Risk early warning method, medium and system based on blast furnace tuyere image data processing

Abstract

The invention provides a risk early warning method based on blast-furnace tuyere image data processing, a medium and a system, and belongs to the technical field of image data processing.The risk early warning method comprises the steps that firstly, video data of a blast-furnace tuyere are collected, and an image sequence is extracted according to a preset time interval; in order to eliminate the brightness difference caused by the environment illumination change, the image sequence is subjected to brightness equalization processing. Then, the color space is converted from RGB to HSV, and a color feature matrix is extracted. And through clustering analysis, the slag adhering state and the large block falling state of the blast furnace tuyere area are recognized. In addition, a time sequence of a color feature matrix is established, a stable component and a fluctuation component are extracted by adopting a matrix decomposition method, and the fluctuation component represents a state change trend. And after dimensionality reduction of the change trend matrix, inputting the change trend matrix into a preset risk discrimination matrix equation set to obtain a slag adhering risk index and a large block falling risk index. And finally, performing threshold analysis according to the risk vector, outputting a final risk result, and realizing risk early warning of slag adhering and large block falling.

Description

Technical Field

[0001] The present invention belongs to the technical field of image data processing, and in particular, relates to a risk warning method, medium and system based on blast furnace tuyere image data processing. Background Art

[0002] The blast furnace is one of the most important smelting equipment in the steel industry, and its safe and stable operation is directly related to the production efficiency and economic benefits of steel enterprises. The blast furnace tuyere is one of the most critical components of the blast furnace. As a channel for supplying air to the furnace, it not only undertakes the important function of conveying hot air to the blast furnace, but also directly affects the smelting conditions of the blast furnace. During the operation of the blast furnace, abnormal phenomena such as slag hanging and large pieces often occur in the blast furnace. If these abnormal conditions are not discovered and handled in time, serious safety accidents and economic losses will occur. Slag hanging and large pieces falling are one of the common risks of blast furnace tuyere. At present, the monitoring of slag hanging status and large piece falling status of blast furnace tuyere mainly relies on the following methods:

[0003] Manual inspection: Operators regularly observe and record the blast furnace tuyere on site. This method has obvious limitations: first, manual observation is subjective and uncertain, and it is difficult to ensure the consistency of the judgment standard; second, due to the harsh blast furnace environment, manual observation has safety hazards; third, manual observation cannot achieve continuous monitoring and is prone to miss the occurrence of key abnormal conditions.

[0004] Traditional parameter monitoring: various sensors are installed to monitor the temperature, pressure and other parameters at the air outlet. Although this method can provide certain reference information, it has the following problems: first, a single parameter is difficult to fully reflect the actual state of the air outlet; second, the installation position of the sensor is limited, and it is difficult to directly reflect the situation inside the air outlet; third, the sensor is easily damaged in harsh environments, and the maintenance cost is high.

[0005] In summary, it is difficult to achieve safe and efficient risk warning of slag hanging state and large piece falling state at the blast furnace tuyere with the existing technology. Summary of the invention

[0006] In view of this, the present invention provides a risk warning method, medium and system based on blast furnace tuyere image data processing, which can solve the technical problem that the prior art is difficult to achieve safe and efficient risk warning of slag hanging state and large piece falling state of blast furnace tuyere.

[0007] The present invention is achieved in that:

[0008] A first aspect of the present invention provides a risk warning method based on blast furnace tuyere image data processing, comprising the following steps:

[0009] S10, collecting video data of blast furnace tuyere, and analyzing the video data according to a preset sampling time interval to obtain a blast furnace tuyere image sequence;

[0010] S20, performing brightness equalization processing on the blast furnace tuyere image sequence to eliminate the image brightness difference caused by the ambient light change, and obtain a brightness equalized image sequence;

[0011] S30, performing color space conversion on the image sequence after brightness equalization, converting the RGB color space into the HSV color space, and extracting the color feature matrix of the blast furnace tuyere area;

[0012] S40, performing cluster analysis on the color feature matrix to obtain a plurality of color cluster centers, and identifying the slag hanging state and the large-piece falling state in the tuyere area of ​​the blast furnace based on the color cluster centers;

[0013] S50, establishing a time series of the color feature matrix, and using a matrix decomposition method to decompose the color feature matrix into a stable component matrix and a fluctuating component matrix;

[0014] S60, establishing a change trend matrix of the blast furnace tuyere state based on the fluctuation component matrix, wherein the change trend matrix includes a change direction and a change amplitude of the blast furnace tuyere state;

[0015] S70, performing dimensionality reduction processing on the change trend matrix by using principal component analysis to obtain a low-dimensional change trend matrix;

[0016] S80, inputting the change trend matrix into a preset risk discrimination matrix equation group to obtain a risk vector; the risk vector includes a slag risk index and a large piece falling risk index;

[0017] S90: Perform threshold analysis according to the risk vector, obtain risk results, and output them.

[0018] On the basis of the above technical solution, the risk warning method based on blast furnace tuyere image data processing of the present invention can also be improved as follows:

[0019] The color feature matrix is ​​specifically expressed as follows:

[0020]

[0021] Where t represents the time series number; h ij represents the hue value of the pixel in the i-th row and j-th column, and its value range is [0,360]; s ij represents the saturation value of the pixel in the i-th row and j-th column, and its value range is [0,1]; v ijRepresents the brightness value of the pixel in the i-th row and j-th column, and its value range is [0,1]. m and n represent the number of rows and columns of the image respectively.

[0022] The matrix decomposition method adopts singular value decomposition, which is specifically expressed as follows:

[0023]

[0024] In the formula, σ i is a singular value, and σ1≥σ2≥…≥σ n ≥0;e ij is the error term, satisfying |e ij |≤0.01.

[0025] The stable component matrix S t and the fluctuation component matrix F t The calculation method is as follows:

[0026]

[0027] F t =C t -S t ;

[0028] In the formula, k is the number of main singular values ​​selected, satisfying

[0029] The change trend matrix T t The calculation method is as follows:

[0030]

[0031] in:

[0032] In the formula, is the fluctuation component matrix F t The element in the i-th row and j-th column; Δt is the sampling time interval; α is the smoothing factor, and its value range is [0.1, 0.5].

[0033] The principal component analysis dimensionality reduction processing method is as follows:

[0034]

[0035] Where p is the dimension after dimensionality reduction, satisfying p≤min(m,n); w ij The elements of the principal component transformation matrix are given by the characteristic equation To obtain.

[0036] The risk discrimination matrix equation group is specifically expressed as follows:

[0037]

[0038] In the formula, [x1 x2 … x p ] T T low Column vector of ij is the weight coefficient; β1 and β2 are the correction coefficients; γ1e jωt and γ2e jωt is the periodic fluctuation term, where T is the blast furnace production cycle.

[0039] The threshold analysis equations are as follows:

[0040]

[0041] In the formula, θ s1 is the slag risk change rate threshold, ranging from [0.05, 0.1]; θ s2 is the slag risk threshold, ranging from [0.7, 0.9]; θ b1 is the risk change rate threshold of the large block, and its value range is [0.08, 0.15]; θ b2 is the risk threshold of large blocks, and its value range is [0.8, 0.95].

[0042] The parameter acquisition method is:

[0043] 1. Weight coefficient a ij The least squares method is used to obtain , as follows:

[0044]

[0045] In the formula, and is the actual risk value of the kth group of historical data; and is the predicted risk value of the kth group of data; N is the number of historical data groups, not less than 1000.

[0046] 2. The correction coefficients β1 and β2 are obtained by gradient descent method, as follows:

[0047]

[0048] In the formula, is the loss function; η is the learning rate, which ranges from [0.01, 0.1]; M is the number of experimental data sets, which is not less than 100.

[0049] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned risk warning method based on blast furnace tuyere image data processing.

[0050] A third aspect of the present invention provides a risk warning system based on blast furnace tuyere image data processing, which includes the above-mentioned computer-readable storage medium.

[0051] Compared with the prior art, the risk warning method, medium and system based on blast furnace tuyere image data processing provided by the present invention have the following beneficial effects:

[0052] 1. Improve monitoring accuracy and reliability. The method of the present invention uses color features as monitoring indicators, which can more accurately reflect the actual state of the blast furnace tuyere. At the same time, through technical means such as brightness equalization and matrix decomposition, interference factors such as ambient light are effectively eliminated, and the stability and reliability of detection are improved.

[0053] 2. Realize timely warning. The method of the present invention can not only monitor the current working status of the blast furnace in real time, but also predict the possible future risks of slag hanging and large pieces falling based on the status change trend, thus providing sufficient response time for production safety warning.

[0054] 3. Reduce computational complexity. The method of the present invention cleverly uses dimensionality reduction techniques such as principal component analysis to compress the original high-dimensional feature space into a low-dimensional space, greatly reducing the amount of calculation of the subsequent risk discrimination model and improving the operating efficiency of the overall system.

[0055] In summary, the present invention solves the technical problem that the prior art is difficult to achieve safe and efficient risk warning of slag hanging state and large piece falling state at the blast furnace tuyere. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0058] like Figure 1 As shown, it is a flow chart of a risk warning method based on blast furnace tuyere image data processing provided by the first aspect of the present invention, and the method comprises the following steps:

[0059] S10, collecting video data of blast furnace tuyere, and analyzing the video data according to a preset sampling time interval to obtain a blast furnace tuyere image sequence;

[0060] S20, performing brightness equalization processing on the blast furnace tuyere image sequence to eliminate the image brightness difference caused by the ambient light change, and obtain a brightness-equalized image sequence;

[0061] S30, performing color space conversion on the image sequence after brightness equalization, converting the RGB color space into the HSV color space, and extracting the color feature matrix of the blast furnace tuyere area;

[0062] S40, performing cluster analysis on the color feature matrix to obtain multiple color cluster centers, and identifying the slag hanging state and the large block falling state in the blast furnace tuyere area based on the color cluster centers;

[0063] S50, establishing a time series of a color feature matrix, and using a matrix decomposition method to decompose the color feature matrix into a stable component matrix and a fluctuating component matrix;

[0064] S60, establishing a change trend matrix of the blast furnace tuyere state based on the fluctuation component matrix, wherein the change trend matrix includes a change direction and a change amplitude of the blast furnace tuyere state;

[0065] S70, performing dimensionality reduction processing on the change trend matrix by using principal component analysis to obtain a low-dimensional change trend matrix;

[0066] S80, inputting the change trend matrix into a pre-set risk discrimination matrix equation group to obtain a risk vector; the risk vector includes a slag risk index and a large piece falling risk index;

[0067] S90. Perform threshold analysis based on the risk vector, obtain risk results, and output them.

[0068] The specific implementation methods of the above steps are described in detail below:

[0069] The specific implementation of step S10 is: first collect the video data of the blast furnace tuyere. For subsequent image processing, it is necessary to analyze the video data frame by frame according to the preset sampling time interval Δt to obtain the blast furnace tuyere image sequence t is the sequence subscript, and T is the total number of frames in the sequence. The purpose of this step is to extract the blast furnace tuyere image that needs to be analyzed from the original video data. Usually, Δt should be set between 1 second and 5 seconds to fully reflect the changes in the blast furnace tuyere state.

[0070] The specific implementation of step S20 is: Perform brightness equalization. Due to the complex working environment of the blast furnace, the ambient lighting conditions often change, which will cause obvious differences in image brightness. In order to eliminate the influence of this brightness difference, it is necessary to use a brightness equalization algorithm to process the image sequence. Commonly used brightness equalization algorithms include histogram equalization and adaptive histogram equalization. Specifically, for the t-th frame image I (t) , first calculate its histogram h (t) (v), then according to the histogram equalization formula Get the image after brightness equalization Where L is the grayscale level and N is the total number of image pixels. The purpose of this step is to eliminate the impact of ambient lighting changes on subsequent color feature extraction.

[0071] The specific implementation of step S30 is: Perform color space conversion. First, convert the RGB color space to the HSV color space. The HSV color space contains three important color features: hue, saturation, and value. Then, extract the hue of the blast furnace tuyere area. Saturation and brightness Values, forming the color feature matrix C t , the specific expression is:

[0072]

[0073] Where, t represents the time series number; Represents the hue value of the pixel in the i-th row and j-th column, and its value range is [0,360]; Represents the saturation value of the pixel in the i-th row and j-th column, and the value range is [0,1]; represents the brightness value of the pixel in the i-th row and j-th column, and its value range is [0,1]; m and n represent the number of rows and columns of the image respectively. The purpose of this step is to extract the color features that can reflect the state of the blast furnace tuyere from the image, and provide a data basis for subsequent state identification and risk analysis.

[0074] The specific implementation of step S40 is: t Cluster analysis is performed to identify the slag hanging state and large pieces falling state in the blast furnace tuyere area. First, the K-means clustering algorithm is used to cluster the features of hue h, saturation s and brightness v to obtain K color cluster centers. The clustering results can be expressed as:

[0075]

[0076] Then, based on the characteristics of these color cluster centers, the slag hanging state and large pieces falling state in the blast furnace tuyere area can be identified. Generally speaking, the color feature value is low in the slag hanging state, while the color feature value is high in the large pieces falling state. The purpose of this step is to use the image color features to identify the current working state of the blast furnace tuyere and provide basic data for subsequent risk analysis.

[0077] The specific implementation of step S50 is: establish a color feature matrix C t Time Series And use matrix decomposition method to decompose it into stable component matrix S t and the fluctuation component matrix F t Here, singular value decomposition (SVD) is used as the matrix decomposition method, which is specifically expressed as follows:

[0078]

[0079] Among them, σ i is a singular value, and σ1≥σ2≥…≥σ n ≥0;e ij is the error term, satisfying |e ij |≤0.01.

[0080] Stable component matrix S t and the fluctuation component matrix F t The calculation method is as follows:

[0081]

[0082] F t =C t -S t ;

[0083] Among them, k is the number of main singular values ​​selected, satisfying The purpose of this step is to separate the stable and fluctuating components in the color feature matrix, providing a basis for subsequent change trend analysis.

[0084] The specific implementation of step S60 is: based on the fluctuation component matrix F t Establish the change trend matrix T of blast furnace tuyere status t . Change trend matrix T t The calculation method is as follows:

[0085]

[0086] in:

[0087]

[0088] In the formula, is the fluctuation component matrix F t The i-th row and j-th column element in ; Δt is the sampling time interval; α is the smoothing factor, and its value range is [0.1, 0.5]. The purpose of this step is to extract the dynamic characteristics of the color feature matrix that reflect the change of the blast furnace tuyere state as the input for subsequent risk analysis.

[0089] The specific implementation of step S70 is: t Perform principal component analysis (PCA) dimensionality reduction processing to obtain a low-dimensional change trend matrix T low The specific calculation method is as follows:

[0090] First calculate T t The covariance matrix of in

[0091] Then solve the characteristic equation |∑-λI|=0 and get the eigenvalues ​​λ1,λ2,…,λ p and the corresponding eigenvectors w1,w2,…,w p , where p≤min(m,n) is the dimension after dimensionality reduction, satisfying

[0092] Finally, calculate the low-dimensional change trend matrix T low :

[0093] T low =T t ·[w1 w2 … w p ];

[0094] The purpose of this step is to further extract the key features in the change trend matrix and provide more effective input for subsequent risk identification.

[0095] The specific implementation of step S80 is: transform the dimension-reduced change trend matrix T low Input it into the pre-set risk judgment matrix equation group to obtain the slag risk index r s and the risk index of falling blocks b The risk discrimination matrix equation group is specifically expressed as follows:

[0096]

[0097] Among them, [x1 x2 … x p ] T T low Column vector of ij is the weight coefficient; β1 and β2 are the correction coefficients; γ1e jωt and γ2e jωt is the periodic fluctuation term, where T is the blast furnace production cycle.

[0098] Weight coefficient a ij The least square method is used to obtain , as follows:

[0099]

[0100] in, and is the actual risk value of the kth group of historical data; and is the predicted risk value of the kth group of data; N is the number of historical data groups, not less than 1000.

[0101] The correction coefficients β1 and β2 are obtained by gradient descent method, as follows:

[0102]

[0103] in, is the loss function; η is the learning rate, ranging from [0.01, 0.1]; M is the number of experimental data sets, not less than 100.

[0104] The purpose of this step is to establish a matrix equation that can predict the risk of slag hanging and large pieces falling according to the characteristics of the changing trend.

[0105] The specific implementation of step S90 is: according to the risk index r obtained in step S80 s and r b , perform threshold analysis, obtain risk results, and output them. The threshold analysis equations are as follows:

[0106]

[0107] Among them, θ s1 is the slag risk change rate threshold, with a value range of [0.05, 0.1]; θ s2 is the slag risk threshold, ranging from [0.7, 0.9]; θ b1 is the risk change rate threshold of large blocks, with a value range of [0.08, 0.15]; θ b2 is the risk threshold of large blocks, and its value range is [0.8, 0.95].

[0108] The purpose of this step is to determine whether there is a risk of slag hanging or large pieces falling at the current blast furnace tuyere based on the pre-set thresholds, and output the final risk warning results.

[0109] Optionally, in order to further improve the computational efficiency, a GPU may be used to perform the matrix calculation process.

[0110] In summary, this risk warning method based on blast furnace tuyere image data processing mainly includes the following core steps:

[0111] The image sequence is extracted from the blast furnace tuyere video and the brightness is equalized.

[0112] Convert the image to HSV color space, extract hue, saturation and brightness features, and construct a color feature matrix.

[0113] The color feature matrix is ​​decomposed into stable components and fluctuating components by matrix decomposition method, and the dynamic characteristics reflecting the changes in the state of blast furnace tuyere are extracted from the fluctuating components.

[0114] The principal component analysis is performed on the dynamic features to reduce the dimension and obtain a low-dimensional change trend matrix.

[0115] A risk identification model is established to predict the slag hanging risk and large piece falling risk at the blast furnace tuyere according to the change trend matrix.

[0116] Set reasonable risk thresholds to achieve online monitoring and early warning of slag hanging risks and large chunk falling risks.

[0117] This method combines image processing, matrix analysis, machine learning and other technologies, can effectively eliminate the interference of environmental factors, extract stable and reliable risk warning features, and provide important support for blast furnace production safety. In the specific implementation, it is necessary to reasonably set the threshold parameters according to the factory environment and production characteristics to improve the accuracy and reliability of risk warning.

[0118] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned risk warning method based on blast furnace tuyere image data processing.

[0119] A third aspect of the present invention provides a risk warning system based on blast furnace tuyere image data processing, which includes the above-mentioned computer-readable storage medium.

[0120] Specifically, the principle of the present invention is:

[0121] Image acquisition and preprocessing. First, collect real-time video data from the camera installed at the blast furnace tuyere, and analyze the video frame by frame at a certain time interval to obtain an image sequence of the blast furnace tuyere. Due to the complex working environment of the blast furnace, there are obvious changes in ambient lighting, which will lead to large differences in image brightness. In order to eliminate this interference factor, it is necessary to perform brightness equalization processing on the image sequence so that each frame of the image has similar overall brightness characteristics. The purpose of this step is to extract stable and reliable image features from the original video data.

[0122] Color feature extraction. Convert the equalized image sequence to the HSV color space, extract the hue, saturation and brightness features of each pixel, and form a color feature matrix. Color features can more intuitively reflect the working status of the blast furnace tuyere area. For example, the color feature value is lower in the slag hanging state, while the color feature value is higher in the large block falling state. The purpose of this step is to obtain key features that can characterize the blast furnace state from the image.

[0123] Analysis of state change trend. For the time series of the color feature matrix, the matrix decomposition technology is used to decompose it into stable components and fluctuating components. Among them, the fluctuating component contains key information reflecting the change of the blast furnace state. Based on the fluctuating component, the change trend matrix of the blast furnace tuyere state is further calculated, which can reflect the direction and amplitude of the state change. The purpose of this step is to extract dynamic features that can predict future risks from the original color features.

[0124] Risk identification and early warning. In order to achieve early warning of slag risk and large-block risk, it is necessary to establish a risk identification model based on the change trend matrix. Specifically, the dimensionality-reduced change trend matrix is ​​input into the pre-trained risk identification equation group to obtain the slag risk index r. s and the risk index of falling blocks b Finally, according to the preset risk threshold, it is determined whether there is a risk of slag hanging or falling large pieces in the current blast furnace, and the warning results are output in time. The purpose of this step is to convert dynamic features into specific risk warning information.

[0125] The time from risk perception to danger occurrence is often short, so the entire detection process needs to be carried out in high speed and real time, and the computational complexity is also relatively high. To this end, the solution of the present invention adopts a number of optimization measures:

[0126] GPU acceleration technology is used to utilize parallel computing capabilities to significantly improve the system's computing speed.

[0127] The feature space is reduced in dimension through principal component analysis, which greatly reduces the computational complexity of the subsequent risk discrimination model.

[0128] Efficient matrix decomposition algorithms, such as singular value decomposition, are used to achieve rapid decomposition of large-scale color feature matrices.

[0129] An optimized gradient descent algorithm is used to quickly solve the correction coefficient in the risk discrimination model.

[0130] Through the comprehensive application of the above-mentioned technical means, the solution of the present invention can greatly improve the operating efficiency of the overall system while ensuring the detection accuracy, and ensure real-time monitoring of the working status of the blast furnace and risk warning.

[0131] In summary, the risk warning method of the present invention makes full use of image processing and machine learning technology to achieve accurate monitoring of the blast furnace tuyere state, and combines the state change trend analysis to effectively predict the possible future risks of slag hanging and falling large pieces. Through algorithm optimization and hardware acceleration, the method can meet the requirements of real-time detection and provide a strong guarantee for blast furnace production safety.

[0132] The following is an example of a specific application scenario of the present invention: a steel company has a blast furnace that produces 300 tons of pig iron per day. In order to ensure production safety, the company decides to install a video monitoring system near the blast furnace tuyere and use the risk warning method based on image recognition proposed in the present invention to monitor the blast furnace status in real time.

[0133] First, the company installed three high-definition cameras at the blast furnace tuyere, covering different angles of vision to ensure comprehensive monitoring of the entire tuyere area. These cameras capture one frame per second and transmit the video data to a centralized monitoring center for processing and analysis.

[0134] According to step S10 of the present invention, the monitoring center analyzes the collected blast furnace tuyere video data frame by frame to obtain an image sequence Where T is the total number of frames. Due to the complex environment around the blast furnace, the lighting conditions often change, resulting in large differences in the brightness of the images captured at different times. To this end, the monitoring center uses the histogram equalization algorithm described in step S20 to perform brightness equalization on the image sequence to obtain a preprocessed image sequence.

[0135] Next, the monitoring center converts the brightness-equalized image sequence into Convert to HSV color space and extract the hue of each pixel Saturation and brightness Features, forming the color feature matrix C t . Take the matrix C at time t=100 100 For example, its specific values ​​are shown in Table 1:

[0136] Table 1 Color feature matrix C at time t=100 100 ;

[0137] h s v 1,1 230.4 0.82 0.65 1,2 235.1 0.77 0.59 … … … … 480,640 245.3 0.91 0.72

[0138] It can be seen from Table 1 that the color characteristic values ​​of the blast furnace tuyere area at this moment are generally low, indicating that the blast furnace may be in a slag hanging state at this time.

[0139] According to step S40, the monitoring center then Cluster analysis is performed on time series data to identify the slag hanging state and large pieces falling state in the blast furnace tuyere area. The K-means algorithm is used to cluster the features of the three dimensions of hue h, saturation s and brightness v, and 5 color cluster centers are obtained. By analyzing the characteristics of these cluster centers, we can find that:

[0140] Cluster center 1 (μ1) corresponds to the dross state, with a lower hue and medium saturation and brightness;

[0141] Cluster center 2 (μ2) and cluster center 3 (μ3) correspond to the normal working state, with high hue, saturation and brightness;

[0142] Cluster center 4 (μ4) and cluster center 5 (μ5) correspond to the large block state, with low hue, saturation and brightness.

[0143] Based on the above analysis results, the monitoring center can identify the current working status of the blast furnace in real time and provide basic data for subsequent risk warnings.

[0144] In order to deeply analyze the changing trend of the blast furnace tuyere status, the monitoring center converts the color feature matrix into The time series is matrix decomposed to obtain the stable component matrix and the fluctuation component matrix Taking t = 100 as an example, the stable component matrix S 100 and the fluctuation component matrix F 100 The values ​​are shown in Table 2 and Table 3:

[0145] Table 2 Stable component matrix S at time t = 100 100 ;

[0146] h s v 1,1 232.7 0.79 0.62 1,2 237.4 0.74 0.56 … … … … 480,640 242.9 0.88 0.69

[0147] Table 3 Fluctuation component matrix F at time t = 100 100 ;

[0148] h s v 1,1 -2.3 0.03 0.03 1,2 -2.3 0.03 0.03 … … … … 480,640 2.4 0.03 0.03

[0149] It can be seen from Tables 2 and 3 that compared with the stable component, the fluctuating component is smaller at each pixel point, indicating that the state of the blast furnace tuyere is relatively stable at this time.

[0150] According to step S60, the monitoring center uses the fluctuation component matrix Calculate the change trend matrix Taking t=100 as an example, the change trend matrix T 100 Some values ​​are shown in Table 4:

[0151] Table 4 The change trend matrix T at time t=100 100 ;

[0152] t 1,1 -0.02 1,2 -0.02 … … 480,640 0.02

[0153] It can be seen from Table 4 that at this time, the overall trend of the color characteristics of the blast furnace tuyere area is relatively gentle, with no obvious increase or decrease.

[0154] Next, the monitoring center calculates the change trend matrix according to step S70. Perform principal component analysis and dimensionality reduction to obtain a low-dimensional change trend matrix Taking t=100 as an example, the change trend matrix after dimensionality reduction is As shown in Table 5:

[0155] Table 5 Low-dimensional change trend matrix at time t=100

[0156] Feature 1 Feature 2 Feature 3 -0.08 0.02 -0.04

[0157] As can be seen from Table 5, after principal component analysis, the original high-dimensional change trend matrix is ​​compressed into three main features, providing more refined input data for subsequent risk identification.

[0158] According to step S80, the monitoring center converts the above low-dimensional change trend matrix Input into the pre-trained risk discrimination model to calculate the slag risk index r s and the risk index of falling blocks b Taking t=100 as an example, the risk index at this moment is:

[0159] r s =0.62;

[0160] r b =0.31;

[0161] From the results, it can be seen that there is a certain risk of slag hanging in the blast furnace at this time, but the risk of large pieces falling is relatively low.

[0162] Finally, the monitoring center compares the above risk index with the preset threshold according to step S90 to obtain the final risk warning result:

[0163] Currently, there is a certain risk of slag hanging in blast furnaces, but the risk of large pieces falling is relatively low.

[0164] Slag risk index r s =0.62, higher than the slag risk threshold θ s2 =0.55, triggering slag risk warning.

[0165] Large block risk index rb =0.31, lower than the risk threshold of falling large blocks θ b2 =0.80, no large-block falling risk warning was triggered.

[0166] The monitoring center immediately issued a slag risk warning, reminding on-site operators to take appropriate emergency measures to ensure the safety of blast furnace production.

[0167] During the subsequent monitoring process, the monitoring center continued to track changes in the status of the blast furnace tuyere.

[0168] Table 6 Risk index changes in the past 24 hours

[0169] time <![CDATA[slag hanging risk index r s > <![CDATA[Risk index r of large block falling b > 00:00 0.52 0.25 01:00 0.55 0.28 02:00 0.58 0.32 … … … 23:00 0.62 0.31

[0170] As can be seen from Table 6, in the past 24 hours, the risk of slag hanging in the blast furnace has shown a gradual upward trend, while the risk of large pieces falling is relatively stable. The monitoring center will continue to pay attention to the changes in the status of the blast furnace and take necessary emergency measures according to the actual situation to ensure production safety.

[0171] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A risk warning method based on blast furnace tuyere image data processing, characterized in that: The following steps are involved: S10, collecting video data of blast furnace tuyere, and analyzing the video data according to a preset sampling time interval to obtain a blast furnace tuyere image sequence; S20, performing brightness equalization processing on the blast furnace tuyere image sequence to eliminate the image brightness difference caused by the ambient light change, and obtain a brightness equalized image sequence; S30, performing color space conversion on the image sequence after brightness equalization, converting the RGB color space into the HSV color space, and extracting the color feature matrix of the blast furnace tuyere area; S40, performing cluster analysis on the color feature matrix to obtain a plurality of color cluster centers, and identifying the slag hanging state and the large-piece falling state in the tuyere area of ​​the blast furnace based on the color cluster centers; S50, establishing a time series of the color feature matrix, and using a matrix decomposition method to decompose the color feature matrix into a stable component matrix and a fluctuating component matrix; S60, establishing a change trend matrix of the blast furnace tuyere state based on the fluctuation component matrix, wherein the change trend matrix includes a change direction and a change amplitude of the blast furnace tuyere state; S70, performing dimensionality reduction processing on the change trend matrix by using principal component analysis to obtain a low-dimensional change trend matrix; S80, inputting the change trend matrix into a preset risk discrimination matrix equation group to obtain a risk vector; The risk vector includes a slag risk index and a large piece falling risk index; S90: Perform threshold analysis according to the risk vector, obtain risk results, and output them.

2. The risk warning method based on blast furnace tuyere image data processing according to claim 1 is characterized in that: The color feature matrix is ​​specifically expressed as follows: Where t represents the time series number; h ij Indicates the hue value of the pixel in the i-th row and j-th column, and its value range is [0,360]; s ij represents the saturation value of the pixel in the i-th row and j-th column, and its value range is [0,1]; v ij Represents the brightness value of the pixel in the i-th row and j-th column, and its value range is [0,1]. m and n represent the number of rows and columns of the image respectively.

3. The risk warning method based on blast furnace tuyere image data processing according to claim 2 is characterized in that: The steps of decomposing the color feature matrix into a stable component matrix and a fluctuating component matrix by using a matrix decomposition method are specifically expressed as follows: In the formula, σ i is a singular value, and σ1≥σ2≥…≥σ n ≥0;e ij is the error term, satisfying |e ij |≤0.01; Among them, the stable component matrix Fluctuation component matrix F t =C t -S t ; In the formula, k is the number of main singular values ​​selected, satisfying 4. The risk warning method based on blast furnace tuyere image data processing according to claim 3 is characterized in that: The change trend matrix is ​​specifically expressed as follows: in: In the formula, is the fluctuation component matrix F t The element in the i-th row and j-th column; Δt is the sampling time interval; α is the smoothing factor, and its value range is [0.1, 0.5].

5. The risk warning method based on blast furnace tuyere image data processing according to claim 4 is characterized in that: The risk discrimination matrix equation group is specifically expressed as follows: In the formula, [x1 x2 … x p ] T T low Column vector of ij is the weight coefficient; β1 and β2 are the correction coefficients; γ1e jωt and γ2e j ωt is the periodic fluctuation term, where T is the blast furnace production cycle, Where p is the dimension after dimensionality reduction, satisfying p≤min(m,n); w ij The elements of the principal component transformation matrix are given by the characteristic equation To obtain.

6. The risk warning method based on blast furnace tuyere image data processing according to claim 5 is characterized in that: The threshold analysis is specifically expressed as follows: In the formula, θ s1 is the slag risk change rate threshold, ranging from [0.05, 0.1]; θ s2 is the slag risk threshold, ranging from [0.7, 0.9]; θ b1 is the risk change rate threshold of the large block, and its value range is [0.08, 0.15]; θ b2 is the risk threshold of large blocks, and its value range is [0.8, 0.95].

7. The risk warning method based on blast furnace tuyere image data processing according to claim 6 is characterized in that: Weight coefficient a ij The least squares method is used to obtain , as follows: In the formula, r s (k) and r b (k) is the actual risk value of the kth group of historical data; and is the predicted risk value of the kth group of data; N is the number of historical data sets, no less than 1000.

8. The risk warning method based on blast furnace tuyere image data processing according to claim 7 is characterized in that: The correction coefficients β1 and β2 are obtained by gradient descent method, as follows: In the formula, is the loss function; η is the learning rate, ranging from [0.01, 0.1]; M is the number of experimental data sets, not less than 100.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the risk warning method based on blast furnace tuyere image data processing as described in any one of claims 1 to 8.

10. The risk warning system based on blast furnace tuyere image data processing is characterized by: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.

Images