Digital Twin Method and Device for Required Amount of Fused Magnesia Based on Edge-Cloud Collaboration

Abstract

The present application discloses a method and device for digital twin of the required amount of fused magnesia based on edge-cloud collaboration, which relates to the field of digital twin technology for complex furnace chambers of different batches. By correcting the non-linear part in the digital twin model of the required amount, it realizes the modeling of industrial process operation indicators whose dynamic characteristics change with the production process. The method includes: generating a dynamic model of the power of a multi-furnace batch group furnace; constructing a digital twin model of the required amount of a multi-furnace batch group furnace, including a linear model and a non-linear dynamic system; constructing an offline deep learning model of the non-linear dynamic system, constructing an online deep learning correction model and a cloud deep learning correction model; obtaining specified production process data to correct the online deep learning correction model and the cloud deep learning correction model, and completing the correction of the non-linear dynamic system; obtaining the sum value of the system output of the non-linear dynamic system and the model output of the linear model as the target required amount data, and sending the target required amount data to a display terminal for display.

Description

Technical Field

[0001] The present application relates to the technical field of digital twins for complex furnace chambers in different batches, and particularly to a digital twin method, device, computer device, and computer-readable storage medium for the required amount of fused magnesia based on edge-cloud collaboration. Background Art

[0002] Fused magnesia has a dense structure, high melting point, strong oxidation resistance, and strong insulation. It is the main raw material for refractory materials required in industries such as manufacturing and aerospace. The equipment for producing fused magnesia is the electric melting magnesia furnace, which is a major energy-consuming equipment. The required amount is the average power consumption value of the electric melting magnesia group furnace within a specified time period during the production process of fused magnesia, used to monitor the power consumption of the group furnace. There are 4 electric melting magnesia furnaces in the electric melting magnesia group furnace, and its daily production process is recorded as one furnace charge.

[0003] Currently, driven by the electricity price policy, fused magnesia production enterprises strive to improve the electricity utilization rate as much as possible on the premise of not exceeding the peak value of the required amount stipulated by the power department. To avoid exceeding the required amount, when it exceeds the set threshold, the control system randomly shuts down one electric melting magnesia furnace to bring the required amount back below the threshold.

[0004] In the process of implementing the present application, the applicant found that the related technology has at least the following problems:

[0005] The existing technology causes unnecessary power cuts during the peak of the required amount, resulting in a decrease in the grade and output of fused magnesia. To improve the electricity utilization rate and product quality, it is necessary to conduct required amount forecasting and intelligent control. However, the verification of forecasting and control methods in a real industrial environment is often restricted by the production environment and safety, and the experimental cost is relatively high. Therefore, it is urgent to establish a digital twin model for the required amount and build an experimental environment in the laboratory with the same dynamic characteristics of the required amount as the industrial site, which is convenient for the research of required amount intelligent control and required amount forecasting. Summary of the Invention

[0006] In view of this, the present application provides a digital twin method, device, computer device, and computer-readable storage medium for the required amount of fused magnesia based on edge-cloud collaboration, mainly aiming to solve the problem that the verification of forecasting and control methods in a real industrial environment is often restricted by the production environment and safety, and the experimental cost is relatively high.

[0007] According to the first aspect of the present application, a digital twin method for the required amount of fused magnesia based on edge-cloud collaboration is provided, and the method includes:

[0008] Establish a dynamic model of the current closed-loop control system for the smelting process of single-furnace fused magnesia, and generate a dynamic model of the power of the group furnace of electric melting magnesia furnaces based on the dynamic model of the current closed-loop control system;

[0009] Construct a multi-batch group furnace demand digital twin model based on the demand definition and the multi-batch group furnace power dynamic model. The multi-batch group furnace demand digital twin model includes a linear model and a non-linear dynamic system;

[0010] Construct an offline deep learning model for the non-linear dynamic system, and construct an online deep learning correction model and a cloud deep learning correction model based on the offline deep learning model;

[0011] Collect production process data of the electrofused magnesia furnace at the edge side, transmit the production process data to the cloud database, and calibrate the online deep learning correction model and the cloud deep learning correction model by obtaining specified production process data from the cloud database according to a preset self-calibration mechanism, so as to complete the calibration of the non-linear dynamic system;

[0012] Obtain the system output of the non-linear dynamic system and the model output of the linear model, use the sum value of the system output and the model output as the target demand data, and send the target demand data to a display terminal for display.

[0013] According to the second aspect of the present application, there is provided an electrofused magnesia demand digital twin device based on edge-cloud collaboration. The device includes:

[0014] A generation module, configured to establish a dynamic model of the current closed-loop control system for the single-furnace electrofused magnesia smelting process, and generate a multi-batch group furnace power dynamic model for the electrofused magnesia furnace based on the dynamic model of the current closed-loop control system;

[0015] A first construction model, configured to construct a multi-batch group furnace demand digital twin model based on the demand definition and the multi-batch group furnace power dynamic model. The multi-batch group furnace demand digital twin model includes a linear model and a non-linear dynamic system;

[0016] A second construction module, configured to construct an offline deep learning model for the non-linear dynamic system, and construct an online deep learning correction model and a cloud deep learning correction model based on the offline deep learning model;

[0017] A calibration module, which collects production process data of the electrofused magnesia furnace at the edge side, transmits the production process data to the cloud database, and calibrates the online deep learning correction model and the cloud deep learning correction model by obtaining specified production process data from the cloud database according to a preset self-calibration mechanism, so as to complete the calibration of the non-linear dynamic system;

[0018] An acquisition module, configured to obtain the system output of the non-linear dynamic system and the model output of the linear model, use the sum value of the system output and the model output as the target demand data, and send the target demand data to a display terminal for display.

[0019] According to the third aspect of the present application, a storage device is provided, on which a computer program is stored, and when the program is executed by a processor, the steps of the method described in any one of the above first aspects are implemented.

[0020] According to the fourth aspect of the present application, a computer storage medium is provided, including a storage device, a processor, and a computer program stored on the storage device and executable on the processor. When the processor executes the program, the steps of the method described in any one of the above first aspects are implemented.

[0021] By means of the above technical solutions, a digital twin method, device, computer device, and computer-readable storage medium for the required amount of fused magnesia based on edge-cloud collaboration are provided in the present application. First, a dynamic model of the current closed-loop control system for the single-furnace fused magnesia smelting process is established, and based on the dynamic model of the current closed-loop control system, a dynamic model of the multi-furnace group furnace power of the electric melting magnesia furnace is generated. Subsequently, a multi-furnace group furnace required amount digital twin model including a linear model and a nonlinear dynamic system is constructed according to the definition of the required amount and the dynamic model of the multi-furnace group furnace power. Next, an offline deep learning model of the nonlinear dynamic system is constructed, and an online deep learning correction model and a cloud deep learning correction model are constructed according to the offline deep learning model. Further, production process data of the electric melting magnesia furnace is collected at the edge side and transmitted to the cloud database. According to a preset self-correction mechanism, specified production process data is obtained from the cloud database to correct the online deep learning correction model and the cloud deep learning correction model, and the correction of the nonlinear dynamic system is completed. Finally, the system output of the nonlinear dynamic system and the model output of the linear model are obtained, and the sum value of the system output and the model output is used as the target required amount data, and the target required amount data is sent to the display terminal for display. The present invention combines mechanism with industrial big data, system identification with deep learning, and offline training with online correction, establishes a multi-furnace required amount digital twin model, and proposes a digital twin method for the required amount based on edge-cloud collaboration to correct the nonlinear part in the multi-furnace required amount digital twin model, realizing the modeling of the operation index of the industrial process whose dynamic characteristics change with the production process.

[0022] The above description is only an overview of the technical solutions of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features, and advantages of the present application more obvious and understandable, the specific embodiments of the present application are specifically given below. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Upon reading the following detailed description of the preferred embodiments, various other advantages and benefits will become apparent to those of ordinary skill in the art. The drawings are only for the purpose of illustrating the preferred embodiments and are not considered to be a limitation of the present application. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0024] Figure 1 Shows a schematic diagram of the process of a digital twin method for the required amount of fused magnesia based on edge-cloud collaboration provided by an embodiment of the present application;

[0025] Figure 2A Shows a schematic diagram of the process of a digital twin method for the required amount of fused magnesia based on edge-cloud collaboration provided by an embodiment of the present application;

[0026] Figure 2B Shows a schematic diagram of the determination of parameters of a digital twin method for the required amount of fused magnesia based on edge-cloud collaboration provided by an embodiment of the present application;

[0027] Figure 2C Shows a schematic diagram of the determination of parameters of a digital twin method for the required amount of fused magnesia based on edge-cloud collaboration provided by an embodiment of the present application;

[0028] Figure 2D Shows a schematic diagram of the determination of parameters of a digital twin method for the required amount of fused magnesia based on edge-cloud collaboration provided by an embodiment of the present application;

[0029] Figure 2E Shows a schematic diagram of the network structure of a digital twin method for the required amount of fused magnesia based on edge-cloud collaboration provided by an embodiment of the present application;

[0030] Figure 2F Shows a schematic diagram of the structure of a digital twin method for the required amount of fused magnesia based on edge-cloud collaboration provided by an embodiment of the present application;

[0031] Figure 2G Shows a schematic diagram of the determination of parameters of a digital twin method for the required amount of fused magnesia based on edge-cloud collaboration provided by an embodiment of the present application;

[0032] Figure 2H Shows a schematic diagram of the test results of a digital twin method for the required amount of fused magnesia based on edge-cloud collaboration provided by an embodiment of the present application;

[0033] Figure 2I Shows a schematic diagram of the test results of a digital twin method for the required amount of fused magnesia based on edge-cloud collaboration provided by an embodiment of the present application;

[0034] Figure 2J Shows a schematic diagram of the test results of a digital twin method for the required amount of fused magnesia based on edge-cloud collaboration provided by an embodiment of the present application;

[0035] Figure 2K Shows a schematic diagram of the test results of a digital twin method for the required amount of fused magnesia based on edge-cloud collaboration provided by an embodiment of the present application;

[0036] Figure 2L Shows a schematic diagram of the test results of a digital twin method for the required amount of fused magnesia based on edge-cloud collaboration provided by an embodiment of the present application;

[0037] Figure 3 Shows a schematic structural diagram of a digital twin device for the required amount of fused magnesia based on edge-cloud collaboration provided by an embodiment of the present application;

[0038] Figure 4 Shows a schematic structural diagram of a device of a computer device provided by an embodiment of the present application. Specific embodiments

[0039] Hereinafter, exemplary embodiments of the present application will be described in more detail with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present application can be more thoroughly understood and the scope of the present application can be fully conveyed to those skilled in the art.

[0040] An embodiment of the present application provides a digital twin method for the required amount of fused magnesia based on edge-cloud collaboration, as Figure 1 shown, the method includes:

[0041] 101. Establish a dynamic model of the current closed-loop control system for the single-furnace fused magnesia smelting process, and generate a dynamic model of the multi-furnace group furnace power of the fused magnesia furnace based on the dynamic model of the current closed-loop control system.

[0042] 102. Construct a multi-furnace group furnace required amount digital twin model according to the required amount definition and the multi-furnace group furnace power dynamic model. The multi-furnace group furnace required amount digital twin model includes a linear model and a non-linear dynamic system.

[0043] 103. Construct an offline deep learning model for the non-linear dynamic system, and construct an online deep learning correction model and a cloud deep learning correction model according to the offline deep learning model.

[0044] 104. Collect production process data of the fused magnesia furnace at the edge side, transmit the production process data to the cloud database, and calibrate the online deep learning correction model and the cloud deep learning correction model according to the preset self-calibration mechanism by obtaining specified production process data from the cloud database to complete the calibration of the non-linear dynamic system.

[0045] 105. Obtain the system output of the nonlinear dynamic system and the model output of the linear model, use the sum value of the system output and the model output as the target demand data, and send the target demand data to the display terminal for display.

[0046] The method provided by the embodiment of the present application can first establish a dynamic model of the current closed-loop control system for the single-furnace fused magnesia smelting process. Based on the dynamic model of the current closed-loop control system, a multi-furnace group furnace power dynamic model of the fused magnesia furnace is generated. Subsequently, according to the demand definition and the multi-furnace group furnace power dynamic model, a multi-furnace group furnace demand digital twin model including a linear model and a nonlinear dynamic system is constructed.

[0047] Next, an offline deep learning model of the nonlinear dynamic system is constructed, and an online deep learning correction model and a cloud deep learning correction model are constructed according to the offline deep learning model. Further, the production process data of the fused magnesia furnace is collected at the edge side and transmitted to the cloud database. According to the preset self-correction mechanism, the specified production process data is obtained from the cloud database to correct the online deep learning correction model and the cloud deep learning correction model, and the correction of the nonlinear dynamic system is completed. Finally, obtain the system output of the nonlinear dynamic system and the model output of the linear model, use the sum value of the system output and the model output as the target demand data, and send the target demand data to the display terminal for display. The present invention combines mechanism with industrial big data, system identification with deep learning, and offline training with online correction, establishes a multi-furnace demand digital twin model, and proposes a demand digital twin method based on edge-cloud collaboration to correct the nonlinear part in the multi-furnace demand digital twin model, realizing the modeling of the operation index of the industrial process whose dynamic characteristics change with the production process.

[0048] The fused magnesia furnace is a large energy-consuming device for producing fused magnesia, and the power consumption cost accounts for more than 60% of the total cost. The fused magnesia furnace adopts the submerged arc method and uses a feedback control system to control the melting current. While melting, materials are added, and magnesite ore is converted into fused magnesia products. In order to avoid the impact of random power-off of the control system on the production process and product quality when the demand exceeds the limit, and also to avoid the demand exceeding the demand peak specified by the power department, it is necessary to establish a demand digital twin model and build an experimental environment in the laboratory with the same demand dynamic characteristics as the industrial site, which is convenient for demand forecasting and demand control research and is of great significance for energy conservation. The fused magnesia production process is multi-furnace, that is, after one furnace production is completed, after a fixed interval, the next furnace production continues. Since the same production system is used and the properties of the raw ore vary with different furnace batches, the smelting characteristics of the current furnace batch have both some characteristics of the historical furnace batches and change with the change of the raw ore of the current furnace batch. Therefore, in order to build an experimental environment with the same demand dynamic characteristics as the industrial site, the embodiment of the present application provides a fused magnesia demand digital twin method based on edge-cloud collaboration, such asFigure 2A As shown in the figure, the method includes:

[0049] 201. Establish a dynamic model of the current closed-loop control system for the single-furnace fused magnesia smelting process. Based on the dynamic model of the current closed-loop control system, generate a dynamic model of the multi-furnace group furnace power for the electric melting magnesia furnace.

[0050] In the embodiment of the present application, establish a current dynamic model of the fused magnesia smelting process represented by a second-order linear term and an unknown high-order non-linear term. Control the industrial process based on the melting current feedback control system, establish a dynamic model of the current closed-loop control system, and then realize the establishment of a dynamic model of the multi-furnace group furnace power with the current of the closed-loop control system as the input and the power as the output under multiple furnaces (the number of furnace times is S) for the electric melting magnesia furnace. The specific process of generating the dynamic model of the multi-furnace group furnace power is as follows:

[0051] First, use the rotation direction and frequency u of the single-furnace electric melting magnesia furnace motor i as the input data, and the electrode current y i as the output data to establish the current dynamic model of the i-th electric melting magnesia furnace:

[0052]

[0053] where L i1 is the winding inductance of the transformer and the reactor; L i2 is the short-circuit inductance; U is the phase voltage; R i1 is the active resistance of the transformer and the impedance; R i2 is the short-circuit active resistance; p is the number of pole pairs of the motor; h(·) is the molten pool height; Δh ie (·) is the change in the electrode height; a and b are arc constants.

[0054] Next, discretize the current dynamic model based on the Euler method, and represent the current dynamic model as the sum of a second-order linear term and a high-order non-linear term, that is, A i (z -1 )y i (k) = B i (z -1 )u i (k - 1) + v' i (k). It should be noted that A i (z -1 ) = 1 + a i1 z -1 + a i2 z -2 ; B i (z -1 ) = b i0 ; y i (k) is the melting current of the i-th electric melting magnesia furnace; ui (k - 1) is the rotation direction and frequency of the i-th frequency conversion motor of the electrofused magnesia furnace; v′ i (k) is an unknown high-order non-linear term.

[0055] Subsequently, a melting current feedback control system is used to control the discretized current dynamic model to obtain the dynamic model of the current closed-loop control system. The melting current feedback control system is expressed as (1 - z -1 )u i (k) = G i (z -1 )e i (k), it should be noted that G i (z -1 ) = g i0 + g i1 z -1 + g i2 z -2 ; e i (k) = y i (k) - y * ; y * is the set value of the melting current. The dynamic model of the current closed-loop control system is specifically shown in Formula 1 below:

[0056] Formula 1: T i (z -1 )y i (k) = B i (z -1 )G i (z -1 )y * + (1 - z -1 )v i '(k)

[0057] Among them, T i (z -1 ) is the closed-loop system characteristic polynomial selected for the design of the feedback control system parameters. Specifically, T i (z -1 ) = 1 + t i1 z -1 + t i2 z -2 + t i3 z -3 .

[0058] According to the mechanism relationship between power and current, the power of a single furnace electrofused magnesia furnace is defined as shown in Formula 2 below:

[0059] Formula 2:

[0060] Among them, U is the smelting voltage, y i(k) is the average value of the smelting currents of the three-phase electrodes of the i-th electrofused magnesia furnace at the k-th moment within a furnace charge, is the power factor.

[0061] Based on the definition of the single-furnace power of the electrofused magnesia furnace and the dynamic model of the current closed-loop control system, a power dynamic model of the i-th electrofused magnesia furnace within a furnace charge with the current output from the dynamic model of the current closed-loop control system as the input is constructed, that is, the single-furnace power dynamic model for a single furnace charge, which can be specifically expressed as:

[0062]

[0063] where, b 0 G i (z -1 ) = b 0 (g i0 + g i1 + g i2 ) = d i0 , p * is the power corresponding to the melting current set value y * . T i * (z -1 ) = T i (z -1 ) - 1.

[0064] Since there are four electrofused magnesia furnaces in the group furnace power monitoring system within each furnace charge, the group furnace power is defined as shown in Formula 3 below:

[0065] Formula 3:

[0066] Furthermore, using the group furnace model parameters t 1 ′, t 2 ′, t 3 ′, d 0 ′ to replace the single-furnace model parameters in the single-furnace power dynamic model for a single furnace charge the power dynamic model of the electrofused magnesia group furnace within a furnace charge can be obtained as:

[0067]

[0068] where, v′(k) consists of the non-linear term and the model error generated by introducing t 1 ′, t 2 ′, t 3 ′, d 0 ′.

[0069] According to the current dynamic model, due to the unevenness of the raw ore, the differences in raw ore between different furnace batches, and the variation of the properties of the raw ore within the same furnace batch during the melting and feeding processes, the model parameters L(·), R(·), h(·), and Δh(·) vary with the furnace batch S, which in turn causes the nonlinear term v i ′(k) to vary with the furnace batch S. In specific application scenarios, constants t 1 , t 2 , t 3 , d 0 that do not vary with the furnace batch are used to replace the model parameters t 1 ′, t 2 ′, t 3 ′, d 0 ′ that vary with the furnace batch, and the multi-furnace batch group furnace power dynamic model is obtained as shown in Equation 4 below:

[0070] Equation 4:

[0071] where ψ(k - 1) = [y(k - 1), y(k - 2), y(k - 3), y * ; θ = [-t 1 , -t 2 , -t 3 , d 0 ; ψ(k - 1)θ is an identifiable model; the unknown nonlinear term v T ′(k) consists of the nonlinear term v′(k) and the model error generated by introducing the constants t S , t 1 , t 2 , t 3 , d 0 that do not vary with the furnace batch. Specifically, the unknown nonlinear term v S ′(k) can be expressed as shown in Equation 5 below:

[0072] Equation 5: v S ′(k) = v′(k) + (t 1 - t 1 ′)y(k - 1) + (t 2 - t 2 ′)y(k - 2) + (t 3 - t 3 ′)y(k - 3) + (d 0 - d 0 ′)y *

[0073] Based on the above steps, using the dynamic model of the current closed-loop control system and the mechanism model between power and current, a multi-furnace power dynamic model of fused magnesia group furnaces consisting of an identifiable model and an unknown nonlinear system is established, and then the power data of the multi-furnace fused magnesia group furnaces is obtained.

[0074] 202. Construct a multi-furnace group furnace demand digital twin model based on the demand definition and the multi-furnace group furnace power dynamic model. The multi-furnace group furnace demand digital twin model includes a linear model and a nonlinear dynamic system.

[0075] In the embodiment of the present application, by estimating the parameters of the identifiable model in the multi-furnace group furnace power dynamic model, and then constructing the unknown nonlinear system of the multi-furnace group furnace power dynamic model, finally realizing the construction of the multi-furnace demand digital twin model. The specific process of establishing the multi-furnace demand digital twin model is as follows:

[0076] First, use the least squares identification algorithm to identify the identifiable model in the multi-furnace group furnace power dynamic model, estimate the identification parameter θ of the identifiable model, and obtain the parameter estimation value of the identification parameter Among them, the parameter identification equation is p(k) = ψ(k - 1)θ. The parameter identification equation is expressed as Y(k) = X(k - 1)θ, where the output vector Y(k) can be expressed as Y(k) = [p(k), p(k - 1),..., p(k - N S + 1)] T , and the input matrix can be expressed as:

[0077]

[0078] The objective function of the least squares algorithm is Among them The estimated value of the parameter θ of the least squares algorithm is

[0079] Subsequently, the model error brought by the parameter estimation value is used as the identification error of the identifiable model. The sum of the identification error and the unknown nonlinear term v′ S (k) of the multi-furnace group furnace power dynamic model is used as the unknown nonlinear system Specifically, it is shown in Formula 6 below:

[0080] Formula 6:

[0081] Demand is calculated based on the group furnace power. The demand value is the average value of the group furnace power p(k) at the current moment and the previous m - 1 moments. The demand definition is specifically shown in Formula 7:

[0082] Formula 7:

[0083] Among them, is the demand value at the k-th moment, and p(k) is the power value of the group of furnaces at the k-th moment.

[0084] Furthermore, the fused magnesia production enterprise calculates the demand value using 30 power sampling periods. Based on the demand definition, unknown nonlinear system, and multi-furnace group furnace power dynamic model, a multi-furnace group furnace demand digital twin model is calculated. It should be noted that the multi-furnace group furnace demand digital twin model is the sum of a nonlinear dynamic system and a linear model. The specific process of constructing the multi-furnace group furnace demand digital twin model can be realized based on the following formula 8:

[0085] Formula 8:

[0086] Where is the linear model of the demand digital twin model ; is the nonlinear dynamic system of the demand digital twin model , and In the specific application process, the unknown nonlinear function f(·) is used to represent Among them, f(·) is a nonlinear function with unknown changes, and n is the unknown order of the system, as shown in the following formula 9:

[0087] Formula 9:

[0088] Through the above steps, the identification algorithm is used to estimate the parameters of the identifiable model in the multi-furnace fused magnesia furnace power dynamic model using the current and power data of one furnace collected at the edge side. Furthermore, the identification error of the identifiable model and the unknown nonlinear terms under multi-furnaces are combined into an unknown nonlinear system with unknown model structure and order. Finally, according to the demand definition, unknown nonlinear system, and power dynamic model under multi-furnaces of fused magnesia group furnaces, a multi-furnace demand digital twin model composed of a linear model and a nonlinear dynamic system is established.

[0089] 203. Construct an offline deep learning model for the nonlinear dynamic system, and construct an online deep learning correction model and a cloud deep learning correction model according to the offline deep learning model.

[0090] In the embodiment of the present application, a long short-term memory multi-layer neural network (LSTM) network architecture is used to establish an offline deep learning model for the unknown nonlinear dynamic system. It should be noted that the output of the offline deep learning model is the predicted value of the nonlinear dynamic system ; Then, an online deep learning calibration model and a cloud deep learning calibration model are established using the same network architecture as the offline deep learning model. The specific processes of establishing the offline deep learning model, the online deep learning calibration model, and the cloud deep learning calibration model are as follows:

[0091] First, select 1 layer for the number of network layers, and use N S groups of input-output data to form sample data. Specifically, the input-output data is y(i), N S which can be 5000. Use a training algorithm to make the label data and the actual output of the offline deep learning model have as small an error as possible. Determine the number of neurons n, the number of nodes in a single neuron h, and the number of network layers L. The objective function of the training algorithm is shown in Equation 10 below:

[0092] Equation 10:

[0093] Where the connection weight is a 1×h row vector, is the bias parameter, and h 1 (k) is the output of the nth neuron.

[0094] Among them, h 1 (k) T = o 1 (k) ⊙ tanh(C 1 (k)), tanh(x) = (1 - e -2x ) / (1 + e -2x ), the output gate o 1 (k) and the state gate C 1 (k) are respectively o 1 (k) = σ(W 1 o ·x 1 (k) T + b 1 o ), in the output gate o 1 (k), σ(x) = 1 / (1 + e -x ), the input of the nth neuron in the state gate C 1 (k), the forget gate f 1 = σ(W 1 f ·x 1 (k) T + b 1 f ), the input gate i1 = σ(W 1 in ·x 1 (k) T + b 1 in ), state candidate value W f , W in , W c is a matrix of h × (h + 3), b 1 f , b 1 in , b 1 c is a column vector of h × 1.

[0095] The connection weights and bias parameters are trained using the gradient descent algorithm, and the output gate connection weight W 1 o and the weight b 1 o The training algorithm for is

[0096] where α is the learning rate. The same training algorithm is then used to determine the other connection weights and biases of the model.

[0097] Next, in order to identify the structure of the deep learning model, the mean absolute error is introduced:[[]]

[0098]

[0099] Specifically, the actual current and demand data of one heat are used as input and output data, and the number of neurons n of the LSTM network is first determined. Let n = 1, 2, 3, 4,... and calculate the mean absolute error (MAE) of the model output error . When the MAE of is the smallest, the value of the number of neurons n is determined. Then, the number of neurons n is fixed, and the number of nodes h = 1, 2, 3, 4,... is set and the mean absolute error of the model output error is calculated. When the MAE of is the smallest, the value of the number of nodes h is determined. Finally, with n and h of the deep learning model fixed, by increasing the number of network layers L = 1, 2, 3, 4,... the mean absolute error of the model output error is calculated respectively. When the MAE of is the smallest, the value of the number of network layers L is determined. Finally, as Figures 2B to 2D shown, it can be obtained that the number of neurons n is 34, the number of nodes h in a single neuron hidden layer is 238, and the number of network layers L is 3, determining the network architecture of the offline deep learning model, specifically asFigure 2E as shown

[0100] Finally, an online deep learning calibration model and a cloud deep learning calibration model are constructed with the same number of single neuron nodes, number of network layers, and number of network neurons as the offline deep learning model. Denote the output of the online deep learning calibration model as Denote the output of the cloud deep learning calibration model as

[0101] Through the above steps, an offline deep learning model is constructed using a long short-term memory multi-layer neural network (LSTM). The actual current and demand data of one furnace are used as input and output data to determine the number of neurons n, the number of unit nodes h, and the number of multi-layer neural network layers L of the offline deep learning model, thereby realizing the construction of the online deep learning calibration model and the cloud deep learning calibration model.

[0102] 204. Collect the production process data of the electrofused magnesia furnace at the edge side, transmit the production process data to the cloud database, and calibrate the online deep learning calibration model and the cloud deep learning calibration model according to the preset self-calibration mechanism by obtaining the specified production process data from the cloud database, thereby completing the calibration of the nonlinear dynamic system.

[0103] In the embodiment of the present application, the demand digital twin model consists of a linear model and a nonlinear dynamic system, and the nonlinear dynamic system consists of an online deep learning calibration model, a self-calibration mechanism, and a cloud deep learning calibration model. The nonlinear dynamic system is calibrated in a way of edge-cloud collaboration. Specifically, the production process data of the electrofused magnesia furnaces are collected in real time at the edge side and uploaded to the cloud database. All layers of weights and bias parameters of the cloud deep learning calibration model are calibrated by extracting the production process data updated in real time for all furnace batches in the cloud database. Then, fix the number of single neuron nodes, the number of network layers, and the number of network neurons of the online deep learning calibration model, and calculate the time series window length of the online deep learning calibration model using the production data of the same offline furnace batch. Subsequently, extract the production process data updated online within the same furnace batch from the cloud database, and calibrate the weights and bias parameters of the fully connected layer of the online deep learning calibration model based on the comparison result between the data volume of the production process data and the time series window length. By calculating the error upper bound of the model output of the nonlinear dynamic system, a preset self-calibration mechanism is obtained according to the error upper bound. Finally, according to the preset self-calibration mechanism, the cloud deep learning calibration model is used to calibrate the online deep learning calibration model.

[0104] In the actual application process, such as Figure 2FAs shown in the figure, the edge side is a data acquisition and transmission device for real-time acquisition of current, power, and demand data; the cloud side is a cloud computing device, such as an NVIDIA server, for correcting all layer weights and bias parameters of the cloud deep learning correction model according to the production process data updated in real time for all furnace batches. The edge side is an edge computing device, such as a Dongtu industrial server, for correcting the weights and bias parameters of the fully connected layer of the online deep learning correction model according to the production process data with an online updated window length of N during the smelting process of the current furnace batch. When the self-correction mechanism condition is met, download all layer weights and biases of the cloud deep learning correction model to replace all layer weights and biases of the online deep learning correction model. It should be noted that in the actual application process, the output after the correction of the non-linear dynamic system is reflected in the output of the online deep learning correction model. Therefore, finally, add the output value of the demand linear model to the output value of the online deep learning correction model to output the current demand value.

[0105] First, the production process data of the fused magnesia group furnace is collected in real time at the edge side and uploaded to the cloud database. Subsequently, all layer weights and bias parameters of the cloud deep learning correction model are corrected by extracting the production process data updated in real time for all furnace batches in the cloud database. Specifically, each furnace batch has N S groups of input-output data, and use N S (k) groups of real-time updated production process data y S-1 (i), y 1 (i), y S (j), to correct all layer weights and biases of the cloud deep learning correction model in real time. As the smelting process progresses, the number of data N S (k) continuously increases.

[0106] It should be noted that the objective function of the training algorithm is where N S (k) represents the number of input-output data pairs used by the cloud deep learning correction model. In fact, N S (k) = (S - 1)N S + k, where S represents the current furnace batch number, k represents the kth sampling moment of the current furnace batch, the sampling period is 7 seconds, and N S (k) continuously increases as the smelting process progresses, increasing by 1 every 7 seconds.

[0107] Further, fix the number of single neuron nodes, the number of network layers, and the number of network neurons in the online deep learning calibration model, and calculate the time series window length using the production data of the same heat in the offline mode. Specifically, fix n, h, and L of the deep learning model, and calculate the output error of the online deep learning calibration model by increasing the time series window length N = 1, 2, 3, 4,... respectively. The average absolute error of is calculated. When Figure 2G the MAE of

[0108] is the smallest, determine the value of the time series window length N. Specifically, the time series window length is 1450, and the training process is as Figure 2G shown.

[0108] Then, extract the production process data updated online within the same heat from the cloud database to calibrate the online deep learning calibration model. Specifically, compare the data volume of the production process data with the time series window length. When the comparison result indicates that the data volume of the production process data within the same heat does not meet the time series window length, do not calibrate the weight and bias parameters of the fully connected layer of the online deep learning calibration model. When the comparison result indicates that the data volume of the production process data meets the time series window length, keep the heat S unchanged, and according to the production process data y(k - 1),..., y(k - N - 1) with the online updated time series window length of N, keep the weight and bias parameters of the hidden layer of the online deep learning calibration model unchanged, and calibrate the weight and bias parameters of the fully connected layer in real time.

[0109] The output of the online deep learning calibration model is In the formula, the connection weight is a 1×h row vector, and is the bias. Fix the connection weights and biases of layers 1 to L unchanged, and only calibrate the connection weights and bias parameters in the output of the online deep learning calibration model online.

[0110] The objective function of the online calibration is: The calibration algorithm for the connection weight is The calibration algorithm for the bias parameter is where

[0111] Next, calculate the upper bound of the error of the model output of the nonlinear dynamic system. According to the upper bound of the error, obtain the preset self - calibration mechanism, and based on the preset self - calibration mechanism, use the cloud - based deep learning calibration model to calibrate the online deep learning calibration model. Specifically, use the model accuracy evaluation indicators RMSE and goodness of fit R 2, the rising trend accuracy TPR and the falling trend accuracy TNR, establish a self - calibration mechanism between the cloud deep - learning calibration model and the online deep - learning calibration model. That is, set the upper bound of the error of the model output of the non - linear dynamic system according to the experimental results (RMSE = δ 1 , R 2 = δ 2 , TPR = δ 3 , TNR = δ 4 ). Use 5000 groups of data in the validation set to calculate the self - calibration mechanism evaluation indexes RMSE, R 2 , TPR and TNR. After calculation, the four indexes are RMSE = 12.72, R 2 = 99.62%, TPR = 0.93, TNR = 0.92. Then, according to Formulas 11 to 14, calculate the model accuracy evaluation indexes of the online deep - learning calibration model.

[0112] Formula 11:

[0113] Formula 12:

[0114] where k is the sampling period (k = 1, 2,...), is the output value of the online deep - learning calibration model, and the output mean value of the online deep - learning calibration model the mean value of the true value

[0115] Formula 13:

[0116] Formula 14:

[0117] where TP 1 (k) represents the number of rising trends in the model output of the online deep - learning calibration model; TN 1 (k) represents the number of falling trends in the model output of the online deep - learning calibration model; ALLP(k) represents the total number of actual rising trends; ALLN(k) represents the total number of actual falling trends. The calculation formulas of TP 1 (k), TN 1 (k), ALLP(k) and ALLN(k) are as follows:

[0118]

[0119]

[0120]

[0121]

[0122] Among them,

[0123]

[0124]

[0125]

[0126]

[0127] The model accuracy evaluation indicators of the deep learning correction model in the cloud are calculated in the same way, denoted as RMSE 2 (k), R 2 2 (k), TPR 2 (k), TNR 2 (k). The model accuracy evaluation indicators of the online deep learning correction model and the deep learning correction model in the cloud are detected respectively, and the model accuracy evaluation indicators of the online deep learning correction model and the deep learning correction model in the cloud are compared with the preset self-correction mechanism to generate a comparison result. When the model accuracy evaluation indicators of the online deep learning correction model (RMSE 1 (k) ≥ 12.72 or R 1 2 (k) ≤ 99.62% or TPR 1 (k) ≤ 0.93 or TNR 1 (k) ≤ 0.92) and the model accuracy evaluation indicators of the deep learning correction model in the cloud (RMSE 2 (k) < 12.72, R 2 2 (k) > 99.62%, TPR 2 (k) > 0.93, TNR 2 (k) > 0.92), a comparison result indicating model correction is generated, and then all layer weights and bias parameters of the deep learning correction model in the cloud are downloaded to replace all layer weights and bias parameters of the online deep learning correction model.

[0128] 205. Obtain the system output of the nonlinear dynamic system and the model output of the linear model Take the sum value of the system output and the model output as the target demand data and send the target demand data to the display terminal for display.

[0129] In the embodiments of the present application, the system output of the non-linear dynamic system and the model output of the linear model are obtained, and the sum value of the system output and the model output is used as the target demand data, and the target demand data is sent to the display terminal for display, realizing the digital twin of the demand and providing an environmental basis for demand forecasting and intelligent control.

[0130] The present application is a method for digital twin of demand in the production process of fused magnesia under multi-hearth based on edge-cloud collaboration. Based on the production process data collected in real time for the same hearth, real-time simulation is performed on the online deep learning calibration model and the cloud deep learning calibration model, and the comparison results are as Figures 2H to 2I shown. Figure 2H This is the accuracy test result of the online deep learning calibration model, and the tested RMSE is 13.678; Figure 2I This is the accuracy test result of the cloud deep learning calibration model, and the tested RMSE is 10.589. According to the test results, it can be seen that the model accuracy of the cloud deep learning calibration model is higher than that of the online deep learning calibration model. Therefore, when the accuracy of the online deep learning calibration model does not meet the requirements, a self-calibration mechanism is adopted to download all the layer weights and biases of the cloud deep learning calibration model to replace all the layer weights and biases of the online deep learning calibration model.

[0131] Based on data from different hearths, real-time simulation is performed on the demand digital twin model, and 5000 groups of data from one hearth collected offline are used to test the accuracy of the model. The test results are as Figures 2J to 2L shown. Figure 2J This is the accuracy trend chart of the demand digital twin model for real-time simulation using data from different hearths. When real-time simulation is performed using single-hearth data and 15-hearth data, the test results of the demand digital twin model are respectively Figure 2K and Figure 2L shown (for clarity, only 150 groups are shown in the figure). From Figure 2J it can be seen that as more hearth data is used for real-time simulation, the root mean square error of the output error of the demand digital twin model shows a downward trend. From Figures 2J to 2L it can be seen that the model accuracy of the demand digital twin model for real-time simulation using multi-hearth data is significantly better than that of the demand digital twin model for real-time simulation using single-hearth data. The RMSE of the output error of the demand digital twin models for real-time simulation using single-hearth data and 15-hearth data are 14.293 and 6.508 respectively, and the model accuracy has increased by 54.46%. Therefore, as the number of hearths increases and the data continues to increase, the model learns more dynamic characteristics, and through edge-cloud collaborative calibration, the model accuracy continues to increase, meeting the requirements of digital twin.

[0132] The method provided by the embodiment of the present application may first establish a dynamic model of the current closed-loop control system for the single-furnace fused magnesia smelting process, and generate a multi-furnace group furnace power dynamic model for the fused magnesia furnace based on the dynamic model of the current closed-loop control system. Subsequently, a multi-furnace group furnace demand digital twin model including a linear model and a non-linear dynamic system is constructed according to the demand definition and the multi-furnace group furnace power dynamic model. Next, an offline deep learning model of the non-linear dynamic system is constructed, and an online deep learning correction model and a cloud deep learning correction model are constructed according to the offline deep learning model. Further, production process data of the fused magnesia furnace is collected at the edge side, and the production process data is transmitted to the cloud database. According to a preset self-correction mechanism, specified production process data is obtained from the cloud database to correct the online deep learning correction model and the cloud deep learning correction model, and the correction of the non-linear dynamic system is completed. Finally, the system output of the non-linear dynamic system and the model output of the linear model are obtained, and the sum value of the system output and the model output is used as the target demand data, and the target demand data is sent to the display terminal for display. The present invention combines mechanism with industrial big data, system identification with deep learning, and offline training with online correction, establishes a multi-furnace demand digital twin model, and proposes a demand digital twin method based on edge-cloud collaboration to correct the non-linear part in the multi-furnace demand digital twin model, realizing the modeling of industrial process operation indicators whose dynamic characteristics change with the production process.

[0133] Further, as Figure 1 a specific implementation of the method, the embodiment of the present application provides an edge-cloud collaborative fused magnesia demand digital twin device, as Figure 3 shown. The device includes: a generation module 301, a first construction module 302, a second construction module 303, a correction module 304, and an acquisition module 305.

[0134] The generation module 301 is used to establish a dynamic model of the current closed-loop control system for the single-furnace fused magnesia smelting process, and generate a multi-furnace group furnace power dynamic model for the fused magnesia furnace based on the dynamic model of the current closed-loop control system;

[0135] The first construction model 302 is used to construct a multi-furnace group furnace demand digital twin model according to the demand definition and the multi-furnace group furnace power dynamic model, and the multi-furnace group furnace demand digital twin model includes a linear model and a non-linear dynamic system;

[0136] The second construction module 303 is used to construct an offline deep learning model of the non-linear dynamic system, and construct an online deep learning correction model and a cloud deep learning correction model according to the offline deep learning model;

[0137] The calibration module 304 is used to collect the production process data of the electric fused magnesia furnace at the edge side, transmit the production process data to the cloud database, and calibrate the online deep learning calibration model and the cloud deep learning calibration model according to a preset self-calibration mechanism by obtaining specified production process data from the cloud database, so as to complete the calibration of the non-linear dynamic system;

[0138] The acquisition module 305 is used to acquire the system output of the non-linear dynamic system and the model output of the linear model, take the sum value of the system output and the model output as the target demand data, and send the target demand data to a display terminal for display.

[0139] In a specific application scenario, the generation module 301 is used to use the rotation direction and frequency of the single-furnace electric fused magnesia furnace motor as input data and the electrode current as output data to establish a current dynamic model; discretize the current dynamic model based on the Euler method, and represent the current dynamic model as the sum value of a second-order linear term and a high-order non-linear term; use a melting current feedback control system to control the current dynamic model to obtain the current closed-loop control system dynamic model; establish a single-furnace single-furnace power dynamic model with the current output by the current closed-loop control system dynamic model as the input; replace the single-furnace model parameters in the single-furnace single-furnace power dynamic model with group-furnace model parameters to obtain a single-furnace single-group-furnace power dynamic model; use a constant that does not change with the furnace number to replace the group-furnace model parameters that change with the furnace number to obtain the multi-furnace group-furnace power dynamic model.

[0140] In a specific application scenario, the first model construction module 302 is used to identify the identifiable model in the multi-furnace group-furnace power dynamic model by using the least squares identification algorithm, and estimate the identification parameters of the identifiable model; calculate the parameter estimation values of the identification parameters, and determine the identification error of the identifiable model generated by the parameter estimation values; take the sum value of the identification error and the unknown non-linear term in the multi-furnace group-furnace power dynamic model as the unknown non-linear system; calculate a multi-furnace group-furnace demand digital twin model based on the demand definition, the unknown non-linear system and the multi-furnace group-furnace power dynamic model, and the multi-furnace group-furnace demand digital twin model is the sum of a non-linear dynamic system and a linear model.

[0141] In a specific application scenario, the second construction module 303 is used to obtain the actual current and demand data of the same furnace number as sample data, and use the sample data to determine the number of single neuron nodes, the number of network layers, and the number of network neurons of a multi-layer long short-term memory (LSTM) network architecture to obtain the offline deep learning model; use the number of single neuron nodes, the number of network layers, and the number of network neurons to construct the online deep learning calibration model and the cloud deep learning calibration model.

[0142] In a specific application scenario, the calibration module 304 is configured to: in real time collect the production process data of the electrofused magnesia cluster furnace at the edge side, and upload the production process data to the cloud database; extract all the production process data updated in real time for all furnace batches in the cloud database to calibrate all layer weights and bias parameters of the cloud deep learning calibration model; fix the number of single neuron nodes, the number of network layers and the number of network neurons corresponding to the online deep learning calibration model, and calculate the time series window length of the online deep learning calibration model by using the production data of the same furnace batch offline; extract the production process data updated online within the same furnace batch in the cloud database to calibrate the weights and bias parameters of the fully connected layer of the online deep learning calibration model; calculate the upper bound of the error of the model output of the non-linear dynamic system, and obtain the preset self-calibration mechanism according to the upper bound of the error; and perform model calibration on the online deep learning calibration model by using the cloud deep learning calibration model according to the preset self-calibration mechanism.

[0143] In a specific application scenario, the calibration module 304 is configured to: when the data volume of the production process data within the same furnace batch meets the time series window length, keep the weights and bias parameters of the hidden layer of the online deep learning calibration model unchanged, and calibrate the weights and bias parameters of the fully connected layer in real time according to the production process data of the time series window length updated online; when the data volume of the production process data within the same furnace batch does not meet the time series window length, do not calibrate the weights and bias parameters of the fully connected layer of the online deep learning calibration model, but still follow the preset self-calibration mechanism.

[0144] In a specific application scenario, the calibration module 304 is configured to: respectively detect the model accuracy evaluation indexes of the online deep learning calibration model and the cloud deep learning calibration model, compare the model accuracy evaluation indexes of the online deep learning calibration model and the cloud deep learning calibration model with the preset self-calibration mechanism, and generate a comparison result; when the comparison result indicates a calibration model, replace all layer weights and bias parameters of the online deep learning calibration model with all layer weights and bias parameters of the cloud deep learning calibration model.

[0145] The device provided by the embodiment of the present application can first establish a dynamic model of the current closed-loop control system for the single-furnace fused magnesia smelting process, and generate a multi-furnace group furnace power dynamic model for the electric melting magnesia furnace based on the current closed-loop control system dynamic model. Subsequently, a multi-furnace group furnace demand digital twin model including a linear model and a non-linear dynamic system is constructed according to the demand definition and the multi-furnace group furnace power dynamic model. Next, an offline deep learning model of the non-linear dynamic system is constructed, and an online deep learning correction model and a cloud deep learning correction model are constructed according to the offline deep learning model. Further, the production process data of the electric melting magnesia furnace is collected at the edge side and transmitted to the cloud database, and according to the preset self-correction mechanism, the specified production process data is obtained from the cloud database to correct the online deep learning correction model and the cloud deep learning correction model, completing the correction of the non-linear dynamic system. Finally, the system output of the non-linear dynamic system and the model output of the linear model are obtained, and the sum value of the system output and the model output is used as the target demand data, and the target demand data is sent to the display terminal for display. The present invention combines mechanism with industrial big data, system identification with deep learning, and offline training with online correction, establishes a multi-furnace demand digital twin model, and proposes a demand digital twin method based on edge-cloud collaboration to correct the non-linear part in the multi-furnace demand digital twin model, realizing the modeling of the operation index of the industrial process whose dynamic characteristics change with the production process.

[0146] It should be noted that for other corresponding descriptions of each functional unit involved in the fused magnesia demand digital twin device based on edge-cloud collaboration provided by the embodiment of the present application, reference can be made to Figure 1 the corresponding description in and Figure 2, which will not be elaborated here.

[0147] In an exemplary embodiment, referring to Figure 4 , a device is further provided. The device includes a communication bus, a processor, a memory, and a communication interface, and may further include an input-output interface and a display device. Among them, each functional unit can complete mutual communication through the bus. The memory stores a computer program, and the processor is used to execute the program stored on the memory to execute the method for fused magnesia demand digital twin based on edge-cloud collaboration in the above embodiment.

[0148] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for fused magnesia demand digital twin based on edge-cloud collaboration are realized.

[0149] Through the description of the above embodiments, those skilled in the art can clearly understand that this application can be implemented by hardware or by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various implementation scenarios of this application.

[0150] Those skilled in the art can understand that the drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the drawings are not necessarily essential for implementing this application.

[0151] Those skilled in the art can understand that the modules in the devices in the implementation scenarios can be distributed in the devices in the implementation scenarios according to the description of the implementation scenarios, or can be correspondingly changed and located in one or more devices different from this implementation scenario. The modules in the above implementation scenarios can be combined into one module, or can be further split into multiple sub-modules.

[0152] The above serial numbers of this application are only for description and do not represent the advantages or disadvantages of the implementation scenarios.

[0153] The above disclosure only shows several specific implementation scenarios of this application. However, this application is not limited thereto, and any changes that can be conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A digital twin method for the required amount of fused magnesia based on edge-cloud-end collaboration, characterized in that, it includes: Establish a dynamic model of the current closed-loop control system for the smelting process of single-furnace fused magnesia. Based on the dynamic model of the current closed-loop control system, generate a dynamic model of the multi-furnace group furnace power of the fused magnesia furnace; Construct a multi-furnace group furnace required amount digital twin model according to the required amount definition and the multi-furnace group furnace power dynamic model. The multi-furnace group furnace required amount digital twin model includes a linear model and a non-linear dynamic system; Construct an offline deep learning model for the non-linear dynamic system, and construct an online deep learning correction model and a cloud deep learning correction model according to the offline deep learning model; Collect production process data of the fused magnesia furnace at the edge side, transmit the production process data to the cloud database, and according to the preset self-correction mechanism, obtain specified production process data from the cloud database to correct the online deep learning correction model and the cloud deep learning correction model, and complete the correction of the non-linear dynamic system; Obtain the system output of the non-linear dynamic system and the model output of the linear model, take the sum value of the system output and the model output as the target required amount data, and send the target required amount data to the display terminal for display.

2. The method according to claim 1, characterized in that, The establishment of the dynamic model of the current closed-loop control system for the smelting process of single-furnace fused magnesia, and based on the dynamic model of the current closed-loop control system, the generation of the dynamic model of the multi-furnace group furnace power of the fused magnesia furnace includes: Use the rotation direction and frequency of the single-furnace fused magnesia furnace motor as input data and the electrode current as output data to establish a current dynamic model; Discretize the current dynamic model based on the Euler method, and represent the current dynamic model as the sum of a second-order linear term and a high-order non-linear term; Use the melting current feedback control system to control the current dynamic model to obtain the dynamic model of the current closed-loop control system; Take the current output by the dynamic model of the current closed-loop control system as input, and establish a single-furnace single-furnace power dynamic model; Replace the single-furnace model parameters in the single-furnace single-furnace power dynamic model with group furnace model parameters to obtain a single-furnace group furnace power dynamic model; Use a constant that does not change with the furnace number to replace the group furnace model parameters that change with the furnace number to obtain the multi-furnace group furnace power dynamic model.

3. The method according to claim 1, characterized in that, The construction of the multi-furnace group furnace required amount digital twin model according to the required amount definition and the multi-furnace group furnace power dynamic model includes: Use the least squares identification algorithm to identify the identifiable model in the multi-furnace group furnace power dynamic model and estimate the identification parameters of the identifiable model; Calculate the parameter estimation value of the identification parameter and determine the identification error of the identifiable model generated by the parameter estimation value; Take the sum value of the identification error and the unknown non-linear term in the multi-furnace group furnace power dynamic model as the unknown non-linear system; Based on the demand definition, the unknown nonlinear system, and the multi-furnace group furnace power dynamic model, a multi-furnace group furnace demand digital twin model is calculated. The multi-furnace group furnace demand digital twin model is the sum of a nonlinear dynamic system and a linear model.

4. The method according to claim 1, wherein, the construction of the offline deep learning model of the nonlinear dynamic system, and the construction of the online deep learning correction model and the cloud deep learning correction model according to the offline deep learning model include: Obtain the actual current and demand data of the same furnace as sample data, and use the sample data to determine the number of single neuron nodes, the number of network layers, and the number of network neurons of the multi-layer long short-term memory (LSTM) network architecture to obtain the offline deep learning model; Construct the online deep learning correction model and the cloud deep learning correction model using the number of single neuron nodes, the number of network layers, and the number of network neurons.

5. The method according to claim 1, wherein, collecting the production process data of the electrofused magnesia furnace at the edge side, transmitting the production process data to the cloud database, and according to a preset self-correction mechanism, obtaining specified production process data from the cloud database to correct the online deep learning correction model and the cloud deep learning correction model, including: Collect the production process data of the electrofused magnesia group furnace in real time at the edge side and upload the production process data to the cloud database; Extract the production process data of all furnaces updated in real time in the cloud database to correct all layer weights and bias parameters of the cloud deep learning correction model; Fix the number of single neuron nodes, the number of network layers, and the number of network neurons corresponding to the online deep learning correction model, and use the production data of the same offline furnace to calculate the time series window length of the online deep learning correction model; Extract the production process data updated online within the same furnace in the cloud database to correct the weights and bias parameters of the fully connected layer of the online deep learning correction model; Calculate the error upper bound of the model output of the nonlinear dynamic system, and obtain the preset self-correction mechanism according to the error upper bound; According to the preset self-correction mechanism, use the cloud deep learning correction model to correct the online deep learning correction model.

6. The method according to claim 5, wherein, extracting the production process data of the time series window length updated online within the same furnace in the cloud database to correct the weights and bias parameters of the fully connected layer of the online deep learning correction model, including: When the data volume of the production process data within the same furnace meets the time series window length, according to the production process data of the updated online time series window length, keep the weights and bias parameters of the hidden layer of the online deep learning correction model unchanged, and correct the weights and bias parameters of the fully connected layer in real time. When the amount of production process data within the same heat does not meet the length of the time series window, the weights and bias parameters of the fully connected layer of the online deep learning correction model are not corrected, but the preset self-correction mechanism is still followed.

7. The method according to claim 5, wherein, the model correction of the online deep learning correction model by using the cloud deep learning correction model according to the preset self-correction mechanism includes: respectively detecting the model accuracy evaluation indexes of the online deep learning correction model and the cloud deep learning correction model, comparing the model accuracy evaluation indexes of the online deep learning correction model and the cloud deep learning correction model with the preset self-correction mechanism, and generating a comparison result; when the comparison result indicates a correction model, replacing the weights and bias parameters of all layers of the online deep learning correction model with the weights and bias parameters of all layers of the cloud deep learning correction model.

8. An electrofused magnesia demand digital twin device based on edge-cloud collaboration, wherein, it includes: a generation module, configured to establish a dynamic model of the current closed-loop control system for the single-furnace electrofused magnesia smelting process, and generate a multi-heat group furnace power dynamic model of the electrofused magnesia furnace based on the dynamic model of the current closed-loop control system; a first model construction module, configured to construct a multi-heat group furnace demand digital twin model according to the demand definition and the multi-heat group furnace power dynamic model, and the multi-heat group furnace demand digital twin model includes a linear model and a nonlinear dynamic system; a second construction module, configured to construct an offline deep learning model of the nonlinear dynamic system, and construct an online deep learning correction model and a cloud deep learning correction model according to the offline deep learning model; a correction module, configured to collect production process data of the electrofused magnesia furnace at the edge side, transmit the production process data to the cloud database, and correct the online deep learning correction model and the cloud deep learning correction model by obtaining specified production process data from the cloud database according to the preset self-correction mechanism, so as to complete the correction of the nonlinear dynamic system; an acquisition module, configured to acquire the system output of the nonlinear dynamic system and the model output of the linear model, use the sum value of the system output and the model output as the target demand data, and send the target demand data to a display terminal for display.

9. A computer device, including a memory and a processor, where the memory stores a computer program, wherein, when the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, on which a computer program is stored, wherein, when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images