CN116580353A

CN116580353A - Automatic early warning method and system for potential safety hazards of thermal power plant

Info

Publication number: CN116580353A
Application number: CN202310471023.XA
Authority: CN
Inventors: 贾顺杰; 赵菊; 米路中; 周越; 张艳萍; 李洋; 樊怡
Original assignee: Guoneng Xinkong Internet Technology Co Ltd
Current assignee: Guoneng Xinkong Internet Technology Co Ltd
Priority date: 2023-04-27
Filing date: 2023-04-27
Publication date: 2023-08-11

Abstract

The application provides an automatic early warning method and system for potential safety hazards of a thermal power plant. Based on the coal conveying belt of a large coal-fired unit, research and development are carried out on the basis of intelligent inspection and safety work normalization, advanced technologies such as deep learning, multi-mode data fusion and reinforcement learning are matched with civil air defense, physical air defense and technical air defense to form multi-dimensional three-dimensional safety protection, intelligent application is established by utilizing video AI intelligent analysis in the aspects of monitoring and early warning of the state of the coal conveying belt, automatic management of anti-violation of coal conveying personnel and the like, the production efficiency of enterprises is improved, the intensity of reducing personnel is reduced, the safety production and safety protection emergency response capability of a power-assisted power plant are realized, and a modern intelligent power plant is created.

Description

Automatic early warning method and system for potential safety hazards of thermal power plant

Technical Field

The application belongs to the field of image recognition, and particularly relates to an automatic early warning method and system for potential safety hazards of a thermal power plant.

Background

The coal conveying belt is adopted to convey the coal material, is an important mode for conveying the fuel of the thermal power plant, has the characteristics of long line, poor environment and the like, is usually adopted to carry out manual inspection, has the problems of high operation and maintenance cost, large workload, bad environment and the like, and can cause great influence on the safety production of enterprises if inaccurate judgment or timely discovery of belt defects is carried out on the current situation. In recent years, the detection and diagnosis of the safety failure of the coal conveying belt are realized by utilizing image analysis, and the method is a relatively effective mode, and is focused by a wide range of enterprises due to simple installation and low application cost.

At present, most of the gas turbine units of the thermal power plant are provided with near 200 paths of high-definition digital cameras, and are provided with independent control rooms of the system, the system usually uses 200 ten thousand pixels of digital cameras, 50 paths of high-definition digital cameras are additionally arranged in key areas for meeting the basic requirements on image shooting equipment, and based on the basic requirements, the system integrates an automatic early warning system for potential safety hazards of the thermal power plant, so that the full-function application of the system can be realized, the construction period can be shortened, the construction cost can be greatly reduced, and the system can be put into use early, so that the system is used for the safety production and the navigation protection of enterprises.

Chinese patent CN114120109a discloses a belt longitudinal tearing detection method based on a neural network, in which a convolutional neural network is constructed, a preset training set is marked with frames, and then image information of a coal conveying belt area is collected and input into a convolutional neural network algorithm model for judgment. However, the method only detects belt tearing, and the application range is too narrow.

Chinese patent CN114860893B discloses an intelligent decision method and apparatus based on multi-modal data fusion and reinforcement learning, which obtains an intelligent decision task including language instruction and visual information, and encodes the language instruction and the visual information to obtain multi-modal data; based on the multi-modal fusion method, multi-modal fusion data are obtained according to the multi-modal data, the multi-modal fusion data are input into a reinforcement learning algorithm, and based on the reinforcement learning algorithm and instant language rewards, actions are output and intelligent decisions are completed. However, the method requires a great deal of training and debugging, has high requirements on data quantity and computing resources, and has certain difficulty and cost for practical application.

Disclosure of Invention

In order to solve the defects in the prior art, the application provides an automatic early warning method and system for potential safety hazards of a thermal power plant. Based on the coal conveying belt of a large coal-fired unit, research and development are carried out on the basis of intelligent inspection and safety work normalization, advanced technologies such as deep learning, multi-mode data fusion and reinforcement learning are matched with civil air defense, physical air defense and technical air defense to form multi-dimensional three-dimensional safety protection, intelligent application is established by utilizing video AI intelligent analysis in the aspects of monitoring and early warning of the state of the coal conveying belt, automatic management of anti-violation of coal conveying personnel and the like, the production efficiency of enterprises is improved, the intensity of reducing personnel is reduced, the safety production and safety protection emergency response capability of a power-assisted power plant are realized, and a modern intelligent power plant is created.

The application adopts the following technical scheme.

An automatic early warning method for potential safety hazards of a thermal power plant comprises the following steps:

step 1, generating a multi-mode data set according to a coal conveying scene and personnel safety indexes;

step 2, fusing the image data sets in different forms by using a convolution automatic encoder to obtain a fused image;

step 3, extracting features of the fused image generated in the step 2 in a mode of fusing an attention mechanism, a circular convolution network and reinforcement learning;

and 4, classifying abnormal conditions of the coal conveying belt by adopting a deep convolutional neural network model based on the characteristics of the fusion image, and detecting the abnormal conditions based on time sequence data acquired by a sensor in the running process of the coal conveying belt.

Preferably, in step 1, the coal conveying scene includes coal conveying belt deviation and fracture data, foreign matter data and smoke data.

The personnel safety indexes comprise safety helmet data, out-of-range data and open flame data.

8. The automatic early warning method for potential safety hazards of a thermal power plant according to claim 1, which is characterized in that:

in step 2, the convolutional self-encoder network structure is divided into an encoding layer and a decoding layer; the coding layer comprises 4 groups of convolution layers, a pooling layer and two groups of full-connection layers, and the decoding layer comprises 4 groups of convolution, an up-sampling layer and one convolution layer.

Preferably, in step 2, the color map and the depth map of the same image are input as convolution from the encoder, and after passing through the encoding layer, two groups of vectors in 1504 dimensions are obtained; and serially combining the two groups of 1504-dimensional vectors to obtain 3008-dimensional vectors, and taking the fused vectors as the input of a decoding layer to obtain a fused image.

Preferably, in step 3, the input image is subjected to linear regression processing to obtain the position information of each pixel point in the image and the extracted significant features; and the fused attention mechanism adopts an adaptive pooling method to generate attention force diagrams of each mode, and performs point multiplication operation on the feature diagrams of each mode and the corresponding attention force diagrams to obtain weighted fused feature diagrams of the mode.

In step 3, the weighted fusion feature map is serialized using a bi-directional LSTM network.

The utility model provides an automatic early warning system of potential safety hazard of steam power plant, includes multimode dataset construction module, fuses module, reinforcement learning module.

The multi-mode data set construction module constructs a coal conveying scene and a personnel safety multi-mode data set, including coal conveying belt deviation and fracture identification; identifying a foreign matter; smoke identification; identifying a safety helmet; out-of-range identification; open flame identification;

the fusion module fuses the images in different forms by using a convolution automatic encoder to obtain a fused image;

the reinforcement learning module performs feature extraction on the fused image in a mode of fusing an attention mechanism, a circular convolution network and reinforcement learning.

Compared with the prior art, the intelligent safety warning system has the advantages that the intelligent safety warning system mainly aims at the potential safety hazards of the gas turbine unit in the safety production process, the intelligent safety warning system is used for preventing and warning in advance by utilizing the intelligent technology of the Internet of things, intelligent dangerous recognition means are arranged, some safety accidents of human negligence are greatly reduced, the intelligent Internet of things coverage and real-time monitoring are carried out on the whole factory, the intelligent sensors are integrated into the background central processing unit for each area with the potential safety hazards, and the automatic recognition of safety and intelligent warning of the system are realized. The system can accurately identify wearing behaviors, fire disaster early warning, illegal actions, regional division, off-duty monitoring, camera abnormality and the like, realizes artificial intelligence in the whole process, realizes scientific and technological management of safe production, and provides an efficient management tool for managers.

Drawings

FIG. 1 is a schematic flow chart of an automatic early warning method for potential safety hazard of a thermal power plant;

FIG. 2 is a schematic diagram of reinforcement learning feature extraction in the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the technical solutions of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. The described embodiments of the application are only some, but not all, embodiments of the application. All other embodiments, which can be made by those skilled in the art without inventive faculty, are within the scope of the application, based on the spirit of the application.

An automatic early warning method for potential safety hazards of a thermal power plant, as shown in fig. 1, comprises the following specific steps:

step 1, constructing a coal conveying scene and a personnel safety multi-mode data set;

in step 1, the coal conveying scene data set includes:

1) Deviation and fracture data of coal conveying belt

Through installing the camera on the coal conveying line, carry out the multi-angle and shoot, carry out real-time data acquisition to coal conveying belt running condition, discern abnormal data to data acquisition center uploads.

2) Foreign matter data

Through easily appearing the foreign matter and drop the position installation camera on the coal line, carry out the analysis to the object that the video image of camera appears, discern when stone, ironware, timber or other foreign matter appear.

3) Smoke data

And carrying out smoke recognition on the lens analysis area, finding out abnormal smoke and fire seedlings in the monitoring area in time, alarming and providing useful information.

The personal safety dataset comprises:

1) Helmet data

And carrying out safety helmet identification in the lens analysis area, supporting the detection of the safety helmet of the head of the pedestrian, and carrying out audible and visual alarm if the safety helmet is not worn.

2) Out-of-range data

The detection and tracking of personnel are carried out in the area, the unsafe area is set in the video shooting range for detection, an alarm is generated when someone invades, meanwhile unsafe behaviors such as a coal conveyor, a dust mask is not worn on the coal conveyor and the like are detected for the video personnel, the faces appearing in the video images are detected and identified in a set time period, the condition that the personnel arrive on duty is automatically recorded, and the size of the pedestrians is not less than 150x60 pixels in a 640x360 picture.

3) Open flame data

And (5) performing flame identification in the lens, and alarming after finding out the flame. In the 640x360 frame, the flame detection size is not less than 40x40 pixels.

Step 2, fusing the images in different forms by using a convolution automatic encoder to obtain a fused image;

in step 2, the convolution is separated from the encoder network structure into an encoding layer and a decoding layer. The coding layer comprises 4 groups of convolution layers and pooling layers and two groups of full connection layers. The decoding layer includes 4 sets of convolution+upsampling layers and one convolution layer. The original images r and d are respectively a color image and a depth image of the same image and have complementary relations. The color map and the depth map are simultaneously input as convolution self-encoder, and after passing through the encoding layer, two groups of vectors with 1504 dimensions are obtained. Two sets of vectors are input to the decoder separately, resulting in two reconstructed images, which can be considered as approximate reconstructions of the image represented by the two sets of vectors. And serially combining the two groups of 1504-dimensional vectors to obtain 3008-dimensional vectors, and taking the fused vectors as the input of a decoding layer to obtain a fused image. Specifically, the vector fusion process includes the steps of:

step 2.1: the vector is standardized to ensure that the similarity between the vectors is comparable, and the two groups of vectors are weighted and fused by utilizing similarity-based fusion;

step 2.2: taking the weighted and fused vector as input, and decoding the vector by a decoder to obtain a fused image;

in step 2.2, the specific implementation of the decoder includes a multi-layer neural network, and the internal parameters of the decoder are obtained through training, so that the layer-by-layer decoding and reconstruction of the input vector can be realized, and finally a synthesized fusion image is obtained. The formula is as follows:

min(r-D_r(E_r(r)))

min(d-D_d(E_d(d)))

taking image r as an example, e_r is the encoding process of image r, features are extracted from r, d_r is the decoding process of image r, and the reconstructed image is decoded by the features. Images r and d respectively learn the respective characteristics, the characteristics are connected in series at the tail end of the coding layer, and the decoding layer shares weight, so that the joint unified expression of the characteristics of different modes is realized. The loss function is the difference between the original image and the reconstructed image, and is formulated as follows:

wherein m represents m groups of data in total, r ⁱ And d ⁱ Representing the original color map and depth map respectively,and->Representing the reconstructed color map and depth map. When multi-mode data fusion is carried out by using convolution self-coding, a full connection layer cannot be deleted or a convolution layer is used for replacing, otherwise, useful characteristics cannot be learned, and images cannot be generated normally.

The multi-mode fusion of the thermal power plant data sets can enhance the picture quality under the condition of poor light effect, and improve the recognition accuracy and precision of the coal conveying belt.

And 3, extracting the characteristics of the fused image generated in the step 2 in a mode of fusing an attention mechanism, a circular convolution network and reinforcement learning.

As shown in fig. 2, linear represents the feature after the Linear regression process, loc represents the position information, fx represents the extracted feature, and TD represents the differential algorithm of reinforcement learning. The specific rules are as follows:

linear regression processing yields positional information and extracts salient features: and carrying out linear regression processing on the input image to obtain the position information of each pixel point in the image and the extracted significant features. These salient features include information on color, texture, shape, etc., which can effectively distinguish and classify images.

The correct positions are rewarded and classified by using a difference algorithm TD of reinforcement learning, all the remarkable characteristics of the whole picture are iterated continuously, the position, color and level space related information are combined and transmitted into a countermeasure network model for training, and abnormal conditions are marked by using a machine translation technology, so that real-time reminding is realized; the step utilizes a reinforcement learning algorithm to classify each pixel of the image, and the correct classification position is rewarded according to a difference algorithm TD, so that the classification result is optimized. And the abnormal conditions are marked by using a machine translation technology, so that real-time reminding of foreign matters, belt breakage, belt deviation and personnel safety feature classification in the coal conveying scene is realized.

And fusing the attention mechanisms, generating attention force diagrams of each mode by adopting an adaptive pooling method, and performing point multiplication operation on the feature diagrams of each mode and the corresponding attention force diagrams to obtain weighted fusion feature diagrams of the mode: the step uses an attention mechanism, and feature maps of different modes can be weighted according to the importance of the feature maps in the fusion process. And generating attention force diagram of each mode through an adaptive pooling method, and performing point multiplication operation on the attention force diagram and the corresponding feature diagram to obtain a weighted fusion feature diagram of the mode.

Serializing the weighted fusion feature graphs by using a bidirectional LSTM network to obtain more discriminative feature representation; the step uses a bidirectional LSTM network to sequence the weighted fusion feature graphs so as to obtain more abstract and more discriminant feature representations. The LSTM network can effectively learn long-term dependencies in the feature map, which helps to further improve the accuracy of classification or regression tasks.

And training an agent by adopting a Q-learning algorithm, so that the agent can select the optimal action according to the current state, and obtain corresponding rewards. The state of the agent is represented by multi-modal features, the actions include modifying neural network structure and parameters, rewarding performance based on classification or regression tasks; the step uses Q-learning algorithm to train an agent, so that the agent can select the optimal action according to the current state and obtain corresponding rewards. The state of the agent is represented by a multi-modal feature and the actions include modifying the neural network structure and parameters. The rewards are based on the performance of classification or regression tasks, and the accuracy of the classification or regression tasks is further improved through iterative optimization of the neural network. Finally, the serialized features are mapped into a full connection layer to complete classification or regression tasks.

And 4, classifying abnormal conditions of the coal conveying belt by adopting a deep convolutional neural network model, and detecting the abnormal conditions based on time sequence data acquired by a sensor in the running process of the coal conveying belt.

By constructing a real-time monitoring system. Specifically, a GPU-accelerated computer cluster may be used to assign multi-modal feature inputs to different computing nodes for processing to increase processing speed. Meanwhile, a Deep Convolutional Neural Network (DCNN) model can be adopted to classify the coal conveying belt, and abnormality detection can be carried out based on time sequence data. In the classification task, the features extracted in the step 3 are directly input into the DCNN network for classification. In the abnormality detection task, time context information, such as the state of the coal conveyor belt in the previous period of time, needs to be considered, so as to more accurately determine whether an abnormality exists at present. Finally, whether the alarm mechanism is triggered or not is judged by setting a threshold value.

Finally, the data and abnormal conditions of each monitoring need to be recorded and stored in a database. Meanwhile, the monitoring result can be fed back to the management department of the thermal power plant so as to help the management department to better master the running condition of the coal conveying belt. In addition, the historical data can be analyzed by adopting data mining and machine learning technologies, and a monitoring algorithm and a neural network model are further optimized, so that the early warning accuracy is improved, and the false alarm rate is reduced.

The present disclosure may be a system, method, and/or computer program product. The computer program product may include a computer readable storage medium having computer readable program instructions embodied thereon for causing a processor to implement aspects of the present disclosure.

The computer readable storage medium may be a tangible device that can hold and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium would include the following: portable computer disks, hard disks, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static Random Access Memory (SRAM), portable compact disk read-only memory (CD-ROM), digital Versatile Disks (DVD), memory sticks, floppy disks, mechanical coding devices, punch cards or in-groove structures such as punch cards or grooves having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media, as used herein, are not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., optical pulses through fiber optic cables), or electrical signals transmitted through wires.

The computer readable program instructions described herein may be downloaded from a computer readable storage medium to a respective computing/processing device or to an external computer or external storage device over a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmissions, wireless transmissions, routers, firewalls, switches, gateway computers and/or edge servers. The network interface card or network interface in each computing/processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium in the respective computing/processing device.

Computer program instructions for performing the operations of the present disclosure can be assembly instructions, instruction Set Architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, c++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer readable program instructions may be executed entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, aspects of the present disclosure are implemented by personalizing electronic circuitry, such as programmable logic circuitry, field Programmable Gate Arrays (FPGAs), or Programmable Logic Arrays (PLAs), with state information of computer readable program instructions, which can execute the computer readable program instructions.

Finally, it should be noted that the above embodiments are only for illustrating the technical solution of the present application and not for limiting the same, and although the present application has been described in detail with reference to the above embodiments, it should be understood by those skilled in the art that: modifications and equivalents may be made to the specific embodiments of the application without departing from the spirit and scope of the application, which is intended to be covered by the claims.

Claims

1. The automatic early warning method for potential safety hazards of the thermal power plant is characterized by comprising the following steps of:

2. The automatic early warning method for potential safety hazards of a thermal power plant according to claim 1, which is characterized in that:

in the step 1, the coal conveying scene comprises coal conveying belt deviation and fracture data, foreign matter data and smoke data.

3. The automatic early warning method for potential safety hazards of a thermal power plant according to claim 1, which is characterized in that:

in the step 1, the personnel safety index comprises safety helmet data, out-of-range data and open flame data.

4. The automatic early warning method for potential safety hazards of a thermal power plant according to claim 1, which is characterized in that:

5. The automatic early warning method for potential safety hazards of a thermal power plant according to claim 1, which is characterized in that:

in step 2, the color image and the depth image of the same image are input as convolution self-encoder, and after passing through the encoding layer, two groups of vectors with 1504 dimensions are obtained; and serially combining the two groups of 1504-dimensional vectors to obtain 3008-dimensional vectors, and taking the fused vectors as the input of a decoding layer to obtain a fused image.

6. The automatic early warning method for potential safety hazards of a thermal power plant according to claim 1, which is characterized in that:

in the step 3 of the method, in the step (3),

performing linear regression processing on the input image to obtain the position information of each pixel point in the image and the extracted significant features;

and the fused attention mechanism adopts an adaptive pooling method to generate attention force diagrams of each mode, and performs point multiplication operation on the feature diagrams of each mode and the corresponding attention force diagrams to obtain weighted fused feature diagrams of the mode.

7. The automatic early warning method for potential safety hazards of a thermal power plant according to claim 1 or 6, which is characterized in that:

8. The utility model provides a steam power plant potential safety hazard automatic early warning system for realize a steam power plant potential safety hazard automatic early warning method according to claim 1-7, includes multimode dataset construction module, fusion module, reinforcement learning module, its characterized in that:

9. A terminal comprising a processor and a storage medium; the method is characterized in that:

the storage medium is used for storing instructions;

the processor being operative according to the instructions to perform the steps of the method according to any one of claims 1-7.

10. Computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the method according to any of claims 1-7.