WO2024060284A1 - 高炉炉顶料面温度分布的识别方法、装置及存储介质 - Google Patents

高炉炉顶料面温度分布的识别方法、装置及存储介质 Download PDF

Info

Publication number
WO2024060284A1
WO2024060284A1 PCT/CN2022/122610 CN2022122610W WO2024060284A1 WO 2024060284 A1 WO2024060284 A1 WO 2024060284A1 CN 2022122610 W CN2022122610 W CN 2022122610W WO 2024060284 A1 WO2024060284 A1 WO 2024060284A1
Authority
WO
WIPO (PCT)
Prior art keywords
temperature measurement
infrared
cross temperature
neural network
data
Prior art date
Application number
PCT/CN2022/122610
Other languages
English (en)
French (fr)
Inventor
严晗
欧燕
叶理德
Original Assignee
中冶南方工程技术有限公司
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 中冶南方工程技术有限公司 filed Critical 中冶南方工程技术有限公司
Publication of WO2024060284A1 publication Critical patent/WO2024060284A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01J—MEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J5/00—Radiation pyrometry, e.g. infrared or optical thermometry
    • G01J5/48—Thermography; Techniques using wholly visual means
    • G01J5/485—Temperature profile
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01J—MEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J5/00—Radiation pyrometry, e.g. infrared or optical thermometry
    • G01J5/0014—Radiation pyrometry, e.g. infrared or optical thermometry for sensing the radiation from gases, flames
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01K—MEASURING TEMPERATURE; MEASURING QUANTITY OF HEAT; THERMALLY-SENSITIVE ELEMENTS NOT OTHERWISE PROVIDED FOR
    • G01K13/00—Thermometers specially adapted for specific purposes
    • G01K13/02—Thermometers specially adapted for specific purposes for measuring temperature of moving fluids or granular materials capable of flow
    • G01K13/024—Thermometers specially adapted for specific purposes for measuring temperature of moving fluids or granular materials capable of flow of moving gases
    • 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
    • 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
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/02—Affine transformations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/20—Image enhancement or restoration using local operators
    • 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
    • G01—MEASURING; TESTING
    • G01J—MEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J5/00—Radiation pyrometry, e.g. infrared or optical thermometry
    • G01J2005/0077—Imaging
    • 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
    • G06T2207/10—Image acquisition modality
    • G06T2207/10048—Infrared image
    • 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]

Definitions

  • the present invention relates to the field of blast furnace ironmaking detection technology, and in particular to a method, device and storage medium for identifying the temperature distribution of the top surface of a blast furnace.
  • the distribution of blast furnace gas flow in the furnace directly affects the temperature distribution in the furnace, the position and shape of the soft melt zone, the gas utilization rate, and the stability of the furnace conditions. It ultimately affects the economic indicators of the blast furnace and obtains the ideal gas flow distribution. It is an important goal in blast furnace operation. Therefore, timely and effective understanding of the distribution of gas flow in the furnace is of great significance to blast furnace operation.
  • the temperature distribution of the blast furnace top surface most directly reflects the distribution of the gas flow. Generally, analysis of the blast furnace production process believes that where the gas flow develops strongly, the gas temperature is higher; conversely, the gas temperature is higher. The temperature is lower.
  • the current indirect detection methods for the furnace top temperature distribution mainly include the furnace throat cross temperature measurement device and the furnace top infrared camera.
  • the former generally installs four temperature measurement arms in four directions on the circumferential surface of the furnace throat, with 5 to 7 temperature sensors distributed on the temperature measurement arms. These temperature measurement points can provide real-time and accurate temperature data and can accurately reflect the distribution of gas flow in the four directions of the furnace throat.
  • the number of temperature measurement points is limited and it is impossible to fully grasp the temperature distribution of the entire material surface; the latter is not Contact infrared temperature measurement device can detect the temperature distribution changes of the entire material surface in real time in the form of infrared images.
  • the main disadvantage of infrared imaging is that the accuracy of the measurement values is difficult to guarantee.
  • the embodiment of the present invention provides a method for identifying the temperature distribution of the blast furnace top material surface, which combines two measurement methods, the cross temperature measurement and the infrared camera, and utilizes the advantages of the two measurement methods, the cross temperature measurement and the infrared camera. To obtain accurate and comprehensive temperature distribution information on the furnace top material surface.
  • a method for identifying the temperature distribution of the blast furnace top surface is provided.
  • a furnace throat cross temperature measurement device and a furnace top infrared camera are installed on the top of the blast furnace. Through the cross temperature measurement device, Collect cross temperature measurement data and collect infrared detection data through the infrared camera.
  • the identification method includes:
  • the deep learning neural network model established includes:
  • a generator comprising an encoder and a decoder, wherein the encoder is used to fuse the input cross temperature measurement data and infrared detection data, and input the fusion result into the decoder, and the decoder calculates and outputs the recognition result of the furnace top material surface temperature distribution according to the fusion result;
  • a discriminator used to distinguish the recognition result output by the generator and the infrared detection data and conduct confrontation training between the recognition result output by the generator and the infrared detection data;
  • training the deep learning neural network model includes:
  • the loss function of the deep learning neural network model meets the preset conditions, it is determined that the deep learning neural network model is trained, the training is stopped and the hyperparameters of the generator at this time are saved;
  • cross temperature measurement data is collected through the cross temperature measurement device, and the step of collecting infrared detection data through the infrared camera includes:
  • the cross temperature measurement data of the cross temperature measurement device at the same time is collected.
  • the cross temperature measurement data is collected through the cross temperature measurement device, and the step of collecting infrared detection data through the infrared camera further includes:
  • the infrared image on the horizontal plane is cropped so that the furnace core is in the center of the picture and the furnace throat outline is close to the edge of the image to obtain a two-dimensional array of m*n size, where m and n are natural numbers, and the sizes of m and n are determined by the infrared
  • the resolution of the camera is determined;
  • the cross temperature measurement data and the infrared detection data input to the deep learning neural network model are the cross temperature measurement array and the infrared array.
  • the cross temperature measuring device includes four temperature measuring arms arranged in four directions, each temperature measuring arm is provided with a predetermined number of temperature measuring sensors, and the interpolation processing of the collected cross temperature measuring data includes:
  • the temperature measurement values in the four directions obtained by the cross temperature measuring device are interpolated by a pre-selected interpolation method to obtain a temperature smooth curve of the temperature measurement values in the four directions.
  • the encoder fuses the cross temperature measurement data and infrared detection data after the same dimension stacking, And output the fusion feature map;
  • the encoder is a densely connected feature structure in DenseNet, including 5 convolutional layers, and the input of each convolutional layer is composed of a cascade of channels output by all layers before the convolutional layer.
  • the decoder includes 4 convolutional layers;
  • each of the convolutional layers is a 3*3 convolution kernel
  • the convolution step size is 1
  • the activation function uses the linear rectification unit ReLU
  • batch normalization BN is performed before the linear rectification unit to convert
  • the input to the linear rectifier unit is normalized to a standard normal distribution with mean 0 and variance 1.
  • the recognition method wherein the loss function of the generator is:
  • G represents the generator, Represents the adversarial loss, L cont is the content loss, and ⁇ is the weight coefficient;
  • E[] represents mathematical expectation
  • log is a logarithmic function
  • D V () represents the output result of the discriminator
  • G(V,I) is the output result of the generator
  • V is the input infrared array
  • I is the input Cross temperature measurement array
  • L cont is calculated using the following formula:
  • m, n is the size of the input array
  • e is the natural index
  • G(V,I) j represents the j-th point in G(V,I)
  • I k represents the k-th point in the cross temperature measurement array I
  • d j,k represents the Euclidean distance between G(V,I) j and I k
  • eta is the weight coefficient
  • TV represents the TV-norm.
  • the discriminator D V includes 3 layers of convolutional layers and 1 layer of fully connected layers, and the loss function of the discriminator D V is:
  • the identification method when the change of the loss function of the generator and the discriminator D V on the verification set drops to a stable level, it is determined that the deep learning neural network model has been trained.
  • a device for identifying the temperature distribution of the blast furnace top surface including a memory and a processor.
  • the memory stores at least one program, and the at least one program is executed by the processor to implement any of the above. method described.
  • a computer-readable storage medium in which at least one program is stored, and the at least one program is executed by a processor to implement any of the identification methods described above.
  • the detection data is used to train the deep learning neural network based on the generative adversarial network, and a neural network model used to identify the temperature distribution of the furnace top material surface is obtained; in After collecting the latest detection data from the furnace throat cross temperature measurement device and the furnace top infrared camera, the two-dimensional array data is processed and input into the trained neural network model. After the model completes data fusion, the temperature distribution of the furnace top material surface can be obtained. Information, thus completing the identification of the temperature distribution of the furnace top material surface.
  • the identification method provided by the embodiment of the present invention can provide the temperature distribution of the furnace top material surface, has the advantages of high accuracy, strong real-time performance, and good visualization effect, and can be effectively used in the detection of the temperature of the blast furnace top material surface, and provides useful information for blast furnace operations. It provides an accurate basis for judgment and furnace condition diagnosis, which has direct guiding significance and contributes to the long-term safe and stable operation of the blast furnace.
  • Figure 1 is a schematic flow chart of a method for identifying the temperature distribution of the blast furnace top material surface according to an embodiment of the present invention
  • Figure 2 is an example of an infrared image of the furnace top in the identification method according to an embodiment of the present invention
  • FIG3 is a schematic diagram of a furnace throat cross temperature measuring device in an identification method according to an embodiment of the present invention.
  • Figure 4 is a schematic diagram of affine transformation used in the identification method according to an embodiment of the present invention.
  • Figure 5 is a schematic structural diagram of the deep learning neural network model established in the recognition method according to the embodiment of the present invention.
  • Figure 6 is a schematic diagram of the working process of the deep learning neural network model in the identification method according to the embodiment of the present invention.
  • Figure 7 is a schematic structural diagram of a device for identifying the temperature distribution of a blast furnace top material surface according to an embodiment of the present invention.
  • Figure 1 is a schematic flowchart of a method for identifying the temperature distribution of the blast furnace top material surface according to an embodiment of the present invention.
  • a furnace throat cross temperature measurement device and a furnace top infrared camera are installed on the top of the blast furnace.
  • the cross temperature measurement data is collected through the cross temperature measurement device, and the infrared detection data is collected through the infrared camera, as shown in Figure 1.
  • the identification method of this embodiment includes:
  • the deep learning neural network model established includes:
  • the generator includes an encoder and a decoder.
  • the encoder is used to fuse the input cross temperature measurement data and infrared detection data, and input the fusion results into the decoder.
  • the decoder calculates based on the fusion results and outputs the identification of the temperature distribution of the furnace top surface. result;
  • the discriminator is used to distinguish the recognition results output by the generator and the infrared detection data and conduct confrontation training between the recognition results output by the generator and the infrared detection data;
  • training the deep learning neural network model includes:
  • the loss function of the deep learning neural network model meets the preset conditions, it is determined that the deep learning neural network model is trained, stops training and saves the hyperparameters of the generator at this time;
  • the implementation of the embodiment of the present invention requires that the top of the blast furnace be equipped with a furnace throat cross temperature measurement device and a furnace top infrared camera.
  • the deep learning neural network model is trained by collecting data measured by the furnace throat cross temperature measurement device and the furnace top infrared camera respectively. and use trained neural network models.
  • the furnace throat cross temperature measurement device and the furnace top infrared camera used are those of the prior art, and their installation method can also adopt the installation method of the prior art.
  • the above collection steps are continuously performed.
  • the instruction to collect infrared images can be issued by a superior control system, such as the PLC controller in the first-level blast furnace system, before the cloth starts, during the cloth distribution, or after the cloth is completed, so as to collect the above-mentioned normal infrared images.
  • the method for identifying the temperature distribution of the blast furnace top material surface in the embodiment of the present invention is mainly realized by combining two means: a cross temperature measurement device and a furnace top infrared camera.
  • the deep learning neural network is used to obtain the temperature distribution through fusion and processing. temperature measurement data to obtain more accurate identification results.
  • the cross temperature measurement data at the same time is collected.
  • the blast furnace top is required to be equipped with a furnace throat cross temperature measuring device and a furnace top infrared camera. Continuously perform the following steps: when the blast furnace is not in operation, collect normal images taken by the infrared camera that are not blocked by the chute and have no obvious interference. Whenever the latest normal infrared image is acquired, the cross temperature measurement data at the same time is collected.
  • Figure 2 gives an example of an infrared image of a stove top.
  • Image data is generally represented as a two-dimensional matrix array, in which the value of each pixel represents the temperature of the pixel.
  • FIG 3 is a schematic diagram of an exemplary furnace throat cross temperature measuring device.
  • the furnace throat cross temperature measurement device includes temperature measurement arms in four directions, of which 3 temperature measurement arms have 5 temperature measurement points, and one temperature measurement arm has 6 temperature measurement points, a total of 21 Temperature measurement point.
  • the furnace throat cross temperature measuring device is only exemplary and is used to illustrate the implementation of the embodiments of the present invention.
  • the implementation of the embodiments of the present invention is not limited to this specific structure. It can also have different numbers and temperature measurement points. location and/or different orientation. For example, there can be 17 temperature measurement points.
  • the temperature measurement value of each temperature measurement point can accurately reflect the airflow temperature at that location.
  • the infrared array After obtaining the infrared image data, use mean filtering to preprocess the infrared image, and perform a certain affine transformation on the array to obtain a two-dimensional array of infrared images on the horizontal plane; crop the image so that the furnace core is in the center of the screen and the furnace throat outline is Close to the edge of the image, a two-dimensional infrared array of m*n size is obtained, referred to as the infrared array.
  • m and n are natural numbers, and the size of m and n is determined by the resolution of the infrared camera. For example, whether the furnace throat contour is close to the image edge can be judged by whether the direct distance between the furnace throat contour and the image edge is less than a predetermined distance threshold.
  • the cross temperature measurement array Generate a two-dimensional array with the same size m*n as the infrared image and filled with 0 values. Based on the infrared image coordinate system after affine transformation, the temperature measurement values obtained by cross temperature measurement in four directions are interpolated. The obtained smooth curve is filled into the corresponding pixel point in the array according to the spatial position of the smooth curve, thus obtaining a two-dimensional cross temperature measurement array that reflects the cross temperature measurement results, referred to as the cross temperature measurement array.
  • the affine transformation steps are as follows: select 4 points on the edge of the same height plane inside the furnace throat, find their corresponding coordinates in the infrared image, and combine the actual positions of these four points in the blast furnace according to the perspective transformation principle coordinates, and their coordinates in the infrared image, the perspective transformation matrix T can be solved, and then the transformation matrix T is used to perform affine transformation on the infrared image to obtain an infrared image looking from top to bottom.
  • This transformation process is shown in Figure 4.
  • the exemplary interpolation process steps are as follows: interpolate the temperature values in the four directions obtained by the cross temperature measurement through a selected interpolation method such as cubic spline interpolation or Akima spline interpolation. , obtain the temperature smooth curve in four directions, and perform numerical conversion calculation on the curve according to the corresponding relationship between the temperature size and the infrared pixel value size in the infrared image array.
  • the selected interpolation method is not limited to the interpolation method exemplified above.
  • a deep learning neural network model for identifying the temperature distribution of the furnace top material surface is established.
  • the transformed infrared array and cross temperature measurement array obtained in step (2) are used as deep learning neural network models.
  • the neural network model can be used to output the identification results of the furnace top temperature distribution.
  • the characteristic of this network is that it does not require real result annotation (Ground Truth) for model training.
  • the model function can be realized through a defined loss function and collected sample training, which is an unsupervised learning network.
  • the structure of the established deep learning neural network model is described below.
  • the network model structure established in the embodiment of the present invention mainly includes a generator G and a discriminator DV.
  • the infrared array V and the cross temperature measurement array I are used as inputs to the generator, and the generator outputs the recognition result, and
  • the infrared array V and the recognition result F are input to the discriminator DV for adversarial training.
  • the generator G is mainly used to fuse two input data, and its network structure includes an encoder and a decoder.
  • the input infrared array V and the cross temperature measurement array I are stacked in the same dimension, and then the stacked arrays are input to the encoder.
  • the encoder will output a fusion feature map, and then the feature map is input to the decoder.
  • the recognition result will be obtained after calculation.
  • the specific working process of the neural network is shown in Figure 6.
  • the encoder is a feature (Short Connection) structure in the densely connected neural network DenseNet, which is composed of 5 convolutional layers.
  • the input of each layer is composed of a stack of channels output by all previous layers; the decoder is composed of Composed of 4 convolutional layers.
  • All convolutional layers in the neural network are 3*3 convolution kernels, the convolution step size is 1, the activation function uses the linear rectification unit ReLU, and batch normalization BN is performed before the linear rectification unit; among them, batch normalization BN
  • the input is normalized to a standard normal distribution with a mean of 0 and a variance of 1.
  • Table 1 The specific channel number configuration of different convolutional layers is as follows: Table 1:
  • the discriminator D V is mainly used to distinguish the array generated by the generator and the input infrared array. Its structure is composed of 3 layers of convolutional layers and 1 layer of fully connected layers, as shown in Table 2 below:
  • the loss function used by the deep learning neural network in this embodiment of the present invention is explained below.
  • the loss function of generator G is:
  • E[] represents the mathematical expectation
  • log is the logarithmic function
  • DV() represents the output result of the discriminator
  • G(V,I) is the output result of the generator
  • V is the input infrared array
  • I is the input cross test Warm array.
  • L cont is calculated using the following formula:
  • m, n is the size of the input array
  • e is the natural index
  • G(V,I) j represents the jth point in G(V,I)
  • I k represents the kth point in the cross temperature measurement array I
  • d j,k represents the Euclidean distance between G(V,I) j and I k
  • eta is the weight coefficient
  • TV represents the TV-norm, which is the total variation norm.
  • the former term of this formula reflects the deviation between the generated result G(V,I) and the cross temperature measurement result near the temperature measurement point, and the latter term reflects the difference in temperature distribution between the generated result G(V,I) and infrared imaging. Similarity of texture details.
  • the loss function of the discriminator DV is:
  • the discriminator is used to distinguish the source input data and the fused data.
  • the loss of the discriminator can be used to calculate the difference between different distributions. Therefore, it can be used to judge the authenticity of the distribution of pixel intensity and texture details and make the distribution of the fused data closer to the real temperature. distributed.
  • the cross temperature measurement array and infrared array are obtained at the selected moment or time period, and the obtained arrays are input into the trained model.
  • the generator can identify the temperature distribution of the blast furnace top surface at a selected moment or a selected time period. This process does not require a discriminator.
  • the discriminator is mainly used to train the neural network model.
  • one way to use it is to follow steps (1) and (2) above to obtain the latest infrared array and cross temperature measurement array, use them as input, and feed them into the trained generator G without the need for the discriminator D V , after generator G completes the calculation, a two-dimensional array is obtained as the data fusion result.
  • This two-dimensional array represents the furnace top temperature distribution information integrating infrared images and cross temperature measurement results. This completes the calculation of the furnace top material surface temperature. Distribution identification.
  • the present invention also provides a device for identifying the temperature distribution of the blast furnace top surface.
  • the device includes a processor 701, a memory 702, a bus 703, and a device stored in the memory 702 and capable of running on the processor 701.
  • a computer program includes one or more processing cores.
  • the memory 702 is connected to the processor 701 through a bus 703.
  • the memory 702 is used to store program instructions. When the processor executes the computer program, the above method of Embodiment 1 of the present invention is implemented. The steps in the example.
  • a device for identifying the temperature distribution of the blast furnace top surface can be a computer unit, and the computer unit can be a computing device such as a desktop computer, a notebook, a handheld computer, and a cloud server.
  • the computer unit may include, but is not limited to, a processor and a memory.
  • Those skilled in the art can understand that the above-mentioned composition structure of the computer unit is only an example of the computer unit and does not constitute a limitation on the computer unit. It may include more or less components than the above, or some components may be combined, or different components may be used. part.
  • the computer unit may also include input and output devices, network access devices, buses, etc., which are not limited in this embodiment of the present invention.
  • the so-called processor can be a central processing unit (Central Processing Unit, CPU), or other general-purpose processor, digital signal processor (Digital Signal Processor, DSP), application-specific integrated circuit ( Application Specific Integrated Circuit (ASIC), off-the-shelf programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • the general processor can be a microprocessor or the processor can be any conventional processor, etc.
  • the processor is the control center of the computer unit and uses various interfaces and lines to connect various parts of the entire computer unit.
  • the memory can be used to store computer programs and/or modules, and the processor implements various functions of the computer unit by running or executing the computer programs and/or modules stored in the memory, and calling data stored in the memory.
  • the memory may mainly include a stored program area and a stored data area, wherein the stored program area may store the operating system and at least one application required for a function; the stored data area may store data created based on the use of the mobile phone, etc.
  • the memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart memory card (Smart Media Card, SMC), secure digital (Secure Digital, SD) card , Flash Card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.
  • non-volatile memory such as hard disk, memory, plug-in hard disk, smart memory card (Smart Media Card, SMC), secure digital (Secure Digital, SD) card , Flash Card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.
  • Embodiment 4 is a diagrammatic representation of Embodiment 4:
  • the present invention also provides a computer-readable storage medium.
  • the computer-readable storage medium stores a computer program.
  • the computer program is executed by a processor, the steps of the above method in the embodiment of the present invention are implemented.
  • the modules/units integrated with the computer unit are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.
  • the present invention can implement all or part of the processes in the methods of the above embodiments, and can also be completed by instructing relevant hardware through a computer program.
  • the computer program can be stored in a computer-readable storage medium, and the computer program can be stored in a computer-readable storage medium.
  • the steps of each of the above method embodiments can be implemented.
  • the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc.
  • Computer-readable media can include: any entity or device that can carry computer program code, recording media, USB flash drives, mobile hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM, Read-Only Memory), random access Memory (RAM, Random Access Memory) and software distribution media, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Computation (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Quality & Reliability (AREA)
  • Radiation Pyrometers (AREA)

Abstract

一种高炉炉顶料面温度分布的识别方法、装置及存储介质，该方法同时利用炉喉十字测温装置和炉顶红外摄像仪的测温数据，包括：S1，建立用于识别炉顶料面温度分布的、基于生成对抗网络的深度学习神经网络模型，并利用炉喉十字测温装置和炉顶红外摄像仪的测温数据进行训练；S2，获取选定时刻或时间段的高炉十字测温数据和红外检测数据，并将所述选定时刻或时间段的十字测温数据和红外检测数据输入训练好的所述深度学习神经网络模型中的生成器，所述生成器输出所述选定时刻或时间段的炉顶料面温度分布的识别结果。利用该方法，可以准确全面地识别高炉炉顶料面温度分布信息。

Description

高炉炉顶料面温度分布的识别方法、装置及存储介质 技术领域
本发明涉及高炉炼铁检测技术领域，尤其涉及一种高炉炉顶料面温度分布的识别方法、装置及存储介质。
背景技术
在高炉生产中，高炉煤气流在炉内的分布状况直接影响了炉内温度分布、软熔带位置形状、煤气利用率以及炉况稳顺等，最终影响到高炉的经济指标，获得理想的煤气流分布是高炉操作中的重要目标。因此，及时有效地了解炉内煤气流的分布状况对高炉操作有着重要意义。在高炉炉内检测信息中，高炉炉顶料面温度分布最直接地反映了煤气流的分布情况，一般高炉生产工艺分析认为，煤气流发展较强的地方，煤气温度较高；反之，则煤气温度较低。
由于炉顶料面温度分布无法直接检测，目前对于炉顶温度分布的间接检测手段主要有炉喉十字测温装置和炉顶红外摄像仪。前者一般在炉喉圆周面上的四个方向安装四个测温臂，在测温臂上分布不等如5～7个的温度传感器。这些测温点能够提供实时准确的温度数据，可以准确地反映煤气流在炉喉4个方向上的分布，但其测温点数量有限，无法全面地掌握整个料面的温度分布；而后者是非接触式红外测温装置，它可以实时以红外图像的形式检测整个料面的温度分布变化，红外成像的主要缺点在于测量值的准确性难以保证。
发明内容
本发明的实施例提供了一种高炉炉顶料面温度分布的识别方法，结合了十字测温和红外摄像仪两种测量手段，通过利用十字测温和红外摄像仪这两种测量手段的优点以得到准确且全面的炉顶料面温度分布信息。
为了实现上述目的，一方面，提供了一种高炉炉顶料面温度分布的识别方法，在高炉的炉顶安装有炉喉十字测温装置和炉顶红外摄像仪，通过所述十字 测温装置采集十字测温数据，通过所述红外摄像仪采集红外检测数据，所述识别方法包括：
S1，建立并训练用于识别炉顶料面温度分布的、基于生成对抗网络的深度学习神经网络模型；
其中，所建立的所述深度学习神经网络模型包括：
生成器，包括编码器和解码器，所述编码器用于融合输入的十字测温数据和红外检测数据，并将融合结果输入所述解码器，所述解码器根据所述融合结果进行计算并输出炉顶料面温度分布的识别结果；
判别器，用于判别所述生成器输出的识别结果和所述红外检测数据并对所述生成器输出的识别结果和所述红外检测数据进行对抗训练；
其中，训练所述深度学习神经网络模型包括：
获取所述十字测温数据和所述红外检测数据；
将所获取的所述十字测温数据和所述红外检测数据分成训练集、验证集和测试集来训练和验证所述深度学习神经网络模型；
当所述深度学习神经网络模型的损失函数满足预设条件时，确定所述深度学习神经网络模型训练好了，停止训练并保存此时所述生成器的超参数；
S2,获取选定时刻或时间段的高炉十字测温数据和红外检测数据，并将所述选定时刻或时间段的十字测温数据和红外检测数据输入训练好的所述深度学习神经网络模型中的生成器，所述生成器输出所述选定时刻或时间段的炉顶料面温度分布的识别结果。
优选地，所述的识别方法，其中通过所述十字测温装置采集十字测温数据，通过所述红外摄像仪采集红外检测数据的步骤包括：
在高炉未进行布料动作的期间采集红外摄像机所拍摄的无溜槽遮挡、无明显干扰的红外图像；
当获取到最新的红外图像时，采集同时刻的十字测温装置的十字测温数据。
优选地，所述的识别方法，其中通过所述十字测温装置采集十字测温数据，通过所述红外摄像仪采集红外检测数据的步骤还包括：
使用均值滤波对所述红外图像进行滤波，并将滤波后的数据进行仿射变 换，获得水平面上的红外图像的二维红外数组；
对所述水平面上的红外图像进行裁切，使炉心在画面中心，且炉喉轮廓贴近图像边缘，得到m*n大小的二维数组，其中m，n为自然数，m，n的大小由红外摄像仪的分辨率确定；
生成一个m*n大小且填充0值的二维数组，以仿射变换后的红外图像坐标系为基准，将采集的十字测温数据进行插值处理后得到的平滑曲线按照其所在空间位置填入至所述填充0值的二维数组的相应像素点中，得到二维十字测温数组；
其中，输入所述深度学习神经网络模型的十字测温数据和所述红外检测数据为所述十字测温数组和所述红外数组。
优选地，所述的识别方法，其中，所述十字测温装置包括四个方向上设置的四个测温臂，每个测温臂上设置有预定数目的测温传感器，其中将采集的十字测温数据进行插值处理包括：
将通过十字测温装置得到的所述四个方向上的测温值，通过预先选定的插值方法对进行插值，得到所述四个方向上测温值的温度平滑曲线。
优选地，所述的识别方法，其中，在输入的十字测温数据和红外检测数据进行同维堆叠后，所述编码器对经过同维堆叠后的十字测温数据和红外检测数据进行融合，并输出融合特征图；其中，所述编码器为密集连接的DenseNet中的特征结构，包括5层卷积层，每层卷积层的输入为该卷积层之前所有层输出的通道级联组成；所述解码器包括4层卷积层；
其中，所述所有卷积层中的每一个都为3*3卷积核，卷积步长为1，激活函数采用线性整流单元ReLU，在线性整流单元之前执行批归一化BN，以将线性整流单元的输入规范化到均值为0、方差为1的标准正态分布。
优选地，所述的识别方法，其中，所述生成器的损失函数为：
Figure PCTCN2022122610-appb-000001
其中，G表示所述生成器，
Figure PCTCN2022122610-appb-000002
表示对抗损失，L cont为内容损失，λ为权重系数；其中，
Figure PCTCN2022122610-appb-000003
用下式计算：
Figure PCTCN2022122610-appb-000004
式中，E[]表示数学期望，log为对数函数，D V()表示判别器输出结果，G(V,I) 为生成器的输出结果，V为输入的红外数组，I为输入的十字测温数组；
L cont用下式计算：
Figure PCTCN2022122610-appb-000005
式中，m,n为输入数组的大小，e为自然指数，G(V,I) j表示G(V,I)中第j点，I k表示十字测温数组I中第k点，d j,k表示G(V,I) j和I k的欧式距离，η为权重系数，|| || TV表示TV-范数。
优选地，所述的识别方法，其中，所述判别器D V包括3层卷积层和1层全连接层，所述判别器D V的损失函数为：
Figure PCTCN2022122610-appb-000006
优选地，所述的识别方法，其中，当所述生成器和所述判别器D V的损失函数在验证集上的变化下降至平稳后，确定所述深度学习神经网络模型训练好了。
另一方面，提供了一种高炉炉顶料面温度分布的识别装置，包括存储器和处理器，所述存储器存储有至少一段程序，所述至少一段程序由处理器执行以实现如上文任一所述的方法。
又一方面，提供了一种计算机可读存储介质，所述存储介质中存储有至少一段程序，所述至少一段程序由处理器执行以实现如上文任一所述的识别方法。
上述技术方案具有如下技术效果：
通过获取炉喉十字测温装置和炉顶红外摄像仪的检测数据，利用检测数据对基于生成对抗网络的深度学习神经网络进行训练，得到用于识别炉顶料面温度分布的神经网络模型；在采集最新的炉喉十字测温装置和炉顶红外摄像仪的检测数据后，经过处理得到二维数组数据输入至训练好的神经网络模型，模型完成数据融合，即可得到炉顶料面温度分布信息，如此完成对炉顶料面温度分布的识别。本发明实施例提供的识别方法可以给出炉顶料面温度分布情况，具有准确度高、实时性强、可视化效果好的优点，能有效应用于高炉炉顶料面温度检测中，并为高炉操作和炉况诊断等提供了准确的判断依据，具有直接的指 导意义，有助于高炉长期安全平稳的运行。
附图说明
图1为本发明一实施例的高炉炉顶料面温度分布的识别方法的流程示意图；
图2为本发明一实施例的识别方法中，炉顶红外图像的一个示例；
图3为本发明一实施例的识别方法中，炉喉十字测温装置的示意图；
图4为本发明一实施例的识别方法中采用的仿射变换的示意图；
图5为本发明实施例的识别方法中建立的深度学习神经网络模型结构示意图；
图6为本发明实施例的识别方法中，深度学习神经网络模型的工作过程示意图；
图7为本发明实施例的高炉炉顶料面温度分布的识别装置的结构示意图。
具体实施方式
为进一步说明各实施例，本发明提供有附图。这些附图为本发明揭露内容的一部分，其主要用以说明实施例，并可配合说明书的相关描述来解释实施例的运作原理。配合参考这些内容，本领域普通技术人员应能理解其他可能的实施方式以及本发明的优点。图中的组件并未按比例绘制，而类似的组件符号通常用来表示类似的组件。
现结合附图和具体实施方式对本发明进一步说明。
实施例一：
图1为本发明一实施例的高炉炉顶料面温度分布的识别方法的流程示意图。该实施例的识别方法中，在高炉的炉顶安装有炉喉十字测温装置和炉顶红外摄像仪，通过十字测温装置采集十字测温数据，通过红外摄像仪采集红外检测数据，如图1，该实施例的识别方法包括：
S1，建立并训练用于识别炉顶料面温度分布的、基于生成对抗网络的深度学习神经网络模型；
其中，所建立的深度学习神经网络模型包括：
生成器，包括编码器和解码器，编码器用于融合输入的十字测温数据和红外检测数据，并将融合结果输入解码器，解码器根据融合结果进行计算并输出炉顶料面温度分布的识别结果；
判别器，用于判别生成器输出的识别结果和红外检测数据并对生成器输出的识别结果和红外检测数据进行对抗训练；
其中，训练深度学习神经网络模型包括：
获取十字测温数据和红外检测数据；
将所获取的十字测温数据和红外检测数据分成训练集、验证集和测试集来训练和验证深度学习神经网络模型；
当深度学习神经网络模型的损失函数满足预设条件时，确定深度学习神经网络模型训练好了，停止训练并保存此时生成器的超参数；
S2,获取选定时刻或时间段的高炉十字测温数据和红外检测数据，并将选定时刻或时间段的十字测温数据和红外检测数据输入训练好的深度学习神经网络模型中的生成器，该生成器输出选定时刻或时间段的炉顶料面温度分布的识别结果。
本发明实施例的实施要求高炉炉顶装有炉喉十字测温装置和炉顶红外摄像仪，通过分别采集炉喉十字测温装置和炉顶红外摄像仪测量的数据来训练深度学习神经网络模型和使用训练好的神经网络模型。所使用的炉喉十字测温装置和炉顶红外摄像仪为现有技术的炉喉十字测温装置和炉顶红外摄像仪，其安装方式也可采用现有技术的安装方式。
为保证炉顶红外摄像仪测量数据的准确，需要在高炉未进行布料动作期间，采集红外摄像机所拍摄的无溜槽遮挡、无明显干扰的红外图像，此场景下的图像可称为正常的红外图像。具体实现中，不断地执行上述采集步骤。高炉工作中，其布料是周期性的。采集红外图像的指令可以是通过上级控制系统如高炉一级系统中的PLC控制器在布料开始前、布料间歇或布料完成后发出，以采集上述正常的红外图像。
本发明实施例的高炉炉顶料面温度分布的识别方法主要通过结合十字测温装置和炉顶红外摄像仪这两个手段来实现，利用深度学习神经网络、通过融合并处理通过两个手段获取的测温数据来获得较准确的识别结果。
具体实现中，每当获取到最新的正常红外图像时，采集同时刻的十字测温数据。
实施例二：
下面对本发明另一实施例的高炉炉顶料面温度分布的识别方法的具体步骤进行描述。
(1)采集炉喉十字测温装置和炉顶红外摄像仪的检测数据
其中，要求高炉炉顶装有炉喉十字测温装置和炉顶红外摄像仪。不断地执行下述步骤：在高炉未进行布料动作期间，采集红外摄像机所拍摄的无溜槽遮挡、无明显干扰的正常图像。每当获取到最新的正常红外图像时，采集同时刻的十字测温数据。
图2给出了炉顶红外图像的一个示例。图像数据一般表示为二维矩阵数组，其中每个像素点的数值大小表示了该像素点的温度大小。
图3为一示例性的炉喉十字测温装置的示意图。如图3，该炉喉十字测温装置包括四个方向上的测温臂，其中3个测温臂上有5个测温点，一个测温臂上有6个测温点，一共21个测温点。该炉喉十字测温装置仅为示例性地，用来说明本发明实施例的实现，本发明实施例的实现并不仅限于该具体结构，其也可以具有不同的测温点数目、测温点位置和/或不同的方向。如可以有17个测温点。如图3所示，每个测温点的测温值可以准确体现该位置气流温度的大小。
(2)数据处理
得到红外图像数据后，使用均值滤波对红外图像进行预处理，并对该数组经过一定仿射变换得到水平面的红外图像二维数组；对图像进行裁切，使炉心在画面中心，且炉喉轮廓贴近图像边缘，得到m*n大小的二维红外数组，简称红外数组。其中m，n为自然数，m，n的大小由红外摄像仪的分辨率确定。示例性地，可以通过炉喉轮廓与图像边缘直接的距离是否小于预定的距离阈值来判断炉喉轮廓是否贴近图像边缘。
生成一个与红外图像同样大小m*n且填充0值的二维数组，以仿射变换后的红外图像坐标系为基准，将十字测温在四个方向上获得的测温值进行插值处理后得到的平滑曲线按照其即平滑曲线所在空间位置填入至数组中的相应像 素点中，如此得到一个体现十字测温结果的二维十字测温数组，简称十字测温数组。
其中，示例性地，仿射变换步骤如下：选取炉喉内部同一高度平面的边缘4点，找到其在红外图像中的对应坐标，根据透视变换原理，结合这四个点在高炉内的实际位置坐标，和它们在红外图像中的坐标，可以求解透视变换矩阵T，然后使用变换矩阵T对红外图像进行仿射变换得到从上向下正视的红外图像，这一变换过程如图4所示。
其中，示例性地插值处理过程步骤如下：将十字测温得到的四个方向的测温值，通过选定的插值方法如三次样条插值或Akima样条插值对四个方向的温度值进行插值，得到四个方向的温度平滑曲线，并按照红外图像数组中温度大小与红外像素点值大小的对应关系对该曲线进行数值转换计算。选定的插值方法不限于上述举例的插值方法。
(3)建立模型
基于生成对抗网络(Generative Adversarial Networks，GAN)建立用于识别炉顶料面温度分布的深度学习神经网络模型，将第(2)步得到的变换后的红外数组和十字测温数组作为深度学习神经网络模型的输入。神经网络模型可用于输出炉顶温度分布的识别结果。该网络的特点在于无需真实结果标注(Ground Truth)用于模型训练，通过定义好的损失函数和采集样本训练即可实现模型功能，即为无监督学习网络。
下面对建立的深度学习神经网络模型结构进行描述。
如图5，本发明实施例中建立的网络模型结构主要包括生成器G和判别器DV，将红外数组V和十字测温数组I作为输入，给入生成器，生成器输出识别结果，并将红外数组V和识别结果F输入至判别器DV中进行对抗训练。
其中，生成器G主要用于融合两种输入数据，其网络结构包括一个编码器和一个解码器。首先将输入的红外数组V和十字测温数组I进行同维堆叠，然后将堆叠后的数组输入至编码器中，编码器经计算后将输出一个融合特征图，再将该特征图输入至解码器中，解码器经计算后将得到识别结果。神经网络的具体工作过程图6所示。
其中，编码器为密集连接的神经网络DenseNet中的特征(Short  Connection)结构，采用了5层卷积层组成，每层的输入为其之前所有层输出的通道级联如堆叠组成；解码器由4层卷积层组成。神经网络中所有卷积层都为3*3卷积核，卷积步长为1，激活函数采用线性整流单元ReLU，在线性整流单元之前执行批归一化BN；其中，批归一化BN将其输入规范化到均值为0、方差为1的标准正态分布，具体不同卷积层的通道数配置如下表1：
卷积层 通道数(特征图)
H 48
L1 128
L2 64
L3 16
L4 1
表1
判别器D V主要用于判别生成器生成数组和输入的红外数组，其结构为3层卷积层和1层全连接层组成，如下表2：
Figure PCTCN2022122610-appb-000007
表2
下面对本发明该实施例中深度学习神经网络采用的损失函数进行说明。
生成器G的损失函数为：
Figure PCTCN2022122610-appb-000008
其中，
Figure PCTCN2022122610-appb-000009
表示对抗损失，L cont为内容损失，λ为权重系数。
Figure PCTCN2022122610-appb-000010
用下式计算：
Figure PCTCN2022122610-appb-000011
式中，E[]表示数学期望，log为对数函数，DV()表示判别器输出结果，G(V,I)为生成器输出结果，V为输入的红外数组，I为输入的十字测温数组。
L cont用下式计算：
Figure PCTCN2022122610-appb-000012
式中，m,n为输入的数组大小即尺寸，e为自然指数，G(V,I) j表示G(V,I)中第j点，I k表示十字测温数组I中第k点，d j,k表示G(V,I) j和I k的欧式距离，η为权重系数，|| || TV表示TV-范数即全变分范数。此式前一项反映了生成结果G(V,I)与十字测温结果在测温点附近的偏差，后一项反映了生成结果G(V,I)与红外成像在温度分布变化上如纹理细节的相似程度。
判别器DV的损失函数为：
Figure PCTCN2022122610-appb-000013
判别器用于区分源输入数据和融合数据，判别器的损失可以用来计算不同分布之间的差异，因此可以用来判断像素强度和纹理细节分布的真实性并促使融合数据的分布更贴近真实温度分布。
(4)数据融合
按照第(1)、(2)步采样大量红外数组和十字测温数组后，对(3)中提到的神经网络模型进行训练和交叉验证。示例性地，随机将数据集的80％作为训练集用来训练神经网络模型，10％的数据作为验证集来挑选超参数，10％的数据作为测试集来测试模型的泛化能力。示例性地，训练中，设定权重系数λ＝0.5，η＝1.2。在神经网络模型的损失函数在验证集上的变化下降至平稳后，停止训练，保存生成器G的超参数，此模型即可投入使用。上述训练集、测试集和验证集的比例可以根据需要调整；权重系数的选择也可以根据需要进行调整。
训练好用来识别高炉炉顶料面温度分布的深度学习神经网络模型后，通过获取选定时刻或选定时间段的十字测温数组和红外数组，将获取的数组输入该训练好的模型中的生成器，可以识别出在选定时刻或选定时间段的高炉炉顶料面温度分布。该过程不需要判别器。本发明实施例中，判别器主要是用来训练神经网络模型的。
例如，一种使用方式为，按照上文的(1)、(2)步获取当前最新的红外数组 和十字测温数组，将其作为输入，给入训练好的生成器G，无需判别器D V，生成器G完成计算后，得到作为数据融合结果的二维数组，此二维数组表示综合红外图像和十字测温结果的炉顶温度分布信息，由此即完成了对炉顶料面温度分布的识别。
实施例三：
本发明还提供一种高炉炉顶料面温度分布的识别装置，如图7所示，该装置包括处理器701、存储器702、总线703、以及存储在存储器702中并可在处理器701上运行的计算机程序，处理器701包括一个或一个以上处理核心，存储器702通过总线703与处理器701相连，存储器702用于存储程序指令，处理器执行计算机程序时实现本发明实施例一的上述方法实施例中的步骤。
进一步地，作为一个可执行方案，一种高炉炉顶料面温度分布的识别装置可以是计算机单元，该计算机单元可以是桌上型计算机、笔记本、掌上电脑及云端服务器等计算设备。计算机单元可包括，但不仅限于，处理器、存储器。本领域技术人员可以理解，上述计算机单元的组成结构仅仅是计算机单元的示例，并不构成对计算机单元的限定，可以包括比上述更多或更少的部件，或者组合某些部件，或者不同的部件。例如计算机单元还可以包括输入输出设备、网络接入设备、总线等，本发明实施例对此不做限定。
进一步地，作为一个可执行方案，所称处理器可以是中央处理单元(Central Processing Unit，CPU)，还可以是其他通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现成可编程门阵列(Field-Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等，处理器是计算机单元的控制中心，利用各种接口和线路连接整个计算机单元的各个部分。
存储器可用于存储计算机程序和/或模块，处理器通过运行或执行存储在存储器内的计算机程序和/或模块，以及调用存储在存储器内的数据，实现计算机单元的各种功能。存储器可主要包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需的应用程序；存储数据区可存储 根据手机的使用所创建的数据等。此外，存储器可以包括高速随机存取存储器，还可以包括非易失性存储器，例如硬盘、内存、插接式硬盘，智能存储卡(Smart Media Card,SMC)，安全数字(Secure Digital,SD)卡，闪存卡(Flash Card)、至少一个磁盘存储器件、闪存器件、或其他易失性固态存储器件。
实施例四：
本发明还提供一种计算机可读存储介质，计算机可读存储介质存储有计算机程序，计算机程序被处理器执行时实现本发明实施例上述方法的步骤。
计算机单元集成的模块/单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本发明实现上述实施例方法中的全部或部分流程，也可以通过计算机程序来指令相关的硬件来完成，的计算机程序可存储于一计算机可读存储介质中，该计算机程序在被处理器执行时，可实现上述各个方法实施例的步骤。其中，计算机程序包括计算机程序代码，计算机程序代码可以为源代码形式、对象代码形式、可执行文件或某些中间形式等。计算机可读介质可以包括：能够携带计算机程序代码的任何实体或装置、记录介质、U盘、移动硬盘、磁碟、光盘、计算机存储器、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)以及软件分发介质等。需要说明的是，计算机可读介质包含的内容可以根据司法管辖区内立法和专利实践的要求进行适当的增减。
尽管结合优选实施方案具体展示和介绍了本发明，但所属领域的技术人员应该明白，在不脱离所附权利要求书所限定的本发明的精神和范围内，在形式上和细节上可以对本发明做出各种变化，均为本发明的保护范围。

Claims (10)

  1. 一种高炉炉顶料面温度分布的识别方法，其特征在于，在高炉的炉顶安装有炉喉十字测温装置和炉顶红外摄像仪，通过所述十字测温装置采集十字测温数据，通过所述红外摄像仪采集红外检测数据，所述识别方法包括：
    S1，建立并训练用于识别炉顶料面温度分布的、基于生成对抗网络的深度学习神经网络模型；
    其中，所建立的所述深度学习神经网络模型包括：
    生成器，包括编码器和解码器，所述编码器用于融合输入的十字测温数据和红外检测数据，并将融合结果输入所述解码器，所述解码器根据所述融合结果进行计算并输出炉顶料面温度分布的识别结果；
    判别器，用于判别所述生成器输出的识别结果和所述红外检测数据并对所述生成器输出的识别结果和所述红外检测数据进行对抗训练；
    其中，训练所述深度学习神经网络模型包括：
    获取所述十字测温数据和所述红外检测数据；
    将所获取的所述十字测温数据和所述红外检测数据分成训练集、验证集和测试集来训练和验证所述深度学习神经网络模型；
    当所述深度学习神经网络模型的损失函数满足预设条件时，确定所述深度学习神经网络模型训练好了，停止训练并保存此时所述生成器的超参数；
    S2,获取选定时刻或时间段的高炉十字测温数据和红外检测数据，并将所述选定时刻或时间段的十字测温数据和红外检测数据输入训练好的所述深度学习神经网络模型中的生成器，所述生成器输出所述选定时刻或时间段的炉顶料面温度分布的识别结果。
  2. 根据权利要求1所述的识别方法，其特征在于，所述通过所述十字测温装置采集十字测温数据，通过所述红外摄像仪采集红外检测数据的步骤包括：
    在高炉未进行布料动作的期间采集红外摄像机所拍摄的无溜槽遮挡、无明显干扰的红外图像；
    当获取到最新的红外图像时，采集同时刻的十字测温装置的十字测温数据。
  3. 根据权利要求2所述的识别方法，其特征在于，
    所述通过所述十字测温装置采集十字测温数据，通过所述红外摄像仪采集红外检测数据的步骤还包括：
    使用均值滤波对所述红外图像进行滤波，并将滤波后的数据进行仿射变换，获得水平面上的红外图像的二维红外数组；
    对所述水平面上的红外图像进行裁切，使炉心在画面中心，且炉喉轮廓贴近图像边缘，得到m*n大小的二维数组，其中m，n为自然数，m，n的大小由红外摄像仪的分辨率确定；
    生成一个m*n大小且填充0值的二维数组，以仿射变换后的红外图像坐标系为基准，将采集的十字测温数据进行插值处理后得到的平滑曲线按照其所在空间位置填入至所述填充0值的二维数组的相应像素点中，得到二维十字测温数组；
    其中，输入所述深度学习神经网络模型的十字测温数据和所述红外检测数据为所述十字测温数组和所述红外数组。
  4. 根据权利要求3所述的识别方法，其特征在于，所述十字测温装置包括四个方向上设置的四个测温臂，每个测温臂上设置有预定数目的测温传感器，其中将采集的十字测温数据进行插值处理包括：
    将通过十字测温装置得到的所述四个方向上的测温值，通过预先选定的插值方法对进行插值，得到所述四个方向上测温值的温度平滑曲线。
  5. 根据权利要求1所述的识别方法，其特征在于，在输入的十字测温数据和红外检测数据进行同维堆叠后，所述编码器对经过同维堆叠后的十字测温数据和红外检测数据进行融合，并输出融合特征图；其中，所述编码器为密集连接的DenseNet中的特征结构，包括5层卷积层，每层卷积层的输入为该卷积层之前所有层输出的通道级联组成；所述解码器包括4层卷积层；
    其中，所述所有卷积层中的每一个都为3*3卷积核，卷积步长为1，激活函数采用线性整流单元ReLU，在线性整流单元之前执行批归一化BN，以将线性整流单元的输入规范化到均值为0、方差为1的标准正态分布。
  6. 根据权利要求3或5所述的识别方法，其特征在于，所述生成器的损失函数为：
    Figure PCTCN2022122610-appb-100001
    其中，G表示所述生成器，
    Figure PCTCN2022122610-appb-100002
    表示对抗损失，L cont为内容损失，λ为权重系数；其中，
    Figure PCTCN2022122610-appb-100003
    用下式计算：
    Figure PCTCN2022122610-appb-100004
    式中，E[]表示数学期望，log为对数函数，D V()表示判别器输出结果，G(V,I)为生成器的输出结果，V为输入的红外数组，I为输入的十字测温数组；
    L cont用下式计算：
    Figure PCTCN2022122610-appb-100005
    式中，m,n为输入数组的大小，e为自然指数，G(V,I) j表示G(V,I)中第j点，I k表示十字测温数组I中第k点，d j,k表示G(V,I) j和I k的欧式距离，η为权重系数，|||| TV表示TV-范数。
  7. 根据权利要求6所述的识别方法，其特征在于，所述判别器D V包括3层卷积层和1层全连接层，所述判别器D V的损失函数为：
    Figure PCTCN2022122610-appb-100006
  8. 根据权利要求7所述的识别方法，其特征在于，当所述生成器和所述判别器D V的损失函数在验证集上的变化下降至平稳后，确定所述深度学习神经网络模型训练好了。
  9. 一种高炉炉顶料面温度分布的识别装置，其特征在于，包括存储器和处理器，所述存储器存储有至少一段程序，所述至少一段程序由处理器执行以实现如权利要求1至8任一所述的方法。
  10. 一种计算机可读存储介质，其特征在于，所述存储介质中存储有至少一段程序，所述至少一段程序由处理器执行以实现如权利要求1至8任一所述的识别方法。
PCT/CN2022/122610 2022-09-21 2022-09-29 高炉炉顶料面温度分布的识别方法、装置及存储介质 WO2024060284A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211148946.3A CN115541030B (zh) 2022-09-21 2022-09-21 高炉炉顶料面温度分布的识别方法、装置及存储介质
CN202211148946.3 2022-09-21

Publications (1)

Publication Number Publication Date
WO2024060284A1 true WO2024060284A1 (zh) 2024-03-28

Family

ID=84727362

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/122610 WO2024060284A1 (zh) 2022-09-21 2022-09-29 高炉炉顶料面温度分布的识别方法、装置及存储介质

Country Status (2)

Country Link
CN (1) CN115541030B (zh)
WO (1) WO2024060284A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115876334B (zh) * 2023-02-22 2023-05-30 瑞纳智能设备股份有限公司 红外热像室内测温方法、系统和计算机可读存储介质
CN117309195B (zh) * 2023-11-30 2024-04-12 中控技术股份有限公司 测温仪表的检测方法、装置及电子设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102732660A (zh) * 2012-06-27 2012-10-17 浙江大学 一种基于多源信息融合的料面温度场检测方法
CN109852748A (zh) * 2019-02-27 2019-06-07 内蒙古科技大学 监测高炉布料周期煤气流发展过程及预测煤气利用率方法
CN112700402A (zh) * 2020-12-09 2021-04-23 浙江工业大学 基于图像和自编码生成网络的锅炉高温氧浓度预测方法
WO2022083155A1 (zh) * 2020-10-22 2022-04-28 中冶南方工程技术有限公司 一种高炉铁口工作状态的自动检测方法、系统及存储介质
CN114581785A (zh) * 2022-05-09 2022-06-03 中国科学院自动化研究所 多源异构数据融合的高炉煤气流分布状态识别方法及系统

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109410252A (zh) * 2018-12-20 2019-03-01 合肥英睿系统技术有限公司 一种热像设备
CN113902947B (zh) * 2021-10-09 2023-08-25 南京航空航天大学 一种自然图像的对空目标红外图像生成式对抗网络构建方法
CN114758783A (zh) * 2022-03-31 2022-07-15 大连理工大学 深度学习的弹性成像方法、装置、计算机设备和存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102732660A (zh) * 2012-06-27 2012-10-17 浙江大学 一种基于多源信息融合的料面温度场检测方法
CN109852748A (zh) * 2019-02-27 2019-06-07 内蒙古科技大学 监测高炉布料周期煤气流发展过程及预测煤气利用率方法
WO2022083155A1 (zh) * 2020-10-22 2022-04-28 中冶南方工程技术有限公司 一种高炉铁口工作状态的自动检测方法、系统及存储介质
CN112700402A (zh) * 2020-12-09 2021-04-23 浙江工业大学 基于图像和自编码生成网络的锅炉高温氧浓度预测方法
CN114581785A (zh) * 2022-05-09 2022-06-03 中国科学院自动化研究所 多源异构数据融合的高炉煤气流分布状态识别方法及系统

Also Published As

Publication number Publication date
CN115541030B (zh) 2024-02-27
CN115541030A (zh) 2022-12-30

Similar Documents

Publication Publication Date Title
WO2024060284A1 (zh) 高炉炉顶料面温度分布的识别方法、装置及存储介质
CN109902677B (zh) 一种基于深度学习的车辆检测方法
CN111080693A (zh) 一种基于YOLOv3的机器人自主分类抓取方法
CN111160269A (zh) 一种人脸关键点检测方法及装置
US11189022B2 (en) Automatic detection, counting, and measurement of logs using a handheld device
CN116664559B (zh) 基于机器视觉的内存条损伤快速检测方法
WO2021139494A1 (zh) 基于单目摄像头的动物体在线核赔方法、装置及存储介质
CN108491850B (zh) 三维牙齿网格模型的特征点自动提取方法及装置
CN113077476B (zh) 一种高度测量方法、终端设备以及计算机存储介质
CN102509304A (zh) 基于智能优化的摄像机标定方法
CN113435282B (zh) 基于深度学习的无人机影像麦穗识别方法
CN111028205B (zh) 一种基于双目测距的眼睛瞳孔定位方法及装置
CN111860587A (zh) 一种用于图片小目标的检测方法
CN110136153A (zh) 一种图像处理方法、设备及存储介质
CN106778587B (zh) 基于无人机影像的飞行状态检测方法及装置
CN103954334A (zh) 一种全自动摄像式水表检定系统及其工作方法
CN210386980U (zh) 一种基于机器视觉的冷床智能控制系统
CN116682106A (zh) 一种基于深度学习的柑橘木虱智能检测方法及装置
CN110007764B (zh) 一种手势骨架识别方法、装置、系统及存储介质
CN113116377A (zh) 超声成像导航方法、超声设备及存储介质
CN114359172A (zh) 一种库存或陈列场景下的条烟多面检测识别方法及系统
CN113008380B (zh) 一种智慧ai体温预警方法、系统及存储介质
CN111369497B (zh) 一种行走式树上果实连续计数方法及装置
CN108875825B (zh) 一种基于图像分块的粮仓害虫检测方法
CN105184736B (zh) 一种窄重叠双视场高光谱成像仪的图像配准的方法

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: 22959286

Country of ref document: EP

Kind code of ref document: A1