Remote control inspection method based on AI operation and power plant supervision system

Through the AI-driven remote control inspection method, the AR intelligent head ring and lightweight AI model are used to quickly determine device abnormalities, and communicate with remote experts in real time, solving the problem of time delay in equipment inspection, improving the fault handling efficiency and Equipment operation efficiency.

CN118938733APending Publication Date: 2024-11-12NATIONAL ENERGY GROUP HEILONGJIANG ELECTRIC POWER CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410925047.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-11
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

In the prior art, during equipment inspection, the time from discovering an abnormality to notifying relevant personnel to deal with it is long, resulting in low fault handling efficiency.

Method used

Using AI-based remote control inspection methods, capture and process on-site data through AR intelligent head rings, use lightweight AI models to determine whether the device is abnormal, and communicate and connect with remote experts through real-time video communication tools to provide decision-making support.

Benefits of technology

It effectively shortens the time from abnormal discovery to processing, improves fault processing efficiency, reduces communication costs, breaks regional restrictions, and improves the efficiency of problem solving.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118938733A_ABST
    Figure CN118938733A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of remote inspection, and particularly discloses a remote control inspection method based on AI operation and a power plant supervision system. The method mainly comprises the following steps: receiving inspection data transmitted by an AR intelligent head ring, wherein the inspection data is data after field processing; receiving help seeking information which is transmitted by the AR intelligent head ring and is based on the inspection data, and judging the professional field of the inspection data; a remote expert is matched based on the professional field of inspection data, and the remote expert and the AR intelligent head ring are in instant communication and connection through a real-time video communication tool; and inputting the inspection data into the trained LSTM model to obtain abnormal reason analysis of the abnormal equipment, and sending the abnormal reason analysis to a remote expert to provide decision support. According to the invention, the abnormity can be rapidly judged locally and the inspection data can be transmitted, so that the remote expert and the AR intelligent head ring can carry out instant communication and connection, and the fault processing efficiency is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of remote inspection technology, and in particular to a remote control inspection method based on AI operation and a remote control power plant supervision system based on AI operation. Background Art

[0002] On-site inspections of power plant equipment typically involve a variety of technologies, including manual inspections, recording systems, sensor monitoring, video surveillance, data analysis, and mobile applications. Operations and maintenance personnel ensure proper equipment operation through on-site inspections, recording inspection results, sensor monitoring, and video surveillance.

[0003] Once an equipment anomaly occurs, a series of steps are required to notify the relevant personnel, including the inspector discovering the anomaly, recording the anomaly, notifying the on-duty personnel or supervisor, and the on-duty personnel or supervisor notifying the maintenance personnel. This series of transmission and communication links may take a long time, resulting in delays in the arrival of relevant personnel on site and the delay in the start of anomaly handling. Summary of the Invention

[0004] This application solves the problem in the prior art of equipment inspection in which it takes a long time from discovering an abnormality in an abnormal device to notifying relevant personnel to handle the abnormality by providing a remote control inspection method and power plant supervision system based on AI operations. It can quickly judge the abnormality locally and transmit inspection data, allowing remote experts to communicate and connect with the AR smart headband in real time, thereby improving the efficiency of fault handling.

[0005] The present application provides a remote control inspection method based on AI operation, which includes the following steps: receiving inspection data transmitted by an AR smart headband, where the inspection data is data processed on site; receiving a help message based on the inspection data transmitted by the AR smart headband, and judging the professional field of the inspection data; matching remote experts based on the professional field of the inspection data, and enabling the remote experts to communicate and connect with the AR smart headband in real time through real-time video communication tools; inputting the inspection data into a trained LSTM model to obtain an abnormal cause analysis of the abnormal equipment, and sending the abnormal cause analysis to the remote experts to provide decision support.

[0006] Furthermore, the on-site processing method includes: the AR smart headband captures a preset device and saves an on-site picture of the captured device, collects on-site data of the captured device and marks the on-site data; uses a feature extraction algorithm to extract features from the on-site data; inputs the extracted features into a lightweight AI model to obtain an output result, and the output result is the probability of normal and abnormal; based on the output result, the on-site data is classified into normal data and abnormal data. If the output result is abnormal data, the on-site data of the device and the picture of the device are transmitted to the platform in sequence. If the output result is normal data, only the on-site data of the device is transmitted to the platform.

[0007] Furthermore, the lightweight AI model is a convolutional neural network model, which includes a convolution layer, a pooling layer and a fully connected layer. Different feature patterns are detected through multiple convolution kernels. Each convolution kernel performs a convolution operation on the input data to generate a new set of feature maps. Each convolution layer can contain multiple convolution kernels, and each convolution kernel produces a feature map. The result of the convolution operation is activated by a nonlinear activation function. The pooling layer is used to reduce the spatial size of the convolution layer output and reduce the number of parameters while retaining important features. Maximum pooling is used to extract the maximum value of each area from the input feature map. The fully connected layer compresses the output of the convolution layer and the pooling layer into a one-dimensional vector, and classifies or detects anomalies through a series of neurons.

[0008] Furthermore, if the abnormal data is greater than a preset threshold, the output result is determined to be abnormal data; if the abnormal data is less than the threshold, the output result is determined to be normal data.

[0009] Furthermore, the specific method for capturing the preset device is: obtaining a real-time image of the scene and preprocessing the real-time image of the scene; calculating the color histogram of the image to describe the color characteristics of the device, using the edge detection algorithm to identify the edge information in the image to describe the contour characteristics of the device, and using the corner detection algorithm to detect the corners in the image to describe the local characteristics of the device; using the target detection algorithm to identify the device in the image, and determine whether there is a device that needs to be inspected in the image. If there is a device that needs to be inspected in the image, the identification and marking results are displayed in real time on the display of the AR smart headband to achieve the capture of the preset device.

[0010] Furthermore, the specific method for saving the on-site picture of the captured device is: when the device is captured, it is identified and determined whether the device has a mark. If there is no mark, the device is marked as having been captured, and the capture time is recorded and the on-site picture is saved. If there is a mark, the time interval between the current time and the last time the device was captured is compared. If the time interval is less than the set time limit, the on-site picture is no longer saved.

[0011] Furthermore, the professional field of the inspection data is determined based on the label of the field data.

[0012] Furthermore, matching remote experts based on the professional field of the inspection data is specifically as follows: when receiving a request for help, a suitable remote expert is screened from the database according to the professional field of the inspection data; a connection request is generated, including a summary of the request for help and an expert matching result; the connection request is sent to the selected remote expert through the communication system of the AR smart headband, and after the remote expert accepts the request, a video call connection is established through the AR smart headband and the remote expert's device.

[0013] Furthermore, the LSTM model includes an input layer, an LSTM layer and an output layer. The input layer receives inspection data. The LSTM layer includes an input gate, a forget gate, a cell state and an output gate. The cell state is controlled by the input gate, the forget gate and the output gate and is used to store and transmit information. The input gate is responsible for controlling the input of new information. The forget gate determines which information in the cell state needs to be forgotten. The output layer makes a final prediction or classification based on the output of the LSTM layer to obtain an analysis of the cause of the abnormality.

[0014] In the second aspect, the present application provides a remote-controlled power plant supervision system based on AI operations, which adopts a remote-controlled inspection method based on AI operations in the first aspect, which includes: a data receiving module, a data analysis module, an expert matching module and a decision support module.

[0015] The data receiving module is used to receive the inspection data transmitted by the AR smart headband. The inspection data is data that has been processed on site.

[0016] The data analysis module is used to receive the help information based on the inspection data transmitted by the AR smart headband and determine the professional field of the inspection data;

[0017] The expert matching module is used to match remote experts based on the professional fields of the inspection data, and enable the remote experts to communicate and connect with the AR smart headband in real time through real-time video communication tools;

[0018] The decision support module is used to input the inspection data into the trained LSTM model to obtain the abnormal cause analysis of the abnormal equipment and send the abnormal cause analysis to the remote expert to provide decision support.

[0019] On the third aspect, the present application provides a remote control power plant monitoring device based on AI operation, including a memory and a processor.

[0020] The memory is used to store computer programs; the processor is used to implement the steps of the remote control inspection method based on AI operations as the first aspect when executing the computer program.

[0021] The technical solution provided by this application has at least the following technical effects or advantages:

[0022] The design uses AR smart headband to determine whether the equipment is abnormal on site and transmit inspection data, allowing remote experts to communicate and connect with the AR smart headband in real time to achieve remote guidance. This effectively solves the problem of long time from discovering abnormal equipment to notifying relevant personnel to handle the abnormality during equipment inspection. It effectively reduces data transmission time, breaks geographical restrictions, reduces communication costs, and improves problem-solving efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is the overall flow chart of the remote control inspection method based on AI operation in this application;

[0024] Figure 2 This is a flowchart for this application platform to determine abnormal devices. DETAILED DESCRIPTION

[0025] This application proposes a method of using an AR smart headband to determine whether the equipment is abnormal on site and transmit inspection data, so that remote experts can communicate and connect with the AR smart headband in real time to achieve remote guidance. It can effectively optimize the fault handling process, improve maintenance efficiency, reduce downtime, and ultimately improve the operating efficiency of the equipment and user satisfaction.

[0026] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0027] Example 1:

[0028] like Figure 1-Figure 2 As shown, this embodiment provides a remote control inspection method based on AI operation, which includes the following steps:

[0029] S100. Receive inspection data transmitted by the AR smart headband, where the data at the inspection site includes the AR smart headband.

[0030] Inspection data is data that has been processed on-site. The on-site processing method includes:

[0031] S110. The AR smart headband captures a preset device and saves a live picture of the captured device, collects live data of the captured device and marks the live data.

[0032] For equipment that needs inspection, the AR smart headband can capture these devices during the inspection and save the on-site pictures at the same time. The sensors on the AR smart headband can collect the vibration, temperature and other data of the equipment in real time, and use the embedded processor on the headband to perform real-time data processing and analysis. Through the predictive ability of the lightweight AI model, it can detect possible abnormal vibration frequency and temperature of the equipment. At the same time, the AR smart headband will temporarily record the on-site pictures of the captured equipment. If the equipment is abnormal, the on-site data will be marked. Different marks correspond to equipment in different professional fields.

[0033] The specific method to capture the preset device is:

[0034] S111. Acquire real-time images on site and pre-process the real-time images on site.

[0035] Use the camera function of the AR smart headband to capture image data of on-site equipment in real time, and then pre-process the image data, including saving the real-time captured images to memory or temporary files, performing image recognition to detect the location and characteristics of the equipment, marking, quality adjustment, cropping, denoising and enhancing the images to improve clarity and accuracy.

[0036] S112. Calculate the color histogram of the image to describe the color features of the device, use an edge detection algorithm to identify edge information in the image to describe the contour features of the device, and use a corner detection algorithm to detect corners in the image to describe the local features of the device.

[0037] When extracting color features, the image is converted to the HSV color space, the HSV image is segmented into separate channels, and a histogram is calculated for each channel. The color range is divided into several intervals (bins), and the histogram of each channel is normalized to compare the color distribution of different images. When extracting edge features, the image is converted to a grayscale image, and an edge detection algorithm (such as the Canny operator) is used to detect edges in the image. The detected edges are further processed, such as non-maximum suppression and edge connection, to obtain a continuous edge contour. After the image is converted to a grayscale image, a corner detection algorithm (such as Harris corner detection) is used to detect corners in the image. The detected corners are screened and located to obtain accurate corner positions, thereby obtaining corner features.

[0038] S113. Use the target detection algorithm to identify the devices in the image and determine whether there are devices that need to be inspected in the image. If there are devices that need to be inspected in the image, the recognition and marking results are displayed in real time on the display of the AR smart headband to capture the preset devices.

[0039] SSD can be used for device identification. The pre-trained SSD model weights are downloaded. Before identifying the device, it is necessary to ensure that the model includes the device category that needs to be identified, or fine-tune it to adapt to the specific device identification task. The preprocessed image is input into the SSD model, and the detection result is obtained through the forward propagation of the model. The detection result includes the detected target category, location box (bounding box) and confidence level and other information. For each detected target, determine whether its category is a device that needs to be inspected. This can be determined based on the category label. For devices that need to be inspected, obtain the coordinate information of its location box, including the pixel coordinates of the upper left corner and the lower right corner. A bounding box can be drawn on the original image based on the coordinate information of the location box to identify the location of the device. The identified device that needs to be inspected and its location information are output. The location box can be marked on the image, and the location information can be saved for subsequent processing.

[0040] The specific method for saving the on-site picture of the captured device is: when the device is captured, it is identified and whether the device has a mark. If there is no mark, the device is marked as having been captured, and the capture time is recorded and the on-site picture is saved. If there is a mark, the time interval between the current time and the last time the device was captured is compared. If the time interval is less than the set time limit, the on-site picture is no longer saved.

[0041] This approach effectively avoids data redundancy, conserves storage resources, and improves data management efficiency. By comparing the current time with the last capture time and setting a time limit to determine whether to save live images, it quickly filters out devices that require recapture, ensuring data timeliness and accuracy. Users can customize time limits based on their actual needs, making the system more flexible and adaptable to different needs, reducing operational errors and improving system stability and data security. This design approach optimizes the data capture process, enhancing overall system performance and data management quality.

[0042] S120. Use a feature extraction algorithm to extract features from the field data.

[0043] Before feature extraction, the processor will preprocess the collected raw data, including denoising, filtering, data calibration, etc., and use feature extraction algorithms in the field of computer vision, such as SIFT, SURF, HOG, etc., to extract key feature information from the image. If feature extraction is performed on speech data, algorithms such as MFCC (Mel-Frequency Cepstral Coefficients) are used.

[0044] S130. Input the extracted features into the lightweight AI model to obtain the output results, which are the probabilities of normal and abnormal.

[0045] The extracted feature data is input into the AI ​​algorithm model for inference. Using a lightweight model, based on the model's inference results, it can output results such as recognized objects, text conversion of speech, sentiment analysis, or generate corresponding alarms and feedback information.

[0046] Lightweight AI models can use a convolutional neural network architecture, consisting of convolutional layers, pooling layers, and fully connected layers. Convolution and pooling operations extract spatial features of the input data, and finally, fully connected layers perform classification or anomaly detection. Convolutional layers are used to extract features from the input data, using multiple convolution kernels to detect different characteristic patterns. Each convolution kernel performs a convolution operation on the input data, generating a new set of feature maps. Each convolution layer can contain multiple convolution kernels, each producing a feature map. The results of the convolution operation are activated using a nonlinear activation function to increase the nonlinear capabilities of the network. Pooling layers are used to reduce the spatial size of the convolutional layer output, reducing the number of parameters while retaining important features. Max pooling is used to extract the maximum value of each region in the input feature map to reduce the spatial dimensionality. Fully connected layers compress the outputs of the convolutional and pooling layers into a one-dimensional vector and pass it through a series of neurons for classification or anomaly detection.

[0047] The model formula can be expressed as:

[0048] Z l =f(W l *X+b l )

[0049] P l =MaxPool(Z l )

[0050] Y=σ1(W fc F+b fc )

[0051] Where X represents the input data, X is a vector of input samples composed of data from different devices, X = [X1, X2, ..., X n ],W l represents the convolution kernel, b l represents the bias term, * represents the convolution operation, f represents the activation function, the activation function can be ReLU, MaxPool represents the maximum pooling operation, W fc represents the weight matrix of the fully connected layer, b fc Represents the bias term, σ1 represents the activation function, which can be a sigmoid function. After the input data X passes through the lth layer of convolution operation, the feature map Z is obtained. l , feature map Z l After the maximum pooling operation, the pooling result P is obtained l , where the feature map Zl After the maximum pooling operation, the pooling result P is obtained l , the features after convolution and pooling are flattened into vector F, and finally the fully connected layer output Y is obtained.

[0052] The loss function can be:

[0053]

[0054] In the formula, m represents the number of samples, y (i) represents the true label, Represents the label predicted by the model.

[0055] After the output of the fully connected layer, the softmax function is applied to convert the output into a probability distribution of each category for classification prediction and loss calculation. The softmax function converts the output into a probability distribution so that the sum of the probability values ​​of all categories is 1.

[0056] S140. Based on the output result, the on-site data is classified into normal data and abnormal data. If the output result is abnormal data, the on-site data of the device and the image of the device saved in S110 are sequentially transmitted to the platform. If the output result has no abnormal probability, only the on-site data of the device is transmitted to the platform. If the abnormal data is greater than a preset threshold, the output result is determined to be abnormal data; if the abnormal data is less than the threshold, the output result is determined to be normal data.

[0057] By accurately classifying field data into normal and abnormal based on the output results, the system can effectively save resources and improve response speed. The image of the equipment is only transmitted when an anomaly is detected, saving transmission bandwidth and storage space. The system can quickly identify and handle abnormal situations, ensuring that staff can take necessary measures in a timely manner, further improving overall efficiency and responsiveness.

[0058] Using edge computing technology, AI algorithms are deployed on the AR smart headband for rapid local processing, reducing remote transmission delays. Furthermore, since information in normal conditions is typically relatively stable and repetitive and can be described through simple records, information in abnormal conditions can be more complex and valuable. Therefore, detailed recording of information in abnormal conditions can better support fault analysis and problem solving. This can help the platform better understand the performance and characteristics of devices under abnormal conditions, focusing on abnormal conditions and improving the efficiency and accuracy of fault diagnosis. This facilitates fault analysis and trend analysis, helping to predict and avoid similar abnormal conditions in the future and improve device reliability and stability. Furthermore, recording detailed information under abnormal conditions provides historical data support for subsequent data analysis and decision-making, helping to better understand the operating status and performance of the device.

[0059] S200. Receive a request for help based on the inspection data transmitted by the AR smart headband, and determine the professional field of the inspection data.

[0060] After determining the professional field of the inspection data, the equipment can be classified according to the professional field, which can ensure the orderly management of equipment information and the standardization of processing procedures. At the same time, problems can be quickly located, resources can be effectively allocated, potential risks can be prevented, and support can be provided for future decision-making based on the accumulated data, ultimately achieving improved user satisfaction.

[0061] The professional field of the inspection data is determined based on the tag of the field data.

[0062] The AR smart headband will mark the professional field of each device, and record the mark of the professional field of the device in the on-site data. Since the AR smart headband transmits the on-site data and pictures of the abnormal device in sequence, the platform can identify the professional field of the abnormal device at the first time so that corresponding measures can be taken immediately, which helps to improve the response speed and efficiency, and ensure that problems in specific professional fields can be handled and resolved in a timely manner. At the same time, it can effectively optimize the fault handling process, improve maintenance efficiency, reduce downtime, and ultimately improve the operating efficiency of the equipment and user satisfaction.

[0063] S300. Match remote experts based on the professional field of the inspection data, and enable the remote experts to communicate and connect with the AR smart headband in real time through real-time video communication tools.

[0064] This not only helps on-site personnel understand historical equipment issues and trends, but also provides valuable reference data for remote experts. When on-site personnel are unable to resolve an equipment anomaly, remote experts can use historical anomaly data to determine the cause, providing faster and more accurate guidance for on-site repairs. Remote expert guidance avoids the time and cost of in-person on-site visits, reducing maintenance and troubleshooting expenses. Furthermore, the real-time data display enables remote experts to more accurately diagnose and resolve issues, improving work efficiency and accuracy, and providing strong support for equipment maintenance and troubleshooting.

[0065] Matching remote experts based on the professional field of the inspection data, and enabling the remote experts to communicate and connect with the AR smart headband in real time through real-time video communication tools is specifically as follows: Matching remote experts based on the professional field of the inspection data is specifically as follows: when receiving a help message, screening suitable remote experts from the database according to the professional field of the inspection data; generating a connection request, including a summary of the help message and the expert matching result; sending the connection request to the selected remote expert through the communication system of the AR smart headband, and after the remote expert accepts the request, establishing a video call connection through the AR smart headband and the remote expert's device.

[0066] By automatically matching experts in the corresponding fields, users do not need to search on their own, thus saving time and energy. After receiving the request for help, the system will screen suitable remote experts from the database based on the professional field of the inspection data and generate a connection request, which includes a summary of the request for help and the expert matching results. The remote expert can understand the abnormal situation before the connection is made and prepare targeted solutions, improving the efficiency and accuracy of problem solving. Through the communication system of the AR smart headband, the system can quickly send the connection request to the selected remote expert, realizing instant communication and rapid advancement of problem solving.

[0067] S400. Input the inspection data into the trained LSTM model to obtain the abnormal cause analysis of the abnormal equipment, and send the abnormal cause analysis to the remote expert to provide decision support.

[0068] The LSTM model includes an input layer, an LSTM layer, and an output layer. The input layer receives inspection data. The LSTM layer includes an input gate, a forget gate, a cell state, and an output gate. The cell state is controlled by the input gate, the forget gate, and the output gate to store and transmit information. The input gate is responsible for controlling the input of new information. The forget gate determines which information in the cell state needs to be forgotten. The output layer makes a final prediction or classification based on the output of the LSTM layer to obtain an analysis of the cause of the abnormality.

[0069] The model formula can be expressed as:

[0070] i t =σ2(W i ·[H t-1 ,X t ]+b i )

[0071] f t =σ2(W if ·[H t-1 ,X t ]+b f )

[0072] C t =f t ·C t-1 +i t tanh(W ic ·[H t-1 ,X t ]+b c )

[0073] o t =σ2(W io ·[H t-1 ,C t ]+b o)

[0074] H t =o t tanh(C t )

[0075] Where, X t =[x1 t ,x2 t ,…,xn t ], represents the input data at time t, x1 t ,x2 t ,…,xn t They are different data of abnormal devices, i t represents the input gate, f t represents the forget gate, C t Represents cell state update, o t Represents the output gate, H t Represents hidden state update, W i 、W if 、W ic and W io represents the weight matrix, b i 、b f 、b c and b o represents the bias term, σ2 represents the Sigmoid activation function, and tanh represents the hyperbolic tangent activation function.

[0076] Analysis of H t and C t The key values ​​in, combined with the input data X t , can infer key sensor readings that may cause anomalies, thereby analyzing the causes of the anomalies and helping them make scientific decisions. These advantages help remote experts more accurately and efficiently determine the causes of anomalies, improving the quality and efficiency of patent application processing.

[0077] Example 2:

[0078] This embodiment provides a remote-controlled power plant supervision system based on AI operations, which adopts the remote-controlled inspection method based on AI operations in Example 1, and includes: a data receiving module, a data analysis module, an expert matching module and a decision support module.

[0079] The data receiving module is used to receive the inspection data transmitted by the AR smart headband. The inspection data is data that has been processed on site.

[0080] The data analysis module is used to receive the help information based on the inspection data transmitted by the AR smart headband, and determine the professional field of the inspection data.

[0081] The expert matching module is used to match remote experts based on the professional field of the inspection data, and enable remote experts to communicate and connect with the AR smart headband in real time through real-time video communication tools.

[0082] The decision support module is used to input the inspection data into the trained LSTM model to obtain the abnormal cause analysis of the abnormal equipment and send the abnormal cause analysis to the remote expert to provide decision support.

[0083] This embodiment has all the advantages of the remote control inspection method based on AI operations, and can automatically implement all steps of the remote control inspection method based on AI operations.

[0084] Example 3

[0085] This embodiment provides a remote control power plant monitoring device based on AI operations, including a memory and a processor.

[0086] The memory is used to store computer programs; the processor is used to implement the steps of the remote control inspection method based on AI operations in Example 1 when executing the computer program.

[0087] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely 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 magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0088] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The present invention is described in detail below. ...

Claims

1. A remote control inspection method based on AI operation, It is characterized in that It includes the following steps: Receive inspection data transmitted by the AR smart headband, which is data processed on site; receiving a request for help based on the inspection data transmitted by the AR smart headband, and determining a professional field of the inspection data; Matching remote experts based on the professional field of the inspection data, and enabling the remote experts to communicate and connect with the AR smart headband instantly through real-time video communication tools; The inspection data is input into the trained LSTM model to obtain the abnormal cause analysis of the abnormal equipment, and the abnormal cause analysis is sent to the remote experts to provide decision support.

2. The remote control inspection method based on AI operation as claimed in claim 1, It is characterized in that The on-site treatment method comprises: The AR smart headband captures the preset device and saves the live picture of the captured device, collects the live data of the captured device and marks the live data; Use feature extraction algorithm to extract features from field data; The extracted features are input into the lightweight AI model to obtain the output results, which are the probabilities of normality and abnormality; Based on the output results, the field data is classified into normal data and abnormal data. If the output result is abnormal data, the field data of the device and the picture of the device are transmitted to the platform in sequence. If the output result is normal data, only the field data of the device is transmitted to the platform.

3. The remote control inspection method based on AI operation as claimed in claim 2, It is characterized in that The lightweight AI model is a convolutional neural network model, which includes a convolution layer, a pooling layer and a fully connected layer. Different feature patterns are detected through multiple convolution kernels. Each convolution kernel performs a convolution operation on the input data to generate a new set of feature maps. Each convolution layer can contain multiple convolution kernels, and each convolution kernel produces a feature map. The result of the convolution operation is activated by a nonlinear activation function. The pooling layer is used to reduce the spatial size of the convolution layer output and reduce the number of parameters while retaining important features. Maximum pooling is used to extract the maximum value of each area from the input feature map. The fully connected layer compresses the output of the convolution layer and the pooling layer into a one-dimensional vector, and classifies or detects anomalies through a series of neurons.

4. The remote control inspection method based on AI operation as claimed in claim 2, It is characterized in that If the abnormal data is greater than a preset threshold, the output result is determined to be abnormal data; if the abnormal data is less than the threshold, the output result is determined to be normal data.

5. The remote control inspection method based on AI operation as claimed in claim 2, It is characterized in that The specific method to capture the preset device is: Acquire on-site real-time images and pre-process the on-site real-time images; Calculate the color histogram of the image to describe the color characteristics of the device, use the edge detection algorithm to identify the edge information in the image to describe the contour characteristics of the device, and use the corner detection algorithm to detect the corner points in the image to describe the local characteristics of the device; Use the target detection algorithm to identify the devices in the image and determine whether there are devices that need to be inspected in the image. If there are devices that need to be inspected in the image, the recognition and marking results are displayed in real time on the display of the AR smart headband to capture the preset devices.

6. The remote control inspection method based on AI operation as claimed in claim 5, It is characterized in that The specific method for saving the on-site picture of the captured device is: when the device is captured, it is identified and determined whether the device has a mark. If there is no mark, the device is marked as having been captured, and the capture time is recorded and the on-site picture is saved. If there is a mark, the time interval between the current time and the last time the device was captured is compared. If the time interval is less than the set time limit, the on-site picture is no longer saved.

7. The remote control inspection method based on AI operation as claimed in claim 2, It is characterized in that The professional field of the inspection data is determined based on the tag of the field data.

8. The remote control inspection method based on AI operation as claimed in claim 1, It is characterized in that The specific matching of remote experts based on the professional field of the inspection data is: When receiving a help request, select appropriate remote experts from the database according to the professional field of the inspection data; generate a connection request, including a summary of the help request and expert matching results; Through the communication system of the AR smart headband, the connection request is sent to the selected remote expert. After the remote expert accepts the request, a video call connection is established through the AR smart headband and the remote expert's device.

9. The remote control inspection method based on AI operation as claimed in claim 1, It is characterized in that The LSTM model includes an input layer, an LSTM layer and an output layer. The input layer receives inspection data. The LSTM layer includes an input gate, a forget gate, a cell state and an output gate. The cell state is controlled by the input gate, the forget gate and the output gate for storing and transmitting information. The input gate is responsible for controlling the input of new information. The forget gate determines which information in the cell state needs to be forgotten. The output layer performs a final prediction or classification based on the output of the LSTM layer to obtain an analysis of the cause of the abnormality.

10. A remote control power plant monitoring system based on AI operation, which adopts a remote control inspection method based on AI operation as described in claims 1-9, wherein include: Data receiving module: It is used to receive the inspection data transmitted by the AR smart headband. The inspection data is data that has been processed on site. Data analysis module: used to receive the help information based on the inspection data transmitted by the AR smart headband, and determine the professional field of the inspection data; Expert matching module: it is used to match remote experts based on the professional field of the inspection data, and enable remote experts to communicate and connect with the AR smart headband in real time through real-time video communication tools; Decision support module: It is used to input the inspection data into the trained LSTM model, obtain the abnormal cause analysis of the abnormal equipment, and send the abnormal cause analysis to the remote expert to provide decision support.

Citation Information

Cited By

  • Intelligent power distribution network fault early warning system and method based on Internet of Things

    CN119813543A