Reduction furnace production process control method, device and equipment and storage medium

Abstract

The invention relates to the technical field of industrial control, in particular to a reduction furnace production process control method, device and equipment and a storage medium. According to the method, a self-adaptive controller is constructed through a reduction furnace system, and according to a preset updating strategy, each composition network of the self-adaptive controller is updated until each composition network meets a preset convergence condition, so that a target control strategy of the reduction furnace system is obtained through the updated self-adaptive controller; and controlling the cost function value of the reduction furnace system to reach a target value. Thus, the control input of the reduction furnace is adjusted through the target control strategy, stable control over the production parameters such as the surface temperature of the silicon rod and the silicon deposition rate is achieved, and the production efficiency and safety of polycrystalline silicon are improved.

Description

Technical Field

[0001] The present application relates to the field of industrial control, and in particular to a reduction furnace production process control method, device, equipment and storage medium. Background Art

[0002] Polysilicon is a key basic material for the electronic information industry and the new energy industry, and is widely used in the manufacture of photovoltaic cells. With the ever-expanding market demand, improving the production efficiency of polysilicon has become a key task in promoting the development of the photovoltaic industry.

[0003] At present, the industrial production of polysilicon is mainly based on the modified Siemens method, which deposits silicon on the surface of high-purity silicon rods through the generation, purification and chemical vapor deposition (CVD) of chlorosilane. The CVD reaction process is complex, involving multi-field coupling phenomena such as fluid flow, heat transfer, and mass transfer, and needs to be completed in a high-temperature, closed reduction furnace. However, during the reaction, the diameter of the silicon rod gradually increases, and the heat loss and deposition rate change dynamically, making it difficult to stably control production parameters such as the surface temperature and deposition rate of the silicon rod, resulting in reduced polysilicon production efficiency and damage to the reduction furnace production equipment, which may even cause serious risks to personal safety.

[0004] Therefore, there is an urgent need for a method to effectively control polysilicon production parameters and improve polysilicon production efficiency and safety. Summary of the invention

[0005] The present invention provides a reduction furnace production process control method, device, equipment and storage medium for effectively controlling polysilicon production parameters and improving polysilicon production efficiency and safety.

[0006] In a first aspect, the present application provides a reduction furnace production process control method, the method comprising:

[0007] Based on the reduction furnace system, an adaptive controller is constructed; the reduction furnace system represents the response relationship between the control strategy and the state parameters in the reduction furnace production process;

[0008] Based on a preset update strategy, updating each component network of the adaptive controller until each component network meets a preset convergence condition;

[0009] Based on the updated adaptive controller, a target control strategy of the reduction furnace system is obtained to control the cost function value of the reduction furnace system to reach a target value.

[0010] In a second aspect, the present application provides a reduction furnace production process control device, the device comprising:

[0011] A construction unit is used to construct an adaptive controller based on a reduction furnace system; the reduction furnace system represents a response relationship between a control strategy and a state parameter in a reduction furnace production process;

[0012] An updating unit, configured to update each component network of the adaptive controller based on a preset updating strategy until each component network satisfies a preset convergence condition;

[0013] The control unit is used to obtain a target control strategy for the reduction furnace system based on the updated adaptive controller to control the cost function value of the reduction furnace system to reach a target value.

[0014] Optionally, the updating unit is specifically used to:

[0015] Based on the component networks, a corresponding loss function is constructed; the loss function represents the deviation between the predicted output value of the component network and the target output value;

[0016] Based on each loss function, calculate the weight update value of the corresponding component network;

[0017] Based on the weight update value, the weights of the component networks are updated until the loss function values ​​meet the preset convergence conditions.

[0018] Optionally, each component network includes an execution network, an evaluation network and a model network, wherein:

[0019] The execution network is used to obtain a corresponding control strategy based on the state parameters of the reduction furnace system;

[0020] The model network is used to predict the state parameters of the reduction furnace system at the next moment based on the control strategy of the reduction furnace system at the current moment;

[0021] The evaluation network is used to obtain corresponding cost function values ​​based on the control strategy and state parameters of the reduction furnace system.

[0022] Optionally, the construction unit is further used for:

[0023] Based on the polysilicon production process of the reduction furnace, a current heating model and a chemical vapor deposition model are constructed; the current heating model characterizes the current-thermal coupling relationship of the silicon rods in the reduction furnace, and the chemical vapor deposition model characterizes the reaction process of generating silicon deposition in the reduction furnace;

[0024] Based on a preset combination strategy, the current heating model and the chemical vapor deposition model are weighted combined to obtain the reduction furnace system.

[0025] Optionally, the construction unit is specifically used for:

[0026] Based on a preset sampling strategy, sample points of the current heating model and the chemical vapor deposition model are obtained; the sample points represent input variables and output variables of the corresponding models;

[0027] Based on a preset proxy modeling strategy, modeling is performed on the sample points to obtain a plurality of proxy sub-models;

[0028] Based on the multiple proxy sub-models, weighted combination processing is performed to obtain the reduction furnace system.

[0029] In a third aspect, the present application provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any one of the reduction furnace production process control methods described in the first aspect is implemented.

[0030] In a fourth aspect, the present application provides a computer storage medium, wherein the computer-readable storage medium stores computer program instructions, and the computer program instructions are executed by a processor to implement any one of the reduction furnace production process control methods described in the first aspect.

[0031] In a fifth aspect, an embodiment of the present application provides a computer program product, comprising computer program instructions, which, when executed by a processor, implement any one of the reduction furnace production process control methods in the first aspect.

[0032] The beneficial effects of the present invention are as follows:

[0033] The embodiment of the present application provides a reduction furnace production process control method, which constructs an adaptive controller through a reduction furnace system, and updates the component networks of the adaptive controller according to a preset update strategy until each component network meets the preset convergence condition, thereby obtaining the target control strategy of the reduction furnace system through the updated adaptive controller to control the cost function value of the reduction furnace system to reach the target value. In this way, the control input of the reduction furnace is adjusted through the target control strategy output by the controller, so as to achieve stable and accurate control of production parameters such as silicon rod surface temperature and silicon deposition rate, thereby improving the efficiency and safety of polysilicon production. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0035] Figure 1A flow chart of a reduction furnace production process control method provided in an embodiment of the present application;

[0036] Figure 2 A schematic diagram of the structure of an adaptive controller provided in an embodiment of the present application;

[0037] Figure 3 A schematic diagram of an update process of an adaptive controller provided in an embodiment of the present application;

[0038] Figure 4 A schematic diagram of a trajectory of a reduction furnace control strategy changing over time provided in an embodiment of the present application;

[0039] Figure 5 A controller output and setting value change process provided in an embodiment of the present application;

[0040] Figure 6 A schematic diagram of a system framework provided for an embodiment of the present application;

[0041] Figure 7 A schematic diagram of a microservice list provided for an embodiment of the present application;

[0042] Figure 8 A schematic diagram of a system generating a target control strategy provided in an embodiment of the present application;

[0043] Fig. 9 A schematic diagram of the structure of a reduction furnace production process control device provided in an embodiment of the present application;

[0044] Fig.10 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme in the embodiment of the present application will be clearly and completely described below in conjunction with the drawings in the embodiment of the present application. Obviously, the described embodiment is only a part of the embodiment of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application. In the absence of conflict, the embodiments in the present application and the features in the embodiments can be arbitrarily combined with each other. In addition, although the logical order is shown in the flow chart, in some cases, the steps shown or described can be performed in an order different from that here.

[0046] The terms "first" and "second" in the specification and claims of the present application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the term "comprising" and any of their variations are intended to cover non-exclusive protection. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices. "Multiple" in the present application can mean at least two, for example, can be two, three or more, and the embodiments of the present application are not limited.

[0047] The term "and / or" in the embodiments of the present application is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0048] It is understandable that in the following specific implementations of the present application, related data such as reduction furnace production data are involved. When the various embodiments of the present application are applied to specific products or technologies, relevant licenses or consents need to be obtained, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions. For example, relevant volunteers can be recruited and relevant agreements on volunteer authorization data can be signed, and then the data of these volunteers can be used for implementation; or, by implementing within the scope of the authorized organization, the following implementation methods are implemented by using the data of internal members of the organization to manage data; or, the relevant data used in the specific implementation are all simulated data, such as simulated data generated in a virtual scene.

[0049] The following is a brief introduction to the design concept of the embodiments of the present application.

[0050] Polysilicon is a key basic material for the electronic information industry and the new energy industry, and is widely used in the manufacture of photovoltaic cells. With the ever-expanding market demand, improving the production efficiency of polysilicon has become a key task in promoting the development of the photovoltaic industry.

[0051] At present, the industrial production of polysilicon is mainly based on the modified Siemens method, which deposits silicon on the surface of high-purity silicon rods through the generation, purification and chemical vapor deposition of chlorosilane. The CVD reaction process is complex, involving multi-field coupling phenomena such as fluid flow, heat transfer, and mass transfer, and needs to be completed in a high-temperature, closed reduction furnace. The reduction furnace reactor is a bell-shaped chamber with a group of slender silicon rods inside, which are heated by a strong electric current. During the reaction, the surface of the silicon rod undergoes a chemical vapor deposition reaction through hydrogen and trichlorosilane, gradually depositing silicon on the surface of the silicon rod, resulting in an increase in the diameter of the silicon rod. In addition, as the diameter of the silicon rod increases, the deposition rate and heat loss of silicon also increase. In order to keep the surface temperature of the silicon rod constant, the current needs to be increased. However, the increase in current is limited by mechanical stress and the melting point temperature of the silicon core. In addition, due to the distribution characteristics of the internal heating source caused by the Joule effect and the influence of heat transfer with the environment. The radial temperature gradient inside the silicon rod is difficult to control. If the temperature of the silicon core of the silicon rod is too high, it may cause the rod to collapse due to internal thermal stress, which is a very serious safety accident. It will not only cause irreversible damage to the reduction furnace production equipment, but also pose a personal safety risk to the operators who open the furnace.

[0052] In view of the above problems, the embodiment of the present application provides a reduction furnace production process control method, which constructs an adaptive controller through the reduction furnace system, and updates the component networks of the adaptive controller according to the preset update strategy until each component network meets the preset convergence condition, thereby obtaining the target control strategy of the reduction furnace system through the updated adaptive controller to control the cost function value of the reduction furnace system to reach the target value. In this way, in view of the significant nonlinear problems existing in the reduction furnace production process, the controller outputs the target control strategy and adjusts the control input of the reduction furnace, so as to achieve stable and accurate control of production parameters such as silicon rod surface temperature and silicon deposition rate, thereby improving the efficiency and safety of polysilicon production.

[0053] The following briefly introduces the application scenarios to which the technical solutions of the embodiments of the present application can be applied. It should be noted that the application scenarios introduced below are only used to illustrate the embodiments of the present application and are not limited. In the specific implementation process, the technical solutions provided by the embodiments of the present application can be flexibly applied according to actual needs.

[0054] The solution provided in the embodiment of the present application can be applied to most application scenarios of process optimization and quality control in the production process of polysilicon, and is used to effectively control the production parameters of the reduction furnace to greatly improve the production efficiency and safety of polysilicon. For example, in industrial production scenarios, by accurately controlling the temperature of the silicon rod and the silicon deposition rate, the silicon rod current heating process and the silicon deposition process are optimized to avoid safety accidents and equipment damage caused by abnormal working conditions, while ensuring the quality of polysilicon products, reducing energy consumption and improving production efficiency, providing strong technical support for the green production and sustainable development of polysilicon production.

[0055] Of course, the method provided in the embodiment of the present application is not limited to the above application scenarios, but can also be used in other possible application scenarios, which are not limited by the embodiment of the present application. The functions that can be realized by each device in the above application scenarios will be described in the subsequent method embodiments, and no further details will be given here.

[0056] Below, in combination with the application scenarios described above, the method provided by the exemplary embodiment of the present application is described with reference to the accompanying drawings. It should be noted that the above application scenarios are only shown to facilitate understanding of the spirit and principles of the present application, and the implementation methods of the present application are not limited in this regard.

[0057] See also Figure 1 As shown, it is a flow chart of a reduction furnace production process control method provided in an embodiment of the present application, and the specific implementation process of the method is as follows:

[0058] Step 101: construct an adaptive controller based on the reduction furnace system.

[0059] In the embodiment of the present application, the reduction furnace simulation system characterizes the response relationship between the production parameters (such as silicon rod temperature, silicon deposition rate, etc.) and the control strategy (such as current, reactant flow rate) in the polysilicon production process, and the adaptive controller dynamically adjusts the control strategy to ensure that the silicon rod surface temperature and silicon deposition rate remain stable during the production process by iteratively optimizing the cost function value of the system, and optimizes the overall energy efficiency of the system.

[0060] Specifically, the dynamic system expression of the reduction furnace system is as follows:

[0061]

[0062] Among them, x(k) represents the state parameter, that is, the current internal physical parameter of the reduction furnace system, which is used to describe the dynamic characteristics of the reduction furnace, such as silicon rod surface temperature, reaction gas concentration, silicon deposition rate, etc.

[0063] u(k) represents the control parameter, which is an adjustable parameter outside the system and can affect the system state, such as input current and reactant flow rate. The surface temperature of the silicon rod can be adjusted by controlling the input current, and the reactant concentration can be controlled by adjusting the gas flow rate, thus affecting the chemical vapor deposition process.

[0064] f(·) and g(·) are unknown and bounded nonlinear smooth functions, C is the system output factor, and y(k) represents the system output.

[0065] In summary, in order to solve the target control strategy u of the reduction furnace system *(x(k)), the embodiment of the present application constructs the dynamic system of the polysilicon reduction furnace as an optimal control problem in an infinite time domain, and expects to determine u(k)∈Ω, Ω is an allowable control set, so that the cost function value of the system reaches the minimum value.

[0066] The expression of the cost function of the reduction furnace system is as follows:

[0067]

[0068] Among them, J(x(k),u(k)) is the cost function value of the reduction furnace system when the state parameter is x(k) and the control parameter is u(k), that is, the sum of the effect function value at this moment and the cost function value at the future moment;

[0069] 0<γ≤1 represents the single-step cost discount coefficient, which means the weight of future income, which is required to be greater than 0 and less than or equal to 1;

[0070] U(x(k),u(k)) is the single-step cost function generated at time step k, and its expression is as follows:

[0071] U(x(k),u(k))=x T (k)Qx(k)+u T (x(k))Ru(x(k))

[0072] Among them, Q and R are R n×n and R m×m The positive definite matrix, x T (k)Qx(k) represents the state utility, u T (x(k))Ru(x(k)) represents the control utility.

[0073] According to the Bellman optimality principle, the optimal cost function can be shown as follows:

[0074]

[0075] Among them, J * (x(k)) represents the optimal cost function, which indicates the cumulative minimum cost value starting from time step k.

[0076] x(i) represents the state parameters of the system at time step i, namely, the state variables such as silicon rod temperature and deposition rate during the reduction furnace production process.

[0077] u(i) represents the control parameter input to the system at time step i, such as current input or reactant flow rate.

[0078] U(x(i),u(i)) represents the single-step cost function, which indicates the instantaneous cost generated when the system state parameter is x(i) and the control parameter is u(i).

[0079] γ represents the discount factor, which indicates the weight of future costs.

[0080] N represents the total simulation time step, indicating the end time of the control process.

[0081] In this way, the optimal cost function defines an optimal cost evaluation criterion by accumulating the current and future single-step costs, with the goal of minimizing the cumulative cost.

[0082] Furthermore, the optimal cost function needs to satisfy the following discrete-time Hamilton-Jacobi-Bellman (HJB) equation:

[0083] J * (x(k))=min u(x(k)) {U(x(k),u(x(k)))+J * (x(k+1))}

[0084] J * (x(k)) represents the optimal cost function at time step k.

[0085] U(x(k),u(k)) represents the single-step instantaneous cost, which is the cost value under the current state x(k) and control input u(k).

[0086] J * (x(k+1)) represents the future cost, that is, the optimal cost function for the next time step k+1.

[0087] The HJB equation is the discrete-time form of the Bellman equation, indicating that the optimal cost function is the minimized cumulative value of the current single-step cost and the future cost.

[0088] However, due to J * The value of (x(k+1)) is unknown in advance, and the controlled object is a non-affine system. Therefore, it is difficult to solve the above HJB equation using traditional methods, and the optimal control strategy u * The expression of (x(k)) is as follows:

[0089]

[0090] Among them, u * (x(k)) represents the optimal control strategy, which is the control input that minimizes the cost given the current state parameter x(k).

[0091] U(x(k),u(x(k)) represents the current single-step cost.

[0092] J *(x(k+1)) represents the optimal cost function value for the next time step.

[0093] This formula is used to calculate the optimal control input u * (x(k)), which minimizes the sum of the current single-step cost and the future cumulative cost. However, this expression is also difficult to solve using traditional methods, and the extended state x e (k) cannot be used directly for feedback.

[0094] In summary, it is difficult for traditional methods to solve the HJB equation corresponding to the cost function, and thus it is impossible to obtain the optimal control parameters corresponding to each state parameter, resulting in the inability to adjust the reduction furnace system state at each time step by inputting the optimal control parameters during the production of polysilicon in the reduction furnace, and it is impossible to achieve precise control of the silicon rod surface temperature and silicon deposition rate. Therefore, the embodiment of the present application constructs an adaptive controller, combines the optimality principle to decompose the optimal control problem into the cumulative form of the immediate cost and future cost under the time step, and converts the HJB equation in the nonlinear discrete time domain into an online iterative optimization problem. Through the approximate learning of the neural network, the execution network, evaluation network and model network of the controller are iteratively updated, and the optimal cost function and optimal control parameters are gradually approached, so that the cost function value of the system is minimized, thereby overcoming the difficulty that the traditional method cannot resolve the HJB equation and realizing the parameter optimization control of the polysilicon reduction furnace production process.

[0095] In a possible implementation manner, the network structure of the adaptive controller provided in the embodiment of the present application is as follows:

[0096]

[0097] Wherein, σ is a neuron activation function, which is used to introduce nonlinear mapping capability. In the embodiment of the present application, the activation function enables the network to have the ability to fit the dynamics of nonlinear systems and to handle the complex nonlinear relationships inside the reduction furnace.

[0098] x(k)=[x1(k),x2(k),…,x L (k)] T ,u(k)=[u1(k),u2(k),…,u K (k)] T is an input vector consisting of the state parameters and control parameters of the k time step, which is used to predict the system state in the next time step.

[0099] It represents the system state of the next time step predicted by the controller. For example, it predicts the state parameters and control parameters of the reduction furnace system at the current time step, and outputs the state parameters of the reduction furnace system such as the silicon rod surface temperature and chemical reaction rate at the next moment, which are used to update the optimal control strategy.

[0100] is the input layer weight, which is used to transform the system state x(k) and control input u of the input layer (i) T(x(k)) is mapped to the hidden layer.

[0101] The output layer weight is used to weight the sum of the results of the output layer neurons of the neural network to obtain the predicted system state. In the embodiment of the present application, the output layer weight is an adjustable parameter of the model network, and the parameter can be updated through training to ensure that the output predicted by the controller is close to the actual system state.

[0102] b1 and b2 represent the hidden layer bias term and the output layer bias term, respectively. b1 is used to bias the linear combination result of the hidden layer to enhance the fitting ability of the network. b2 is used to bias the final network output to adjust the accuracy of the network output to ensure that the predicted state of the next time step is closer to the actual value. In the embodiment of the present application, the bias term can be used together with the weight matrix to adjust the controller's prediction result of the system state.

[0103] In this way, the adaptive controller can adapt to the complex dynamic characteristics of the reduction furnace through its nonlinear mapping capability. It can predict the system state of the reduction furnace system at the next time step by inputting the state parameters and control parameters of the reduction furnace system at the current time step, and its model parameters are updated through network training to improve the prediction accuracy.

[0104] In a possible implementation, the embodiment of the present application may first construct a current heating model representing the current-thermal coupling relationship of silicon rods in the reduction furnace and a chemical vapor deposition model representing the reaction process of generating silicon deposition in the reduction furnace according to the polysilicon production process of the reduction furnace. Then, according to a preset combination strategy, a weighted combination process is performed on the current heating model and the chemical vapor deposition model to obtain a reduction furnace simulation system representing the multi-field coupling relationship between the silicon rod current heating process and the chemical vapor deposition reaction in the reduction furnace.

[0105] Specifically, in the embodiment of the present application, a reduction furnace simulator is constructed, including constructing a current heating model and a chemical vapor deposition model, and combining the current heating model and the chemical vapor deposition model to construct the reduction furnace simulator. Two major processes, so as to efficiently simulate the complex physical process in the reduction furnace, provide a basic interactive environment for subsequent optimization control. Taking the modified Siemens method for preparing polysilicon as an example, its preparation process mainly relies on a high-temperature closed reduction furnace reactor to achieve, first using an electric current to heat the silicon rod to about 1373K, and then the pre-heated trichlorosilane and hydrogen mixture are introduced into the reactor, and a vapor deposition reaction is carried out on the surface of the silicon rod. As the deposition process proceeds, the diameter of the silicon rod gradually increases. Since this polysilicon production process involves processes such as fluid flow, convective heat transfer, radiation heat transfer, mass transfer, chemical reaction, and phase change, the embodiment of the present application analyzes the current silicon rod heating process and the chemical vapor deposition process of the reduction furnace respectively through thermodynamics, chemical reaction kinetics, fluid mechanics, heat transfer and other aspects of theory, so as to construct a current heating model and a chemical vapor deposition model.

[0106] Specifically, in view of the skin effect and radial temperature gradient phenomena that occur in the silicon rod during the current heating process, the embodiment of the present application studies and analyzes the thermal effects of heat transfer phenomena such as convective heat transfer, radiation heat transfer, and chemical reaction on the surface of the silicon rod, and describes and calculates the thermal behavior of the silicon rod under current heating based on electromagnetic theory and solid heat transfer theory, and performs thermoelectric coupling calculation, and finally establishes a current heating model, and the model building formula is as follows:

[0107] Under normal reduction process conditions, the surface temperature of the polysilicon rod is generally 1373K, while the reactor wall is cooled by cooling water, and the temperature is generally controlled at about 373K. In this way, strong radiation heat exchange will inevitably occur between the high-temperature silicon rod surface and the cold reactor wall. The rod surface radiation heat transfer calculation formula in the embodiment of the present application can be shown as follows:

[0108] q convection =h·(T S -T g )

[0109] Among them, q convection represents the convective heat transfer, the heat exchange between the silicon rod and the surrounding gas through convective heat transfer. h represents the convective heat transfer coefficient, the heat transfer per unit area under unit temperature difference, which is affected by air flow speed and temperature. Ts and Tg are the silicon rod surface temperature and gas temperature respectively.

[0110] Under normal reduction process conditions, the surface temperature of the polysilicon rod is generally 1373K, while the reactor wall is cooled by cooling water, and the temperature is generally controlled at about 373K. In this way, strong radiation heat exchange will inevitably occur between the high-temperature silicon rod surface and the cold reactor wall. The rod surface radiation heat transfer calculation formula in the embodiment of the present application can be shown as follows:

[0111]

[0112] q radiation It represents the radiation heat transfer amount, that is, the amount of heat exchange between the silicon rod and the surrounding environment through thermal radiation.

[0113] ε r Represents the emissivity of the silicon rod surface, and the value range is between 0-1.

[0114] σ B Stands for the Stefan-Boltzmann constant, which is the relationship between thermal radiation power per unit area and absolute temperature.

[0115] Ts, Tw represent the silicon rod surface temperature and ambient temperature.

[0116] The energy required by the silicon rods in the reactor is all provided by electric current heating. When the silicon rods are heated to the reaction temperature by direct current, the electric field inside the silicon rods satisfies the following relationship under steady-state conditions:

[0117]

[0118] The electric field distribution formula represents the electric potential Poisson equation satisfied by the electric field inside the silicon rod, describes the distribution law of the electric field in space, and is the basis for the subsequent calculation of current density and electric heating. Among them, E is the electric field intensity, which represents the spatial distribution of the electric field inside the silicon rod.

[0119] Under steady-state conditions, the heat transfer equation inside the silicon rod in the Siemens reactor can be expressed as follows:

[0120]

[0121] The heat transfer equation represents the relationship between the heat source and heat conduction. It is used to calculate the temperature distribution inside the silicon rod and determine the equilibrium state of the Joule heat source and heat loss. Among them, T represents the temperature distribution, which describes the temperature field inside the silicon rod. k represents the thermal conductivity, which is related to the thermal conductivity of the material and changes with temperature. q is the internal heat source density, which is mainly composed of Joule heat and chemical reaction heat, representing the heat inside the silicon rod through heat conduction. and heat source q reaches dynamic equilibrium.

[0122] Considering the axial symmetry of the silicon rod geometric model, the above electric field distribution formula and heat transfer equation can be applied to cylindrical coordinates to obtain the electric potential radial distribution equation and the temperature distribution equation inside the silicon rod, which are shown as follows:

[0123]

[0124] The electric potential radial distribution equation represents the radial distribution law of the electric field strength inside the silicon rod, and is used to determine the relationship between the silicon rod current and the heat source distribution. Among them, r represents the radial distance, that is, the radial distance from the center of the silicon rod to the outside, and E represents the electric field strength, that is, the change of the electric field strength inside the silicon rod along the radial direction, which is used to determine the electric field distribution law.

[0125]

[0126] The main equation for the temperature distribution inside the silicon rod is the main equation for describing heat transfer (conduction, radiation, convection), where T represents the temperature distribution, r represents the radial position, and k(T) represents the thermal conductivity, which changes with temperature. k′(T) represents the derivative of thermal conductivity with respect to temperature, describing its rate of change. σ(T) represents electrical conductivity, which indicates the material's ability to conduct electric current. dV / dz represents the axial component of the electric field intensity, which is related to the heating power. Thus, the balance of heat transfer and heat generation inside the silicon rod is described by combining Joule heating (represented by σ(T) and dV / dz) and thermal conductivity characteristics.

[0127] For the electric potential radial distribution equation, its boundary conditions include the electric field boundary conditions and the total current calculation formula, which are as follows:

[0128]

[0129] This formula represents the electric field boundary condition, where E is the electric field intensity, and r=0 represents the center position of the silicon rod, that is, the electric field boundary condition represents that the electric field at the center of the silicon rod has no radial component, which conforms to physical symmetry.

[0130]

[0131] This formula is the total current calculation formula. The total current passing through the entire silicon rod is calculated by integrating the current density (σE), providing a basis for the evaluation of the electric heating power. Among them, I represents the total current, σ represents the conductivity, E represents the electric field intensity, and Ω represents the integration domain, that is, the cross-sectional area of ​​the silicon rod.

[0132] For the temperature distribution equation inside the silicon rod, the boundary conditions are as follows:

[0133]

[0134] This formula means that the temperature gradient at the center of the silicon rod is zero and the temperature distribution is symmetrical.

[0135]

[0136] This formula is the boundary heat balance formula, which means that the temperature gradient on the silicon rod surface is determined by the superposition of heat flux from convection, radiation and chemical reaction. It is used to accurately calculate the surface temperature of the silicon rod and combine heat transfer and reaction heat for overall thermodynamic balance. convection ,q radiation and q reaction They are the average convection, radiation and reaction heat losses per unit area on the silicon rod surface, R represents the outer diameter of the silicon rod, and k represents the thermal conductivity.

[0137] Specifically, for chemical vapor deposition reactions, the embodiment of the present application constructs a polysilicon chemical vapor deposition transfer-kinetic model under multi-field coupling based on gas phase reactions and surface phase reactions. By analyzing and comparing the effects of different mechanisms on the deposition rate, the chemical reaction mechanism suitable for Siemens polysilicon reduction furnace is determined, thereby establishing a chemical vapor deposition model. The reaction mechanism is shown in Tables 1 and 2 below:

[0138] Table 1 Gas phase reaction mechanism of SiHCl3-H2 system

[0139] Serial number Chemical reaction expressions A Ea(cal / mol) α G1 <![CDATA[SiCl3H=HCl+SiCl2]]> <![CDATA[3.162×10 14 ]]> 72,900 0 G2 <![CDATA[SiCl2H2=SiCl2+H2]]> <![CDATA[3.162×10 13 ]]> 69,300 0 G3 <![CDATA[SiCl2H2=HSiCl+HCl]]> <![CDATA[6.918×10 14 ]]> 75,800 0 G4 <![CDATA[H2ClSiSiCl3=SiCl4+SiH2]]> <![CDATA[1.585×10 13 ]]> 55,500 0 G5 <![CDATA[H2ClSiSiCl3=SiCl3H+HSiCl]]> <![CDATA[3.162×10 13 ]]> 49,800 0 G6 <![CDATA[H2ClSiSiCl3=SiCl2H2+SiCl2]]> <![CDATA[6.310×10 13 ]]> 44,300 0 G7 <![CDATA[Si2Cl5H=SiCl4+HSiCl]]> <![CDATA[5.012×10 13 ]]> 52,300 0 G8 <![CDATA[Si2Cl5H=SiCl3H+SiCl2]]> <![CDATA[7.943×10 13 ]]> 45,900 0 G9 <![CDATA[Si2Cl6=SiCl4+SiCl2]]> <![CDATA[1.585×10 14 ]]> 48,800 0

[0140] Among them, the above table 1 represents multiple groups of chemical reactions mainly occurring in the gas phase of reactants (SiHCl3-H2), which are used to describe the chemical changes between gas components, to determine the reaction path of reactants in the gas phase, and to provide a basis for the types and concentrations of chemical substances transferred to the surface. Among them, A represents the pre-reaction factor, which represents the magnitude of the reaction rate constant, which is affected by the collision frequency and spatial arrangement of the reactants. In an embodiment of the present application, A can be used to calculate the rate constant of gas phase or surface phase reaction, and provide parameter support for the kinetic model. Ea represents activation energy, in units of cal / mol, i.e., the minimum energy required for the reaction, which determines whether the reaction can occur at a specific temperature. In an embodiment of the present application, it is used to control the rate of deposition reaction by temperature change and optimize the operating temperature of the reduction furnace. α represents the reaction order, which represents the degree of dependence of the reaction rate on the concentration of the reactants. In an embodiment of the present application, it can be used to determine the relationship between the reaction rate and the concentration of the reactants, and to optimize the gas ratio.

[0141] Table 2 Surface reaction mechanism of SiHCl3-H2 system

[0142]

[0143] Among them, Table 2 above represents 10 surface phase reactions of the reaction mechanism system, that is, the reactions that occur after the gas phase substances reach the surface of the silicon rod, which are used to describe the deposition and surface chemical reaction processes, thereby determining the deposition rate of silicon and the generation of surface products, the synergistic effect between the reaction surface species and the gas phase species, and gives a silicon surface adsorption phase reaction model, which mainly includes three surface substances SiH(S), SiCl(S), and Si(S), among which the surface density of Si(S) is 1.162E-9mol / cm-2.

[0144] In a possible implementation, before constructing the current heating model and the chemical vapor deposition model, the embodiment of the present application may use the optimal Latin hypercube experimental design method to obtain sample points, for example, by performing random sampling in the parameter variable space through the optimal Latin hypercube experimental design method, to obtain parameters such as silicon rod current, reaction gas concentration, and temperature. Thus, the sample points are evenly distributed within the variable range to ensure the spatial filling of the sample point distribution, thereby improving the accuracy of the reduction furnace simulation system. Then, through the acquired sample points, the radial basis function models corresponding to the silicon rod current heating process and the chemical vapor deposition process are respectively constructed, thereby obtaining the current heating model and the chemical vapor deposition model. The radial basis function model can exhibit local or global convergence characteristics according to the selected radial function. The embodiment of the present application can select commonly used kernel functions such as Gaussian kernel function, polyhedral kernel function, inverse polyhedral kernel function, and thin plate spline kernel function. Different kernel functions have different approximation characteristics in different scenarios.

[0145] In a possible implementation, considering that a single current heating model and chemical vapor deposition model will have a long interaction time due to the complexity of calculation, thus reducing the efficiency of obtaining the control strategy. Therefore, after constructing the current heating model and the chemical vapor deposition model, the embodiment of the present application obtains sample points representing the input variables and output variables of the current heating model and the chemical vapor deposition model through a preset sampling strategy, and obtains the corresponding proxy sub-model by modeling the sample points, thereby performing weighted combination processing on multiple proxy sub-models to obtain a reduction furnace system, thereby performing high-fidelity approximation on the mechanism model to improve the calculation speed.

[0146] Specifically, the embodiment of the present application can obtain sample points through the optimal Latin hypercube experimental design method, so that the sample points are evenly distributed within the variable range, ensuring the spatial filling of the sample point distribution, thereby improving the accuracy of the model. Then, the RBF proxy model is constructed through the obtained sample points, that is, the corresponding current heating model and chemical vapor deposition model are generated. The RBF model can exhibit local or global convergence characteristics according to the selected radial function. For example, commonly used kernel functions including Gaussian kernel function, polyhedral kernel function, inverse polyhedral kernel function and thin plate spline kernel function can be selected, and different kernel functions have different approximation characteristics in different scenarios. After obtaining the current heating model and the chemical vapor deposition model, the weighted (PRESS Weighted Surrogate, PWS) algorithm based on PRESS can be used to combine the weight distribution of the current heating model and the chemical vapor deposition model to construct a combined proxy model. The PWS algorithm can determine the weight of each model in the combined proxy model by calculating the error value (PRESS value) of each single proxy model, thereby improving the robustness and accuracy of the prediction. In this way, through the PWS algorithm, the embodiment of the present application can integrate the current heating model and the chemical vapor deposition model to generate a combined proxy model, wherein the current heating model and the chemical vapor deposition model are assigned different weights according to their respective PRESS values, and finally form a combined proxy model with better performance, thereby improving the fitting ability of a single physical process, ensuring that the prediction results of each model are accurate and reliable, and providing high-quality input data for the subsequent construction of the reduction furnace simulation system. Then, the embodiment of the present application can use the output of the current heating model as the input of the chemical vapor deposition model to construct a joint proxy model and complete the construction of the reduction furnace simulation system. The reduction furnace simulation system constructed in this way can efficiently simulate the complex physical processes in the reduction furnace and provide a basic interactive environment for subsequent optimization control.

[0147] Step 202: Based on a preset update strategy, update each component network of the adaptive controller until the adaptive controller meets a preset convergence condition.

[0148] In the embodiment of the present application, a preset update strategy can be used to gradually optimize the component networks of the adaptive controller, thereby improving the accuracy of its control strategy generation, system state prediction and cost function estimation, so that the sum of the system's immediate cost and future cost gradually approaches the minimum value to adapt to the control requirements of nonlinear complex systems.

[0149] In a possible implementation, the constituent network of the adaptive controller in the embodiment of the present application may include an execution network, an evaluation network, and a model network. The execution network is used to obtain the corresponding control strategy through the state parameters of the reduction furnace system; the model network is used to predict the state parameters of the reduction furnace system at the next moment through the control strategy of the reduction furnace system at the current moment; the evaluation network is used to obtain the corresponding cost function value through the control strategy and state parameters of the reduction furnace system.

[0150] Specifically, refer to Figure 2 FIG. 1 is a schematic diagram of a structure of an adaptive controller provided in an embodiment of the present application. Figure 2 It reflects the synergy of the execution network, the model network and the two evaluation networks A and B. The execution network receives the current state parameter x(k) as input and outputs an approximate control strategy u(k) to approximate the optimal control strategy, which is used to control the input variables (such as current or power) of the reduction furnace, dynamically adjust the input parameters during the silicon rod heating process, and ensure that the surface temperature is maintained in the optimal range. The model network is used to receive the current state x(k) and the control input u(k), and output the predicted next state x(k+1), which is used to simulate the complex dynamic behavior of the reduction furnace and provide real-time state prediction. When the real system is difficult to model, neural network learning replaces traditional physical modeling. Evaluation network A is used to estimate the cost function value J(x(k)) of the current state as a reference for executing network optimization. By analyzing x(k) and u(k), it evaluates the short-term utility of the current strategy and helps the adaptive controller determine whether the control strategy needs to be adjusted. Evaluation network B receives the predicted state parameter x(k+1) of the model network and estimates the cost function J(x(k+1) of the next time step. Combined with evaluation network A, it is used to calculate the time difference error and provide an evaluation of future costs to ensure that the control strategy takes into account both short-term and long-term optimization.

[0151] In one possible implementation, the embodiment of the present application can obtain the deviation between the predicted output value and the target output value of each component network, that is, the loss value, according to the loss function corresponding to each component network, and obtain the weight update value of the corresponding component network through each loss value, so as to update the weight of each component network through the weight update value until each loss function value meets the preset convergence condition.

[0152] In a possible implementation, the embodiment of the present application can predict and process the state parameters of the reduction furnace system through the execution network to obtain the corresponding control strategy valuation, and thus obtain the loss value of the execution network through the control strategy valuation and the loss function.

[0153] Specifically, for the execution network, the purpose of adjusting the execution network weight is to generate a control strategy to minimize the control cost function. By optimizing the loss function, the execution network can adjust the weight and gradually approach the optimal control strategy. In the embodiment of the present application, the adjustable weight of the execution network is represented as W a (k), the corresponding loss function is as follows:

[0154]

[0155] When the loss function of the execution network is 0, the weight of the execution network reaches the optimal value, and its weight increment can be expressed as:

[0156]

[0157] Where, ΔW α (k) represents the increment of the execution network weight. The weight is adjusted by the gradient descent method to gradually reduce the loss function.

[0158] α α (k) represents the learning rate, which controls the step size of weight update.

[0159] According to the loss function of the execution network and the chain rule, the weight increment of the execution network is as follows:

[0160]

[0161] The error term quantifies the effect of the control input u(k) on the immediate cost function U(k) and the future cost function J(k+1)

[0162] It is the core error of the execution network loss function, which is used to guide the weight update of the execution network.

[0163] γ discount factor, used to balance the weights of immediate and long-term costs

[0164] The intermediate multiplier is written as The internal neuron state of the execution network is denoted as s a (k), then the weight increment of the execution network can be expressed as:

[0165]

[0166] s α (k) represents the internal state of the neurons in the execution network and is used to describe the current dynamic characteristics of the network.

[0167] In this way, the control strategy is comprehensively optimized through learning rate, error gradient and neuron state.

[0168] In the weight increment expression of the execution network, U(k) is a single-step cost function defined according to the actual problem, so it is easy to get And the second term in the weight increment expression can be calculated by the following formula:

[0169]

[0170] In one possible implementation, the embodiment of the present application can predict the control strategy valuation through the model network, obtain the state parameters of the reduction furnace system at the next moment, and obtain the loss value of the model network through the state parameters of the reduction furnace system at the next moment and the loss function.

[0171] Specifically, for the model network, the embodiment of the present application represents the output weight of the model network as W m (k), the loss function of the model network is as follows:

[0172]

[0173] Among them, e m (k) = y pre (k)-y m (k), y pre (k) is the predicted output of the model network, y m (k) is the measured value of the system output.

[0174] Furthermore, the weight update formula of the model network is as follows:

[0175] ΔW m (k) = -α m (k)e m (k)s m (k)

[0176] Where, ΔW m (k) represents the increment of the model network weight. The weight is adjusted by the gradient descent method to gradually reduce the loss function.

[0177] α m (k) represents the learning rate, which controls the step size of weight update.

[0178] s m (k) represents the internal state of the model network neurons and is used to describe the current dynamic characteristics of the network.

[0179] Comprehensively optimize model predictions through learning rate, error gradient, and neuron state.

[0180] In a possible implementation, the control strategy estimate and state parameters can be processed for cost prediction through the evaluation network to obtain the corresponding cost function prediction value, thereby obtaining the loss value of the evaluation network based on the cost function prediction value and the loss function.

[0181] Specifically, for the evaluation network, the output weight of the evaluation network is defined as W C , construct its loss function as follows:

[0182]

[0183] e c (k) = J(k) - U(k) - γJ(k+1)

[0184] Among them, L c (k) represents the loss function value of the evaluation network, which is used to measure the error between the predicted value of the evaluation network and the target value;

[0185] e c (k) represents the prediction error, which is used to calculate the prediction error under the current state parameters, measure whether the cost function output by the evaluation network is accurate, and provide error input for the loss function to optimize the parameters of the evaluation network.

[0186] J(k) represents the cost function prediction value of the evaluation network for the current state parameters.

[0187] U(k) represents the instantaneous cost function of the system under the current state parameters.

[0188] γ represents the discount factor, which weighs the impact of current costs against future costs.

[0189] J(k+1) represents the cost function prediction of the evaluation network for the state parameters at the next moment.

[0190] Furthermore, the weight increment can be calculated by the gradient descent method, the weight of the evaluation network can be updated, and the loss function can be gradually optimized. The weight update formula is as follows:

[0191] W c (k+1)=W c (k)+ΔW c (k)

[0192]

[0193] W c (k) represents the current weight of the evaluation network. The weight of the evaluation network is updated by the gradient descent method to gradually optimize the loss function.

[0194] ΔW c (k) represents the weight increment of the evaluation network. According to the error and gradient calculation, the weight increment calculation formula uses the learning rate and the gradient of the error to adjust the weight to ensure that the weight update direction can reduce the prediction error.

[0195] α c(k) is the learning rate of the evaluation network, and the neuron state inside the evaluation network is s c (k), according to the input-output relationship expression, the weight update formula of the evaluation network can be further simplified as:

[0196] ΔW c (k)==-α c (k)e c (k)s c (k)

[0197] In summary, according to the input-output relationship between the model network and the evaluation network, we can further obtain:

[0198]

[0199] where σ(·) is the activation function of the neuron, θ c (k) and θ m (k) are the inputs of the dynamic neurons of the evaluation network and the model network respectively. c (k) and W m (k) are the output weight matrices of the evaluation network and the model network respectively. in,c (k) is the input weight vector of the evaluation network, W in,m (k) is the component of the input weight vector of the model network that is related to u.

[0200] In one possible implementation, reference Figure 3 The figure shows a schematic diagram of an update process of an adaptive controller provided by an embodiment of the present application. By setting the system simulation step, the upper limit of the simulation step iteration is defined, and the weights of the constituent networks of the adaptive controller are updated during the iteration process of each time step to improve the model performance. Among them, the execution network generates an estimated value of the optimal control strategy under the current state according to the current state parameters. The evaluation network A evaluates the current system state and control strategy, and calculates the relevant effect function to measure the performance. The model network predicts the state of the next time step and supports the iterative optimization of the control strategy. The evaluation network B evaluates the state of the next time step to further verify and optimize the model. At the same time, the weight increment of each network is calculated for each iteration to update the weights, improve the accuracy and robustness of the controller, and judge whether the current number of iterations reaches the maximum number of iterations. If so, the process is terminated. In this way, through continuous evaluation, prediction and iterative update, the optimal solution of the system is gradually approached.

[0201] Specifically, in the embodiment of the present application, the surface temperature of the silicon rod in the polysilicon production process can be used as the state parameter x(k) of the reduction furnace system to initialize the network parameters of the adaptive controller, for example, the adjustable parameters of the evaluation network, the execution network and the model network are set to be θ c ,θ a and θm , select the total time step N of the system simulation, for example, covering the entire reduction furnace operation cycle. The execution network outputs according to the current state parameter x(k) That is, the estimated value of the optimal control strategy in the current state. The evaluation network A outputs the cost function J(k) according to the current state parameter x(k), which reflects the impact of the current system state on the system performance, provides a basis for the next step of optimization, and uses the utility function to estimate the current control strategy. and state parameters x(k), calculate the instantaneous cost U(k). The instantaneous cost is usually directly related to indicators such as the energy consumption, production efficiency, and silicon rod quality of the reduction furnace, and is used to measure the immediate impact of the current control strategy on system performance. The model network predicts the state parameters x(k+1) of the next time step k+1 based on the current state parameters and the current control strategy valuation, providing time scalability for control strategy optimization. The evaluation network B uses the future state x(k+1) predicted by the model network to calculate the future cost function J(k+1), reflecting the potential impact of the current control strategy on the future system state. According to the current control strategy u(k), update the system state from x(k) to x(k+1). Next, the current control strategy Apply it to the system to get x(k+1), which updates the state parameters of the system and pushes the system into the next time step to provide a new state for the next round of optimization. At the same time, the gradient descent method is used to adjust the weights of the evaluation network, execution network, and model network. The weight update is as follows:

[0202] θ c (k+1)←θ c (k)+Δθ c (k)

[0203] θ a (k+1)←θ a (k)+Δθ a (k)

[0204] θ m (k+1)←θ m (k)+Δθ m (k)

[0205] Among them, Δθ c (k) represents the weight increment of the evaluation network, which is determined by the cost function error.

[0206] Δθ a (k) represents the weight increment of the execution network, which is determined by the control strategy error.

[0207] Δθ m (k) represents the weight increment of the model network, which is determined by the prediction error.

[0208] In this way, the parameters of the neural network are gradually optimized through weight updating to improve the accuracy and reliability of the control strategy.

[0209] Furthermore, check whether the number of iterations reaches the set value N. If the total time step N is not reached, continue to optimize the constituent networks of the adaptive controller and adjust the control strategy. As the simulation time step k increases, the output of the evaluation network will gradually approach the optimal cost function, and the total cost will also decrease with the improvement of the strategy and finally converge to a constant value, and the output of the action network will gradually converge to the optimal parameters to generate the optimal control strategy. Therefore, when the time step reaches N, the constituent networks of the adaptive controller meet the convergence conditions and output the final optimized target control strategy, thereby ensuring that the controller completes the optimization within a limited time and generates the target control strategy.

[0210] Specifically, refer to Figure 4 FIG. 1 is a schematic diagram showing the trajectory of a reduction furnace control strategy (U_Trajectory) changing over time provided in an embodiment of the present application. Figure 4 In the figure, the horizontal axis represents the time step. In the reduction furnace dynamic system, each time step represents an update of the control strategy on the system, usually starting from t=0 and increasing over time. The vertical axis represents the output value of the control strategy, which is the target control strategy generated by the adaptive controller at each time step, affecting the state variables such as the silicon rod temperature of the reduction furnace. Figure 4 It can be seen that the control strategy has the ability to adjust quickly in the initial stage, which means that the adaptive controller of the embodiment of the present application can efficiently respond to the deviation of the initial state of the system, which helps to reduce the time the system deviates from the target state and improve operational efficiency. The small fluctuations in the intermediate stage represent the controller's ability to strike a balance between optimizing the immediate cost and the future cost function. The small fluctuation range indicates that the optimization process of the controller has a certain robustness and can maintain a high degree of accuracy in a dynamic environment. The control strategy eventually stabilizes, which means that the control strategy of the embodiment of the present application has good convergence characteristics, and the stable control input provides a reliable guarantee for the efficient and stable operation of the subsequent reduction furnace. Further, refer to Figure 5 The figure shows the change process of the controller output and the set value provided by an embodiment of the present application. Figure 5 In the figure, the horizontal axis (Time Step) represents the time step, that is, the discrete time point in the system operation process. Each time step corresponds to an iteration of the control strategy or system state update. The vertical axis (Ts) represents the silicon rod surface temperature (Ts), which is a key production indicator of the reduction furnace production of polysilicon and is determined by the control strategy. Figure 5It can be clearly seen that Ts rises rapidly in the initial stage, which means that the controller can quickly respond to the deviation of the initial state and reduce the time to reach the set value, so that the reduction furnace can reach the optimal operating conditions in a shorter time. The small oscillation after the initial overshoot means that the controller can quickly correct the initial overshoot and converge to the target value through dynamic adjustment to ensure the dynamic stability of the system. The final trajectory completely coincides with the set value, proving that the control strategy generated by the controller can maintain the silicon rod temperature stable at the set value.

[0211] Step 103: Based on the updated adaptive controller, a target control strategy of the reduction furnace system is obtained.

[0212] In the embodiment of the present application, the system state parameters of the current reduction furnace are input into the adaptive controller to iteratively update its constituent network. After multiple rounds of iteration and verification, the target control strategy output by the adaptive controller can be obtained, that is, the optimal control parameter set that the cost function value reaches the target value. Through the target control strategy, key production parameters such as silicon rod surface temperature and silicon deposition rate can be stably controlled, the system energy consumption and cost can be reduced, the efficiency and product quality of polysilicon production can be improved, and the stable control of the reduction furnace system in a dynamically changing environment can be ensured.

[0213] In a possible implementation, the embodiment of the present application can build an adaptive dynamic programming system for the polysilicon production process through Vue.js and SpringBoot. First, the front end implements the interface display and interaction logic based on the Vue.js framework, and uses Axios as the HTTP communication tool to interact with the back end for data. Webpack is used as a module packaging tool to package JavaScript, CSS, HTML and other resource files into optimized static resources to improve the performance and loading speed of the front-end application. The back end is built with the Java SpringBoot framework, which provides a RESTful API interface for the front end to call, realize business logic processing, database operations and docking with machine learning models. Redis is used as a cache database to store frequently accessed data and implement session management to improve system response speed and scalability. Real-time communication functions such as real-time message push are realized through WebSocket to enhance user experience.

[0214] Specifically, refer to Figure 6The diagram shows a schematic diagram of a system framework provided by an embodiment of the present application, which includes a web page system, a service gateway, a service cluster, a registration center, a configuration center, a message queue, a machine learning model, a database, a distributed cache, and a distributed search. Among them, the web page system can provide a user operation interface and a system interaction entrance, and can realize the monitoring, configuration management, and data visualization operations of the polysilicon reduction furnace. The service gateway can receive the user's operation request and is responsible for load balancing, distribute the request to the corresponding service module, ensure the fast routing and reliable processing of the user request in the dynamic programming system, and avoid performance degradation due to excessive concurrency. The service cluster consists of multiple microservices, which handle different business logics in the production process respectively. Realize key functions such as real-time data analysis, dynamic programming calculation, and machine learning model reasoning in the polysilicon production process. The registration center dynamically manages the registration and discovery of services to ensure efficient and reliable calls between services. The configuration center centrally manages configuration parameters, supports dynamic updates without restarting services, and provides flexible service expansion and parameter adjustment capabilities for the adaptive dynamic programming system. The message queue realizes asynchronous communication between services to ensure data processing capabilities under high concurrency. In the production process of the reduction furnace, status update information, optimization control instructions, and machine learning results are transmitted in real time. The machine learning model provides prediction and decision support for production optimization, including dynamic programming, temperature control, chemical deposition simulation and other functions. Through model training and reasoning, it provides data support and precise control for the production optimization of the reduction furnace. Distributed cache accelerates access to high-frequency data, such as real-time monitoring data and intermediate calculation results. Distributed search provides efficient production data retrieval services. The database stores the core data of the system, including historical records, optimization results and production status, providing data support for adaptive dynamic programming, while supporting long-term production optimization and traceability analysis.

[0215] Next, the embodiment of the present application can use Nacos as a service discovery and configuration center on the microservice architecture, support dynamic registration and discovery of services, and centralized management of configurations, and improve the maintainability and scalability of the system. Use Docker containerization technology to encapsulate and deploy each microservice to ensure application consistency and convenience in different environments. At the same time, use Docker for continuous integration and continuous deployment to speed up the development process and deployment efficiency. Figure 7 The figure shows a schematic diagram of a microservice list provided in an embodiment of the present application. Different modules in the system can exist in the form of microservices, and the microservice platform ensures that these modules can be discovered and communicated in real time after deployment.

[0216] In one possible implementation, reference Figure 8The schematic diagram of a system-generated target control strategy provided by an embodiment of the present application is shown. The system provides a control index selection function, such as silicon rod surface temperature, deposition rate, etc. Through the flexible index selection function, users can set key parameters according to different production goals (such as improving quality or increasing output). And provide different control frameworks for users to choose, such as heuristics, dual heuristics, global quadratic heuristics, control-dependent heuristics, etc. After the user selects the appropriate control index and dynamic programming framework, the system can generate the target control strategy online, and provide performance index monitoring diagrams and optimal control strategy diagrams, showing the control path generated by dynamic programming, reflecting the operating status of the optimized system, so that users can monitor the actual change trends of different indicators and ensure the realization of control goals.

[0217] Specifically, heuristic dynamic programming is the most classic part of adaptive dynamic programming. It uses an evaluation network to approximate the optimal cost function, has low computational complexity, and is easy to implement. It has a wide range of applications. In dual heuristic dynamic programming, it estimates the gradient of the cost function rather than the cost itself, which improves the estimation accuracy of the gradient of the cost function, and is especially suitable for scenarios where dependency analysis of the gradient of the cost function is required. Complex control problems with fast dynamic changes and high optimization accuracy. In global quadratic heuristic dynamic programming, the output of the evaluation network is an estimate of the cost function and its gradient, which combines the advantages of heuristic and dual heuristics, and shows high accuracy and robustness in highly nonlinear and complex scenarios. Control-dependent heuristic dynamic programming adds control input as an evaluation indicator on the basis of heuristic dynamic programming. The relationship between control input and cost function directly affects the optimization process. The dependency analysis of control input can optimize specific control objectives, and is suitable for scenarios with strong constraints on control input, such as systems with limited resources or strict requirements on control variables. Control-dependent dual heuristic planning further increases the dependence on control input, so that the optimization of the cost function gradient can be combined with specific control input to achieve a more accurate control strategy. It combines the dual optimization of control input and cost function gradient, and is suitable for scenarios with high requirements for gradient optimization and control variable coupling analysis. Control-dependent global quadratic heuristic planning adds control input as an evaluation indicator, and globally estimates and optimizes the cost function and its gradient. It can simultaneously meet the requirements of high precision and strict constraints on control input. It is suitable for scenarios with extremely high complexity or high precision requirements, high dynamic changes, and extremely high control precision requirements.

[0218] Please refer to Fig. 9 Based on the same inventive concept, the embodiment of the present application further provides a reduction furnace production process control device 90, which includes:

[0219] The construction unit 901 is used to construct an adaptive controller based on a reduction furnace system; the reduction furnace system represents the response relationship between the control strategy and the state parameters in the reduction furnace production process;

[0220] An updating unit 902 is used to update each component network of the adaptive controller based on a preset updating strategy until each component network meets a preset convergence condition;

[0221] The control unit 903 is used to obtain a target control strategy of the reduction furnace system based on the updated adaptive controller to control the cost function value of the reduction furnace system to reach a target value.

[0222] Optionally, the updating unit 902 is specifically configured to:

[0223] Based on each component network, a corresponding loss function is constructed; the loss function represents the deviation between the predicted output value of the component network and the target output value;

[0224] Based on each loss function, calculate the weight update value of the corresponding component network;

[0225] Based on the weight update value, the weights of each component network are updated until the value of each loss function meets the preset convergence conditions.

[0226] Optionally, each component network includes an execution network, an evaluation network and a model network, wherein:

[0227] An execution network is used to obtain a corresponding control strategy based on state parameters of the reduction furnace system;

[0228] The model network is used to predict the state parameters of the reduction furnace system at the next moment based on the control strategy of the reduction furnace system at the current moment;

[0229] The evaluation network is used to obtain the corresponding cost function value based on the control strategy and state parameters of the reduction furnace system.

[0230] Optionally, the construction unit 901 is further used for:

[0231] Based on the polysilicon production process of the reduction furnace, a current heating model and a chemical vapor deposition model are constructed; the current heating model characterizes the current-thermal coupling relationship of the silicon rods in the reduction furnace, and the chemical vapor deposition model characterizes the reaction process of generating silicon deposition in the reduction furnace;

[0232] Based on the preset combination strategy, the current heating model and the chemical vapor deposition model are weighted combined to obtain the reduction furnace system.

[0233] Optionally, the construction unit 901 is specifically used for:

[0234] Based on a preset sampling strategy, sample points of the current heating model and the chemical vapor deposition model are obtained; the sample points represent the input variables and output variables of the corresponding models;

[0235] Based on the preset proxy modeling strategy, the sample points are modeled to obtain multiple proxy sub-models;

[0236] Based on multiple proxy sub-models, weighted combination processing is performed to obtain the reduction furnace system.

[0237] For the convenience of description, the above parts are divided into each unit module (or module) according to the function and described separately. Of course, when implementing the present application, the functions of each unit (or module) can be implemented in the same one or more software or hardware. It can be understood by those skilled in the art that various aspects of the present application can be implemented as a system, method or program product. Therefore, various aspects of the present application can be specifically implemented in the following forms, namely: a complete hardware implementation method, a complete software implementation method (including firmware, microcode, etc.), or an implementation method combining hardware and software, which can be collectively referred to as "circuit", "module" or "system" here.

[0238] The device can be used to execute the methods shown in the various embodiments of the present application. Therefore, for the functions that can be implemented by the various functional modules of the device, reference can be made to the description of the aforementioned embodiments and no further details will be given.

[0239] See also Fig.10 As shown, based on the same technical concept, an embodiment of the present application further provides a computer device 100 , which may include a memory 1001 and a processor 1002 .

[0240] The so-called memory 1001 is used to store computer programs executed by the processor 1002. The memory 1001 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of the computer device, etc. The processor 1002 may be a central processing unit (CPU), or a digital processing unit, etc. The specific connection medium between the above-mentioned memory 1001 and the processor 1002 is not limited in the embodiments of the present application. The embodiments of the present application are Fig.10 In the embodiment, the memory 1001 and the processor 1002 are connected via a bus 1003. The bus 1003 is connected to the processor 1002 via a bus 1003. Fig.10 The connection between other components is shown by bold lines, and is not intended to be limiting. The so-called bus 1003 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.10 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0241] The memory 1001 may be a volatile memory, such as a random-access memory (RAM); the memory 1001 may also be a non-volatile memory, such as a read-only memory, a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD), or the memory 1001 is any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 1001 may be a combination of the above memories.

[0242] The processor 1002 is used to execute the reduction furnace production process control method executed by the equipment in each embodiment of the present application when calling the computer program stored in the so-called memory 1001.

[0243] In some possible embodiments, various aspects of the reduction furnace production process control method provided in the present application can also be implemented in the form of a program product, which includes a program code. When the program product is run on a computer device, the program code is used to enable the computer device to execute the steps of the reduction furnace production process control method according to various exemplary embodiments of the present application described above in this specification. For example, the computer device can execute the steps of each embodiment.

[0244] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0245] The program product of the embodiment of the present application may adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on a computing device. However, the program product of the present application is not limited thereto. In the present application, the readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with a command execution system, device or device.

[0246] The readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, wherein the readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable signal medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with a command execution system, apparatus, or device.

[0247] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0248] Program code for performing the operations of the present application may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user equipment, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device via any type of network including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0249] It should be noted that, although several units or subunits of the device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided into multiple units to be embodied.

[0250] In addition, although the operations of the method of the present application are described in a specific order in the drawings, this does not require or imply that the operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0251] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0252] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0253] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A reduction furnace production process control method, characterized in that: The method comprises: Based on the reduction furnace system, an adaptive controller is constructed; the reduction furnace system represents the response relationship between the control strategy and the state parameters in the reduction furnace production process; Based on a preset update strategy, updating each component network of the adaptive controller until each component network meets a preset convergence condition; Based on the updated adaptive controller, a target control strategy of the reduction furnace system is obtained to control the cost function value of the reduction furnace system to reach a target value.

2. The method according to claim 1, characterized in that The updating process of each component network of the adaptive controller based on a preset updating strategy until each component network satisfies a preset convergence condition includes: Based on the loss functions corresponding to the component networks, respectively obtain the loss values ​​of the component networks; the loss values ​​represent the deviation between the predicted output value and the target output value of the corresponding component network; Based on each loss value, obtain the weight update value of the corresponding component network; Based on the weight update value, the weights of the component networks are updated until the loss function values ​​meet the preset convergence conditions.

3. The method according to claim 2, characterized in that Each of the component networks includes an execution network, and based on the loss functions corresponding to each of the component networks, the loss values ​​of each of the component networks are obtained respectively, including: Based on the execution network, predicting and processing the state parameters of the reduction furnace system to obtain corresponding control strategy estimates; Based on the control strategy valuation and the loss function, a loss value of the execution network is obtained.

4. The method according to claim 3, characterized in that Each of the component networks also includes a model network, and based on the loss functions corresponding to each of the component networks, the loss values ​​of each of the component networks are obtained respectively, including: Based on the model network, predicting the control strategy valuation, and obtaining the state parameters of the reduction furnace system at the next moment; Based on the state parameters and loss function of the reduction furnace system at the next moment, the loss value of the model network is obtained.

5. The method according to claim 4, characterized in that Each of the component networks also includes an evaluation network, and based on the loss functions corresponding to each of the component networks, the loss values ​​of each of the component networks are obtained respectively, including: Based on the evaluation network, cost prediction processing is performed on the control strategy estimation value and the state parameter to obtain a corresponding cost function prediction value; Based on the cost function prediction value and the loss function, a loss value of the evaluation network is obtained.

6. The method according to claim 1, characterized in that Before constructing an adaptive controller based on the reduction furnace system, the method further includes: Based on the polysilicon production process of the reduction furnace, a current heating model and a chemical vapor deposition model are constructed; the current heating model characterizes the current-thermal coupling relationship of the silicon rods in the reduction furnace, and the chemical vapor deposition model characterizes the reaction process of generating silicon deposition in the reduction furnace; Based on a preset combination strategy, the current heating model and the chemical vapor deposition model are weighted combined to obtain the reduction furnace system.

7. A reduction furnace production process control device, characterized in that: The device comprises: A construction unit is used to construct an adaptive controller based on a reduction furnace system; the reduction furnace system represents a response relationship between a control strategy and a state parameter in a reduction furnace production process; An updating unit, configured to update each component network of the adaptive controller based on a preset updating strategy until each component network satisfies a preset convergence condition; The control unit is used to obtain a target control strategy for the reduction furnace system based on the updated adaptive controller to control the cost function value of the reduction furnace system to reach a target value.

8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising computer program instructions, characterized in that When the computer program instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images