A length self-adaptive control method for a seamless steel pipe hot rolling tension unit

Abstract

This invention discloses an adaptive length control method for a hot-rolling tension reduction mill for seamless steel pipes. It utilizes historical data to establish a length calculation model for steel pipes of the same steel grade and specification. The process control system tracks materials in real time. If the current signal indicates the completion of tension reduction steel pipe length measurement, model error and the average historical tension reduction coefficient are processed. If the current signal indicates steel tapping from the reheat furnace, it determines whether the steel grade or specification of the next steel pipe to be rolled has changed compared to the previous rolled steel pipe, thus achieving differentiated setting control of the tension reduction coefficient. Based on the established steel pipe length calculation model, this invention further considers the current rolling process state and historical differences, calculating the impact of the tension coefficient on the rolling length under the current state, achieving online dynamic control of the steel pipe length in the tension reduction mill.

Description

Technical Field

[0001] This invention relates to metallurgical production technology, and more specifically, to a length adaptive control method for a hot rolling tension reduction mill for seamless steel pipes. Background Technology

[0002] The basic production process of hot-rolled seamless steel pipes is as follows: after the billet is heated in a furnace, it passes through a piercing mill and a continuous rolling mill to obtain a steel pipe with an intermediate wall thickness and length. Then, it is rolled by a tension reduction mill to obtain the finished hot-rolled steel pipe. Hot-rolled seamless steel pipes are managed and produced in batches, with each batch corresponding to the same product specifications and generally using the same rolling process parameters for production control. In actual control, fluctuations in equipment status and process control, such as inconsistent oxidation loss due to heating time and temperature fluctuations, and uneven wall thickness during rolling, can affect the final length of the finished steel pipe. When online pipe length measurement devices are lacking, on-site production mainly relies on sampling inspection, i.e., randomly selecting a portion of the steel pipes in a batch and measuring their length offline to ensure they are within acceptable limits, thus preventing batches of steel pipes from failing to meet finished product requirements. Now, advanced hot-rolled seamless steel pipe production lines install online laser length measuring devices. When on-site production technicians detect abnormal pipe lengths, they adjust the rolling parameters based on production experience to achieve effective control over the pipe length. Existing control methods cannot achieve online dynamic control of the length of each individual steel pipe. In particular, the intervention and adjustment of length control by different technicians based on production experience introduces human differences, which is not conducive to precise product control. Furthermore, the current trend in seamless steel pipe production is shifting from large-scale centralized production to multi-variety, small-batch production, making the necessity of online dynamic control of steel pipe length even more urgent. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the purpose of this invention is to provide an adaptive length control method for hot-rolled tension reduction mills of seamless steel pipes. Based on the established steel pipe length calculation model, this method further considers the current rolling process status and historical differences, calculates the influence of the tension coefficient on the rolling length under the current status, and realizes online dynamic control of the steel pipe length of the tension reduction mill.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] An adaptive length control method for hot rolling tension reduction mills of seamless steel pipes;

[0006] Using historical data, a calculation model for the length of steel pipes of the same steel type and specification was established;

[0007] The process control system tracks the material in real time. If the current signal is the completion signal of the length measurement of the tension reduction steel pipe, the model error and the average value of the historical tension reduction coefficient are processed. If the current signal is the steel tapping signal of the reheat furnace, it is determined whether the steel grade or specification of the next steel pipe to be rolled and the previous rolled steel pipe of the tension reduction unit have changed, so as to realize the differentiated setting control of the tension reduction coefficient.

[0008] Preferably, the historical data includes feeding length, steel pipe empty reduction length, steel pipe continuous rolling length, steel pipe tension reduction length, and tension reduction adjustment coefficient; and

[0009] Heating time of billet heating furnace, billet exit temperature, and reheating furnace temperature.

[0010] Preferably, the steel pipe length calculation model is modeled using a multivariate nonlinear fitting method;

[0011] First, outlier handling is performed, removing samples with missing data and those whose data deviates from a reasonable range.

[0012] Ideally, the processing of model error and the average historical tension coefficient includes:

[0013] 1) When the deviation between the measured length and the theoretical length of the tension reduction is within the allowable error range, the model error is calculated using the following method:

[0014] ΔL new =α×ΔL old +(1-α)×(M new -C new )

[0015] Where, ΔL new This represents the model error value obtained after the current steel pipe rolling is completed;

[0016] ΔL old This represents the model error value obtained after the previous batch of steel pipes of the same type and specification was rolled out;

[0017] M new This indicates the current measured value of the tension reduction rolling length of the steel pipe;

[0018] C new This represents the calculated value of the current steel pipe tension reduction rolling length model;

[0019] α represents the historical weighting coefficient of the model error;

[0020] 2) Processing of historical tension reduction coefficient average value

[0021] If the deviation between the measured tension reduction length and the theoretical tension reduction length is within the allowable error range, the control process is considered stable, and the historical average tension coefficient is updated. Otherwise, the historical average value is kept unchanged. The formula for calculating the historical average tension coefficient is as follows:

[0022] WTCA new =(WTCA) old ×N+wtca) / (N+1)

[0023] Among them, WTCA new This indicates the average tension coefficient corresponding to the current steel pipe specification after the current steel pipe is rolled out;

[0024] WTCA old This represents the average tension coefficient of the corresponding steel pipe specification before rolling.

[0025] wtca represents the tension coefficient currently used in steel pipes;

[0026] N represents the corresponding WTCA old Quantity of steel pipes rolled.

[0027] After the calculation is complete, use WTCA. new Update WTCA old The corresponding number of rolled steel pipes is updated to N+1;

[0028] WTCA old Used to set the tension coefficient of the first steel pipe in a batch.

[0029] Preferably, the calculated value C of the current steel pipe tension reduction rolling length model is... new The XGBoost fitting model was used for calculation;

[0030] The historical error weighting coefficient α of the model takes values ​​of [0.2, 0.5].

[0031] Ideally, if the steel grade or specification of the next steel pipe to be rolled and the previous steel pipe changed in the tension reduction unit, the tension coefficient setting logic of the first steel pipe was adopted, that is, the historical average tension coefficient consistent with the current steel pipe specification was used for setting.

[0032] Ideally, if the steel grade or specification of the next steel pipe to be rolled in the tension reduction mill unit has not changed from that of the previous rolled steel pipe, that is, the current steel pipe to be rolled is not the first pipe of the current rolling batch, then dynamic setting is adopted, and the tension coefficient of the next steel pipe to be rolled is calculated iteratively.

[0033] Preferably, the iterative calculation method is as follows:

[0034] The tension coefficient wtca used in the previous steel pipe tension reduction rolling was the initial value wtca(1), which was substituted into the XGBoost regression model to calculate L(1);

[0035] If the target deviation ΔL(1)=L(1)+ΔL new -L Tar If the value is greater than 0, then wtca is increased by the minimum adjustment step size;

[0036] Conversely, the minimum adjustment step size is used to decrease wtca to obtain wtca(2), which is then substituted into XGBoost to obtain L(2).

[0037] Preferably, during the iterative calculation process, when ΔL(i) and ΔL(i-1) are reversed, the iterative calculation is completed, and the wtca corresponding to the target deviation > 0 is taken as the preset value of the tension reduction coefficient of the next steel pipe.

[0038] The present invention provides a length adaptive control method for a hot rolling tension reduction mill for seamless steel pipes. It uses statistical parameters of the stable rolling process to achieve length control of the first steel pipe in the rolling batch. For subsequent steel pipes in the same rolling batch, a high-precision steel pipe tension reduction length fitting model is used, and the model is adaptively corrected according to the rolling performance to achieve dynamic setting of the tension coefficient of subsequent steel pipes, thereby achieving precise and stable control of the steel pipe length. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of the steel pipe length control logic of the tension reduction unit in the length adaptive control method of the present invention;

[0040] Figure 2 This is a schematic diagram of the calculation process for the tension coefficient of the steel pipe in the tension reduction unit of the length adaptive control method of the present invention;

[0041] Figure 3 This is a schematic diagram illustrating the establishment and prediction process of the XGBoost fitting algorithm for the length adaptive control method of this invention.

[0042] Figure 4 This is a schematic diagram of the XGBoost fitting model effect in an embodiment of the length adaptive control method of the present invention. Detailed Implementation

[0043] To better understand the above-mentioned technical solutions of the present invention, the technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0044] In the hot rolling process of seamless steel pipes, the tension reduction mill is a key process for controlling the pipe length. The tension reduction mill is a multi-stand continuous rolling process. The basic principle of continuous rolling is flow rate balance, meaning that the reduction of area and rolling speed of the steel pipe must remain balanced across different stands. The control process characteristics of the tension reduction mill are: the stand pass shape controls the outer diameter, and the longitudinal tension between the series stands controls the wall thickness. Changes in the outer diameter and wall thickness determine the pipe length. Since the stand pass shape is fixed during production, length control is mainly achieved through tension adjustment. In this invention, the smaller the tension reduction coefficient, the greater the tension between stands. That is, when the tension reduction coefficient < 0, the speed of the preceding stand is reduced based on the base speed corresponding to the standard tension, increasing the speed difference between the preceding and following stands, thereby increasing the overall tension between stands; when the tension reduction coefficient > 0, the speed of the preceding stand is increased based on the base speed corresponding to the standard tension, decreasing the speed difference between the preceding and following stands, thereby decreasing the overall tension between stands.

[0045] This invention provides an adaptive length control method for a hot-rolling tension reduction mill for seamless steel pipes. Based on the established steel pipe length calculation model, it further considers the current rolling process state and historical differences, calculates the influence of the tension coefficient on the rolling length under the current state, and realizes online dynamic control of the steel pipe length of the tension reduction mill.

[0046] Combination Figure 1 and Figure 2 As shown, the length adaptive control method of the present invention specifically includes the following steps:

[0047] 1. Using historical data, establish a calculation model for the length of steel pipes of the same steel type and specification;

[0048] The steel pipe length calculation model employs a multivariate nonlinear fitting method. This invention utilizes the XGBoost fitting algorithm, which adds a regularization term to mitigate overfitting and improve generalization ability. This algorithm is mature and can be directly implemented using Python software packages; therefore, it will not be discussed in detail here.

[0049] The key feature variables considered mainly include length-related data collected during on-site production: feeding length, steel pipe empty reduction length, steel pipe continuous rolling length, steel pipe tension reduction length, tension reduction adjustment coefficient, etc. In addition, temperature information such as billet heating furnace heating time, billet tapping temperature, and reheating furnace temperature can be added to improve model fitting accuracy. When modeling the actual data collected, outlier handling is also required to remove missing samples and samples whose data deviates from reasonable ranges, such as... Figure 3 As shown.

[0050] 2. Then, the material is tracked in real time through the process control system. If the current signal is the completion signal of the length measurement of the tension-reducing steel pipe, the model error and the average value of the historical tension-reducing coefficient are processed.

[0051] (1) When the deviation between the measured length and the theoretical length of the tension reduction is within the allowable error range, the model error is calculated using the following method:

[0052] ΔL new =α×ΔL old +(1-α)×(M new -C new )

[0053] Where, ΔL new This represents the model error value obtained after the current steel pipe rolling is completed;

[0054] ΔL old This represents the model error value obtained after the previous batch of steel pipes of the same type and specification was rolled out;

[0055] M new This indicates the current measured value of the tension reduction rolling length of the steel pipe;

[0056] C new This represents the calculated value of the current steel pipe tension reduction rolling length model, calculated using the XGBoost fitting model;

[0057] α represents the historical weighting coefficient of the model error;

[0058] It should be noted that the model error is also calculated separately for layers of the same steel type and specification.

[0059] Considering the differences in equipment status and process control corresponding to different production plans, the historical weighting coefficient α of the model error is taken to be a small value, generally between [0.2, 0.5].

[0060] (2) Processing of historical tension reduction coefficient average value

[0061] If the deviation between the measured tension reduction length and the theoretical tension reduction length is within the allowable error range, the control process is considered stable, and the historical average tension coefficient is updated. Otherwise, the historical average value is kept unchanged. The formula for calculating the historical average tension coefficient is as follows:

[0062] WTCA new =(WTCA) old ×N+wtca) / (N+1)

[0063] Among them, WTCA new This indicates the average tension coefficient corresponding to the current steel pipe specification after the current steel pipe is rolled out;

[0064] WTCAold This represents the average tension coefficient of the corresponding steel pipe specification before rolling.

[0065] wtca represents the tension coefficient currently used in steel pipes;

[0066] N represents the corresponding WTCA old Quantity of steel pipes rolled.

[0067] After the calculation is complete, use WTCA. new Update WTCA old The corresponding number of rolled steel pipes is updated to N+1;

[0068] WTCA old Used to set the tension coefficient of the first steel pipe in a batch.

[0069] 3. If the current signal is the reheat furnace tapping signal, then determine whether the steel grade or specification of the next steel pipe to be rolled and the previous rolled steel pipe of the tension reduction unit have changed, so as to realize the differentiated setting control of the tension reduction coefficient.

[0070] (1) If the steel type or specification of the next steel pipe to be rolled and the previous steel pipe to be rolled in the tension reduction unit change, the tension coefficient setting logic of the first steel pipe is adopted, that is, the historical average tension coefficient consistent with the current steel pipe specification is used for setting.

[0071] (2) If the steel type or specification of the next steel pipe to be rolled and the previous steel pipe to be rolled in the tension reduction unit have not changed, that is, the current steel pipe to be rolled is not the first one in the current batch, dynamic setting is adopted, and the tension coefficient of the next steel pipe to be rolled is calculated iteratively.

[0072] The iterative calculation method is as follows:

[0073] The tension coefficient wtca used in the previous steel pipe tension reduction rolling was the initial value wtca(1), which was substituted into the XGBoost regression model to calculate L(1);

[0074] If the target deviation ΔL(1)=L(1)+ΔL new -L Tar If the value is greater than 0, then wtca is increased by the minimum adjustment step size;

[0075] Conversely, the minimum adjustment step size is used to decrease wtca to obtain wtca(2), which is then substituted into XGBoost to obtain L(2).

[0076] During the iterative calculation, when ΔL(i) and ΔL(i-1) are reversed, the iterative calculation is completed, and the wtca corresponding to the target deviation > 0 is taken as the preset value of the tension reduction coefficient of the next steel pipe. In order to avoid incorrect adjustment caused by poor length measurement accuracy and model calculation accuracy, wtca is limited, and the deviation of wtca before and after does not exceed the preset threshold.

[0077] Example

[0078] This embodiment provides a length adaptive control method for a hot rolling mill unit for seamless steel pipes. Without loss of generality, it is illustrated using the first two steel pipes of a rolled billet as an example. Assume the batch information is as follows:

[0079] billet diameter Finished product outer diameter Finished wall thickness Continuous rolling outer diameter Continuous rolling wall thickness Feeding length Theoretical length of continuous rolling Zhang Jian Theoretical Length 178 73.03 5.6 151.5 6.75 3.95 31.702 81.62

[0080] The units for variables such as billet diameter, finished product outer diameter, finished product wall thickness, continuous rolling outer diameter, and continuous rolling wall thickness are all mm; the units for fields such as feeding length, continuous rolling theoretical length, and tension reduction theoretical length are all m.

[0081] 1. The theoretical length of continuous rolling is calculated based on the elongation rate:

[0082] Theoretical length of continuous rolling = (1 - oxidation loss rate) × feeding length × billet cross-sectional area ÷ continuous rolling tube cross-sectional area

[0083] = (1-0.01)×3.95×178×178÷(151.5×151.5-(151.5-6.75×2)×(151.5-6.75×2))

[0084] =31.702

[0085] The oxidation loss rate during the heating process is approximated here as 0.01.

[0086] Theoretical length of the rolled tube = (1 - rolling loss rate) × theoretical length of the continuously rolled tube × cross-sectional area of ​​the continuously rolled tube ÷ cross-sectional area of ​​the finished tube

[0087] = (1-0.005)×31.702×(151.5×151.5-(151.5-6.75×2)×(151.5-6.75×2))÷(73.03×73.03-(73.03-5.6×2)×(73.03-5.6×2))

[0088] =81.62

[0089] The rolling loss rate mainly consists of head and tail trimming and reheat furnace burn-off, which is approximated here as 0.005.

[0090] The initial information for the first two steel pipes before they enter the tension reduction unit is as follows:

[0091]

[0092] 2. In this embodiment, the XGBoost fitting algorithm is first used to establish the steel grade specification model with a sample size of 15394. In this embodiment, the independent variables considered include the fields: feeding length, empty reduction measurement length, continuous rolling measurement length, and tension reduction coefficient, and the dependent variable is the tension reduction measurement length.

[0093]

[0094] 15349 rows × 43 columns

[0095] The XGBoost algorithm includes not only model parameters but also manually tuned hyperparameters. These hyperparameters are not derived from the data, but they have a significant impact on the accuracy of the algorithm's predictions. Inappropriate hyperparameter selection leads to poor fitting results. Therefore, optimizing the hyperparameters can ensure the algorithm achieves optimal performance on the validation set. The relevant XGBoost hyperparameters and their ranges are shown in the table below.

[0096]

[0097] XGBoost is an ensemble algorithm based on gradient boosting trees. Its core components are the ensemble algorithm, weak estimators, and other processes. The ensemble algorithm corresponds to parameters `n_estimators`, `learning_rate`, and `subsample`; the weak estimator corresponds to parameters such as `max_depth`, `booster`, `gamma`, `min_child_weight`, `colsample_bytree`, `reg_alpha`, and `reg_lambda`. The shape of the validation curves on the training and validation sets is compared to determine whether the XGBoost prediction model fits the target dataset. By combining different hyperparameters, the model's hyperparameters are: `n_estimators` = 108, `learning_rate` = 0.1, and `max_depth` = 5. Using the default values ​​for the remaining parameters yields good prediction accuracy for the current sample. Figure 4 As shown.

[0098] 3. The process control system enables real-time material tracking, acquires real-time tracking signals, and performs corresponding processing.

[0099] In this embodiment, the continuous rolling measurement lengths of the first two steel pipes are 32.05m and 31.85m, respectively. When calculating the model error and statistically analyzing the historical tension reduction coefficient, the error range is set to 1% of the theoretical length, which satisfies the following formula:

[0100] |Measured tension reduction length - Theoretical tension reduction length| ≤ 0.01 × Theoretical tension reduction length

[0101] The corresponding tension reduction coefficient is included in the calculation of the historical tension reduction coefficient average.

[0102] Generally, a smaller error range setting is more conducive to improving the accuracy of subsequent steel pipe rolling length control. In this case, the steel grade and specification correspond to a rolling quantity of 15,394 steel pipes, the number of samples that meet the error range is 9,000, the current historical average tension coefficient is -70, and the current model error for calculating the tension reduction length is 0.58m.

[0103] (1) When the obtained signal is the reheat furnace tapping signal, determine whether the steel type or specification of the next steel pipe to be rolled and the previous rolled steel pipe of the tension reduction unit have changed. If they have changed, use the tension coefficient setting logic of the first steel pipe.

[0104] Therefore, when the obtained furnace exit signal is the first steel pipe of this rolling batch, the average tension coefficient of the most recent rolling batch corresponding to the steel grade and specification is taken as -70 for control, and the tension reduction coefficient of -70 is sent to the L1 system for tension reduction unit length control.

[0105] (2) When the obtained signal is the reheat furnace tapping signal, determine whether the steel type or specification of the next steel pipe to be rolled and the previous rolled steel pipe of the tension reduction unit have changed. If there is no change, the tension reduction coefficient adopts dynamic setting logic.

[0106] Therefore, when the second steel pipe of this batch is tapped from the reheating furnace, a dynamic setting logic will be used.

[0107] Substituting parameters such as feeding length 3.95m, empty reduction measurement length 10.949m, continuous rolling measurement length 31.85m, and tension coefficient -70 into the XGBoost regression model, the calculated length for the first tension reduction is L(1) = 81.15m. The judgment is as follows:

[0108] ΔL(1)=L(1)+ΔLnew-LTar=81.15+0.265-81.62=-0.205m

[0109] The current model error ΔLnew = 0.265, and the specific calculation is explained below; the target length is taken as the tension reduction theoretical length corresponding to the feeding length, which is 81.62.

[0110] ΔL(1)<0 indicates that the current tension coefficient may result in a shorter steel pipe length. It is necessary to reduce the tension coefficient and increase the tension between the stands to increase the rolling length of the steel pipe.

[0111] Without loss of generality, we assume here that the minimum adjustment amplitude of the tension coefficient is 1, then in the next iteration:

[0112] Tension coefficient wtca(2)=wtca(1)-1=-70-1=-71

[0113] Substituting the parameters such as feeding length 3.95, empty reduction measurement length 10.949, continuous rolling measurement length 31.85, and tension coefficient -71 into the XGBoost regression model, the calculated length for the second tension reduction is L(2) = 81.27m. The judgment is as follows:

[0114] ΔL(2)=L(2)+ΔLnew-LTar=81.27+0.265-81.62=-0.085<0

[0115] The complete iterative calculation process is shown in the table below:

[0116]

[0117] During the 4th iteration, the iteration calculation conditions are met, namely ΔL(3)<0 and ΔL(4)>0. Therefore, wtca(4)=-73 is sent to L1 as the tension coefficient to perform dynamic control of the current steel pipe tension reduction rolling.

[0118] (3) When the obtained signal is the length measurement completion signal after tension reduction rolling, the model error and the average value of the historical tension reduction coefficient are processed.

[0119] After the tension reduction and length measurement of the first steel pipe in the rolling batch is completed, the corresponding tension reduction length of the steel pipe is 82.05m. According to the tension coefficient setting of the first pipe, the tension coefficient wtca is -70. The current model error is known to be 0.58m. In this case, the historical weighting coefficient of model error α = 0.25 is taken.

[0120] First, determine whether the deviation between the current measured length and the theoretical length of the tension reduction is within the allowable error range.

[0121] |Measured length of tension reduction - Theoretical length of tension reduction| = |82.05 - 81.62| = 0.46m

[0122] 0.01 × Zhang's theoretical length = 0.01 × 81.62 = 0.8162m > 0.46m

[0123] If the conditions are met, process the model error and historical tension coefficient.

[0124] By inputting the information such as the feeding length of 3.95m, the air reduction measurement length of 11.09m, the continuous rolling measurement length of 32.05m, and the actual value of the tension reduction coefficient of -70 into the XGBoost fitting model, the current calculated length C of the steel pipe is obtained. new =81.89m.

[0125] ΔL new =α×ΔL old +(1-α)×(M new -C new )=0.25×0.58+(1-0.25)×(82.05-81.89)=0.265m

[0126] Processing of historical tension reduction coefficient average value:

[0127] WTCA new =(WTCA) old ×N+wtca) / (N+1)=(-70×9000-70) / (9000+1)=-70

[0128] Obviously, because the first steel pipe used the historical average tension coefficient, and the current steel pipe will not change after rolling, the updated historical tension coefficient WTCA is used. old =-70, quantity N=90001.

[0129] After the tension reduction and length measurement of the second steel pipe in the rolling batch is completed, the same processing procedure is adopted. The tension reduction measurement length of the second steel pipe is 81.78m, the tension coefficient wtca is -73, and the current model error obtained after the first steel pipe is rolled is 0.265m. In this case, the historical weighting coefficient of model error α = 0.25 is taken.

[0130] First, determine whether the deviation between the current measured length and the theoretical length of the tension reduction is within the allowable error range.

[0131] |Measured length of tension reduction - Theoretical length of tension reduction| = |81.78 - 81.62| = 0.16m

[0132] 0.01 × Zhang's theoretical length = 0.01 × 81.62 = 0.8162m > 0.16m

[0133] If the conditions are met, process the model error and historical tension coefficient.

[0134] ΔL new =α×ΔL old +(1-α)×(M new -C new )=0.25×0.265+0.75×(81.78-81.44)=0.321m

[0135] Processing of historical tension reduction coefficient average value:

[0136] WTCA new =(WTCA) old×N+wtca) / (N+1)=(-70×9001-73) / (9001+1)=-70

[0137] Additionally, please note:

[0138] 1. For steel grades with particularly large rolling volumes, the average tension coefficient tends to a stable value, and short-term tension adjustments will not significantly change it. Therefore, to better adapt to changes in production processes and equipment, the average tension coefficient can be calculated per rolling batch. Corresponding to the example above, before the first steel pipe is rolled, the average tension coefficient is -70, and the number of steel pipes is 9000. If calculated per rolling batch, before the first steel pipe is rolled, the average tension coefficient is -70, and the number of steel pipes is 0. The current rolling batch will restart counting when calculating the average tension coefficient.

[0139] 2. When calculating model error, the choice of the historical weighting coefficient α of the model error is also related to the accuracy of the tension reduction measurement length. Generally speaking, if the measurement accuracy is high, it is recommended to take a smaller value for α, and if the measurement accuracy is low, it is recommended to take a larger value for α. This will avoid frequent fluctuations in model correction due to measurement error.

[0140] 3. In this embodiment, the XGBoost fitting algorithm was used to establish the tensile length calculation model. In fact, as long as a certain fitting accuracy is met, it is feasible to use neural networks, multivariate nonlinear regression and other algorithms for modeling.

[0141] Those skilled in the art should recognize that the above embodiments are merely illustrative of the present invention and are not intended to limit the present invention. Any variations or modifications to the above embodiments that are within the spirit and essence of the present invention will fall within the scope of the claims of the present invention.

Claims

1. A length adaptive control method for a hot rolling tension reduction mill for seamless steel pipes, characterized in that: Using historical data, a calculation model for the length of steel pipes of the same steel type and specification was established; The process control system tracks the material in real time. If the current signal is the completion signal of the length measurement of the tension reduction steel pipe, the model error and the average value of the historical tension reduction coefficient are processed. If the current signal is the steel tapping signal of the reheat furnace, it is determined whether the steel grade or specification of the next steel pipe to be rolled and the previous rolled steel pipe of the tension reduction unit have changed, so as to realize the differentiated setting control of the tension reduction coefficient.

2. The length adaptive control method for a hot rolling tension reduction mill for seamless steel pipes according to claim 1, characterized in that: The historical data includes feeding length, steel pipe empty reduction length, steel pipe continuous rolling length, steel pipe tension reduction length, and tension reduction adjustment coefficient; and Heating time of billet heating furnace, billet exit temperature, and reheating furnace temperature.

3. The length adaptive control method for a seamless steel pipe hot rolling stretch reducing mill train according to claim 1, characterized by: The steel pipe length calculation model is modeled using a multivariate nonlinear fitting method. First, outlier handling is performed, removing samples with missing data and those whose data deviates from a reasonable range.

4. The length adaptive control method for a seamless steel pipe hot rolling stretch reducing mill train according to claim 1, characterized by, The processing of model error and the average historical tension coefficient includes the following: 1) When the deviation between the measured length and the theoretical length of the tension reduction is within the allowable error range, the model error is calculated using the following method: ΔL new = a x ΔL old + (1 - a) x (M new - C new ) wherein ΔL new represents the model error value obtained after the current steel pipe rolling is completed; ΔL old represents the model error value obtained after the rolling of the previous branch of steel pipes of the same steel grade and specification is completed; M new This indicates the current measured value of the tension reduction rolling length of the steel pipe; C new This represents the calculated value of the current steel pipe tension reduction rolling length model; α represents the historical weighting coefficient of the model error; 2) Processing of historical tension reduction coefficient average value If the deviation between the measured tension reduction length and the theoretical tension reduction length is within the allowable error range, the control process is considered stable, and the historical average tension coefficient is updated. Otherwise, the historical average value is kept unchanged. The formula for calculating the historical average tension coefficient is as follows: WTCA new = (WTCA old × N + wtca) / (N + 1) Wherein, WTCA new represents the updated average tension coefficient corresponding to the current steel pipe specification after the current steel pipe rolling is completed; WTCA old represents the average value of the tension coefficient corresponding to the steel pipe specification before the current steel pipe rolling; wtca represents the tension coefficient currently used in steel pipes; N denotes the WTCA corresponding to old Steel pipe rolling quantity. After the calculation is completed, the WTCA new The WTCA is updated old The number of steel pipe rolling corresponding to the updated N+1; WTCA old The tension factor setting for the first pass of the rolling batch is implemented.

5. The length adaptive control method for a seamless steel pipe hot rolling stretch reducing mill train according to claim 4, characterized in that: The current steel pipe tension reduction rolling length model calculation value C new Calculated using an XGBoost fitted model; The historical error weighting coefficient α of the model takes values ​​of [0.2, 0.5].

6. The length adaptive control method for a seamless steel pipe hot rolling stretch reducing mill train according to claim 1, characterized by: If the steel type or specification of the next steel pipe to be rolled and the previous steel pipe to be rolled in the tension reduction unit change, the tension coefficient setting logic of the first steel pipe is adopted, that is, the historical average tension coefficient consistent with the current steel pipe specification is used for setting.

7. The length adaptive control method for a seamless steel pipe hot rolling stretch reducing mill train according to claim 1, characterized by: If the steel type or specification of the next steel pipe to be rolled in the tension reduction mill unit has not changed from that of the previous rolled steel pipe, that is, the current steel pipe to be rolled is not the first pipe of the current rolling batch, then dynamic setting is adopted, and the tension coefficient of the next steel pipe to be rolled is calculated iteratively.

8. The length adaptive control method for a hot rolling tension reduction mill for seamless steel pipes according to claim 7, characterized in that, The iterative calculation method is as follows: The tension coefficient wtca used in the previous steel pipe tension reduction rolling was the initial value wtca(1), which was substituted into the XGBoost regression model to calculate L(1); If the target deviation ΔL(1)=L(1)+ΔL new -L Tar If the value is greater than 0, then wtca is increased by the minimum adjustment step size; Conversely, the minimum adjustment step size is used to decrease wtca to obtain wtca(2), which is then substituted into XGBoost to obtain L(2).

9. The length adaptive control method for a hot rolling tension reduction mill for seamless steel pipes according to claim 8, characterized in that: During the iterative calculation, when ΔL(i) and ΔL(i-1) are reversed, the iterative calculation is completed, and the wtca corresponding to the target deviation > 0 is taken as the preset value of the tension reduction coefficient of the next steel pipe.

Images