A Fast Calculation Method for Molten Steel Level in Converter Based on Evolutionary Learning

Abstract

This invention provides a rapid calculation method for molten steel level in a converter based on evolutionary learning, relating to the field of converter smelting technology in the steel industry. The method first constructs a physical modeling-based calculation model for molten steel level, calculating the corresponding molten steel level for different molten steel volumes and converter tilt angles. Then, the data on molten steel level heights for different molten steel volumes and converter tilt angles is saved as a dataset. Next, a rapid calculation model for molten steel level based on evolutionary learning is constructed. Finally, based on the constructed rapid calculation model, the molten steel level for different molten steel volumes and converter tilt angles is quickly calculated. This method can provide real-time output of molten steel level heights for different molten steel volumes and converter tilt angles to on-site operators, allowing them to flexibly adjust the converter tilt angle according to actual operating conditions and control the optimal distance between the molten steel level and the converter opening.

Description

Technical Field

[0001] This invention relates to the field of converter smelting technology in the iron and steel industry, and in particular to a method for rapidly calculating the height of molten steel in a converter based on evolutionary learning. Background Technology

[0002] As a crucial step in the steelmaking process, accurately measuring the molten steel level in the converter is essential for determining the size of the reaction space and adjusting the working position of the top-blown oxygen lance as needed. The position of the oxygen lance during converter blowing is a key factor in controlling slag formation and achieving optimal blowing performance. An improper oxygen lance position can easily cause foamy slag overflow and molten metal splashing, even leading to lance burnout and affecting its lifespan. Real-time measurement of the molten steel level is also crucial for controlling the optimal tilting angle of the converter during tapping, thereby suppressing slag flow into the ladle, improving the purity of the molten steel, and reducing slag erosion of the taphole and sliding nozzle lining. A schematic diagram of the converter molten steel tapping process is shown below. Figure 1 As shown, excessive slag during tapping not only increases the amount of sulfur and phosphorus reversion in the molten steel, and increases the amount and processing time of synthetic slag in subsequent refining processes, but also increases inclusions in the steel, affecting the quality of steel billets and the service life of refractory materials such as ladles.

[0003] Currently, there are three main methods for measuring the molten steel level in converters: The first method uses an oxygen lance. The process involves inserting a long iron rod into the oxygen lance nozzle, lowering the lance to the slag-splashing position so that part of the iron rod enters the molten steel, then lifting the lance and removing the rod. The mark left by the high-temperature molten steel on the rod is measured, and the molten steel level is determined using a calculation formula. The second method uses an auxiliary lance to measure the molten steel level. The basic principle is that the initial position of the auxiliary lance is known as the clearance height from the bottom of the converter. The first method involves a gun with uniform lowering and raising speeds. When the sub-gun probe is lowered into the molten steel at a uniform speed, it generates sudden switching signals such as temperature and electromotive force. The programmable logic controller (PLC) connected to the sub-gun counts these switching signals and, combined with the gun speed and clearance height, calculates the molten steel level. The second method directly uses sensors to measure the molten steel level. Its basic principle is to use specific sensors combined with a drive device and a programmable logic controller (PLC) to process and calculate signals from different measurement positions, thereby indirectly obtaining the molten steel level.

[0004] While existing methods for measuring the height of molten steel in converters can all achieve this, they all have some shortcomings. For example, the method using an oxygen lance requires inserting a long iron rod and then manually measuring the length of the molten steel mark, making the operation cumbersome. For instance, the method proposed in Chinese patent CN105274280 A requires a measuring rod welded with a long iron nail and a measuring device consisting of a conical wooden plug. The conical wooden plug is used to fix the long iron nail on the measuring rod, and then the measuring device is fixed at the oxygen lance nozzle to achieve the measurement of the molten steel level. The method using a secondary lance requires a specific high-temperature sensor probe and a programmable logic controller (PLC), resulting in a complex measurement system and high equipment costs. For example, the molten steel level measurement system proposed in Chinese Patent CN103290165 A not only includes a secondary lance instrument, a secondary lance PLC, a secondary lance height encoder, an oxygen lance height encoder, and a converter PLC, but also adds a liquid level computer, an operation computer, and a display screen computer; the molten steel level measurement system proposed in Chinese Patent CN203034044 U not only has a secondary lance and a programmable logic controller (PLC), but also includes a compensation wire connected to the secondary lance, a signal processor connected to the compensation wire, and other devices. The method of directly using sensors to measure the molten steel level and the method of using secondary lance measurement have the same drawbacks. They not only require specific measuring sensors and programmable logic controllers (PLCs), but also require dedicated drive devices for the sensors, resulting in complex system composition and high equipment installation and maintenance costs. For example, Chinese patent CN202246735 U proposes an intelligent detection device for the liquid level of molten steel in a converter. This device includes many components such as sensors, moving motors, sliding racks, waveguides, programmable logic controllers (PLCs), and computers. The measuring device in the method proposed by Chinese patent CN104388623 A also includes a driving device consisting of a stirring head sensor, a programmable logic controller (PLC), and a lifting motor. The system structure is equally complex. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for rapid calculation of the height of molten steel in a converter based on evolutionary learning, which addresses the shortcomings of the prior art.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for rapid calculation of the height of molten steel in a converter based on evolutionary learning, comprising the following steps:

[0007] Step 1: Construct a physical model-based calculation model for the height of molten steel in a converter;

[0008] Step 1.1: Construct the converter model dimensions in a rectangular coordinate system and establish the curve equations for each part of the converter body;

[0009] The coordinates of the endpoints of the four parts of the converter model—the upper frustum, the middle straight cylinder, the lower frustum, and the spherical bottom—in a rectangular coordinate system with the x-axis and z-axis as references are set as (r1, h1), (-r1, h1), (r2, h2), (-r2, h2), (r3, h3), (-r3, h3), (r4, h4), and (-r4, h4), respectively. Using the two-point linear equation formula, the two converter body curves l4 and l5 corresponding to the upper frustum, middle straight cylinder, and lower frustum about the z-axis are obtained respectively. _4 l3, l _3 l2, l _2 They are respectively:

[0010]

[0011]

[0012]

[0013]

[0014]

[0015]

[0016] Where x and z represent the coordinate values ​​of the converter body curve about the x-axis and z-axis, respectively;

[0017] From the known radius R of the spherical notch bottom in the converter model, the curve equation l1 of the spherical notch bottom is obtained as follows:

[0018] l1:x 2 +(zR) 2 =R 2 (7)

[0019] Step 1.2: Establish the functional relationship between the volume of molten steel in the furnace and the molten steel surface line under different tilting angles of the converter;

[0020] Given the volume of molten steel, integrate along the z-axis to establish the intersection point (x, y) of the molten steel volume and the molten steel surface line with the converter body. s ,z s The integral equation of ) is then solved to obtain the vertex (x) where the molten steel surface line intersects the converter body. s ,z s The specific coordinates of the coordinates are as follows:

[0021] Given the volume of molten steel, integrating along the z-axis yields two cases:

[0022] Case 1: The volume of molten steel in the converter is V1, the tilting angle of the converter is α1, and the bottom point (x) of the intersection of the molten steel surface line equation L1 and the converter body is... d ,z d In the region above the negative half of the x-axis, i.e., x d <0; In this case, the integral is performed along the z-axis over the molten steel, and the upper and lower limits of the integral are the intersecting vertices (x, y, z) respectively. s ,z s The z-axis value corresponding to the origin (0,0); in this case, since the expression for the coordinates of the vertex where the molten steel surface line intersects the converter body is known, is (x s ,z s Given that the tilting angle of the converter is α1, the expression for the steel liquid level equation L1 is established as follows:

[0023] L1:z=tanα1·(xx s )+z s (8)

[0024] Case 2: The volume of molten steel in the converter is V2, the tilting angle of the converter is α2, and the bottom point (x) of the intersection of the molten steel surface line equation L2 and the converter body is... d ,z d In the region above the positive x-axis, i.e., x d ≥0; In this case, the integral is performed along the z-axis over the molten steel, and the upper and lower limits of the integral are the intersecting vertices (x, y, z) respectively. s ,z s ) and intersecting base points (x) d ,z d The corresponding z-axis value; in this case, since the expression for the coordinates of the intersecting vertices is known as (x s ,z s Given that the tilting angle of the converter is α2, the expression for the steel liquid level equation L2 is established as follows:

[0025] L2:z=tanα2·(xx s )+z s (9)

[0026] Next, by combining the equations for the molten steel surface line L2 and the converter body curves l1, l2, and l3, the intersection point (x) under different tilting angles of the converter is solved. d ,z d ) by intersecting vertices (x s ,z s The coordinate expression represented by )

[0027]

[0028] In equation (10), b = 2(x) s tan2 α2+(Rz s )tanα2), a=tan 2 α2+1;

[0029] Once the upper and lower limits of the steel volume integral are determined, the steel volume can be integrated along the z-axis. Since integrating the steel volume within the converter involves two cases, and the converter body is a rotating body, the integrand for integrating the steel volume along the z-axis is the segmental surface S corresponding to the integration variable dz. _1 Or a circular surface S _2 The integral expression for the volume of molten steel is:

[0030]

[0031] For case 1, the volume of molten steel V1 is divided into upper and lower parts for integration. First, the lower half is integrated circularly, and its integral expression is:

[0032]

[0033] In equation (12), the upper limit of integration z d Based on the steel liquid level equation L1, the converter body curve l is simultaneously established. _1 l _2 l _3 and l _4 Solving the equation for the intersection point (x) d ,z d Therefore, the intersection point (x) of the converter at different tilting angles is obtained. d ,z d ) by intersecting vertices (x s ,z s The coordinate expression represented by ) is:

[0034]

[0035] In equation (13), b = 2(x) s tan 2 α1+(Rz s )tanα1), a=tan 2 α1+1;

[0036] In equation (11), S _2 Let the circular surface corresponding to the integration node of the lower half of the molten steel volume be expressed as:

[0037] S _2 =πr 2 (14)

[0038] In equation (14), r is the converter radius corresponding to the volume of molten steel at the integration node;

[0039] The expression for the converter radius r corresponding to the volume of molten steel at the integration node is:

[0040]

[0041] In equation (15),

[0042] Then, the upper half of the molten steel volume V1 is integrated, and its integral expression is:

[0043]

[0044] S-shaped surface _1 The expression is:

[0045]

[0046] In equation (17), θ is the angle of the segment corresponding to the segment, as shown in the following formula:

[0047] θ=π-2arcsin(1-e / r) (18)

[0048] In equation (18), e is the height of the arc-shaped surface, as shown in the following formula:

[0049] e = r - (zz) s ) / tanα-r s (19)

[0050] In equation (19), r s Let be the converter radius corresponding to the upper limit of integration, and z be the z-axis value corresponding to the arcuate surface at the integration node;

[0051] In summary, under case 1, using equations (12), (13), (14), (15), (16), (17), (18), and (19) with the molten steel volume V1 known, we establish the molten steel volume corresponding to different tilting angles of the converter and the intersection vertex (x). s ,z s Integral equations;

[0052] For case 2, integrating the molten steel volume V2 is a special case of integrating only the upper half in case 1. Therefore, by combining equations (10), (15), (16), (17), (18), and (19), given the molten steel volume V2, we can establish the molten steel volume corresponding to different tilting angles of the converter and the intersection vertex (x). s ,z s Integral equations;

[0053] Step 1.3: Based on the initial volume of molten steel poured into the converter, use the bisection method to solve for the specific coordinates of the intersection point of the molten steel surface line and the converter body under different tilting angles of the converter;

[0054] Let the initial volume of molten steel poured into the converter be V. i The tilting angle of the converter is α3. The bisection method is used to solve for the vertex (x) where the molten steel surface line intersects the converter body. s ,z s The specific coordinate values ​​are calculated as follows:

[0055] Step 1.3.1: Calculate the critical molten steel volume V3 for two cases when the converter tilt angle is α3, distinguishing between the two cases. The expression for the molten steel surface line L3 equation is as follows:

[0056] L3:z=tanα3·x (20)

[0057] Solve the equations of the molten steel surface line L3 and the converter body curves l1, l2, l3, and l4 simultaneously, and solve for the vertex (x) where the molten steel surface line L3 intersects the converter body curves at a converter tilting angle of α3. s ,z s The specific coordinates of )

[0058] Step 1.3.2: Based on the solved intersecting vertices (x... s ,z s The specific coordinate values ​​and equations (11), (17), (18), and (19) are used to solve for the specific critical molten steel volume V3.

[0059] Step 1.3.3: Compare the initial volume V of molten steel poured into the converter. i The value of the critical molten steel volume V3 determines the calculation of the molten steel volume integral; when V i When >V3, the corresponding case x d <0; V i When ≤V3, the corresponding case x d ≥0, and then determine the integral calculation formula for the volume of molten steel according to formula (11);

[0060] Step 1.3.4: Use the binary search method to search for the upper limit of integration z. s The volume of molten steel V is then calculated using equation (11). d Equal to the initial volume V of molten steel poured into the converter i The vertex where the molten steel surface line intersects with the converter body (x) s ,z s Specific coordinate values;

[0061] Step 1.4: Based on the specific coordinates of the intersection of the molten steel surface line and the converter body, solve for the specific equation of the molten steel surface line;

[0062] Step 1.5: Based on the solved equation of the molten steel surface line and the equation of the molten steel surface line passing through the converter mouth and parallel to the molten steel surface line, solve for the height of the molten steel from the converter mouth;

[0063] Based on the specific equation of the liquid level line L1 obtained in step 1.4, a straight line L parallel to the molten steel liquid level line L1 is drawn through the converter mouth coordinates (r4, h4). p Its expression is:

[0064] L p :z=tanα1·(x-r4)+h4 (21)

[0065] Finally, the height H of the molten steel surface from the converter mouth is obtained using the formula for the distance between two parallel lines:

[0066]

[0067] Step 2: Based on the constructed physical model for calculating the molten steel level in the converter, calculate the corresponding molten steel level for different molten steel volumes and converter tilting angles, and then save the data of molten steel level for different molten steel volumes and converter tilting angles as a dataset.

[0068] Based on the specific structural parameters of the converter model, and using the steel volume integral calculation formula in step 1.2, the maximum steel volume of the converter is obtained when the converter tilt angle is 0 and the upper limit of the integral is the converter mouth coordinates (r4, h4), denoted as V. m To obtain a dataset of molten steel level heights corresponding to different molten steel volumes and converter tilting angles, the maximum molten steel volume V of the converter is first... m One-tenth of the volume is used as the increment of molten steel volume for each round of molten steel level calculation, and the increment of the converter tilting angle is set to 1°. Then, using the physics-based molten steel level calculation model constructed in step 1, the molten steel level corresponding to different molten steel volumes and converter tilting angles is calculated. Finally, the data of different molten steel volumes, converter tilting angles, and the calculated molten steel level are saved as a dataset.

[0069] Step 3: Construct a fast calculation model for the height of molten steel in a converter based on evolutionary learning;

[0070] Step 3.1: Based on the structure of the dataset generated in Step 2, construct a neural network framework for generating a fast calculation model for the height of the molten steel level in a converter;

[0071] A neural network framework based on a multilayer feedforward neural network is used to generate a fast calculation model for the height of molten steel in a converter. This network framework consists of an input layer, a hidden layer, and an output layer. The number of neurons in the input layer should be set to 2, and the number of neurons in the output layer should be set to 1. The structure of the hidden layer is set according to the actual converter model parameters and the characteristics of the converter model generated data in step 2.

[0072] Step 3.2: Divide the dataset generated in Step 2 into a training dataset and a test dataset for neural network training and testing according to a certain ratio;

[0073] Step 3.3: Set the initial hyperparameters and structural parameters of the neural network constructed in Step 3.1;

[0074] Step 3.4: Use the training dataset generated in Step 3.2 to train the neural network established in Step 3.1, and use the multi-objective differential evolution algorithm to optimize the structural parameters and hyperparameters of the neural network during the training process to generate a fast calculation model for the height of molten steel in the converter that meets the performance requirements.

[0075] Step 3.4.1: Train the neural network; After setting the hyperparameters and structural parameters of the neural network, use the training dataset generated in step 3.2 to train the neural network established in step 3.1, and output the predicted loss value of the training model in real time during the training process, that is, the degree of difference between the model's predicted data and the actual data.

[0076] Step 3.4.2: Optimize the structural parameters and hyperparameters of the neural network using the multi-objective differential evolution algorithm;

[0077] Step 3.4.2.1: Initialize the population P in the multi-objective differential evolution algorithm t This refers to the parameter search space of the neural network, with the number of generations set to t=0; the parameter search space includes the structural parameters and hyperparameters of the neural network.

[0078] Step 3.4.2.2: Calculate population P t The fitness value of each individual in the population is defined as the neural network performance corresponding to the set of individual parameters in the search space. Two optimization objective functions are designed to measure the quality of individuals in the population. These two optimization objective functions are:

[0079]

[0080] In equation (23), H y' (y) represents the loss value of the model, where y' i y represents the actual height of the molten steel level. i The height of the molten steel surface predicted by the model;

[0081]

[0082] In equation (24), N represents the sum of the number of hidden layers and the number of neurons, where l is the number of hidden layers and n is the number of neurons. i The number of neurons in the i-th hidden layer;

[0083] Step 3.4.2.3: First, for population P t Perform differential evolution to produce a population Q of offspring. t That is, finding new parameter combinations in the neural network parameter search space; then, comparing the population P t and Q t The fitness values ​​of individuals are used to generate the next generation population P through selection operations. t+1 ;

[0084] Step 3.4.2.4: Determine whether the evolutionary algorithm meets the set termination condition. If it does, stop running and execute step 3.4.2.5; if it does not, t = t + 1, and return to step 3.4.2.3.

[0085] Step 3.4.2.5: Output population P t+1 The optimal set of solutions that satisfy the optimization objective conditions, i.e., the optimal set of neural network structure parameters and hyperparameters;

[0086] Step 3.5: Use the test set to test the rapid calculation model of converter molten steel level height generated in Step 3.4 to determine the calculation accuracy of the final model;

[0087] Step 4: Based on the rapid calculation model for the molten steel level height in the converter obtained in Step 3.4, quickly calculate the molten steel level height corresponding to different molten steel volumes and converter tilting angles.

[0088] The beneficial effects of adopting the above technical solution are as follows: The present invention provides a rapid calculation method for the molten steel level in a converter based on evolutionary learning. First, a physical model for calculating the molten steel level in a converter based on physical modeling is constructed. Based on this physical model, a rapid calculation model for the molten steel level in a converter based on evolutionary learning is then constructed. This method directly improves the calculation speed while maintaining the accuracy of the molten steel level calculation. Compared to the physical model based on physical modeling, the rapid calculation model for the molten steel level in a converter based on evolutionary learning proposed in this invention improves the calculation speed from hundreds of seconds to milliseconds, while the calculation accuracy decreases by only 0.91%. When applied to actual converter steelmaking production, this method can provide real-time output to on-site operators of the molten steel level corresponding to different molten steel volumes and converter tilting angles. This allows operators to flexibly adjust the tilting angle of the converter according to actual working conditions, controlling the optimal distance between the molten steel level and the converter opening, thereby helping steel companies improve the production quality of molten steel and the service life of refractory materials such as ladles. Attached Figure Description

[0089] Figure 1 A schematic diagram of the converter tapping process control provided in the background art of this invention;

[0090] Figure 2 A flowchart illustrating a rapid calculation method for converter molten steel level height based on evolutionary learning, provided in an embodiment of the present invention;

[0091] Figure 3 This is a converter model diagram constructed in a rectangular coordinate system provided in an embodiment of the present invention;

[0092] Figure 4 This is a schematic diagram illustrating the different volume integrals of molten steel under different volumes of molten steel in the converter, as provided in an embodiment of the present invention.

[0093] Figure 5 The diagram shows an integral function for integrating the volume of molten steel along the z-axis, as provided in an embodiment of the present invention. (a) is a diagram of an arcuate integrand, and (b) is a diagram of a circular integrand.

[0094] Figure 6 A schematic diagram of the critical molten steel volume for distinguishing two cases of molten steel volume integral calculation provided in an embodiment of the present invention;

[0095] Figure 7 This is a schematic diagram showing the height of the molten steel surface from the converter opening, provided in an embodiment of the present invention.

[0096] Figure 8 A flowchart illustrating the construction of a rapid calculation model for converter molten steel level based on evolutionary learning, provided in an embodiment of the present invention.

[0097] Figure 9 A neural network structure diagram provided in an embodiment of the present invention;

[0098] Figure 10 This is a flowchart illustrating the optimization of neural network parameters using the MODE algorithm provided in an embodiment of the present invention.

[0099] Figure 11 This is a schematic diagram of the population structure in the MODE algorithm provided in this embodiment of the invention;

[0100] Figure 12 The results of testing the fast calculation model of molten steel level in converter based on evolutionary learning using a test set are provided in the embodiments of the present invention. Detailed Implementation

[0101] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0102] In this embodiment, a method for rapidly calculating the height of molten steel in a converter based on evolutionary learning is described, such as... Figure 2 As shown, it includes the following steps:

[0103] Step 1: Construct a physical model-based calculation model for the height of molten steel in a converter;

[0104] Step 1.1: Construct the converter model dimensions in a Cartesian coordinate system, such as... Figure 3 As shown, curve equations for each part of the converter furnace body are established;

[0105] Depend on Figure 3 It can be seen that the converter model can be divided into four parts: upper frustum S4, middle straight cylinder S3, lower frustum S2, and spherical cap bottom S1. The coordinates of the endpoints of the four parts in a rectangular coordinate system with the x-axis and z-axis as references are (r1,h1), (-r1,h1), (r2,h2), (-r2,h2), (r3,h3), (-r3,h3), (r4,h4), and (-r4,h4), respectively. According to the two-point linear equation formula, the two converter body curves l4 and l2 corresponding to the upper frustum S4, middle straight cylinder S3, and lower frustum S2 about the z-axis can be obtained respectively. _4 l3, l _3 l2, l _2 They are respectively:

[0106]

[0107]

[0108]

[0109]

[0110]

[0111]

[0112] Where x and z represent the coordinate values ​​of the converter body curve about the x-axis and z-axis, respectively;

[0113] From the known radius R of the spherical notch bottom S1 in the converter model, the curve equation l1 of the spherical notch bottom S1 is obtained as follows:

[0114]

[0115] Step 1.2: Establish the functional relationship between the volume of molten steel in the furnace and the molten steel surface line under different tilting angles of the converter;

[0116] The different volume integrals of molten steel corresponding to different amounts of molten steel in the converter are as follows: Figure 4 As shown, to solve the equation of the molten steel surface line given the converter tilt angle α, it is necessary to know the vertex (x) where the molten steel surface line intersects the converter body. s ,z s ) and bottom point (x) d ,z d One of the points in ). According to Figure 4 Analysis shows that, given the volume of molten steel, we can integrate along the z-axis to establish the relationship between the volume of molten steel and the intersecting vertex (x). s ,z s The integral equation of ) is then solved to obtain the intersecting vertex (x). s ,z s The specific coordinates of the coordinates are as follows, and the detailed steps are as follows:

[0117] Given the volume of molten steel, we can integrate along the z-axis, which can be divided into two cases:

[0118] Case 1: The volume of molten steel in the converter is V1, the tilting angle of the converter is α1, and the bottom point (x) of the intersection of the molten steel surface line equation L1 and the converter body is... d ,z d In the region above the negative half of the x-axis, i.e., x d <0. In this case, the integral is performed along the z-axis over the molten steel, and the upper and lower limits of the integral are the intersecting vertices (x, y, z) and (x, y, z). s ,z s The coordinates of the intersection vertices are (x) and the z-axis value corresponding to the origin (0,0). In this case, since the expression for the coordinates of the intersection vertices is known as (x) s ,z s Given that the tilting angle of the converter is α1, the expression for the steel liquid level equation L1 is established as follows:

[0119] L1:z=tanα1·(xx s )+z s (32)

[0120] Case 2: The volume of molten steel in the converter is V2, the tilting angle of the converter is α2, and the bottom point (x) of the intersection of the molten steel surface line equation L2 and the converter body is... d ,z d In the region above the positive x-axis, i.e., x d ≥0. In this case, the integral is performed along the z-axis over the molten steel, and the upper and lower limits of the integral are the intersecting vertices (x, y, z) respectively. s ,z s ) and intersecting base points (x) d ,z d The z-axis value corresponding to ). In this case, since the expression for the coordinates of the intersecting vertices is known as (x s ,z s Given that the tilting angle of the converter is α2, the expression for the steel liquid level equation L2 is established as follows:

[0121] L2:z=tanα2·(xx s )+z s (33)

[0122] Next, by simultaneously solving the equations for the molten steel surface line L2 and the converter body curves l1, l2, and l3, the intersection point (x) under different tilting angles of the converter can be solved. d ,z d ) by intersecting vertices (x s ,z s The coordinate expression represented by )

[0123]

[0124] In equation (10), b = 2(x) s tan 2 α2+(Rz s )tanα2), a=tan 2 α2+1.

[0125] Once the upper and lower limits of the steel volume integration are determined, the steel volume can be integrated along the z-axis. Since integrating the steel volume within the converter involves two cases, and the converter body is a rotating body, the integrand for integrating the steel volume along the z-axis is the segmental surface S corresponding to the integration variable dz. _1 Or a circular surface S _2 ,like Figure 5 As shown, the integral expression for the volume of molten steel is:

[0126]

[0127] Regarding scenario 1, by Figure 4 It can be seen that by dividing the volume of molten steel V1 into two parts and integrating them separately, that is, first integrating the dashed line L... * The lower half of the integral is circularly integrated, and its integral expression is:

[0128]

[0129] In equation (12), the upper limit of integration z d The converter body curve l can be established based on the steel liquid level equation L1. _1 l _2 l _3 and l _4 Solving the equation for the intersection point (x) d ,z d Therefore, the intersection point (x) of the converter at different tilting angles is obtained. d ,z d ) by intersecting vertices (x s ,z s The coordinate expression represented by ) is:

[0130]

[0131] In equation (13), b = 2(x) s tan 2 α1+(Rz s )tanα1), a=tan 2 α1+1.

[0132] In equation (11), S _2 The dashed line L * The lower half of the molten steel volume is at the integration node (the integration node is...). Figure 4 In this context, dz represents a small change in z along the z-axis when integrating, which can be considered as the circular surface corresponding to a point on the axis. Figure 5 The geometric relationship shown in (b) is expressed as follows:

[0133] S _2 =πr 2 (38)

[0134] In equation (14), r is the converter radius corresponding to the volume of molten steel at the integral node.

[0135] Since the converter consists of four parts: the upper frustum S4, the middle straight cylinder S3, the lower frustum S2, and the spherical cap bottom S1, the expression for the converter radius r corresponding to the volume of molten steel at the integration node is:

[0136]

[0137] In equation (15),

[0138] Then, for the dashed line L * Integrating the upper part of the equation, its integral expression is:

[0139]

[0140] according to Figure 5 The geometric relationship shown in (a), the arc surface S _1 The expression is:

[0141]

[0142] In equation (17), θ is the angle of the segment corresponding to the segmental surface, derived from... Figure 5 (a) The expression for the angle θ of the segment is:

[0143] θ=π-2arcsin(1-e / r) (42)

[0144] In equation (18), e is the height of the arc-shaped surface, and the expression for e is:

[0145] e = r - (zz) s ) / tanα-r s (43)

[0146] In equation (19), r s Let be the converter radius corresponding to the upper limit of integration, and z be the z-axis value corresponding to the arcuate surface at the integration node.

[0147] In summary, under case 1, equations (12), (13), (14), (15), (16), (17), (18), and (19) can be used to establish the relationship between the molten steel volume V1 and the corresponding molten steel volume at different tilting angles of the converter and the intersecting vertex (x). s ,z s The integral equation of ).

[0148] Regarding scenario 2, by Figure 4 It can be seen that integrating over the volume of molten steel V2 is equivalent to integrating only over the dashed line L in case 1. * The special case of integrating the upper part is given by equations (10), (15), (16), (17), (18), and (19). Therefore, given the molten steel volume V2, we can establish the relationship between the molten steel volume corresponding to different tilting angles of the converter and the intersecting vertex (x). s ,z s The integral equation of ).

[0149] Step 1.3: Based on the initial volume of molten steel poured into the converter, use the bisection method to solve for the specific coordinates of the intersection point of the molten steel surface line and the converter body under different tilting angles of the converter;

[0150] Assume the initial volume of molten steel poured into the converter is V. i The tilting angle of the converter is α3. The bisection method is used to solve for the vertex (x) where the molten steel surface line intersects the converter body. s ,z s The specific steps for obtaining the coordinate values ​​are as follows:

[0151] Step 1.3.1: Calculate the critical molten steel volume V3 for two cases when the converter tilt angle is α3, distinguishing between the two cases of molten steel volume integral calculation, such as... Figure 6 As shown, the expression for equation L3 of the molten steel level at this time is:

[0152] L3:z=tanα3·x (44)

[0153] By simultaneously solving the equations for the molten steel surface line L3 and the converter body curves l1, l2, l3, and l4, the intersection point (x) of the molten steel surface line L3 and the converter body at the converter tilting angle α3 can be determined. s ,z s The specific coordinates of ).

[0154] Step 1.3.2: Based on the solved intersecting vertices (x... s ,z s The specific coordinate values ​​and equations (11), (17), (18), and (19) are used to solve for the specific critical molten steel volume V3.

[0155] Step 1.3.3: Compare the initial volume V of molten steel poured into the converter. i The value of the critical molten steel volume V3 determines the calculation of the molten steel volume integral; when V i When >V3, the corresponding case x d <0; V i When ≤V3, the corresponding case x d ≥0, and then determine the integral calculation formula for the volume of molten steel according to formula (11);

[0156] Step 1.3.4: Use the binary search method to search for the upper limit of integration z. s The volume of molten steel V is then calculated using equation (11). d Equal to the initial volume V of molten steel poured into the converter i The vertex where the molten steel surface line intersects with the converter body (x) s ,z s Specific coordinate values;

[0157] Step 1.4: Based on the specific coordinates of the intersection of the molten steel surface line and the converter body, solve for the specific equation of the molten steel surface line;

[0158] The vertex (x) where the molten steel level line intersects with the converter body s ,z s Substitute the specific coordinate values ​​into the equation L1 or L2 of the molten steel surface line to solve for the specific equation L1 or L2 of the molten steel surface line.

[0159] Step 1.5: Based on the equation of the molten steel surface line and the equation of the straight line passing through the converter mouth and parallel to the molten steel surface line, calculate the height of the molten steel surface from the converter mouth.

[0160] Assuming the specific equation for the liquid level line L1 has been obtained according to step 1.4, as follows: Figure 7 As shown, then through the converter opening coordinates (r4, h4), draw a straight line L parallel to the molten steel surface line L1. p Its expression is:

[0161] L p z = tanα1·(x-r4) + h4 (45)

[0162] Finally, based on the formula for the distance between two parallel lines, the height H of the molten steel surface from the converter mouth can be calculated as follows:

[0163]

[0164] Step 2: Based on the constructed physical model for calculating the molten steel level in the converter, calculate the corresponding molten steel level for different molten steel volumes and converter tilting angles, and then save the data of molten steel level for different molten steel volumes and converter tilting angles as a dataset.

[0165] Based on the specific structural parameters of the converter model, and using the steel volume integral calculation formula described in step 1.2, the maximum steel volume of the converter can be obtained when the converter tilt angle is 0 and the upper limit of the integral is the converter mouth coordinates (r4, h4), denoted as V. m In actual production, the tilting angle of the converter during tapping typically ranges from 0 to 110°. To obtain a dataset of molten steel level heights corresponding to different molten steel volumes and converter tilting angles, the maximum molten steel volume V of the converter is first... m One-tenth of the volume is used as the increment of molten steel volume for each round of molten steel level calculation, and the increment of the converter tilting angle is set to 1°. Then, the molten steel level calculation model based on physical modeling constructed in step 1 is used to calculate the molten steel level corresponding to different molten steel volumes and converter tilting angles. Finally, the data of different molten steel volumes, converter tilting angles, and the calculated molten steel level are saved as a dataset.

[0166] Step 3: Construct a fast calculation model for the height of molten steel in a converter based on evolutionary learning;

[0167] Although the physical model for calculating the molten steel level in a converter, based on physical modeling, can accurately determine the molten steel level under different molten steel volumes and converter tilting angles, as shown in step 1.3, the search process for matching the calculated molten steel volume with the initial molten steel volume is based on a bisection method. During this bisection search, the physical model needs to continuously calculate the integral of the molten steel volume and then compare it with the initial molten steel volume to see if they are equal. In experiments, this search process is extremely time-consuming, typically requiring 300–500 seconds. However, in actual converter steelmaking production, real-time acquisition of the molten steel level is crucial for controlling the optimal tilting angle of the converter during tapping, thereby suppressing slag inflow into the ladle, improving the purity of the molten steel, and enhancing the quality of the produced billets. The requirement for real-time acquisition of the molten steel level indicates that the constructed physical model for calculating the molten steel level in a converter is clearly unsuitable for actual production. Therefore, a rapid molten steel level calculation model is needed to replace the physical calculation model. The basic construction process is as follows: Figure 8 As shown.

[0168] Step 3.1: Based on the structure of the dataset generated in Step 2, construct a neural network framework for generating a fast calculation model for the height of the molten steel level in a converter;

[0169] In this embodiment, a neural network framework based on a multi-layer feedforward neural network is used to generate a fast calculation model for the height of molten steel in a converter. This network framework consists of an input layer, hidden layers (intermediate layers), and an output layer. The number of neurons in the input layer is the same as the dimension of the input data, the number of neurons in the output layer is the same as the number of data points to be fitted, and the number of neurons in the hidden layers and the number of layers need to be set according to certain rules and objectives. The network structure is as follows: Figure 9 As shown.

[0170] In this embodiment, since the corresponding molten steel level is calculated based on different molten steel volumes and converter tilting angles during the converter steelmaking process, it can be concluded that the number of neurons in the input layer of the neural network should be set to 2, and the number of neurons in the output layer should be set to 1. The structure of the hidden layer needs to be specifically set according to the actual converter model parameters and the characteristics of the converter model generated in step 2.

[0171] Step 3.2: Divide the dataset generated in Step 2 into a training dataset and a test dataset for neural network training and testing according to a certain ratio;

[0172] In neural network applications, the original dataset needs to be divided into a training dataset and a test dataset. In this embodiment, the dataset generated in step 2 is divided into a training dataset and a test dataset according to the conventional dataset division ratio of 8:2. The training dataset is mainly used for parameter training and optimization of the fast calculation model for converter molten steel level height, while the test dataset is used to evaluate the generated fast calculation model for converter molten steel level height.

[0173] Step 3.3: Set the initial hyperparameters and structural parameters of the neural network constructed in Step 3.1;

[0174] Before training a neural network, various hyperparameters and structural parameters involved in the network need to be set. These hyperparameters include: network learning rate, hidden layer activation functions, network optimization strategy, and number of training iterations. In feedforward neural networks, the learning rate is typically set within the range of [0.001, 0.1]. Commonly used activation functions include the Sigmoid function, ReLU function, tanh function, and ELU (Exponential Linear Units) function. Network optimization strategies typically include BGD (Batch Gradient Descent), SGD (Stochastic Gradient Descent), Momentum, and Adam (Adaptive Moment Estimation). The number of training iterations is generally set based on a comprehensive consideration of factors such as network training accuracy and training time. In this embodiment, the initial learning rate of the neural network is set to 0.01; the initial activation function of each hidden layer is set to the Sigmoid function; the initial network optimization strategy is the SGD algorithm; and the initial number of training iterations is set to 10,000. The structural parameters involved include the number of hidden layers and the number of neurons in each hidden layer. In this embodiment, the parameter values ​​for the initial number of neurons and the number of layers in the neural network are set to [10, 10], that is, the number of hidden layers is 2, and the number of neurons in each hidden layer is 10 and 10 respectively.

[0175] Step 3.4: Use the training dataset generated in Step 3.2 to train the neural network established in Step 3.1, and use the Multi-objective Differential Evolution (MODE) algorithm to optimize the structural parameters and hyperparameters of the neural network during the training process to generate a fast calculation model for the height of molten steel in the converter that meets the performance requirements.

[0176] Step 3.4.1: Training the Neural Network; After setting the hyperparameters and structural parameters of the neural network, the neural network established in Step 3.1 is trained using the training dataset generated in Step 3.2. During training, the predicted loss value of the trained model is output in real time, representing the degree of difference between the model's predicted data and the actual data. The smaller the loss value, the more accurate the model's prediction and the better its performance. In this embodiment, the predicted loss value of the trained model is the degree of difference between the molten steel level height predicted by the fast calculation model of the converter molten steel level height generated by the neural network and the actual molten steel level height.

[0177] Step 3.4.2: Optimize the structural parameters and hyperparameters of the neural network using the MODE algorithm. During neural network training, it is necessary to continuously adjust the network's structural parameters and hyperparameters to improve the performance of the network training model. Commonly used parameter tuning methods are mainly divided into two types: manual empirical parameter tuning and algorithm-adaptive parameter tuning. In this embodiment, the classic MODE algorithm is used to optimize the multi-objective parameters of the neural network used to generate a fast calculation model for the height of molten steel in a converter. The specific implementation steps are as follows, and the flowchart is shown below. Figure 10 As shown.

[0178] Step 3.4.2.1: Initialize the population P in the MODE algorithm t This refers to the parameter search space of the neural network, with the number of generations set to t = 0. As described in step 3.3, this parameter search space includes the structural parameters and hyperparameters of the neural network. The specific search ranges for these parameters are set as follows: learning rate [0.001, 0.1]; activation function {Sigmoid, ReLu, tanh, ELU}; network optimization strategy {BGD, SGD, Momentum, Adam}; number of training iterations [5000, 20000]; number of hidden layer neurons [5, 30]; number of hidden layers [1, 5]. The structural diagram of the population is shown below. Figure 11 As shown.

[0179] Step 3.4.2.2: Calculate population P t The fitness value of each individual in the population represents the neural network performance corresponding to the set of individual parameters in the search space. In this embodiment, two optimization objective functions are designed to measure the quality of individuals in the population. These two optimization objective functions are:

[0180]

[0181] In equation (23), H y' (y) represents the loss value of the model, where y' i y represents the actual height of the molten steel level. i This is the predicted height of the molten steel level from the model.

[0182]

[0183] In equation (24), N represents the sum of the number of hidden layers and the number of neurons, where l is the number of hidden layers and n is the number of neurons. i denoted as the number of neurons in the i-th hidden layer.

[0184] Step 3.4.2.3: First, for population P t Perform differential evolution operations (mutation and crossover) to produce a progeny population Q. t That is, finding new parameter combinations in the neural network parameter search space; then, comparing the population P t and Q t The fitness values ​​of individuals are used to generate the next generation population P through selection operations. t+1 .

[0185] Step 3.4.2.4: Determine whether the evolutionary algorithm meets the set termination condition. If it does, stop running and execute step 3.4.2.5; if it does not, t = t + 1, and return to step 3.4.2.3.

[0186] Step 3.4.2.5: Output population P t+1 The optimal solution set that satisfies the optimization objective conditions is the optimal set of neural network structure parameters and hyperparameters.

[0187] After the MODE algorithm is optimized, the optimal converter molten steel level calculation model can be trained based on the generated optimal neural network structure parameters and hyperparameters.

[0188] Step 3.5: Use the test set to test the rapid calculation model of converter molten steel level height generated in Step 3.4 to determine the calculation accuracy of the final model;

[0189] The generated rapid calculation model for converter molten steel level was tested using the test set obtained in step 3.2. The average calculation accuracy of the model was 99.09%. Based on the actual situation in converter steelmaking production and the experience of operators, this calculation accuracy can well meet the needs of steelmaking production. To demonstrate the performance of the generated rapid calculation model for converter molten steel level, this embodiment randomly selected the model test results under six different molten steel volumes, such as... Figure 12 As shown. Figure 12 The horizontal axis represents the converter's tilt angle, and the vertical axis represents the molten steel level inside the converter. When the molten steel level becomes 0, it indicates that the converter has reached its maximum tilt angle with that volume of molten steel, and also indicates that the converter is currently in the tapping state. Figure 12As can be seen, the molten steel level height predicted by the rapid calculation model of converter molten steel level height generated in step 3.4 can fit the results in the test set very well. Furthermore, since the actual molten steel level height cannot be negative, the negative values ​​predicted by the model can be processed as 0 values ​​in practical applications.

[0190] Step 4: Based on the rapid calculation model for the molten steel level height in the converter obtained in Step 3.4, quickly calculate the molten steel level height corresponding to different molten steel volumes and converter tilting angles;

[0191] The computational efficiency of the generated fast converter molten steel level calculation model was tested using the test set obtained in step 3.2. The test results show that the average time taken by the evolutionary learning-based fast converter molten steel level calculation model to calculate the molten steel level for different molten steel volumes and converter tilt angles is only 1.3 ms. Compared with the physics-based converter molten steel level calculation model, its computational efficiency is improved by more than 300,000 times. Combined with the actual converter steelmaking production process, this calculation speed can effectively achieve real-time acquisition of molten steel level, thus providing a reliable basis for subsequent production processes.

[0192] Therefore, by applying this evolutionary learning-based rapid calculation model for molten steel level in actual converter steelmaking production, it is possible to quickly calculate the molten steel level corresponding to different molten steel volumes and converter tilting angles. The calculation results are then fed back to the on-site operators in real time, allowing them to flexibly adjust the converter tilting angle according to actual working conditions and control the optimal distance between the molten steel level and the converter opening. This, in turn, helps steel companies improve the quality and efficiency of molten steel production.

[0193] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.

Claims

1. A converter liquid level height rapid calculation method based on evolutionary learning, characterized in that: The method comprises the following steps: Step 1: constructing a physical modeling-based converter liquid level calculation model; Step 2: calculating the corresponding liquid level of the molten steel under different molten steel volumes and converter tilting angles according to the constructed converter molten steel liquid level calculation physical model, and then saving the corresponding liquid level of the molten steel under different molten steel volumes and converter tilting angles as a data set; Step 3: constructing an evolutionary learning-based converter molten steel liquid level rapid calculation model; Step 3.1: constructing a neural network framework for generating the converter molten steel liquid level rapid calculation model according to the structure form of the data set generated in step 2; The converter molten steel liquid level rapid calculation model is generated using a neural network framework based on a multilayer feedforward neural network, which is composed of an input layer, a hidden layer and an output layer; the number of input layer neurons should be set to 2, and the number of output layer neurons should be set to 1; the structure of the hidden layer is set according to the actual converter model parameters and the characteristics of the data generated in step 2; Step 3.2: dividing the data set generated in step 2 into a training data set and a test data set for neural network training and testing according to a certain proportion; Step 3.3: setting the initial hyperparameters and structure parameters of the neural network constructed in step 3.1; Step 3.4: training the neural network established in step 3.1 using the training data set generated in step 3.2, and optimizing the structure parameters and hyperparameters of the neural network using a multi-objective differential evolution algorithm during the training process to generate a converter molten steel liquid level rapid calculation model that meets the performance requirements; Step 3.5: testing the converter molten steel liquid level rapid calculation model generated in step 3.4 using the test set to determine the calculation accuracy of the final model; Step 4: rapidly calculating the corresponding liquid level of the molten steel under different molten steel volumes and converter tilting angles according to the constructed converter molten steel liquid level rapid calculation model.

2. The method for rapid calculation of the liquid level height of converter steel based on evolutionary learning according to claim 1, characterized in that: The specific method of step 1 is: Step 1.1: constructing the converter model size in a rectangular coordinate system and establishing the curve equations of each part of the converter shaft; Step 1.2: establishing the functional relationship between the molten steel volume and the molten steel liquid surface line under different tilting angles of the converter; Step 1.3: using the bisection method to solve the specific coordinate values of the intersection points of the molten steel liquid surface line and the converter shaft under different tilting angles of the converter based on the initial molten steel volume poured into the converter; Step 1.4: solving the specific equation of the molten steel liquid surface line based on the solved specific coordinate values of the intersection points of the molten steel liquid surface line and the converter shaft; Step 1.5: solving the height of the molten steel from the converter mouth based on the solved equation of the molten steel liquid surface line and the equation passing through the converter mouth and parallel to the molten steel liquid surface line.

3. The method for rapid calculation of the liquid level height of converter steel based on evolutionary learning according to claim 1, characterized in that: The specific method of step 1.1 is: The coordinates of the end points of the upper cone, the middle straight cylinder, the lower cone and the spherical bottom of the converter model in the rectangular coordinate system with x-axis and z-axis as the reference are respectively (r1, h1), (-r1, h1), (r2, h2), (-r2, h2), (r3, h3), (-r3, h3), (r4, h4), (-r4, h4); the two-point straight line equation formula is used to obtain two converter body curves l4, l _4 3, l _3 2, l _2 respectively: ； ； ； ； ； ； Wherein, x and z represent the coordinate values of the converter shaft curve with respect to the x-axis and z-axis, respectively; The curve equation l1 of the spherical bottom is obtained as follows based on the known radius R of the spherical bottom in the converter model: 。 4. The method for quickly calculating the liquid level height of converter steel based on evolutionary learning according to claim 3, characterized in that: The specific method of step 1.2 is: In the case of known molten steel volume, the molten steel is integrated along the z-axis to establish an integral equation of the molten steel volume and the intersection point (x s , z s ) of the molten steel liquid surface line and the converter shaft, and then the specific coordinate values of the intersection point (x s , z s ) of the molten steel liquid surface line and the converter shaft are solved, and the specific process is as follows: In the case of known molten steel volume, the molten steel is integrated along the z-axis, which is divided into two cases: Case 1: the volume of molten steel in the converter is V1, the tilting angle of the converter is α1, the molten steel liquid surface line equation L1 intersects the converter shaft at a bottom point (x d , z d ) in the area above the negative half of the x-axis, i.e. ; in this case, the molten steel is integrated along the z-axis, and the upper and lower limits of the integration are the z-axis values corresponding to the intersection vertex (x s , z s ) and the coordinate origin (0, 0), respectively; in this case, since the expression of the intersection vertex coordinates of the molten steel liquid surface line and the converter shaft is (x s , z s ) and the tilting angle of the converter is α1, the expression of the molten steel liquid surface line equation L1 is established as follows: ； Case 2: the volume of molten steel in the converter is V2, the tilting angle of the converter is a2, the intersection point (x d , z d ) of the molten steel surface line equation L2 and the converter body is in the area above the positive half-axis of the x-axis, i.e. ; in this case, the molten steel is integrated along the z-axis, and the upper and lower limits of the integration are the z-axis values corresponding to the intersection vertex (x s , z s ) and the intersection bottom point (x d , z d ), respectively; in this case, since the expression of the intersection vertex coordinates is (x s , z s ) and the tilting angle of the converter is a2, the expression of the molten steel surface line equation L2 is established as follows: ； Next, by combining the equations for the molten steel surface line L2 and the converter body curves l1, l2, and l3, the intersection point (x) under different tilting angles of the converter is solved. d , z d ) by intersecting vertices (x s , z s The coordinate expression represented by ) ； In formula (10), , , , ; After the upper and lower limits of the volume of liquid steel are determined, the volume of liquid steel is integrated along the z-axis; since the volume of liquid steel in the converter is integrated in two cases, and the body of the converter is a rotary body, the integrand for the integration of the volume of liquid steel along the z-axis is the arc surface S corresponding to the integral variable dz _1 or the circular surface S _2 The integral expression of the volume of liquid steel is: ； For case 1, the molten steel volume V1 is divided into two parts to be integrated, i.e. the lower part is firstly integrated, and the integral expression is as follows: ； In formula (12), the upper limit of integration z d According to the molten steel liquid surface line equation L1 and the converter shaft curve l _1 , _2 , _3 and l _4 Equation, the intersection bottom point (x d , z d ) is obtained, so the intersection bottom point (x d , z d ) under different tilting angles of the converter is expressed by the coordinate expression of the intersection vertex (x s , z s ). ； In formula (13), , , , , , ; In formula (11), S _2 is the volume of the lower half of the molten steel at the integral node corresponding to the circular surface, and its expression is: ； In formula (14), r is the radius of the converter corresponding to the integral node of the molten steel volume; The expression of the radius r of the converter corresponding to the integral node of the molten steel volume is as follows: ； in formula (15), ; Then, the upper part of the molten steel volume V1 is integrated, and the integral expression is as follows: ； Arcuate surface S _1 The expression is: ； In formula (17), theta is the arc angle corresponding to the arc surface, and the expression is as follows: ； In formula (18), e is the height of the arc surface, and the expression is as follows: ； In formula (19), r s is the radius of the converter corresponding to the upper limit of integration, and z is the z-axis value corresponding to the arc at the integration point. In summary, under case 1, using equations (12), (13), (14), (15), (16), (17), (18), and (19) with the molten steel volume V1 known, we establish the molten steel volume corresponding to different tilting angles of the converter and the intersection vertex (x). s , z s Integral equations; For case 2, integrating the molten steel volume V2 is a special case of integrating only the upper half in case 1. Therefore, by combining equations (10), (15), (16), (17), (18), and (19), given the molten steel volume V2, we can establish the molten steel volume corresponding to different tilting angles of the converter and the intersection vertex (x). s , z s The integral equation of ).

5. The method for rapid calculation of the liquid level height of converter steel based on evolutionary learning according to claim 4, characterized in that: The specific method of step 1.3 is as follows: The volume of molten steel initially poured into the converter is set as V i , the tilting angle of the converter is α3, and the process for solving the specific coordinate values of the intersection point (x s , z s ) of the molten steel liquid surface line and the converter shaft is as follows: Step 1.3.1: Calculate the critical molten steel volume V3 for distinguishing the two cases of the integral calculation of the molten steel volume when the tilting angle of the converter is alpha3, and the expression of the molten steel liquid surface line L3 at this time is as follows: ； The intersection point (x, z) of the molten steel liquid level line L3 and the converter shaft curve l1, l2, l3, l4 is solved by simultaneously solving the equations of the molten steel liquid level line L3 and the converter shaft curve l1, l2, l3, l4 when the tilting angle of the converter is α3. s , z s ) of the intersection point of the molten steel liquid level line L3 and the converter shaft curve l1, l2, l3, l4 is solved by simultaneously solving the equations of the molten steel liquid level line L3 and the converter shaft curve l1, l2, l3, l4 when the tilting angle of the converter is α Step 1.3.2: According to the solved intersection vertex (x s , z s ) specific coordinate value and formula (11), (17), (18), (19), the specific critical molten steel volume V3 is solved; Step 1.3.3: Compare the initial volume V of molten steel poured into the converter. i The value of the critical molten steel volume V3 determines the calculation of the molten steel volume integral; when V i > V3, corresponding situation V i When ≤ V3, the corresponding situation Then, the integral calculation formula for the volume of molten steel is determined according to formula (11); Step 1.3.4: Binary search for the upper limit of integration z s , and then the volume of molten steel V d calculated from equation (11) is equal to the volume of molten steel V i initially poured into the converter. The specific coordinate values of the intersection point (x s , z s ) of the molten steel liquid surface line and the converter shaft at the time when the molten steel liquid surface line reaches the intersection point 6. The method for rapid calculation of the liquid level height of converter steel based on evolutionary learning according to claim 5, characterized in that: The specific method of step 1.5 is as follows: The specific equation of the liquid surface line L1 obtained according to step 1.4 is set, and then a straight line L parallel to the liquid surface line L1 of the molten steel is drawn through the converter mouth coordinate (r4, h4) p whose expression is: ； Finally, the height H of the molten steel liquid surface from the converter mouth is obtained according to the distance formula between two parallel lines: 。 7. The method for rapid calculation of the liquid level height of converter steel based on evolutionary learning according to claim 6, characterized in that: The step 2 obtains the maximum molten steel volume of the converter when the tilting angle of the converter is 0 and the upper limit of the integral is the coordinate (r4, h4) of the converter mouth according to the specific structural parameters of the converter model and the integral calculation formula of the molten steel volume in step 1.2, and is recorded as V m To obtain the data set of the molten steel liquid level corresponding to different molten steel volumes and tilting angles of the converter, first, 1 / 120 of the maximum molten steel volume V m of the converter is taken as the molten steel volume increment for each round of molten steel liquid level calculation, and the increment of the tilting angle of the converter is set to 1°; then, the molten steel liquid level corresponding to different molten steel volumes and tilting angles of the converter is calculated through the molten steel liquid level calculation model of the converter based on physical modeling constructed in step 1; finally, the data of different molten steel volumes, tilting angles of the converter and especially the calculated molten steel liquid level are saved as a data set.

8. The method for quickly calculating the liquid level height of converter steel based on evolutionary learning according to claim 7, characterized in that: The specific method of step 3.4 is as follows: Step 3.4.1: Train the neural network; after the hyperparameters and structure parameters of the neural network are set, the training data set generated in step 3.2 is applied to train the neural network established in step 3.1, and the prediction loss value of the training model is output in real time during the training process, i.e. the difference between the model prediction data and the actual data; Step 3.4.2: Optimize the structure parameters and hyperparameters of the neural network using the multi-objective differential evolution algorithm; Step 3.4.2.1: initialize the population P in the multi-objective differential evolution algorithm t i.e., the parameter search space of the neural network, and set the evolutionary generation t = 0; the parameter search space includes the structure parameters and hyperparameters of the neural network; Step 3.4.2.2: Calculate population P t The fitness value of each individual, i.e. the performance of the neural network corresponding to the individual's parameter set in the search space; two optimization objective functions are designed to measure the pros and cons of individuals in the population, which are: ； In formula (23), represents a loss value of the model, wherein is an actual molten steel liquid level value, is a molten steel liquid level value predicted by the model; ； In formula (24), N represents the sum of the number of hidden layers and the number of neurons, where l is the number of hidden layers, and n i is the number of neurons in the i-th hidden layer. Step 3.4.2.3: First, for population P t Perform differential evolution to produce a population Q of offspring. t That is, finding new parameter combinations in the neural network parameter search space; then, comparing the population P t and Q t The fitness values ​​of individuals are used to generate the next generation population P through selection operations. t+1 ; Step 3.4.2.4: Determine whether the evolution algorithm meets the set termination condition, if yes, stop running and execute step 3.4.2.5; if not, t = t + 1, return to step 3.4.2.3; Step 3.4.2.5: output population P t+1 The optimal solution set that satisfies the optimization objective condition, i.e., the optimal neural network structure parameter and hyperparameter set.

Images