WO2022160413A1 - 电力生产异常监控方法、装置、计算机设备和存储介质 - Google Patents

电力生产异常监控方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2022160413A1
WO2022160413A1 PCT/CN2021/079282 CN2021079282W WO2022160413A1 WO 2022160413 A1 WO2022160413 A1 WO 2022160413A1 CN 2021079282 W CN2021079282 W CN 2021079282W WO 2022160413 A1 WO2022160413 A1 WO 2022160413A1
Authority
WO
WIPO (PCT)
Prior art keywords
detection
abnormality
image
leakage
scene image
Prior art date
Application number
PCT/CN2021/079282
Other languages
English (en)
French (fr)
Inventor
汪志强
张豪
陈满
卢勇
刘涛
李建辉
吕志鹏
林恺
韩玉麟
巩宇
陆传德
黄发满
景增明
Original Assignee
南方电网调峰调频发电有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 南方电网调峰调频发电有限公司 filed Critical 南方电网调峰调频发电有限公司
Publication of WO2022160413A1 publication Critical patent/WO2022160413A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Definitions

  • the present application relates to the technical field of power production, and in particular, to a method, device, computer equipment and storage medium for monitoring abnormality of power production.
  • a pumped storage power plant with an installed capacity of 1200MW has more than 200 cameras in the production area.
  • a 500kV substation production area has more than 100 cameras.
  • automated sensors cannot fully sense abnormal conditions such as oil leakage, water leakage, and parts falling off equipment. Relying on strict management and conscientious and responsible inspectors cannot achieve the ideal inspection effect of full-scale and full-time coverage.
  • a method for monitoring abnormality of power production comprising:
  • the detection frequency of the ledger object corresponding to the work scene image input the work scene image into a pre-trained detection model to obtain an abnormal detection result
  • the corresponding secondary abnormality detection model is invoked according to the abnormality type to perform secondary detection on the work site;
  • the work scene image is input into a pre-trained detection model to obtain an abnormality detection result, including:
  • the working scene image is input into the detection model, the detection model is used to detect whether there is an abnormality in the power production site, and the classification result is output and when the classification result is abnormal, the
  • calling the corresponding secondary abnormality detection model to perform secondary detection on the work site according to the abnormality type including:
  • the working scene image is input into the secondary identification model of oil leakage and water leakage.
  • the secondary identification model of oil leakage and water leakage adopts a residual network structure.
  • the recognition model outputs secondary anomaly detection results.
  • calling the corresponding secondary abnormality detection model to perform secondary detection on the work site according to the abnormality type including:
  • the abnormality type is steam leakage, obtain the infrared image collected by the infrared camera of the accounting object corresponding to the working scene image;
  • the infrared image is input into the infrared image recognition model, the infrared image features are extracted through the feature extraction network of the infrared image recognition model, and the infrared image features are input into the classifier to obtain the secondary abnormality detection result of steam leakage.
  • acquiring the detection frequency of the ledger object corresponding to the work scene image includes:
  • the method of training the detection model includes:
  • the image sample set includes a sample, and a segmentation label marked on each pixel of the sample;
  • the feature extraction network and the segmentation network of the detection model are trained
  • the classification label of the sample is obtained
  • the detection model is trained according to the classification labels.
  • the classification label of the sample is obtained according to the trained feature extraction network and segmentation network, including:
  • a power production abnormality monitoring device comprising:
  • the image acquisition module is used to acquire the working scene image collected by the camera set at the power production site;
  • a detection module configured to input the work scene image into a detection model according to the detection frequency of the corresponding ledger object in the work scene image to obtain an abnormality detection result
  • the secondary detection module is used to call the corresponding secondary abnormality detection model according to the abnormality type to perform secondary detection on the work site when the abnormality detection result of the power production site is obtained according to the detection model;
  • the alarm module is used to issue an abnormal alarm if the secondary detection is still abnormal.
  • a computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:
  • the detection frequency of the ledger object corresponding to the work scene image input the work scene image into a pre-trained detection model to obtain an abnormal detection result
  • the corresponding secondary abnormality detection model is invoked according to the abnormality type to perform secondary detection on the work site;
  • the detection frequency of the ledger object corresponding to the work scene image input the work scene image into a pre-trained detection model to obtain an abnormal detection result
  • the corresponding secondary abnormality detection model is invoked according to the abnormality type to perform secondary detection on the work site;
  • the above-mentioned method, device, computer equipment and storage medium for monitoring abnormality of electric power production by collecting working scene images collected by cameras on the production site, using a pre-trained detection model, after detecting abnormality, call the secondary abnormality detection model corresponding to the abnormality type.
  • this method for abnormality detection it is only necessary to arrange cameras at the power production site. The cameras have a wide collection range, and complex arrangements are not required, which can reduce management costs.
  • FIG. 1 is an application environment diagram of a method for monitoring abnormality of power production in one embodiment
  • FIG. 2 is a schematic flowchart of a method for monitoring abnormal power production in one embodiment
  • 3A is an image of a working scene of water leakage in one embodiment
  • 3B is a result of segmenting the water leakage area from the work scene image in one embodiment
  • 4A is a working scene image of steam leakage in one embodiment
  • 4B is a result of segmenting the vapor leakage region from the working scene image in one embodiment
  • 5A is an image of a working scene of oil leakage in one embodiment
  • FIG. 5B is a segmentation result of oil leakage area segmented from a working scene image in one embodiment
  • FIG. 6 is a schematic flowchart of the steps of training the detection model in one embodiment
  • FIG. 7 is a schematic structural diagram of a detection model in one embodiment
  • FIG. 8 is a schematic structural diagram of a segmentation network in one embodiment
  • FIG. 9 is a schematic structural diagram of a classification network in one embodiment
  • FIG. 10 is a structural block diagram of a power production abnormality monitoring device in one embodiment
  • Figure 11 is a diagram of the internal structure of a computer device in one embodiment.
  • the method for monitoring abnormality of power production can be applied to the application environment shown in FIG. 1 .
  • a camera 102 and an infrared camera 104 for the ledger object are set at the power production site.
  • the camera 102 and the infrared camera 104 are connected to the monitoring terminal 106 in the control room, and the monitoring terminal receives the working scene image collected by the camera and the infrared image collected by the infrared camera. .
  • the monitoring terminal obtains the working scene image collected by the camera set at the power production site; according to the detection frequency of the corresponding ledger object in the working scene image, the working scene image is input into the detection model to obtain the abnormal detection result; when the power production site is obtained according to the detection model When the abnormal detection result is found, the corresponding secondary abnormal detection model is called according to the abnormal type to perform secondary detection on the work site; if the secondary detection is still abnormal, an abnormal alarm is issued.
  • the monitoring terminal 102 may be, but is not limited to, various personal computers, notebook computers, smart phones or tablet computers.
  • a method for monitoring abnormal power production is provided, and the method is applied to the monitoring terminal in FIG. 1 as an example for description, including the following steps:
  • Step 202 acquiring a working scene image collected by a camera set at the power production site.
  • cameras are installed at the power production site to collect images of the work site.
  • the images collected by the camera are stored in the database, the file name is set to the camera name + preset position + time, and the name is stored in the database.
  • the name of a camera is "middle aisle of unit #1 on the 6.5-meter floor of the main building (dome camera)”
  • the preset position is "1”
  • the time is "2020-09-24 16:13:14”
  • the image is "Middle aisle (ball machine) 1 2020-09-24 16:13:14 of #1 unit on the 6.5-meter floor of the main building.”
  • the video collected by the camera is switched and patrolled according to the set interval in the monitoring frame.
  • Step 204 Input the working scene image into a pre-trained detection model according to the detection frequency of the ledger object corresponding to the working scene image to obtain an abnormality detection result.
  • the ledger object refers to the power production equipment. Different power production equipment has different detection frequencies. For example, if the ledger object is #1 unit, #2 unit, #3 unit, #4 unit, etc., the set algorithm frequency is 30 minutes/time, and the ledger object is 400V#1 unit power plate, 400V#1 unit power plate, 400V The algorithm frequency set by #1 unit power plate, 400V #1 unit power plate, etc. is 60 minutes/time, and the ledger object is the algorithm frequency set by #2 unit governor fuel tank, #2 unit governor pressure oil tank, etc. for 10 minutes/time.
  • the detection frequency of the ledger equipment when the detection time is reached, the corresponding working scene image is input into the detection model to obtain the abnormal detection result.
  • the detection frequency of the work scene image corresponding to the ledger object is input into the detection model, and the abnormal detection result is obtained, including: obtaining the detection frequency of the work scene image corresponding to the ledger object; when the detection time corresponding to the detection frequency is reached
  • the detection model is used to detect whether there is an abnormality in the power production site, and the output classification result and the abnormal area segmentation result segmented from the working scene image when the classification result is abnormal; the classification results include: normal , oil leakage, steam leakage and water leakage.
  • oil leakage, steam leakage and water leakage are the three abnormal conditions that exist.
  • the camera name and preset position of the working scene image are obtained; the ledger object is matched according to the camera name and the preset position; if a single ledger object is matched, the detection frequency corresponding to the ledger object is obtained; If the ledger object is used, the minimum detection frequency among multiple ledger objects is obtained.
  • the name of the camera is usually related to its installation location and type of camera.
  • the name of the camera is "the middle aisle of the #1 unit on the 6.5-meter floor of the main building (ball camera)", which means that its installation location is in the middle aisle of the #1 unit on the 6.5-meter floor of the main building, which is the type of dome camera.
  • Presets are the way in which the focus area being monitored is linked to the performance of the dome.
  • the operator can set a preset position for the current monitoring target, such as a moving point PTZ, which can be rotated and monitored 365 or 360 degrees in all directions; the operator can put a The location of a certain power production equipment is set as a preset position; the preset preset position can be saved on the decoder of the terminal monitoring PTZ through the operation of the control equipment software.
  • the preset preset position can be saved on the decoder of the terminal monitoring PTZ through the operation of the control equipment software.
  • the algorithm frequency set by the ledger object is set to the frequency at which the camera's data is sent to the algorithm for detection. If the matching ledger objects are multiple objects, the frequency at which the data of the camera is sent to the algorithm for detection is set according to the smallest frequency among the matching ledger objects.
  • server resource allocation estimation method can be calculated according to the frequency of camera data feeding into the algorithm, and the calculation method is as formula (1)
  • S represents the estimated value of the resources required by the server
  • N represents the total number of times the algorithm is executed during the peak period in the cycle (the product of the peak period duration and the algorithm frequency)
  • A represents the ratio of the peak period time in the cycle to the total cycle time
  • T represents the The duration of the peak period in the cycle
  • B represents the ratio of the actual operation of the algorithm to the complexity of the test environment
  • C represents the redundancy reserved for the future development of the system, which needs to be estimated according to the application system
  • D represents the optimal utilization of the system, because the system Excessive utilization results in system bottlenecks, and when the utilization is at 75%, it is in the optimal state of utilization.
  • E is the amount of resources required by a single algorithm in the test environment.
  • the detection model is pre-trained based on the target data, detects whether there is an abnormality in the work scene, and outputs the classification result and the abnormal area segmentation result segmented from the work scene image when it is classified as abnormal; the classification results include: normal, oil leakage, leakage Steam and water leaks.
  • water leakage refers to abnormal situations such as leakage, jetting or ground flooding of water-related components such as reservoirs or water pipes.
  • Oil leakage is similar to water leakage, which refers to abnormal situations such as leakage, spraying or ground oil immersion in oil-containing parts such as oil storage tanks or oil pipes.
  • Steam leaks include high temperature steam leaks and white steam leaks.
  • the abnormal area segmentation result that is, the abnormal part displayed in the image, is also segmented from the working scene image.
  • FIG. 3A it is a working scene image of water leakage, the water leakage area is in the box, and FIG. 3B is the segmentation result of the water leakage area segmented from the working scene image.
  • FIG. 4A it is a working scene image of steam leakage, and the frame is a steam leakage area, and FIG. 4B is the segmentation result of the steam leakage area segmented from the working scene image.
  • FIG. 5A is an image of a working scene of oil leakage, the oil leakage area is in the box, and FIG. 5B is the segmentation result of the oil leakage area segmented from the working scene image.
  • Step 206 when the abnormal detection result of the work site is obtained according to the detection model, the corresponding secondary abnormality detection model is invoked according to the abnormality type to perform secondary detection on the work site.
  • the classification result obtained by the detection model is any one of oil leakage, steam leakage and water leakage
  • secondary detection is performed on the work site according to the secondary abnormality detection model corresponding to the abnormality type. That is to say, different anomaly types correspond to different secondary anomaly detection models. For anomaly types, different secondary anomaly detection models can be set according to anomalies, which can further improve detection accuracy.
  • Step 208 if the secondary detection is still abnormal, an abnormal alarm is issued.
  • the output alarm result includes the name of the camera, the monitored object, the spatial location path, etc. Otherwise, return to start to continue monitoring.
  • the classification result is oil leakage or water leakage
  • the output alarm result includes (camera name, monitored object, spatial location path, etc.), otherwise return to start to continue monitoring.
  • the result of classification is steam leakage
  • the output alarm result includes (camera name, monitored object, spatial location path, etc.), otherwise return to start to continue monitoring.
  • the above-mentioned method for monitoring abnormality of power production by collecting the working scene images collected by the cameras on the production site, using the pre-trained detection model, after detecting the abnormality, calling the secondary abnormality detection model corresponding to the abnormality type to perform secondary detection on the work site , using the neural network model to extract higher features, and further through the secondary detection and identification, improve the detection accuracy and reduce the false alarm rate.
  • this method for abnormality detection it is only necessary to arrange cameras at the power production site. The cameras have a wide collection range, and complex arrangements are not required, which can reduce management costs.
  • the way of training the detection model includes:
  • S602 Acquire an image sample set, where the image sample set includes a sample and a segmentation label marked on each pixel of the sample.
  • the structure of the detection model 70 is shown in FIG. 7 , including a feature extraction network 701 , a segmentation network 702 , a spatial attention network 703 , and a classification network 704 .
  • the feature extraction network 701 is respectively connected with the segmentation network 702, the spatial attention network 703, and the spatial attention network 703 is connected with the classification network 704.
  • the parameters of the spatial attention module and the classification module are fixed, and the samples of the image sample set are input to the feature extraction network 701 to extract image features.
  • the feature annotation extraction network is a multi-layer convolutional neural network, which is used to extract the high-level features F ⁇ R c,w,h of the input image.
  • convolutional neural networks such as VGG19 and MobileNet can be used.
  • the image features are input to the segmentation network, and the segmentation results of abnormal regions are output.
  • the abnormal area refers to the area of oil leakage, water leakage and steam leakage determined by characteristics.
  • the segmentation module is a multi-layer convolutional neural network, the last layer of the network is connected to the sigmoid activation function, and the mask w ⁇ R c,w of the water leakage, oil leakage and steam leakage area is output.
  • a branch of water leakage, oil leakage and steam leakage segmentation module is added to output the regional mask of water leakage, oil leakage and steam leakage.
  • the output spatial attention mask of this branch is combined with the classification branch. feature to improve the classification accuracy.
  • the segmentation network can use the network structure shown in Figure 8, where conv3 ⁇ 3 represents a convolution block with a convolution kernel size of 3 ⁇ 3 and a stride of 1, including convolution operations and nonlinear unit rule activation functions .
  • the loss function can use loss functions such as L1 and L2.
  • the classification label of the sample is obtained.
  • the samples in the sample set have no classification labels, and the classification labels are obtained by using the trained feature extraction network and segmentation network.
  • the sample image is input into the detection model, the image features are extracted through the feature extraction network of the detection model, the image features are input into the segmentation network, the segmentation result of the abnormal area is output, and the classification label of the sample is generated according to the segmentation result.
  • the samples of the image sample set are input into the detection model, and the image features are extracted through the feature extraction network 701 .
  • the feature annotation extraction network is a multi-layer convolutional neural network, which is used to extract the high-level features F ⁇ R c,w,h of the input image.
  • convolutional neural networks such as VGG19 and MobileNet can be used.
  • the abnormal area refers to the area of oil leakage, water leakage and steam leakage determined by characteristics.
  • the segmentation module is a multi-layer convolutional neural network, the last layer of the network is connected to the sigmoid activation function, and the mask w ⁇ R c,w of the water leakage, oil leakage and steam leakage area is output.
  • the segmentation network can use the network structure shown in Figure 8, where conv3 ⁇ 3 represents a convolution block with a convolution kernel size of 3 ⁇ 3 and a stride of 1, including convolution operations and nonlinear unit rule activation functions .
  • Categorical label data needs to be dynamically generated during training.
  • the generation method is as follows: binarize the output result w of the segmentation module to obtain w b .
  • the binarization threshold can be set as required, the range is (0, 1), and it can be 0.5. Then calculate the classification label of water leakage, oil leakage and steam leakage according to the following formula.
  • T is the threshold for judging water leakage, oil leakage and steam leakage, which can be set as required.
  • the classification model can be trained using the dynamically generated labels described above.
  • classification labels are obtained using the trained feature extraction network and segmentation network. Specifically, a two-stage classification method is proposed, which uses a segmentation network to generate candidate regions, and then classifies the candidate regions to balance the number of positive and negative samples to improve the accuracy of the classifier.
  • the detection model is trained according to the classification labels and the segmentation labels of the samples marked in the image sample set.
  • the segmentation results and image features are input into the spatial attention network to obtain the attention features
  • the attention features are input into the classification network to obtain the predicted abnormal classification results. According to the difference between the predicted classification results and the labeled classification results, adjust the Detect the model, and iteratively train until the training end condition is met, and a trained detection model is obtained.
  • the spatial attention network is a feature pixel-level operation.
  • W [w,w,...,w]
  • W ⁇ R c,w,h . ⁇ is the pixel multiplication, pixel addition.
  • the classification network is composed of multiple convolutional layers, pooling layers, fully connected layers, and output layers, which are used to output the scoring result s of water leakage detection.
  • s (s 1 , s 2 , s 3 ) ⁇ R 3 is a vector with a dimension of 3. The value of each dimension represents the score of water leakage, oil leakage and steam leakage. The higher the value, the greater the confidence rate of the corresponding state. .
  • the classification network can use the network structure shown in Figure 9, where conv3 ⁇ 3 represents a convolution block with a convolution kernel size of 3 ⁇ 3 and a stride of 1, including convolution operations and nonlinear unit rule activation functions.
  • Global pooling means using a global pooling operation to pool the input feature resolution to a specific size, such as 7 ⁇ 7.
  • MLP is a multi-layer perceptron network model that expands the pooled features into a 1-dimensional vector, which is then input into a two-layer fully connected layer neural network. Finally, the classification result is obtained through the softmax function.
  • the difference between the predicted classification result and the labeled classification result is back-propagated, the detection model is adjusted, and the training is iteratively trained until the training end condition is satisfied, and a trained detection model is obtained.
  • the training end condition can be that the number of iterations reaches the maximum number of iterations, or the model accuracy meets the requirements. Among them, iteration refers to re-executing the above training process on the samples. During the training process, you can use the test data set for testing to observe the training effect.
  • the method using convolutional neural network has stronger ability to extract features, and the classification network combining segmentation module and spatial attention module can focus on possible water leakage. It can improve the ability of the network to extract features and improve the robustness of the network model.
  • calling a corresponding secondary abnormality detection model to perform secondary detection on the work site according to the abnormality type including: if the abnormality type is oil leakage and/or water leakage, inputting the work scene image into the oil and water leakage secondary
  • the identification model, the secondary identification model of oil leakage and water leakage adopts the residual network structure, and outputs the secondary abnormal detection results through the secondary identification model of oil leakage and water leakage.
  • the image of the detected oil leakage and water leakage will be input into the secondary identification model of oil leakage and water leakage for secondary identification.
  • ResNet50 50-layer residual network
  • the number of hidden units in the final fully connected layer is set to 3, which are respectively classified into three types: oil leakage, water leakage, and normal.
  • ResNet50 is trained with three categories of data of oil leakage, water leakage and normal, and is trained by stochastic gradient descent. In order to prevent the network from overfitting, the method of dropout (randomly discarding neurons) is used for network training.
  • the infrared image collected by the infrared camera corresponding to the ledger object of the working scene image is obtained; the infrared image is input into the infrared image recognition model, and the feature extraction model of the infrared image recognition model is used to extract the image.
  • the infrared image features are extracted, and the infrared image features are input into the classifier to obtain the secondary abnormality detection results of whether there is steam leakage.
  • the infrared camera monitoring the ledger object is matched, and the infrared image data corresponding to the time is obtained; the infrared image recognition module is used to obtain the corresponding infrared camera image for identification.
  • the infrared image module consists of a feature extraction module and a classification module, wherein the feature extraction module adopts the traditional directional gradient histogram (HOG feature) extraction method, and the classification module is a support vector machine (SVM) classifier, and the present invention adopts the steam leakage image.
  • HOG feature traditional directional gradient histogram
  • SVM support vector machine
  • Positive samples infrared image samples of various poses including steam leakage
  • negative samples any infrared image samples without steam leakage
  • the result of the classifier output classification is the result of the secondary identification of steam leakage.
  • a multi-task learning network combined with spatial attention mechanism is proposed, which can simultaneously detect three abnormal states of oil leakage, water leakage and steam leakage.
  • a matching method is proposed from the polling camera shooting content and the monitoring object, and by setting the frequency to be recognized for the monitoring object, the camera can automatically match the processing frequency.
  • An estimation formula of computing resources is proposed, which can calculate the computing resources required by the algorithm configuration of the whole scene, and has strong practicability.
  • a branch of water leakage, oil leakage and steam leakage segmentation module is added to output the regional mask of water leakage, oil leakage and steam leakage.
  • the output spatial attention mask of this branch is combined with the classification branch. features to improve the classification accuracy.
  • a power production abnormality monitoring device including:
  • the image acquisition module 1001 is configured to acquire a working scene image collected by a camera set at a power production site.
  • the detection module 1002 is configured to input the work scene image into a detection model according to the detection frequency of the corresponding ledger object in the work scene image to obtain an abnormality detection result.
  • the secondary detection module 1003 is configured to call the corresponding secondary abnormality detection model according to the abnormality type to perform secondary detection on the work site when the abnormality detection result of the power production site is obtained according to the detection model.
  • the alarm module 1004 is configured to issue an abnormal alarm if the secondary detection is still abnormal.
  • the above-mentioned power production abnormality monitoring device collects the working scene images collected by the cameras on the production site, and uses the pre-trained detection model to detect the abnormality, and then calls the secondary abnormality detection model corresponding to the abnormality type to perform secondary detection on the work site.
  • the model using the neural network extracts higher features, and further through the secondary detection and identification, the detection accuracy is improved and the false alarm rate is reduced.
  • this method for abnormality detection it is only necessary to arrange cameras at the power production site. The cameras have a wide collection range, and complex arrangements are not required, which can reduce management costs.
  • the detection module includes:
  • a detection frequency determination module configured to obtain the detection frequency of the ledger object corresponding to the working scene image
  • the prediction module is used to input the working scene image into the detection model when the detection time corresponding to the detection frequency is reached, and detect whether there is an abnormality in the power production site through the detection model, and output the classification result and the classification result as The abnormal area segmentation result segmented from the working scene image when abnormal; the classification result includes: normal, oil leakage, steam leakage and water leakage.
  • a secondary detection module is configured to input the working scene image into a secondary identification model for oil leakage and water leakage if the abnormal type is oil leakage and/or water leakage, and the secondary identification model for oil leakage and water leakage Using the residual network structure, the secondary abnormality detection result is output through the secondary identification model of oil leakage and water leakage.
  • the secondary detection module is configured to obtain an infrared image collected by an infrared camera corresponding to the ledger object in the working scene image if the abnormality type is steam leakage; input the infrared image into the infrared image recognition model, and pass The feature extraction network of the infrared image recognition model extracts infrared image features, and inputs the infrared image features into a classifier to obtain a secondary abnormality detection result of whether there is steam leakage.
  • a detection frequency determination module is used to obtain the camera name and preset position of the working scene image; match the ledger object according to the camera name and preset position; if a single ledger object is matched, The detection frequency corresponding to the ledger object is acquired; if multiple ledger objects are matched, the minimum detection frequency among the plurality of ledger objects is acquired.
  • the power production abnormality monitoring device further includes:
  • the sample set processing module is used to obtain an image sample set, where the image sample set includes samples and segmentation labels marked on each pixel of the samples.
  • the primary training module is used to train the feature extraction network and the segmentation network of the detection model according to the segmentation labels of the samples marked in the image sample set.
  • the classification module is used for extracting the network and segmenting the network according to the trained feature to obtain the classification label of the sample.
  • the secondary training module is used for training the detection model according to the classification label.
  • the classification module is configured to input the samples into the detection model, extract image features through the feature extraction network trained in the detection model; input the image features into the trained segmentation network, and output abnormal regions The segmentation result; according to the segmentation result, the classification label of the sample is generated.
  • Each module in the above-mentioned power production abnormality monitoring device may be implemented in whole or in part by software, hardware, and combinations thereof.
  • the above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
  • a computer device is provided, and the computer device may be a monitoring terminal, and its internal structure diagram may be as shown in FIG. 11 .
  • the computer equipment includes a processor, memory, a communication interface, a display screen, and an input device connected by a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium, an internal memory.
  • the nonvolatile storage medium stores an operating system and a computer program.
  • the internal memory provides an environment for the execution of the operating system and computer programs in the non-volatile storage medium.
  • the communication interface of the computer device is used for wired or wireless communication with an external terminal, and the wireless communication can be realized by WIFI, operator network, NFC (Near Field Communication) or other technologies.
  • the computer program when executed by the processor, implements a power production abnormality monitoring method.
  • the display screen of the computer equipment may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment may be a touch layer covered on the display screen, or a button, a trackball or a touchpad set on the shell of the computer equipment , or an external keyboard, trackpad, or mouse.
  • FIG. 11 is only a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer equipment to which the solution of the present application is applied. Include more or fewer components than shown in the figures, or combine certain components, or have a different arrangement of components.
  • a computer device including a memory and a processor, where a computer program is stored in the memory, and the processor implements the steps in the foregoing method embodiments when the processor executes the computer program.
  • a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps in the foregoing method embodiments are implemented.
  • Non-volatile memory may include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory, or optical memory, and the like.
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • the RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Multimedia (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)
  • Secondary Cells (AREA)

Abstract

本申请涉及一种电力生产异常监控方法、装置、计算机设备和存储介质。该方法包括：获取设置在电力生产现场的摄像头采集的工作场景图像；根据工作场景图像对应台账对象的检测频率，将工作场景图像输入到预先训练的检测模型，得到异常检测结果；在根据检测模型得到电力生产现场异常的检测结果时，根据异常类型调用对应的二次异常检测模型对工作现场进行二次检测；若二次检测仍为异常，则发出异常报警。该方法能够提高检测的准确性和降低了虚警率，其中漏水漏油的二次识别采用与第一次识别的监测网络不同，比用相同网络结构识别具有更高的可靠性，而漏汽的二次识别通过匹配到监测对象对应的红外图像数据进行识别，从而提高了蒸汽的检测准确性。

Description

电力生产异常监控方法、装置、计算机设备和存储介质 技术领域
本申请涉及电力生产技术领域，特别是涉及一种电力生产异常监控方法、装置、计算机设备和存储介质。
背景技术
一座装机容量为1200MW的抽水蓄能电厂生产区域有超过200路摄像头。一个500kV变电站生产区域有超过100路摄像头。过去仅能依靠人工进行现场视频查阅，海量视频数据无法应用于数据分析，工业电视系统也仅为事后分析或查看确认的系统。同时自动化传感器无法完全感知设备漏油、漏水、零件脱落等异常情况。依赖于严格的管理，依靠认真负责的巡检员也无法做到全范围全时段覆盖的理想巡检效果。
电厂/变电站设备种类众多，包含油、水、气、电设备。由于密封性能下降、零部件缺损脱落等机械缺陷原因，生产过程中设备漏水、漏油、漏气的“三漏”时有发生。生产设备出现漏油、漏水和漏汽等异常情况不及时发现，往往会导致缺陷影响升级导致主设备停运，甚至带来其它安全事故。燃气电厂/火电厂中高温蒸汽泄露往往会因为热质泄露带来机组发电效率降低进而影响到整个机组的经济性指标。此外高温蒸汽泄漏还可能会给运维人员带来人身伤害。因此，在电厂生产中急需应用光学/红外摄像头对漏油、漏水、漏汽进行主动检测以及时发现异常。
发明内容
基于此，有必要针对上述技术问题，提供一种管理简单且检测精度高的电力生产异常监控方法、装置、计算机设备和存储介质。
一种电力生产异常监控方法，所述方法包括：
获取设置在电力生产现场的摄像头采集的工作场景图像；
根据所述工作场景图像对应台账对象的检测频率，将所述工作场景图像输入到预先训练的检测模型，得到异常检测结果；
在根据所述检测模型得到电力生产现场异常的检测结果时，根据异常类型调用对应的二次异常检测模型对工作现场进行二次检测；
若二次检测仍为异常，则发出异常报警。
在其中一个实施例中，根据所述工作场景图像对应台账对象的检测频率，将所述工作场景图像输入到预先训练的检测模型，得到异常检测结果，包括：
获取所述工作场景图像对应台账对象的检测频率；
当达到所述检测频率对应的检测时间时，将所述工作场景图像输入到检测模型，通过所述检测模型对电力生产现场是否出现异常进行检测，输出分类结果和分类结果为异常时从所述工作场景图像分割出来的异常区域分割结果；所述分类结果包括：正常、漏油、漏汽和漏水。
在其中一个实施例中，根据异常类型调用对应的二次异常检测模型对工作现场进行二次检测，包括：
若异常类型为漏油和/或漏水，则将所述工作场景图像输入漏油漏水二次识别模型，所述漏油漏水二次识别模型采用残差网络结构，通过所述漏油漏水二次识别模型输出二次异常检测结果。
在其中一个实施例中，根据异常类型调用对应的二次异常检测模型对工作现场进行二次检测，包括：
若异常类型为漏汽，则获取工作场景图像对应台账对象的红外摄像头采集的红外图像；
将所述红外图像输入红外图像识别模型，通过所述红外图像识别模型的特征提取网络，提取红外图像特征，将所述红外图像特征输入分类器，得到是否漏汽的二次异常检测结果。
在其中一个实施例中，获取所述工作场景图像对应台账对象的检测频率，包括：
获取所述工作场景图像的摄像头名称和预置位；
根据所述摄像头名称和预置位匹配台账对象；
若匹配到单个台账对象，则获取所述台账对象对应的检测频率；
若匹配到多个台账对象，则获取多个所述台账对象中最小的检测频率。
在其中一个实施例中，训练检测模型的方式，包括：
获取图像样本集，所述图像样本集包括样本，以及对样本各像素点标注的分割标签；
根据所述图像样本集中标注的样本的分割标签，训练所述检测模型的特征提取网络和分割网络；
根据训练的所述特征提取网络和分割网络，得到样本的分类标签；
根据所述分类标签，对检测模型进行训练。
在其中一个实施例中，根据训练的所述特征提取网络和分割网络，得到样本的分类标签，包括：
将样本输入到检测模型，通过所述检测模型中训练的特征提取网络，提取图像特征；
将所述图像特征输入到训练的分割网络，输出异常区域的分割结果；
根据所述分割结果，生成所述样本的分类标签。
一种电力生产异常监控装置，所述装置包括：
图像获取模块，用于获取设置在电力生产现场的摄像头采集的工作场景图像；
检测模块，用于根据所述工作场景图像对应台账对象的检测频率，将所述工作场景图像输入到检测模型，得到异常检测结果；
二次检测模块，用于在根据所述检测模型得到电力生产现场异常的检测结果时，根据异常类型调用对应的二次异常检测模型对工作现场进行二次检测；
报警模块，用于若二次检测仍为异常，则发出异常报警。
一种计算机设备，包括存储器和处理器，所述存储器存储有计算机程序，所述处理器执行所述计算机程序时实现以下步骤：
获取设置在电力生产现场的摄像头采集的工作场景图像；
根据所述工作场景图像对应台账对象的检测频率，将所述工作场景图像输入到预先训练的检测模型，得到异常检测结果；
在根据所述检测模型得到电力生产现场异常的检测结果时，根据异常类型调用对应的二次异常检测模型对工作现场进行二次检测；
若二次检测仍为异常，则发出异常报警。
一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现以下步骤：
获取设置在电力生产现场的摄像头采集的工作场景图像；
根据所述工作场景图像对应台账对象的检测频率，将所述工作场景图像输入到预先训练的检测模型，得到异常检测结果；
在根据所述检测模型得到电力生产现场异常的检测结果时，根据异常类型调用对应的二次异常检测模型对工作现场进行二次检测；
若二次检测仍为异常，则发出异常报警。
上述电力生产异常监控方法、装置、计算机设备和存储介质，通过采集生产现场的摄像头采集的工作场景图像，利用预先训练的检测模型，检测到异常后，再调用异常类型对应的二次异常检测模型对工作现场进行二次检测，利用神经网络的模型提取特征更高，进一步通过二次检测识别，提高检测的准确性和降低了虚警率。另外，采用该方法进行异常检测，只需在电力生产现场布置摄像头即可，摄像头的采集范围广，无需进行复杂布置，能够降低管理成本。
附图说明
图1为一个实施例中电力生产异常监控方法的应用环境图；
图2为一个实施例中电力生产异常监控方法的流程示意图；
图3A为一个实施例中漏水的工作场景图像；
图3B为一个实施例中从工作场景图像分割出漏水区域分割结果；
图4A为一个实施例中漏汽的工作场景图像；
图4B为一个实施例中从工作场景图像分割出漏汽区域分割结果；
图5A为一个实施例中漏油的工作场景图像；
图5B为一个实施例中从工作场景图像分割出漏油区域分割结果；
图6为一个实施例中训练所述检测模型的步骤的流程示意图；
图7为一个实施例中检测模型的结构示意图；
图8为一个实施例中分割网络的结构示意图；
图9为一个实施例中分类网络的结构示意图；
图10为一个实施例中电力生产异常监控装置的结构框图；
图11为一个实施例中计算机设备的内部结构图。
具体实施方式
以下结合附图和实例对本发明的具体实施作进一步说明，但本发明的实施和保护不限于此。需指出的是，以下若有未特别详细说明之过程，均是本领域技术人员可参照现有技术实现或理解的。本申请提供的电力生产异常监控方法，可以应用于如图1所示的应用环境中。在电力生产现场设置了针对台账对象的摄像头102和红外摄像头104，摄像头102和红外摄像头104连接到控制室的监控终端106，监控终端接收摄像头采集的工作场景图像，接收红外摄像头采集的红外图像。
监控终端获取设置在电力生产现场的摄像头采集的工作场景图像；根据工作场景图像对应台账对象的检测频率，将工作场景图像输入到检测模型，得到异常检测结果；在根据检测模型得到电力生产现场异常的检测结果时，根据异常类型调用对应的二次异常检测模型对工作现场进行二次检测；若二次检测仍为异常，则发出异常报警。
其中，监控终端102可以但不限于是各种个人计算机、笔记本电脑、智能手机或平板电脑。
在一个实施例中，如图2所示，提供了一种电力生产异常监控方法，以该方法应用于图1中的监控终端为例进行说明，包括以下步骤：
步骤202，获取设置在电力生产现场的摄像头采集的工作场景图像。
具体地，在电力生产现场设置有摄像头，采集工作现场的图像。摄像头采集 的图像存储在数据库中，文件名设置为摄像头名称+预置位+时间，名字存储进入数据库。如一个摄像头名称为“主厂房6.5米层#1机组中间过道(球机)”，预置位为“1”，时间为“2020-09-24 16：13：14”，则对应的工作场景图像为“主厂房6.5米层#1机组中间过道(球机)1 2020-09-24 16：13：14”。其中，摄像头采集的视频在监视框中根据设定的间隔时间进行切换轮巡。
步骤204，根据工作场景图像对应台账对象的检测频率，将工作场景图像输入到预先训练的检测模型，得到异常检测结果。
其中，台账对象指的是电力生产设备。不同的电力生产设备的检测频率不同。如台账对象为#1机组、#2机组、#3机组、#4机组等设置的算法频率为30分钟/次，台账对象为400V#1机组动力盘、400V#1机组动力盘、400V#1机组动力盘、400V#1机组动力盘等设置的算法频率为60分钟/次，台账对象为#2机组调速器集油箱、#2机组调速器压油罐等设置的算法频率为10分钟/次等。
根据台账设备的检测频率，在达到检测时间时，将相应的工作场景图像输入到检测模型，得到异常检测结果。
其中，根据工作场景图像对应台账对象的检测频率，将工作场景图像输入到检测模型，得到异常检测结果，包括：获取工作场景图像对应台账对象的检测频率；当达到检测频率对应的检测时间时，将工作场景图像输入到检测模型，通过检测模型对电力生产现场是否出现异常进行检测，输出分类结果和分类结果为异常时从工作场景图像分割出来的异常区域分割结果；分类结果包括：正常、漏油、漏汽和漏水。
其中，漏油、漏汽和漏水就是存在的三种异常情况。
具体地，获取工作场景图像的摄像头名称和预置位；根据摄像头名称和预置位匹配台账对象；若匹配到单个台账对象，则获取台账对象对应的检测频率；若匹配到多个台账对象，则获取多个台账对象中最小的检测频率。
摄像头名称通常与其安装位置和摄像头类型有关，如摄像头名称“主厂房6.5米层#1机组中间过道(球机)”表示其安装位置在主厂房6.5米层#1机组中间过道，是球机类型的摄像头。预置位将被监视的重点区域与球机的运行状况联系在一起的方式。当云台运行到需要重点监视的地方，向球机发出设置预置点的 命令，球机则将此时的云台的方位和摄像机的状态记录下来，并与该预置点的号码联系起来。当用户通过控制设备操作终端的监控云台监视目标时，操作人员可以把当前监视目标设置一个预置位，比如一个动点云台，可以365或360度全方位旋转监视；操作人员可以把一个某个电力生产设备的地点设置为预置位；设置好的预置位可以通过控制设备软件操作把当前位置保存在终端监控云台的解码器上。当用户需要快速监视某个监视目标时候；可以通过控制设备的调用命令来调出需要监视的位置。可以理解的是，不同的预置位对应不同的生产区域。
根据摄像头名称及其预置位匹配台账对象，如果若摄像头匹配的台账对象为单个对象，则将该台账对象设置的算法频率设置为该摄像头的数据送入算法检测的频率，若摄像头匹配的台账对象为多个对象，则按照其匹配的台账对象中最小的那一个频率设置为该摄像头的数据送入算法检测的频率。
另外根据摄像头的数据送入算法的频率可计算出服务器资源配置估算方法，计算方式如公式(1)
S＝(N×A)/T×B×C/D×E     (1)
其中，S表示服务器所需资源估计值，N表示周期内高峰期执行算法总次数(周期内高峰期时长与算法频率的乘积)，A表示周期内高峰期时间与周期总时间的比值，T表示周期内高峰期持续时间，B表示算法实际运行相对于测试环境下的复杂程度比值，C表示为系统未来发展冗余预留，需要根据应用系统估算，D表示为系统最佳利用率，因为系统利用率过高产生系统瓶颈，而利用率处于75％时，是处于利用率最佳状态。此值一般设定为C＝75％，E为测试环境下单次算法所需的资源量。
检测模型是预先根据标数据训练得到的，对工作场景是否出现异常进行检测，输出分类结果和分类为异常时从工作场景图像分割出来的异常区域分割结果；分类结果包括：正常、漏油、漏汽和漏水。其中，漏水是指蓄水池或水管等涉水部件发生渗漏、喷射或地面水浸等异常情况。漏油与漏水的情况类似，是指蓄油池或油管等含油部件发生渗漏、喷射或地面油浸等异常情况。漏汽包含高温蒸汽泄漏和白色蒸汽泄漏。
具体地，若分类结果为漏油、漏汽和漏水中的任一种，还从工作场景图像分割出异常区域分割结果，即图像所显示的异常部分。
如图3A所示，为漏水的工作场景图像，方框内为漏水区域，图3B则为从工作场景图像分割出漏水区域分割结果。如4A所示，为漏汽的工作场景图像，方框内为漏汽区域，图4B则为从工作场景图像分割出漏汽区域分割结果。如图5A为漏油的工作场景图像，方框内为漏油区域，图5B则为从工作场景图像分割出漏油区域分割结果。
步骤206，在根据检测模型得到工作现场异常的检测结果时，根据异常类型调用对应的二次异常检测模型对工作现场进行二次检测。
具体地，若检测模型得到的分类结果为漏油、漏汽和漏水中的任一种，则根据异常类型对应的二次异常检测模型对工作现场进行二次检测。也就是说，不同的异常类型对应不同的二次异常检测模型，针对异常类型，能够根据异常特别分别设置不同的二次异常检测模型，能够进一步提高检测的精确度。
步骤208，若二次检测仍为异常，则发出异常报警。
具体地，若二次检查仍为异常，则输出结果并进行报警，输出的报警结果包含摄像头名称、监测的对象，空间位置路径等，否则返回开始继续监测。
在识别到漏油、漏水和漏汽三种情况后，不单单只有报警提示，还会输出报警的摄像头名称及所监测的对象内容等。例如分类结果为漏油或者漏水，则输出结果并进行报警，输出的报警结果包含(摄像头名称、监测的对象，空间位置路径等)，否则返回开始继续监测。例如分类的结果为漏汽则输出结果并进行报警，输出的报警结果包含(摄像头名称、监测的对象，空间位置路径等)，否则返回开始继续监测。
上述的电力生产异常监控方法，通过采集生产现场的摄像头采集的工作场景图像，利用预先训练的检测模型，检测到异常后，再调用异常类型对应的二次异常检测模型对工作现场进行二次检测，利用神经网络的模型提取特征更高，进一步通过二次检测识别，提高检测的准确性和降低了虚警率。另外，采用该方法进行异常检测，只需在电力生产现场布置摄像头即可，摄像头的采集范围广，无需进行复杂布置，能够降低管理成本。
在一个实施例中，训练检测模型的方式，如图6所示，包括：
S602，获取图像样本集，图像样本集包括样本，以及对样本各像素点标注的分割标签。
具体地，收集大量工作现场的图像，包括漏油、漏水、漏汽、正常等场景的图像数据，然后对所有图像进行像素级标注，制作分割标签，把所有像素分为4个类别：正常、漏水、漏油、漏汽。把图像随机划分两个部分，一部分作为训练的图像样本集，另一部分作为测试数据集。
S604，根据图像样本集中标注的样本的分割标签，训练检测模型的特征提取网络和分割网络。
一个实施例中，检测模型70的结构如图7所示，包括特征提取网络701，分割网络702，空间注意力网络703，分类网络704。其中，特征提取网络701分别与分割网络702，空间注意力网络703连接，空间注意力网络703与分类网络704连接。
首先，固定空间注意力模块和分类模块的参数，将图像样本集的样本输入到特征提取网络701提取图像特征。其中，特征注提取网络为多层卷积神经网络，用于提取输入图像的高级特征F∈R c,w,h。一般地，可以选用VGG19、MobileNet等卷积神经网络。
其次，将图像特征输入到分割网络，输出异常区域的分割结果。
异常区域是指通过特征确定的漏油漏水漏汽区域。分割模块为多层卷积神经网络，网络的最后一层接sigmoid激活函数，输出漏水漏油漏汽区域的掩码w∈R c,w。通过在一般的分类神经网络基础上，外加一条漏水漏油漏汽分割模块支路，用以输出漏水漏油漏汽的区域掩码，该支路输出空间注意力掩码结合到分类支路的特征上，提高分类的准确性。
具体地，分割网络可以使用如图8所示的网络结构，其中conv3×3表示一个卷积核尺寸为3×3步长为1的卷积块，包含卷积操作和非线性单元rule激活函数。
最后，根据预测的分割结果和标注的分割标签的差异，调整检测模型的分割网络和特征提取网络，损失函数可使用L1、L2等损失函数，当模型损失收敛且 不下降时，再同时训练检测模型的参数。
S606，根据训练的特征提取网络和分割网络，得到样本的分类标签。
样本集中的样本没有分类标签，利用已训练的特征提取网络和分割网络得到分类标签。具体地，将样本图像输入到检测模型，通过检测模型的特征提取网络，提取图像特征，将图像特征输入到分割网络，输出异常区域的分割结果，根据分割结果，生成样本的分类标签。
具体地，将图像样本集的样本输入到检测模型，通过特征提取网络701提取图像特征。其中，特征注提取网络为多层卷积神经网络，用于提取输入图像的高级特征F∈R c,w,h。一般地，可以选用VGG19、MobileNet等卷积神经网络。
异常区域是指通过特征确定的漏油漏水漏汽区域。分割模块为多层卷积神经网络，网络的最后一层接sigmoid激活函数，输出漏水漏油漏汽区域的掩码w∈R c,w。
具体地，分割网络可以使用如图8所示的网络结构，其中conv3×3表示一个卷积核尺寸为3×3步长为1的卷积块，包含卷积操作和非线性单元rule激活函数。
样本集中没有制作分类标签。分类标签数据需要在训练的过程中动态生成。其生成方法如下：将分割模块的输出结果w进行二值化得到w b。二值化阈值可根据需要设置，范围为(0，1)，可取0.5。然后根据下述公式计算漏水漏油漏汽的分类标签。
Figure PCTCN2021079282-appb-000001
Figure PCTCN2021079282-appb-000002
Figure PCTCN2021079282-appb-000003
其中
Figure PCTCN2021079282-appb-000004
分别为漏水、漏油、漏汽的标签，T为判定漏水漏油漏汽的阈值，可根据需要设置。分类模型可使用上述动态生成的标签训练。
在实际场景中，漏油、漏水和漏汽发生的频率较低，收集漏水漏油漏汽的正 样本数据较难。如何针对只有少量正样本的数据，设计和训练一个检测模型是一个难点。基于此，利用已训练的特征提取网络和分割网络得到分类标签。具体地，提出一种二阶段的分类方法，使用分割网络生成候选区域，再对候选区域进行分类，平衡正负样本的数量，提高分类器的准确性。
S608，根据分类标签以及图像样本集中标注的样本的分割标签，对检测模型进行训练。
具体地，将分割结果和图像特征输入到空间注意力网络，得到注意力特征，将注意力特征输入分类网络，得到预测的异常分类结果，根据预测的分类结果与标注的分类结果的差异，调整检测模型，并迭代训练，直到满足训练结束条件，得到训练好的检测模型。
漏油、漏水在图像上呈现的变化常常较为细微和缓慢，且实际应用场景环境复杂，背景和地面材质不一，漏油漏水和漏汽的成像差异较大。针对不同的复杂场景，保证检测系统的准确性和鲁棒性是一大难点。针对这一难点，结合空间注意力机制，使网络把焦点聚集在特定区域，降低复杂背景对检测算法的干扰，同时提高算法对鲁棒性和准确率。
具体地，空间注意力网络是一个针对特征像素级的操作。可以用公式
Figure PCTCN2021079282-appb-000005
Figure PCTCN2021079282-appb-000006
其中W＝[w,w，…,w]，且W∈R c,w,h.⊙为像素点乘，
Figure PCTCN2021079282-appb-000007
像素加法。
分类网络为多个卷积层、池化层和全连接层、输出层组成，用以输出漏水检测的得分结果s。s＝(s 1，s 2，s 3)∈R 3是一个维度为3的向量，每个维度的值分别表示漏水漏油漏汽的得分，值越高，说明对应状态的置信率越大。
具体地，分类网络可以使用如图9网络结构，其中conv3×3表示一个卷积核尺寸为3×3步长为1的卷积块，包含卷积操作和非线性单元rule激活函数。Global pooling表示使用全局池化操作将输入特征分辨率池化到特定尺寸，如7×7。MLP是一个多层感知机网络模型，其将池化后的特征展开成1维向量，然后输入到两层全连接层神经网络中。最后，通过softmax函数得到分类结果。
具体地，预测的分类结果与标注的分类结果的差异，反向传播，调整检测模型，并迭代训练，直到满足训练结束条件，得到训练好的检测模型。训练结束条 件可以为迭代次数达到最大迭代次数，或是模型精度达到要求。其中，迭代是指将样本重新执行上述训练过程。训练过程中，可使用测试数据集进行测试，观测训练效果。
上述的电力生产异常监控方法，相比使用手工提取特征的传统方法，使用卷积神经网络的方法提取特征的能力更强，结合了分割模块和空间注意力模块的分类网络能聚焦于可能漏水漏油漏汽的区域，且能提高网络提取特征的能力，提高网络模型的鲁棒性。
在另一个实施例中，根据异常类型调用对应的二次异常检测模型对工作现场进行二次检测，包括：若异常类型为漏油和/或漏水，则将工作场景图像输入漏油漏水二次识别模型，漏油漏水二次识别模型采用残差网络结构，通过漏油漏水二次识别模型输出二次异常检测结果。
若检测模型检测的异常类型为漏油和/或漏水，则将将检测出漏油漏水的图像，输入到漏油漏水二次识别模型进行二次识别。
其中，采用ResNet50(50层的残差网络)作为漏油漏水二次识别模型的网络结构，并将最后的全连接层隐藏单元数设置为3分别对应分类为：漏油、漏水、正常三种情况的分类分数。其中，利用漏油漏水及正常三个类别的数据训练ResNet50，并采用随机梯度下降的方式进行训练，为了防止网络过拟合，采用dropout(随机丢弃神经元)的方式进行网络训练。
在另一个实施例中，若异常类型为漏汽，则获取工作场景图像对应台账对象的红外摄像头采集的红外图像；将红外图像输入红外图像识别模型，通过红外图像识别模型的特征提取模型，提取红外图像特征，将红外图像特征输入分类器，得到是否漏汽的二次异常检测结果。
具体地，根据检测出漏汽的图像的文件名对应的台账对象，匹配出监视该台账对象的红外摄像头，获取对应时间的红外图像数据；采用红外图像识别模块对获取到对应对的红外图像进行识别。
红外图像模块由特征提取模块和分类模块，其中特征提取模块是采用传统的方向梯度直方图(HOG特征)提取方法，而分类模块是支持向量机(SVM)分类器，其中本发明采用漏汽图像正样本(包含蒸汽泄漏的各种姿态形式的红外 图像样本)和负样本(不含有蒸汽泄漏的任意红外图像样本)，通过提取HOG特征后送入SVM分类器，最终形成训练模型分类使用。分类器输出分类的结果即为漏汽二次识别的结果。
本申请的电力生产异常监控方法，具有以下效果：
1、提出了一种结合空间注意力机制的多任务学习网络，能够同时检测漏油、漏水和漏汽三种异常状态。
2、结合实际场景(电厂场景)提出了一种从轮询摄像头拍摄内容与监视对象的匹配方法，并且通过对监视对象设置需要识别的频率，摄像头就能自动匹配到处理频率。
3、提出了一种计算资源的估计公式，能够计算得到整个场景的算法配置所需要的计算资源，具有很强的实用性。
4、在一般的分类神经网络基础上，外加一条漏水漏油漏汽分割模块支路，用以输出漏水漏油漏汽的区域掩码，该支路输出空间注意力掩码结合到分类支路的特征上，提高分类的准确性。
5、在首次通过漏油、漏水和漏汽检测网络检测后，设置一个二次识别模块进行二次校验，降低检测的误报率，并且将漏油漏水二次识别和漏汽二次识别分开，其中漏水漏油的二次识别采用与前述的监测网络不同，采用其他的深度网络，比用相同网络结构识别具有更高的可靠性，而漏汽的二次识别通过图片的名称匹配到监测对象，再而匹配到对应的红外图像数据，并且设计了一个红图像识别方法进行识别，从而提高了蒸汽的检测准确性。
6、在识别到漏油、漏水和漏汽三种情况后，不单单只有报警提示，还会输出报警的摄像头名称及所监测的对象内容等。
应该理解的是，虽然上述各流程图中的各个步骤按照箭头的指示依次显示，但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，这些步骤可以以其它的顺序执行。而且，上述各流程图中的至少一部分步骤可以包括多个步骤或者多个阶段，这些步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，这 些步骤或者阶段的执行顺序也不必然是依次进行，而是可以与其它步骤或者其它步骤中的步骤或者阶段的至少一部分轮流或者交替地执行。
在一个实施例中，如图10所示，提供了一种电力生产异常监控装置，包括：
图像获取模块1001，用于获取设置在电力生产现场的摄像头采集的工作场景图像。
检测模块1002，用于根据所述工作场景图像对应台账对象的检测频率，将所述工作场景图像输入到检测模型，得到异常检测结果。
二次检测模块1003，用于在根据所述检测模型得到电力生产现场异常的检测结果时，根据异常类型调用对应的二次异常检测模型对工作现场进行二次检测。
报警模块1004，用于若二次检测仍为异常，则发出异常报警。
上述电力生产异常监控装置，通过采集生产现场的摄像头采集的工作场景图像，利用预先训练的检测模型，检测到异常后，再调用异常类型对应的二次异常检测模型对工作现场进行二次检测，利用神经网络的模型提取特征更高，进一步通过二次检测识别，提高检测的准确性和降低了虚警率。另外，采用该方法进行异常检测，只需在电力生产现场布置摄像头即可，摄像头的采集范围广，无需进行复杂布置，能够降低管理成本。
在另一个实施例中，检测模块，包括：
检测频率确定模块，用于获取所述工作场景图像对应台账对象的检测频率；
预测模块，用于当达到所述检测频率对应的检测时间时，将所述工作场景图像输入到检测模型，通过所述检测模型对电力生产现场是否出现异常进行检测，输出分类结果和分类结果为异常时从所述工作场景图像分割出来的异常区域分割结果；所述分类结果包括：正常、漏油、漏汽和漏水。
在另一个实施例中，二次检测模块，用于若异常类型为漏油和/或漏水，则将所述工作场景图像输入漏油漏水二次识别模型，所述漏油漏水二次识别模型采用残差网络结构，通过所述漏油漏水二次识别模型输出二次异常检测结果。
在另一个实施例中，二次检测模块，用于若异常类型为漏汽，则获取工作场 景图像对应台账对象的红外摄像头采集的红外图像；将所述红外图像输入红外图像识别模型，通过所述红外图像识别模型的特征提取网络，提取红外图像特征，将所述红外图像特征输入分类器，得到是否漏汽的二次异常检测结果。
在另一个实施例中，检测频率确定模块，用于获取所述工作场景图像的摄像头名称和预置位；根据所述摄像头名称和预置位匹配台账对象；若匹配到单个台账对象，则获取所述台账对象对应的检测频率；若匹配到多个台账对象，则获取多个所述台账对象中最小的检测频率。
在另一个实施例中，电力生产异常监控装置，还包括：
样本集处理模块，用于获取图像样本集，所述图像样本集包括样本，以及对样本各像素点标注的分割标签。
一次训练模块，用于根据所述图像样本集中标注的样本的分割标签，训练所述检测模型的特征提取网络和分割网络。
分类模块，用于根据训练的所述特征提取网络和分割网络，得到样本的分类标签。
二次训练模块，用于根据所述分类标签，对检测模型进行训练。
在另一个实施例中，分类模块，用于将样本输入到检测模型，通过所述检测模型中训练的特征提取网络，提取图像特征；将所述图像特征输入到训练的分割网络，输出异常区域的分割结果；根据所述分割结果，生成所述样本的分类标签。
关于电力生产异常监控装置的具体限定可以参见上文中对于电力生产异常监控方法的限定，在此不再赘述。上述电力生产异常监控装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
在一个实施例中，提供了一种计算机设备，该计算机设备可以是监控终端，其内部结构图可以如图11所示。该计算机设备包括通过系统总线连接的处理器、存储器、通信接口、显示屏和输入装置。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统和计算机程序。该内存储器为非易失性存储 介质中的操作系统和计算机程序的运行提供环境。该计算机设备的通信接口用于与外部的终端进行有线或无线方式的通信，无线方式可通过WIFI、运营商网络、NFC(近场通信)或其他技术实现。该计算机程序被处理器执行时以实现一种电力生产异常监控方法。该计算机设备的显示屏可以是液晶显示屏或者电子墨水显示屏，该计算机设备的输入装置可以是显示屏上覆盖的触摸层，也可以是计算机设备外壳上设置的按键、轨迹球或触控板，还可以是外接的键盘、触控板或鼠标等。
本领域技术人员可以理解，图11中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
在一个实施例中，提供了一种计算机设备，包括存储器和处理器，存储器中存储有计算机程序，该处理器执行计算机程序时实现上述各方法实施例中的步骤。
在一个实施例中，提供了一种计算机可读存储介质，其上存储有计算机程序，计算机程序被处理器执行时实现上述各方法实施例中的步骤。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的计算机程序可存储于一非易失性计算机可读取存储介质中，该计算机程序在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和易失性存储器中的至少一种。非易失性存储器可包括只读存储器(Read-Only Memory，ROM)、磁带、软盘、闪存或光存储器等。易失性存储器可包括随机存取存储器(Random Access Memory，RAM)或外部高速缓冲存储器。作为说明而非局限，RAM可以是多种形式，比如静态随机存取存储器(Static Random Access Memory，SRAM)或动态随机存取存储器(Dynamic Random Access Memory，DRAM)等。
以上实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实 施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对发明专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。

Claims (10)

  1. 一种电力生产异常监控方法，所述方法包括：
    获取设置在电力生产现场的摄像头采集的工作场景图像；
    根据所述工作场景图像对应台账对象的检测频率，将所述工作场景图像输入到预先训练的检测模型，得到异常检测结果；
    在根据所述检测模型得到电力生产现场异常的检测结果时，根据异常类型调用对应的二次异常检测模型对工作现场进行二次检测；
    若二次检测仍为异常，则发出异常报警。
  2. 根据权利要求1所述的方法，其特征在于，根据所述工作场景图像对应台账对象的检测频率，将所述工作场景图像输入到预先训练的检测模型，得到异常检测结果，包括：
    获取所述工作场景图像对应台账对象的检测频率；
    当达到所述检测频率对应的检测时间时，将所述工作场景图像输入到检测模型，通过所述检测模型对电力生产现场是否出现异常进行检测，输出分类结果和分类结果为异常时从所述工作场景图像分割出来的异常区域分割结果；所述分类结果包括：正常、漏油、漏汽和漏水。
  3. 根据权利要求1所述的方法，其特征在于，根据异常类型调用对应的二次异常检测模型对工作现场进行二次检测，包括：
    若异常类型为漏油和/或漏水，则将所述工作场景图像输入漏油漏水二次识别模型，所述漏油漏水二次识别模型采用残差网络结构，通过所述漏油漏水二次识别模型输出二次异常检测结果。
  4. 根据权利要求1所述的方法，其特征在于，根据异常类型调用对应的二次异常检测模型对工作现场进行二次检测，包括：
    若异常类型为漏汽，则获取工作场景图像对应台账对象的红外摄像头采集的红外图像；
    将所述红外图像输入红外图像识别模型，通过所述红外图像识别模型的特征提取网络，提取红外图像特征，将所述红外图像特征输入分类器，得到是否漏汽的二次异常检测结果。
  5. 根据权利要求2所述的方法，其特征在于，获取所述工作场景图像对应台账对象的检测频率，包括：
    获取所述工作场景图像的摄像头名称和预置位；
    根据所述摄像头名称和预置位匹配台账对象；
    若匹配到单个台账对象，则获取所述台账对象对应的检测频率；
    若匹配到多个台账对象，则获取多个所述台账对象中最小的检测频率。
  6. 根据权利要求1所述的方法，其特征在于，训练检测模型的方式，包括：
    获取图像样本集，所述图像样本集包括样本，以及对样本各像素点标注的分割标签；
    根据所述图像样本集中标注的样本的分割标签，训练所述检测模型的特征提取网络和分割网络；
    根据训练的所述特征提取网络和分割网络，得到样本的分类标签；
    根据所述分类标签，对检测模型进行训练。
  7. 根据权利要求6所述的方法，其特征在于，根据训练的所述特征提取网络和分割网络，得到样本的分类标签，包括：
    将样本输入到检测模型，通过所述检测模型中训练的特征提取网络，提取图像特征；
    将所述图像特征输入到训练的分割网络，输出异常区域的分割结果；
    根据所述分割结果，生成所述样本的分类标签。
  8. 一种电力生产异常监控装置，其特征在于，所述装置包括：
    图像获取模块，用于获取设置在电力生产现场的摄像头采集的工作场景图像；
    检测模块，用于根据所述工作场景图像对应台账对象的检测频率，将所述工作场景图像输入到检测模型，得到异常检测结果；
    二次检测模块，用于在根据所述检测模型得到电力生产现场异常的检测结果时，根据异常类型调用对应的二次异常检测模型对工作现场进行二次检测；
    报警模块，用于若二次检测仍为异常，则发出异常报警。
  9. 一种计算机设备，包括存储器和处理器，所述存储器存储有计算机程序，其特征在于，所述处理器执行所述计算机程序时实现权利要求1至7中任一项所述方法的步骤。
  10. 一种计算机可读存储介质，其上存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现权利要求1至7中任一项所述的方法的步骤。
PCT/CN2021/079282 2021-01-29 2021-03-05 电力生产异常监控方法、装置、计算机设备和存储介质 WO2022160413A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110130443.2A CN112669316B (zh) 2021-01-29 2021-01-29 电力生产异常监控方法、装置、计算机设备和存储介质
CN202110130443.2 2021-01-29

Publications (1)

Publication Number Publication Date
WO2022160413A1 true WO2022160413A1 (zh) 2022-08-04

Family

ID=75414987

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/079282 WO2022160413A1 (zh) 2021-01-29 2021-03-05 电力生产异常监控方法、装置、计算机设备和存储介质

Country Status (3)

Country Link
CN (1) CN112669316B (zh)
LU (1) LU502731B1 (zh)
WO (1) WO2022160413A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115420761A (zh) * 2022-11-07 2022-12-02 安徽启新明智科技有限公司 汽油与水区分网络的训练方法、汽油与水区分方法
CN115905807A (zh) * 2022-11-18 2023-04-04 中国市政工程西南设计研究总院有限公司 一种基于深度学习的粗格栅优化运行方法
CN116596934A (zh) * 2023-07-18 2023-08-15 华东交通大学 接地网腐蚀检测方法、系统、存储介质及移动终端
CN117312929A (zh) * 2023-11-29 2023-12-29 国网浙江省电力有限公司杭州供电公司 站房改造类型的辨识方法、装置、终端设备及存储介质
CN117612095A (zh) * 2023-11-27 2024-02-27 中国南方电网有限责任公司 一种电力安全控制方法及装置

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112907138B (zh) * 2021-03-26 2023-08-01 国网陕西省电力公司电力科学研究院 一种从局部到整体感知的电网场景预警分类方法及系统
CN114596308A (zh) * 2022-04-02 2022-06-07 卡奥斯工业智能研究院(青岛)有限公司 基于5g网络的信息处理方法、装置、设备及介质
CN118175428A (zh) * 2024-05-11 2024-06-11 江西政高技术服务有限公司 输电全景监控方法、系统及设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2010037261A1 (zh) * 2008-09-26 2010-04-08 中联绿盟信息技术(北京)有限公司 网络异常流量分析设备和方法
CN106226050A (zh) * 2016-07-15 2016-12-14 北京航空航天大学 一种tfds故障自动识别方法
CN205899784U (zh) * 2016-07-01 2017-01-18 深圳华盛创力科技有限公司 火力发电厂无线监测系统
CN111523659A (zh) * 2020-04-14 2020-08-11 北京华控智加科技有限公司 一种基于三级神经网络建模的机器故障预测诊断方法

Family Cites Families (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102053596B (zh) * 2010-09-26 2012-09-12 陕西宇宏国际建机有限公司 风力电厂和风电机舱的智能分析监控装置和方法
CN102722166B (zh) * 2012-06-29 2015-03-04 山东电力集团公司电力科学研究院 变电站设备智能视觉检测系统及状态检测方法
GB201711412D0 (en) * 2016-12-30 2017-08-30 Maxu Tech Inc Early entry
CN107945458A (zh) * 2017-12-11 2018-04-20 北京奇虎科技有限公司 摄像头场景下的跌倒检测方法、装置及计算设备
CN110443969B (zh) * 2018-05-03 2021-06-04 中移(苏州)软件技术有限公司 一种火点检测方法、装置、电子设备及存储介质
CN108921013B (zh) * 2018-05-16 2020-08-18 浙江零跑科技有限公司 一种基于深度神经网络的视觉场景识别系统及方法
CN109029731A (zh) * 2018-05-24 2018-12-18 河海大学常州校区 一种基于多目视觉的电力设备异常监测系统及方法
CN109101906A (zh) * 2018-07-27 2018-12-28 中国南方电网有限责任公司超高压输电公司贵阳局 一种变电站电力设备红外图像异常实时检测方法及装置
CN109543607A (zh) * 2018-11-22 2019-03-29 广州小狗机器人技术有限公司 目标物异常状态检测方法、系统、监护系统及存储介质
CN109993082B (zh) * 2019-03-20 2021-11-05 上海理工大学 卷积神经网络道路场景分类与道路分割方法
CN110163099B (zh) * 2019-04-17 2022-04-01 中国电子科技网络信息安全有限公司 一种基于电磁泄漏信号的异常行为识别装置与方法
CN110006435A (zh) * 2019-04-23 2019-07-12 西南科技大学 一种基于残差网络的变电站巡检机器人视觉辅助导航方法
CN110414313B (zh) * 2019-06-06 2024-02-13 平安科技（深圳）有限公司 异常行为告警方法、装置、服务器及存储介质
CN110334661A (zh) * 2019-07-09 2019-10-15 国网江苏省电力有限公司扬州供电分公司 基于深度学习的红外输变电异常发热点目标检测方法
CN110909630B (zh) * 2019-11-06 2023-04-18 腾讯科技（深圳）有限公司 一种异常游戏视频检测方法和装置
CN110889377A (zh) * 2019-11-28 2020-03-17 深圳市丰巢科技有限公司 广告物异常识别方法、装置、服务器设备及存储介质
CN111125405A (zh) * 2019-12-19 2020-05-08 国网冀北电力有限公司信息通信分公司 电力监控图像异常检测方法和装置、电子设备及存储介质
CN111209980B (zh) * 2019-12-25 2024-02-09 深圳供电局有限公司 环境检测方法和装置、电子设备、计算机可读存储介质
CN111126292A (zh) * 2019-12-25 2020-05-08 上海民办打一外国语小学 一种安全监控方法、装置及系统
CN111340843A (zh) * 2020-02-19 2020-06-26 山东大学 基于环境自适应和小样本学习的电力场景视频检测方法
CN111429304A (zh) * 2020-02-28 2020-07-17 鄂尔多斯市斯创网络科技有限责任公司 一种食品安全监管平台
KR102143231B1 (ko) * 2020-03-26 2020-08-10 주식회사 동진에스앤디 얼굴 감시 및 열화상 감시를 이용한 건강 이상 감지방법, 건강 이상 감지장치 및 컴퓨터 프로그램

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2010037261A1 (zh) * 2008-09-26 2010-04-08 中联绿盟信息技术(北京)有限公司 网络异常流量分析设备和方法
CN205899784U (zh) * 2016-07-01 2017-01-18 深圳华盛创力科技有限公司 火力发电厂无线监测系统
CN106226050A (zh) * 2016-07-15 2016-12-14 北京航空航天大学 一种tfds故障自动识别方法
CN111523659A (zh) * 2020-04-14 2020-08-11 北京华控智加科技有限公司 一种基于三级神经网络建模的机器故障预测诊断方法

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115420761A (zh) * 2022-11-07 2022-12-02 安徽启新明智科技有限公司 汽油与水区分网络的训练方法、汽油与水区分方法
CN115420761B (zh) * 2022-11-07 2023-02-24 安徽启新明智科技有限公司 汽油与水区分网络的训练方法、汽油与水区分方法
CN115905807A (zh) * 2022-11-18 2023-04-04 中国市政工程西南设计研究总院有限公司 一种基于深度学习的粗格栅优化运行方法
CN115905807B (zh) * 2022-11-18 2023-10-20 中国市政工程西南设计研究总院有限公司 一种基于深度学习的粗格栅优化运行方法
CN116596934A (zh) * 2023-07-18 2023-08-15 华东交通大学 接地网腐蚀检测方法、系统、存储介质及移动终端
CN116596934B (zh) * 2023-07-18 2023-09-26 华东交通大学 接地网腐蚀检测方法、系统、存储介质及移动终端
CN117612095A (zh) * 2023-11-27 2024-02-27 中国南方电网有限责任公司 一种电力安全控制方法及装置
CN117312929A (zh) * 2023-11-29 2023-12-29 国网浙江省电力有限公司杭州供电公司 站房改造类型的辨识方法、装置、终端设备及存储介质
CN117312929B (zh) * 2023-11-29 2024-02-13 国网浙江省电力有限公司杭州供电公司 站房改造类型的辨识方法、装置、终端设备及存储介质

Also Published As

Publication number Publication date
CN112669316B (zh) 2023-05-30
LU502731B1 (en) 2022-10-31
CN112669316A (zh) 2021-04-16

Similar Documents

Publication Publication Date Title
WO2022160413A1 (zh) 电力生产异常监控方法、装置、计算机设备和存储介质
CN108022235B (zh) 高压输电铁塔关键部件缺陷识别方法
CN111222423B (zh) 基于作业区域的目标识别方法、装置、计算机设备
CN110084165B (zh) 基于边缘计算的电力领域开放场景下异常事件的智能识别与预警方法
CN112380952A (zh) 基于人工智能的电力设备红外图像实时检测及识别方法
Guo et al. An improved AlexNet for power edge transmission line anomaly detection
Lestari et al. Fire hotspots detection system on CCTV videos using you only look once (YOLO) method and tiny YOLO model for high buildings evacuation
CN109858367B (zh) 工人通过支撑不安全行为的视觉自动化检测方法及系统
CN111582235A (zh) 用于实时监控站内异常事件的警报方法、系统及设备
CN113343779B (zh) 环境异常检测方法、装置、计算机设备和存储介质
CN110458794B (zh) 用于轨道列车的配件质量检测方法及装置
CN111815576B (zh) 金属部件腐蚀状况检测方法、装置、设备和存储介质
CN113723184A (zh) 基于智能网关的场景识别系统、方法、装置和智能网关
CN113569736A (zh) 刀闸状态识别方法、装置、计算机设备和存储介质
Rayhana et al. Automated defect-detection system for water pipelines based on CCTV inspection videos of autonomous robotic platforms
JP6954242B2 (ja) 定置式ガス検知装置の設置箇所の調査方法
CN113569956A (zh) 一种基于ai算法的山火灾害勘查识别方法
CN111010544B (zh) 配电房监控方法、装置、计算机设备和存储介质
CN117372954A (zh) 充电站安全监测方法、装置、系统和计算机设备
Xiao et al. Monitoring the Abnormal Human Behaviors in Substations based on Probabilistic Behaviours Prediction and YOLO-V5
Dai et al. Cascade centernet: Robust object detection for power line surveillance
CN116403162A (zh) 一种机场场面目标行为识别方法、系统及电子设备
CN113284103B (zh) 基于空间变换Faster R-CNN模型的变电站设备缺陷在线检测方法
CN114241400A (zh) 电网系统的监控方法及其装置、计算机可读存储介质
Olivatti et al. Analysis of artificial intelligence techniques applied to thermographic inspection for automatic detection of electrical problems

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21922012

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21922012

Country of ref document: EP

Kind code of ref document: A1