A polysilicon reduction furnace adaptive control method and system

Abstract

This invention provides an adaptive control method and system for a polysilicon reduction furnace, relating to the field of polysilicon production. The method includes the following steps: establishing an offline multiphysics simulation model based on the physical structure of the target reduction furnace; running simulation calculations under different preset process conditions to fit a mapping relationship model characterizing the correspondence between silicon rod diameter / surface area, furnace flow field conditions, and mass transfer conditions, and storing this model in a database; calling the corresponding mapping relationship model based on production plan parameters and performing operating condition calibration; performing iterative calculations starting from the initial silicon rod state and process settings to generate a process control data table; controlling the operation of the reduction furnace according to the process control data table and collecting actual operating data in real time; comparing this data with the corresponding predicted values ​​in the process control data table and dynamically adjusting the process data. This solution reduces reliance on human experience and improves system adaptability, process stability, reproducibility, and production efficiency.

Description

Technical Field

[0001] This invention relates to the field of polysilicon production, and more specifically, to an adaptive control method and system for a polysilicon reduction furnace. Background Technology

[0002] In polysilicon production, the modified Siemens process is a core technology that uses chemical vapor deposition to grow silicon rods in a reduction furnace. Precise control of the reduction furnace is crucial for ensuring product quality, improving production efficiency, and reducing energy consumption. Current mainstream automatic control systems primarily rely on real-time measurement and feedback of the total resistance of the silicon rod. However, this single signal of total resistance is simultaneously affected by two variables: the diameter of the silicon rod and temperature. Neither parameter can be uniquely determined, making it difficult for the control system to accurately distinguish between diameter growth and temperature fluctuations, resulting in control lag and blind spots.

[0003] Existing technologies attempt to assist with direct measurement (such as infrared thermography) or indirect estimation, but these methods are prone to errors in the complex furnace environment and lack the ability to provide proactive control. This results in long process debugging cycles, poor process stability and reproducibility across different furnace types or batches, high dependence on operator experience, and potential waste of energy and materials. Therefore, there is an urgent need for an intelligent control scheme that can deeply integrate the physicochemical mechanisms within the reduction furnace and possess predictive and adaptive adjustment capabilities. Summary of the Invention

[0004] The purpose of this invention is to provide an adaptive control method and system for a polysilicon reduction furnace. This system can establish a digital twin model of the reduction furnace by constructing a three-layer intelligent control architecture consisting of offline simulation modeling, online predictive optimization, and real-time feedback calibration, thereby achieving a fundamental shift from passive control based on apparent parameters to predictive feedforward control based on internal mechanism parameters.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0006] An adaptive control method and system for a polycrystalline silicon reduction furnace includes the following steps:

[0007] Based on the physical structure of the target reduction furnace, an offline multiphysics simulation model was established; and simulation calculations were performed under different preset process conditions to obtain a mapping relationship model that characterizes the relationship between silicon rod diameter / surface area, furnace flow field conditions and mass transfer conditions, and stored in the database.

[0008] Based on the production plan parameters, the corresponding mapping relationship model is retrieved from the database, and the mapping relationship model is calibrated under operating conditions. Starting from the initial silicon rod state and process setting value, iterative calculations are performed in combination with the calibrated mapping relationship model to generate a process control data table that changes over time.

[0009] The reduction furnace is controlled according to the process control data table, and the actual operating data is collected in real time. The actual operating data is compared with the corresponding predicted values ​​in the process control data table to dynamically adjust the process data.

[0010] Furthermore, the mapping relationship model is a model that characterizes the correspondence between the silicon rod diameter / surface area and the surface gas velocity and boundary layer deviation coefficient; wherein, the surface gas velocity is defined as the average airflow velocity at a specific location on the silicon rod surface, and the boundary layer deviation coefficient is used to characterize the degree of non-uniformity of the actual flow field distribution.

[0011] Further, the step of calibrating the mapping relationship model under operating conditions includes:

[0012] Based on the feed quantity, proportion, and temperature parameters in the feed schedule, the mapping relationship model is adjusted using a preset calibration formula; wherein, the calibration formula includes:

[0013] Surface gas velocity data calibration formula: ;

[0014] in: The surface gas velocity after calibration, in m / s; The reference surface velocity is calculated by simulation software, and the unit is m / s. Total molar amount of reference material, unit: mol; This represents the change in the molar amount of hydrogen, in mol. This represents the change in the molar amount of trichlorosilane, in mol. Feed temperature, unit: K;

[0015] Boundary layer deviation coefficient calibration formula: ;

[0016] in: To calibrate the boundary layer deviation coefficient; The baseline boundary layer deviation coefficient is calculated by simulation software. R represents the feed rate of trichlorosilane, in kg; R is the molar ratio of hydrogen.

[0017] Furthermore, the step of iteratively calculating based on the calibrated mapping model to generate a time-varying process control data table includes:

[0018] Based on the silicon rod diameter at the current time point, the corresponding surface gas velocity and boundary layer deviation coefficient are obtained from the calibrated mapping relationship model;

[0019] Based on the process setting temperature and feed ratio, and combined with the obtained surface gas velocity and boundary layer deviation coefficient, the silicon deposition amount is calculated through the surface chemical reaction equilibrium model, and the silicon rod diameter for the next time node is updated accordingly.

[0020] Further, the step of calculating the silicon deposition amount using a surface chemical reaction equilibrium model includes:

[0021] Based on the temperature and pressure conditions set in the process, the equilibrium equations for multiple independent chemical reactions, including the reaction between trichlorosilane and hydrogen, are solved to obtain the concentrations of each gas phase component under equilibrium conditions and the theoretical maximum silicon deposition rate (Si). max .

[0022] Based on the current silicon rod diameter, obtain the preset deposition efficiency coefficient δ, and calculate the actual silicon deposition amount Si = δ × Si max .

[0023] Furthermore, the generation of the time-varying process control data table also includes:

[0024] After the iterative calculation at each time point is completed, the diameter difference Dd, which characterizes the uniformity of silicon rod growth, is calculated based on the boundary layer deviation coefficient and the current silicon rod diameter increment.

[0025] When the diameter difference Dd exceeds a preset threshold, a risk warning signal for the inverted rod is generated and output.

[0026] Furthermore, the generation of the time-varying process control data table also includes:

[0027] For each time point, based on the reduction furnace parameters, silicon rod parameters, and fluid material parameters of that time point, the convective heat transfer coefficient and the radiative heat transfer coefficient are calculated, and the overall heat transfer coefficient is determined.

[0028] Based on the overall heat transfer coefficient, the corresponding gas phase outlet temperature is determined by iteratively solving the energy balance equation.

[0029] Based on the gas phase outlet temperature, the gas heating power and radiative heat dissipation power are calculated, summed, and losses are included in a preset proportion to obtain the total operating power P at that time point. 总 and unit power consumption.

[0030] Furthermore, the generation of the time-varying process control data table also includes:

[0031] Based on the total operating power P 总 The required current and voltage settings for this time node are calculated based on the electric power formula, along with the resistivity ρ, length L, and cross-sectional area A of the silicon rod at the corresponding temperature.

[0032] An adaptive control system for a polysilicon reduction furnace, used to implement the method, the system comprising:

[0033] The simulation modeling and database module is used to establish the offline multiphysics simulation model, fit and store the mapping relationship model;

[0034] The process optimization and prediction module is used to call the model, perform operating condition calibration and iterative calculation, and generate the process control data table.

[0035] The real-time control and feedback module is used to execute the process control data table and is connected to the sensors and actuators to collect the actual operating data and complete the dynamic adjustment.

[0036] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned adaptive control method for a polysilicon reduction furnace.

[0037] The present invention has at least the following advantages or beneficial effects:

[0038] This invention establishes an offline multiphysics simulation model based on the physical structure of the target reduction furnace and fits and stores a mapping relationship model. This structure digitizes and productizes the complex multiphysics coupling mechanism within the furnace and expert experience, constructing a portable and reusable digital process knowledge base. This allows for immediate access and rapid optimization of control parameters when a new furnace type is put into production or when processes are changed, significantly shortening the lengthy debugging cycle required by the traditional trial-and-error correction model. Furthermore, by calling and calibrating the mapping relationship model based on production plan parameters, and then iteratively generating a process control data table, this invention enables full-cycle analysis of silicon rod growth trajectory, energy consumption, and electrical parameters based on the mechanism model before production begins. High-precision simulation prediction and optimization have enabled a fundamental shift from lagging control (production first, adjustment later) to feedforward control (prediction first, parameter pre-optimization), reducing material and energy waste caused by blind process adjustments at the source. By executing control based on process control data sheets and dynamically adjusting by collecting data in real time to compare with predicted values, a complete closed loop of feedforward prediction and feedback calibration can be formed in actual operation. This allows the system to automatically sense and compensate for real-time interference such as furnace differences and raw material fluctuations, significantly improving the adaptability, long-term stability, and consistency of process effects between different furnace batches and furnace types, and reducing the reliance on the personal experience of core operators. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 A flowchart illustrating the steps of the adaptive control method for a polycrystalline silicon reduction furnace provided in the embodiments of the application;

[0041] Figure 2 This is an example of a cloud map showing the temperature and flow rate distribution inside a reduction furnace obtained from a multiphysics simulation of the application embodiment;

[0042] Figure 3 This is a residual convergence graph from the simulation calculations of the application embodiment;

[0043] Figure 4 This is a schematic diagram illustrating the data calculated using simulation of a silicon rod with a diameter of 40mm in the embodiment of the application.

[0044] Figure 5 The figure shows the fitting relationship curve between the diameter of the silicon rod, the surface gas velocity, and the boundary layer deviation coefficient in the embodiment of the application.

[0045] Figure 6 This is the feature model selection interface in the online prediction and process parameter optimization steps of the application embodiment;

[0046] Figure 7 The graphs showing the silicon rod diameter, surface gas velocity, and boundary layer deviation coefficient before and after calibration are shown in the example provided in the application.

[0047] Figure 8 The thermodynamic equilibrium curve of the Si-Cl-H ternary system at 0.52 MPa is shown in the application example.

[0048] Figure 9 This is a graph showing the relationship between the diameter of the silicon rod and δ in the application embodiment;

[0049] Figure 10 This is a time-varying curve of the silicon rod diameter data from the application embodiment;

[0050] Figure 11 This is a time-varying curve of the current data in the application embodiment;

[0051] Figure 12 This is an interface diagram of the silicon rod temperature infrared thermometer used in the application embodiment. Detailed Implementation

[0052] Example

[0053] Please refer to Figure 1The figure shown is a flowchart of the adaptive control method for polycrystalline silicon reduction furnace in an embodiment of the present invention.

[0054] This embodiment provides an adaptive control method for a polycrystalline silicon reduction furnace. Its core lies in establishing an accurate mechanistic model through multiphysics simulation, and then using this model for feedforward prediction and online calibration. The method mainly includes the following three steps:

[0055] S110. Establish an offline multiphysics simulation model library and feature database: Based on the physical structure of the target reduction furnace, establish an offline multiphysics simulation model; and run simulation calculations under different preset process conditions to fit and obtain a mapping relationship model that characterizes the correspondence between silicon rod diameter / surface area, furnace flow field conditions and mass transfer conditions, and store it in the database.

[0056] S120, Online Prediction and Process Parameter Optimization: Based on production plan parameters, the corresponding mapping relationship model is called from the database, and the mapping relationship model is calibrated under operating conditions; Starting from the initial silicon rod state and process setting value, iterative calculation is performed in combination with the calibrated mapping relationship model to generate a process control data table that changes over time;

[0057] S130. Real-time closed-loop control and data calibration based on process control data: Control the operation of the reduction furnace according to the process control data table and collect actual operating data in real time; compare the actual operating data with the corresponding predicted values ​​in the process control data table to dynamically adjust the process data.

[0058] Step S110 above establishes an offline multiphysics simulation model library and feature database, specifically including: First, modeling is performed based on the physical structure of the target reduction furnace, such as the number of rods, gas inlet layout, and silicon rod arrangement, to establish a high-fidelity CFD-CVD (Computational Fluid Dynamics-Chemical Vapor Deposition) coupled simulation model. This model comprehensively solves for multiphysics processes such as fluid flow, heat transfer, mass transfer, and surface chemical reactions. Second, simulation calculations are performed by running simulations under a large number of different preset process conditions, such as different gas inlet flow rates, ratios, power, and silicon rod diameters. Multi-point calculations are performed according to the production stage (such as nodes with different silicon rod diameters). The more calculation points, the higher the data accuracy. The gas velocity distribution, temperature distribution, and component concentration distribution inside the furnace are calculated. Then, the core relationship curve is fitted. By extracting key feature parameters, a mapping relationship model that can characterize the intrinsic correspondence between silicon rod diameter / surface area and key flow fields and mass transfer conditions inside the furnace is finally fitted. Finally, the relationship curves or response surface models fitted under different furnace types and operating conditions are stored in the furnace type feature database; and they are classified and numbered according to different models so that on-site operators can make selections.

[0059] It should be noted that an appropriate convergence factor should be selected during the calculation to prevent unstable convergence. When the residual is below 0.001 and the imbalance of mass, heat transfer, and radiation is below 2%, the model calculation results are highly consistent with the actual results. At this point, the calculation should be stopped and the corresponding data should be read.

[0060] As an example, the above mapping model characterizes the relationship between silicon rod diameter / surface area and surface gas velocity and boundary layer deviation coefficient. When fitting the core relationship curve, the focus is on extracting the surface gas velocity and boundary layer thickness / concentration gradient at specific locations on the silicon rod surface (e.g., center, top, bottom). The surface gas velocity is defined as the average gas velocity parallel to the surface at a point 1 mm above the silicon rod surface; the boundary layer deviation coefficient is defined as the ratio of the actual boundary layer thickness to the theoretical boundary layer thickness under ideal fully developed flow. In other words, for each furnace type, a relationship curve or response surface model is fitted to the relationship between silicon rod diameter / surface area, surface gas velocity, and boundary layer deviation coefficient.

[0061] This model reveals the bridge between macroscopically measurable / controllable parameters and the microscopic reaction environment, and is categorized and stored in a feature database, forming a reusable digital process knowledge base for enterprises. By digitizing and systematizing scattered expert experience and physical mechanisms, it provides a readily available basic model for the commissioning of new furnace types or process changes, significantly shortening the debugging cycle required by traditional trial-and-error methods.

[0062] The above step S120, online prediction and process parameter optimization (before furnace start-up / production planning stage), specifically includes: When actual production begins, the system calls the corresponding mapping relationship model based on production planning parameters (such as the feed schedule). These production planning parameters mainly include: running time, chlorosilane feed rate, hydrogen feed rate or ratio, and silicon rod surface temperature (which can be set uniformly or separately for each phase). Based on these parameters, the specific furnace type number used for this production is selected, and the system retrieves the corresponding feature model from the database.

[0063] Secondly, since the actual feed rate and proportions differ from the simulation baseline, the system will use the silicon rod diameter-surface gas velocity-boundary layer deviation coefficient fitting curve data stored in the feature database as the benchmark, and the feed rate input by the on-site operator as the target, to perform operating condition calibration on the silicon rod diameter-surface gas velocity-boundary layer deviation coefficient fitting curve through the following equations. The calibration formulas include:

[0064] Silicon rod diameter / surface area - surface gas velocity data calibration formula: ;in: The surface gas velocity after calibration, in m / s; The reference surface velocity is calculated by simulation software, and the unit is m / s. Total molar amount of reference material, unit: mol; This represents the change in the molar amount of hydrogen, in mol. This represents the change in the molar amount of trichlorosilane, in mol. The feed temperature is expressed in Kelvin (K).

[0065] Silicon rod diameter / surface area - boundary layer deviation coefficient calibration formula: ;in: To calibrate the boundary layer deviation coefficient; The baseline boundary layer deviation coefficient is calculated by simulation software. R represents the feed rate of trichlorosilane, in kg; R is the molar ratio of hydrogen.

[0066] Subsequently, starting with the initial silicon rod (silicon core) state and process settings (such as target temperature), the system performs closed-loop iterative calculations using a calibrated mapping model. The final output is optimal process control data for parameters such as time, silicon rod diameter, deposition rate, current, voltage, silicon powder yield, total power, and power loss. This dynamically generates a process control data table covering the entire production cycle. It also provides predicted results for final yield, power consumption, and quality, with deviations from actual results typically less than 5%. If abnormal predicted conditions occur, the system will issue an alert, allowing process engineers to adjust the feed rate in a timely manner. This forms the basis of predictive feedforward control.

[0067] The optimal control data table generated above mainly includes data such as silicon rod diameter and silicon deposition amount, rod collapse risk warning, heat exchange power and gas phase temperature, current, and voltage. The above data are calculated, integrated and processed, and then output to form a data table.

[0068] As an example, the calculation process for silicon rod diameter and silicon deposition amount in the above-mentioned process of generating the optimal control data table includes: at each time point, the system obtains the corresponding surface gas velocity and boundary layer deviation coefficient from the calibration model based on the current silicon rod diameter; then, based on the process set temperature and feed ratio, the silicon deposition amount is calculated through the surface chemical reaction equilibrium model. The specific calculation process is as follows:

[0069] Currently, the polycrystalline silicon CVD process mainly considers 10 important gaseous substances: SiHCl3, H2, HCl, SiCl4, SiCl3, SiCl2, SiCl, SiH2Cl2, SiH3Cl, and SiH4, and one solid substance: Si. Only silicon exists as a solid phase; the rest are gaseous. The chemical vapor deposition reaction equations are shown in Table 1 below.

[0070]

[0071] Based on the Brinklev method, the chemical equilibrium equations for eight independent reactions, including the reaction between trichlorosilane and hydrogen, are first solved according to the temperature and pressure conditions set for the process. This yields the equilibrium concentrations of each component and the theoretical maximum silicon deposition rate (Si). max Subsequently, a nonlinear equation system incorporating the mole fractions of all key gaseous components was established and solved using numerical methods (such as the Newton-Raphson method) to obtain the precise mole fractions of each gaseous component at chemical equilibrium. The Si / Cl ratio in the gas phase at equilibrium was then calculated, denoted as (Si / Cl)eq. This ratio was compared with the (Si / Cl)feed of the feed gas, allowing for the determination of the Si / Cl ratio using the formula... The theoretical equilibrium silicon yield η is calculated, and then the theoretical maximum silicon deposition rate Si under the current feed rate is derived. max This step determined the theoretical upper limit of silicon deposition under this condition from a purely chemical thermodynamic perspective.

[0072] However, the actual deposition process is limited by the mass transfer rate. Since reactants need to pass through the gas boundary layer on the silicon rod surface via convection and diffusion to reach the rod surface for reaction, the actual deposition rate is lower than the theoretical maximum. To quantify this effect, a deposition efficiency coefficient δ is introduced. This coefficient δ is a parameter closely related to the silicon rod diameter, essentially encapsulating the combined influence of the specific surface area determined by the silicon rod diameter, the surface flow field (gas velocity), and the boundary layer state on the reactant mass transfer efficiency. Its specific value is predetermined through the aforementioned offline simulation and data analysis, forming a curve showing the relationship between δ and the silicon rod diameter, as shown below. Figure 9 As shown, it is stored in the system.

[0073] Finally, the system calculates the actual silicon deposition amount Si at the current time point using the formula Si = δ × Si max This formula combines thermodynamic equilibrium (determining the upper limit) with transport kinetics (determining efficiency), achieving a precise physical simulation of the actual production process. Based on the calculated actual silicon deposition amount Si and known parameters such as the initial length of the silicon rod and the density of silicon, the system can use the cylinder volume formula to back-calculate and update the silicon rod diameter for the next time point, using this as a new starting point for the next round of iterative calculations. Through this process, a precise and forward-looking simulation of the silicon rod growth trajectory based on physicochemical mechanisms is achieved. It not only predicts how large it will grow but also reveals why it grows to such a large size.

[0074] As an example, the steps for generating the optimal control data table, including the risk warning of silicon rod collapse, specifically include: after the iterative calculation at each time point, based on the calculated boundary layer deviation coefficient (reflecting flow field inhomogeneity) and silicon rod diameter increment, calculating the silicon rod diameter range Dd using the formula Dd=(D1-D0)×boundary layer deviation coefficient×20%. When Dd exceeds a threshold (e.g., 0.25mm), the system alarms, significantly improving the safety of the production process and allowing process engineers to adjust parameters in advance to avoid furnace collapse accidents caused by uneven silicon rod growth. Here, 20% is a correction coefficient, which can be adjusted according to production stability. Under the premise of no furnace collapse, this value can be appropriately increased to improve silicon yield. D0 represents the initial silicon rod diameter at each time point, D1 represents the initial silicon rod diameter at the next time point, and Dd refers to the difference between the calculated average silicon rod diameter increment and the minimum silicon rod diameter increment, reflecting the diameter difference between the thinnest and thickest silicon rods affected by the flow field distribution within the furnace. The Dd value will be calculated every hour in the third step. This value is set with a warning threshold of 0.25mm. That is, when Dd≥0.25mm, the difference in the diameter of the surface silicon rod is too large, the process control data calculation system will issue a warning, and the process engineer needs to adjust the process parameters to prevent the rod from turning over.

[0075] It should be noted that during the polycrystalline silicon production process, the deposition of silicon atoms on the surface of the silicon rod is not uniform; there are localized phenomena of excessively fast and slow deposition. This is due to the uneven distribution of the flow field within the furnace. Excessive non-uniform deposition can lead to excessively thin silicon rods in certain areas, significantly increasing the risk of rod collapse during production. Therefore, the calculation system uses a coupled calculation of boundary layer non-uniformity parameters and the average diameter of the silicon rod to output the difference in silicon rod diameter, ensuring the stability of the production process.

[0076] As an example, to generate complete control commands, the system also needs to calculate the energy and electrical parameters required to maintain the process. The calculations of heat exchange power and gas phase temperature, as well as current and voltage calculations, during the generation of the optimal control data table, specifically include:

[0077] For each time point, based on the reduction furnace parameters, silicon rod parameters, and fluid material parameters for that point—namely, the determined silicon rod diameter, surface gas velocity, set temperature Ts, wall temperature Tw, and other material properties—the convective heat transfer coefficient hc and the radiative heat transfer coefficient hr are calculated, thus yielding the overall heat transfer coefficient U = hc + hr. The calculation formula is as follows:

[0078] The convective heat transfer coefficient hc = Nu × k / Dh = (0.023 × Re) ^0.8 ×Pr ^0.4×0.050) / Dh; where Nu is the Nusselt number; K is the thermal conductivity of the fluid; Pr is the Prandtl number; Re is the Reynolds number; Dh is the diameter of the reduction furnace channel, Dh=2×(silicon rod spacing-silicon rod diameter);

[0079] Radiative heat transfer coefficient Where σ = 5.67 × 10 -8 , is the Stefan-Boltzmann constant; ε=0.75, is the surface emissivity; Ts is the silicon rod set temperature; Tw is the inner wall surface temperature.

[0080] Then, based on the overall heat transfer coefficient U, the energy balance equation MCp(T) is applied. out -T in )=UA∆T lm T in Given the feed temperature, the logarithmic mean temperature difference ∆Tlm = (T out -T in ) / ln(T s -T in T s -T out Iteratively solve for the gas phase outlet temperature T. out .

[0081] Based on the gas phase outlet temperature, the gas heating power Q and radiative heat dissipation power P can be calculated. After summing these and considering a small amount of additional losses, the total operating power P is obtained. 总 and unit power consumption.

[0082] Based on total operating power P 总 Given the resistivity ρ, length L, and cross-sectional area A of the silicon rod at the corresponding temperature, the required current and voltage setpoints for that time node are calculated based on the power formula. The required current I and voltage U setpoints are derived by reverse calculation. This achieves a precise conversion from process objectives to specific execution parameters (current and voltage), providing direct setpoints for automatic control.

[0083] Finally, all iterations and calculation results are integrated into a detailed process control data sheet, including time, silicon rod diameter, silicon yield, current, voltage, total power, unit power consumption, risk status, etc., as a blueprint for automatic control.

[0084] In step S130, the real-time closed-loop control and data calibration based on process control data specifically includes: During production operation, the system controls the reduction furnace (execution control) according to the data table through the power regulator and mass flow controller (MFC). Simultaneously, it collects real-time operating data such as infrared temperature measurement of the silicon rod surface, diameter measurement, current, and voltage (real-time feedback), and compares the measured values ​​with the predicted values ​​in the data table in real time. If a deviation occurs, the system dynamically fine-tunes the setpoints of current and feed rate to achieve adaptive calibration. This forms a complete closed loop of prediction-execution-monitoring-calibration, enabling the system to automatically adapt to disturbances such as raw material fluctuations and equipment status changes, maintaining the process in optimal operation over the long term.

[0085] This invention integrates an offline simulation layer (establishing a mechanism model), a static knowledge base layer (storing furnace type feature mapping relationships), an online optimization layer (personalized parameter optimization before production), and a real-time control layer (dynamic calibration during operation). It systematizes, digitizes, and productizes scattered expert experience, physical mechanisms, and operational data, constructing a core process brain for the enterprise and forming a complete, self-learning intelligent control ecosystem. This completely changes the traditional trial-and-error process transfer model, ultimately realizing an intelligent upgrade of polysilicon reduction production from experience-driven to data- and mechanism-driven hybrid approaches. When a new furnace type is put into production or a process is changed, control parameters can be readily used and quickly optimized, significantly shortening the debugging cycle and ensuring the consistency and replicability of process effects between different furnace types (the knowledge base can be continuously iterated and optimized as operational data accumulates).

[0086] Compared with existing technologies, the system offers several advantages: Firstly, it introduces a multiphysics simulation model into the control closed loop, enabling the control system to understand the microscopic reaction processes within the furnace. The control objective shifts from apparent parameters to mechanistic parameters, significantly improving the scientific rigor and accuracy of the control. Before production begins, process results can be predicted and parameters optimized through digital simulation, avoiding the lag and material and energy waste associated with traditional methods that involve production first and adjustments later. This proactive prediction reduces blind adjustments. Furthermore, by using real-time data to infer key states and compare them with predictions, the system possesses online calibration and adaptive adjustment capabilities. It can automatically adapt to disturbances such as furnace differences and raw material fluctuations, significantly improving the stability and replicability of processes across different furnace batches and types. This forms a reusable and inheritable "digital process knowledge base" for the enterprise, reducing the technical risks associated with personnel turnover.

[0087] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings and a specific example. These examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0088] I. Establish an offline multiphysics simulation model library and feature database

[0089] Taking a certain type of 48-pair rod reduction furnace as an example, full-size or symmetrical models incorporating physical fields such as turbulence, chemical reaction, mass transfer, heat transfer, and radiative heat transfer are established using the mainstream reduction furnace simulation software PolySim3D for simulation. Reduction furnace models with different polycrystalline silicon rod diameters are also established.

[0090] Step 1: Modeling: A full-size model was constructed to scale using the structural data of a 48-pair reduction furnace. The silicon rod diameter is 40mm, the air inlet distribution is 3-12-12, the air inlet diameter is 16-16-16 (mm), and the silicon core height is 2m. The mesh size is 0.01m, the first layer mesh thickness is 0.005m, and the transition factor is 1.5. Figure 2 The image shows an example of temperature and flow velocity distribution cloud maps obtained from multiphysics simulation inside a reduction furnace.

[0091] The second step is boundary conditions: After modeling is completed, input data such as pressure, absorption rate, silicon rod temperature, cooling water inlet and outlet temperatures, feed rate, feed temperature, and current. The above data are process data and need to be measured and calculated to ensure consistency with reality.

[0092] The third step is data acquisition: During calculation, an appropriate convergence factor should be selected to prevent unstable convergence. When the residual is below 0.001 and the imbalance of mass, heat transfer, and radiation is below 2%, the model calculation results are highly consistent with reality. At this point, the calculation should be stopped, and the corresponding data should be acquired. For example... Figure 3 The diagram shown is a residual convergence plot.

[0093] The above simulation calculated the average gas velocity and boundary layer deviation data on the silicon rod surface at a node with a silicon rod diameter of 40 mm, such as... Figure 4 The image shown is a screenshot of the core data for this node. Then, calculations are performed for other silicon rod diameter nodes. The calculation method is the same as for the 40mm node, but different silicon rod diameters need to be selected during modeling. In this embodiment, calculations were performed for six nodes with silicon rod diameters of 40mm, 60mm, 80mm, 100mm, 120mm, and 140mm.

[0094] Through the above simulation calculations, the average gas velocity and boundary layer deviation data of the silicon rod surface at six nodes were obtained. Then, the relationship curve of silicon rod diameter-surface gas velocity-boundary layer deviation coefficient of the model was obtained by fitting, as shown in the figure. Figure 5 As shown. The above curves are classified as follows: 48 pairs of rods reduction - air inlet distribution 3-12-12 - air inlet diameter 16-16-16 (mm) - silicon core height 2m.

[0095] It should be noted that for other furnace types and different air inlets, recalculation is required, and corresponding relationship curves need to be fitted for classification. Finally, a corresponding feature database should be established for classification and storage. Simulation software such as COMSOL Multiphysics and ANSYS can also be used for calculation, but the model must include modules for turbulence, chemical reaction, mass transfer, heat transfer, and radiation. The calculation results must converge to ensure accuracy.

[0096] II. Online Prediction and Process Parameter Optimization (Before Furnace Start-up / Production Planning Stage)

[0097] 1. Bill of Materials (BOM) Formulation and Model Selection

[0098] On-site operators select the furnace batch to run in the process control data calculation system. The example production uses a characteristic model with 48 pairs of rods, a 3-12-12 inlet distribution, an inlet diameter of 16-16-16 (mm), and a silicon core height of 2m. Then, the feed schedule is entered into the system. The main process parameters are: running time, chlorosilane feed rate, hydrogen feed rate or ratio, silicon rod surface temperature (which can be set uniformly or separately for each phase), and feed temperature. Figure 6 The image shows the feature model selection interface.

[0099] 2. Calibration of the fitting curve of silicon rod diameter-surface gas velocity-boundary layer deviation coefficient

[0100] The process control data calculation system uses the silicon rod diameter-surface gas velocity-boundary layer deviation coefficient fitting relationship curve data stored in the feature database as a benchmark, and takes the feed rate input by the on-site operator as the target. It calibrates the silicon rod diameter-surface gas velocity-boundary layer deviation coefficient fitting relationship curve through the silicon rod diameter / surface area-surface gas velocity data calibration formula and the silicon rod diameter / surface area-boundary layer deviation coefficient calibration formula, respectively.

[0101] like Figure 7 The figures show the curves before and after calibration of silicon rod diameter, surface gas velocity, and boundary layer deviation coefficient. After system calibration, the fitted curves show that the surface gas velocity deviates from the actual value by less than 0.1 m / s, and the boundary layer deviation deviates from the actual value by less than 5 percentage points, demonstrating good accuracy.

[0102] 3. Fitting calculation of silicon deposition amount and silicon rod diameter

[0103] In the first hour, a 20mm diameter silicon core was used (the initial silicon rod diameter). The control temperature set in the material list was 1300K, the TCS (trichlorosilane) feed rate was 2t, and the hydrogen ratio was 2.5. The process control data calculation system read the boundary layer deviation and average gas velocity data calibrated when the diameter was 20mm, which were 54% and 1.5m / s, respectively. Based on these two data, combined with the control temperature T of 1300K, the polycrystalline silicon deposition rate was calculated using the following logic.

[0104] Currently, the polycrystalline silicon CVD process mainly considers 10 important gaseous substances: SiHCl3, H2, HCl, SiCl4, SiCl3, SiCl2, SiCl, SiH2Cl2, SiH3Cl, and SiH4, and one solid substance: Si. Only silicon exists as a solid phase; the rest are gaseous. It involves 8 chemical reaction equations, which are listed in Table 1 above.

[0105] First, based on the thermodynamic data of the relevant pure substances, calculate the Gibbs free energy of each reaction at temperatures between 1000K and 2000K using the following formula:

[0106]

[0107]

[0108]

[0109]

[0110] After obtaining the Gibbs free energy of each reaction, the thermodynamic formula lnK is used. P =-ΔG0 / RT, calculate the standard equilibrium constant of the reaction at the reaction temperature; where K P The standard pressure equilibrium constant, ΔG0 is the standard molar Gibbs free energy change, R is the molar gas constant, and T is the thermodynamic temperature.

[0111] The system then assigns each component to yx, and the mole fraction of each component is shown in Table 2 below:

[0112]

[0113] Based on current calculations for pressure reduction furnaces, and using the formula for calculating the reaction equilibrium constant, the following formulas for calculating the equilibrium molar ratios of each component are derived:

[0114] Under specific temperature, pressure, and Cl / H ratio conditions (temperature T = 1300 K, furnace pressure 0.52 MPa, molar ratio of feed hydrogen to trichlorosilane (TCS) r = 2.5), the Cl / H atomic ratio of the feed can be calculated. =0.5), when the reaction reaches equilibrium, the above nonlinear equations can be solved using the Newton-Raphson method to obtain y1 and y2. Then, the equilibrium mole fractions of other gas phase components can be calculated to obtain the content of each component. Thus, the Si / Cl ratio in the gas phase at equilibrium can be calculated, denoted as (Si / Cl)eq, and a formula can be developed as follows. Figure 8 The curve shown. The feed ratio, denoted as (Si / Cl)feed, can be calculated from the feed gas, and the equilibrium silicon yield (η) can then be calculated using the following formula:

[0115] After the system reaction reaches full completion in the first hour, the maximum silicon deposition Si max for:

[0116] Si max =28×η×TCS / 135.45=28×16.21%×2000kg / 135.45=67.02kg; where TCS is the feed mass of trichlorosilane.

[0117] In the actual reaction process, the contact time between the material and the silicon rod is limited due to the influence of the specific surface area (diameter of the silicon rod). Therefore, the actual silicon yield Si = δ × Si max δ is a fixed value related to the diameter of the silicon rod, and its data curve is shown in the figure. Figure 9 As shown.

[0118] The δ data is built into the system. When the diameter of the silicon rod (silicon core) is 20mm, δ=0.38. After the system reads this data, it is substituted into the silicon yield formula:

[0119] Si = δ × Si max =0.38 × 67.02 kg = 25.4676 kg

[0120] The initial rod length L = 203 m, the initial diameter D0 = 20 mm = 0.02 m, and the polycrystalline silicon density ρ 硅 =2330kg / m, the increased silicon mass due to deposition Δm=25.4676kg, the initial mass m0=148.61kg.

[0121] Total mass after sedimentation:

[0122]

[0123] Total volume after sedimentation:

[0124]

[0125] Based on the formula for the volume of a cylinder The calculated new diameter D1 = 2 × r1 ≈ 0.02165 m = 21.65 mm.

[0126] Based on the above calculations, the system obtains three key values: polysilicon yield (25.4676 kg), silicon rod diameter (21.65 mm), and equilibrium gas phase composition percentage after 1 hour of operation. Following this calculation logic, the silicon rod diameter after 1 hour is used as the starting diameter for the second hour. Combined with the feed table from the second hour, the data for the third hour is calculated, and so on, to calculate the total polysilicon yield and average silicon rod diameter within the operating cycle. Figure 10 The figure shown is a time curve of silicon rod diameter data.

[0127] 4. Risk warning of inverted rod

[0128] The calculation process is as follows: Simulation results show that the boundary layer inhomogeneity parameter is 54% at the first hour; the process control data calculation system has calculated the average silicon rod diameter to be 21.65 mm at the second hour using the above steps. The calculation system uses the following equation:

[0129] Dd=(D1-D0)×boundary layer inhomogeneity parameter×20%=(21.65-20)×54%×20%=0.1782mm.

[0130] The 20% is a correction factor that can be adjusted based on production stability. Provided no furnace collapse occurs, this value can be appropriately increased to improve silicon yield. The Dd value will be calculated hourly during the third step. This value has a warning threshold of 0.25mm. That is, when Dd ≥ 0.25mm, the difference in surface silicon rod diameter is too large, and the process control data calculation system will issue a warning. The process engineer needs to adjust the process parameters to prevent rod collapse.

[0131] 5. Calculation of heat exchange power and gas phase temperature

[0132] 1) Calculate structural parameters:

[0133] Channel area: Ac = δ1 × π × (D0 + δ1) = π × 0.218 × (0.02 + 0.218) = 0.1629 m² 2 Where, D0 = silicon rod diameter = 0.02m, δ1 = silicon rod spacing - silicon rod diameter = 238 - 20 = 0.218m

[0134] Flow channel diameter: D h =2×δ1=0.436m

[0135] Rod heat exchange area: A=πD0L 单 ×15=π×0.02×2×15=1.885m 2 ; where L 单 =Height of a single silicon rod=2m.

[0136] 2) Calculate physical properties and heat transfer coefficient

[0137] Material: gas phase density ρ 气 =0.542kg / m 3 Prandtl number Pr = 0.68, surface gas velocity of silicon rod v = 1.5 m / s;

[0138] Mass flow rate: M=ρ 气 ×v×Ac×Cp=0.528×1.5×0.1629×1060=139.7; where Cp is the specific heat capacity.

[0139] Convection heat transfer coefficient: hc = Nu × k / Dh = (0.023 × Re) ^0.8 ×Pr ^0.4 ×0.050) / Dh;

[0140] Re=(ρ 气 ×v×D h ) / μ=(0.524×1.5×0.436) / 3.2×10 -5 =11260.

[0141] Therefore, hc = 3.9 W / (m²·K); where Nu is the Nusselt number, k is the thermal conductivity of the fluid = 0.050 W / (m·K), and μ is the dynamic viscosity = 3.2 × 10⁻⁶. -5 .

[0142] Radiative heat transfer coefficient: hr = σε s (T s 2 +T w 2 (T) s +T w = 38.51W / (m 2 ·k); where σ is the Stefan-Boltzmann constant = 5.67 × 10 -8 ε s For a surface emissivity of 0.75, T s For silicon rod temperature = 1300K, T w The inner wall surface temperature is 473.15K.

[0143] The overall heat transfer coefficient is: U = hc + hr = 3.84 + 38.51 = 42.41 W / (m²). 2 ·k)

[0144] 3) Solving for energy balance to determine the gas phase temperature T out

[0145] Energy equation: MCp(T) out -T in )=UA∆T lm Tin =463.15K, logarithmic mean temperature difference ∆Tlm=(T out -T in ) / ln(T s -T in T s -T out )

[0146] The system first assumes T out Using 873.15K as the initial value, a stable T was calculated through multiple iterations. out =828.15K. Where U is the overall heat transfer coefficient; T in Feed temperature; M is mass flow rate; C p is the specific heat capacity; A is the heat transfer area of ​​the rod.

[0147] 4) Power and unit power consumption

[0148] Gas heating power Q = 27 × M × C p ×(T out -T in =27 × 50990 W = 1376.73 kW;

[0149] Radiative heat dissipation power P

[0150]

[0151] Total power P 总 =Q + P + (Q + P) × 1.5% = 2985.77kW

[0152] Unit power consumption = Total power / Silicon yield = 2985.77 / 25.4676 = 117.23 kWh / kg

[0153] Where σ is the Stefan-Boltzmann constant; A S ε is the surface area of ​​the silicon rod; s ε is the emissivity of the silicon rod surface; w A represents the emissivity of the inner surface of the furnace wall. w This represents the area of ​​the inner surface of the furnace wall.

[0154] 5) Current and voltage calculations

[0155] The resistivity of polycrystalline materials varies with temperature, and therefore can be determined based on the total power P. 总 =UI=I 2 R=I 2 ρL 单 A 截 The current and voltage values ​​required to maintain the corresponding power are calculated. Where ρ is the resistivity of the silicon rod; L... 单 It is the length of a single silicon rod; A 截It refers to the cross-sectional area of ​​a single silicon rod. For example... Figure 11 The image shows a time curve of the current data.

[0156] 6) Data integration, processing, and output

[0157] Based on the above calculation logic, the process control data calculation system obtains the following data and outputs it:

[0158] Through iterative calculations, the table will sequentially output the changes in core process data within the 1-X running time cycle. Engineers will optimize and adjust the material list based on the results. After the adjustment is satisfactory, clicking "confirm" will allow the system to precisely control the current and feed rate via the PLC server. The high-precision mass flow controller receives the flow setpoint signal from the process optimization layer and forms a closed-loop control with a control deviation of less than ±0.5%. The multi-loop intelligent power regulator receives the current setpoint signal from the process optimization layer and forms a closed-loop control with a control accuracy better than ±0.5%.

[0159] During operation, the real-time feedback system monitors data such as flow rate, current, and temperature in real time, feeding this data back to the process control data calculation system for real-time analysis. When data deviations occur, timely calibration is performed. The real-time feedback system also uses infrared thermography and measuring devices to measure the surface temperature and diameter of the silicon rod in real time. Figure 12 The image shows the interface of the silicon rod temperature infrared thermometer. The process control data calculation system compares and analyzes the silicon rod temperature and diameter data with the pre-running plan data. If any deviation occurs, it will be calibrated in a timely manner.

[0160] This application also provides an adaptive control system for a polysilicon reduction furnace, including a simulation modeling and database module, a process optimization and prediction module, and a real-time control and feedback module. The simulation modeling and database module is used to establish the offline multiphysics simulation model, fit and store the mapping relationship model. The process optimization and prediction module is used to call the model, perform operating condition calibration and iterative calculations, and generate the process control data table. The real-time control and feedback module is used to execute the process control data table and connect to sensors and actuators to collect the actual operating data and complete the dynamic adjustment.

[0161] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the adaptive control method for a polysilicon reduction furnace provided in any embodiment of this application.

[0162] In other words, when the program is executed by the processor, it can: establish an offline multiphysics simulation model based on the physical structure of the target reduction furnace; run simulation calculations under different preset process conditions to fit a mapping relationship model characterizing the correspondence between silicon rod diameter / surface area, furnace flow field conditions and mass transfer conditions, and store it in a database; call the corresponding mapping relationship model from the database based on production plan parameters, and perform operating condition calibration on the mapping relationship model; starting from the initial silicon rod state and process setting values, perform iterative calculations in combination with the calibrated mapping relationship model to generate a process control data table that changes over time; control the operation of the reduction furnace according to the process control data table, and collect actual operating data in real time; compare the actual operating data with the corresponding predicted values ​​in the process control data table to dynamically adjust the process data.

[0163] Any combination of one or more computer-readable media may be used. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium that contains or stores a program that may be used by or in connection with an instruction execution system, apparatus, or device.

[0164] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0165] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the operator's computer, partially on the operator's computer, as a standalone software package, partially on the operator's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the operator's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider). The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably.

[0166] The above are merely preferred embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive control method for a polycrystalline silicon reduction furnace, characterized in that, Includes the following steps: Based on the physical structure of the target reduction furnace, an offline multiphysics simulation model was established; and simulation calculations were performed under different preset process conditions to obtain a mapping relationship model that characterizes the relationship between silicon rod diameter / surface area, furnace flow field conditions and mass transfer conditions, and stored in the database. Based on the production plan parameters, the corresponding mapping relationship model is retrieved from the database, and the mapping relationship model is calibrated for operating conditions. Starting from the initial silicon rod state and process settings, iterative calculations are performed using a calibrated mapping model to generate a process control data table that changes over time. The reduction furnace is controlled according to the process control data sheet, and actual operating data is collected in real time. The actual operating data is compared with the corresponding predicted values ​​in the process control data table to dynamically adjust the process data.

2. The adaptive control method for a polycrystalline silicon reduction furnace according to claim 1, characterized in that, The mapping model is a model that characterizes the correspondence between the diameter / surface area of ​​the silicon rod and the surface gas velocity and boundary layer deviation coefficient; wherein, the surface gas velocity is defined as the average airflow velocity at a specific location on the surface of the silicon rod, and the boundary layer deviation coefficient is used to characterize the degree of non-uniformity of the actual flow field distribution.

3. The adaptive control method for a polycrystalline silicon reduction furnace according to claim 2, characterized in that, The step of calibrating the mapping relationship model under operating conditions includes: Based on the feed quantity, proportion, and temperature parameters in the feed schedule, the mapping relationship model is adjusted using a preset calibration formula; wherein, the calibration formula includes: Surface gas velocity data calibration formula: ; in: The surface gas velocity after calibration, in m / s; The reference surface velocity is calculated by simulation software, and the unit is m / s. Total molar amount of reference material, unit: mol; This represents the change in the molar amount of hydrogen, in mol. This represents the change in the molar amount of trichlorosilane, in mol. Feed temperature, unit: K; Boundary layer deviation coefficient calibration formula: ; in: To calibrate the boundary layer deviation coefficient; The baseline boundary layer deviation coefficient is calculated by simulation software. R represents the feed rate of trichlorosilane, in kg; R is the molar ratio of hydrogen.

4. The adaptive control method for a polycrystalline silicon reduction furnace according to claim 2, characterized in that, The step of generating a time-varying process control data table includes: Based on the silicon rod diameter at the current time point, the corresponding surface gas velocity and boundary layer deviation coefficient are obtained from the calibrated mapping relationship model; Based on the process setting temperature and feed ratio, and combined with the obtained surface gas velocity and boundary layer deviation coefficient, the silicon deposition amount is calculated through the surface chemical reaction equilibrium model, and the silicon rod diameter for the next time node is updated accordingly.

5. The adaptive control method for a polycrystalline silicon reduction furnace according to claim 4, characterized in that, The step of calculating the amount of silicon deposition using a surface chemical reaction equilibrium model includes: Based on the temperature and pressure conditions set in the process, the equilibrium equations for multiple independent chemical reactions, including the reaction between trichlorosilane and hydrogen, are solved to obtain the concentrations of each gas phase component under equilibrium conditions and the theoretical maximum silicon deposition rate (Si). max； Based on the current silicon rod diameter, obtain the preset deposition efficiency coefficient δ, and calculate the actual silicon deposition amount Si = δ × Si max .

6. The adaptive control method for a polycrystalline silicon reduction furnace according to claim 4, characterized in that, The generation of the time-varying process control data table also includes: After the iterative calculation at each time point is completed, the diameter difference Dd, which characterizes the uniformity of silicon rod growth, is calculated based on the boundary layer deviation coefficient and the current silicon rod diameter increment. When the diameter difference Dd exceeds a preset threshold, a risk warning signal for the inverted rod is generated and output.

7. The adaptive control method for a polycrystalline silicon reduction furnace according to claim 4, characterized in that, The generation of the time-varying process control data table also includes: For each time point, based on the reduction furnace parameters, silicon rod parameters, and fluid material parameters of that time point, the convective heat transfer coefficient and the radiative heat transfer coefficient are calculated, and the overall heat transfer coefficient is determined. Based on the overall heat transfer coefficient, the corresponding gas phase outlet temperature is determined by iteratively solving the energy balance equation. Based on the gas phase outlet temperature, the gas heating power and radiative heat dissipation power are calculated, summed, and losses are included in a preset proportion to obtain the total operating power P at that time point. 总 and unit power consumption.

8. The adaptive control method for a polycrystalline silicon reduction furnace according to claim 7, characterized in that, The generation of the time-varying process control data table also includes: Based on the total operating power P 总 The required current and voltage settings for this time node are calculated based on the electric power formula, along with the resistivity ρ, length L, and cross-sectional area A of the silicon rod at the corresponding temperature.

9. An adaptive control system for a polycrystalline silicon reduction furnace, characterized in that, The system for implementing the method of any one of claims 1 to 8 comprises: The simulation modeling and database module is used to establish the offline multiphysics simulation model, fit and store the mapping relationship model; The process optimization and prediction module is used to call the model, perform operating condition calibration and iterative calculation, and generate the process control data table. The real-time control and feedback module is used to execute the process control data table and is connected to the sensors and actuators to collect the actual operating data and complete the dynamic adjustment.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the adaptive control method for polysilicon reduction furnace as described in any one of claims 1 to 8.

Images