Crystallizer caking early warning method and system based on characteristic regions

Abstract

The application provides a crystallizer sticking early warning method and system based on a feature area, and relates to the technical field of breakout, and the method comprises the following steps: S110, data acquisition is performed on a single thermocouple, and basic feature data of upper and lower thermocouples is determined according to the collected data; S120, a motion viscosity coefficient and the basic feature data of the upper and lower thermocouples form a multi-dimensional sticking feature vector; S130, the multi-dimensional sticking feature vector is mapped to a high-dimensional feature space, and a sticking feature area is obtained by taking the multi-dimensional sticking feature vector as a feature point; S140, the to-be-judged sticking feature vector is compared with the sticking feature area, and if the to-be-judged sticking feature vector is not in the sticking feature area, a crystallizer sticking early warning is issued. By using the application, the boundary of the feature area can be adaptively defined, and the off-line data analysis link of the traditional early warning system is avoided, so that the application cycle of the system is shortened.

Description

Technical Field

[0001] This invention relates to the field of iron and steel metallurgy technology, and in particular to a method and system for early warning of crystallizer adhesion based on characteristic regions. Background Technology

[0002] Blistering occurs when the thinner initial billet shell near the meniscus cracks during continuous casting, allowing molten steel to seep out and adhere to the copper plates of the mold. As the mold vibrates and the billet moves downwards, the adhered mass repeatedly tears and heals, continuing to move downwards. When it reaches the lower part of the mold, the presence of air gaps causes it to lose the support and constraint of the copper plates, resulting in molten steel overflowing from the cracks, causing leaks. Leaks not only endanger the safety of on-site operators and severely damage continuous casting equipment, but also force the interruption of continuous casting production, significantly increasing equipment maintenance and production costs. Metallurgists both domestically and internationally have developed various early warning products to prevent leaks and reduce production losses.

[0003] Currently, existing methods for predicting steel leakage mainly involve embedding thermocouples in the copper plate of the crystallizer to monitor and determine whether adhesion occurs between the billet and the copper plate based on changes in the thermocouple temperature.

[0004] Based on the practical application of existing leakage prediction technologies, leakage prediction methods based on logical judgment are highly dependent on continuous casting equipment, process parameters, and physical property parameters. When the process is adjusted and the casting speed is increased, the parameters change significantly, leading to a substantial increase in the false alarm rate and the missed alarm rate. Neural network methods have high requirements for learning and training samples. Incomplete or invalid samples will seriously affect their prediction effect, and the model has low generalization ability.

[0005] In summary, the adhesion warning system generally suffers from low accuracy and high false alarm rate. During use, parameters need to be constantly adjusted according to different process conditions or even the deterioration of production equipment. Parameter adjustment is difficult and hard to achieve. Developing an adaptive system with high generalization ability and no parameter adjustment is the current focus and is of great significance for reducing losses in the production process.

[0006] Therefore, there is an urgent need for an action pose recognition method that can improve detection speed without sacrificing detection accuracy. Summary of the Invention

[0007] In view of the above problems, the present invention provides a crystallizer adhesion early warning method and system based on feature regions, the main purpose of which is to improve the generalization ability and realize an adaptive early warning system without adjustment parameters.

[0008] To achieve the above objectives, the present invention provides a crystallizer adhesion early warning method based on feature regions, applied to electronic devices. The method includes: S110, collecting data for a single thermocouple, and performing linear fitting on the collected data to determine the basic feature data of the upper and lower thermocouples; S120, determining the kinematic viscosity coefficient of the upper and lower thermocouples using a formula based on the basic feature data of the upper and lower thermocouples, and forming a multidimensional adhesion feature vector based on the kinematic viscosity coefficient and the basic feature data of the upper and lower thermocouples; S130, obtaining the adhesion feature region by mapping the multidimensional adhesion feature vector to a high-dimensional feature space, using the multidimensional adhesion feature vector as feature points; S140, comparing the adhesion feature vector to be determined formed based on the thermocouple data to be determined with the adhesion feature region, and if the adhesion feature vector to be determined is not within the adhesion feature region, a crystallizer adhesion early warning occurs.

[0009] Preferably, in step S130, the bonding feature region is obtained by determining the boundary values ​​of the bonding feature region. The method for determining the boundary of the bonding feature region includes: S210, mapping the multidimensional bonding feature vector to a high-dimensional feature space, and calculating the distribution function F(X) with the multidimensional bonding feature vector as feature points in the high-dimensional feature space, where F represents the feature space and X is the multidimensional bonding feature vector of the feature points; X = {x i |i=1...k}, k is the dimension of the feature vector, x i Let be the i-th component of the eigenvector; S220, take the second derivative of the distribution function F(X) to obtain the second derivative of F(X), and determine the maxima and minima of each dimension as critical values ​​based on the second derivative of F(X); F″(Xi) <=L imax ;F″(Xi)>=L imin Where Li is the critical parameter of the i-th dimension component.

[0010] Preferably, the adhesion characteristic data of the upper thermocouple includes at least: the upper thermocouple rising and falling mode Umode, the upper thermocouple rising slope Uslap, the upper thermocouple temperature rise index UMaxTCoeff, and the maximum falling slope USlapNeg in the upper thermocouple rising and falling mode; the adhesion characteristic data of the lower thermocouple includes at least: the lower thermocouple rising and falling mode Dmode, the lower thermocouple rising slope Dslap, the lower thermocouple temperature rise index DMaxTCoeff, and the maximum falling slope DSlapNeg in the lower thermocouple rising and falling mode.

[0011] Preferably, in step S120, the formula for obtaining the kinematic viscosity coefficients of the upper and lower thermocouples is as follows:

[0012]

[0013]

[0014] Wherein, the y-coordinate of the upper thermocouple is Yu, the y-coordinate of the lower thermocouple is Yd, the current billet pulling speed is v (m / min), the time of the highest temperature point of the upper thermocouple is UMaxPoint, the time of the highest temperature point of the lower thermocouple is DMaxPoint, and HotPoint V is the hot spot movement speed.

[0015] Preferably, the temperature rise index is obtained by the following formula:

[0016] MaxTCoeff=Tmax / average(T,M)-1;

[0017] Where M is the number of data points, average(T,M) is the average temperature of the M data points, and Tmax is the highest temperature among the M data points.

[0018] Furthermore, to achieve the above objectives, the present invention also provides a crystallizer adhesion early warning system based on feature regions, comprising: a data acquisition module for acquiring data for a single thermocouple; a feature vector acquisition module for linearly fitting the acquired data to determine the basic feature data of the upper and lower thermocouples; then, based on the basic feature data of the upper and lower thermocouples, determining the kinematic viscosity coefficients of the upper and lower thermocouples using a formula, and forming a multidimensional adhesion feature vector based on the kinematic viscosity coefficients and the basic feature data of the upper and lower thermocouples; a feature region formation module for obtaining adhesion feature regions by mapping the multidimensional adhesion feature vectors to a high-dimensional feature space, using the multidimensional adhesion feature vectors as feature points; and an adhesion early warning module for comparing the adhesion feature vector to be determined formed based on the thermocouple data to be determined with the adhesion feature region, and issuing a crystallizer adhesion early warning if the adhesion feature vector to be determined is not within the adhesion feature region.

[0019] Preferably, the feature region forming module includes: a distribution function determination unit, used to map the multidimensional bonding feature vector to a high-dimensional feature space, and calculate the distribution function F(X) with the multidimensional bonding feature vector as feature points in the high-dimensional feature space, where F represents the feature space and X is the multidimensional bonding feature vector of the feature points; the X = {x i |i=1...k}, k is the dimension of the feature vector, x i Let F(X) be the i-th component of the eigenvector; the boundary determination unit is used to perform second-order differentiation on the distribution function F(X) to obtain the second derivative of F(X), and determine the maxima and minima of each dimension as critical values ​​based on the second derivative of F(X); F″(X)i ) <= L imax ;F″(X i )>=L imin , where Li is the critical parameter of the i-th dimension component.

[0020] To achieve the above objectives, the present invention also provides an electronic device, which includes a memory and a processor. The memory stores a crystallizer adhesion warning program based on feature regions. When the processor executes the crystallizer adhesion warning program based on feature regions, it performs the following steps: S110, data is collected for a single thermocouple, and the collected data is linearly fitted to determine the basic feature data of the upper and lower thermocouples; S120, based on the basic feature data of the upper and lower thermocouples, the kinematic viscosity coefficients of the upper and lower thermocouples are determined by a formula, and a multidimensional adhesion feature vector is formed based on the kinematic viscosity coefficients and the basic feature data of the upper and lower thermocouples; S130, the adhesion feature region is obtained by mapping the multidimensional adhesion feature vector to a high-dimensional feature space and using the multidimensional adhesion feature vector as feature points; S140, the adhesion feature vector to be determined formed based on the data of the thermocouple to be determined is compared with the adhesion feature region. If the adhesion feature vector to be determined is not within the adhesion feature region, a crystallizer adhesion warning is issued.

[0021] Preferably, the bonding feature region is obtained by determining the boundary values ​​of the bonding feature region. The method for determining the boundary of the bonding feature region includes: S210, mapping a multidimensional bonding feature vector to a high-dimensional feature space, and calculating a distribution function F(X) with the multidimensional bonding feature vector as feature points in the high-dimensional feature space, where F represents the feature space and X is the multidimensional bonding feature vector of the feature points; X = {x i |i=1...k}, k is the dimension of the feature vector, x i Let be the i-th component of the eigenvector; S220, take the second derivative of the distribution function F(X) to obtain the second derivative of F(X), and determine the maxima and minima of each dimension as critical values ​​based on the second derivative of F(X); F″(Xi) <=L imax ;F″(Xi)>=L imin Where Li is the critical parameter of the i-th dimension component.

[0022] The present invention proposes a crystallizer adhesion early warning method and system based on feature regions. This method collects invariant features of crystallizer adhesion in time and space, and uses these invariant features to construct a static model to describe the adhesion development process. The beneficial effects are as follows:

[0023] 1. Because the early warning judgment is based on the invariant characteristics of crystallizer adhesion, it is more in line with the law of adhesion occurrence and development, thus achieving the effect of high early warning accuracy and low false alarm rate.

[0024] 2. The process of billet bonding change is described as point motion (high-dimensional point vibration) in a high-dimensional abstract space; no parameter adjustment is required in application, and the system adapts to changes in equipment, process and materials through adaptive system parameters;

[0025] 3. It solves the problem of poor generalization ability of other models, avoids the adjustment and waiting period of system parameter adjustment, and can enter the crystallizer adhesion early warning monitoring as soon as the system goes online;

[0026] 4. No offline data is required; the boundaries of the feature area are adaptively defined, avoiding the offline data analysis required by traditional early warning systems, thus shortening the system application cycle. Attached Figure Description

[0027] Figure 1 This is a flowchart of a preferred embodiment of the crystallizer adhesion early warning method based on feature regions of the present invention;

[0028] Figure 2 This is a schematic diagram illustrating the principle of a preferred embodiment of the crystallizer adhesion early warning method based on feature regions of the present invention;

[0029] Figure 3 The multidimensional bonding feature vector of this invention is used as a state diagram of feature points distributed in a high-dimensional feature space;

[0030] Figure 4 This is a logic structure diagram of a crystallizer adhesion early warning system based on feature regions according to an embodiment of the present invention;

[0031] Figure 5 This is a schematic diagram illustrating the application environment of the crystallizer adhesion early warning method based on feature regions according to an embodiment of the present invention.

[0032] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0033] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0034] This invention provides a method for early warning of crystallizer adhesion based on feature regions. It describes the process of billet adhesion change as point movement (high-dimensional point vibration) in a high-dimensional abstract space. It is a method to judge whether the crystallizer is adhered based on the invariant characteristics of crystallizer adhesion in time and space.

[0035] It should be noted that, compared to the large-scale production process of cast billets, adhesion is a rare abnormal phenomenon. Adhesion is significantly different from normal production conditions; therefore, its occurrence signifies a transition from normal to abnormal production conditions, and this abnormal state is sudden and rapid. If, in the spatial dimension, the process of billet adhesion is described as the distribution of points in a high-dimensional abstract space, then the abnormal state manifests as an abnormal distribution of points. In the temporal dimension, if the process of billet adhesion is described as the movement of points in a high-dimensional abstract space, then the abnormal state manifests as movement from one distribution area to another.

[0036] This invention provides a method for early warning of crystallizer adhesion based on characteristic regions. Figure 1 A flowchart of a crystallizer adhesion early warning method based on a feature region according to an embodiment of the present invention is shown. This method can be performed by a device, which can be implemented in software and / or hardware.

[0037] The quality of continuously cast billets and smooth production are two major issues that must be addressed to achieve efficient continuous casting; however, sticking and leakage are the main forms of leakage during continuous casting. Leakage prediction technology is divided into two types: leakage prediction systems and thermal imaging systems. Both utilize thermocouples embedded in the copper plate of the crystallizer to measure the temperature and temperature change trend at various points on the copper plate of the crystallizer during the casting process. Based on certain rules, it identifies possible sticking points (the billet shell sticking to the copper plate of the crystallizer), issues an alarm, and automatically reduces speed or stops the process according to certain rules.

[0038] Under normal casting conditions, the temperature of the upper thermocouple is higher than that of the lower thermocouple due to the continuous downward movement of the newly formed high-temperature billet shell within the crystallizer. When the billet shell bonds and breaks, the added molten steel comes into direct contact with the copper plate, causing the temperature of the upper thermocouple to rise. A weak billet shell forms at the break point and continues to move downward, adhering tightly to the copper wall under the hydrostatic pressure of the molten steel, thus raising the temperature of the lower thermocouple as well. When bonding is severe, the temperature rise of both the upper and lower thermocouples can reach a certain value. It should be noted that the upper thermocouple refers to the thermocouples in the upper row, and the lower thermocouple refers to the thermocouples in the lower row.

[0039] Based on the temperature characteristics of the upper and lower thermocouples in the above-mentioned adhesion phenomenon, the steel leakage prediction technology often measures the temperature in the longitudinal direction of the thermocouples. That is, after the temperature of the upper row of thermocouples is detected, the temperature of the lower row of thermocouples in the same group is immediately detected.

[0040] like Figure 1 As shown, the present invention provides a crystallizer adhesion early warning method based on a feature region, which is applied to an electronic device. The method includes: S110-S140.

[0041] S110. Data is collected for a single thermocouple, and linear fitting is performed on the collected data to determine the basic characteristic data of the upper and lower thermocouples. The upper thermocouple is the upper row of thermocouples, and the lower thermocouple is the lower row of thermocouples.

[0042] In order to collect the invariant characteristics of the crystallizer bonded in time and space, it is necessary to collect data for individual thermocouples; that is, to collect the temperature data of all thermocouples in the copper plate of the crystallizer at a specific moment.

[0043] Key factors influencing crystallizer adhesion include those that reflect the effects of crystallizer adhesion under the same experimental conditions. Crystallizer adhesion sample points with identical values ​​for these key factor attributes are distributed in relatively independent regions, forming adhesion characteristic areas.

[0044] Linear fitting of the collected data is performed using the least squares linear regression algorithm to perform multivariate fitting, thereby obtaining basic feature data. Specifically, a scalar curve is formed based on the collected data point values ​​using the least squares method, and then feature analysis is performed on the formed scalar curve, such as whether the curve is in an upward or downward pattern, if it is in an upward pattern, then the slope of the upward movement, and if it is in a downward pattern, then the maximum slope of the downward movement, etc.

[0045] Further, basic characteristic data include rising and falling patterns, rising fitted slope (i.e., rising slope), maximum falling slope, highest temperature time point, and temperature rise index.

[0046] The process of acquiring basic feature data includes: First, obtaining the slope features of multiple data points through linear fitting of the data points; segmenting the collected data based on the positive or negative slope features; where a positive slope indicates an upward trend in the temperature curve, and a negative slope indicates a downward trend in the temperature curve; Second, calculating the slope of the temperature curve in the upward or downward trend; and calculating the maximum downward slope of the temperature curve in the downward trend; recording the time of occurrence of the highest temperature point, i.e., the highest temperature time point MaxPoint; calculating the temperature rise index. Assuming the number of data points is M, calculating the average temperature average(T, M) of the M data points, where the data point with the highest temperature among these M data points is selected and its highest temperature is recorded as Tmax, then the temperature rise index MaxTCoeff = Tmax / average(T, M) - 1.

[0047] Based on the above data processing, basic characteristic data are obtained for the upper and lower thermocouples respectively. The basic bonding characteristic data for the upper thermocouple includes at least: the upper thermocouple rising and falling modes Umode, the upper thermocouple rising slope Uslap, the upper thermocouple temperature rise index UMaxTCoeff, and the maximum falling slope USlapNeg in the upper thermocouple rising and falling modes. The basic bonding characteristic data for the lower thermocouple includes at least: the lower thermocouple rising and falling modes Dmode; the lower thermocouple rising slope Dslap, the lower thermocouple temperature rise index DMaxTCoeff, and the maximum falling slope DSlapNeg in the lower thermocouple rising and falling modes.

[0048] It should be noted that D** represents the lower thermocouple and U** represents the upper thermocouple.

[0049] S120. Based on the basic characteristic data of the upper and lower thermocouples, determine the kinematic viscosity coefficients of the upper and lower thermocouples using formulas, and form a multidimensional adhesion feature vector based on the kinematic viscosity coefficients and the basic characteristic data of the upper and lower thermocouples.

[0050] It should be noted that in the construction of the bonding judgment feature vector (i.e. bonding feature vector), the features of a single thermocouple are used as components, and the features of the combination also need to be used as components; therefore, it is necessary to calculate the combined feature data through the feature data of a single thermocouple.

[0051] In a specific embodiment, the formula for obtaining the kinematic viscosity coefficient UDMovementCoeff of the upper and lower thermocouples is as follows:

[0052]

[0053]

[0054] Wherein, the y-coordinate of the upper thermocouple is Yu, the y-coordinate of the lower thermocouple is Yd, the current billet pulling speed is v (m / min), the time of the highest temperature point of the upper thermocouple is UMaxPoint, the time of the highest temperature point of the lower thermocouple is DMaxPoint, and HotPoint V is the hot spot movement speed.

[0055] It should be noted that the thermocouples are installed according to a pre-designed thermocouple arrangement scheme, and the coordinates of the upper and lower thermocouples can be determined by the thermocouple arrangement scheme.

[0056] The kinematic viscosity coefficients of the upper and lower thermocouples, combined with the basic characteristic data of the upper and lower thermocouples, form a multidimensional adhesion feature vector, that is, the invariant features constitute a static model used to describe the adhesion development process; in a specific embodiment, the multidimensional feature vector P is a nine-dimensional feature vector:

[0057] P=[Umode, Uslap, UMaxTCoeff, USlapNeg, Dmode, Dslap, DMaxTCoeff, DSlapNeg, UDMovementCoeff].

[0058] The dimensional description of the feature vector P includes: the upper thermocouple's rising and falling mode Umode (true if it's a rising + falling mode, false otherwise); the upper thermocouple's rising slope Uslap (calculated from the temperature curve of the upper thermocouple); the upper thermocouple's temperature rise index UMaxTCoeff (calculated using UMaxTCoeff = Tmax / average(T, M) - 1); USlapNeg (maximum falling slope in rising and falling modes); and the lower thermocouple's rising and falling mode Dmode (true if it's a rising + falling mode). Otherwise, it is false; the rising slope of the lower thermocouple, Dslap, is calculated based on the temperature curve of the lower thermocouple; the temperature rise index of the lower thermocouple, DMaxTCoeff, is calculated using DMaxTCoeff = Tmax / average(T, M)-1; DSlapNeg: the maximum falling slope in rising and falling modes; in addition, the kinematic viscosity coefficients of the upper and lower thermocouples are calculated according to the above calculation formulas; then, the above feature data are used as data of each dimension to form a multi-dimensional feature vector, that is, a nine-dimensional feature vector describing the combined bonding signal.

[0059] S130. By mapping the multidimensional bonding feature vector to a high-dimensional feature space, the bonding feature region is obtained using the multidimensional bonding feature vector as feature points.

[0060] In other words, each eigenvector is considered as a high-dimensional point, and these high-dimensional points are distributed in a high-dimensional feature space to form a hyperconvex polyhedron. This hyperconvex polyhedron is a region in each dimension. Therefore, by checking whether a certain component of each multidimensional bonding eigenvector falls within the defined feature domain region of that dimension, it can be determined whether the bonding eigenvector is within the hyperconvex polyhedron, that is, within the bonding feature region.

[0061] In a specific embodiment, the bonding feature region is determined by calculating and determining the boundary values ​​of the bonding feature region in each dimension. The method for determining the boundary of the bonding feature region, that is, the method for calculating the feature interval in each dimension, includes: S210-S220.

[0062] S210. Map the multidimensional bonding feature vector to a high-dimensional feature space, and calculate the distribution function F(X) with the multidimensional bonding feature vector as the feature point in the high-dimensional feature space, where F represents the feature space, and X is the multidimensional bonding feature vector of the feature point; X = {xi|i = 1...k}, k is the dimension of the feature vector, and xi is the i-th dimension component of the feature vector.

[0063] S220. Take the second derivative of the distribution function F(X) along each dimension to obtain the second derivative of F(X). Determine the maxima and minima of each dimension as critical values ​​based on the second derivative of F(X); F″(Xi) ≤ L imax ;F″(Xi)>=L imin Where Li is the critical parameter of the i-th dimension component.

[0064] It should be noted that F(X) is a function of a vector. Taking the second derivative of the distribution function F(X) is equivalent to solving for the directional (a certain dimension) derivative of the multidimensional bonding feature vector, which is the process of determining the range of regional values ​​for each dimension.

[0065] S140. The bonding feature vector to be determined, formed based on the thermocouple data to be determined, is compared with the bonding feature region. If the bonding feature vector to be determined is not within the bonding feature region, a crystallizer bonding warning is issued.

[0066] Within each dimension, the value of the feature vector to be determined for adhesion is compared with the interval value of the feature region within that dimension. If the component of the feature vector to be determined for adhesion in that dimension is within the interval value of the feature region in that dimension, it is determined to be a feature point in a normal state. Otherwise, if the component of the feature vector to be determined for adhesion in that dimension is not within the interval value of the feature region in that dimension, it is determined to be a feature point in an abnormal state, and an adhesion warning is issued.

[0067] In summary, finding the extreme point by taking the second derivative does not require adjusting system parameters. It is based on the idea of ​​dynamic boundary adjustment of a superconvex polyhedron to dynamically adjust the feature region, thereby making the crystallizer adhesion early warning method based on the feature region of this invention adaptable to the complexity and variability of the production process and operating conditions, as well as the scalability of production specifications.

[0068] Figure 2 A schematic diagram illustrating the principle of a crystallizer adhesion early warning method based on a feature region according to an embodiment of the present invention is shown, such as... Figure 2As shown in step S110, the features of a single pair are first abstracted. These features include an ascending / descending pattern, an ascending fitting slope (i.e., an ascending slope), a temperature rise index, the time point of the highest temperature, and the maximum descending slope. Feature extraction is first performed on the single pair; if any of these features are not present, the extraction process ends. Then, the features of the upper and lower single pairs are combined to form combined feature data (i.e., the kinematic viscosity coefficient). The single-pair features and combined features are integrated to form a static description structure of the pattern (i.e., a multi-dimensional bonding feature vector). High-dimensional feature point distribution calculations are performed on the static description structure of the pattern to obtain the self-adjusting boundary of the bonding feature region. By determining whether the current feature point is within the bonding feature region, the classification process of whether or not bonding has occurred is completed (i.e., determining whether the current feature point is a bonding warning feature point).

[0069] In summary, this invention is based on the set of key influencing factors of crystallizer adhesion to form adhesion characteristic regions; and uses the intervals of the adhesion characteristic regions to make adhesion early warning judgments for characteristic points.

[0070] Figure 3 The diagram illustrates a state diagram of multidimensional bonding feature vectors as feature points distributed in a high-dimensional feature space according to an embodiment of the present invention, such as... Figure 3 As shown, each eigenvector is considered as a high-dimensional point, and these high-dimensional points are distributed in the high-dimensional feature space to form a hyperconvex polyhedron. For the bonding phenomenon, there is an independent distribution space in the high-dimensional feature space, called the bonding feature domain, which is represented as a closed hyperbody in the high-dimensional feature space (i.e., the aforementioned hyperconvex polyhedron). This closed hyperbody belongs to the eigenspace of bonding.

[0071] In the diagram, black dots represent the normal distribution of feature points, while black dots within the boxes represent the distribution of bonded feature points. It can be seen that after bonding occurs, feature points move from the normally densely distributed area to the boundary bonding feature domain, forming outliers. Bonding determination aims to identify these outliers and accurately determine whether the current location of an outlier is within the bonding feature space, i.e., the bonding feature domain.

[0072] This invention provides a crystallizer adhesion early warning method based on characteristic regions. It finds the extreme points in each dimension by calculating the second derivative of the distribution function F(X) with multidimensional adhesion feature vectors as feature points, thereby determining the regional space in each dimension. Within each dimension, the value of the adhesion feature vector to be determined is compared with the interval value of the feature region within that dimension. If the component of the adhesion feature vector in that dimension falls within the interval value of the feature region in that dimension, it is determined to be a feature point in a normal state. Otherwise, it is determined to be a feature point in an abnormal state, and adhesion early warning is issued.

[0073] The multi-dimensional adhesion feature vector of the crystallizer adhesion early warning method based on feature regions in this invention adopts a description based on the intrinsic properties of adhesion, which is more in line with the laws of adhesion occurrence and development, and has a high accuracy rate in early warning. Furthermore, the feature region description method requires no parameter adjustment steps in application, has adaptive system parameters, and can adapt to changes in equipment, processes, and raw materials.

[0074] In addition, the present invention also protects a crystallizer adhesion early warning system based on feature regions. Figure 4 The logical structure of a crystallizer adhesion early warning system based on a feature region according to an embodiment of the present invention is shown, such as... Figure 4 As shown,

[0075] The crystallizer adhesion early warning system based on feature regions includes: a data acquisition module 41, a feature vector acquisition module 42, a feature region formation module 43, and an adhesion early warning module 44.

[0076] The data acquisition module 41 is used to acquire data for a single thermocouple.

[0077] The feature vector acquisition module 42 is used to perform linear fitting on the collected data to determine the basic feature data of the upper and lower thermocouples; then, based on the basic feature data of the upper and lower thermocouples, the kinematic viscosity coefficients of the upper and lower thermocouples are determined by formula, and a multidimensional bonding feature vector is formed based on the kinematic viscosity coefficients and the basic feature data of the upper and lower thermocouples.

[0078] The feature region forming module 43 is used to obtain the bonding feature region by mapping the multidimensional bonding feature vector to a high-dimensional feature space and using the multidimensional bonding feature vector as feature points.

[0079] The adhesion warning module 44 is used to compare the adhesion feature vector to be determined formed based on the thermocouple data to be determined with the adhesion feature region. If the adhesion feature vector to be determined is not in the adhesion feature region, a crystallizer adhesion warning is issued.

[0080] The feature region forming module 43 includes: a distribution function determination unit, used to map a multidimensional bonding feature vector to a high-dimensional feature space, and to calculate a distribution function F(X) with the multidimensional bonding feature vector as feature points in the high-dimensional feature space, where F represents the feature space and X is the multidimensional bonding feature vector of the feature points; the X = {x i |i=1...k}, k is the dimension of the feature vector, x i Let F(X) be the i-th component of the eigenvector; the boundary determination unit is used to perform second-order differentiation on the distribution function F(X) to obtain the second derivative of F(X), and determine the maxima and minima of each dimension as critical values ​​based on the second derivative of F(X); F″(X)i ) <= L imax ;F″(X i )>=L imin , where Li is the critical parameter of the i-th dimension component.

[0081] The formula for obtaining the kinematic viscosity coefficients of the upper and lower thermocouples is as follows:

[0082]

[0083]

[0084] Wherein, the y-coordinate of the upper thermocouple is Yu, the y-coordinate of the lower thermocouple is Yd, the current billet pulling speed is v (m / min), the time of the highest temperature point of the upper thermocouple is UMaxPoint, the time of the highest temperature point of the lower thermocouple is DMaxPoint, and HotPoint V is the hot spot movement speed.

[0085] The specific implementation of the above-mentioned module of the present invention is largely the same as the specific implementation of the crystallizer adhesion early warning method based on the feature region, and will not be repeated here.

[0086] The crystallizer adhesion early warning system based on feature regions of this invention solves the problem of poor generalization ability of existing models; the crystallizer adhesion early warning system based on feature regions of this invention can be put into normal monitoring immediately after going online, without the need for a system parameter adjustment period of dozens of days; the crystallizer adhesion early warning system based on feature regions of this invention does not require offline data, nor does it require pre-definition of feature region boundaries or classification parameter training, thus avoiding the offline data analysis required by traditional early warning systems, thereby shortening the engineering application cycle of the early warning system.

[0087] This invention provides a crystallizer adhesion early warning method based on characteristic regions, applied to an electronic device 5. (Refer to...) Figure 5 The diagram shown is an application environment schematic of the crystallizer adhesion early warning method based on feature regions according to an embodiment of the present invention.

[0088] In this embodiment, the electronic device 5 can be a terminal device with computing capabilities, such as a server, smartphone, tablet computer, portable computer, or desktop computer.

[0089] The electronic device 5 includes: a processor 52, a memory 51, a communication bus 53, and a network interface 54.

[0090] The memory 51 includes at least one type of readable storage medium. The at least one type of readable storage medium may be a non-volatile storage medium such as flash memory, hard disk, multimedia card, card-type memory 51, etc. In some embodiments, the readable storage medium may be an internal storage unit of the electronic device 5, such as the hard disk of the electronic device 5. In other embodiments, the readable storage medium may also be an external memory 51 of the electronic device 5, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 5.

[0091] In this embodiment, the readable storage medium of the memory 51 is typically used to store a crystallizer adhesion warning program 50 based on a feature region installed on the electronic device 5. The memory 51 can also be used to temporarily store data that has been output or will be output.

[0092] In some embodiments, processor 52 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 51 or process data, such as executing a crystallizer adhesion warning program 50 based on feature regions.

[0093] The communication bus 53 is used to enable communication between these components.

[0094] The network interface 54 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface), which is typically used to establish communication connections between the electronic device 5 and other electronic devices.

[0095] Figure 5 Only electronic device 5 with components 51-53 is shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0096] In one specific embodiment of the present invention, the electronic device 5 may further include a user interface, which may include an input unit such as a keyboard, a voice input device such as a microphone or other device with voice recognition function, a voice output device such as a speaker or headphones, etc. Optionally, the user interface may also include a standard wired interface or a wireless interface.

[0097] Furthermore, the electronic device 5 may also include a display, which may also be referred to as a screen or display unit. In some embodiments, it may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an organic light-emitting diode (OLED) touchscreen, etc. The display is used to display information processed in the electronic device 5 and to display a visual user interface.

[0098] In addition, the electronic device 5 also includes a touch sensor. The area provided by the touch sensor for user touch operation is called the touch area. Furthermore, the touch sensor described here can be a resistive touch sensor, a capacitive touch sensor, etc. Moreover, the touch sensor includes not only contact-type touch sensors but also proximity-type touch sensors, etc. Furthermore, the touch sensor can be a single sensor or, for example, multiple sensors arranged in an array.

[0099] Furthermore, the area of ​​the display of the electronic device 5 can be the same as or different from the area of ​​the touch sensor. Optionally, the display and the touch sensor can be stacked to form a touch display screen. The device detects touch operations triggered by the user based on the touch display screen.

[0100] Optionally, the electronic device 5 may also include radio frequency (RF) circuits, sensors, audio circuits, etc., which will not be described in detail here.

[0101] exist Figure 5 In the illustrated device embodiment, the memory 51, which serves as a computer storage medium, may include an operating system and a crystallizer adhesion warning program 50 based on feature regions; when the processor 52 executes the crystallizer adhesion warning program 50 based on feature regions stored in the memory 51, it performs the following steps:

[0102] S110. Data is collected for a single thermocouple, and the collected data is linearly fitted to determine the basic characteristic data of the upper and lower thermocouples; S120. Based on the basic characteristic data of the upper and lower thermocouples, the kinematic viscosity coefficients of the upper and lower thermocouples are determined by formula, and a multidimensional bonding feature vector is formed based on the kinematic viscosity coefficients and the basic characteristic data of the upper and lower thermocouples; S130. By mapping the multidimensional bonding feature vector to a high-dimensional feature space, the bonding feature region is obtained using the multidimensional bonding feature vector as feature points; S140. The bonding feature vector to be determined formed based on the data of the thermocouple to be determined is compared with the bonding feature region. If the bonding feature vector to be determined is not within the bonding feature region, a crystallizer bonding warning is issued.

[0103] In other embodiments, the crystallizer adhesion warning program 50 based on the feature region can also be divided into one or more modules, which are stored in memory 51 and executed by processor 52 to complete the present invention. The module referred to in the present invention is a series of computer program instruction segments capable of performing a specific function.

[0104] The crystallizer adhesion early warning program 50 based on feature regions can be divided into: a data acquisition module, a feature vector acquisition module, a feature region formation module, and an adhesion early warning module. The specific implementation of the above modules of the present invention is roughly the same as the specific implementation of the crystallizer adhesion early warning method based on feature regions, and will not be repeated here.

[0105] The specific implementation of the computer-readable storage medium of the present invention is largely the same as the specific implementation of the crystallizer adhesion early warning method and electronic device based on the above-mentioned feature region, and will not be described again here.

[0106] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0107] The sequence numbers of the above embodiments of the present invention are merely for description and do not represent the superiority or inferiority of the embodiments. Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases, the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0108] As per the above reference Figures 1 to 5The crystallizer adhesion early warning method, system, apparatus, and computer-readable storage medium based on feature regions according to the present invention are described by way of example. The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A crystallizer adhesion early warning method based on characteristic regions, applied to electronic devices, characterized in that, The method includes the following steps: S110. Data is collected for a single thermocouple, and linear fitting is performed on the collected data to determine the basic characteristic data of the upper and lower thermocouples. S120. Based on the basic characteristic data of the upper and lower thermocouples, determine the kinematic viscosity coefficients of the upper and lower thermocouples using a formula, and form a multidimensional adhesion feature vector based on the kinematic viscosity coefficients and the basic characteristic data of the upper and lower thermocouples. S130. By mapping the multidimensional bonding feature vector to a high-dimensional feature space, the bonding feature region is obtained using the multidimensional bonding feature vector as feature points; wherein, the bonding feature region is obtained by determining the boundary values ​​of the bonding feature region, and the method for determining the boundary of the bonding feature region includes: S210. Map the multidimensional bonding feature vector to a high-dimensional feature space, and calculate the distribution function F(X) with the multidimensional bonding feature vector as feature points in the high-dimensional feature space, where F represents the feature space and X is the multidimensional bonding feature vector of the feature points; X = {x i |i=1...k}, k is the dimension of the feature vector, x i Let i be the i-th component of the eigenvector; S220. Take the second derivative of the distribution function F(X) to obtain the second derivative of F(X), and determine the maximum and minimum values ​​of each dimension as critical values ​​based on the second derivative of F(X). F”(Xi)<=L imax ;F”(Xi)>=L imin Where Li is the critical parameter of the i-th dimension component; S140. The bonding feature vector to be determined, formed based on the thermocouple data to be determined, is compared with the bonding feature region. If the bonding feature vector to be determined is not within the bonding feature region, a crystallizer bonding warning is issued.

2. The crystallizer adhesion early warning method based on characteristic regions according to claim 1, characterized in that, The bonding characteristic data of the upper thermocouple includes at least: the upper thermocouple rising and falling mode Umode, the upper thermocouple rising slope Uslap, the upper thermocouple temperature rise index UMaxTCoeff, and the maximum falling slope USlapNeg in the upper thermocouple rising and falling modes. The adhesion characteristic data of the lower thermocouple includes at least: the lower thermocouple rising and falling modes Dmode, the lower thermocouple rising slope Dslap, the lower thermocouple temperature rise index DMaxTCoeff, and the maximum falling slope DSlapNeg in the lower thermocouple rising and falling modes.

3. The crystallizer adhesion early warning method based on characteristic regions according to claim 1, characterized in that, In step S120, the formula for obtaining the kinematic viscosity coefficients of the upper and lower thermocouples is as follows: Wherein, the y-coordinate of the upper thermocouple is Yu, the y-coordinate of the lower thermocouple is Yd, the current billet pulling speed is v (m / min), the time of the highest temperature point of the upper thermocouple is UMaxPoint, the time of the highest temperature point of the lower thermocouple is DMaxPoint, and HotPoint V is the hot spot movement speed.

4. The crystallizer adhesion early warning method based on characteristic regions according to claim 2, characterized in that, The temperature rise index is obtained using the following formula: MaxTCoeff=Tmax / average(T,M)-1; Where M is the number of data points, average(T,M) is the average temperature of the M data points, and Tmax is the highest temperature among the M data points.

5. A crystallizer adhesion early warning system based on characteristic regions, characterized in that, include: The data acquisition module is used to acquire data from a single thermocouple. The feature vector acquisition module is used to perform linear fitting on the collected data to determine the basic feature data of the upper and lower thermocouples; then, based on the basic feature data of the upper and lower thermocouples, the kinematic viscosity coefficients of the upper and lower thermocouples are determined by formula, and a multidimensional bonding feature vector is formed based on the kinematic viscosity coefficients and the basic feature data of the upper and lower thermocouples. A feature region forming module is used to obtain bonding feature regions by mapping the multidimensional bonding feature vector to a high-dimensional feature space, using the multidimensional bonding feature vector as feature points; wherein, the feature region forming module includes: a distribution function determining unit, used to map the multidimensional bonding feature vector to a high-dimensional feature space, and calculate the distribution function F(X) in the high-dimensional feature space using the multidimensional bonding feature vector as feature points, where F represents the feature space, and X is the multidimensional bonding feature vector of the feature points; X = {x i |i=1...k}, k is the dimension of the feature vector, x i Let i be the i-th component of the eigenvector; A boundary determination unit is used to perform second-order differentiation on the distribution function F(X) to obtain the second derivative of F(X), and to determine the maxima and minima of each dimension as critical values ​​based on the second derivative of F(X); F(X) i ) <= L imax ;F"(X i )>=L imin , where Li is the critical parameter of the i-th dimension component; The adhesion warning module is used to compare the adhesion feature vector to be determined formed based on the thermocouple data to be determined with the adhesion feature region. If the adhesion feature vector to be determined is not within the adhesion feature region, a crystallizer adhesion warning is issued.

6. The crystallizer adhesion early warning system based on feature regions according to claim 5, characterized in that, The formula for obtaining the kinematic viscosity coefficient of the upper and lower thermocouples is: Wherein, the y-coordinate of the upper thermocouple is Yu, the y-coordinate of the lower thermocouple is Yd, the current billet pulling speed is v (m / min), the time of the highest temperature point of the upper thermocouple is UMaxPoint, the time of the highest temperature point of the lower thermocouple is DMaxPoint, and HotPoint V is the hot spot movement speed.

7. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a crystallizer adhesion warning program based on a feature region. When executed by the processor, the feature region-based crystallizer adhesion warning program performs the following steps: S110. Data is collected for a single thermocouple, and linear fitting is performed on the collected data to determine the basic characteristic data of the upper and lower thermocouples. S120. Based on the basic characteristic data of the upper and lower thermocouples, determine the kinematic viscosity coefficients of the upper and lower thermocouples using a formula, and form a multidimensional adhesion feature vector based on the kinematic viscosity coefficients and the basic characteristic data of the upper and lower thermocouples. S130. By mapping the multidimensional bonding feature vector to a high-dimensional feature space, the bonding feature region is obtained using the multidimensional bonding feature vector as feature points; wherein, the bonding feature region is obtained by determining the boundary values ​​of the bonding feature region, and the method for determining the boundary of the bonding feature region includes: S210. Map the multidimensional bonding feature vector to a high-dimensional feature space, and calculate the distribution function F(X) with the multidimensional bonding feature vector as feature points in the high-dimensional feature space, where F represents the feature space and X is the multidimensional bonding feature vector of the feature points; X = {x i |i=1...k}, k is the dimension of the feature vector, x i Let i be the i-th component of the eigenvector; S220. Take the second derivative of the distribution function F(X) to obtain the second derivative of F(X), and determine the maximum and minimum values ​​of each dimension as critical values ​​based on the second derivative of F(X). F”(Xi)<=L imax ;F”(Xi)>=L imin Where Li is the critical parameter of the i-th dimension component; S140. The bonding feature vector to be determined, formed based on the thermocouple data to be determined, is compared with the bonding feature region. If the bonding feature vector to be determined is not within the bonding feature region, a crystallizer bonding warning is issued.

Images