CN108364095B - Diagnosis method of molten steel quality in steelmaking production process based on data mining - Google Patents

Abstract

The invention provides a method for diagnosing molten steel quality in a steelmaking production process based on data mining, and belongs to the technical field of molten steel quality diagnosis. According to the method, data are collected and screened, then the data are subjected to standardized processing, steel moisture and temperature control targets of all processes based on clustering are determined, reasonable modes of molten steel (molten iron) classification and process operation process are extracted, process molten steel quality is predicted, a process operation scheme is provided, and finally system early warning is achieved. The method fills the gap that most of the existing researches only aim at the quality research of casting blanks, and can also regulate and control the quality of molten steel in time in the steelmaking production process, thereby ensuring the stability and narrow window control in the production process and effectively improving the product quality.

Description

Diagnosis method of molten steel quality in steelmaking production process based on data mining

technical field

The invention relates to the technical field of molten steel quality diagnosis, in particular to a method for diagnosing molten steel quality in a steelmaking production process based on data mining.

Background technique

At present, the outstanding problems of my country's iron and steel production are manifested in the overproduction of conventional steel and a certain gap between the quality of high-quality steel and the international advanced level, which often needs to be realized by importing. When some key steel products have problems, quality positioning and diagnosis are required. Therefore, it is urgent to use the method combining modern data mining and metallurgical mechanism to realize the whole process process quality diagnosis. Many domestic and foreign experts have made research on it. The online monitoring and offline analysis system for metallurgical whole process quality developed by the National Engineering Research Center of High Efficiency Rolling of University of Science and Technology Beijing proposes to comprehensively collect, monitor and online rate the process quality parameters of the whole process for the quality control of automobile plate and substrate. It is necessary to quickly trace, analyze, optimize and improve the whole process. Peng Kaixiang of Beijing University of Science and Technology and others proposed a fault diagnosis method and device for the quality of hot continuous strip rolling, which solves the problem that the product quality in the prior art is often controlled by more skilled operators with their own experience, so that once a fault occurs, It is difficult to guarantee the quality of products only by the feedback control strategy of delay and lag.

However, the current quality diagnosis technology is difficult to realize the diagnosis and early warning of iron and steel products in the production process. It is only after the production of a certain pouring time that it is sampled and checked until the next pouring time, and then the quality adjustment is carried out, which is difficult to prevent quality problems. The emergence of batches has brought great limitations to improving production efficiency. At the same time, most of the quality diagnosis in the metallurgical industry is for the quality diagnosis of the cast slab, and there are few quality diagnosis for the three production processes of molten iron pretreatment, converter steelmaking and out-of-furnace refining. At the same time, most of the quality diagnosis in the iron and steel industry is put forward on the premise that the process variables are independent of each other from the perspective of the diagnosis object, while the present invention adopts a comprehensive quality diagnosis technology for dependent variables. There is a certain relationship with the ingredients, and it is a method of comprehensive consideration of multiple factors.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide a method for diagnosing molten steel quality in the steelmaking production process based on data mining.

The method includes the following steps:

(1) Data collection and screening:

Use the secondary system of the steel plant to collect the production process data of the steel plant, use the heat number, the production data of the four processes of series hot metal pretreatment, converter steelmaking, LF furnace refining, and RH furnace refining, and generate a joint table to filter out valid data in it;

(2) Standardization of data:

Standardize the valid data screened out in step (1);

(3) Determine the composition and temperature control target of molten steel or molten iron in each process based on clustering:

From the standardized data in step (2), screen out the sets of heats processed by different processes, and further screen out all sets of heats whose outbound components and temperatures of each process meet the operating rules; analyze the temperature and composition of molten steel or molten iron The influence law of parameters on the treatment of different processes, the temperature and composition parameters of molten steel or molten iron that have an important impact on the quality of molten steel or molten iron are selected as clustering factors, and the selected outbound components and temperatures of each process are set for all heats that meet the operating regulations. The heat data inside is clustered by K-Means clustering algorithm to determine the target of molten steel or molten iron composition and temperature control in each process;

(4) Classification of molten steel or molten iron and extraction of a reasonable mode of process operation technology:

All heats screened out in step (3) where the outbound components and temperatures of each process meet the operating rules are set as hit heats, and the rest of the heats are non-hit heats. Analyze and compare the different processes between the hit heats and the non-hit heats. Processing technology, summarize the processing technology characteristics of the hit heat and the processing technology characteristics of the non-hit heat, the processing technology of the hit heat is the reasonable corresponding process process mode of this type of molten steel, analyze and collect the process modes of different processes, And enter the database management;

(5) Prediction of molten steel quality in process:

The temperature and composition of incoming molten steel or molten iron for a certain heat are collected through the secondary system. If it exceeds the control target of entering the station for the corresponding process, but is within the controllable range (that is, the composition temperature of all hit heats is traced back to the time of entering the station) The set of information is within the controllable range of the process), then according to the temperature and composition of molten steel or molten iron, it is classified into the corresponding category after processing in step (3), and then according to the hit of the category Predict the probability that the temperature and composition of the molten steel in this heat or molten iron after the corresponding process will reach the end point control target, the value is equal to the hit rate;

(6) The process operation plan provides:

According to the prediction of molten steel quality in step (5), a process operation mode scheme is provided for the corresponding heat of the corresponding process to ensure that the control target requirements are finally achieved;

(7) Warning:

If the temperature and composition of incoming molten steel in a heat exceeds the controllable range, the system will alarm.

Among them, the valid data in step (1) is the data of a certain heat in the whole steelmaking process, from KR preprocessing to the end of refining, the temperature, composition, various process operations and parameter information are complete and within a reasonable range.

The standardization process in step (2) is to use the Min-Max standardization method to convert all clustering factors into scalars and map them in the interval [0,1].

The beneficial effects of the above-mentioned technical solutions of the present invention are as follows:

The method of the invention fills the gap that most of the existing researches only focus on the quality of the cast slab, and at the same time, the quality of molten steel can be regulated in time during the steelmaking production process, thereby ensuring the stability of the production process and narrow window control, and can Effectively improve product quality.

Description of drawings

Fig. 1 is the technical flow chart of the molten steel quality diagnosis method of steelmaking production process based on data mining of the present invention;

Fig. 2 is the flow chart of the system diagnosing method of molten steel temperature and composition in the steelmaking production process provided by the embodiment of the present invention during practical application;

3 is a flowchart of the K-Means clustering algorithm used when the production data is clustered and divided according to an embodiment of the present invention;

FIG. 4 is a flowchart of a decision tree algorithm provided by an embodiment of the present invention.

Detailed ways

In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

The present invention provides a method for diagnosing molten steel quality in a steel-making production process based on data mining. As shown in FIG. 1 , the method includes the following steps:

(1) Data collection and screening:

Use the secondary system of the steel plant to collect the production process data of the steel plant, use the heat number, the production data of the four processes of series hot metal pretreatment, converter steelmaking, LF furnace refining, and RH furnace refining, and generate a joint table to filter out valid data in it;

(2) Standardization of data:

Standardize the valid data screened out in step (1);

(3) Determine the composition and temperature control target of molten steel or molten iron in each process based on clustering:

From the standardized data in step (2), screen out the sets of heats processed by different processes, and further screen out all sets of heats whose outbound components and temperatures of each process meet the operating regulations; analyze the temperature and composition of molten steel or molten iron The influence law of parameters on the treatment of different processes, the temperature and composition parameters of molten steel or molten iron that have an important impact on the quality of molten steel or molten iron are selected as clustering factors, and the selected outbound components and temperatures of each process are set for all heats that meet the operating regulations. The heat data inside is clustered by K-Means clustering algorithm to determine the target of molten steel or molten iron composition and temperature control in each process;

(4) Classification of molten steel or molten iron and extraction of a reasonable mode of process operation technology:

All heats screened out in step (3) where the outbound components and temperatures of each process meet the operating rules are set as hit heats, and the rest of the heats are non-hit heats. Analyze and compare the different processes between the hit heats and the non-hit heats. Processing technology, summarize the processing technology characteristics of the hit heat and the processing technology characteristics of the non-hit heat, the processing technology of the hit heat is the reasonable corresponding process process mode of this type of molten steel, analyze and collect the process modes of different processes, And enter the database management;

(5) Prediction of molten steel quality in process:

The temperature and composition of incoming molten steel or molten iron for a certain heat are collected through the secondary system. If it exceeds the control target of entering the station for the corresponding process, but is within the controllable range (all hit heats are traced back to the composition temperature information at the time of entering the station) within the set), then according to the temperature and composition of molten steel or molten iron, classify it into the corresponding category after processing in step (3), and then predict the molten steel or molten iron in this heat according to the hit rate of the category. The probability that the temperature and composition of molten steel processed by the corresponding process reach the end point control target, the value is equal to the hit rate;

(6) The process operation plan provides:

According to the prediction of molten steel quality in step (5), a process operation mode scheme is provided for the corresponding heat of the corresponding process to ensure that the control target requirements are finally achieved;

(7) Warning:

If the temperature and composition of incoming molten steel in a heat exceeds the controllable range, the system will alarm.

As shown in Figure 2, the specific implementation process is as follows:

(1) Data collection and screening

Collect historical sample data of a certain type of molten steel in the production process of the steelmaking plant, the data includes KR molten iron pretreatment data, converter steelmaking data, LF refining process data, RH refining process data, slab continuous casting production control Records, ladle process query records, use the database to associate these production data based on the heat number, form a joint data table for the whole process of the steelmaking plant, and filter out the valid data. It mainly studies the composition and temperature data of molten steel that have an important influence on the quality of molten steel when it enters and exits each station. For different steel grades, all the processed heats of this steel grade are screened from the historical sample data and recorded as set S, and in the set S of all heats, all the heats whose outbound composition and temperature meet the operating regulations are hit heats , denoted as the set P.

(2) Standardization of data

Since these temperature and composition data have different distribution intervals and units, it is unreasonable to directly cluster them, so they must be standardized first. The processing method used here is the Min-Max normalization processing method. This method is a linear transformation of the original data. Let minA and maxA be the maximum and minimum values of attribute A, respectively, and an original value x of A is mapped to the value x` in the interval [0,1] through Min-Max normalization. The formula is:

(3) Determination of composition and temperature control target of molten steel (hot metal) in each process based on clustering

First trace the information of the set P back to the time of entering the station, and then use the K-Means clustering algorithm to perform cluster analysis on the historical sample data of its composition and temperature information. As shown in Figure 3, the composition of a certain furnace of molten steel and Multiple variables such as temperature information are stored in the form of a matrix. The rows of the matrix are used to represent the molten steel of each heat, and the columns are used to represent the information of the molten steel of this heat. Then the set of p pieces of information about the molten steel of n furnaces can be represented by an n× The p-dimensional matrix is represented, and the jth information of the molten steel in the ith heat is represented as x _ij in the matrix, and the data matrix is as follows:

Then the dissimilarity matrix is used to store the differences between different heats of the same steel grade. The dissimilarity matrix of n heats of molten steel is represented as an n×n-dimensional matrix, and d(A, B) is used to represent A The dissimilarity with B, the dissimilarity matrix of the set X={x ₁ ,x ₂ ,...,x _n } containing n heats is expressed as follows:

Here, d(x _i ,x _j ) is a similarity measure function, when x _i is similar or similar, the value of d(x _i ,x _j ) is close to 0, and when the value of d(x _i ,x _j ) is close to 0 When it is larger, it means that the heats x _i and x _j are very different. The most commonly used similarity measure function is Euclidean distance, which is defined as:

x _i and x _j represent any two furnaces of molten steel, p is the variable number of molten steel, and n is the number of furnaces.

Determine the dataset X={x ₁ ,x ₂ ,...,x _n } and the number of clusters according to the historical sample data;

①Initialization: randomly assign k cluster centers (m ₁ , m ₂ ,...,m _k );

②Assign _xi : For each sample _xi , find the cluster center closest to it and assign it to this class;

③ Recalculate the center of each cluster:

④Calculation deviation:

⑤ Judgment convergence: if the J value converges, the algorithm terminates; otherwise, return to the second step.

⑥ Obtaining modes: k modes can be obtained by repeated operations until convergence

These k modes correspond to k cluster centers, and the k centers are used as the cluster centers of all data sets S to perform clustering again, so that all production data are divided into k categories, and for each type of data The qualified rate after the end of the process is excavated, and it is considered that the set with high hit rate is the operation procedure when the process enters the station, and the molten steel temperature value m and each component value n _i of the cluster center of this type of molten steel are the optimal control values. , according to the requirements of smelting steel grades, determine the target range of the end point ΔT`, and then combine the molten steel temperature drop value ΔT between processes, the target temperature at the end of the previous process is m+ΔT±ΔT`. According to the requirements of smelting steel grades, the target range Δn of the end point is determined, and the composition does not change much between processes, so the target composition at the end point of the previous process is n _i + Δn.

(4) Classification of molten steel (hot metal) and extraction of reasonable models of process operation technology

In the previous step, k types of molten steel were obtained. In each type of molten steel, some of the heats belong to the hit heats. The set of all the hit heats is the controllable range of the molten steel in this process. The proportion of the middle hit heats in all the heats of this type is recorded as the hit rate j of this type of molten steel entering the station in this process.

Analyze and compare the difference between the hit heat and the non-hit heat in this process, and summarize and extract the processing technology characteristics of the hit heat. This process is done using the decision tree method:

As shown in Figure 4, the continuous attribute is processed by the decision tree method. Assuming that the continuous attribute a has n different values in the sample set D, the proportion of qualified and unqualified samples is p _k (k=1, 2,…,|y|)

① Sort the n different values of a continuous attribute that appear in the sample set from small to large, denoted as {a ¹ ,a ² ,...,a ⁿ }

作为可能的分裂点t，将数据集分为两部分，计算每个可能的分裂点的信息增益：②The midpoint between two adjacent values a ⁱ and a ⁱ⁺¹ of an attribute

As a possible split point t, divide the dataset into two parts and calculate the information gain for each possible split point:

where information entropy

③ Select the split point with the largest information gain after correction as the best split point for the feature

④ Calculate the information gain rate of the best split point and modify it as the information gain of the attribute

in

⑤Comparing the information gain rate of each attribute to construct a decision tree, the one with the largest gain rate is used as the root node, and the decision tree is extended in turn and the decision tree is pruned.

The treatment process of the hit heat is the reasonable treatment process mode of this type of molten steel in this process. Analyze, collect and record the processing mode of all k-type molten steel in a certain process, and enter it into database management.

(5) Prediction of molten steel quality in process

The incoming molten steel temperature and composition of a certain heat in a certain process are collected through the secondary system. If it exceeds the incoming control target of the process, but is within the controllable range, the temperature and composition will be taken according to the value of the temperature and composition. It is classified into one of the k categories, and then predicts the probability that the molten steel temperature and composition of the molten steel after RH treatment in this heat will reach the end point control target according to the hit rate of this category, and the value is equal to the hit rate.

(6) Provide process operation plan

According to the process mode corresponding to the k-type incoming molten steel of a certain process, a reasonable process operation mode scheme is provided for this heat, and the process is guided to carry out reasonable treatment to ensure that the control target requirements are finally achieved.

(7) Early warning

If the temperature and composition of the incoming molten steel in a certain process of a certain heat exceeds the controllable range, the system will alarm, warning the process that the temperature and composition of the molten steel in this heat exceeds the temperature and composition range of all hit heats in the historical data. The hit rate of this process is very low, and it is necessary to focus on it and choose an unconventional processing process.

The above are the preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, without departing from the principles of the present invention, several improvements and modifications can be made. These improvements and modifications It should also be regarded as the protection scope of the present invention.

Claims (2)

1. a steelmaking production process molten steel quality diagnosis method based on data mining, is characterized in that: comprise steps as follows: (1) Data collection and screening: Use the secondary system of the steel plant to collect the production process data of the steel plant, use the heat number, the production data of the four processes of series hot metal pretreatment, converter steelmaking, LF furnace refining, and RH furnace refining, and generate a joint table to filter out valid data in it; (2) Standardization of data: Standardize the valid data screened out in step (1); Using the Min-Max normalization processing method, let minA and maxA be the maximum and minimum values of attribute A, respectively, and an original value x of A is mapped to the value x` in the interval [0,1] through Min-Max normalization, and its The formula is:

(3) Determination of composition and temperature control target of molten steel or molten iron in each process based on clustering: From the standardized data in step (2), screen out the sets of heats processed by different processes, and further screen out all sets of heats whose outbound components and temperatures of each process meet the operating rules; analyze the temperature and composition of molten steel or molten iron The influence law of parameters on the treatment of different processes, the temperature and composition parameters of molten steel or molten iron that have an important impact on the quality of molten steel or molten iron are selected as clustering factors, and the selected outbound components and temperatures of each process are set for all heats that meet the operating regulations. The heat data inside is clustered by K-Means clustering algorithm to determine the target of molten steel or molten iron composition and temperature control in each process; Specifically: The composition and temperature information variables of a certain heat of molten steel are stored in the form of a matrix, the rows of the matrix represent the molten steel of each heat, and the columns represent the information of the molten steel of this heat, then the collection of n heats of molten steel p pieces of information It can be represented by an n×p-dimensional matrix. The jth information of molten steel in the i-th heat is represented as x _ij in the matrix, and the data matrix is as follows:

Then the dissimilarity matrix is used to store the differences between different heats of the same steel grade. The dissimilarity matrix of n heats of molten steel is represented as an n×n-dimensional matrix, and d(A, B) is used to represent A The dissimilarity with B, the dissimilarity matrix of the set X={x ₁ ,x ₂ ,...,x _n } containing n heats is expressed as follows:

d(x _i ,x _j ) is a similarity measure function. When x _i is similar or similar, the value of d(x _i ,x _j ) is close to 0, and when the value of d(x _i ,x _j ) is larger , which means that the heats x _i and x _j are very different; the most commonly used similarity measure function is the Euclidean distance, which is defined as:

x _i and x _j represent any two heats of molten steel, p is the information number of molten steel, and n is the number of heats; Determine the dataset X={x ₁ ,x ₂ ,...,x _n } and the number of clusters according to the historical sample data: ①Initialization: randomly assign k cluster centers (m ₁ , m ₂ ,...,m _k ); ②Assign _xi : For each sample _xi , find the cluster center closest to it and assign it to this class; ③ Recalculate the center of each cluster:

④Calculation deviation:

⑤ Judgment convergence: if the J value converges, the algorithm terminates; otherwise, return to step ②; ⑥ Obtaining modes: k modes can be obtained by repeated operations until convergence; These k modalities correspond to k clustering centers, and the k centers are used as the clustering centers of all data sets S to perform clustering again, and all production data are divided into k categories, and the mining process for each category of data is completed. It is considered that the set of classes with a high hit rate is the operation procedure at the time of entry of the process, and the molten steel temperature value m and each component value n _i of the cluster center of this class of molten steel are the optimal control values, according to the smelting According to the requirements of the steel grade, the target range of the end point ΔT` is determined, and combined with the temperature drop value ΔT of molten steel between processes, the target temperature at the end point of the previous process is m+ΔT±ΔT`; according to the requirements of the smelting steel grade, the target range of the end point Δn is determined , then the target component at the end of the previous process is n _i +Δn; (4) Classification of molten steel or molten iron and extraction of a reasonable mode of process operation technology: In step (3), the outbound components of each process and all the heats whose temperature meets the operating rules are set as hit heats, and the rest of the heats are non-hit heats. The decision tree method is used to analyze and compare the hit and non-hit heats. The processing technology of the different processes of the second heat is summarized, and the processing technology characteristics of the hit heat and the non-hit heat are summarized. Analyze, collect, and enter database management; (5) Prediction of molten steel quality in process: The temperature and composition of incoming molten steel or molten iron for a certain heat are collected through the secondary system. If it exceeds the entry control target of the corresponding process, but is within the controllable range, the temperature and composition of molten steel or molten iron will be taken according to the value. It is classified into the corresponding category after the treatment in step (3), and then the probability that the molten steel temperature and composition of the molten steel or molten iron processed by the corresponding process will reach the end point control target is predicted according to the hit rate of the category, and the value is equal to the hit rate. Rate; (6) The process operation plan provides: According to the prediction of molten steel quality in step (5), a process operation mode scheme is provided for the corresponding heat of the corresponding process to ensure that the control target requirements are finally achieved; (7) Warning: If the temperature and composition of incoming molten steel for a certain heat exceeds the controllable range, the system will alarm; The valid data in the step (1) is the complete temperature, composition, various technological operations and parameter information of the data of a certain heat in the whole steelmaking process from KR preprocessing to the end of refining; The standardization process in the step (2) is to use the Min-Max standardization method to convert all the clustering factors into scalars and map them in the interval [0,1]. 2. The method for diagnosing the quality of molten steel in a steelmaking production process based on data mining according to claim 1, wherein the controllable range in the step (5) is that all hit heats are traced back to the composition at the time of entering the station A collection of temperature information.

Images