CN115909212A - Real-time early warning method for typical violation behaviors of power operation - Google Patents
Real-time early warning method for typical violation behaviors of power operation Download PDFInfo
- Publication number
- CN115909212A CN115909212A CN202211565004.5A CN202211565004A CN115909212A CN 115909212 A CN115909212 A CN 115909212A CN 202211565004 A CN202211565004 A CN 202211565004A CN 115909212 A CN115909212 A CN 115909212A
- Authority
- CN
- China
- Prior art keywords
- image
- real
- early warning
- violation
- warning method
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
Images
Classifications
-
- 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
Landscapes
- Image Analysis (AREA)
Abstract
A real-time early warning method for typical violation behaviors of power operation comprises the following steps: arranging a ball control distribution system on an operation site, and acquiring and storing a site video; meanwhile, according to a deep learning algorithm, performing neural network modeling on the operators; the acquisition host computer samples and identifies image information according to the collected video decoding and the Shannon theorem, analyzes the area and transmits the area to an edge calculation module and a network cloud monitoring platform of the monitoring host computer; processing the recognized image, confirming the abnormality of the operation site, comparing the abnormality through face recognition, and finally obtaining a violation conclusion; the violation behavior condition of the operation personnel at the operation site is recorded in time and fed back to the safety management personnel, and the violation behavior result is used for reminding the violation behavior worker to make adjustment in time in a mode of sharing the mobile phone terminal APP and the device host by the network cloud monitoring platform.
Description
Technical Field
The invention belongs to the technical field of monitoring and alarming, and particularly relates to a real-time early warning method for typical violation behaviors of electric power operation based on edge calculation and deep learning.
Background
At present, electric power operators often need to operate on a special operation site, but before the operation process, the operators are easy to be paralyzed and directly enter the site to work under the condition of not meeting the safety management standard of electric power operation, so that personal harm and public property loss are caused. The current method for improving the safety of workers is to install a large number of video monitoring devices at different angles of a construction site, and the video monitoring devices shoot the site operation image information of the current workers. The image information is transmitted back to the video monitoring center through the network, and the personnel in the video monitoring center identifies the current behavior of the operator and studies and judges whether the behavior of the operator violates the regulations.
However, such conventional manual screening methods have serious drawbacks. Firstly, the traditional method relies on manual visual identification, and human resources are limited, so that workers cannot be timely and effectively reminded when dealing with a large number of simultaneously-performed construction sites. Secondly, the conditions of large workload and line blockage can exist by utilizing the camera for monitoring, identifying, tracking and screening, a certain delay time can occur during alarming, and the personnel safety and equipment potential safety hazards can be caused because the operating personnel do not receive prompt in time within the delay time.
Disclosure of Invention
The invention aims to solve the problems that the monitoring alarm time period of the electric power operation site is long, the labor occupation ratio is high, the operation environment detection is single, and the operation is difficult to monitor at the same time, and provides a real-time early warning method for typical violation behaviors of the electric power operation, which can be used for various types, most quantities, multiple states and scenes for real-time monitoring of the construction operation site at the same time, so that the monitoring efficiency is improved, the accident rate is reduced, and the potential safety hazard of people and equipment is reduced.
The technical scheme of the invention is as follows:
a real-time early warning method for typical violation behaviors of power operation is characterized by comprising the following steps:
step 1, arranging a ball distribution and control system on an operation site, monitoring the site by using the ball distribution and control system, and acquiring and storing videos; meanwhile, according to a deep learning algorithm, neural network modeling is carried out on the operators, and the working state and the standard state of the operators are modeled and compared through long-time training to form a sample library;
step 2, the acquisition host samples image information according to the Shannon theorem according to video decoding collected by the ball distribution and control system, performs image recognition according to the dressing characteristics of field constructors, analyzes the region, and transmits the analyzed region to an edge calculation module and a network cloud monitoring platform of the device host in a digital signal coding mode;
step 3, the device host processes the recognized image through an edge calculation module, confirms the operation site abnormity, compares the operation site abnormity through face recognition, and finally obtains a violation conclusion;
and 4, timely recording the violation behaviors of the operating personnel at the operating site, feeding the violation behaviors back to the safety management personnel, and reminding the violation behaviors workers of timely adjustment in a mode of the mobile phone terminal APP and the device host by utilizing a network cloud monitoring platform according to a signal of a recorded result.
Further, the process of performing neural network modeling on the operator according to the deep learning algorithm in the step 1 is as follows:
the deep neural network in deep learning automatically extracts the target characteristics,
let w k 、w k+1 For k and k +1 trainsThe network weight vector, the weight adjustment quantity is
Δw k ＝w k+1 -w k (3)
According to the Gauss-Newton method, the following results are obtained:
wherein e (w) k ) Represents the weight error vector, J (w) k ) A jacobian matrix representing the error versus weight differential.
Further, the edge calculation of the edge calculation module in step 3 includes noise addition, filtering, enhancement and detection, and specifically includes:
(1) First, a gaussian-compliant distribution is added to the acquired image:
(2) Detecting the place with most obvious gray level change on the image, differentiating possible edge pixel points, judging whether the edge pixel points are edge pixel points or not by second order differentiation, dividing the edge types of the image into a step shape, a roof shape and a pulse shape according to the characteristics of the image, and utilizing a Gaussian filtering algorithm:
filtering the edges of three different shapes to remove redundant pixel points;
(3) Using an index enhancement algorithm for the rest pixel points, using a canny operator to carry out edge detection processing, transmitting the processed result to a cloud monitoring server in real time, sending the processed result to a neural network model trained by using abnormal data to obtain the probability of the abnormal data, and triggering an alarm device to remind workers on site when the probability is greater than a set threshold value; the cloud end feeds back the data transmitted from the physical end through the abnormal data obtained by calculation of the neural network, and triggers the alarm device when the risk of the abnormal data is high.
Further, when the image information is sampled according to shannon's theorem in step 2, when the sampling frequency satisfies:
2u max ≤|u s | (3)
sampling is performed in time, where u s 、v s For the frequency domain of the recovered analogue signal in both directions in the case where the signal is not true, u max 、v max Respectively, the highest angular frequency of the effective spectrum of the signal in the frequency domains of both directions.
Further, in the step 2, when the image information is sampled, the collected image information is firstly subjected to color separation by a signal processing system according to the principle of a subtractive method by utilizing the selective absorption characteristics of red, green and blue color filters on color lights with different wavelengths, and then the image information is sampled according to the Shannon theorem.
Furthermore, in the step 2, the collection host collects the images to be identified into image video data samples before the image identification, and automatically labels the samples by contrasting with a sample library. For the traditional manual labeling mode of identifying by naked eyes, manpower and material resources are consumed under the condition that the number of detection workers is small, once the number of detection workers is increased, the working efficiency is low, and missing or wrong results are easy to occur.
Further, when the image is identified in the step 2, firstly, a convolutional neural network is utilized to extract the features of the image to be identified, a primary equipment feature map is generated, then, a candidate area generation network is utilized to process the primary equipment feature map and output primary equipment feature candidate areas with various scales and aspect ratios, and finally, a classification regression network is utilized to perform judgment and output according to the features in the candidate areas.
Further, the violation conclusion includes the behavior of an operator who does not wear a safety helmet, an operator who does not wear a safety belt, and an operator who does not remove a grounding point.
Further, the image video data samples comprise power field operators, power equipment, safety helmets and insulating gloves, and normal operation samples and illegal operation samples are collected.
The invention has the beneficial effects that:
the invention combines the models of edge calculation and deep learning algorithm, analyzes the monitored signals, overcomes the defects of the traditional manual discrimination method on the quantity scale and time delay of the operators through target tracking and target identification, can greatly improve the monitoring efficiency, reduce the task load of the monitoring device, process the behavior images of the operators in time and remind in real time, and effectively reduce the accident rate of the operators on site and the loss of public property.
Compared with other monitoring systems, the invention has the advantages that: firstly, application scenes are enriched, an image recognition algorithm is optimized, manual operation time is shortened, and the proportion of manual operation is reduced, so that errors and omissions of manual operation are avoided. Secondly, the monitoring and early warning system is optimized by using double insurance of the edge calculation module and the deep learning algorithm, so that accurate calculation, real-time monitoring and timely feedback are achieved. Thirdly, neural network modeling is carried out through a deep learning algorithm, and violation behaviors such as no safety helmet wearing, violation operation and the like are detected and tracked. And fourthly, the obtained violation results are subjected to real-time information feedback in a dual mode of a mobile phone APP and a device host, and typical violation is alarmed in real time according to different operation items to ensure that violation personnel can be adjusted in time.
Drawings
FIG. 1 is a flow chart of the operation of the present invention;
FIG. 2 is a schematic diagram of the system architecture of the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and examples:
a real-time early warning method for typical violation behaviors of power operation is shown in figures 1 and 2 and comprises the following steps:
step 1, arranging a ball distribution and control system on an operation site, monitoring the site by using the ball distribution and control system, and acquiring and storing videos; meanwhile, according to a deep learning algorithm, neural network modeling is carried out on the operators, and the working state and the standard state of the operators are modeled and compared through long-time training to form a sample library;
step 2, the collection host carries out color separation on collected image information through a signal processing system according to a color reduction method principle by utilizing the selective absorption characteristics of red, green and blue color filters on color lights with different wavelengths, then samples the image information according to Shannon's theorem, collects images to be identified into image video data samples, wherein the image video data samples comprise power field operating personnel, power equipment, safety helmets and insulating gloves, collects normal operation samples and illegal operation samples, compares the samples with a sample library, and automatically carries out sample marking; then, image recognition is carried out according to the dressing characteristics of field constructors, a convolutional neural network is used for carrying out feature extraction on an image to be recognized to generate a primary equipment feature map, then a candidate area generation network is used for processing the primary equipment feature map and outputting primary equipment feature candidate areas with various scales and aspect ratios, finally a classification regression network is used for carrying out judgment and output according to features in the candidate areas to analyze the areas where the primary equipment feature map is located, and an acquisition host transmits the analyzed areas to an edge calculation module and a network cloud monitoring platform of a device host in a digital signal coding mode;
step 3, the device host processes the recognized image through an edge computing module, confirms the abnormity of the operation site, compares the abnormity through face recognition and finally obtains a violation conclusion; the violation conclusion includes the behavior of an operator who does not wear a safety helmet, an operator who does not wear a safety belt, and an operator who does not remove a grounding point.
And 4, the device host timely records the violation behaviors of the operating personnel at the operating site and feeds the violation behaviors back to the safety manager, and the violation behavior result is timely adjusted by the mobile phone terminal APP and the device host in a mode of warning the violation behavior workers in an alarm by utilizing a network cloud monitoring platform built based on the WiFi module in a signal recording result.
Further, the process of performing neural network modeling on the operator according to the deep learning algorithm in the step 1 is as follows:
the deep neural network in deep learning automatically extracts the target characteristics,
let w k 、w k+1 For the network weight vector of k times and k +1 times training, the weight adjustment amount is
Δw k ＝w k+1 -w k (5)
According to the Gauss-Newton method, the following results are obtained:
wherein e (w) k ) Represents the weight error vector, J (w) k ) A jacobian matrix representing the error versus weight differential.
Further, the edge calculation of the edge calculation module in step 3 includes noise addition, filtering, enhancement and detection, and specifically includes:
(1) First, a gaussian-compliant distribution is added to the acquired image:
(2) Detecting the place with most obvious gray level change on the image, differentiating possible edge pixel points, judging whether the edge pixel points are edge pixel points or not by second order differentiation, dividing the edge types of the image into a step shape, a roof shape and a pulse shape according to the characteristics of the image, and utilizing a Gaussian filtering algorithm:
filtering the edges of three different shapes to remove redundant pixel points;
(3) Using an index enhancement algorithm for the rest pixel points, using a canny operator to carry out edge detection processing, transmitting the processed result to a cloud monitoring server in real time, sending the processed result to a neural network model trained by using abnormal data to obtain the probability of the abnormal data, and triggering an alarm device to remind workers on site when the probability is greater than a set threshold value; the cloud end feeds back the data transmitted from the physical end through the abnormal data obtained by calculation of the neural network, and triggers the alarm device when the risk of the abnormal data is high.
Further, when the image information is sampled according to shannon's theorem in step 2, when the sampling frequency satisfies:
2u max ≤|u s | (3)
2v max ≤|v s sampling in case of | 4, where u s 、v s For the frequency domain of the recovered analogue signal in both directions in the case where the signal is not true, u max 、v max Respectively, the highest angular frequency of the effective spectrum of the signal in the frequency domains of both directions.
The main operation and comparison work is completed by the edge calculation module, so the module is mainly programmed, a deep reinforcement learning algorithm is designed through python language, the edge calculation module is trained, then the calculation module compares the dressing and behavior of operators with the standard, and finally the result is obtained by labeling in the image. For example, in the case of an illegal act of designing whether to wear a crash helmet, the worker who did not wear the crash helmet was designed to display a "person" character with the training data <0.8, and the worker who worn the crash helmet displayed a "hat" character with the training data >0.8.
The present invention is not limited to the above embodiments, but various modifications and changes can be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (9)
1. A real-time early warning method for typical violation behaviors of power operation is characterized by comprising the following steps:
step 1, arranging a ball distribution and control system on an operation site, monitoring the site by using the ball distribution and control system, and acquiring and storing videos; meanwhile, according to a deep learning algorithm, neural network modeling is carried out on the operators, and the working state and the standard state of the operators are modeled and compared through long-time training to form a sample library;
step 2, the acquisition host samples image information according to the Shannon theorem according to video decoding collected by the ball distribution and control system, performs image recognition according to the dressing characteristics of field constructors, analyzes the region, and transmits the analyzed region to an edge calculation module and a network cloud monitoring platform of the device host in a digital signal coding mode;
step 3, the device host processes the recognized image through an edge calculation module, confirms the operation site abnormity, compares the operation site abnormity through face recognition, and finally obtains a violation conclusion;
and 4, timely recording the violation behaviors of the workers at the operation site, feeding the violation behaviors back to a safety manager, and reminding the violation behaviors of workers to make adjustment in time in a common mode of the mobile phone terminal APP and the device host by using a network cloud monitoring platform according to a signal of a recorded result.
2. The real-time early warning method for typical violation behaviors in power operation as claimed in claim 1, wherein the process of modeling the neural network of the operator according to the deep learning algorithm in the step 1 is as follows:
the deep neural network in deep learning automatically extracts the target characteristics,
let w k 、w k+1 Network weight vector for k times and k +1 times of training, then weightThe value is adjusted by
Δw k ＝w k+1 -w k (1)
According to the Gauss-Newton method, the following results are obtained:
Δw k ＝-[J T (w k )J(w k )] -1 J(w k )e(w k ) (2)
wherein e (w) k ) Represents the weight error vector, J (w) k ) A jacobian matrix representing the error versus weight differential.
3. The real-time early warning method for typical violation behaviors in power operation as claimed in claim 1, wherein the edge calculation of the edge calculation module in step 3 comprises the steps of adding noise, filtering, enhancing and detecting, and specifically comprises the following steps:
1) First, a gaussian-compliant distribution is added to the acquired image:
of the Gaussian noise of, where μ, σ 2 The parameters of the distribution respectively represent the expectation and the variance of the Gaussian distribution;
2) Detecting the place with most obvious gray change on the image, differentiating possible edge pixel points, judging whether the edge pixel points are edge pixel points or not by second order differentiation, dividing the edge types of the image into a ladder shape, a roof shape and a pulse shape according to the characteristics of the image, and utilizing a Gaussian filter algorithm:
filtering the edges of three different shapes to remove redundant pixel points;
3) Using an index enhancement algorithm for the rest pixel points, using a canny operator to carry out edge detection processing, transmitting the processed result to a cloud monitoring server in real time, sending the processed result to a neural network model trained by using abnormal data to obtain the probability of the abnormal data, and triggering an alarm device to remind workers on site when the probability is greater than a set threshold value; the cloud end feeds back the data transmitted from the physical end through the abnormal data obtained by calculation of the neural network, and triggers the alarm device when the risk of the abnormal data is high.
4. The real-time early warning method for typical violation behaviors in power operation as claimed in claim 1, wherein in step 2, when the image information is sampled according to shannon's theorem, when the sampling frequency satisfies:
2u max ≤u s (3)
2v max ≤v s (4) Sampling is performed in time, where u s 、v s For the frequency domain of the recovered analogue signal in both directions in the case where the signal is not true, u max 、v max Respectively, the highest angular frequency of the effective spectrum of the signal in the frequency domains of both directions.
5. The real-time early warning method for typical violation behaviors in power operation as claimed in claim 1, wherein in the step 2, the collected image information is subjected to color separation by a signal processing system according to a subtractive principle by utilizing selective absorption characteristics of red, green and blue color filters on color lights with different wavelengths, and then the image information is sampled according to shannon's theorem.
6. The real-time early warning method for typical violation behaviors in power operation as claimed in claim 1, wherein in step 2, the collection host collects images to be identified into image video data samples before image identification, and automatically labels the samples against a sample library.
7. The real-time early warning method for typical violation behaviors in power operation as claimed in claim 1, wherein in the step 2, when the image is identified, a convolutional neural network is used for extracting the characteristics of the image to be identified to generate a primary equipment characteristic diagram, then a candidate area generating network is used for processing the primary equipment characteristic diagram and outputting primary equipment characteristic candidate areas with various scales and aspect ratios, and finally a classification regression network is used for distinguishing and outputting according to the characteristics in the candidate areas.
8. The real-time early warning method for typical violation behaviors of power operation as claimed in claim 1, wherein the violation conclusion in step 3 comprises the behaviors of an operator without wearing a safety helmet, an operator without wearing a safety belt and an action without removing a grounding point.
9. The real-time early warning method for typical violation behaviors in power operation as claimed in claim 6, wherein the image video data samples comprise power field operators, power equipment, safety helmets and insulating gloves, and normal operation and violation operation samples are collected.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211565004.5A CN115909212A (en) | 2022-12-07 | 2022-12-07 | Real-time early warning method for typical violation behaviors of power operation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211565004.5A CN115909212A (en) | 2022-12-07 | 2022-12-07 | Real-time early warning method for typical violation behaviors of power operation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115909212A true CN115909212A (en) | 2023-04-04 |
Family
ID=86476380
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211565004.5A Pending CN115909212A (en) | 2022-12-07 | 2022-12-07 | Real-time early warning method for typical violation behaviors of power operation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115909212A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117115923A (en) * | 2023-10-16 | 2023-11-24 | 罗普特科技集团股份有限公司 | Intelligent agriculture personnel behavior recognition system based on image recognition algorithm |
-
2022
- 2022-12-07 CN CN202211565004.5A patent/CN115909212A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117115923A (en) * | 2023-10-16 | 2023-11-24 | 罗普特科技集团股份有限公司 | Intelligent agriculture personnel behavior recognition system based on image recognition algorithm |
| CN117115923B (en) * | 2023-10-16 | 2023-12-22 | 罗普特科技集团股份有限公司 | Intelligent agriculture personnel behavior recognition system based on image recognition algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110826538B (en) | Abnormal off-duty identification system for electric power business hall | |
| CN112396658B (en) | Indoor personnel positioning method and system based on video | |
| CN106056079B (en) | A kind of occlusion detection method of image capture device and human face five-sense-organ | |
| CN106412501A (en) | Construction safety behavior intelligent monitoring system based on video and monitoring method thereof | |
| CN110096945B (en) | Indoor monitoring video key frame real-time extraction method based on machine learning | |
| CN114155614B (en) | Method and system for identifying anti-violation behavior of operation site | |
| CN111667655A (en) | Infrared image-based high-speed railway safety area intrusion alarm device and method | |
| CN115909212A (en) | Real-time early warning method for typical violation behaviors of power operation | |
| CN112036327A (en) | SSD-based lightweight safety helmet detection method | |
| CN115223249A (en) | Quick analysis and identification method for unsafe behaviors of underground personnel based on machine vision | |
| CN113743725B (en) | Ascending qualification detection method and system for ascending operators based on edge calculation | |
| CN113313186B (en) | Method and system for identifying irregular wearing work clothes | |
| CN112598865B (en) | Monitoring method and system for preventing cable line from being damaged by external force | |
| CN108776453B (en) | Building safety monitoring system based on computer | |
| CN106355187A (en) | Application of visual information to electrical equipment monitoring | |
| CN113688713A (en) | Cable protection method, device, system and medium based on dangerous behavior recognition | |
| CN116665419B (en) | Intelligent fault early warning system and method based on AI analysis in power production operation | |
| CN113179389A (en) | System and method for identifying crane jib of power transmission line dangerous vehicle | |
| CN111652128A (en) | High-altitude power operation safety monitoring method and system and storage device | |
| CN111339970A (en) | Smoking behavior detection method suitable for public environment | |
| CN116385925A (en) | Intelligent safety management method and system for production site | |
| CN115953815A (en) | Monitoring method and device for infrastructure site | |
| CN115860979A (en) | Artificial intelligence management system for field operation of power grid | |
| CN115830294A (en) | Identification method and device applied to personnel management and control in infrastructure construction operation field | |
| CN112419091B (en) | Intelligent video safety control method for field operation of power distribution network driven by knowledge graph |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |