective device and stores it in the storage layer; S5: The different device image groups of the storage layer are retrieved to the data processing and storage module; S5: The TPA technology platform statistically checks the automated process of defective devices in the data until the defective devices are manually removed from the target device defect graph group.

2. According to claim 1, the intelligent identification model for substation equipment defects based on "RPA+AI" technology is characterized by: The high-definition camera, picture collection server, camera controller and 5G communication module are connected, and several picture collection servers are equipped with 5G modules and can communicate with each other.

3. According to claim 2, the intelligent identification model for substation equipment defects based on "RPA+AI" technology is characterized by: The substation high-definition video inspection system adopts an image intelligent recognition and transmission method. The substation high-definition video inspection system image intelligent recognition and transmission method sets image interval numbers. When taking a photo of the device, the image is automatically attached with an interval number. The interval number can be used to represent the arrangement of the image collection server in the communication domain. When constructing a server communication domain, the image collection servers in different communication domains are connected for communication, and each communication domain is provided with a communication domain self-inspection system with different periods.

4. According to claim 3, the intelligent identification model for substation equipment defects based on "RPA+AI" technology is characterized by: The artificial intelligence recognition model can divide the target image into a target area and a background area; The influence factors of the background area are eliminated, and the target area and the background area are distinguished in the target image.

5. According to claim 4, the intelligent identification model for substation equipment defects based on "RPA+AI" technology is characterized by: The artificial intelligence recognition module extracts features from each target image sample, obtains the target image sample to obtain a feature group, and compares it with the standard control group data stored in the database to obtain the target device defects.

6. According to claim 5, the intelligent identification model for substation equipment defects based on "RPA+AI" technology is characterized by: Through the artificial intelligence recognition model, deep learning is performed, training is performed by inputting different feature groups, storing new equipment defect feature groups, and comparing AI components with new equipment defect images.

7. According to claim 6, the intelligent identification model for substation equipment defects based on "RPA+AI" technology is characterized by: The defect focus inspection group within the communication domain is established by the number of different defect types in the defect group in different communication domains, and the periodic inspection frequency is automatically shortened.

8. The method for using the intelligent identification model for substation equipment defects based on "RPA+AI" technology is characterized in that: Using the substation equipment defect intelligent identification model based on "RPA+AI" technology as claimed in any one of claims 1 to 7, the steps are as follows: S1: The user enters the operation interface through the browser and issues inspection instructions or sets the system self-inspection cycle; S2: The substation high-definition video inspection system takes pictures of target equipment according to business processing requirements and transmits them to the TPA technical center; S3: Based on the received target device images, the robot module in the TPA technology center simulates the manual process operation, recognizes the target device images obtained by the high-definition camera, annotates the images based on the recognized image information, and organizes them into different device image groups according to the annotated classification; S4: The RPA robot module passes the processed image groups of different devices to the AI ​​analysis module; S5: The AI ​​technology center compares the input image with the standard control group stored in the sample library according to the artificial intelligence recognition model in the AI ​​technology center to obtain the target device defect image group; S6: The AI ​​technology center transfers the processed image groups of different devices to the data processing and storage module; S7: the data processing and storage module summarizes the information of the obtained target device defect group, and calls the target device result disposal method from the material library to form an automatic output report, and outputs the automatic output report to the user end; S8: The user end obtains the automatic output report and conducts manual inspection of defective equipment.

Images