Continuous casting machine fan-shaped section pulling-straightening force prediction method based on MAML-LSTM

Abstract

The invention discloses a continuous casting machine fan-shaped section pulling-straightening force prediction method based on MAML-LSTM. The method comprises the steps that 1, the pulling-straightening force of a continuous casting machine fan-shaped section is analyzed; step 2, according to an analysis result of the step 1, constructing a fan-shaped section pulling-straightening force sample data set; 3, constructing a fan-shaped section tension-straightening force prediction model based on MAML-LSTM, and training and testing the prediction model based on the fan-shaped section tension-straightening force sample data set in the step 2; and 4, inputting a fan-shaped section pulling-straightening force data set acquired in real time in the actual production process into the trained pulling-straightening force prediction model in the step 3, and performing real-time prediction on the pulling-straightening force of the fan-shaped section of the continuous casting machine. According to the method, the influence factors influencing the withdrawal and straightening force of the fan-shaped section of the continuous casting machine are analyzed, the fan-shaped section withdrawal and straightening force data set is constructed, the continuous casting machine fan-shaped section withdrawal and straightening force prediction model based on the MAML-LSTM is established, the problem that withdrawal and straightening force samples are insufficient under the condition that the fan-shaped section of the continuous casting machine is abnormal or breaks down can be solved, and the accuracy of predicting the withdrawal and straightening force of the fan-shaped section in advance is improved.

Description

Technical Field

[0001] The invention relates to the technical field of prediction of straightening force of a continuous casting machine segment, and in particular to a method for predicting straightening force of a continuous casting machine segment based on MAML-LSTM. Background Art

[0002] During the continuous casting production process, the straightening force of the continuous casting machine's fan-shaped segment is manifested as the billet drawing resistance that needs to be overcome when the billet passes through the straightening roller array. The straightening force time series can effectively characterize the fault characteristics of the fan-shaped segment. When the fan-shaped segment fails, the billet drawing resistance will increase, which in turn affects the load of the motor. This change will be reflected in the straightening force. The straightening force of the fan-shaped segment is the basis for the identification and prediction of the fan-shaped segment faults of the continuous casting machine. By monitoring the straightening force of the fan-shaped segment, analyzing the performance characteristics of the straightening force under the normal state and fault state of the fan-shaped segment, and exploring the correlation between the straightening force and the fan-shaped segment fault, the efficiency and accuracy of the fan-shaped segment fault troubleshooting can be improved.

[0003] In actual production, continuous casting machines usually maintain normal operation for a long time, the frequency of failure is low, and the number of failure samples is relatively small, which leads to insufficient samples of the straightening force of the fan-shaped segment under abnormal or faulty conditions, thereby limiting the quality and timeliness of the extraction of fault features by the fan-shaped segment fault prediction model. In order to accurately predict the straightening force of the fan-shaped segment of the continuous casting machine in advance and solve the problem of insufficient prediction model training and poor prediction performance caused by the scarcity of fault samples of the fan-shaped segment of the continuous casting machine, a straightening force prediction model based on the long short-term memory network (LSTM) is established by analyzing the change law of the straightening force of the fan-shaped segment of the continuous casting machine, and the model-agnostic metalearning algorithm (MAML) is used to optimize the parameters, thereby realizing the prediction of the straightening force of the fan-shaped segment. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art. The present invention proposes a method for predicting the straightening force of a continuous casting machine segment based on MAML-LSTM, which is used to improve the generalization ability of the segment straightening force prediction model under different working conditions and realize the straightening force prediction in the case of a small sample of segment faults.

[0005] The present invention provides a method for predicting the straightening force of a continuous casting machine segment based on MAML-LSTM, comprising the following steps:

[0006] Step 1: Analyze the straightening force of the continuous casting machine segment, including the force analysis of the segment driving roller, the straightening force analysis of the continuous casting machine flow, and the change law analysis of the straightening force of the segment, and determine the factors affecting the straightening force of the segment;

[0007] Step 2: Based on the factors affecting the straightening force of the fan segment, the time series data of n groups of fan segments are used to construct a sample data set of the straightening force of the fan segment, and the original data in the data set are normalized;

[0008] Step 3: Build a fan-segment straightening force prediction model based on MAML-LSTM, and use the fan-segment straightening force sample data set obtained in step 2 to train and test the prediction model;

[0009] Step 4: Input the fan-shaped segment straightening force data set collected in real time during the actual production process into the trained fan-shaped segment straightening force prediction model to realize the real-time prediction of the fan-shaped segment straightening force of the continuous casting machine.

[0010] Furthermore, the specific operation steps of step 1 are:

[0011] Step 1.1: Perform force analysis on the driving roller of the fan segment, and the straightening force of the fan segment can be obtained as follows:

[0012] F refi =T D × i / r i

[0013] Where:

[0014]

[0015] T i =F i (μ0r i +μ1)

[0016] Among them, F refi is the tension straightening force of sector i; T D is the theoretical torque of the motor; z i is the transmission ratio of sector i; r i F is the driving roller radius of sector i; i is the resultant force of the billet reaction force and the hydraulic cylinder on the driving roller of sector i; μ0 is the sliding friction coefficient between the roller and the billet; μ1 is the rolling friction coefficient between the billet and the roller;

[0017] Step 1.2: Perform force analysis on the n driving rollers of all the segments in each stream of the continuous casting machine, and calculate the tension and straightening force F of each segment i. refi , the total straightening force F of the continuous casting machine flow is obtained ref for:

[0018] F refi =M refi × i / r i

[0019] Fref =F ref1 +F ref2 +…+F refn

[0020] Among them, M refi is the measured torque of the motor;

[0021] Step 1.3: Analyze the changes in the straightening force data of the fan segment at each stage, and determine the factors affecting the straightening force of the fan segment based on the changing law of the straightening force of the fan segment.

[0022] Furthermore, the factors affecting the straightening force of the fan-shaped segment determined in step 1.3 include the billet drawing speed, the billet temperature, the billet surface roughness, the continuous casting slab water nozzle flow rate, the continuous casting slab water nozzle water pressure, the crystallizer vibration amplitude, and the crystallizer vibration frequency.

[0023] Furthermore, the specific operation steps of step 3 are:

[0024] Step 3.1: Construct meta-training set and meta-test set;

[0025] Step 3.2: Construct a MAML-LSTM-based straightening force prediction model based on the meta-training set and the meta-test set.

[0026] Furthermore, the steps of step 3.1 include:

[0027] Step 3.1.1: Randomly extract the fan-shaped segment straightening force sample data set established in step 2 in a ratio of 4:1 as the meta-training set and meta-test set respectively;

[0028] Step 3.1.2: Set the length of each extracted data to l, and randomly extract two sets of continuous data from the meta-training set and meta-test set as the support set and query set respectively;

[0029] Step 3.1.3: The support set and query set obtained each time together form a task, and finally m and n tasks are generated in the meta-training set and meta-test set respectively.

[0030] Further, step 3.2 includes the following steps:

[0031] Step 3.2.1: Establish an LSTM straightening force prediction model, and perform meta-training and meta-testing on the straightening force prediction model;

[0032] Step 3.2.2: Using MAE, RMSE, and R 2 The straightening force prediction model is evaluated on the meta-test set;

[0033] Step 3.2.3: The tensile force prediction model that has passed the evaluation is used as the tensile force prediction model of MAML-LSTM.

[0034] Furthermore, the specific steps of step 3.2.1 are:

[0035] Step 3.2.1.1: In the inner loop, for each task in the meta-training set, first use the support set of the task to train the LSTM model, and then use the following formula to obtain the optimal parameters for the task and gradient

[0036]

[0037] Among them, α represents the learning rate of the inner loop, L represents the function loss, and LSTM θ represents an LSTM model with randomly initialized parameters θ, Indicates the calculation of the gradient of θ to the loss function L;

[0038] Step 3.2.1.2: In the outer loop, use the parameters obtained in the inner loop on the query set of the current meta-training set task Perform outer gradient descent, and then update the optimal initialization parameter θ of the LSTM model using the following formula:

[0039]

[0040] in, Indicates that the parameter is LSTM model, β represents the learning rate of the outer loop;

[0041] Step 3.2.1.3: Input the support set of each task in the meta-test set into the straightening force prediction model trained in step 3.2.2.2, and fine-tune the training parameters; then test the query set of each task in the meta-test set.

[0042] Therefore, the present invention adopts the above-mentioned MAML-LSTM-based continuous casting machine sector straightening force prediction method, which has the following beneficial effects:

[0043] The present invention constructs a sample data set of straightening force of a sector segment by analyzing the influence of the straightening force of a sector segment of a continuous casting machine and the force analysis. On this basis, a MAML algorithm is introduced into a LSTM model, and the parameters of the LSTM model are optimized by the MAML algorithm, so that the LSTM can learn the characteristics of the straightening force data under different working conditions during the process of gradient descent and parameter updating, and a straightening force prediction model of a sector segment is constructed, thereby improving the prediction accuracy of the straightening force data.

[0044] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Analyze the roller force;

[0046] Figure 2 This is a graph showing changes in straightening force data within a cycle;

[0047] Figure 3 A process for building a fan-segment straightening force prediction model based on MAML-LSTM.

[0048] Figure 4 Comparison of the optimization effects of MAML, TSO, and HPO on the LSTM model. DETAILED DESCRIPTION

[0049] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art make various changes or modifications to the present invention, and these equivalent forms also fall within the scope limited by the claims attached to the application.

[0050] The present invention proposes a method for predicting the straightening force of a continuous casting machine segment based on MAML-LSTM, which comprises the following steps:

[0051] Step 1: Sector straightening force analysis;

[0052] Step 1 includes the following specific steps:

[0053] Step 1.1: Force analysis of the sector drive roller;

[0054] Assuming that the driving roller moves in an ideal pure rolling manner, the force on the roller is as follows: Figure 1 As shown in the figure. During the billet drawing process, the driving roller is in close contact with the billet surface and moves forward under the action of the straightening force. The straightening force is mainly generated by the static sliding friction between the billet and the driving roller.

[0055] Combination Figure 1 It can be seen that the driving roller is subjected to forces in both the horizontal and vertical directions, and is also affected by torque, the direction of which is consistent with the moving direction of the billet. By taking the torque at the center of the driving roller, its driving torque T i It can be expressed as:

[0056] T i =F i (μ0r i +μ1)(1)

[0057] Among them, F iis the resultant force of the billet reaction force and the hydraulic cylinder on the driving roller of sector i; μ0 is the sliding friction coefficient between the roller and the billet, μ0=0.3; μ1 is the rolling friction coefficient between the billet and the roller, μ1=0.003m; r i is the driving roller radius of sector i.

[0058] In order to ensure the force balance of the driving roller, the motor torque T D With T i The relationship between must satisfy the following expression:

[0059]

[0060] Among them, T D is the theoretical torque of the motor; z is the transmission ratio; η is the reducer efficiency.

[0061] The straightening force of sector i is expressed as:

[0062] F refi =T D × i / r i (3)

[0063] Among them, F refi is the tension straightening force of sector i; z i is the transmission ratio of sector i.

[0064] In order to make the continuous casting process smooth, the straightening force required by each sector is related to the continuous casting production conditions, such as slab thickness, slab width, continuous casting roller radius, etc., so as to determine the motor torque of the sector. By collecting the motor torque in real time, the actual straightening force of each sector can be obtained, which provides a data source for analyzing the status of the sector of the continuous casting machine.

[0065] Step 1.2: Continuous casting machine straightening force analysis;

[0066] Assuming that each stream of the continuous casting machine contains n (n=14) segments, and each segment is driven by a variable frequency motor, the straightening force of segment i can actually be obtained through the motor torque. The specific calculation formula is as follows:

[0067] F refi =M refi × i / r i (4)

[0068] Among them, M refi is the measured torque of the motor.

[0069] From the above formula, we can know that to obtain the accurate value of the straightening force of the fan segment, the key is to measure the torque of the motor in real time. By installing a torque sensor in the motor system, the torque on the motor shaft can be directly measured.

[0070] The total straightening force of the continuous casting machine flow is equal to the sum of the straightening forces of each fan-shaped segment. If there are n driving rollers, the total straightening force is calculated as follows:

[0071] F ref =F ref1 +F ref2 +…+F refn (5)

[0072] Among them, F ref Represents the total straightening force of the continuous casting machine stream.

[0073] Step 1.3: Analysis of the variation law of the straightening force of the fan segment;

[0074] The stability of the straightening force is an important indicator of the continuous casting machine status. It is mainly affected by factors such as the drawing speed, casting temperature, casting surface roughness, slab water nozzle flow, slab water nozzle water pressure, mold vibration amplitude, and mold vibration frequency. These factors can be used as parameter indicators of the straightening force sample data set.

[0075] At different stages of the continuous casting process, the straightening force of the fan segment will show different forms. The continuous casting process can be divided into four key stages: start casting, change tundish, normal casting and stop casting. The straightening force will show different trends and characteristics at different stages. Figure 2 As shown in the figure, it shows the change of the straightening force data of the fan segment within one cycle. The time period 0-t1 is the beginning of casting, and the straightening force gradually increases from 0 to the normal range; the time period t1-t2 and the time period t3-t4 are the stages of changing the tundish, and the straightening force surges to a certain range and then drops sharply to 0; the time period t2-t3 and the time period t4-t5 are the normal casting stages, and the straightening force of the continuous casting machine will show a relatively stable and continuous change trend, and the value of the straightening force will fluctuate within a relatively stable range. This is because the continuous casting machine is in a stable working state, and all components and parameters are in the normal working range, and the motion state and shape correction of the billet in the continuous casting machine are also in a stable state; the time period t5-t6 is the stage of stopping casting, and the straightening force drops rapidly to near 0.

[0076] Step 2: Constructing the sample dataset of the fan-shaped segment straightening force;

[0077] The main sample parameters that affect the straightening force of the fan segment that can be collected in the actual production process of the continuous casting machine are shown in Table 1. The straightening force sample parameter values ​​collected within time t are used to construct the fan segment straightening force sample data set.

[0078] Table 1 Sector straightening force sample parameters

[0079]

[0080] In order to eliminate the influence of different dimensions and distribution characteristics of data, the original vector needs to be normalized. Max-Min normalization is a data scaling technology based on linear transformation. Through this method, the value of each feature is mapped to the range of 0 to 1, so as to achieve the unification of data of different indicators at the same scale. The normalization formula is shown in (6):

[0081]

[0082] Among them, min(x) is the minimum value of each parameter (p0~p6) of the sample; max(x) is the maximum value of each parameter (p0~p6) of the sample.

[0083] Step 3: Build a fan-segment straightening force prediction model based on MAML-LSTM, and use the fan-segment straightening force sample data set for training and testing;

[0084] In order to achieve accurate prediction of the straightening force data of the fan-shaped segment of the continuous casting machine, the present invention takes the LSTM model as the basic model and introduces the MAML algorithm to establish the MAML-LSTM model, so that the LSTM can learn the characteristics of the straightening force data under different working conditions during the gradient descent and parameter updating process, which helps the model to better adapt to various complex and changing working conditions, thereby improving the prediction accuracy of the straightening force data. Figure 3 The process of establishing a straightening force prediction model based on MAML-LSTM includes the following steps:

[0085] Step 3.1: Construction of training task set and test task set;

[0086] The MAML learning object is data that satisfies the independent and identically distributed conditions randomly extracted from the data set. The data of each task is subdivided into a support set and a query set. In order to ensure the continuity of the data and not destroy the relevance of the data, a section of continuous data is randomly extracted to form a task sample. In MAML, the learning data is presented in the form of task distribution. The task distribution generation process of the tension prediction model is as follows:

[0087] Step 3.1.1: Preparation of the tension force dataset.

[0088] Based on the sample parameters of the tension straightening force in step 2, the time series data sets of n groups of fan segments are used as candidate data sources;

[0089] Step 3.1.2: Allocation of the tension force dataset.

[0090] The meta-training set and meta-test set are randomly selected from the candidate data sources in a ratio of 4:1.

[0091] Step 3.1.3: Task sample extraction.

[0092] The length of each extracted data is set to 40. Two sets of continuous data are randomly extracted from the meta-training set as the support set and the query set, which together form a task. A total of 800 tasks (T i ); Generate 200 tasks in the meta-test set in the same way.

[0093] Step 3.2: Establish a fan-segment straightening force prediction model based on MAML-LSTM.

[0094] Step 3.2.1: Meta-training of the straightening force prediction model.

[0095] Step 3.2.1.1: Establish an LSTM straightening force prediction model and optimize the initialization parameter θ of the LSTM model using the MAML algorithm.

[0096] The MAML algorithm repeatedly performs inner and outer loop training, allowing the model to quickly adapt to new tasks and have good generalization capabilities. This inner loop improvement method can help the model learn common features that adapt to different tasks, so that it can converge and adapt faster when faced with new tasks.

[0097] Step 3.2.1.1.1: In the inner loop, for each task T in the meta-training set i (i≤800), use the support set of this task to train the LSTM model, and then use formula (7) to calculate the optimal parameters for this task. With gradient

[0098]

[0099] Among them, α represents the inner loop learning rate, which determines the learning speed each time it adapts to a new task; L represents the function loss, that is, the difference between the model prediction and the actual output; LSTM θ It is an LSTM model with randomly initialized parameters θ, which is used to process sequence data; It represents the calculation of the gradient of θ to the loss function L, which is used to update the model parameters.

[0100] Step 3.2.1.1.2: In the outer loop, in the current task T of the meta-training set i On the query set, using the inner loop to get Perform outer gradient descent and use formula (8) to update the optimal initialization parameter θ of the LSTM model;

[0101]

[0102] in, Indicates that the parameter is For the LSTM model, β is the outer loop learning rate.

[0103] Step 3.2.2: Element test of the straightening force prediction model;

[0104] Each task T in the meta-test set in step 3.1.3 i The support set of (i≤200) is input into the fan-segment MAML-LSTM straightening force prediction model trained in step 3.2.1 for testing, and the training parameters are fine-tuned at the same time. The training parameters include the gradient descent learning rates α and β of the inner and outer loops in the MAML algorithm, and the random initialization parameter θ of the LSTM model; then the query set of each task in the meta-test set is input into the new parameter model for performance verification testing, and the accuracy of the query set is calculated.

[0105] Step 3.2.3: Evaluation of the tension prediction model;

[0106] The mean absolute error (MAE), root mean square error (RMSE) and coefficient of determination (R-Square, R 2 ) is used as the evaluation index of the tension prediction model, and the tension prediction model is evaluated on the meta-test set. MAE represents the difference between the predicted value and the true value, and RMSE can keep the error result and the input data at the same level, which is convenient for observing the deviation between the predicted value and the true value. 2 Used to evaluate the fit between the predicted value and the true value, that is, the strength of the linear relationship between the predicted value and the true value, MAE, RMSE and R 2 The calculation formulas are:

[0107]

[0108] In the formula, y i , are the actual sample value, predicted value, and actual average value, respectively.

[0109] Among the model evaluation indicators, the smaller the calculation results of MAE and RMSE, the higher the prediction model accuracy. 2 The evaluation range is [0, 1], R 2 The larger it is, the better the fitting effect of the prediction model is.

[0110] Step 4: Real-time prediction of the straightening force of the fan segment;

[0111] The fan-shaped segment straightening force data collected during the actual production process of the continuous casting machine are used to construct a fan-shaped segment straightening force sample data set according to step 2, and it is input into the fan-shaped segment straightening force prediction model based on MAML-LSTM. The model outputs the predicted value of the straightening force of the corresponding fan-shaped segment in real time.

[0112] Example

[0113] Taking a SMS continuous casting machine of a steel plant as the object, the variation law of the straightening force of the fan-shaped segment of the continuous casting machine was analyzed, a sample of the straightening force data set of the fan-shaped segment was constructed, and the prediction model of the straightening force of the fan-shaped segment based on MAML-LSTM was verified. The experimental steps are:

[0114] Step 1: Construct a sample dataset of fan-shaped segment straightening force.

[0115] Since the magnitudes of the data parameters collected in actual production are different, in order to ensure that the neural network can be balanced when receiving the input of these different features and avoid the problem of gradient explosion during model training due to excessive weights of some feature vectors, the input sample data is normalized using formula (5). Some data after sample normalization are shown in Table 2.

[0116]

[0117] Step 2: Prediction of fan-shaped segment straightening force based on MAML-LSTM.

[0118] Step 2.1: Parameter setting of MAML-LSTM straightening force prediction model.

[0119] The MAML algorithm includes two hyperparameters: the inner loop learning rate is set to 0.04, and the outer loop learning rate is set to 0.001. The model is debugged according to the prediction results, and the final model parameter settings are: the input layer contains 5 nodes, the number of hidden layers is 3, the output layer is set to 1 node, and the ReLU activation function is used. The batch size of each training is set to 32, the number of training rounds is 300, the inner loop performs 1 gradient update, and the outer loop performs 5 gradient updates.

[0120] Step 2.2: Construction of training task set and test task set;

[0121] The sample data set in step 1 is randomly selected in a ratio of 4:1 as the meta-training set and meta-test set respectively.

[0122] Step 2.3: Meta-training and meta-testing of the straightening force prediction model;

[0123] The meta-training set and test set in step 2.2 are respectively input into the MAML-LSTM straightening force prediction model initialized in step 2.1 to perform training, learning and testing of the prediction model.

[0124] Step 3: Prediction model performance evaluation.

[0125] In order to verify the optimization effect of MAML under small sample data, it is compared with the Tuna Swarm Optimization (TSO) and Hunter-Prey Optimization (HPO). Figure 4 As shown in the figure, the prediction errors of different models change during the iterative training process. The error of the MAML-LSTM prediction model drops rapidly at the beginning of the iteration and stabilizes after about 60 rounds of iterative learning. In comparison, the TSO-LSTM model and the HPO-LSTM model are trained more slowly, and stabilize at about 110 and 150 rounds respectively.

[0126] The comparison of the accuracy and stability evaluation results of the MAML-LSTM, TSO-LSTM and HPO-LSTM straightening force prediction models is shown in Table 3. From the data analysis, it can be seen that the MAE and RMSE of MAML-LSTM are the smallest, indicating that the error between the predicted value and the true value of MAML-LSTM is the smallest, and the R 2 The largest value indicates that the prediction result of MAML-LSTM has the best fitting effect.

[0127]

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for predicting the straightening force of the continuous casting machine segment based on MAML-LSTM. It is characterized in that The following steps are involved: Step 1: Analyze the straightening force of the continuous casting machine segment, including the force analysis of the segment driving roller, the straightening force analysis of the continuous casting machine flow, and the change law analysis of the straightening force of the segment, and determine the factors affecting the straightening force of the segment; Step 2: Based on the factors affecting the straightening force of the fan segment, the time series data of n groups of fan segments are used to construct a sample data set of the straightening force of the fan segment, and the original data in the data set are normalized; Step 3: Build a fan-segment straightening force prediction model based on MAML-LSTM, and use the fan-segment straightening force sample data set obtained in step 2 to train and test the prediction model; Step 4: Input the fan-shaped segment straightening force data set collected in real time during the actual production process into the trained fan-shaped segment straightening force prediction model to realize the real-time prediction of the fan-shaped segment straightening force of the continuous casting machine.

2. A method for predicting the straightening force of a continuous casting machine segment based on MAML-LSTM as claimed in claim 1, It is characterized in that The specific steps of step 1 are: Step 1.1: Perform force analysis on the driving roller of the fan segment, and the straightening force of the fan segment can be obtained as follows: F refi =T D ×z i / r i Where: T i =F i (m 0 r i +m 1 ) Among them, F refi is the tension straightening force of sector i; T D is the theoretical torque of the motor; z i is the transmission ratio of sector i; r i F is the driving roller radius of sector i; i is the resultant force of the billet reaction force on the driving roller of sector i and the force in the hydraulic cylinder; μ 0 is the sliding friction coefficient between the roller and the billet; μ 1 is the rolling friction coefficient between the billet and the roller; Step 1.2: Perform force analysis on the n driving rollers of all the segments in each stream of the continuous casting machine, and calculate the tension and straightening force F of each segment i. refi , the total straightening force F of the continuous casting machine flow is obtained ref for: F refi =M refi ×z i / r i F ref =F ref1 +F ref2 +…+F refn Among them, M refi is the measured torque of the motor; Step 1.3: Analyze the changes in the straightening force data of the fan segment at each stage, and determine the factors affecting the straightening force of the fan segment based on the changing law of the straightening force of the fan segment.

3. A method for predicting the straightening force of a continuous casting machine segment based on MAML-LSTM as claimed in claim 2, It is characterized in that The factors affecting the straightening force of the fan-shaped segment determined in step 1.3 include the billet drawing speed, the billet temperature, the billet surface roughness, the continuous casting slab water nozzle flow rate, the continuous casting slab water nozzle water pressure, the crystallizer vibration amplitude, and the crystallizer vibration frequency.

4. A method for predicting the straightening force of a continuous casting machine segment based on MAML-LSTM as claimed in claim 3, It is characterized in that The specific steps of step 3 are as follows: Step 3.1: Construct meta-training set and meta-test set; Step 3.2: Construct a MAML-LSTM-based straightening force prediction model based on the meta-training set and the meta-test set.

5. A method for predicting the straightening force of a continuous casting machine segment based on MAML-LSTM as claimed in claim 4, It is characterized in that The steps of step 3.1 include: Step 3.1.1: Randomly extract the fan-shaped segment straightening force sample data set established in step 2 in a ratio of 4:1 as the meta-training set and meta-test set respectively; Step 3.1.2: Set the length of each extracted data to l, and randomly extract two sets of continuous data from the meta-training set and meta-test set as the support set and query set respectively; Step 3.1.3: The support set and query set obtained each time together form a task, and finally m and n tasks are generated in the meta-training set and meta-test set respectively.

6. A method for predicting the straightening force of a continuous casting machine segment based on MAML-LSTM as claimed in claim 5, It is characterized in that Step 3.2 includes the following steps: Step 3.2.1: Establish an LSTM straightening force prediction model, and perform meta-training and meta-testing on the straightening force prediction model; Step 3.2.2: Using MAE, RMSE, and R 2 The straightening force prediction model is evaluated on the meta-test set; Step 3.2.3: The tensile force prediction model that has passed the evaluation is used as the tensile force prediction model of MAML-LSTM.

7. A method for predicting the straightening force of a continuous casting machine segment based on MAML-LSTM as claimed in claim 6, It is characterized in that The specific steps of step 3.2.1 are: Step 3.2.1.1: In the inner loop, for each task in the meta-training set, first use the support set of the task to train the LSTM model, and then use the following formula to obtain the optimal parameters for the task and gradient Among them, α represents the learning rate of the inner loop, L represents the function loss, and LSTM θ represents an LSTM model with randomly initialized parameters θ, Indicates the calculation of the gradient of θ to the loss function L; Step 3.2.1.2: In the outer loop, use the parameters obtained in the inner loop on the query set of the current meta-training set task Perform outer gradient descent, and then update the optimal initialization parameter θ of the LSTM model using the following formula: in, Indicates that the parameter is LSTM model, β represents the learning rate of the outer loop; Step 3.2.1.3: Input the support set of each task in the meta-test set into the straightening force prediction model trained in step 3.2.2.2, and fine-tune the training parameters; then test the query set of each task in the meta-test set.

Images