An ultrahigh-purity nanometer graphite impurity migration prediction control method and system

Abstract

The application discloses an ultra-high-purity nanometer graphite impurity migration prediction control method and system, and relates to the technical field of ultra-high-purity nanometer graphite preparation. In the purification process, the furnace temperature, temperature change rate, furnace cavity air pressure, heating power and tail gas discharge flow are synchronously collected; the impurity migration driving force and the impurity residual coefficient are calculated according to a micro area, and the impurity release potential field is obtained by multiplication; the potential field is continuously updated and evolution characteristics are extracted, and the prediction model is input to obtain a predicted potential field; the migratable impurity reserves are calculated in combination with the impurity residual coefficient, the purification yield function is obtained by comparing the system energy consumption, and the purification temperature, the furnace cavity air pressure and the heating power are adjusted in stages according to the purification yield function, so that the purification endpoint is adaptively judged; and the driving force weight and the prediction model are iteratively updated according to the detection error after the purification is completed. The application has the foresight prediction and adaptive control ability of the impurity migration trend in the purification process, and improves the product purity stability and the purification efficiency.

Description

Technical Field

[0001] This application relates to the field of ultra-high purity nano-graphite preparation technology, specifically to a method and system for predicting and controlling the migration of impurities in ultra-high purity nano-graphite. Background Technology

[0002] Ultra-high purity nano-graphite is a crucial raw material for preparing high-purity silicon carbide powder, silicon carbide single crystals, third-generation semiconductor materials, and high-end electronic packaging materials. Its purity level directly affects the electrical, thermal, and reliability properties of subsequent products. With the development of the third-generation semiconductor industry, the purity requirements for nano-graphite are constantly increasing. Currently, industrial applications typically require a purity of 5N or higher, and some high-end applications even require 6N or higher. Existing processes for preparing ultra-high purity nano-graphite usually employ high-temperature vacuum purification techniques. By controlling process parameters such as heating temperature, holding time, and vacuum level, metallic and non-metallic impurities are volatilized, diffused, or migrated, thereby achieving impurity removal.

[0003] However, in actual purification processes, impurity removal is not a simple linear relationship with temperature or vacuum level. It is influenced by multiple factors, including impurity type, particle size, crystal structure, and current impurity distribution. Therefore, the impurity migration process exhibits significant dynamic characteristics. Currently, industrial control systems typically employ preset process curve control, executing the purification process according to a fixed heating curve, fixed holding time, and fixed vacuum conditions. When the process environment fluctuates or raw material batches change, the control system cannot obtain real-time information about the impurity migration status and can only continue operating based on preset parameters. This lack of dynamic awareness of impurity migration trends leads to deviations between control parameters and actual impurity removal requirements, resulting in insufficient or excessive purification, which affects product purity stability and production efficiency. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for predicting and controlling the migration of impurities in ultra-high purity nano-graphite, so as to solve the problems mentioned in the background art.

[0005] In a first aspect, one embodiment of this application provides a method for predicting and controlling impurity migration in ultra-high purity nano-graphite. The method includes: during the purification process in a high-temperature vacuum purification furnace for ultra-high purity nano-graphite, synchronously collecting furnace temperature, temperature change rate, furnace chamber pressure, heating power, and exhaust gas flow rate according to a preset sampling period; dividing the effective heating area of ​​the purification furnace into several micro-regions; for each micro-region, calculating the impurity migration driving force based on temperature, temperature change rate, furnace chamber pressure, and the ratio of exhaust gas flow rate to heating power; calculating the impurity residue coefficient for each micro-region based on historical batch purification data and current process parameters, whereby the impurity residue coefficient characterizes the concentration level of migratable impurities still present in the current micro-region; multiplying the impurity migration driving force of each micro-region by the corresponding impurity residue coefficient to obtain the impurity release potential energy of each micro-region; and constructing an impurity release potential field from the impurity release potential energy of all micro-regions; continuously updating the impurity release potential field according to the sampling period during the purification process, retaining a potential field time series for several consecutive moments, and analyzing the potential field... Potential field evolution features are extracted from the time series data. The potential field time series and evolution features are input into the impurity migration prediction model to predict the predicted potential field for the next sampling period. Based on the predicted potential field and the impurity residue coefficients of each micro-region, the amount of migratable impurities is calculated. The amount of migratable impurities is compared with the system energy consumption in the current sampling period to obtain the purification benefit function. The purification benefit function is compared with a preset first benefit threshold and a second benefit threshold. When the purification benefit function is greater than the first benefit threshold, the purification temperature is increased and the furnace pressure is decreased. When the purification benefit function is between the first and second benefit thresholds, the current process parameters are maintained. When the purification benefit function is lower than the second benefit threshold, a cooling program is initiated to end the purification process. After purification, the impurity content of this batch of products is tested. Based on the prediction error between the test results and the predicted value corresponding to the predicted potential field, the weights of the impurity migration driving force and the impurity migration prediction model are iteratively updated. The data for this batch is stored in a historical database for use in subsequent batches.

[0006] In conjunction with the first aspect, in some implementations of the first aspect, the effective heating area of ​​the purification furnace is divided into several micro-regions. For each micro-region, the impurity migration driving force is calculated based on temperature, temperature change rate, furnace chamber pressure, and the ratio of exhaust gas flow rate to heating power. This includes: dividing the effective heating area into several micro-regions using the purification furnace axis as the main dividing dimension, combined with the distribution of heating elements and temperature gradient characteristics within the furnace, with each micro-region corresponding to a temperature measurement position; for each micro-region, the impurity migration driving force is obtained by adding the product of temperature and temperature weight coefficient, the product of temperature change rate and change rate weight coefficient, the product of vacuum driving factor obtained by mapping furnace chamber pressure and vacuum weight coefficient, and the product of energy release factor obtained by mapping the ratio of exhaust gas flow rate to heating power and energy weight coefficient; the initial values ​​of temperature weight coefficient, change rate weight coefficient, vacuum weight coefficient, and energy weight coefficient are obtained by regression from historical batch purification data and are continuously corrected during iterative updates.

[0007] In conjunction with the first aspect, in certain implementations of the first aspect, based on historical batch purification data and current process parameters, the impurity residue coefficient of each micro-region is calculated. The impurity residue coefficient characterizes the concentration level of migratable impurities still present in the current micro-region. This includes: retrieving historical batches with specifications similar to the current batch of raw materials from the historical database, interpolating the estimated a priori residue value of each micro-region based on the current purification time; calculating the process deviation correction amount based on the deviation between the current process parameters and the corresponding process parameters of the historical batch, and using the process deviation correction amount to correct the estimated a priori residue value to obtain the corrected residue value; using the corrected residue value as the impurity residue coefficient of each micro-region, the impurity residue coefficient is normalized based on the initial impurity content of the corresponding micro-region of the current batch of raw materials, and the value range is 0 to 1; wherein, an impurity residue coefficient of 0 indicates that the impurity content of the micro-region has dropped below the preset detection limit, and an impurity residue coefficient of 1 indicates that the impurity content of the micro-region is equal to the initial impurity content of the corresponding micro-region of the current batch of raw materials.

[0008] In conjunction with the first aspect, in some implementations of the first aspect, the calculation of the impurity residual coefficient of each microregion based on historical batch purification data and current process parameters further includes: if there is measured data on the content of intermediate process impurities during the purification process, the corrected residual estimate is further corrected based on the measured data to obtain the measured corrected value; the prior residual estimate, the corrected residual estimate, and the measured corrected value are weighted and fused according to their respective confidence levels to obtain the final impurity residual coefficient of each microregion; when there is no measured data on intermediate processes, only the prior residual estimate and the corrected residual estimate are weighted and fused to obtain the final impurity residual coefficient of each microregion.

[0009] In conjunction with the first aspect, in certain implementations of the first aspect, the impurity release potential field is continuously updated according to the sampling period during the purification process, and a potential field time series of several consecutive moments is retained. Potential field evolution features are extracted from the potential field time series. The potential field time series and potential field evolution features are input into the impurity migration prediction model to predict the predicted potential field for the next sampling period. This includes: calculating the potential field change between adjacent moments in the potential field time series to obtain a potential field change series; extracting the potential field decay rate, potential field fluctuation amplitude, and potential field convergence trend from the potential field change series to form potential field evolution features; wherein, the potential field decay rate is the average decrease rate of the potential field change within the time window corresponding to the potential field time series, the potential field fluctuation amplitude is the standard deviation of the potential field change within the time window, and the potential field convergence trend is the slope obtained by linearly fitting the potential field mean sequence within the time window; and inputting the potential field change series and potential field evolution features together into the impurity migration prediction model to predict the predicted potential field for each micro-region in the next sampling period.

[0010] In conjunction with the first aspect, in some implementations of the first aspect, the amount of movable impurities is calculated based on the predicted potential field and the impurity residual coefficient of each micro-region, including: extracting the predicted potential energy corresponding to each micro-region from the predicted potential field, multiplying the predicted potential energy by the impurity residual coefficient of the corresponding micro-region at the current time to obtain the predicted amount of movable impurities for each micro-region; and summing the predicted amounts of movable impurities for all micro-regions to obtain the amount of movable impurities.

[0011] In conjunction with the first aspect, in some implementations of the first aspect, the purification benefit function is obtained by comparing the amount of migratory impurities with the system energy consumption in the current sampling period, including: calculating the system energy consumption in the current sampling period based on the heating power and the sampling period; dividing the amount of migratory impurities by the system energy consumption to obtain the purification benefit function; the purification benefit function generally decreases as purification progresses, and is used to adaptively determine the purification endpoint of each batch.

[0012] In conjunction with the first aspect, in some implementations of the first aspect, the effective heating area of ​​the purification furnace includes several segmented heating areas with independently adjustable heating power, each segmented heating area corresponding to several micro-regions; the purification benefit function is compared with a preset first benefit threshold and a second benefit threshold, including: the first benefit threshold being higher than the second benefit threshold; when the purification benefit function is greater than the first benefit threshold, the purification temperature is increased, the furnace chamber pressure is decreased, and several micro-regions with the highest predicted mobility are identified among the predicted mobility of each micro-region, and the heating power ratio of the segmented heating areas corresponding to these micro-regions is increased; when the purification benefit function is between the first benefit threshold and the second benefit threshold, the current temperature, furnace chamber pressure, and heating power are maintained unchanged; when the purification benefit function is lower than the second benefit threshold, the heating power is gradually reduced, a controlled cooling program is initiated, and the purification of this batch is terminated; the initial values ​​of the first benefit threshold and the second benefit threshold are respectively taken as the high quantile and low quantile values ​​of the historical batch purification benefit function curves, and are gradually optimized and updated with the accumulation of historical batches.

[0013] In conjunction with the first aspect, in certain implementations of the first aspect, impurity content detection is performed on the current batch of products. Based on the prediction error between the detection results and the corresponding predicted values ​​of the predicted potential field, the calculation weights of the impurity migration driving force and the impurity migration prediction model are iteratively updated. This includes: converting the predicted potential field into predicted values ​​for each impurity element according to a pre-established potential-concentration mapping relationship; comparing the impurity content detection results with the converted predicted values, calculating the prediction error for each impurity element, and summarizing them to obtain a comprehensive prediction error; iteratively updating the temperature weight coefficient, rate of change weight coefficient, vacuum weight coefficient, and energy weight coefficient based on the comprehensive prediction error, so that the temperature weight coefficient, rate of change weight coefficient, vacuum weight coefficient, and energy weight coefficient gradually converge to the actual laws of the current purification furnace and raw material system; using the potential field time series and impurity content detection results of the current batch as new training samples, when the comprehensive prediction error is lower than a preset error threshold, incremental training and updating of the impurity migration prediction model is performed; when the comprehensive prediction error is higher than the preset error threshold, the impurity migration prediction model is fully retrained using all valid batch data in the historical database.

[0014] Secondly, one embodiment of this application provides an impurity migration prediction and control system for ultra-high purity nano-graphite. The system includes: a data acquisition module for synchronously acquiring furnace temperature, temperature change rate, furnace chamber pressure, heating power, and exhaust gas flow rate during the purification process in an ultra-high purity nano-graphite high-temperature vacuum purification furnace, according to a preset sampling cycle; a driving force calculation module for dividing the effective heating area of ​​the purification furnace into several micro-regions, and for each micro-region, calculating the impurity migration driving force based on temperature, temperature change rate, furnace chamber pressure, and the ratio of exhaust gas flow rate to heating power; a residual coefficient calculation module for calculating the impurity residual coefficient of each micro-region based on historical batch purification data and current process parameters, whereby the impurity residual coefficient characterizes the concentration level of migratable impurities still existing in the current micro-region; a potential field construction module for multiplying the impurity migration driving force of each micro-region by the corresponding impurity residual coefficient to obtain the impurity release potential energy of each micro-region, and constructing an impurity release potential field from all micro-regions; and a potential field prediction module for continuously updating the impurity release potential field during the purification process according to the sampling cycle, and retaining a potential field time series for several consecutive moments. The system employs several mechanisms: First, it extracts potential field evolution features from the potential field time series and inputs these features into an impurity migration prediction model to predict the potential field for the next sampling period. Second, it calculates the amount of migratable impurities based on the predicted potential field and the impurity residue coefficients of each micro-region. Third, it calculates the purification benefit function by comparing the amount of migratable impurities with the system energy consumption in the current sampling period. Fourth, it compares the purification benefit function with preset first and second benefit thresholds. When the purification benefit function is greater than the first benefit threshold, it increases the purification temperature and decreases the furnace pressure. When the purification benefit function is between the first and second benefit thresholds, it maintains the current process parameters. When the purification benefit function is lower than the second benefit threshold, it initiates a cooling program to end the purification process. Fifth, it updates the model by detecting the impurity content of the batch after purification, iterating the weights of the impurity migration driving force and the impurity migration prediction model based on the prediction error between the detection results and the predicted value of the predicted potential field, and storing the batch data in a historical database for subsequent batches.

[0015] Compared with the prior art, the beneficial effects of this application are: 1. This application constructs an impurity release potential field that combines spatial distribution information with the physical meaning of impurity migration, and uses the potential field evolution characteristics to drive an impurity migration prediction model to make a forward prediction of the potential field for future sampling periods. This enables the purification control system to perceive the evolution trend of impurity migration before significant changes occur, overcoming the shortcomings of existing technologies that perform purification according to fixed heating curves, fixed holding times, and fixed vacuum conditions, and cannot obtain the impurity migration status in real time. It avoids the phenomenon of insufficient or over-purification caused by the mismatch between control parameters and actual impurity removal requirements, and improves the stability of product purity.

[0016] 2. This application constructs a purification benefit function based on the ratio of the amount of migratable impurities to the system energy consumption, and adaptively determines the purification endpoint for each batch based on this function. Compared with the existing technology that determines the purification endpoint by using a fixed time or a fixed process curve, this application can adaptively determine the optimal purification endpoint for each batch based on the differences in the initial impurity content of different batches of raw materials, avoiding energy waste or substandard purity caused by uniform time parameters. At the same time, combined with the impurity content detection results of each batch of products, the calculation weight of the impurity migration driving force and the impurity migration prediction model are iteratively updated, so that the control system gradually converges to the actual laws of the current purification furnace and raw material system, thereby improving the production efficiency and energy efficiency of the purification process. Attached Figure Description

[0017] Figure 1 A schematic flowchart of a method for predicting and controlling the migration of impurities in ultra-high purity nano-graphite provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of an ultra-high purity nano-graphite impurity migration prediction and control system provided in an embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] Figure 1 This is a schematic flowchart illustrating a method for predicting and controlling impurity migration in ultra-high purity nano-graphite provided in an embodiment of this application. Figure 1 As shown, an embodiment of this application provides a method for predicting and controlling the migration of impurities in ultra-high purity nano-graphite, which includes the following steps: Step 101: During the purification process of the ultra-high purity nano-graphite high-temperature vacuum purification furnace, the furnace temperature, temperature change rate, furnace chamber pressure, heating power, and exhaust gas flow rate are collected synchronously according to the preset sampling cycle.

[0020] It should be understood that temperature is the real-time thermodynamic state parameter of the corresponding temperature measurement position in the furnace at the current moment, and it is the basic condition for driving the dissociation of impurity lattice and surface desorption.

[0021] The furnace chamber pressure is the real-time pressure reading inside the purification furnace chamber, reflecting the external driving conditions for impurity volatilization.

[0022] The exhaust gas flow rate is the gas flow rate at the exhaust gas outlet of the purification furnace, reflecting the rate at which impurity volatilization products are discharged with the gas per unit time.

[0023] Step 102: Divide the effective heating area of ​​the purification furnace into several micro-zones. For each micro-zone, calculate the impurity migration driving force based on temperature, temperature change rate, furnace chamber pressure, and the ratio of tail gas emission flow rate to heating power.

[0024] It should be understood that the effective heating area is the actual space within the purification furnace that participates in the heating and purification process.

[0025] A micro-region is the smallest computational unit obtained by dividing the effective heating area according to its spatial location. Each micro-region corresponds to a local impurity migration state.

[0026] Impurity migration driving force is a quantitative indicator that characterizes the overall impurity migration capability of the corresponding micro-region under the current operating conditions. The higher the value, the stronger the current impurity migration driving capability of the micro-region.

[0027] Step 103: Based on historical batch purification data and current process parameters, calculate the impurity residue coefficient for each microregion. The impurity residue coefficient represents the concentration level of migratable impurities still present in the current microregion.

[0028] It should be understood that historical batch purification data refers to the process parameters and test results recorded for batches that have already been purified, and is used as a reference for estimating the status of the current batch.

[0029] The current process parameters are the process data such as temperature and furnace pressure collected in real time during the purification process of this batch.

[0030] Step 104: Multiply the impurity migration driving force of each micro-region by the corresponding impurity residual coefficient to obtain the impurity release potential energy of each micro-region, and combine the impurity release potential energy of all micro-regions to form the impurity release potential field.

[0031] It should be understood that the impurity release potential field is the overall state quantity composed of the spatial distribution of the impurity release potential energy of all micro-regions, used to characterize the overall distribution characteristics of the impurity migration state in the purification furnace.

[0032] Step 105: During the purification process, the impurity release potential field is continuously updated according to the sampling cycle. The potential field time series is retained for a continuous number of moments. The potential field evolution features are extracted from the potential field time series. The potential field time series and potential field evolution features are input into the impurity migration prediction model to predict the predicted potential field for the next sampling cycle.

[0033] It should be understood that the potential field time series is time-series data composed of several historical moments of impurity release potential fields recorded continuously according to the sampling period.

[0034] Potential field evolution characteristics are feature quantities extracted from the potential field time series to characterize the trend and stability of potential field changes.

[0035] The impurity migration prediction model is a computational model based on historical potential field data, used to predict the potential field for impurity release at future moments.

[0036] The predicted potential field is a state quantity composed of the predicted values ​​of the potential energy released by impurities in each micro-region in the next sampling period, as output by the impurity migration prediction model.

[0037] Step 106: Calculate the amount of movable impurities based on the predicted potential field and the impurity residual coefficient of each micro-region.

[0038] It should be understood that the migratable impurity reserve is an indicator that characterizes the total amount of effective impurity migration expected to occur in the purification region during the next sampling period, calculated based on the predicted potential field and the impurity residual coefficient.

[0039] Step 107: Compare the amount of mobile impurities with the system energy consumption during the current sampling period to obtain the purification benefit function.

[0040] It should be understood that the system energy consumption refers to the actual energy consumed by the purification furnace during the current sampling cycle.

[0041] The purification benefit function is the ratio of the amount of mobile impurities to the system energy consumption, which is used to characterize the expected impurity removal benefit per unit of energy input.

[0042] Step 108: Compare the purification benefit function with a preset first benefit threshold and a second benefit threshold. When the purification benefit function is greater than the first benefit threshold, increase the purification temperature and decrease the furnace chamber pressure. When the purification benefit function is between the first and second benefit thresholds, maintain the current process parameters. When the purification benefit function is lower than the second benefit threshold, initiate a cooling program to end the purification process.

[0043] It should be understood that the first benefit threshold is a pre-set critical value used to determine that the purification benefit is at a high level.

[0044] The second benefit threshold is a pre-set critical value that is lower than the first benefit threshold and is used to determine that the purification benefit is at a low level.

[0045] Step 109: After purification, the impurity content of this batch of products is tested. Based on the prediction error between the test results and the predicted value corresponding to the predicted potential field, the weight of the impurity migration driving force and the impurity migration prediction model are iteratively updated, and the data of this batch is stored in the historical database for use in subsequent batches.

[0046] It should be understood that the impurity content detection refers to the determination of the impurity element content of this batch of products after purification.

[0047] The historical database is a data storage unit that stores historical batch process parameters, test results, and model parameters for subsequent batches to query and access.

[0048] This embodiment constructs a physically meaningful impurity release potential field by coupling the impurity migration driving force with the impurity residue coefficient, and makes a forward-looking prediction of the future impurity migration trend based on the evolution characteristics of the potential field. Then, based on the purification benefit function, it adaptively adjusts the purification temperature, furnace pressure and heating power and automatically determines the purification endpoint. At the same time, it iteratively learns the driving force calculation weight and impurity migration prediction model by combining the test results of each batch. This overcomes the problem of insufficient or over-purification caused by the purification of existing technologies according to fixed process curves and the difficulty in dynamically sensing the impurity migration state. It improves the stability of the purity of ultra-high purity nano-graphite products and the production efficiency and energy efficiency of the purification process.

[0049] Furthermore, using the furnace axis as the primary dividing dimension, and considering the distribution of heating elements and temperature gradient characteristics within the furnace, the effective heating area is divided into several segments along the axis. If the radial temperature difference of the furnace exceeds a preset threshold, preferably between 2℃ and 5℃ (e.g., 3℃), it indicates that the unevenness of radial temperature distribution within the same axial segment is not negligible. Radial stratification is then introduced to improve the spatial resolution of the micro-region division. Therefore, radial stratification is introduced simultaneously with the axial segmentation, forming a two-dimensional micro-region grid along the axial diameter. The effective heating area is divided into several micro-regions, each corresponding to a temperature measurement position. The spatial range of each micro-region extends from the corresponding temperature measurement position towards adjacent temperature measurement positions, with the boundary at the midpoint between the two adjacent positions. After the micro-region division is completed, a micro-region spatial index table is established, recording the micro-region number, spatial coordinate range, corresponding temperature measurement position number, and the initial impurity distribution estimate determined by raw material batch information and historical batch purification data. The micro-region division remains fixed after completion and is not adjusted with changes in purification batches.

[0050] For each micro-region, the impurity migration driving force is calculated using the following formula: ; in, For the first Micro-regions The driving force of impurity migration at any moment The temperature measurement location corresponding to this micro-area is at Temperature at any moment Temperature weighting coefficient; For this temperature measurement location at The rate of temperature change at any given time The rate of change weighting coefficient; The vacuum drive factor is obtained by mapping the furnace cavity gas pressure. This is the vacuum weighting coefficient. Calculate using the following formula: ; in, for The furnace chamber pressure at any given time, The preset reference pressure, The smallest positive number set to prevent the denominator from being zero. For the first The position correction coefficient corresponding to the micro-zone is obtained from the simulation results or experimental calibration data of the gas flow distribution in the furnace. It is used to correct the local pressure unevenness caused by the difference in the relative position between each micro-zone and the exhaust port, so that the vacuum driving factor is larger when the furnace pressure is lower, thereby reflecting the law of enhanced impurity volatilization driving force under low pressure conditions. The energy release factor is obtained by mapping the ratio of exhaust gas flow rate to heating power. Energy weighting coefficient, Calculate using the following formula: ; in, for Exhaust emission flow rate at any given time for Heating power at any time For the first The contribution weight of a micro-region to gas emissions is determined by the proportion of its volume to the total volume of the effective heating area and its relative position. The product of temperature and temperature weighting coefficient, the product of temperature change rate and change rate weighting coefficient, the product of vacuum drive factor and vacuum weighting coefficient, and the product of energy release factor and energy weighting coefficient are added together to obtain the first... Micro-regions Momentary impurity migration driving force .

[0051] The aforementioned linear weighted model of impurity migration driving force is a commonly used multi-factor comprehensive evaluation method in engineering. Temperature, temperature change rate, vacuum driving factor, and energy release factor respectively represent four independent physical driving mechanisms: thermodynamic desorption driving, dynamic excitation driving, gas phase mass transfer driving, and discharge rate driving. After normalization, the contribution of each component to the overall impurity migration capability can be reasonably approximated by linear superposition. This form facilitates subsequent regression based on historical data and online iteration to continuously correct the weight coefficients, balancing model accuracy and online update efficiency. Furthermore, the energy release factor is constructed based on the ratio of exhaust gas flow rate to heating power, which characterizes the discharge rate of impurity volatilization products corresponding to a unit heating energy. A larger ratio indicates higher impurity discharge efficiency per unit energy input. This relationship has been verified using historical batch data. Additionally, the contribution weight of the micro-region to gas emissions... The determination method is supplemented as follows: it is determined by the proportion of the volume of the micro-region to the total volume of the effective heating area, combined with the weighted factor of the relative position of the micro-region to the exhaust gas outlet.

[0052] The initial values ​​of the temperature weighting coefficient, rate of change weighting coefficient, vacuum weighting coefficient, and energy weighting coefficient were obtained by performing multiple linear regression on purification data from multiple historical batches in the historical database. Using the measured impurity removal rate at each moment of the historical batch as the regression target, and the temperature, temperature change rate, vacuum driving factor, and energy release factor at the corresponding moment as independent variables, the combination of weighting coefficients that minimizes the regression error was solved, and these were used as the initial values ​​of the four weighting coefficients. Since the contribution ratios of various factors to impurity migration differ under different purification furnace structures and raw material systems, the initial values ​​of the above four weighting coefficients have corresponding process physics meanings and are continuously corrected in iterative updates, gradually converging to values ​​that match the current purification furnace structure and raw material system.

[0053] Furthermore, historical batches with similar raw material specifications to the current batch are retrieved from the historical database. These specifications include particle size ranges and initial impurity content ranges. The decay curves of impurity concentration in each micro-region of similar historical batches under similar process conditions are statistically analyzed as a function of purification time. Based on the purification time already achieved by the current batch, the decay curves are interpolated to obtain the prior residual estimates for each micro-region. .

[0054] Based on the deviations between the current process parameters, including temperature and furnace pressure, and the corresponding process parameters of historical batches, the process deviation correction amount is calculated using the following formula. : ; in, , These are the temperature parameters and furnace pressure parameters corresponding to historical batches. , These are the current batch numbers. micro-area in The temperature and furnace pressure at any given time , These are the normalized ranges for temperature and furnace chamber pressure, respectively. To correct the intensity coefficient, this is used to adjust the degree of influence of process parameter deviations on the residual coefficient correction. When the actual temperature of the current batch is higher than the corresponding temperature of historical batches, the process deviation correction is negative, indicating that the impurity removal rate is faster than that of historical batches, and the prior residual estimate is lowered accordingly.

[0055] The sign of the process deviation correction is determined by the overall calculation result of the weighted sum of the temperature difference term and the furnace pressure difference term using the formula, and not solely by the sign of the temperature difference term. When the temperature deviation and pressure deviation are in the same direction, or when the temperature deviation term is significantly larger than the pressure deviation term, the sign of the correction tends to be consistent with the sign of the temperature difference term. That is, when the actual temperature of the current batch is higher than the corresponding temperature of the historical batch and the furnace pressure deviation is insufficient to reverse the overall sign, the correction is negative overall, indicating that the impurity removal rate is faster than that of the historical batch, and the prior residual estimate is lowered accordingly. Conversely, when the actual temperature of the current batch is lower than the corresponding temperature of the historical batch and the furnace pressure deviation is insufficient to reverse the overall sign, the correction is positive overall, and the prior residual estimate is raised accordingly. When the temperature deviation and pressure deviation are in opposite directions and their values ​​are equivalent, the actual calculation result of the formula should be used to determine the sign and magnitude of the correction, and it should not be judged solely by the sign of a single variable.

[0056] The corrected residual estimate is used as the impurity residual coefficient for each micro-region. The impurity residual coefficient is normalized based on the initial impurity content of the corresponding micro-region of this batch of raw materials, and the value range is 0 to 1. Among them, an impurity residual coefficient of 0 indicates that the impurity content of the micro-region has dropped below the preset detection limit, and an impurity residual coefficient of 1 indicates that the impurity content of the micro-region is equal to the initial impurity content of the corresponding micro-region of this batch of raw materials.

[0057] Furthermore, if there is measured data on the content of intermediate process impurities during the purification process, the measured data is used as a benchmark, and the corrected residual estimate is further corrected according to the following formula to obtain the measured corrected value. : ; in, This represents the impurity content measured in the most recent intermediate process. This represents the initial impurity content of this batch of raw materials. If the purification furnace is not equipped with intermediate sampling conditions or there is no measured data on the intermediate process impurity content during purification, this step should not be performed.

[0058] The prior residual estimate, the corrected residual estimate, and the measured correction value are weighted and fused according to their respective confidence levels to obtain the final impurity residual coefficient for each micro-region: ; in, , , The weighting coefficients are the prior residual estimate, the corrected residual estimate, and the measured correction value, respectively, and their sum is 1. The weighting coefficients are dynamically allocated based on the reliability of each source, which is determined by the amount of historical data and the availability of intermediate measured data. When intermediate measured data is unavailable, only the prior residual estimate and the corrected residual estimate are weighted and fused to obtain the final impurity residual coefficient for each micro-region. Take 0, and The sum is 1.

[0059] Furthermore, by continuously executing the steps of calculating the impurity migration driving force, calculating the impurity residual coefficient, and constructing the impurity release potential field according to the sampling period, a continuous... Potential field time series at each moment ,in The default historical window length is [value]. This means retaining the potential field history corresponding to the most recent 20 sampling cycles, which corresponds to a time window of about 10 minutes. This can be adjusted within the range of 10-60 according to the thermal response characteristics of the purification furnace.

[0060] Calculate the change in potential field between adjacent time points in the potential field time series to obtain the potential field change series: ; in, For the first A vector consisting of the change in the potential energy released by impurities in each micro-region within each sampling period; A negative value overall indicates that the potential field of each micro-region is in a state of decay, that is, impurities are being continuously removed; a positive value locally indicates that the potential field of the corresponding micro-region has an abnormal increase in that period, which may be caused by a sudden increase in local temperature or an error in the estimation of the impurity residual coefficient.

[0061] The potential field decay rate, potential field fluctuation amplitude, and potential field convergence trend are extracted from the potential field change sequence to form the potential field evolution characteristics. Potential field decay rate The average rate of decrease of the potential field change within the time window corresponding to the potential field time series is calculated using the following formula: ; in, The change in potential field The One portion, The total number of microregions is represented by the potential field decay rate, which is usually a large negative number in the early stage of purification, indicating that the potential field decays rapidly. In the late stage of purification, it gradually approaches 0, indicating that the potential field tends to stabilize.

[0062] Potential field fluctuation amplitude The standard deviation of the potential field change within the time window is calculated using the following formula: ; The amplitude of potential field fluctuations reflects the stability of the impurity migration process: a continuously large value indicates irregular changes in the potential field, which may indicate process disturbances such as vacuum fluctuations or local overheating, or abnormalities in raw material batches; a small value indicates that the purification process is progressing smoothly.

[0063] The convergence trend of the potential field is the slope obtained by linearly fitting the potential field mean sequence within the time window: First, calculate the mean of the potential field in each micro-region at each sampling time within the time window to obtain the potential field mean sequence: ; Then, perform linear least squares fitting on the mean sequence of the potential field, and take the slope of the fitted line as the convergence trend of the potential field: ; The convergence trend of the potential field reflects whether the potential field is still in a downward trend within the current window. ), tending to level ( Or will there be a rebound? (This is an abnormal situation), and its approach to 0 usually occurs earlier than the purification benefit function, which can provide an advance for subsequent control decisions.

[0064] The potential field change sequence, along with the potential field evolution characteristics consisting of the potential field decay rate, potential field fluctuation amplitude, and potential field convergence trend, are input into the impurity migration prediction model to predict the potential field of each micro-region in the next sampling period. ; in, The potential field evolution feature vector is composed of the potential field decay rate, the potential field fluctuation amplitude, and the potential field convergence trend. The predicted potential field is formed by the predicted potential energy of each micro-region in the next sampling period. This is an impurity migration prediction model. The impurity migration prediction model employs a lightweight time-series prediction structure combining vector autoregression and feature enhancement: the evolution of the impurity release potential field is considered as... 3D time series, with the most recent The potential field change sequence of each step is used as the autoregressive input, and the potential field evolution feature vector is used as the covariate input; this aims to achieve stable generalization under limited historical batch conditions and support subsequent online incremental updates. Specifically, the model switching trigger logic is as follows: a preset batch number threshold is set. When the number of historical batches accumulated in the historical database has not reached the preset batch number threshold, linear regression is used for modeling; when the number of historical batches reaches or exceeds the preset batch number threshold, the modeling method is switched to gradient boosting tree to further improve the nonlinear fitting ability as the sample size increases. The fusion method of the vector autoregression module and feature enhancement is as follows: the potential field evolution feature vector is used as the covariate feature, and it is concatenated with the potential field change sequence of the most recent few steps (autoregressive input feature) in the feature dimension to form an extended feature vector. This extended feature vector is then uniformly input into the currently used linear regression model or gradient boosting tree model for mapping to obtain the predicted potential field of each micro-region in the next sampling period.

[0065] Furthermore, the predicted potential energy corresponding to each micro-region is extracted from the predicted potential field. The predicted potential energy is multiplied by the impurity residue coefficient of the corresponding micro-region at the current moment to obtain the predicted mobility of each micro-region. Then, the predicted mobility of all micro-regions is summed to obtain the movable impurity reserves. ; in, To predict the predicted potential energy corresponding to the i-th micro-region in the potential field, For the first The impurity residue coefficient of each micro-region at the current moment.

[0066] The amount of migratable impurities generally decreases monotonically as purification progresses: in the early stages of purification, a large amount of impurities remain unremoved, resulting in a high amount of migratable impurities; towards the end of purification, the impurity residue coefficients in each micro-region approach zero, and the amount of migratable impurities also approaches zero. The rate of decrease reflects the speed of purification progress. In the summation structure of the migratable impurity amount, different micro-regions contribute differently to the predicted migratability. Micro-regions with higher predicted migratability represent the main areas for impurity removal in the current stage, allowing for targeted enhancement of process parameters in these regions for subsequent control decisions. Because the migratable impurity amount is calculated using the predicted potential field rather than the current potential field as input, it possesses a forward-looking nature, representing the amount of impurities expected to be removed in the next sampling period, rather than the amount of impurities already removed at the current moment, which aligns with the decision logic of the purification benefit function.

[0067] Furthermore, based on the heating power and sampling period, the system energy consumption within the current sampling period is calculated: ; in, for Heating power at any time The sampling period is This represents the system energy consumption during the current sampling period.

[0068] Dividing the amount of mobile impurities by the system energy consumption yields the purification benefit function. : ; The purification benefit function is measured as the ratio of the amount of migratory impurities to the unit of energy consumption. It characterizes the expected amount of migratory impurities corresponding to a unit of energy input, i.e., the amount of impurities expected to migrate in the next sampling period, rather than those already determined to be completely removed. It reflects the marginal benefit of continuing purification under current process conditions. The purification benefit function generally decreases as purification progresses, allowing for temporary fluctuations due to local process disturbances (such as localized unevenness of raw materials, instantaneous temperature or pressure fluctuations). A higher value indicates sufficient migratory impurities and high marginal benefit from continued purification, suggesting that process parameters should be maintained or strengthened. A lower value indicates a decrease in migratory impurities or a reduction in marginal benefit. A consistently low value indicates that the purification process is approaching saturation, and the actual benefit of continued purification is extremely low. The purification benefit function is used to adaptively determine the purification endpoint for each batch: Compared to determining the purification endpoint based on fixed holding time or a fixed process curve, the purification benefit function shifts the basis for endpoint determination from time measurement to benefit evaluation. It can adaptively determine the optimal purification endpoint for each batch based on the differences in initial impurity content of different batches of raw materials, avoiding under-purification or over-purification caused by uniform time parameters.

[0069] Furthermore, the first benefit threshold is higher than the second benefit threshold. When the purification benefit function is greater than the first benefit threshold, it is determined that there are still a large number of migratable impurities in the current purification area, the migration driving force is sufficient, and the purification efficiency is high. Therefore, the purification temperature is increased and the furnace chamber pressure is reduced: the target temperature is increased by a preset temperature increment. The temperature increment is adaptively determined based on the margin between the current actual temperature and the rated maximum temperature of the purification furnace, avoiding exceeding the upper limit of the furnace's rated temperature. The preset temperature increment is preferably 5℃–15℃, for example, 10℃. When the margin between the current actual temperature and the rated maximum temperature of the purification furnace is less than the preset temperature increment, the actual adjustment is based on a preset proportion of this margin (e.g., 80%), to avoid exceeding the upper limit of the furnace's rated temperature. The vacuum pump speed is increased to reduce the target value of the furnace chamber pressure, reduce the gas molecule density inside the furnace, and enhance the gas-phase mass transfer driving force for impurity volatilization. Meanwhile, several micro-regions with the highest predicted mobility are identified from the predicted mobility of each micro-region. The number of these micro-regions is taken as a preset proportion of the total number of micro-regions, such as one-quarter of the total number of micro-regions and rounded down. If the purification furnace supports segmented heating control, the heating power ratio of the heating areas corresponding to these micro-regions is increased to achieve spatial refinement and enhancement of impurity removal.

[0070] When the purification benefit function is between the first benefit threshold and the second benefit threshold, it is determined that the current purification process is in effect, the impurity migration rate is well matched with the process parameters, the current temperature, furnace pressure and heating power are kept unchanged, the current purification process continues to be executed, and the updated result of the purification benefit function is awaited in the next sampling cycle.

[0071] When the purification benefit function falls below the second benefit threshold, it is determined that a large amount of migratory impurities in the current purification area have been removed, and the benefit of continuing purification is extremely low. The heating power is gradually reduced according to the preset power reduction rate, and a controlled cooling program is started. The preset power reduction rate is preferably 5%-10% of the current heating power in each sampling cycle, gradually reducing it to the preset lower limit of the heating power (e.g., 10% of the rated heating power) before entering the heat preservation and cooling stage to avoid thermal stress damage to the furnace body and product caused by a sudden drop in furnace temperature, and the purification of this batch ends. At the same time, the excessive gas extraction of the vacuum pump is stopped, and the protective atmosphere filling mode is restored, such as filling high-purity argon to a slightly positive pressure state. The purification end time of this batch is recorded for use in subsequent detection feedback processes.

[0072] The initial values ​​of the first and second revenue thresholds are taken from the high and low percentile values ​​of the historical batch purification revenue function curves, respectively. For example, the 75th and 25th percentile values ​​of the historical batch purification revenue function curves can be taken as the initial thresholds. The first and second revenue thresholds are gradually optimized and updated as historical batch data is accumulated.

[0073] Further, the impurity content of this batch of products is tested according to the preset testing procedure to obtain the final measured content of each impurity element. The impurity content test results are compared with the predicted values ​​calculated from the predicted potential field. The predicted values ​​are calculated from the last valid predicted potential field through a pre-established potential-concentration mapping relationship, which is obtained by regression analysis of historical batch purification data. For each impurity element, the difference between the measured content and the predicted value is calculated as the prediction error for that element. ; in, Number the impurity elements, No. The final measured content of each impurity element For the first The predicted values ​​for each impurity element are calculated. The prediction errors for each impurity element are then summarized as follows to obtain the overall prediction error: ; in, The total number of impurity elements detected is represented by the overall prediction error, which reflects the overall relative error of the impurity migration prediction model in this batch.

[0074] Based on the comprehensive prediction error, the weighting coefficients for temperature, rate of change, vacuum, and energy are iteratively updated. The iterative update method is as follows: using the loss function corresponding to the comprehensive prediction error as the optimization objective, the values ​​of each weighting coefficient are gradually adjusted along the negative gradient direction of the weighting coefficients. ; in, For batch iteration rounds, Let be the loss function targeting the prediction error of impurity content. Update the learning rate for the weights. The gradient of the loss function with respect to the temperature weighting coefficient, rate of change weighting coefficient, vacuum weighting coefficient, and energy weighting coefficient is calculated by inversely expanding the chain propagation relationship between the calculation relationship of the impurity migration driving force and the calculation relationship of the impurity residue coefficient; the learning rate is... The initial value can be preset to 0.01, and gradually decreases as the number of batches participating in the iteration increases. The decay method can be to reduce the learning rate to 90% of the original value every 10 batches, with a lower limit of no less than 0.001, to avoid overfitting to a small number of historical batches of data. Through the above iterative updates, the temperature weight coefficient, the rate of change weight coefficient, the vacuum weight coefficient, and the energy weight coefficient gradually converge to the actual laws of the current purification furnace and raw material system, overcoming the dependence of the initial regression value on historical similar furnaces.

[0075] Using the potential field time series and impurity content detection results of this batch as new training samples, the impurity migration prediction model is updated as follows: The data of this batch is organized in the format of input-output pairs, with the potential field change of several consecutive sampling periods as input and the potential field of the corresponding next sampling period as output, and expanded into the historical training dataset; when the comprehensive prediction error is lower than the preset error threshold, the preset error threshold is preferably 5%-10%, for example, 8%; when the comprehensive prediction error is lower than this threshold, it indicates that the fitting degree of the impurity migration prediction model to the current purification furnace and raw material system is still within an acceptable range, and incremental training can meet the accuracy requirements; when the comprehensive prediction error is higher than this threshold, it indicates that the model deviates significantly, and full retraining is required to ensure prediction accuracy. Incremental learning is employed to incrementally train and update the impurity migration prediction model. This includes matrix updates for the linear regression model and the addition of new decision trees to the gradient boosting tree model, integrating the latest information from the current batch into the impurity migration prediction model without retraining all historical batch data. When the overall prediction error exceeds a preset error threshold, indicating a significant deviation between the impurity migration prediction model and the current purification furnace and raw material system, the model is fully retrained using all valid batch data from the historical database. After iterative updates, the updated temperature weighting coefficient, rate of change weighting coefficient, vacuum weighting coefficient, energy weighting coefficient, impurity migration prediction model parameters, and the detection results of the current batch are stored in the historical database for use in subsequent batches.

[0076] Figure 2 This is a schematic diagram of the structure of an ultra-high purity nano-graphite impurity migration prediction and control system provided in an embodiment of this application. Figure 2 As shown in the embodiment of this application, an ultra-high purity nano-graphite impurity migration prediction and control system is provided. The output of the data acquisition module 201 is connected to the input of the driving force calculation module 202 and the input of the residual coefficient calculation module 203, respectively. The temperature, temperature change rate, furnace pressure, and the ratio of exhaust gas flow rate to heating power synchronously acquired by the data acquisition module 201 are transmitted to the driving force calculation module 202 to calculate the impurity migration driving force of each micro-region. The temperature and furnace pressure synchronously acquired by the data acquisition module 201 are transmitted to the residual coefficient calculation module 203 as current process parameters. The residual coefficient calculation module 203 is also connected to a historical database to read historical batch purification data to calculate the prior residual estimate value of each micro-region and the corresponding process deviation correction amount.

[0077] The output terminals of the driving force calculation module 202 and the residual coefficient calculation module 203 are respectively connected to the input terminals of the potential field construction module 204. The impurity migration driving force of each micro-region calculated by the driving force calculation module 202 and the impurity residual coefficient of each micro-region calculated by the residual coefficient calculation module 203 are input together into the potential field construction module 204. The potential field construction module 204 multiplies the two to obtain the impurity release potential energy of each micro-region, and the impurity release potential energy of all micro-regions is used to form the impurity release potential field.

[0078] The output of the potential field construction module 204 is connected to the input of the potential field prediction module 205. The impurity release potential field continuously output by the potential field construction module 204 according to the sampling period is received by the potential field prediction module 205 and retained as a potential field time series for a continuous number of time periods. The potential field prediction module 205 extracts the potential field evolution features based on the potential field time series and inputs the potential field time series and potential field evolution features into the impurity migration prediction model to predict the predicted potential field for the next sampling period.

[0079] The output of the potential field prediction module 205 and the output of the residual coefficient calculation module 203 are respectively connected to the input of the reserve calculation module 206. The reserve calculation module 206 multiplies the predicted potential energy of each micro-region in the predicted potential field with the impurity residual coefficient of the corresponding micro-region at the current time and sums it over all micro-regions to obtain the relocatable impurity reserves.

[0080] The output of the storage calculation module 206 and the output of the data acquisition module 201 (heating power data) are respectively connected to the input of the profit function calculation module 207. The profit function calculation module 207 calculates the system energy consumption in the current sampling period based on the heating power and the sampling period, and compares the migratable impurity storage with the system energy consumption to obtain the purification profit function.

[0081] The output of the profit function calculation module 207 is connected to the input of the control decision module 208. The control decision module 208 compares the received purification profit function with the preset first profit threshold and second profit threshold, and generates corresponding control commands based on the comparison results. The output of the control decision module 208 is connected to the temperature control unit, vacuum control unit and heating power control unit of the purification furnace, respectively, to increase or maintain the purification temperature, decrease or maintain the furnace chamber pressure, and adjust the heating power ratio, thereby realizing the adjustment of the purification furnace process parameters.

[0082] The input terminals of the model update module 209 are connected to the output terminals of the impurity content detection device and the potential field prediction module 205, respectively, to obtain the impurity content detection results and the predicted values ​​corresponding to the predicted potential field of the current batch of products, and to calculate the prediction error accordingly. The output terminals of the model update module 209 are connected to the driving force calculation module 202 and the potential field prediction module 205, respectively, to feed back the calculated weights of the impurity migration driving force after iterative updates based on the prediction error to the driving force calculation module 202, and to feed back the updated impurity migration prediction model parameters to the potential field prediction module 205. The model update module 209 is also connected to the historical database to store the purification data of the current batch, the updated calculated weights, and the prediction model parameters into the historical database for use by the residual coefficient calculation module 203 and the potential field prediction module 205 during subsequent batch purification.

[0083] This application achieves a hierarchical connection and feedback between data acquisition, driving force calculation, residual coefficient calculation, potential field construction, potential field prediction, reserve calculation, profit function calculation, control decision, and model update modules. It couples the impurity migration driving force and the impurity residual coefficient into a physically meaningful impurity release potential field. Based on the potential field evolution characteristics, it enables forward-looking prediction of future impurity migration trends. Then, based on the purification profit function, it adaptively adjusts the purification temperature, furnace pressure, and heating power, and automatically determines the purification endpoint. Simultaneously, it iteratively learns the driving force calculation weights and impurity migration prediction model by combining the test results of each batch. This overcomes the problems of insufficient or over-purification caused by existing technologies that purify according to a fixed process curve and struggle to dynamically perceive the impurity migration state. It improves the stability of the purity of ultra-high purity nano-graphite products and the production efficiency and energy efficiency of the purification process.

[0084] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this application. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for predicting and controlling the migration of impurities in ultra-high purity nano-graphite, characterized in that, include: During the purification process of ultra-high purity nano-graphite high-temperature vacuum purification furnace, the furnace temperature, temperature change rate, furnace chamber pressure, heating power and exhaust gas flow rate are collected synchronously according to the preset sampling cycle. The effective heating area of ​​the purification furnace is divided into several micro-regions, each micro-region corresponding to a temperature measurement position. For each micro-region, the impurity migration driving force of each micro-region is obtained by adding the product of the temperature and the temperature weighting coefficient, the product of the temperature change rate and the change rate weighting coefficient, the product of the vacuum driving factor obtained by mapping the furnace cavity gas pressure and the vacuum weighting coefficient, and the product of the energy release factor obtained by mapping the ratio of the exhaust gas flow rate to the heating power and the energy weighting coefficient. Before adding, the temperature, the temperature change rate, the vacuum driving factor, and the energy release factor are normalized respectively. Based on historical batch purification data and current process parameters, the impurity residue coefficient of each microregion is calculated, including: retrieving historical batches with similar raw material specifications to the current batch from the historical database, interpolating the estimated prior residue value of each microregion based on the current purification time; calculating the process deviation correction amount based on the deviation between the current process parameters and the corresponding process parameters of the historical batch, and using the process deviation correction amount to correct the estimated prior residue value to obtain the corrected residue value; using the corrected residue value as the impurity residue coefficient of each microregion, where the impurity residue coefficient characterizes the concentration level of migratory impurities still present in the microregion, and normalizing it based on the initial impurity content of the microregion corresponding to the current batch of raw material, with a value range of 0 to 1. Multiply the impurity migration driving force of each micro-region by the corresponding impurity residual coefficient to obtain the impurity release potential energy of each micro-region, and combine the impurity release potential energy of all micro-regions to form the impurity release potential field. During the purification process, the impurity release potential field is continuously updated according to the sampling period, and the potential field time series at several consecutive moments is retained. The potential field evolution features are extracted from the potential field time series, and the potential field time series and the potential field evolution features are input into the impurity migration prediction model to predict the predicted potential field for the next sampling period. Based on the predicted potential field and the impurity residue coefficient of each micro-region, the amount of movable impurities is calculated, including: extracting the predicted potential energy corresponding to each micro-region from the predicted potential field, multiplying the predicted potential energy by the impurity residue coefficient of the corresponding micro-region at the current time to obtain the predicted amount of movable impurities for each micro-region; and summing the predicted amounts of movable impurities for all micro-regions to obtain the amount of movable impurities. The purification benefit function is obtained by comparing the amount of the migratable impurities with the system energy consumption during the current sampling period. The purification benefit function is compared with a preset first benefit threshold and a second benefit threshold. When the purification benefit function is greater than the first benefit threshold, the purification temperature is increased and the furnace pressure is decreased. When the purification benefit function is between the first benefit threshold and the second benefit threshold, the current process parameters are maintained. When the purification benefit function is lower than the second benefit threshold, a cooling program is started to end the purification process. After purification, the impurity content of the current batch of products is detected. Based on the prediction error between the detection result and the predicted value corresponding to the predicted potential field, the calculation weight of the impurity migration driving force and the impurity migration prediction model are iteratively updated, and the current batch data is stored in the historical database for use in subsequent batches.

2. The method according to claim 1, characterized in that, Using the axial direction of the purification furnace as the main dividing dimension, and combining the distribution of heating elements and temperature gradient characteristics inside the furnace, the effective heating area is divided into several micro-regions; The initial values ​​of the temperature weighting coefficient, rate of change weighting coefficient, vacuum weighting coefficient, and energy weighting coefficient are obtained by regression from historical batch purification data and are continuously corrected in the iterative update.

3. The method according to claim 1, characterized in that, The calculation of the impurity residue coefficient for each micro-region also includes: If there is measured data on the content of intermediate process impurities during the purification process, the corrected residual estimate is further corrected based on the measured data to obtain the measured corrected value. The prior residual estimate, the corrected residual estimate, and the measured correction value are weighted and fused according to their respective confidence levels to obtain the final impurity residual coefficient of each micro-region; when there is no intermediate process measured data, only the prior residual estimate and the corrected residual estimate are weighted and fused to obtain the final impurity residual coefficient of each micro-region.

4. The method according to claim 1, characterized in that, The process of continuously updating the impurity release potential field according to the sampling period during purification, retaining a potential field time series for a continuous number of moments, extracting potential field evolution features from the potential field time series, and inputting the potential field time series and the potential field evolution features into the impurity migration prediction model to predict the predicted potential field for the next sampling period includes: Calculate the change in potential field between adjacent moments in the potential field time series to obtain the potential field change sequence; The potential field decay rate, potential field fluctuation amplitude, and potential field convergence trend are extracted from the potential field change sequence to form potential field evolution characteristics; wherein, the potential field decay rate is the average rate of decrease of the potential field change within the time window corresponding to the potential field time series, the potential field fluctuation amplitude is the standard deviation of the potential field change within the time window, and the potential field convergence trend is the slope obtained by linearly fitting the potential field mean sequence within the time window. The potential field change sequence and the potential field evolution characteristics are input together into the impurity migration prediction model to predict the predicted potential field of each micro-region in the next sampling period.

5. The method according to claim 1, characterized in that, The step of comparing the amount of mobile impurities with the system energy consumption during the current sampling period to obtain the purification benefit function includes: Calculate the system energy consumption within the current sampling period based on the heating power and the sampling period; Dividing the amount of migratory impurities by the system energy consumption yields the purification benefit function; the purification benefit function generally decreases as purification progresses, and is used to adaptively determine the purification endpoint for each batch.

6. The method according to claim 1, characterized in that, The effective heating area of ​​the purification furnace includes several segmented heating areas with independently adjustable heating power, each segmented heating area corresponding one-to-one with the several micro-regions; the comparison of the purification benefit function with a preset first benefit threshold and a second benefit threshold includes: The first revenue threshold is higher than the second revenue threshold; When the purification benefit function is greater than the first benefit threshold, the purification temperature is increased, the furnace chamber pressure is reduced, and several micro-regions with the highest predicted mobility are identified in the predicted mobility of each micro-region. The heating power ratio of the segmented heating areas corresponding to the several micro-regions is increased. When the purification benefit function is between the first benefit threshold and the second benefit threshold, the current temperature, furnace chamber pressure and heating power are kept constant. When the purification benefit function is lower than the second benefit threshold, the heating power is gradually reduced, a controlled cooling program is started, and the purification of this batch is ended. The initial values ​​of the first revenue threshold and the second revenue threshold are taken as the high quantile and low quantile values ​​of the historical batch purification revenue function curve, respectively, and are gradually optimized and updated as historical batches accumulate.

7. The method according to claim 2, characterized in that, The step of detecting the impurity content of the current batch of products, and iteratively updating the calculation weight of the impurity migration driving force and the impurity migration prediction model based on the prediction error between the detection result and the predicted value corresponding to the predicted potential field, includes: According to the pre-established potential energy-concentration mapping relationship, the predicted potential field is converted to obtain the predicted value of each impurity element; the impurity content detection result is compared with the converted predicted value, the prediction error of each impurity element is calculated, and the comprehensive prediction error is obtained by summarizing. Based on the comprehensive prediction error, the temperature weight coefficient, rate of change weight coefficient, vacuum weight coefficient, and energy weight coefficient are iteratively updated so that the temperature weight coefficient, rate of change weight coefficient, vacuum weight coefficient, and energy weight coefficient gradually converge to the actual laws of the current purification furnace and raw material system. Using the potential field time series of the current batch and the impurity content detection results as new training samples, when the comprehensive prediction error is lower than the preset error threshold, the impurity migration prediction model is incrementally trained and updated; when the comprehensive prediction error is higher than the preset error threshold, the impurity migration prediction model is fully retrained using all valid batch data in the historical database.

8. A prediction and control system for impurity migration in ultra-high purity nano-graphite, characterized in that, include: The data acquisition module is used to synchronously collect furnace temperature, temperature change rate, furnace chamber pressure, heating power and exhaust gas flow rate according to a preset sampling cycle during the purification process of ultra-high purity nano-graphite high temperature vacuum purification furnace. The driving force calculation module is used to divide the effective heating area of ​​the purification furnace into several micro-regions. For each micro-region, the product of the temperature and the temperature weighting coefficient, the product of the temperature change rate and the change rate weighting coefficient, the product of the vacuum driving factor obtained by mapping the furnace cavity gas pressure and the vacuum weighting coefficient, and the product of the energy release factor obtained by mapping the ratio of the exhaust gas flow rate to the heating power and the energy weighting coefficient are added together to obtain the impurity migration driving force of each micro-region. Before the addition, the temperature, the temperature change rate, the vacuum driving factor and the energy release factor are normalized respectively. The residual coefficient calculation module is used to calculate the impurity residual coefficient of each micro-region based on historical batch purification data and current process parameters. This includes: retrieving historical batches with similar raw material specifications to the current batch from the historical database; interpolating the estimated prior residual value for each micro-region based on the current purification duration; calculating the process deviation correction amount based on the deviation between the current process parameters and the corresponding process parameters of the historical batch; correcting the estimated prior residual value using the process deviation correction amount to obtain the corrected residual estimate value, which serves as the impurity residual coefficient for each micro-region. The impurity residual coefficient characterizes the concentration level of migratory impurities still present in the micro-region, and is normalized based on the initial impurity content of the micro-region corresponding to the current batch of raw material, with a value range of 0 to 1. The potential field construction module is used to multiply the impurity migration driving force of each micro-region by the corresponding impurity residual coefficient to obtain the impurity release potential energy of each micro-region, and to form the impurity release potential field by combining the impurity release potential energy of all micro-regions. The potential field prediction module is used to continuously update the impurity release potential field according to the sampling period during the purification process, retain the potential field time series for a continuous number of moments, extract the potential field evolution features from the potential field time series, and input the potential field time series and the potential field evolution features into the impurity migration prediction model to predict the predicted potential field for the next sampling period. The reserve calculation module is used to calculate the movable impurity reserve based on the predicted potential field and the impurity residue coefficient of each micro-region, including: extracting the predicted potential energy corresponding to each micro-region from the predicted potential field, multiplying the predicted potential energy by the impurity residue coefficient of the corresponding micro-region at the current time to obtain the predicted movable amount of each micro-region; and summing the predicted movable amounts of all micro-regions to obtain the movable impurity reserve. The profit function calculation module is used to compare the amount of the migratable impurities with the system energy consumption in the current sampling period to obtain the purification profit function; The control decision module is used to compare the purification benefit function with a preset first benefit threshold and a second benefit threshold. When the purification benefit function is greater than the first benefit threshold, the purification temperature is increased and the furnace pressure is decreased. When the purification benefit function is between the first benefit threshold and the second benefit threshold, the current process parameters are maintained. When the purification benefit function is lower than the second benefit threshold, a cooling program is started to end the purification. The model update module is used to detect the impurity content of the current batch of products after purification. Based on the prediction error between the detection result and the predicted value corresponding to the predicted potential field, the module iteratively updates the calculation weight of the impurity migration driving force and the impurity migration prediction model, and stores the current batch data in the historical database for use in subsequent batches.

Images