A crystallizer liquid level control method, device, electronic device and storage medium

By obtaining the influence data of the crystallizer liquid level and establishing a fluctuation factor model, adjusting the influencing factors to finely control the liquid level, the problem of inaccurate liquid level control in the existing technology is solved, and the quality and safety of the casting billet are improved.

CN115430815BActive Publication Date: 2024-12-03HENGYANG RAMON SCI & TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210951137.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-08-09
Publication Date
2024-12-03
Estimated Expiration
2042-08-09

AI Technical Summary

Technical Problem

The existing crystallizer liquid level control system only relies on liquid level height detection, resulting in inaccurate control, affecting the quality of the cast billet and may lead to accidents such as bonding and steel leakage.

Method used

By obtaining the influence data of the crystallizer liquid level, establish a model for determining the liquid level fluctuation factor, adjust the influencing factors to finely control the liquid level, and avoid relying solely on liquid level height detection.

Benefits of technology

It improves the accuracy of liquid level control, improves the quality and safety of cast billets, and avoids major accidents such as bonding and steel leakage.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115430815B_ABST
    Figure CN115430815B_ABST
Patent Text Reader

Abstract

The present application relates to a method, device, electronic device and storage medium for controlling the liquid level of a crystallizer, wherein the method comprises: obtaining influencing data of the liquid level of the crystallizer, wherein the influencing data is determined according to influencing factors of the liquid level of the crystallizer; inputting the influencing data into a crystallizer liquid level fluctuation factor determination model to determine the fluctuation state of the liquid level of the crystallizer; adjusting the influencing factors of the crystallizer according to the fluctuation state to control the liquid level of the crystallizer, wherein the influencing data is obtained through the determined influencing factors, and the fluctuation state of the liquid level of the crystallizer is determined based on the influencing data, thereby avoiding the case where the fluctuation state of the liquid level of the crystallizer is determined by only detecting the height of the molten steel level; at the same time, the influencing factors are adjusted according to the fluctuation state determined by the influencing data, thereby avoiding the problem of inaccurate control caused by only detecting the height of the molten steel level to control the liquid level of the crystallizer, thereby improving the control accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of crystallizer technology, and in particular to a crystallizer liquid level control method, device, electronic equipment and storage medium. Background Art

[0002] In the steel continuous casting production process, the liquid level control of the continuous casting machine crystallizer is an important part of production automation, and its control accuracy directly affects the quality of the ingot. If the liquid level fluctuates by more than ±5mm, it will have a certain impact on the quality of the ingot, and even cause the unmelted protective slag to be drawn into the molten steel, forming surface defects of the ingot; if the liquid level fluctuates by more than ±10mm, it may cause the automatic casting mode to be interrupted, affecting the normal production rhythm, and in severe cases, even major accidents such as bonding and steel leakage may occur.

[0003] The existing crystallizer liquid level control system mainly relies on the liquid level detection device to detect the molten steel level height for control, but this method only makes judgments based on the detected liquid level height. In fact, the molten steel level height is affected by many factors. For example, the molten steel flow rate, angle, and height at the crystallizer outlet will affect the liquid level fluctuation. Only detecting the molten steel level height to control the crystallizer liquid level will lead to inaccurate control. Therefore, how to more finely control the crystallizer liquid level has become an urgent problem to be solved.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0005] The present application provides a crystallizer liquid level control method, device, electronic device and storage medium to solve the problem in the related art that the crystallizer liquid level is controlled by only detecting the height of the molten steel level, resulting in inaccurate control.

[0006] In a first aspect, the present application provides a crystallizer liquid level control method, the crystallizer liquid level control method comprising: obtaining influencing data of the crystallizer liquid level, the influencing data being determined based on influencing factors of the crystallizer liquid level; inputting the influencing data into a crystallizer liquid level fluctuation factor determination model to determine the fluctuation state of the crystallizer liquid level; and adjusting the influencing factors of the crystallizer based on the fluctuation state to control the liquid level of the crystallizer.

[0007] In some examples, before inputting the impact data into the crystallizer liquid level fluctuation factor determination model, the method also includes: acquiring historical impact data, and determining normal samples and fluctuation samples based on the historical impact data; determining a training set and a test set based on the normal samples and the fluctuation samples; and modeling the training set and the test set through a clustering algorithm to obtain the crystallizer liquid level fluctuation factor determination model.

[0008] In some examples, the historical impact data includes: actual liquid level and set liquid level; wherein, determining normal samples and fluctuation samples based on the historical impact data includes: determining the difference between the actual liquid level and the set liquid level at each moment in the historical impact data; when the difference between the actual liquid level and the set liquid level at any moment in the historical impact data exceeds a first threshold, using the data at any moment as the fluctuation sample; when the difference between the actual liquid level and the set liquid level at any moment in the historical impact data is lower than a second threshold, using the data at any moment as the normal sample, and the second threshold is lower than the first threshold.

[0009] In some examples, determining a training set and a test set based on the normal samples and the fluctuation samples includes: mixing the normal samples and the fluctuation samples to obtain a total training set; and dividing the total training set based on a preset ratio to obtain the training set and the test set.

[0010] In some examples, the training set and the test set are modeled by a clustering algorithm to obtain the crystallizer liquid level fluctuation factor determination model, including: extracting the time domain characteristics and frequency domain characteristics of the changing variables in the training set and the test set respectively to obtain the training set data dimension and the test set data dimension; performing dimensionality reduction processing on the training set data dimension and the test set data dimension to obtain training set reduced dimension data and test set reduced dimension data; modeling the training set reduced dimension data and the test set reduced dimension data by a clustering algorithm to obtain the crystallizer liquid level fluctuation factor determination model.

[0011] In some examples, the training set dimensionality reduction data and the test set dimensionality reduction data are modeled by a clustering algorithm to obtain the crystallizer liquid level fluctuation factor determination model, including: modeling the training set dimensionality reduction data and the test set dimensionality reduction data by at least two clustering algorithms to obtain at least two initial crystallizer liquid level fluctuation factor determination models; evaluating at least two of the initial crystallizer liquid level fluctuation factor determination models based on the normal samples to obtain the optimal initial crystallizer liquid level fluctuation factor determination model; and using the optimal initial crystallizer liquid level fluctuation factor determination model as the crystallizer liquid level fluctuation factor determination model.

[0012] In some examples, after the optimal initial crystallizer liquid level fluctuation factor determination model is used as the crystallizer liquid level fluctuation factor determination model, the method also includes: classifying the clustering of the training set dimensionality reduction data and the test set dimensionality reduction data according to the crystallizer liquid level fluctuation factor determination model to obtain multiple preset fluctuation states; and setting an influencing factor adjustment method corresponding to the preset fluctuation state.

[0013] In the second aspect, the present application provides a crystallizer liquid level control device, which includes: an acquisition module, which is used to acquire influencing data of the crystallizer liquid level, and the influencing data is determined according to the influencing factors of the liquid level of the crystallizer; an input module, which is used to input the influencing data into a crystallizer liquid level fluctuation factor determination model to determine the fluctuation state of the liquid level of the crystallizer; and a control module, which is used to adjust the influencing factors of the crystallizer according to the fluctuation state to control the liquid level of the crystallizer.

[0014] In a third aspect, an electronic device is provided, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0015] Memory, used to store computer programs;

[0016] The processor is used to implement the steps of the crystallizer liquid level control method described in any embodiment of the first aspect when executing the program stored in the memory.

[0017] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the crystallizer liquid level control method as described in any embodiment of the first aspect are implemented.

[0018] The above technical solution provided by the embodiment of the present application has the following advantages compared with the prior art:

[0019] The crystallizer liquid level control method provided in the embodiment of the present application includes: obtaining influencing data of the crystallizer liquid level, wherein the influencing data is determined according to the influencing factors of the crystallizer liquid level; inputting the influencing data into a crystallizer liquid level fluctuation factor determination model to determine the fluctuation state of the crystallizer liquid level; adjusting the influencing factors of the crystallizer according to the fluctuation state to control the liquid level of the crystallizer, wherein the influencing data is obtained through the determined influencing factors, and the fluctuation state of the crystallizer liquid level is determined based on the influencing data, thereby avoiding only detecting the height of the molten steel liquid level to determine the fluctuation state of the crystallizer liquid level. At the same time, the influencing factors are adjusted according to the fluctuation state determined by the influencing data to achieve accurate control of the crystallizer liquid level, thereby avoiding the problem of inaccurate control caused by only detecting the height of the molten steel liquid level to control the crystallizer liquid level, thereby improving the control accuracy, thereby improving the quality of the ingot, and avoiding major accidents such as sticking and steel leakage, thereby improving safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0022] Figure 1 A basic flow chart of an optional crystallizer liquid level control method provided in an embodiment of the present application;

[0023] Figure 2 A basic flow chart of an optional crystallizer liquid level control device provided in an embodiment of the present application;

[0024] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0026] In order to solve the problem of inaccurate control of the mold liquid level by only detecting the molten steel level in the related art, please refer to Figure 1 , Figure 1 A crystallizer liquid level control method provided in this embodiment, such as Figure 1 As shown, the crystallizer liquid level control method comprises:

[0027] S101, obtaining influencing data of the liquid level of the crystallizer, wherein the influencing data is determined according to influencing factors of the liquid level of the crystallizer;

[0028] S102, inputting the influencing data into a crystallizer liquid level fluctuation factor determination model to determine the fluctuation state of the crystallizer liquid level;

[0029] S103, adjusting the influencing factors of the crystallizer according to the fluctuation state to control the liquid level of the crystallizer.

[0030] The main factors affecting the crystallizer liquid level fluctuation can be divided into two directions: above the crystallizer (including the crystallizer) and below the crystallizer for analysis, with the crystallizer as the dividing point. The direction above the crystallizer (including the crystallizer) can be considered from three aspects: the molten steel flow rate, angle, and height of the crystallizer outlet;

[0031] Among them, taking the flow rate at the steel outlet as an example, the main factors affecting the flow rate at the steel outlet include but are not limited to: (1) the quality of molten steel affects the fluidity of molten steel; (2) the opening of the stopper rod affects the outflow speed of molten steel, thereby affecting the pulling speed of the straightening machine; 3) the change of the weight of the tundish affects the change of molten steel flow rate; (4) the nodules and corrosion of the stopper rod and the nozzle also have a certain impact on the molten steel flow rate at the steel outlet; (5) the gap and blockage of the stopper rod mechanism also affect the liquid level control, thereby affecting the flow rate at the steel outlet.

[0032] Taking the angle of the steel outlet as an example, when the aluminum content in the steel is high, it is easy to react with oxygen to form aluminum oxide. Aluminum oxide is a high-melting-point inclusion. As the casting time increases, it will slowly accumulate on the stopper rod head, water nozzle, bowl and inner wall, affecting the angle of the steel outlet. As the casting time increases, the stopper rod and water nozzle will corrode, which will also affect the angle of the steel outlet and cause liquid level fluctuations in the crystallizer.

[0033] Taking the height of the steel tapping port as an example, the zero-point drift of the crystallizer liquid level detection system and the thickness of the crystallizer protective slag layer further cause the crystallizer liquid level fluctuation by affecting the height of the steel tapping port. It is understandable that the bulging of the ingot is also likely to cause the liquid level fluctuation of the crystallizer. Before the ingot is completely solidified, the deviation of the roller surface of one or some sectors is large, resulting in uneven force on the liquid core in the ingot and causing liquid level fluctuation. In addition, the unstable pulling speed of the straightening machine is also likely to cause liquid level fluctuation;

[0034] Therefore, combined with the above examples, it can be seen that the stopper rod, tundish, straightening machine, bulging of the ingot, etc. are all influencing factors that will affect the liquid level of the crystallizer. The influencing data affecting the crystallizer determined based on the above influencing factors include but are not limited to: actual liquid level, set liquid level, actual pulling speed, given pulling speed, actual position of the stopper rod, given position of the stopper rod, tundish weight, section number, bulging compensation amount, actual liquid level (electromagnetic) electromagnetic instrument N value, electromagnetic instrument NP value, sensor temperature, steel grade number, and pouring time.

[0035] Continuing with the above example, obtaining the influencing data of the crystallizer liquid level includes: obtaining the influencing data of the crystallizer liquid level in real time, wherein the influencing factors are monitored in real time, thereby obtaining the influencing data of the crystallizer liquid level in real time. In some examples of this embodiment, obtaining the influencing data of the crystallizer liquid level includes: periodically obtaining the influencing data of the crystallizer liquid level, wherein the influencing factors are monitored periodically, thereby obtaining the influencing data of the crystallizer liquid level in real time.

[0036] In some examples of this embodiment, before the impact data is input into the crystallizer liquid level fluctuation factor determination model, the method further includes: obtaining historical impact data, determining normal samples and fluctuation samples according to the historical impact data; determining a training set and a test set based on the normal samples and the fluctuation samples; modeling the training set and the test set through a clustering algorithm to obtain the crystallizer liquid level fluctuation factor determination model. It should be understood that the historical impact data is data within one or more time periods, which includes the impact data within one or more time periods. Multiple normal samples and fluctuation samples can be determined through the historical impact data, and then the training set and the test set can be determined based on the normal samples and the fluctuation samples. Finally, modeling is performed based on the training set and the test set to obtain the crystallizer liquid level fluctuation factor determination model.

[0037] Continuing with the above example, the historical impact data includes: the actual liquid level and the set liquid level; wherein, determining the normal sample and the fluctuation sample based on the historical impact data includes: determining the difference between the actual liquid level and the set liquid level at each moment in the historical impact data; when the difference between the actual liquid level and the set liquid level at any moment in the historical impact data exceeds the first threshold, the data at any moment is used as the fluctuation sample; when the difference between the actual liquid level and the set liquid level at any moment in the historical impact data is lower than the second threshold, the data at any moment is used as the normal sample, and the second threshold is lower than the first threshold. wherein, the actual liquid level is the actual value of the current liquid level of the crystallizer, the set liquid level is a liquid level value set for the crystallizer, and the difference between the actual liquid level and the set liquid level can be obtained by calculating the difference between the actual liquid level and the set liquid level; the first threshold and the second threshold are a value set by relevant personnel according to actual needs, and the second threshold is lower than the first and the first thresholds, for example, the first threshold is 5mm, and the second threshold is 3mm. Among them, when the absolute value of the deviation between the actual liquid level and the set liquid level at the current moment of historical impact data is greater than 5mm, the data at the current moment is used as a fluctuation sample. If the absolute value of the deviation between the actual liquid level and the set liquid level is not higher than 3mm, the data at the current moment is used as a normal sample.

[0038] Continuing with the above example, in some examples, when the difference between the actual liquid level and the set liquid level at any moment in the historical impact data exceeds the first threshold, the data at any moment is used as the fluctuation sample, including: when the difference between the actual liquid level and the set liquid level at any moment in the historical impact data exceeds the first threshold, the time at any moment is used as the time point to obtain a sample interval, and the data of the sample interval is used as the fluctuation sample, wherein the sample interval is determined according to the set first preset time and second preset time length; specifically, taking the first preset time length of five minutes and the second preset time length of one minute as an example, when it is determined that the difference between the actual liquid level at the current moment and the set liquid level exceeds the first threshold, the interval between five minutes before the current moment and one minute after the current moment is used as the sample interval, and the data of the sample interval is obtained as the fluctuation sample; similarly, when the actual liquid level at any moment in the historical impact data is When the difference between the actual liquid level and the set liquid level is lower than the second threshold, the data at any moment is used as the normal sample, including: when the difference between the actual liquid level and the set liquid level at any moment in the historical impact data is lower than the second threshold, the time at any moment is used as the time point to obtain the sample interval, and the data of the sample interval is used as the fluctuation sample, wherein the sample interval is determined according to the set third preset time and fourth preset duration; specifically, taking the third preset duration of 4 minutes and the fourth preset duration of 4 minutes as an example, when it is determined that the difference between the actual liquid level at the current moment and the set liquid level is lower than the second threshold, the interval between five minutes before the current moment and one minute after the current moment is used as the sample interval, and the data of the sample interval is obtained as the normal sample; it should be understood that in some examples, the first preset duration to the fourth preset duration are different; in some examples, the first preset duration to the fourth preset duration are the same. And so on, all historical impact data are divided.

[0039] Following the above example, it can be understood that when the difference between the actual liquid level and the set liquid level at any moment in the historical impact data exceeds the first threshold, the time of any moment is taken as the time point to obtain the sample interval, and the data of the sample interval is taken as the fluctuation sample. If in addition to the any moment a, there is another moment b between the first preset time and the any moment in the sample interval where the difference between the actual liquid level and the set liquid level exceeds the first threshold, then the two moments are merged into the same fluctuation sample, and the first preset time before the other moment b to the second preset time after any moment a is taken as the sample interval, and the data of the sample interval is taken as the fluctuation sample; for example, the first preset time is five minutes, and the second preset time is one minute, if the absolute value of the liquid level deviation at the current moment a is greater than 5mm, and there is a moment b with an absolute value of the liquid level deviation greater than 5mm within 5 minutes before the current moment a, then the moment a and the moment b are merged into the same sample, and the 5 minutes before the moment b and the 1 minute after the moment a are taken as the sample interval, and the data of the sample interval is taken as the sample data. By analogy, all historical impact data are divided.

[0040] In some examples of this embodiment, determining a training set and a test set based on the normal sample and the fluctuation sample includes: mixing the normal sample and the fluctuation sample to obtain a total training set; dividing the total training set based on a preset ratio to obtain the training set and the test set. The preset ratio is the ratio of the training set to the test set, and the preset ratio is a ratio set by relevant personnel according to actual needs. For example, the preset ratio can be 3:1, that is, training set: test set = 3:1. The fluctuation sample is mixed with the normal sample to obtain a total training set, and then the total training set is divided into a training set and a test set according to the preset ratio.

[0041] In some examples of this embodiment, the training set and the test set are modeled by a clustering algorithm to obtain the crystallizer liquid level fluctuation factor determination model, including: extracting the time domain characteristics and frequency domain characteristics of the variable in the training set and the test set respectively to obtain the training set data dimension and the test set data dimension; performing dimensionality reduction processing on the training set data dimension and the test set data dimension to obtain the training set reduced dimension data and the test set reduced dimension data; modeling the training set reduced dimension data and the test set reduced dimension data by a clustering algorithm to obtain the crystallizer liquid level fluctuation factor determination model. Among them, the time domain characteristics include but are not limited to: mean, root mean square value, peak-to-peak value, kurtosis, skewness, peak factor, kurtosis factor, margin factor, waveform factor, pulse factor; frequency domain characteristics include but are not limited to: maximum amplitude frequency, power spectrum sum, power spectrum mean, power spectrum variance, power spectrum skewness, power spectrum kurtosis, power spectrum relative peak value. It should be understood that in some examples, the time domain features and frequency domain features of the changing variables in the training set and the test set are extracted respectively to obtain feature samples, and missing values ​​are deleted from the obtained feature samples to obtain the training set data dimensions and the test set data dimensions.

[0042] It should be understood that the training set data dimension and the test set data dimension have more feature samples and higher dimensions, so the training set data dimension and the test set data dimension are subjected to dimensionality reduction processing. The dimensionality reduction processing method is not limited, for example, the dimensionality reduction processing can be performed by a kernel function, and the above-mentioned kernel function includes but is not limited to: one of the linear kernel function, the poly kernel function, the RBF kernel function, the sigmoid kernel function and the cosine kernel function, and preferably, the kernel function adopts the RBF kernel function.

[0043] Continuing with the above example, the training set data dimension and the test set data dimension are subjected to dimensionality reduction processing to obtain training set dimensionality reduction data and test set dimensionality reduction data, including but not limited to: the training set data dimension and the test set data dimension are subjected to dimensionality reduction processing by at least two kernel functions to obtain at least two training set dimensionality reduction data and at least two test set dimensionality reduction data. It should be understood that after obtaining at least two training set dimensionality reduction data and at least two test set dimensionality reduction data, various training set dimensionality reduction data are respectively judged for their quality, and the best training set dimensionality reduction data and test set dimensionality reduction data are used as the final determined training set dimensionality reduction data and test set dimensionality reduction data. Among them, how to judge the quality of various training set dimensionality reduction data and test set dimensionality reduction data to select the best kernel function will be described in detail later, and will not be repeated here.

[0044] Among them, considering that the fluctuation samples are unlabeled samples, a clustering algorithm in machine learning is selected to model the samples, aiming to distinguish different types of liquid level fluctuations. The above clustering algorithms include but are not limited to: at least one of Gaussian mixture model, variational Bayesian Gaussian mixture model, K-Means clustering, and hierarchical clustering. Preferably, the clustering algorithm adopts Gaussian mixture model; at the same time, since the specific classification of the dimensionality reduction data of the training set and the dimensionality reduction data of the test set is not known before modeling, the classification can be set by relevant personnel. Preferably, it is set to 6-15 categories, with a total of 10 classifications;

[0045] Continuing from the above example, the training set dimensionality reduction data and the test set dimensionality reduction data are modeled by a clustering algorithm to obtain the crystallizer liquid level fluctuation factor determination model, including: modeling the training set dimensionality reduction data and the test set dimensionality reduction data by at least two clustering algorithms to obtain at least two initial crystallizer liquid level fluctuation factor determination models; evaluating at least two of the initial crystallizer liquid level fluctuation factor determination models based on the normal samples to obtain the optimal initial crystallizer liquid level fluctuation factor determination model; using the optimal initial crystallizer liquid level fluctuation factor determination model as the crystallizer liquid level fluctuation factor determination model. Wherein, when evaluating at least two of the initial crystallizer liquid level fluctuation factor determination models based on the normal samples, the quality of each initial crystallizer liquid level fluctuation factor determination model can be evaluated by the recall rate and precision rate of the normal samples, wherein the precision rate is: the correct proportion of predictions in the predicted category, and the denominator is the number of samples in the predicted category. Recall rate: the correct proportion of predictions in the true category, and the denominator is the number of samples in the true category. For example, the dimensionality reduction data of the training set and the dimensionality reduction data of the test set are modeled by four methods, namely, Gaussian mixture model, variational Bayesian Gaussian mixture model, K-Means clustering, and hierarchical clustering, to obtain four initial crystallizer liquid level fluctuation factor determination models. Then, the four initial crystallizer liquid level fluctuation factor determination models are judged by the recall rate and precision rate of normal samples to determine the advantages and disadvantages of the four initial crystallizer liquid level fluctuation factor determination models, and the optimal initial crystallizer liquid level fluctuation factor determination model is used as the crystallizer liquid level fluctuation factor determination model.

[0046] Continuing with the above example, when there are multiple kernel functions and multiple training set dimensionality reduction data and test set dimensionality reduction data are obtained based on the multiple kernel functions, modeling is performed based on each training set dimensionality reduction data and test set dimensionality reduction data and each clustering algorithm, and then the recall rate and precision rate of normal samples are used to judge the determination models of the four initial crystallizer liquid level fluctuation factors to determine the advantages and disadvantages of the determination models of the four initial crystallizer liquid level fluctuation factors, and the optimal initial crystallizer liquid level fluctuation factor determination model is used as the crystallizer liquid level fluctuation factor determination model, and the kernel function corresponding to the crystallizer liquid level fluctuation factor determination model is used as the optimal kernel function.

[0047] In some embodiments of the present invention, after the optimal initial crystallizer liquid level fluctuation factor determination model is used as the crystallizer liquid level fluctuation factor determination model, the method further includes: classifying the clustering of the training set dimensionality reduction data and the test set dimensionality reduction data according to the crystallizer liquid level fluctuation factor determination model to obtain a plurality of preset fluctuation states; and setting the influencing factor adjustment method corresponding to the preset fluctuation state. Among them, the preset fluctuation state includes: liquid level fluctuation within 5 minutes of pouring, low-frequency liquid level fluctuation, high-frequency liquid level fluctuation, liquid level fluctuation caused by water change port, divergent liquid level fluctuation, sudden rise type liquid level fluctuation, liquid level fluctuation caused by flocculation, and periodic liquid level fluctuation, and locates the specific influencing factors of each liquid level fluctuation category based on the preset fluctuation state, and presets the influencing factor adjustment method corresponding to the fluctuation state, and then when the crystallizer liquid level fluctuation factor determination model determines the fluctuation state of the liquid level of the crystallizer, the corresponding influencing factors can be adjusted according to the influencing factor adjustment method corresponding to the set fluctuation state, so as to achieve precise control of the liquid level of the crystallizer.

[0048] The crystallizer liquid level control method provided in this embodiment includes: obtaining influencing data of the crystallizer liquid level, wherein the influencing data is determined according to the influencing factors of the crystallizer liquid level; inputting the influencing data into a crystallizer liquid level fluctuation factor determination model to determine the fluctuation state of the crystallizer liquid level; adjusting the influencing factors of the crystallizer according to the fluctuation state to control the liquid level of the crystallizer, wherein the influencing data is obtained through the determined influencing factors, and the fluctuation state of the crystallizer liquid level is determined based on the influencing data, thereby avoiding only detecting the molten steel liquid level height to determine the fluctuation state of the crystallizer liquid level. At the same time, the influencing factors are adjusted according to the fluctuation state determined by the influencing data to achieve accurate control of the crystallizer liquid level, thereby avoiding the problem of inaccurate control caused by only detecting the molten steel liquid level height to control the crystallizer liquid level, thereby improving the control accuracy, thereby improving the quality of the ingot, and avoiding major accidents such as bonding and steel leakage, thereby improving safety.

[0049] In order to better understand the present invention, this embodiment provides a more specific example to illustrate the present invention;

[0050] This embodiment provides a crystallizer liquid level control method, which includes the following steps:

[0051] 1) Collect the impact data of the liquid level fluctuation of the crystallizer,

[0052] Among them, the influencing data include: actual liquid level, set liquid level, actual casting speed, given casting speed, actual stopper rod position, given stopper rod position, tundish weight, section number, bulging compensation, actual liquid level (electromagnetic), electromagnetic instrument N value, electromagnetic instrument NP value, sensor temperature, steel grade number, pouring time, a total of 14 variables;

[0053] Specifically, for example, first collect historical impact data for a period of time. Excluding the variables that have remained unchanged and irrelevant signal variables, we get 14 variables, including actual liquid level, set liquid level, actual pulling speed, given pulling speed, actual stopper rod position, given stopper rod position, middle bag weight, section number, bulge compensation, actual liquid level (electromagnetic) electromagnetic instrument N value, electromagnetic instrument NP value, sensor temperature, and steel grade number. Considering that the frequency of liquid level fluctuations tends to increase with the increase of casting time, the pouring time label is manually added, and the above data are collected to obtain historical impact data.

[0054] 2) Data preprocessing;

[0055] Process the historical impact data. When the absolute value of the deviation between the actual liquid level and the set liquid level is greater than 5mm, take the data 5 minutes before the current moment a and 1 minute after the moment a as the fluctuation sample; if there is a moment b with an absolute value of the liquid level deviation greater than 5mm within 5 minutes before the current moment a, merge the moment a and moment b into the same sample, and take the data 5 minutes before the moment b and 1 minute after the moment a as the fluctuation sample;

[0056] 3) In each pouring in the data, the liquid level fluctuation is within ±3mm and maintained for about 8 minutes as a normal sample, and the rest are fluctuation samples; the fluctuation samples are mixed with the normal samples, and the training set and the test set are divided according to the ratio of training set: test set = 3:1, and the time domain characteristics of the continuously changing variables of the training set and the test set are extracted respectively: mean, root mean square value, peak-to-peak value, kurtosis, skewness, peak factor, kurtosis factor, margin factor, waveform factor, pulse factor, and frequency domain characteristics: maximum amplitude frequency, power spectrum, power spectrum mean, power spectrum variance, power spectrum skewness, power spectrum kurtosis, power spectrum relative peak value, a total of 17 characteristic factors, and the missing values ​​of the obtained characteristic samples are deleted to obtain the training set data dimension and the test set data dimension; as shown in Table 1 below:

[0057] Table 1. Training set data dimensions and test set data dimensions

[0058]

[0059] 4) Data dimensionality reduction and cluster analysis,

[0060] According to Table 1 above, the training set data dimension and the test set data dimension are 175, which is relatively high, so dimensionality reduction is considered. Therefore, dimensionality reduction is required;

[0061] Among them, different kernel functions are applied to the training set data dimension and the test set data dimension for principal component dimensionality reduction, and after the data set is classified, a clustering algorithm in machine learning is selected to model the samples, and the recall rate and precision rate of normal samples are used to evaluate the quality of the clustering model, from which the best dimensionality reduction kernel function is selected;

[0062] Among them, principal component analysis is suitable for linear dimensionality reduction of data. Kernel principal component analysis can achieve nonlinear dimensionality reduction of data and is used to process linearly inseparable data sets. The general idea of ​​kernel principal component analysis is: for the matrix X in the input space, first use a nonlinear mapping to map all samples in X to a high-dimensional or even infinite-dimensional space to make it linearly separable, and then perform principal component dimensionality reduction in this high-dimensional space. Therefore, the kernel functions provided in this example include but are not limited to Linear kernel, Poly kernel, RBF kernel, Sigmoid kernel, and Cosine kernel; among them, the number of principal components extracted by different kernel functions is shown in Table 2:

[0063] Table 2. Number of principal components extracted by different kernel functions

[0064]

[0065] It should be understood that clustering algorithms include, but are not limited to: Gaussian mixture model, K-Means model, variational Bayesian Gaussian mixture model, hierarchical clustering model;

[0066] Among them, when the data has multiple classes and it is expected to divide the data into some clusters, it can be assumed that the samples in different clusters each obey different Gaussian distributions. The resulting clustering algorithm is called a Gaussian mixture model. The core idea of ​​the Gaussian mixture model is to assume that the data can be regarded as generated from multiple Gaussian distributions. Under this assumption, each individual sub-model is a standard Gaussian model, and its mean and variance are the parameters to be estimated. In addition, each sub-model has a parameter, which can be understood as a weight or the probability of generating data. The EM algorithm is usually used to solve the parameters of the Gaussian mixture model.

[0067] The variational Bayesian Gaussian mixture model is a variant of the Gaussian mixture model with a variational inference algorithm, which can avoid the singularities that often appear in the expectation maximization EM algorithm, but it will also bring slight deviations to the model.

[0068] K-Means is a clustering algorithm based on Euclidean distance, which believes that the closer the distance between two targets, the greater the similarity. The steps are: first divide the data into K groups, randomly select K objects as the initial cluster centers, then calculate the distance between each object and each seed cluster center, and assign each object to the cluster center closest to it. Each time a sample is assigned, the cluster center will be recalculated based on the existing objects in the cluster. This process will be repeated until no (or a minimum number of) objects are reassigned to different clusters, or no (or a minimum number of) cluster centers change again, and the error sum of squares is locally minimized.

[0069] The hierarchical clustering model is: Hierarchical clustering assumes that there is a hierarchical structure between categories and clusters samples into hierarchical classes. Hierarchical clustering has two methods: aggregation and divisive. Aggregation clustering divides each sample into one class, calculates the distance between each sample, and merges the two closest classes into a new class. At this time, it is necessary to calculate the distance between the new class and other classes, and select the two classes with the shortest distance from all distances to merge, and repeat this operation until the stopping condition is met. Aggregation clustering can also be called bottom-up clustering. Divisive clustering divides all classes into one class, and then divides the two classes with the farthest distance into two new classes, and repeats the operation until the stopping condition is met. Divisive clustering is also called top-down clustering.

[0070] Considering that the fluctuation samples are unlabeled samples, the clustering algorithm in machine learning is selected to model the samples, aiming to distinguish different types of liquid level fluctuations.

[0071] Since the labels of normal samples are known, the recall and precision of normal samples can be used to evaluate the quality of the clustering model.

[0072] Precision: The proportion of correct predictions in the predicted category, the denominator is the number of samples in the predicted category. Recall: The proportion of correct predictions in the true category, the denominator is the number of samples in the true category.

[0073] Cluster analysis was performed based on the principal components extracted by 5 kernel functions. The cluster analysis algorithms used in this paper are Gaussian mixture model, variational Bayesian Gaussian mixture model, K-Means clustering, and hierarchical clustering. Since the specific classification of the data set is not known in advance, and it is not known how many categories are appropriate, 6-15 categories are selected, a total of 10 cases. The recall and precision of normal samples based on different clustering models of the training set under different kernel function dimensionality reduction conditions are shown in the following Tables 3, 4, 5, and 6:

[0074] Table 3 Recall and precision of Gaussian mixture model under different clustering conditions

[0075]

[0076] Table 4 Recall and precision of the variational Bayesian Gaussian mixture model under different clustering conditions

[0077]

[0078] Table 5 Recall and precision of K-Means model under different clustering conditions

[0079]

[0080] Table 6 Recall and precision of hierarchical clustering under different clustering conditions

[0081]

[0082] In the selection of dimensionality reduction methods, it can be found that the principal component dimensionality reduction effect based on RBF kernel function and Sigmoid kernel function is significantly better than other kernel functions. If the model with recall and precision rates above 60% on normal samples is defined as a qualified model, it can be seen that hierarchical clustering performs the worst, and there is no qualified model; the Gaussian mixture model performs better than the variational Bayes Gaussian mixture model and K-Means clustering.

[0083] In the Gaussian mixture model, we can see that when the principal component dimensionality reduction is performed based on the Sigmoid kernel function, when the training samples are divided into 11 categories, the recall rate of 67.024% and the precision rate of 99.365% with the best comprehensive evaluation can be achieved; when the principal component dimensionality reduction is performed based on the RBF kernel function, when the training samples are divided into 9 categories, the recall rate of 67.88% and the precision rate of 92.42% with the best comprehensive evaluation can be achieved.

[0084] By calculating the recall and precision of normal samples, it is found that when the Gaussian mixture model obtained by principal component dimensionality reduction based on the Sigmoid kernel function is also divided into 11 categories, the comprehensive evaluation is the best, with a recall rate of 41.509% and a precision rate of 98.507%; when the principal component dimensionality reduction is based on the RBF kernel function, the training samples are also divided into 9 categories, and the comprehensive evaluation is the best, with a recall rate of 57.233% and a precision rate of 94.792%. From the change in recall rate from the training set to the test set, it can be seen that when principal component dimensionality reduction is based on the RBF kernel function, the generalization ability of the model is better than that of the Gaussian mixture model based on the Sigmoid kernel function.

[0085] 5) Classify the liquid level fluctuation clustering situations according to the model constructed by the optimal dimensionality reduction method, and set corresponding solutions for various situations;

[0086] By querying the actual data curve, it can be found that in the test set, the level fluctuation clustering based on the RBF kernel function for principal component dimension reduction can be roughly divided into the following 8 categories: level fluctuation within 5 minutes of pouring, low-frequency level fluctuation, high-frequency level fluctuation, level fluctuation caused by water change, divergent level fluctuation, sudden level fluctuation, level fluctuation caused by flocculation, and periodic level fluctuation. After obtaining the specific clustering results, the specific influencing factors of each level fluctuation category can be located according to the actual situation and combined with the experience of senior engineers, and the treatment measures that can be taken can be obtained. If a fluctuation sample is classified as a level fluctuation caused by flocculation, the level fluctuation can be reduced by increasing jitter.

[0087] 6) The detected data is input into the constructed model in real time for monitoring. When the data is abnormal, the corresponding solution is output and provided to the operator or the control center for control.

[0088] In the above step 1), the liquid level, section number and steel grade number are set as the system setting values; the given pulling speed, the given position of the stopper rod, the bulging compensation amount, the pouring time and the NP value of the electromagnetic instrument are the theoretical values ​​calculated by the system; the actual liquid level, the actual pulling speed, the actual position of the stopper rod, the weight of the middle bag, the actual liquid level (electromagnetic) electromagnetic instrument N value and the sensor temperature are the acquired values ​​of the corresponding sensors or equipment.

[0089] With the continuous maturity of the new generation of information technology represented by big data and artificial intelligence, more and more difficult industrial problems have found new solutions and methods. In this paper, a cluster analysis is performed on the problem of crystallizer liquid level fluctuation using machine learning methods. It is found that the Gaussian mixture model based on the Sigmoid kernel function for principal component dimensionality reduction has good precision and recall when the sample data is divided into 11 categories; the Gaussian mixture model based on the RBF kernel function for principal component dimensionality reduction has good precision and recall when the sample data is divided into 9 categories; and the generalization ability of the Gaussian mixture model based on the RBF kernel function for principal component dimensionality reduction is better than that of the Gaussian mixture model based on the Sigmoid kernel function for principal component dimensionality reduction. By comparing the historical curves, it is found that the clustering results can reflect the actual situation to a certain extent, providing a new solution for locating the specific influencing factors of liquid level fluctuation, avoiding the problem of inaccurate control caused by only detecting the height of the molten steel liquid level to control the crystallizer liquid level, thereby improving the accuracy of the crystallizer liquid level control.

[0090] Based on the same concept, this embodiment provides a crystallizer liquid level control device, such as Figure 2 As shown, the crystallizer liquid level control device comprises:

[0091] An acquisition module 1, wherein the acquisition module 1 is used to acquire influencing data of the liquid level of the crystallizer, wherein the influencing data is determined according to influencing factors of the liquid level of the crystallizer;

[0092] An input module 2, the input module 2 is used to input the influencing data into a crystallizer liquid level fluctuation factor determination model to determine the fluctuation state of the crystallizer liquid level;

[0093] A control module 3 is used to adjust the influencing factors of the crystallizer according to the fluctuation state to control the liquid level of the crystallizer.

[0094] It should be understood that the various modules of the crystallizer liquid level control device provided in this embodiment can be combined to implement the various steps of the above-mentioned crystallizer liquid level control method to achieve the same technical effects as the various steps of the above-mentioned crystallizer liquid level control method, and therefore, they will not be repeated here.

[0095] like Figure 3 As shown, an embodiment of the present application provides an electronic device, including a processor 111, a communication interface 112, a memory 113 and a communication bus 114, wherein the processor 111, the communication interface 112, and the memory 113 communicate with each other through the communication bus 114.

[0096] Memory 113, used for storing computer programs;

[0097] In one embodiment of the present application, the processor 111 is used to implement the steps of the crystallizer liquid level control method provided by any of the aforementioned method embodiments when executing the program stored in the memory 113.

[0098] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the crystallizer liquid level control method provided in any of the aforementioned method embodiments are implemented.

[0099] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0100] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for controlling a crystallizer liquid level, characterized in that: The crystallizer liquid level control method comprises: Acquiring influencing data of a crystallizer liquid level, wherein the influencing data is determined according to influencing factors of the crystallizer liquid level; Inputting the influencing data into a crystallizer liquid level fluctuation factor determination model to determine the fluctuation state of the crystallizer liquid level; Adjusting the influencing factors of the crystallizer according to the fluctuation state to control the liquid level of the crystallizer; Before inputting the influencing data into the crystallizer liquid level fluctuation factor determination model, the method further includes: Obtain historical impact data, and determine normal samples and fluctuation samples based on the historical impact data; Determine a training set and a test set based on the normal samples and the fluctuation samples; Modeling the training set and the test set by using a clustering algorithm to obtain a crystallizer liquid level fluctuation factor determination model; The historical impact data includes: actual liquid level and set liquid level; wherein determining normal samples and fluctuation samples according to the historical impact data includes: Determine the difference between the actual liquid level and the set liquid level at each moment in the historical impact data; When the difference between the actual liquid level and the set liquid level at any moment in the historical impact data exceeds the first threshold, the data at any moment is used as a fluctuation sample, and the sample interval of the fluctuation sample is the interval corresponding to the period from 5 minutes before any moment to 1 minute after any moment; When the difference between the actual liquid level and the set liquid level at any moment in the historical impact data is lower than the second threshold, the data at any moment is taken as the normal sample, the second threshold is lower than the first threshold, and the sample interval of the normal sample is the interval corresponding to the period from 5 minutes before any moment to 1 minute after any moment; Wherein, determining a training set and a test set based on the normal samples and the fluctuation samples includes: Mixing the normal samples and the fluctuation samples to obtain a total training set; Dividing the total training set based on a preset ratio to obtain the training set and the test set; The training set and the test set are modeled by a clustering algorithm to obtain a crystallizer liquid level fluctuation factor determination model, including: Respectively extracting the time domain features and frequency domain features of the changing variables in the training set and the test set to obtain the training set data dimension and the test set data dimension; Using the RBF kernel function to perform dimensionality reduction processing on the training set data dimension and the test set data dimension to obtain training set dimensionality reduction data and test set dimensionality reduction data; Modeling the training set dimensionality reduction data and the test set dimensionality reduction data by using a clustering algorithm to obtain the crystallizer liquid level fluctuation factor determination model; The training set dimensionality reduction data and the test set dimensionality reduction data are modeled by a clustering algorithm to obtain the crystallizer liquid level fluctuation factor determination model, including: Modeling the training set dimensionality reduction data and the test set dimensionality reduction data by at least two clustering algorithms to obtain at least two initial crystallizer liquid level fluctuation factor determination models; Evaluating at least two of the initial crystallizer liquid level fluctuation factor determination models based on the normal samples to obtain the optimal initial crystallizer liquid level fluctuation factor determination model; The optimal initial crystallizer liquid level fluctuation factor determination model is used as the crystallizer liquid level fluctuation factor determination model.

2. The method for controlling the crystallizer liquid level according to claim 1, characterized in that: After using the optimal initial crystallizer liquid level fluctuation factor determination model as the crystallizer liquid level fluctuation factor determination model, the method further includes: Determine the clustering of the training set dimensionality reduction data and the test set dimensionality reduction data according to the crystallizer liquid level fluctuation factors to obtain a plurality of preset fluctuation states; A method for adjusting the influencing factors corresponding to the preset fluctuation state is set.

3. A crystallizer liquid level control device, characterized in that: The crystallizer liquid level control device comprises: An acquisition module, the acquisition module is used to acquire influencing data of the liquid level of the crystallizer, wherein the influencing data is determined according to influencing factors of the liquid level of the crystallizer; An input module, the input module is used to input the influencing data into a crystallizer liquid level fluctuation factor determination model to determine the fluctuation state of the crystallizer liquid level; A control module, the control module is used to adjust the influencing factors of the crystallizer according to the fluctuation state to control the liquid level of the crystallizer; Wherein, the device is also used for: Obtain historical impact data, and determine normal samples and fluctuation samples based on the historical impact data; Determine a training set and a test set based on the normal samples and the fluctuation samples; Modeling the training set and the test set by using a clustering algorithm to obtain a crystallizer liquid level fluctuation factor determination model; Wherein, the historical impact data includes: actual liquid level and set liquid level; the device is also used for: Determine the difference between the actual liquid level and the set liquid level at each moment in the historical impact data; When the difference between the actual liquid level and the set liquid level at any moment in the historical impact data exceeds the first threshold, the data at any moment is used as a fluctuation sample, and the sample interval of the fluctuation sample is the interval corresponding to the period from 5 minutes before any moment to 1 minute after any moment; When the difference between the actual liquid level and the set liquid level at any moment in the historical impact data is lower than the second threshold, the data at any moment is taken as the normal sample, the second threshold is lower than the first threshold, and the sample interval of the normal sample is the interval corresponding to the period from 5 minutes before any moment to 1 minute after any moment; Wherein, determining a training set and a test set based on the normal samples and the fluctuation samples includes: Mixing the normal samples and the fluctuation samples to obtain a total training set; Dividing the total training set based on a preset ratio to obtain the training set and the test set; The training set and the test set are modeled by a clustering algorithm to obtain a crystallizer liquid level fluctuation factor determination model, including: Respectively extracting the time domain features and frequency domain features of the changing variables in the training set and the test set to obtain the training set data dimension and the test set data dimension; Using the RBF kernel function to perform dimensionality reduction processing on the training set data dimension and the test set data dimension to obtain training set dimensionality reduction data and test set dimensionality reduction data; Modeling the training set dimensionality reduction data and the test set dimensionality reduction data by using a clustering algorithm to obtain the crystallizer liquid level fluctuation factor determination model; The training set dimensionality reduction data and the test set dimensionality reduction data are modeled by a clustering algorithm to obtain the crystallizer liquid level fluctuation factor determination model, including: Modeling the training set dimensionality reduction data and the test set dimensionality reduction data by at least two clustering algorithms to obtain at least two initial crystallizer liquid level fluctuation factor determination models; Evaluating at least two of the initial crystallizer liquid level fluctuation factor determination models based on the normal samples to obtain the optimal initial crystallizer liquid level fluctuation factor determination model; The optimal initial crystallizer liquid level fluctuation factor determination model is used as the crystallizer liquid level fluctuation factor determination model.

4. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor is used to implement the steps of the crystallizer liquid level control method described in claim 1 or 2 when executing the program stored in the memory.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the crystallizer liquid level control method according to claim 1 or 2 are implemented.