Coiling temperature calculation method and system based on neural network and mechanism model

Abstract

The invention provides a coiling temperature calculation method and system based on a neural network and a mechanism model.The calculation method comprises the steps that firstly, strip steel slabs with the high coiling temperature hit rate are screened in a historical data set, abnormal values are removed, and layers are divided; segmenting the heat output roller way, and classifying and preprocessing data; performing offline training on the preprocessed data based on a BP neural network to obtain a water-cooling heat transfer coefficient prediction model of each layer; in the actual production process of the strip steel, input parameters of online operation are obtained, the water-cooling heat exchange coefficient of the real-time working condition is predicted online through the water-cooling heat exchange coefficient model, and then the mechanism model is further used for calculating the temperature on the strip steel reheating output roller way.

Description

Technical Field

[0001] The invention relates to the technical field of coiling temperature calculation for metallurgical hot rolling, and in particular to a coiling temperature calculation method and system based on a neural network and a mechanism model. Background Art

[0002] The coiling temperature of hot rolled strip is an important factor that determines the processing performance, mechanical properties and physical properties of the finished strip. The strip needs to control the number of laminar flow manifold switches according to the target coiling temperature after passing through the finishing mill and the hot output roller. Therefore, predicting the laminar cooling coiling temperature is the key to controlling the coiling temperature.

[0003] The laminar cooling device is mainly composed of an upper header and a lower header. Figure 1 The figure shows a laminar cooling production line for hot rolled strip steel, 101 and 102 in the figure represent the upper header and the lower header, and there are 14 sets of header banks in total. Figure 1 , FDT temperature is the final rolling temperature, and CT temperature is the coiling temperature.

[0004] At present, there are many types and specifications of steel in hot rolling mills, and there are different process requirements. The control range of laminar cooling is a serious test, and the heat exchange characteristics in the production process are not fixed. There are three main ways of heat exchange on the heat output roller, which are divided into water cooling, air cooling, and heat radiation. Water cooling accounts for most of the entire heat exchange process. The main factors affecting the water cooling heat transfer coefficient are: strip temperature, thickness, speed, cooling medium temperature, water volume, and production environment temperature. In addition, the on-site production environment is complex, and the heat transfer coefficient has the characteristics of strong nonlinearity, time-varying, and instability. It is impossible to establish an accurate mathematical model for calculation. At present, statistical experience data is mainly used. Therefore, accurate prediction of the heat transfer coefficient is a key link in coiling temperature prediction.

[0005] With the continuous development of control theory, new comprehensive control methods are generally used to deal with systems with strong nonlinear characteristics such as laminar cooling coiling temperature, such as fuzzy control, neural networks, predictive control, etc. For example, the Chinese patent application document with application number CN202011311991.7 discloses a temperature calculation method that optimizes the artificial neural network through the whale optimization algorithm, obtains the predicted upper limit and predicted lower limit of the relative deviation between the measured coiling temperature and the target coiling temperature according to the input, and can perform point prediction and interval prediction at the same time. The Chinese patent application document with application number CN202110336422.6 discloses a temperature control method. On the basis of the existing strip pre-coiling predicted temperature Tpre, the control method introduces the compensation temperature Tcompensated unit length temperature k and the fine-tuning temperature t, and continuously calculates and reassigns them in each sampling cycle, so that the actual coiling temperature of the strip can be controlled more accurately. A Chinese patent application document with application number CN202311464172.X proposes a coiling temperature control method for a hot rolling production line based on an ARO-BAT-LSTM neural network, including selecting control parameters that affect the coiling temperature, constructing an LSTM neural network, obtaining an ARO-BAT-LSTM prediction model, and using the constructed prediction model to obtain a predicted value of the coiling temperature.

[0006] However, the above existing technologies directly predict and control the coiling temperature, or only use intelligent algorithms for prediction. In actual production, there are many parameters that affect the coiling temperature and they are coupled with each other. The control stability of a single algorithm needs to be optimized. These methods do not directly predict the water cooling heat transfer coefficient, nor do they combine intelligent algorithms with mechanism models. Therefore, these methods are still insufficient in practical applications. Summary of the invention

[0007] The present invention provides a coiling temperature calculation method and system based on a neural network and a mechanism model to solve the problem of online prediction of a water cooling heat transfer coefficient under instant working conditions, and uses the predicted water cooling heat transfer coefficient in combination with a traditional mechanism model to predict temperature.

[0008] To achieve the above object, the present invention proposes the following technical solutions:

[0009] The present invention provides a coiling temperature calculation method based on a neural network and a mechanism model, which is characterized by comprising the following steps:

[0010] Step S1: Divide the steel strip into sample segments of equal length, each sample segment is a control object, and divide the heat output roller into a number of control sub-segments according to the header tube bundle;

[0011] Step S2: Obtaining historical production data corresponding to each control sub-segment and performing validity processing on the data, wherein the historical production data includes the historical water cooling heat transfer coefficient and the water cooling heat transfer coefficient influencing factor of the strip in the control sub-segment;

[0012] Step S3: Taking the water cooling heat transfer coefficient influencing factor as input and the water cooling heat transfer coefficient of each control sub-segment as output, a neural network model is constructed, and the historical production data of each control sub-segment after effectiveness processing is used to train the model respectively. After the training is completed, a water cooling heat transfer coefficient prediction model corresponding to each control sub-segment is obtained;

[0013] Step S4: obtaining the current value of the water cooling heat transfer coefficient influencing factor of the first control sub-segment corresponding to each sample segment when the current strip steel is running on the hot output roller table;

[0014] Step S5: inputting the current value of the water cooling heat transfer coefficient influencing factor obtained in step S4 into the water cooling heat transfer coefficient prediction model of the control sub-segment to predict the water cooling heat transfer coefficient corresponding to the control sub-segment;

[0015] Step S6: input the water cooling heat transfer coefficient obtained in step S5 into the mechanism model to calculate the strip temperature at the outlet of each header tube bundle.

[0016] Furthermore, in the coiling temperature calculation method based on neural network and mechanism model provided by the present invention, it can also have the following characteristics: wherein, the influencing factors of water cooling heat transfer coefficient include steel type, thickness, strip entry temperature, speed, water temperature, ambient temperature, and flow rate.

[0017] Furthermore, in the coiling temperature calculation method based on neural network and mechanism model provided by the present invention, it can also have the following characteristics: wherein, the data validity processing in step S2 is to determine whether the data is within the parameter range of normal operation of the strip on the hot output roller, and the [μ-3σ,μ+3σ] confidence interval rule is used to detect and eliminate abnormal data.

[0018] Furthermore, in the coiling temperature calculation method based on neural network and mechanism model provided by the present invention, it can also have the following characteristics: wherein, the neural network model in step S3 is a BP neural network, the BP neural network includes an input layer, a hidden layer and an output layer, each layer contains nodes, the parameters are input in the input layer, and the output value of each unit is obtained through the output of the hidden layer. If the output value exceeds the preset error value, reverse error transmission is performed, and the corresponding weight is adjusted according to the error value.

[0019] Furthermore, in the coiling temperature calculation method based on the neural network and the mechanism model provided by the present invention, it can also have such a feature: wherein, the input parameters reach the nodes of the hidden layer from the input layer, pass through the activation function of each node, and then send the calculation information to the output layer through the hidden layer to obtain the result, which is specifically as follows:

[0020] Initially set N samples, where node i is the output node O i , corresponding to the input node x k , the node corresponding to the output layer is y k , the output of node i is O ik , now calculate the jth node of the lth layer. When the kth sample is input, the input of node j is:

[0021]

[0022] In formula (1) and (2), After the kth data is input into the l-1th layer, the calculation result of the jth node is obtained; After the kth data is input into the lth layer, the calculation result of the jth node is obtained; O is the weight of node i; = is the input of node i in layer l;

[0023] The error function used is:

[0024]

[0025] In formula (3), E k = indicates the deviation between the actual value and the predicted value; Represents the actual calculation result of unit j; y lk Represents the corresponding network output;

[0026] After calculation, the error is:

[0027]

[0028] make

[0029]

[0030] So

[0031]

[0032] In formula (4), (5), (6), E is the sum of calculated errors; N is the number of samples; It is the derivative of the actual value and the deviation value relative to the node input;

[0033] Assume that j is the output node unit, then

[0034]

[0035] In formula (7), It is the input passed to the next layer, i.e. the l+1th layer; is the actual calculation result of unit j; y k is the predicted value of node j; is the measured value of node j; is the output function of the node;

[0036] calculate To calculate back from the (l+1) layer; the calculation result at the mth unit in the (l+1) layer is:

[0037]

[0038] In formula (8), is the input of node m when it is the kth sample; is the weight of node m when it is the jth sample; is the derivative of the error of node m relative to the node input when it is the kth sample;

[0039] get

[0040]

[0041] Then there is

[0042]

[0043] Furthermore, in the coiling temperature calculation method based on the neural network and the mechanism model provided by the present invention, it can also have the following characteristics: wherein, the steps of the BP neural network algorithm are as follows:

[0044] 1) Select the initial value of the weight coefficient;

[0045] 2) Repeat the forward and reverse process until the error index meets the accuracy requirement, where ε is the accuracy, that is:

[0046]

[0047] In formula (8), E is the total error; E k < is the error of the kth sample; ε is the set model accuracy; N is the number of samples; k is the sample number;

[0048] Forward process calculation for k = 1 to N: Calculate the nodes of each layer and k=2,...,N; Reverse process calculation: for each layer (l=L-1 to 2), for each node in each layer, calculate

[0049] Correction weights:

[0050]

[0051] In formula (12) and (13), μ is the step size, w ij is the weight, and E calculates the accumulated error of the data.

[0052] Furthermore, in the coiling temperature calculation method based on neural network and mechanism model provided by the present invention, it can also have the following characteristics: wherein, the number of nodes in the input layer of the BP neural network is 8, which are steel type, thickness, entry temperature, target temperature, speed, water temperature, ambient temperature, and flow rate; the number of nodes in the output layer is 1, that is, the water cooling heat transfer coefficient; the number of nodes in the hidden layer is determined by an empirical formula, which is:

[0053]

[0054] In formula (14), h is the number of hidden layer nodes, m is the number of input layer nodes, n is the number of output layer nodes, and a is a tuning constant between 1 and 10.

[0055] Furthermore, in the coiling temperature calculation method based on neural network and mechanism model provided by the present invention, it can also have the following characteristics: wherein, the weight adjustment rate of the BP neural network is 0.0035, the threshold adjustment rate is 0.009, the maximum number of training times is 800, the allowable error of a single sample is 0.001, and the allowable error of each iteration is 0.004.

[0056] Furthermore, in the coiling temperature calculation method based on the neural network and the mechanism model provided by the present invention, it can also have the following characteristics: wherein the mechanism model in step S6 is a one-dimensional non-steady-state heat conduction process:

[0057]

[0058] In formula (15), T is the strip temperature; τ is the time; λ is the thermal conductivity of the strip; ρ is the material density; C is the specific heat capacity; x is the coordinate in the thickness direction of the strip;

[0059] The initial boundary condition established is the final rolling temperature distribution:

[0060] τ=0,T=T0 (16)

[0061] In formula (16), T0 is the final rolling temperature;

[0062] Considering the real-time working conditions of the strip on the hot output roller table to solve the boundary condition problem, the third type of boundary conditions known for air cooling radiation and water cooling convection are adopted:

[0063]

[0064] In formula (17), α is the heat transfer coefficient of the strip surface; T f is the surface cooling medium temperature of the strip; m is the total number of samples of the strip;

[0065] The water cooling heat transfer coefficient model of the strip running on the hot output roller is:

[0066]

[0067] In formula (18), z 12 , z2 is the self-learning item of heat transfer coefficient; a1, a2, a3, V0 are model parameters; V is the real-time speed of the strip segment; V base is the finishing rolling speed; h is the strip thickness; T ω for water temperature;

[0068] The heat transfer model without water cooling is:

[0069] α=k air ·ε·σ·[(T+273) 2 +(T α +273) 2 ]·[(T+273)+(T α +273)] (19)

[0070] In formula (19), K air is the correction coefficient; ε is the blackness coefficient of the steel plate; σ is the Stefan-Boltzmann constant; T a is the temperature of the air medium.

[0071] The present invention also provides a coiling temperature calculation system based on a neural network and a mechanism model, which is applied to the above-mentioned coiling temperature calculation method based on a neural network and a mechanism model, and is characterized in that it includes: a historical production data acquisition and storage module, which is used to acquire and store historical production data, including historical water-cooling heat transfer coefficients and water-cooling heat transfer coefficient influencing factors; a current heat transfer coefficient influencing factor acquisition module, which is used to collect current heat transfer coefficient influencing factor data in real time; a water-cooling heat transfer coefficient prediction module, which inputs the current value of the heat transfer coefficient influencing factor through a trained BP neural network model, and outputs the corresponding water-cooling heat transfer coefficient prediction value; a mechanism model operation module, which inputs the water-cooling heat transfer coefficient prediction value through the mechanism model, and calculates the strip temperature.

[0072] Beneficial effects of the present invention:

[0073] The coiling temperature calculation method based on neural network and mechanism model of the present invention first screens the slabs with higher coiling temperature hit rate in the historical data set, removes abnormal values, and divides them into layers. The hot output roller is segmented, and the data is classified and preprocessed. The preprocessed data is trained offline based on the BP neural network to obtain a water-cooling heat transfer coefficient prediction model for each layer. In the actual production process of strip steel, the input parameters of the online operation are obtained, and the water-cooling heat transfer coefficient of the real-time working condition is predicted online through the water-cooling heat transfer coefficient prediction model, and then the mechanism model is further used to calculate the temperature on the strip reheat output roller. The temperature predicted by the method of the present invention is closer to the actual value and the target value, and a relatively ideal effect is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 is a schematic diagram of a laminar cooling line for hot rolled strip;

[0075] Figure 2 It is a flow chart of the coiling temperature calculation method based on the neural network and the mechanism model of the present invention;

[0076] Figure 3 It is a schematic diagram of the execution flow of the coiling temperature control method based on the intelligent algorithm and the mechanism model in the embodiment of the present invention;

[0077] Figure 4 It is a comparison chart of the predicted temperature of the coiling inlet calculated in the embodiment of the present invention, the actual temperature tested by the pyrometer, and the target temperature. DETAILED DESCRIPTION

[0078] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the following embodiments are combined with the accompanying drawings to specifically illustrate the technical solutions of the present invention.

[0079] <Example>

[0080] See also Figure 2 and Figure 3 The embodiment of the present invention provides a coiling temperature calculation method based on a neural network and a mechanism model, comprising the following steps:

[0081] Step S1: According to different finishing exit thicknesses, the strip is divided into sample segments of equal length along the roller running direction, each sample segment is a control object, and the heat output roller manifold is divided into several control sub-segments by tube bundle (bank).

[0082] See also Figure 1 Each group of Banks corresponds to a sub-segment, and the 14 groups of Banks on the heat output roller table in this embodiment are divided into 14 control sub-segments.

[0083] Step S2: Obtain the historical production data corresponding to each control sub-segment and perform validity processing on the data, wherein the historical production data includes the historical water cooling heat transfer coefficient and the water cooling heat transfer coefficient influencing factor of the strip in the control sub-segment.

[0084] In this embodiment, the factors affecting the water cooling heat transfer coefficient include steel type, thickness, strip steel entry temperature, speed, flow rate, water temperature, and ambient temperature. Among them, steel type is the material of the slab; thickness is the thickness of the slab at the finish rolling exit; speed is the average speed of the slab in the control sub-segment; flow rate is the pressure at the manifold pneumatic valve; entry temperature is the average temperature of the slab at the entrance of each manifold Bank in the slab control sample section; water temperature is the temperature of the laminar cooling water; and ambient temperature is the ambient temperature corresponding to the production workshop.

[0085] Data validity processing is to determine whether the data is within the parameter range of normal strip operation on the hot output roller. Among them, the steel type and the finishing exit thickness are constants and do not require data validity processing. The average running speed, strip entry temperature, water temperature, ambient temperature, and flow rate require data validity determination.

[0086] The [μ-3σ,μ+3σ] confidence interval rule is used to detect and eliminate abnormal data. The method for determining whether the entry temperature data is valid is as follows: if the entry temperature is significantly higher than the normal temperature of this section of steel on the hot output roller, the data is determined to be unqualified. The determination of the remaining data speed, water temperature, ambient temperature, and flow rate is the same as above. If the parameters are no longer within the predetermined parameter range, the data is determined to be unqualified and the abnormal data is eliminated.

[0087] The following table is the effective historical production data of this embodiment:

[0088] Table 1 Historical production data

[0089]

[0090] The processed data is organized into a data set for subsequent model training.

[0091] Step S3: Taking the water-cooling heat transfer coefficient influencing factor as input and the water-cooling heat transfer coefficient of each control sub-segment as output, a neural network model is constructed, and the historical production data of each control sub-segment after effectiveness processing is used to train the model respectively. After the training is completed, the water-cooling heat transfer coefficient prediction model corresponding to each control sub-segment is obtained.

[0092] The neural network model of this embodiment self-learns the above data, establishes neural network weights and threshold parameters that meet the standard accuracy, and establishes a water cooling heat exchange coefficient prediction model.

[0093] The neural network model is a BP neural network, including an input layer, a hidden layer and an output layer. Each layer contains nodes. The parameters are input in the input layer and the output value of each unit is obtained through the output of the hidden layer. If the output value exceeds the preset error value, reverse error transmission is performed and the corresponding weight is adjusted according to the error value.

[0094] The input parameters go from the input layer to the nodes of the hidden layer, pass through the activation function of each node, and then send the calculation information to the output layer through the hidden layer to get the result, as follows:

[0095] Initially set N samples, where node i is the output node O i , corresponding to the input node x k , the node corresponding to the output layer is y k , the output of node i is O ik , now calculate the jth node of the lth layer. When the kth sample is input, the input of node j is:

[0096]

[0097] In formula (1) and (2), After the kth data is input into the l-1th layer, the calculation result of the jth node is obtained; After the kth data is input into the lth layer, the calculation result of the jth node is obtained; O is the weight of node i; = is the input of node i in layer l;

[0098] The error function used is:

[0099]

[0100] In formula (3), E k = indicates the deviation between the actual value and the predicted value; Represents the actual calculation result of unit j; y lk Represents the corresponding network output;

[0101] After calculation, the error is:

[0102]

[0103] make

[0104]

[0105] So

[0106]

[0107] In formula (4), (5), (6), E is the sum of calculated errors; N is the number of samples; It is the derivative of the actual value and the deviation value relative to the node input;

[0108] Assume that j is the output node unit, then

[0109]

[0110] In formula (7), It is the input passed to the next layer, i.e. the l+1th layer; is the actual calculation result of unit j; yk is the predicted value of node j; is the measured value of node j; is the output function of the node;

[0111] calculate To calculate back from the (l+1) layer; the calculation result at the mth unit in the (l+1) layer is:

[0112]

[0113] In formula (8), is the input of node m when it is the kth sample; is the weight of node m when it is the jth sample; is the derivative of the error of node m relative to the node input when it is the kth sample;

[0114] get

[0115]

[0116] Then there is

[0117]

[0118] The steps of the neural network algorithm are as follows:

[0119] 1) Select the initial value of the weight coefficient;

[0120] 2) Repeat the forward and reverse process until the error index meets the accuracy requirement, where ε is the accuracy, that is:

[0121]

[0122] In formula (8), E is the total error; E k is the error of the kth sample; ε is the set model accuracy; N is the number of samples; k is the sample number;

[0123] Forward process calculation for k = 1 to N: Calculate the nodes of each layer and k=2,...,N; Reverse process calculation: for each layer (l=L-1 to 2), for each node in each layer, calculate

[0124] Correction weights:

[0125]

[0126] In formula (12) and (13), μ is the step size, w ij is the weight, and E calculates the accumulated error of the data.

[0127] The number of nodes in the input layer of the BP neural network is 8, which are steel type, finishing exit thickness, entry temperature, target temperature, speed, water temperature, ambient temperature, and flow rate;

[0128] The number of nodes in the output layer is 1, which is the water cooling heat transfer coefficient;

[0129] The number of nodes in the hidden layer is determined by an empirical formula:

[0130]

[0131] In formula (14), h is the number of hidden layer nodes, m is the number of input layer nodes, n is the number of output layer nodes, and a is a tuning constant between 1 and 10.

[0132] The weight adjustment rate of the BP neural network is 0.0035, the threshold adjustment rate is 0.009, the maximum number of training times is 800, the allowable error for a single sample is 0.001, and the allowable error for each iteration is 0.004.

[0133] In this embodiment, after model training, 14 neural network prediction models corresponding to 14 banks are obtained, which are used to predict the water cooling heat transfer coefficient of each Bank according to steel type, thickness, entry temperature, speed, target temperature, water temperature, ambient temperature, and flow rate.

[0134] The BP model parameters after training are shown in Table 2:

[0135] Table 2 BP model parameters after training

[0136]

[0137]

[0138] Step S4: obtaining the current value of the water cooling heat transfer coefficient influencing factor of the first control sub-segment corresponding to each sample segment when the current strip steel is running on the hot output roller.

[0139] Step S5: input the current value of the water cooling heat transfer coefficient influencing factor obtained in step S4 into the water cooling heat transfer coefficient prediction model of the control sub-segment to predict the water cooling heat transfer coefficient corresponding to the control sub-segment.

[0140] The relevant parameters of this embodiment are as follows in Table 3:

[0141] Table 3 Water cooling heat transfer coefficient and its influencing factors

[0142]

[0143] Step S6: Input the water cooling heat transfer coefficient obtained in step S5 into the mechanism model to calculate the strip temperature at the outlet of each header tube bundle. The strip temperature of the current control sub-segment is used to obtain the end outlet temperature of the control sub-segment. When arriving at the next sub-segment, the heat transfer coefficient is predicted using the latest calculated heat transfer coefficient influencing factor, and then the average temperature of the strip on the hot output roller is calculated.

[0144] The specific calculation process is as follows: when the strip runs on the heat output roller, the heat dissipation in the thickness direction accounts for the main part, and the heat dissipation in the length and width directions can be excluded from the calculation. The same phase change latent heat can be considered in the specific heat during the cooling process. Therefore, the mechanism model is a one-dimensional non-steady-state heat conduction process:

[0145]

[0146] In formula (15), T is the strip temperature; τ is the time; λ is the thermal conductivity of the strip; ρ is the material density; C is the specific heat capacity; x is the coordinate in the thickness direction of the strip;

[0147] The initial boundary condition established is the final rolling temperature distribution:

[0148] τ=0,T=T0 (16)

[0149] In formula (16), T0 is the final rolling temperature of the strip;

[0150] Considering the real-time working conditions of the strip on the hot output roller table to solve the boundary condition problem, air cooling radiation and water cooling convection boundary conditions are adopted:

[0151]

[0152] In formula (17), α is the non-central heat transfer parameter of the strip; T f is the temperature of cooling water; m is the total number of samples of the strip;

[0153] The water cooling heat transfer coefficient model of the strip running on the hot output roller is:

[0154]

[0155] In formula (18), z 12 , z2 is the self-learning item of heat transfer coefficient; a1, a2, a3, V0 are model parameters; V is the real-time speed of the strip segment; V base is the finishing rolling speed; h is the strip thickness; T ω for water temperature;

[0156] The heat transfer model without water cooling is:

[0157] α=k air ·ε·σ·[(T+273) 2 +(T α +273) 2 ]·[(T+273)+(T α +273)] (19)

[0158] In formula (19), K air is the correction coefficient; ε is the blackness coefficient of the steel plate; σ is the Stefan-Boltzmann constant; T a is the temperature of the air medium.

[0159] See also Figure 4 The figure illustrates that the coiling inlet temperature of the strip is calculated by using the coiling temperature based on the neural network and mechanism model of the embodiment of the present invention, and the calculated temperature is compared with the target temperature and the actual temperature measured by the pyrometer. It can be seen that the method of the present invention has achieved a relatively ideal effect.

[0160] This embodiment also provides a coiling temperature calculation system based on neural network and mechanism model, which is applied in the coiling temperature calculation method of neural network and mechanism model. The system is an intelligent control system, and a corresponding operating program is set in the system. The system includes the following functional modules: historical production data acquisition and storage module, current heat transfer coefficient influencing factor acquisition module, water cooling heat transfer coefficient prediction module, and mechanism model operation module.

[0161] The historical production data acquisition and storage module is used to acquire and store historical production data, including historical water cooling heat transfer coefficients and water cooling heat transfer coefficient influencing factors.

[0162] The current heat transfer coefficient influencing factor acquisition module is used to collect the current heat transfer coefficient influencing factor data in real time.

[0163] The water cooling heat transfer coefficient prediction module uses the trained BP neural network model to input the current value of the heat transfer coefficient influencing factor and output the corresponding water cooling heat transfer coefficient prediction value.

[0164] The mechanism model operation module inputs the predicted value of the water cooling heat transfer coefficient through the mechanism model and calculates the strip temperature.

[0165] The above embodiments are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the scope of protection of the present invention.

Claims

1. A coiling temperature calculation method based on neural network and mechanism model, characterized in that: The following steps are involved: Step S1: Divide the steel strip into sample segments of equal length, each sample segment is a control object, and divide the heat output roller into a number of control sub-segments according to the header tube bundle; Step S2: Obtaining historical production data corresponding to each control sub-segment and performing validity processing on the data, wherein the historical production data includes the historical water cooling heat transfer coefficient and the water cooling heat transfer coefficient influencing factor of the strip in the control sub-segment; Step S3: Taking the water cooling heat transfer coefficient influencing factor as input and the water cooling heat transfer coefficient of each control sub-segment as output, a neural network model is constructed, and the historical production data of each control sub-segment after effectiveness processing is used to train the model respectively. After the training is completed, a water cooling heat transfer coefficient prediction model corresponding to each control sub-segment is obtained; Step S4: obtaining the current value of the water cooling heat transfer coefficient influencing factor of the first control sub-segment corresponding to each sample segment when the current strip steel is running on the hot output roller table; Step S5: inputting the current value of the water cooling heat transfer coefficient influencing factor obtained in step S4 into the water cooling heat transfer coefficient prediction model of the control sub-segment to predict the water cooling heat transfer coefficient corresponding to the control sub-segment; Step S6: input the predicted value of the water cooling heat transfer coefficient obtained in step S5 into the mechanism model to calculate the strip temperature at the outlet of each header tube bundle.

2. The coiling temperature calculation method based on neural network and mechanism model according to claim 1, characterized in that: in, The factors affecting the water cooling heat transfer coefficient include steel type, thickness, strip steel entry temperature, speed, water temperature, ambient temperature, and flow rate.

3. The coiling temperature calculation method based on neural network and mechanism model according to claim 1, characterized in that: in, The data validity processing in step S2 is to determine whether the data is within the parameter range of normal strip operation on the hot output roller, and use the [μ-3σ,μ+3σ] confidence interval rule to detect and eliminate abnormal data.

4. The coiling temperature calculation method based on neural network and mechanism model according to claim 1, characterized in that: in, The neural network model in step S3 is a BP neural network, which includes an input layer, a hidden layer and an output layer. Each layer contains nodes. Parameters are input in the input layer, and the output value of each unit is obtained through the output of the hidden layer. If the output value exceeds the preset error value, reverse error transmission is performed, and the corresponding weight is adjusted according to the error value.

5. The coiling temperature calculation method based on neural network and mechanism model as claimed in claim 4, Features: Among them, the input parameters reach the nodes of the hidden layer from the input layer, pass through the activation function of each node, and then send the calculation information to the output layer through the hidden layer to get the result, as follows: Initially set N samples, where node i is the output node O i , corresponding to the input node x k , the node corresponding to the output layer is y k , the output of node i is O ik , now calculate the jth node of the lth layer. When the kth sample is input, the input of node j is: In formula (1) and (2), After the kth data is input into the l-1th layer, the calculation result of the jth node is obtained; After the kth data is input into the lth layer, the calculation result of the jth node is obtained; O is the weight of node i; = is the input of node i in layer l; The error function used is: In formula (3), E k = indicates the deviation between the actual value and the predicted value; Represents the actual calculation result of unit j; y lk Represents the corresponding network output; After calculation, the error is: make So In formula (4), (5), (6), E is the sum of calculated errors; N is the number of samples; It is the derivative of the actual value and the deviation value relative to the node input; Assume that j is the output node unit, then In formula (7), It is the input passed to the next layer, i.e. the l+1th layer; is the actual calculation result of unit j; y k is the predicted value of node j; is the measured value of node j; is the output function of the node; calculate To calculate back from the (l+1) layer; the calculation result at the mth unit in the (l+1) layer is: In formula (8), is the input of node m when it is the kth sample; is the weight of node m when it is the jth sample; is the derivative of the error of node m relative to the node input when it is the kth sample; get Then there is 6. The coiling temperature calculation method based on neural network and mechanism model according to claim 1, characterized in that: in, The steps of the BP neural network algorithm are as follows: 1) Select the initial value of the weight coefficient; 2) Repeat the forward and reverse process until the error index meets the accuracy requirement, where ε is the accuracy, that is: In formula (8), E is the total error; E k is the error of the kth sample; ε is the set model accuracy; N is the number of samples; k is the sample number; Forward process calculation for k = 1 to N: Calculate the nodes of each layer and k=2,...,N; Reverse process calculation: for each layer (l=L-1 to 2), for each node in each layer, calculate Correction weights: In formula (12) and (13), μ is the step size, w ij is the weight, and E calculates the accumulated error of the data.

7. The method for calculating coiling temperature based on a neural network and a mechanism model according to claim 6, characterized in that: in, The number of nodes in the input layer of the BP neural network is 8, which are steel type, thickness, entry temperature, target temperature, speed, water temperature, ambient temperature, and water flow rate; The number of nodes in the output layer is 1, which is the water cooling heat transfer coefficient; The number of nodes in the hidden layer is determined by an empirical formula: In formula (14), h is the number of hidden layer nodes, m is the number of input layer nodes, n is the number of output layer nodes, and a is a tuning constant between 1 and 10.

8. The method for calculating coiling temperature based on a neural network and a mechanism model according to claim 7, characterized in that: in, The weight adjustment rate of the BP neural network is 0.0035, the threshold adjustment rate is 0.009, the maximum number of training times is 800, the allowable error of a single sample is 0.001, and the allowable error of each iteration is 0.

004.

9. The coiling temperature calculation method based on neural network and mechanism model as claimed in claim 1, Features: The mechanism model in step S6 is a one-dimensional non-steady-state heat conduction process: In formula (15), T is the strip temperature; τ is the time; λ is the thermal conductivity of the strip; ρ is the material density; C is the specific heat capacity; x is the coordinate in the thickness direction of the strip; The initial boundary condition established is the final rolling temperature distribution: τ=0,T=T0 (16) In formula (16), T0 is the final rolling temperature; Considering the real-time working conditions of the strip on the hot output roller table to solve the boundary condition problem, the third type of boundary conditions known for air cooling radiation and water cooling convection are adopted: In formula (17), α is the heat transfer coefficient of the strip surface; T f is the surface cooling medium temperature of the strip; m is the total number of samples of the strip; The water cooling heat transfer coefficient model of the strip running on the hot output roller is: In formula (18), z 12 , z2 is the self-learning item of heat transfer coefficient; a1, a2, a3, V0 are model parameters; V is the real-time speed of the strip segment; V base is the finishing rolling speed; h is the strip thickness; T ω for water temperature; The heat transfer model without water cooling is: a=k air ·e·s·[(T+273) 2 +(T α +273) 2 ]·[(T+273)+(T α +273)] (19) In formula (19), K air is the correction coefficient; ε is the blackness coefficient of the steel plate; σ is the Stefan-Boltzmann constant; T a is the temperature of the air medium.

10. A coiling temperature calculation system based on a neural network and a mechanism model, applied to the coiling temperature calculation method based on a neural network and a mechanism model as claimed in any one of claims 1 to 9, characterized in that: include: A historical production data acquisition and storage module is used to acquire and store historical production data, including historical water cooling heat transfer coefficients and water cooling heat transfer coefficient influencing factors; The current heat transfer coefficient influencing factor acquisition module is used to collect the current heat transfer coefficient influencing factor data in real time; The water cooling heat transfer coefficient prediction module uses the trained BP neural network model to input the current value of the heat transfer coefficient influencing factor and output the corresponding water cooling heat transfer coefficient prediction value; The mechanism model operation module uses the mechanism model to input the predicted value of the water cooling heat transfer coefficient and calculate the strip temperature.

Images