Thick plate rough rolling speed control method and system based on current prediction of main transmission motor

Abstract

The invention provides a thick plate rough rolling speed control method and system based on main transmission motor current prediction. The thick plate rough rolling speed control method comprises the steps that rough rolling model setting data and rolling actual performance data are collected; data cleaning and equalization processing are carried out on the collected data, and a thick plate rough rolling main transmission motor current prediction model based on the random forest is constructed; setting data are input into the current prediction model in real time through rolling pass completion event triggering, and a current prediction result is obtained; the rough rolling model setting speed is adjusted according to the current prediction result; and the adjusted rough rolling model set speed is issued to a rolling line automatic control system to be executed. According to the method, the current of the thick plate rough rolling main transmission motor is predicted by utilizing a machine learning algorithm, and the rough rolling speed is optimized and adjusted on the basis of current prediction; and on the premise of ensuring stable operation of current of the main transmission motor, the rolling speed can still meet the requirement of rough rolling fast-paced production.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology for thick plate roughing, and more specifically, to a method and system for controlling the rolling speed of thick plate roughing based on the prediction of the main drive motor current. Background Technology

[0002] With the current capacity continuously increasing, the roughing rolling process can no longer fully match the current rolling capacity. This is specifically reflected in the current overload of the main drive motor during the roughing rolling process. According to statistics, the current of the main drive motor of the roughing rolling on a certain production line is greater than 90% 551 times per month. The current on-site handling measures for the current overload of the main drive motor of the roughing rolling are: monitoring the current value of the main drive motor of the roughing rolling at the L1 level of basic automation. When the current value is overloaded, a pop-up window prompts the screen. After the prompt appears, the on-site operator manually modifies the current rolling speed of the steel plate, reduces the speed, and slows down the rolling rhythm until the current overload prompt disappears. This handling measure has certain shortcomings: (1) the proportion of manual intervention is high and the automation rate is low; (2) the adjustment of the rolling speed relies only on manual experience and is not supported by data. If the adjustment is small, the current overload phenomenon cannot be alleviated, and if the adjustment is large, it will seriously affect the rolling rhythm; (3) this handling measure is a post-event handling and cannot predict the current value of the main drive motor, so it cannot significantly reduce the current overload of the main drive motor of the roughing rolling.

[0003] By analyzing the roughing rolling process, the basic situation of overcurrent in the main drive motor of the roughing mill was summarized: the overcurrent problem of the main drive motor of the roughing mill mainly occurs in the constant power speed regulation mode of the motor. In this mode, the motor operates at the rated power. If the main drive motor operates at a large rolling speed and a large rolling load, the main drive motor control system needs to continuously increase the current until the control requirements are met. This eventually leads to excessive current and causes the main drive motor to trip.

[0004] Based on the operating principle of the main drive motor, reducing the rolling reduction or the rolling speed can reduce the current of the main drive motor. However, since reducing the rolling reduction will lead to an increase in the number of rolling passes, the impact on the rolling rhythm is greater than the impact of reducing the rolling speed. Furthermore, the complexity of manually adjusting the rolling reduction is greater than the complexity of adjusting the rolling speed. Therefore, on-site operators generally reduce the rolling speed to adjust the current of the main drive motor.

[0005] The original roughing rolling speed model uses manually set maximum rolling speed, bite speed, ejection speed, acceleration, and steel plate length as input conditions to calculate the roughing rolling speed. Based on the steel plate length, the model can accelerate the roughing rolling speed to the manually set maximum rolling speed level. However, the model lacks consideration for the current factor of the main drive motor of the roughing mill, so it still issues a large rolling speed under heavy load, resulting in motor current overload.

[0006] Patent document CN114682631B discloses a method for adjusting the current load of a stand in a cold rolling mill. This method controls the rolling speed of the cold rolling mill to 490 m / min-510 m / min; acquires the current load percentage of each stand in the cold rolling mill in real time; identifies stands whose current load percentage exceeds a threshold as target stands; obtains the correspondence between inter-stand tension and current load, and determines the adjacent stands of the target stand based on this correspondence; if the number of stands of the target stand is less than a threshold, the tension between the target stand and the adjacent stands is adjusted to regulate the current load of the target stand. Although this method involves the control of both current load and rolling speed, the control of the two is not directly related, nor does it involve the application of intelligent algorithms such as machine learning.

[0007] Patent document CN103056170A discloses a continuous rolling synchronous control system for a wide lead strip production line. The system includes motors, several roughing mills, and a finishing mill located after the last roughing mill. The control system also includes a PLC control system and pin-type microswitches. Each motor individually controls one roughing mill or one finishing mill, and all motors are connected to the PLC control system. The pin-type microswitches are installed on all roughing mills except the last one to detect whether lead strip accumulates between the roughing mills and provide feedback to the PLC control system. Although this system controls the rolling speed by reading the roughing mill current load, it does not predict the roughing mill current load, nor does it involve the application of intelligent algorithms such as machine learning.

[0008] Patent application CN103817157A discloses a variable acceleration rolling control system for a roughing mill, including a predicted rolling force calculation module, a load judgment module, a rolling force comparison module, and a mill acceleration adjustment module. The predicted rolling force calculation module calculates the predicted rolling force data for the workpiece. The load judgment module determines whether the roughing mill is under load and transmits the signal to the mill acceleration adjustment module. The rolling force comparison module receives the predicted rolling force data calculated by the predicted rolling force calculation module and compares it with a pre-set threshold value for rolling force data, transmitting the comparison result to the mill acceleration adjustment module. The mill acceleration adjustment module adjusts the maximum acceleration of the mill based on the received signal indicating whether the roughing mill is under load and the comparison result between the predicted rolling force data for the workpiece and the pre-set threshold value for rolling force data. However, this patent cannot completely solve the existing technical problems and cannot meet the needs of this invention. Summary of the Invention

[0009] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for controlling the rolling speed of thick plate roughing based on the prediction of the main drive motor current.

[0010] The method for controlling the rolling speed of thick plate roughing based on the prediction of the main drive motor current, provided by the present invention, includes:

[0011] Step S1: Collect roughing rolling model setting data and actual rolling performance data;

[0012] Step S2: Perform data cleaning and equalization on the collected data, construct a current prediction model for the main drive motor of the thick plate roughing mill based on random forest, and form a model file;

[0013] Step S3: Trigger the rolling pass completion event to input the set data into the current prediction model in real time and obtain the current prediction result;

[0014] Step S4: Adjust the setting speed of the roughing rolling model according to the current prediction results. If the preset control standard is not met, adjust it cyclically until it is met.

[0015] Step S5: Send the adjusted roughing rolling model speed setting to the automatic control system of the rolling line for execution.

[0016] Preferably, in step S1:

[0017] The roughing rolling model settings include: roughing slab dimensions, target slab dimensions, slab entry temperature, target exit temperature, target rolling end temperature, average exit temperature, reduction per pass, reduction rate per pass, rolling force per pass, torque per pass, rolling speed per pass, exit dimensions per pass, rolling stage type, controlled rolling type, rolling strategy type, number of passes, steel grade, heating mode, roll radius, and furnace time.

[0018] The rolling performance data includes: measured rolling force, measured torque, measured rolling speed, measured current of the main drive motor of the roughing mill, measured current phase of the main drive motor of the roughing mill, measured speed of the main drive motor of the roughing mill, and overload alarm signal of the current of the roughing mill.

[0019] The roughing rolling model setting data and rolling performance data are right-joined to form a new data table using slab number, BASID number, number of passes, and pass time as conditions.

[0020] Preferably, in step S2:

[0021] Data cleaning includes: deleting missing data; deleting model setting data that exceeds the maximum rolling force, maximum rolling torque, and maximum rolling speed of the process control model; and deleting abnormal rolling data where the actual number of passes is less than the set number of passes.

[0022] The equalization process includes: undersampling, which deletes normal current samples based on a set torque benchmark; and oversampling, which takes each abnormal sample point X0 as the center, selects k nearest neighbor sample points as the basis, randomly selects neighboring points Xk, multiplies the difference by a threshold α in the range of [0, 1], and synthesizes a new overload current sample Xnew: Xnew=X0+α(X0-Xk), until the number of overload and normal samples are equal.

[0023] Random search is used to optimize hyperparameters, including the number of decision trees, the maximum depth of decision trees, the minimum number of samples required for internal node re-splitting, the minimum number of samples for leaf nodes, the preset number of optimal segmentation features, and the sampling method.

[0024] During the model training phase, multiple subsets are randomly sampled from the training set, and each subset corresponds to a decision tree.

[0025] During the model prediction phase, each decision tree makes an independent prediction and then outputs the result through a voting mechanism.

[0026] Preferably, in step S3:

[0027] The rolling pass completion event is triggered by the increase in the number of completed passes within the automatic control system of the rolling line;

[0028] The data is configured to be transmitted to the current prediction model via TCP / IP messages.

[0029] Preferably, in step S4:

[0030] If the current prediction result is overload, the rolling model setting speed is reduced by a fixed step size;

[0031] After adjusting the speed, re-enter the model to predict the current, and repeat the cycle until the current is normal or the maximum number of cycles is reached.

[0032] During the adjustment process, the rolling stage type, slab width, set reduction amount, set rolling torque, set rolling force, rolling target width, target furnace exit temperature, and average furnace exit temperature should remain unchanged.

[0033] The thick plate roughing rolling speed control system based on main drive motor current prediction provided by the present invention includes:

[0034] Module M1: Collects roughing rolling model setting data and actual rolling performance data;

[0035] Module M2: Performs data cleaning and equalization on the collected data, constructs a current prediction model for the main drive motor of the thick plate roughing mill based on random forest, and forms a model file;

[0036] Module M3: Triggered by the rolling pass completion event, the set data is input into the current prediction model in real time to obtain the current prediction result;

[0037] Module M4: Adjusts the setting speed of the roughing rolling model based on the current prediction results. If the preset control standard is not met, it will be adjusted cyclically until it is met.

[0038] Module M5: Sends the adjusted roughing rolling model speed setting to the automatic control system of the rolling line for execution.

[0039] Preferably, in module M1:

[0040] The roughing rolling model settings include: roughing slab dimensions, target slab dimensions, slab entry temperature, target exit temperature, target rolling end temperature, average exit temperature, reduction per pass, reduction rate per pass, rolling force per pass, torque per pass, rolling speed per pass, exit dimensions per pass, rolling stage type, controlled rolling type, rolling strategy type, number of passes, steel grade, heating mode, roll radius, and furnace time.

[0041] The rolling performance data includes: measured rolling force, measured torque, measured rolling speed, measured current of the main drive motor of the roughing mill, measured current phase of the main drive motor of the roughing mill, measured speed of the main drive motor of the roughing mill, and overload alarm signal of the current of the roughing mill.

[0042] The roughing rolling model setting data and rolling performance data are right-joined to form a new data table using slab number, BASID number, number of passes, and pass time as conditions.

[0043] Preferably, in module M2:

[0044] Data cleaning includes: deleting missing data; deleting model setting data that exceeds the maximum rolling force, maximum rolling torque, and maximum rolling speed of the process control model; and deleting abnormal rolling data where the actual number of passes is less than the set number of passes.

[0045] The equalization process includes: undersampling, which deletes normal current samples based on a set torque benchmark; and oversampling, which takes each abnormal sample point X0 as the center, selects k nearest neighbor sample points as the basis, randomly selects neighboring points Xk, multiplies the difference by a threshold α in the range of [0, 1], and synthesizes a new overload current sample Xnew: Xnew=X0+α(X0-Xk), until the number of overload and normal samples are equal.

[0046] Random search is used to optimize hyperparameters, including the number of decision trees, the maximum depth of decision trees, the minimum number of samples required for internal node re-splitting, the minimum number of samples for leaf nodes, the preset number of optimal segmentation features, and the sampling method.

[0047] During the model training phase, multiple subsets are randomly sampled from the training set, and each subset corresponds to a decision tree.

[0048] During the model prediction phase, each decision tree makes an independent prediction and then outputs the result through a voting mechanism.

[0049] Preferably, in module M3:

[0050] The rolling pass completion event is triggered by the increase in the number of completed passes within the automatic control system of the rolling line;

[0051] The data is configured to be transmitted to the current prediction model via TCP / IP messages.

[0052] Preferably, in module M4:

[0053] If the current prediction result is overload, the rolling model setting speed is reduced by a fixed step size;

[0054] After adjusting the speed, re-enter the model to predict the current, and repeat the cycle until the current is normal or the maximum number of cycles is reached.

[0055] During the adjustment process, the rolling stage type, slab width, set reduction amount, set rolling torque, set rolling force, rolling target width, target furnace exit temperature, and average furnace exit temperature should remain unchanged.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] (1) This invention integrates machine learning algorithms with the original system framework, which not only maintains the advantages of the original framework, but also introduces the current trend of machine algorithms to realize intelligent prediction of the current of the main drive motor of the roughing mill, realize automatic speed control, the model prediction accuracy reaches more than 90%, the automation rate reaches more than 90%, and promotes the intelligent production of the production line.

[0058] (2) The control model of this invention intelligently analyzes data collected in real time, such as rolling torque, rolling force, rolling speed, reduction, pass exit width, furnace exit temperature, planned furnace exit temperature, rolling target width, and main drive motor current of thick plate roughing. It analyzes the influence of different parameters on the main drive motor current and calculates the optimal rolling speed setting value of thick plate roughing in real time. By automatically executing the optimal rolling speed calculated by the control model, it replaces the existing manual control measures, ultimately reducing the probability of overload of the main drive motor current in roughing, without affecting the rolling rhythm on site. Attached Figure Description

[0059] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0060] Figure 1 This is a system architecture diagram of an automatic control method for the rolling speed of thick plate roughing based on the prediction of the main drive motor current.

[0061] Figure 2 This is a flowchart of an automatic control method for the rolling speed of thick plate roughing based on the prediction of the main drive motor current. Detailed Implementation

[0062] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0063] Example 1

[0064] The purpose of this invention is to provide a method for controlling the rolling speed of thick plate roughing. By using machine learning algorithms, the current of the main drive motor of the thick plate roughing is predicted, and the rolling speed is optimized and adjusted based on the current prediction. Under the premise of ensuring stable operation of the main drive motor current, the rolling speed can still meet the fast-paced production of roughing.

[0065] To achieve the aforementioned technical objectives, this invention focuses on adjusting the current of the main drive motor in thick plate roughing by controlling the rolling speed. It proposes an automatic control method for the rolling speed of thick plate roughing based on the prediction of the main drive motor current. Figure 2 This includes the following steps:

[0066] Step S1: Collect the setting data of the roughing rolling model and the actual rolling data for subsequent analysis and modeling work.

[0067] Step S2: After data collection, the collected data is cleaned and equalized, and a current prediction model for the main drive motor of the thick plate roughing mill based on random forest is constructed, forming a model file. The model adopts the random forest algorithm structure, integrating multiple decision trees. During the training phase, the model randomly samples the training set, dividing it into multiple subsets, each subset corresponding to a decision tree. During the prediction phase, each decision tree independently predicts the input data. Finally, the model uses a voting mechanism to take the prediction result of the decision tree with the most votes as the model's output.

[0068] Step S3: After completing the model construction step, when the rolling pass is completed, the number of completed rolling passes inside the automatic control system of the rolling line increases, thereby generating a rolling pass completion event. This further triggers the communication module to input the set data into the current prediction model of the main drive motor of the thick plate roughing mill in real time via TCP / IP message. The model uses each decision tree inside to independently predict the set data, and then uses a voting mechanism to select the prediction result of the decision tree with the most votes, and finally obtains and outputs the current prediction result.

[0069] Step S4: Based on the current prediction result from the model, if the current prediction result indicates overload, reduce the rolling model setting speed by a fixed step size. After the rolling model setting speed is adjusted, re-enter the model to perform current prediction and check whether the current prediction result after adjusting the rolling model setting speed is normal. If it is still overload, continue to reduce the rolling model setting speed by a fixed step size until the current prediction result is normal.

[0070] Step S5: After the rolling model setting speed is adjusted, it is sent to the basic automation system of the rolling line for execution.

[0071] Furthermore, in step S1, the data collected for the roughing rolling model setting includes, but is not limited to, the dimensions of the roughing slab (length, width, thickness), the target dimensions of the roughing slab (length, width, thickness), the slab entry temperature, the target exit temperature, the target rolling end temperature, the average exit temperature, the reduction per pass, the reduction rate per pass, the rolling force per pass, the torque per pass, the rolling speed per pass, the exit dimensions (length, width, thickness) per pass, the rolling stage type, the controlled rolling type, the rolling strategy type, the number of passes, the steel grade, the heating mode, the roll radius, and the time in the furnace.

[0072] Rolling performance data includes, but is not limited to, the actual rolling force, torque, speed, current of the main drive motor, phase of the main drive motor, speed of the main drive motor, and overload alarm signal of the current in the roughing mill.

[0073] The two types of data are right-joined using slab number, BASID number, number of passes, and pass time as conditions to form a new data table.

[0074] Furthermore, in step S2, a current prediction model for the main drive motor of the thick plate roughing mill is constructed based on random forest, including:

[0075] First, collect relevant factors affecting the current of the main drive motor of the thick plate roughing mill, namely, historical data involving the thick plate roughing mill model and the thick plate roughing mill main drive motor system.

[0076] Secondly, the current data of the main drive motor of the thick plate roughing mill is preprocessed: missing value data is deleted; abnormal business logic data is deleted, that is, model setting data that are greater than the maximum rolling force, maximum rolling torque, and maximum rolling speed of the process control model are deleted; abnormal rolling data (except for rolling abnormalities caused by current overload) are deleted.

[0077] Next, after the data cleaning step is completed, the current data of the main drive motor of the thick plate roughing mill is subjected to equalization processing.

[0078] In one possible implementation, the ratio of the overload current samples to the normal current samples is approximately 1:146, indicating an extremely unbalanced sample distribution. Therefore, to prevent model overfitting, a combination of synthetic minority class oversampling and undersampling is used to process the current samples of the roughing mill main drive motor, addressing this sample imbalance: (1) Undersampling: Based on historical data, a set torque benchmark is selected. Under this benchmark, there will be no overload of the roughing mill main drive motor current, and the normal current samples under the benchmark are deleted; (2) Oversampling: By inputting overload current samples, new overload current samples are generated iteratively based on the k nearest neighbor sample points (with the shortest Euclidean distance) selected from each abnormal sample point X0, until the number of overload current samples and the number of normal current samples are equal.

[0079] Then, using the random forest algorithm as the model foundation, the equalized current data of the main drive motor of the thick plate roughing mill was used as training data. A hyperparameter optimization algorithm based on random search was employed to optimize hyperparameters in the random forest algorithm, including the number of decision trees, maximum depth of decision trees, minimum number of samples required for internal node re-partitioning, minimum number of samples for leaf nodes, preset optimal number of splitting features, and sampling method. During this process, each type of hyperparameter is freely combined to form several hyperparameter combinations. These hyperparameter combinations are then randomly and independently sampled, and the model is trained using the hyperparameters from these combinations. Finally, the optimal hyperparameter combination for the model training result is output.

[0080] In one possible implementation, a set of hyperparameters for the random forest model is set up, in which each type of hyperparameter is freely combined. A hyperparameter optimization algorithm based on random search is used to randomly and independently sample the hyperparameters, thereby continuously adjusting the hyperparameters of the random forest model until they are optimized.

[0081] After that, the quality of the adjusted model is evaluated using test set data, the model is serialized into binary data and saved to the model file. The above completes the construction of the current prediction model for the main drive motor of the thick plate roughing mill based on random forest.

[0082] Furthermore, in step S3, during actual production, when the current rolling pass is completed, the rolling pass completion trigger event is used to input the rolling model setting data for the next pass into the current prediction model of the main drive motor of the thick plate roughing mill via electronic message transmission. After the model calculation, the current prediction result is obtained.

[0083] Furthermore, in step S4, using the critical safety value of the main drive motor current in the roughing mill as the standard, while keeping the model setting parameters such as rolling stage type, slab width, set reduction amount, set rolling torque, set rolling force, rolling target width, target furnace exit temperature, and average furnace exit temperature unchanged, the model recalculates the rolling setting speed proportionally. After the rolling setting speed calculation is completed, the model replaces the original rolling setting speed with the optimized rolling setting speed and predicts whether the current is overloaded again. If it is still overloaded, the rolling setting speed is optimized until the conditions of predicting that the current is not overloaded or the maximum number of cycles are met.

[0084] Furthermore, in step S5, the optimized rolling setting speed is sent to the rolling line process control system via electronic message transmission, and the rolling line process control system sends the rolling setting speed to the rolling line basic automation system via rolling setting message.

[0085] like Figure 1 The entire architecture consists of a rolling speed automatic control model server, a rolling line process control system, and a rolling line basic automation system. The rolling speed automatic control model server has four major modules: data acquisition, data processing, data analysis, and communication. The three systems interact with each other via TCP / IP messages, and the messages of the three systems are sent and received by their respective communication modules.

[0086] Example 2

[0087] Example 2 is a preferred example of Example 1.

[0088] This invention provides an automatic control method for the rolling speed of thick plate roughing based on the prediction of the main drive motor current. This method is used to automatically control the rolling speed in the roughing system of a thick plate production line.

[0089] Step S1: Construct a current prediction model for the main drive motor of the thick plate roughing mill:

[0090] First, collect the roughing model setting data and the field measured data, including rolling stage type, slab width, set reduction, set rolling torque, set rolling force, set rolling speed, rolling target width, target furnace exit temperature, average furnace exit temperature, measured rolling force, measured torque, and other data.

[0091] Secondly, the data was preprocessed to remove missing values ​​and model setting data where the rolling force was greater than the model's maximum rolling force of 63MN, the rolling torque was greater than the maximum rolling torque of 3400KN / m, and the maximum rolling speed was greater than 5m / s. Rolling anomaly data where the actual number of passes was less than the set number of passes (except for rolling anomalies caused by current overload) were also removed.

[0092] Next, the data is equalized: undersampling is implemented to delete normal current samples with torque Torque < 2500 KN / m; oversampling is implemented to input overload current samples, and based on the k nearest neighbor sample points (shortest Euclidean distance) selected for each abnormal sample point X0, a neighboring point Xk is randomly selected and the difference is multiplied by a threshold α in the range [0, 1] to synthesize a new overload current sample Xnew: Xnew = X0 + α(X0 - Xk). The above process is repeated until the number of overload current samples and the number of normal current samples are equal.

[0093] Then, using the random forest algorithm as the model basis, the equalized data is divided into training and test sets in a 7:3 ratio. The hyperparameters in the random forest algorithm are optimized using a random search hyperparameter optimization algorithm. The optimization results involve the number of decision trees, the maximum depth of the decision trees, the minimum number of samples required for internal node re-splitting, the minimum number of samples for leaf nodes, the preset number of optimal segmentation features, and the sampling method.

[0094] Following this, the model quality was evaluated using test set data, the current prediction model for the main drive motor of the thick plate roughing mill was completed, the model was serialized, and output as a model file.

[0095] Step S2: Collect the roughing model setting data and roughing measured data in real time, and perform a right join with slab number, BASID number, number of passes and pass time as conditions to form a new data table and store it in the database.

[0096] Step S3: When the slab is rolled to the 6th pass in the roughing mill, the following parameters are input into the current prediction model of the main drive motor of the thick plate roughing mill: number of passes = 6, slab width = 2.157m, furnace exit temperature = 1161℃, target furnace exit temperature = 1150℃, target rolling width = 2.563m, target rolling thickness = 2.02mm, rolling reduction = 25.583mm, rolling force = 42.821MN, rolling torque = 2662KN / m, pass exit width = 2.556m, and initial rolling speed = 2.7m / s. The prediction result is current overload.

[0097] Step S4: Based on the current prediction results, the rolling speed is optimized and the calculated rolling speed is 2.656 m / s. The model predicts the current of the main drive motor again, and the prediction result shows that the current is normal.

[0098] Step S5: Send the optimized rolling speed setting to the basic automation system of the rolling line via electronic message.

[0099] The main differences between this embodiment and the traditional method are:

[0100]

[0101]

[0102] Advantages of this embodiment compared to traditional production methods:

[0103] This invention Traditional methods Advantage 1 Machine learning model calculates set values Human estimation Advantage 2 Less human intervention, higher automation rate High degree of human intervention and low rate of automation Advantage 3 Optimal rolling speed The rolling speed was not optimal.

[0104] Example 3

[0105] The present invention also provides a rolling speed control system for thick plate roughing based on main drive motor current prediction. The rolling speed control system for thick plate roughing based on main drive motor current prediction can be implemented by executing the process steps of the rolling speed control method for thick plate roughing based on main drive motor current prediction. That is, those skilled in the art can understand the rolling speed control method for thick plate roughing based on main drive motor current prediction as a preferred embodiment of the rolling speed control system for thick plate roughing based on main drive motor current prediction.

[0106] The system includes: Module M1: collecting roughing rolling model setting data and actual rolling data; Module M2: cleaning and equalizing the collected data, constructing a current prediction model for the main drive motor of the thick plate roughing rolling mill based on random forest, and forming a model file; Module M3: triggering the rolling pass completion event to input the setting data into the current prediction model in real time and obtain the current prediction result; Module M4: adjusting the roughing rolling model setting speed according to the current prediction result, and cyclically adjusting until the preset control standard is met if it is not met; Module M5: sending the adjusted roughing rolling model setting speed to the automatic control system of the rolling line for execution.

[0107] In module M1: the roughing rolling model setting data includes: roughing slab incoming dimensions, roughing slab rolling target dimensions, slab furnace entry temperature, slab target furnace exit temperature, slab target rolling end temperature, slab average furnace exit temperature, reduction per pass, reduction rate per pass, rolling force per pass, torque per pass, rolling speed per pass, exit dimensions per pass, rolling stage type, controlled rolling type, rolling strategy type, number of passes, steel grade, heating mode, roll radius, and furnace time; the rolling performance data includes: roughing rolling measured rolling force, roughing rolling measured torque, roughing rolling measured rolling speed, roughing rolling main drive motor measured current, roughing rolling main drive motor measured current phase, roughing rolling main drive motor measured speed, and roughing rolling current overload alarm signal; the roughing rolling model setting data and rolling performance data are right-joined using slab number, BASID number, number of passes, and pass time as conditions to form a new data table.

[0108] In module M2: data cleaning includes: deleting missing data; deleting model setting data that are greater than the maximum rolling force, maximum rolling torque, and maximum rolling speed of the process control model; deleting abnormal rolling data where the actual number of passes is less than the set number of passes; equalization processing includes: undersampling, deleting normal current samples based on the set torque benchmark; oversampling, taking each abnormal sample point X0 as the center, selecting k nearest neighbor sample points as the basis, randomly selecting neighboring points Xk, multiplying the difference by a threshold α in the range of [0, 1], and synthesizing a new overload current sample Xnew: Xnew = X0 + α(X0 - Xk), until the number of overload and normal samples are equal; using random search to optimize hyperparameters, including the number of decision trees, the maximum depth of decision trees, the minimum number of samples required for internal node repartition, the minimum number of samples for leaf nodes, the preset number of optimal segmentation features, and the sampling method; during the model training phase, multiple subsets are randomly sampled from the training set, and each subset corresponds to a decision tree; during the model prediction phase, each decision tree makes independent predictions and outputs the results through a voting mechanism.

[0109] In module M3: the rolling pass completion event is triggered by the increase of the number of completed passes in the automatic control system of the rolling line; the setting data is transmitted to the current prediction model via TCP / IP message.

[0110] In module M4: If the current prediction result is overload, the rolling model setting speed is reduced by a fixed step size; after the speed is adjusted, the model prediction current is re-entered, and the cycle is repeated until the current is normal or the maximum number of cycles is reached; during the adjustment process, the rolling stage type, slab width, set reduction amount, set rolling torque, set rolling force, rolling target width, target furnace exit temperature and average furnace exit temperature remain unchanged.

[0111] Those skilled in the art will understand that, in addition to implementing the system, apparatus, and their modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and their modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.

[0112] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for controlling the rolling speed of thick plate roughing based on the prediction of the main drive motor current, characterized in that, include: Step S1: Collect roughing rolling model setting data and actual rolling performance data; Step S2: Perform data cleaning and equalization on the collected data, construct a current prediction model for the main drive motor of the thick plate roughing mill based on random forest, and form a model file; Step S3: Trigger the rolling pass completion event to input the set data into the current prediction model in real time and obtain the current prediction result; Step S4: Adjust the setting speed of the roughing rolling model according to the current prediction results. If the preset control standard is not met, adjust it cyclically until it is met. Step S5: Send the adjusted roughing rolling model speed setting to the automatic control system of the rolling line for execution.

2. The method for controlling the rolling speed of thick plate roughing based on the prediction of the main drive motor current as described in claim 1, characterized in that, In step S1: The roughing rolling model settings include: roughing slab dimensions, target slab dimensions, slab entry temperature, target exit temperature, target rolling end temperature, average exit temperature, reduction per pass, reduction rate per pass, rolling force per pass, torque per pass, rolling speed per pass, exit dimensions per pass, rolling stage type, controlled rolling type, rolling strategy type, number of passes, steel grade, heating mode, roll radius, and furnace time. The rolling performance data includes: measured rolling force, measured torque, measured rolling speed, measured current of the main drive motor of the roughing mill, measured current phase of the main drive motor of the roughing mill, measured speed of the main drive motor of the roughing mill, and overload alarm signal of the current of the roughing mill. The roughing rolling model setting data and rolling performance data are right-joined to form a new data table using slab number, BASID number, number of passes, and pass time as conditions.

3. The method for controlling the rolling speed of thick plate roughing based on the prediction of the main drive motor current as described in claim 1, characterized in that, In step S2: Data cleaning includes: deleting missing data; deleting model setting data that exceeds the maximum rolling force, maximum rolling torque, and maximum rolling speed of the process control model; and deleting abnormal rolling data where the actual number of passes is less than the set number of passes. The equalization process includes: undersampling, which deletes normal current samples based on a set torque benchmark; and oversampling, which takes each abnormal sample point X0 as the center, selects k nearest neighbor sample points as the basis, randomly selects neighboring points Xk, multiplies the difference by a threshold α in the range of [0, 1], and synthesizes a new overload current sample Xnew: Xnew=X0+α(X0-Xk), until the number of overload and normal samples are equal. Random search is used to optimize hyperparameters, including the number of decision trees, the maximum depth of decision trees, the minimum number of samples required for internal node re-splitting, the minimum number of samples for leaf nodes, the preset number of optimal segmentation features, and the sampling method. During the model training phase, multiple subsets are randomly sampled from the training set, and each subset corresponds to a decision tree. During the model prediction phase, each decision tree makes an independent prediction and then outputs the result through a voting mechanism.

4. The method for controlling the rolling speed of thick plate roughing based on the prediction of the main drive motor current as described in claim 1, characterized in that, In step S3: The rolling pass completion event is triggered by the increase in the number of completed passes within the automatic control system of the rolling line; The data is configured to be transmitted to the current prediction model via TCP / IP messages.

5. The method for controlling the rolling speed of thick plate roughing based on the prediction of the main drive motor current according to claim 1, characterized in that, In step S4: If the current prediction result is overload, the rolling model setting speed is reduced by a fixed step size; After adjusting the speed, re-enter the model to predict the current, and repeat the cycle until the current is normal or the maximum number of cycles is reached. During the adjustment process, the rolling stage type, slab width, set reduction amount, set rolling torque, set rolling force, rolling target width, target furnace exit temperature, and average furnace exit temperature should remain unchanged.

6. A rolling speed control system for thick plate roughing based on main drive motor current prediction, characterized in that, include: Module M1: Collects roughing rolling model setting data and actual rolling performance data; Module M2: Performs data cleaning and equalization on the collected data, constructs a current prediction model for the main drive motor of the thick plate roughing mill based on random forest, and forms a model file; Module M3: Triggered by the rolling pass completion event, the set data is input into the current prediction model in real time to obtain the current prediction result; Module M4: Adjusts the setting speed of the roughing rolling model based on the current prediction results. If the preset control standard is not met, it will be adjusted cyclically until it is met. Module M5: Sends the adjusted roughing rolling model speed setting to the automatic control system of the rolling line for execution.

7. The rolling speed control system for thick plate roughing based on main drive motor current prediction according to claim 6, characterized in that, In module M1: The roughing rolling model settings include: roughing slab dimensions, target slab dimensions, slab entry temperature, target exit temperature, target rolling end temperature, average exit temperature, reduction per pass, reduction rate per pass, rolling force per pass, torque per pass, rolling speed per pass, exit dimensions per pass, rolling stage type, controlled rolling type, rolling strategy type, number of passes, steel grade, heating mode, roll radius, and furnace time. The rolling performance data includes: measured rolling force, measured torque, measured rolling speed, measured current of the main drive motor of the roughing mill, measured current phase of the main drive motor of the roughing mill, measured speed of the main drive motor of the roughing mill, and overload alarm signal of the current of the roughing mill. The roughing rolling model setting data and rolling performance data are right-joined to form a new data table using slab number, BASID number, number of passes, and pass time as conditions.

8. The rolling speed control system for thick plate roughing based on main drive motor current prediction according to claim 6, characterized in that, In module M2: Data cleaning includes: deleting missing data; deleting model setting data that exceeds the maximum rolling force, maximum rolling torque, and maximum rolling speed of the process control model; and deleting abnormal rolling data where the actual number of passes is less than the set number of passes. The equalization process includes: undersampling, which deletes normal current samples based on a set torque benchmark; and oversampling, which takes each abnormal sample point X0 as the center, selects k nearest neighbor sample points as the basis, randomly selects neighboring points Xk, multiplies the difference by a threshold α in the range of [0, 1], and synthesizes a new overload current sample Xnew: Xnew=X0+α(X0-Xk), until the number of overload and normal samples are equal. Random search is used to optimize hyperparameters, including the number of decision trees, the maximum depth of decision trees, the minimum number of samples required for internal node re-splitting, the minimum number of samples for leaf nodes, the preset number of optimal segmentation features, and the sampling method. During the model training phase, multiple subsets are randomly sampled from the training set, and each subset corresponds to a decision tree. During the model prediction phase, each decision tree makes an independent prediction and then outputs the result through a voting mechanism.

9. The rolling speed control system for thick plate roughing rolling based on main drive motor current prediction according to claim 6, characterized in that, In module M3: The rolling pass completion event is triggered by the increase in the number of completed passes within the automatic control system of the rolling line; The data is configured to be transmitted to the current prediction model via TCP / IP messages.

10. The rolling speed control system for thick plate roughing based on main drive motor current prediction according to claim 6, characterized in that, In module M4: If the current prediction result is overload, the rolling model setting speed is reduced by a fixed step size; After adjusting the speed, re-enter the model to predict the current, and repeat the cycle until the current is normal or the maximum number of cycles is reached. During the adjustment process, the rolling stage type, slab width, set reduction amount, set rolling torque, set rolling force, rolling target width, target furnace exit temperature, and average furnace exit temperature should remain unchanged.

Images