A system and method for preparing a hard carbon negative electrode material

Abstract

The application discloses a preparation system and method of hard carbon negative electrode material, and relates to the technical field of material preparation and process control, and comprises a data acquisition and fusion module, temperature parameter data and pressure parameter data are collected from a hard carbon material preparation device in real time through a sensor network, an initial parameter interaction data set characterized by the interrelation of temperature parameters and pressure parameters is generated, and an arranged parameter mapping table is obtained; the preparation system and method of the hard carbon negative electrode material fully utilize historical data and real-time process data, form a sustainable correction temperature-pressure coupling control model, guarantee the reliability of material performance, take into account the improvement of production efficiency, and provide a technical scheme with high stability, strong adaptability and obvious industrial application value for fine and intelligent control of the hard carbon material preparation process.

Description

Technical Field

[0001] This invention relates to the field of materials preparation and process control technology, specifically to a preparation system and method for hard carbon anode materials. Background Technology

[0002] Hard carbon materials, as important functional materials, have wide applications in new energy storage, electrochemical devices, and related industrial fields. Their preparation process has a decisive influence on the material's microstructure, mechanical properties, and stability. In the industrial preparation of hard carbon materials, the formation of the material structure and the control of its properties typically require complex operating conditions such as high temperature and high pressure. Therefore, the stability and controllability of the preparation process parameters directly affect the final product quality and production efficiency.

[0003] In existing hard carbon material preparation processes, temperature and pressure are generally considered key process parameters affecting material properties. Temperature directly influences the structural evolution of hard carbon materials, such as changes in crystallinity and pore structure formation, while pressure significantly impacts the densification degree, structural stability, and internal defect formation at high temperatures. However, in actual production, temperature and pressure do not act independently but exhibit a clear interrelationship. When process temperature fluctuates, the effect of pressure on material forming and performance regulation also changes, revealing a complex coupling relationship between the two. Current preparation technologies often employ empirical settings or staged adjustments to control temperature and pressure, focusing on the stability of individual parameters while lacking systematic analysis and dynamic coordination of the interaction between temperature and pressure. This control method can meet basic requirements when production conditions are relatively stable, but in actual industrial environments, due to changes in equipment operating conditions, raw material differences, and external disturbances, temperature parameters inevitably fluctuate, leading to uncertainty in pressure control effectiveness and increased performance dispersion across different batches of hard carbon materials. Furthermore, as the application fields of hard carbon materials increasingly demand higher requirements for performance consistency and reliability, traditional process control methods that rely on manual experience or static parameter settings are insufficient to simultaneously optimize material performance and improve production efficiency. On the one hand, the lack of effective characterization of the pressure regulation response under temperature fluctuations results in a lag in process parameter adjustments; on the other hand, the large amount of historical and process data generated during production is not fully utilized, making it difficult to continuously correct and optimize the temperature-pressure coupling relationship, thus limiting the development of the preparation process towards refined and intelligent control. Summary of the Invention

[0004] The purpose of this invention is to provide a preparation system and method for hard carbon anode materials, thereby solving the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a preparation system for hard carbon anode material, comprising: The data acquisition and fusion module collects temperature and pressure parameter data in real time from the hard carbon material preparation equipment through a sensor network, generates an initial parameter interaction dataset characterized by the correlation between temperature and pressure parameters, and obtains a sorted parameter mapping table. The parameter correlation analysis module analyzes the interaction between temperature and pressure parameters based on the organized parameter mapping table, preprocesses fluctuating data, weights key parameters, and constructs a preliminary parameter correlation matrix that reflects the influence of temperature changes on pressure response. The temperature-pressure coupling regulation module uses temperature parameters as the main control disturbance parameters and pressure parameters as the controlled regulation parameters in the initial parameter correlation matrix. When the fluctuation of temperature parameters exceeds the preset threshold, the pressure parameters corresponding to the temperature parameters are iteratively adjusted through a genetic algorithm. The iterative convergence conditions and population initialization methods are set, and the adjusted pressure parameter set coupled with temperature changes is generated by combining the crossover mutation probability and fitness function design. The parameter optimization processing module obtains the adjusted pressure parameter set, performs secondary processing on the data range mapping and standardized transformation formula, and incorporates a missing value filling method to complete incomplete data, resulting in an optimized parameter pairing list for temperature and pressure dual-parameter coordinated control.

[0006] On the other hand, the present invention also provides a method for preparing a hard carbon anode material, using the aforementioned hard carbon anode material preparation system, the method comprising the following steps: S1. Real-time acquisition of temperature and pressure parameters from the hard carbon material preparation equipment via sensor network, generating an initial parameter interaction dataset characterized by the correlation between temperature and pressure parameters, and obtaining a sorted parameter mapping table; S2. Based on the sorted parameter mapping table, analyze the interaction between temperature and pressure parameters, preprocess the fluctuating data, weight the key parameters, and construct a preliminary parameter correlation matrix that reflects the influence of temperature changes on pressure response. S3. In the preliminary parameter correlation matrix, the temperature parameter is used as the main control perturbation parameter and the pressure parameter is used as the controlled adjustment parameter. When the fluctuation of the temperature parameter exceeds the preset threshold, the pressure parameter corresponding to the temperature parameter is iteratively adjusted through a genetic algorithm. The iterative convergence condition and population initialization method are set, and the adjusted pressure parameter set coupled with the crossover mutation probability and fitness function is designed to generate the set of pressure parameters coupled with the temperature change. S4. Obtain the adjusted pressure parameter set, perform secondary processing on the data range mapping and standardized conversion formula, and incorporate the missing value filling method to complete the incomplete data, thereby obtaining an optimized parameter pairing list for temperature and pressure dual-parameter coordinated control. S5. Based on the optimized parameter pairing list, conduct virtual simulation tests on the crystallinity changes of hard carbon materials in the microstructure, and filter the simulation results using a selection and elimination mechanism to determine whether the material performance meets the preset standards under the corresponding temperature and pressure coupling control conditions, thereby obtaining the performance verification dataset. S6. Based on the performance verification dataset, a feedback loop mechanism is used to dynamically update the parameter correlation matrix, so that the adjustment result of the pressure parameter has a reverse effect on the temperature-pressure coupling relationship model. The historical data constraints and parameter adjustment step size in the verification data are processed to determine the refined temperature-pressure dual-parameter coupling control scheme. S7. When the refined temperature and pressure dual-parameter coupling control scheme shows that it can make the pressure control index better than the historical average level under temperature fluctuation conditions, the production efficiency index is integrated by data aggregation technology, and the real-time processing delay problem is handled to generate a comprehensive benefit evaluation result. Based on this, the final optimization path of hard carbon material preparation process is determined, and a temperature and pressure dual-parameter coupling precision control execution scheme for actual production is obtained.

[0007] The beneficial effects of this invention are as follows: By constructing a control mechanism centered on the interrelation between temperature and pressure parameters, this invention achieves dynamic and precise adjustment of pressure parameters under temperature fluctuation conditions. On the one hand, through multi-source data acquisition and fusion, parameter correlation analysis, and feedback update mechanisms, this invention can systematically characterize the coupling law of temperature change to pressure response, overcoming the problems of parameter adjustment lag and control instability caused by the separate control of temperature and pressure and reliance on manual experience settings in existing technologies. On the other hand, by combining genetic algorithm optimization, simulation verification, and performance screening processes, the pressure regulation results are matched with the microstructure evolution and performance targets of hard carbon materials, effectively reducing the performance dispersion between different production batches and improving the consistency of material crystallinity and structural stability.

[0008] Furthermore, this invention makes full use of historical data and real-time process data to form a continuously correctable temperature-pressure coupling control model, which ensures the reliability of material performance while also improving production efficiency. It provides a technical solution with high stability, strong adaptability and significant industrial application value for the refined and intelligent control of the hard carbon material preparation process. Attached Figure Description

[0009] Figure 1 This is a block diagram of the system structure of the present invention; Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] like Figure 1 As shown, the present invention provides a technical solution: a preparation system for hard carbon anode material, comprising: The data acquisition and fusion module collects temperature and pressure parameter data in real time from the hard carbon material preparation equipment through a sensor network, generates an initial parameter interaction dataset characterized by the correlation between temperature and pressure parameters, and obtains a sorted parameter mapping table. The parameter correlation analysis module analyzes the interaction between temperature and pressure parameters based on the organized parameter mapping table, preprocesses fluctuating data, weights key parameters, and constructs a preliminary parameter correlation matrix that reflects the influence of temperature changes on pressure response. The temperature-pressure coupling regulation module uses temperature parameters as the main control disturbance parameters and pressure parameters as the controlled regulation parameters in the initial parameter correlation matrix. When the fluctuation of temperature parameters exceeds the preset threshold, the pressure parameters corresponding to the temperature parameters are iteratively adjusted through a genetic algorithm. The iterative convergence conditions and population initialization methods are set, and the adjusted pressure parameter set coupled with temperature changes is generated by combining the crossover mutation probability and fitness function design. The parameter optimization processing module obtains the adjusted pressure parameter set, performs secondary processing on the data range mapping and standardized transformation formula, and incorporates a missing value filling method to complete the incomplete data, resulting in an optimized parameter pairing list for temperature and pressure dual-parameter coordinated control. The performance simulation and verification module performs virtual simulation tests on the crystallinity changes of hard carbon materials in the microstructure based on the optimized parameter pairing list, and filters the simulation results by combining the selection and elimination mechanism to determine whether the material performance meets the preset standards under the corresponding temperature and pressure coupling control conditions, and obtains the performance verification dataset. The feedback update module, based on the performance verification dataset, uses a feedback loop mechanism to dynamically update the parameter correlation matrix, so that the adjustment result of the pressure parameter has a reverse effect on the temperature-pressure coupling relationship model, processes the historical data constraints and parameter adjustment step size in the verification data, and determines the refined temperature-pressure dual-parameter coupling control scheme. In the execution decision module, when the refined temperature and pressure dual-parameter coupling control scheme shows that it can make the pressure control index better than the historical average level under temperature fluctuation conditions, the production efficiency index is integrated through data aggregation technology, and the real-time processing delay problem is handled to generate a comprehensive benefit evaluation result. Based on this, the final optimization path of hard carbon material preparation process is determined, and a temperature and pressure dual-parameter coupling precision control execution scheme for actual production is obtained.

[0012] In the above implementation, the system first constructs a multi-source sensor network using a data acquisition and fusion module. This sensor network is deployed at key workstations in the hard carbon material preparation equipment, including the carbonization furnace, heat treatment chamber, and pressure control chamber, to collect real-time temperature and pressure parameter data reflecting the process status. The collected data is then fused after timestamp alignment and consistency verification to form an initial parameter interaction dataset characterized by the correspondence between temperature and pressure parameters. This dataset is further organized to generate a parameter mapping table, providing a structured data foundation for subsequent analysis.

[0013] Based on this, the parameter correlation analysis module processes the parameter mapping table. This involves smoothing, filtering, outlier removal, and normalization preprocessing of abnormal fluctuation data. Simultaneously, based on process experience and statistical weighting strategies, different weight coefficients are assigned to key temperature ranges and corresponding pressure ranges, thereby constructing a preliminary parameter correlation matrix reflecting the influence of temperature changes on pressure response. This correlation matrix is ​​used to characterize the degree of influence of temperature disturbances on pressure control behavior under different operating conditions.

[0014] The temperature-pressure coupled regulation module uses the preliminary parameter correlation matrix as the control basis, setting the temperature parameter as the main control disturbance parameter and the pressure parameter as the controlled regulation parameter. When the real-time fluctuation of the temperature parameter exceeds the preset threshold range, the genetic algorithm solution process is triggered to iteratively optimize the pressure parameter corresponding to the temperature parameter. In this process, the population initialization method and iterative convergence conditions are preset, and the crossover probability, mutation probability, and fitness function with pressure stability and response speed as the core are combined to generate an adjusted pressure parameter set coupled with temperature changes.

[0015] Subsequently, the parameter optimization processing module performs secondary processing on the adjusted pressure parameter set, unifies the parameter scale through data range mapping and standardization conversion formula, and introduces a missing value filling method to complete the incomplete data generated during the acquisition or calculation process, forming an optimized parameter pairing list that can be directly used for control decision-making, realizing the collaborative expression of temperature and pressure parameters in a unified parameter space.

[0016] The performance simulation and verification module constructs a virtual simulation model of the hard carbon material preparation process based on the optimized parameter pairing list, simulates and analyzes the crystallinity changes of the microstructure of hard carbon materials under different temperature and pressure coupling conditions, and filters the simulation results by combining a selection and elimination mechanism to determine whether the material performance obtained under the corresponding temperature and pressure coupling control conditions meets the preset process standards, and outputs a performance verification dataset.

[0017] Based on the performance verification dataset, the feedback update module introduces a feedback loop mechanism to apply the verification results back to the parameter correlation matrix, dynamically correcting the correlation weights and adjustment step sizes. At the same time, it combines historical data constraints to suppress excessive fluctuations in parameter adjustments, thereby forming a refined temperature and pressure dual-parameter coupling control scheme.

[0018] Finally, the execution decision module comprehensively evaluates the refined temperature and pressure dual-parameter coupling control scheme. When the evaluation results show that the pressure control index is better than the historical average level under temperature fluctuation conditions, the production efficiency index is integrated through data aggregation technology, and the real-time processing delay is compensated and analyzed to generate a comprehensive benefit evaluation result. Based on this, the final optimized path of hard carbon material preparation process is determined, and a precise temperature and pressure dual-parameter coupling control execution scheme for actual production is output.

[0019] The data acquisition and fusion module collects real-time temperature and pressure parameter data from the hard carbon material preparation equipment via a sensor network. It generates an initial parameter interaction dataset characterized by the correlation between temperature and pressure parameters, resulting in a refined parameter mapping table. This table includes the acquisition of real-time temperature and pressure data streams from the hard carbon material preparation equipment, combined with historical production batch records to construct an original multi-source mixed data sequence. The original multi-source mixed data sequence is then input into a multi-dimensional data fusion model for anomaly removal and noise reduction, yielding a clean production environment data stream. Feature extraction is performed on the clean production environment data stream to quantify the coupling degree between temperature and pressure, generating a parameter dependency matrix. Finally, the module analyzes the strongly correlated features in the parameter dependency matrix and establishes an index, resulting in the refined parameter mapping table.

[0020] In this embodiment, a sensor network is first established and the locations are defined. Temperature sensors are deployed in the feeding section, heating section, holding section, and discharging section of the carbonization furnace, while pressure sensors are deployed in key locations such as the reaction chamber, vacuum pipeline, inert gas supply pipeline, and back pressure regulation unit. Each sensor is bound to a unique equipment location number and process segment identifier. The equipment location number is determined based on the physical connection relationship between the equipment pipeline and the chamber, as well as the process segment division results that are sensitive to the microstructure of the material. The process segment identifier is determined based on the stage boundaries of heating, holding, and cooling in the production formula and the segment division of the equipment control program. After the point definition is completed, the acquisition end synchronously acquires data at a sampling period of 1 second. The 1-second sampling period is determined based on the matching relationship between the shortest effective fluctuation period of temperature and pressure, the sensor response time, and the controller refresh period. Specifically, the shortest identifiable fluctuation period of temperature and pressure under stable operating conditions in historical batches is first counted, then the sensor response time and controller refresh period are checked, and the smallest period that satisfies real-time performance and does not introduce significant quantization noise is selected. The stability of this period is verified by measuring the jitter amplitude of the sampled data during continuous operation in the field. A timestamp is added to each data point during acquisition. The timestamp comes from a unified time synchronization module with the same clock source. The selection of the time synchronization module is based on the existing clock synchronization system and network latency characteristics of the production system to ensure that the temperature and pressure data streams are expressed under the same time reference.

[0021] After acquiring real-time temperature and pressure data streams, historical production batch records are used to construct an original multi-source mixed data sequence. The selection of these historical production batch records follows a batch screening rule, determined based on product qualification results, equipment operational status integrity, and process formula consistency. Specifically, only batches meeting quality inspection standards and without shutdown alarms are included, while also requiring consistent process formula versions for each batch. When constructing the original multi-source mixed data sequence, time alignment is first performed on the real-time temperature and pressure data streams using a 5-second alignment window. This 5-second window is determined based on the relationship between the maximum transmission jitter statistics of the production network under high load and the integer multiple of the sensor sampling period. Specifically, the upper bound of the arrival delay of statistical data is calculated in the historical operation logs, and the smallest integer second above this upper bound is selected as the window to ensure that temperature and pressure data within the same window can be considered as the process status at the same moment. The alignment process employs a sliding window aggregation method. If multiple sample values ​​exist within a window, deduplication and merging are performed. Deduplication is based on timestamp duplication, and merging is based on the median within the window as the representative value. The median is chosen because it is insensitive to sudden spike noise and facilitates subsequent anomaly identification. If missing sample points appear within the window, missing data is marked and its length is recorded. This missing length record is used in the missing data processing branch of the subsequent fusion model to avoid misclassifying missing data as stationary changes. After time alignment is completed, real-time data and selected historical batch data are concatenated using the same field structure. The field structure includes temperature sample value, pressure sample value, process section identifier, equipment location number, and batch identifier. The concatenation order is arranged in ascending order of timestamps, and boundary markers are inserted at the boundaries of adjacent batches. These boundary markers prevent subsequent denoising steps from treating abrupt changes in different batches as the same continuous process.

[0022] When inputting the original multi-source mixed data sequence into the multi-dimensional data fusion model for anomaly removal and denoising, the multi-dimensional data fusion model first performs physical boundary verification, then dynamic change verification, and finally multi-source consistency verification. Physical boundary verification judges the upper temperature threshold, lower temperature threshold, upper pressure threshold, and lower pressure threshold. The upper and lower thresholds are determined based on the intersection of the equipment nameplate allowable range, the process formula target range, and the sensor range redundancy. Specifically, the allowable range of the equipment and the sensor range are read first, then the process target range is superimposed on them, and the narrowest range that simultaneously meets safety and process requirements is taken as the boundary threshold. The boundary threshold is recalculated each time the formula version changes and is verified under no-load and standard operating conditions before going live. Dynamic change verification assesses temperature and pressure change rate thresholds. These thresholds are determined based on the statistical upper bound of the maximum change per unit time in historical qualified batches and the maximum response rate of the equipment actuator. Specifically, the difference between adjacent sampling points in qualified batches is extracted, and its maximum value is calculated. This value is then compared with the maximum achievable change in the same sampling period calculated using the actuator's response rate. The smaller of the two is taken as the change rate threshold to avoid mistaking physically impossible abrupt changes in the actuator as real processes. Multi-source consistency verification assesses temperature and pressure difference thresholds between adjacent points within the same process section. These difference thresholds are determined based on the normal gradient range caused by the equipment structure and the upper bound of the gradient distribution in qualified batches. Specifically, the normal gradient range is first determined based on cavity volume, airflow path, and heating zone layout. Then, the difference between adjacent points in qualified batches is extracted, and its upper bound is calculated. The upper bound of the intersection of these two values ​​is taken as the difference threshold. Data points identified as abnormal through these three types of verification are removed, and an anomaly marker is retained at the same timestamp. This anomaly marker is used for subsequent index tracing and maintenance diagnosis to prevent the disappearance of anomalies from making the cause untraceable.

[0023] The noise reduction process employs a segmented smoothing strategy, with segmentation based on process segment identifiers to prevent the true changes in the heating and holding stages from being weakened by the same smoothing parameter. The smoothing window length is set to 9 sampling points. This length is determined by the conversion between the period corresponding to the main noise frequency band and the sampling period, as well as constraints on control response delay. Specifically, the high-frequency fluctuation period range of temperature and pressure is first statistically analyzed in qualified batches, then converted into the number of sampling points. The minimum odd number of points covering one complete fluctuation period without exceeding the allowable delay is selected as the window length to ensure that the smoothing result does not produce phase shift. During smoothing, a representative value is calculated for the data within each window. This representative value uses a truncated average method, with the truncated ratio determined by the upper limit of the outlier percentage. The upper limit of the outlier percentage is determined by the upper bound of the sensor's fault statistics frequency and network jitter frequency during long-term operation. Specifically, the frequency of outlier markers within a unit of time is statistically analyzed in the operation log, converted into the maximum possible number of outliers within the window, and the maximum number of samples at both ends to be removed is determined accordingly. The remaining samples are then averaged to obtain a clean production environment data stream. Missing data is not directly filled in during the denoising stage; only missing data markers are retained and the start and end times of missing data are recorded synchronously to prevent continuous missing data from being mistakenly processed as stationary segments by the smoothing algorithm.

[0024] When extracting features from a clean production environment data stream to quantify the coupling between temperature and pressure, the feature extraction revolves around synchronous changes, hysteresis response, and fluctuation consistency. First, temperature and pressure difference sequences are calculated. The calculation process for the difference sequences involves determining the change in sampled values ​​at adjacent time points and generating them separately for each process segment to avoid the impact of artificial abrupt changes at segment boundaries. The difference sequences are used to characterize the change trend within a unit sampling period. Subsequently, the intensity of synchronous changes is calculated within a 60-second analysis window. This 60-second analysis window is determined based on the effective response timescale caused by thermal inertia during the hard carbon material preparation process and the stabilization time of the pressure regulation mechanism. Specifically, it involves statistically analyzing the time required for pressure to reach a new stable range after a temperature disturbance occurs in qualified batches, and selecting the minimum window length that covers this time without introducing excessive hysteresis. The calculation process for the synchronous change intensity involves counting the number of times the temperature change and pressure change occur in the same direction and the number of times they occur in opposite directions within the window, taking into account the magnitude of the change. During the statistics, it is first determined whether the signs of the temperature change and pressure change are consistent, and then the magnitude of the change is used as the weight for accumulation. The weight is determined based on the relative magnitude ranking of the change magnitude within the window, so that large changes contribute more to the intensity and avoids small fluctuations dominating the judgment.

[0025] Beyond synchronous changes, the hysteresis response intensity is further calculated. This hysteresis response intensity is achieved through time offset matching, with the time offset range set from 0 to 30 seconds. The upper limit of the time offset range, 30 seconds, is determined based on the maximum response delay of the pressure regulating mechanism and the upper limit of allowable control hysteresis within the process segment. Specifically, it combines the slowest response from valve opening change to stable cavity pressure, using the slowest response observed in the qualified batch as a reference. The time offset matching process involves progressively delaying the pressure difference sequence, calculating the synchronous change intensity of each delay within the same window as the temperature difference sequence, and selecting the delay with the largest synchronous change intensity as the hysteresis response time. This hysteresis response time is then used as the coupling time characteristic of the process segment and incorporated into the parameter dependency matrix. To ensure stable results, an intensity difference threshold of 0.1 is set. This threshold is determined based on the upper bound of the natural range of intensity fluctuations between adjacent delay values ​​in the qualified batch. Specifically, it involves statistically analyzing the intensity difference distribution of each delay value in the qualified batch and selecting the smallest difference above the upper bound of the natural fluctuation as the threshold. This is used to avoid frequent jumps in the optimal delay value due to noise. If the difference between the maximum and second-largest intensity does not reach this threshold, a smaller delay value is used as the hysteresis response time. The selection of the smaller delay value is based on the principle that a shorter control response is more conducive to stability.

[0026] After feature extraction, a parameter dependency matrix is ​​generated. The construction process of the parameter dependency matrix consists of binning, statistical analysis, and normalization. The binning process divides the temperature range into several continuous intervals, and the pressure range into several continuous intervals. The number of temperature intervals and the number of pressure intervals are set to 12. The number of intervals is determined based on the scale of historical batch data, the process target resolution, and computational load constraints. Specifically, it aims to improve resolution while ensuring sufficient sample points in each interval, and simultaneously ensure that the online computation latency meets the real-time requirements of the control system. The temperature and pressure sample values ​​at each time point are mapped to the corresponding interval numbers and located in the matrix. The statistical process accumulates the contribution values ​​of the synchronous change intensity and the lag response time corresponding to each unit. When accumulating contribution values, missing labeled data are skipped, and the number of skips is recorded. The number of skips is used to measure the data integrity of that unit. The normalization process averages the accumulated results of each unit according to the number of samples, and reduces the weight of units with data integrity below a threshold of 0.8. The data integrity threshold of 0.8 is determined based on the minimum requirement for the proportion of reliable samples in online control and the upper bound of the missing rate in qualified batches. Specifically, the missing rate of qualified batches is statistically analyzed, its upper bound is determined, and then the minimum integrity proportion above this upper bound that meets the control reliability requirement is taken as the threshold. The weight reduction process involves multiplying the dependency strength of the unit by a weight proportional to its integrity, thus suppressing the influence of units with more missing data on the mapping relationship. The resulting parameter dependency matrix is ​​used to express the coupling strength distribution and typical hysteresis characteristics between the temperature and pressure ranges.

[0027] When analyzing the strong correlation features in the parameter dependency matrix and building an index, the dependency strength of all pressure intervals corresponding to each temperature interval in the matrix is ​​first sorted, based on the dependency strength from high to low. A strong correlation threshold is set to a dependency strength greater than 0.7. This threshold of 0.7 is determined by the lower bound of the dependency strength corresponding to stable pressure control and stable material crystallinity in qualified batches. Specifically, within qualified batches, time periods with small pressure fluctuations and crystallinity meeting standards are marked, the dependency strength of the corresponding matrix units is extracted, and the lower bound is calculated. A value slightly higher than this lower bound is then used as the threshold to avoid including weak coupling relationships in the mapping. An index is built for units that meet the strong correlation threshold. The index content includes the temperature interval number, the corresponding pressure interval number, the dependency strength level, the hysteresis response time, and the process segment identifier. The dependency strength level is divided into three levels, and the threshold for these three levels is determined by segmenting the historical distribution of dependency strength. Specifically, on the qualified batch samples, the dependency strength distribution is segmented according to the boundaries between dense and sparse areas, and high, medium, and low level intervals are formed using the boundary points, ensuring that the index balances accuracy and stability during retrieval. Finally, the indexes are aggregated by temperature range number to form a sorted parameter mapping table. Each record in the parameter mapping table corresponds to a mapping relationship between the temperature range number and the pressure range number, and is accompanied by the hysteresis response time and the dependence strength level. This is used by the subsequent parameter correlation analysis module to quickly locate the pressure regulation target range that is most closely related to the temperature disturbance in different process sections.

[0028] The parameter correlation analysis module analyzes the interaction between temperature and pressure parameters based on the organized parameter mapping table. It preprocesses fluctuating data, weights key parameters, and constructs a preliminary parameter correlation matrix reflecting the influence of temperature changes on pressure response. This includes extracting temperature and pressure parameter sequences from the organized parameter mapping table; exponentially smoothing fluctuating data segments in the temperature and pressure parameter sequences where the gradient exceeds a preset gradient to obtain smoothed interaction sequences; performing time-series lag analysis on the smoothed interaction sequences to align parameters; assigning feature weights based on the contribution of temperature to pressure in the smoothed interaction sequences to generate a weighted feature dataset; calculating mutual information values ​​based on the weighted feature dataset; and filling the mutual information values ​​into a preset structure to construct a preliminary parameter correlation matrix reflecting the influence of temperature changes on pressure response.

[0029] In this implementation, data extraction and structured organization are first performed based on the organized parameter mapping table. The module reads the temperature range number, pressure range number, process segment identifier, timestamp, and the temperature and pressure sampling values ​​corresponding to the mapping relationship from the mapping table. Then, the data is segmented and organized according to the process segment identifier, so that each segment of data only covers the temperature and pressure interaction process within the same process stage, avoiding confusion of interaction relationships caused by differences in operating conditions across stages. After segmentation, the module arranges the records within each segment in ascending order of timestamps and performs consistency checks: if timestamps are duplicated, one data record marked as the master record is retained, and the remaining records are converted into redundant records with the redundancy count recorded. The redundancy count is used for subsequent data quality assessment. If the timestamp interval deviates from the sampling period by 1 second, the deviation position is resampled and aligned. The 1-second sampling period is determined based on the combined constraints of the actual configuration of the acquisition end, the sensor response time, and the controller refresh period, taking the smallest integer second that satisfies real-time requirements and is consistent with the controller refresh. The resampling alignment uses a fixed alignment window of 5 seconds. The 5-second alignment window is determined based on the upper bound statistical results of network transmission delay and data arrival jitter in the historical operation log, taking the smallest integer second that covers this upper bound, thereby ensuring that temperature data and pressure data express the same process state within the same time slice. If a continuous missing marker exceeds 15 seconds in length, the missing range will be marked as a long missing segment and removed from the current association calculation. The 15-second limit is determined by the statistical results of the maximum effective window time allowed by the control system and the upper limit of the continuous missing length in the qualified batch. The smallest integer seconds that meets the online calculation stability requirements and covers the upper limit is selected to avoid the offset caused by long missing segments in subsequent lag alignment and dependency statistics.

[0030] After extracting the temperature and pressure parameter sequences, the module calculates the gradient of change for each sequence to locate fluctuating data segments that require smoothing. The gradient calculation process is as follows: for each time point, the difference between its value and the corresponding time point one second prior is calculated to obtain the change at that time point, and the change is used as the gradient representation value for that time point; if the sequence is missing at a certain time, the gradient at that time is marked as unusable and skipped, and is not included in the threshold determination. Subsequently, the gradient characterization values ​​were compared point by point with the preset gradient thresholds. The temperature change gradient threshold was set to 3, and the pressure change gradient threshold was set to 5. The temperature change gradient threshold of 3 was determined based on the upper bound statistical results of the maximum natural temperature change intensity per unit time in each process section of the qualified batch, and compared with the maximum change intensity per unit time that can be generated by a step change in the heating power of the equipment. The smaller value of the two was taken and rounded down to obtain 3, thereby excluding abrupt changes that are difficult to form by physical mechanisms. The pressure change gradient threshold of 5 was determined based on the upper bound statistical results of the maximum natural temperature change intensity per unit time in the pressure stable section of the qualified batch, and compared with the maximum change intensity converted to a 1-second sampling scale by the response rate of the pressure regulating actuator. The smaller value of the two was taken and rounded down to obtain 5, in order to suppress the interference of actuator jitter or transient impact on the judgment of the interaction relationship. When determining the threshold, the module merges the time points that continuously exceed the threshold into the same fluctuation segment, and introduces a segment termination condition of 10 consecutive seconds without exceeding the threshold to close the segment boundary. The 10 seconds is determined based on the lower bound statistical result of the time required to recover to the stable fluctuation zone after the pressure adjustment action in the qualified batch, and takes the smallest integer seconds slightly higher than the lower bound to avoid the segment being split into multiple fragments due to a short drop.

[0031] For each fluctuating data segment, the module performs exponential smoothing to obtain a smoothed interactive sequence. The exponential smoothing process is completed recursively point-by-point in chronological order: the smoothed value at the starting point of the segment is taken as the original sampled value at that point as the initial smoothing benchmark; the smoothed value for each subsequent second within the segment is obtained by fusing the current sampled value and the smoothed value of the previous second at a fixed ratio. During the fusion process, a smaller proportion is assigned to the current sampled value, and a larger proportion is assigned to the smoothed value of the previous second, thereby suppressing high-frequency jitter and preserving the trend term. The smoothing coefficient is set to 0.2. The determination of 0.2 is based on the statistical results of the proportion of high-frequency jitter energy in the temperature and pressure signals in the qualified batch and the response delay constraint of the actual step change. Specifically, the intensity level of high-frequency jitter is first statistically analyzed in the qualified batch, and then the peak offset and response tail of the smoothed sequence are compared under different coefficients. The smallest coefficient that can significantly reduce jitter intensity without significantly lengthening the trend change establishment time is selected, and its stability is verified using continuous production data before going live. Exponential smoothing is performed on the temperature segment and the pressure segment respectively. After the execution is completed, the two smoothed segment sequences are paired and merged according to the same timestamp to form a smoothed interactive sequence. If a timestamp has a valid value on the temperature side but is missing on the pressure side, or has a valid value on the pressure side but is missing on the temperature side, the timestamp record is removed and counted in the pairing removal count. The pairing removal count is used for subsequent weight reliability correction to avoid the contribution being misjudged as high due to unilateral missing values.

[0032] After obtaining the smoothed interaction sequence, the module performs time-series hysteresis analysis to align the interaction between temperature and pressure. The hysteresis analysis employs a candidate hysteresis scanning method: the candidate hysteresis range is set from 0 to 30 seconds. The 30-second range is determined based on the statistical results of the maximum response delay from the triggering of the pressure regulating actuator to the pressure entering the stable zone in the chamber, combined with the smaller upper bound of the allowable control hysteresis limit imposed by the process. The candidate hysteresis step size is set to 1 second, determined based on the consistency requirement between the sampling period and the controller refresh period, ensuring that the alignment granularity is consistent with the sampling granularity. During the scanning process, the module performs an alignment evaluation for each candidate hysteresis: shifting the entire pressure smoothing sequence backward by the corresponding number of seconds to align the pressure response with the temperature disturbance in time; calculating the degree of synchronization within a fixed evaluation window of 60 seconds. The 60-second evaluation window is determined based on the statistical results of the effective response timescale caused by the thermal inertia of the hard carbon material preparation process and the stabilization time of the pressure regulating mechanism, taking the smallest integer second that covers the main response process without introducing excessively long statistical hysteresis. The synchronization calculation process consists of three steps: Step 1 transforms the smoothed temperature and pressure sequences within the window into sequences of changing directions, categorized into three types: rising, falling, and stable. Direction determination is based on the sign of the change within an adjacent second and whether it is zero. Step 2 counts the number of times the temperature and pressure directions occur in the same direction and the number of times they occur in opposite directions. Step 3 introduces amplitude weights to improve the distinguishability of effective changes. The amplitude weights are determined by the pressure change amplitude level, which is divided into three levels. The threshold for each level is determined by the densely distributed boundary point of the pressure change amplitude distribution in the qualified batch. Using this densely distributed boundary point as the level boundary ensures that the level classification is insensitive to abnormal peaks but distinguishes typical changes. For stable and consistent events with stable temperature and pressure, the module assigns a lower weight. The lower weight is determined because the stable segment contributes limited to the distinguishing effect of coupling identification; the weight is no higher than 1 / 3 of the weight of rising or falling events, ensuring that stable and consistent events participate in the statistics but do not dominate the results. After calculating the synchronization degree of all candidate lags, the module selects the candidate lag with the highest synchronization degree as the alignment lag and applies it to the smoothed interaction sequence to form the aligned interaction sequence. To suppress frequent jumps in lag caused by noise, a difference threshold of 0.1 is introduced. When the difference between the highest and second highest synchronization degree is less than 0.1, the smaller lag is taken as the alignment lag. The determination of 0.1 is based on the statistical results of the natural fluctuation of the upper bound of the synchronization degree difference between adjacent candidate lags in the qualified batch. The smallest difference higher than the upper bound is taken, so that the control lag is reduced first when the synchronization degree is similar.

[0033] After completing the time-series alignment, the module assigns feature weights based on the aligned interaction sequence and forms a weighted feature dataset. The feature weights are centered on the contribution of temperature to pressure. The contribution is calculated using an error reduction evaluation method, and the calculation process is divided into two paths: baseline prediction and enhanced prediction. The baseline prediction path uses only the short-term inertia of pressure itself to form a pressure baseline prediction sequence: the pressure smoothing data of 10 seconds prior to the current moment is used as a reference window. The 10-second reference window is determined based on the statistical results of the lower bound of the autocorrelation duration of pressure in the stable adjustment phase in the qualified batch, taking the smallest integer second slightly higher than this lower bound to make the inertial prediction representative and not introduce too much historical lag. Within this window, the pressure change trend is calculated. The trend extraction uses a method that gives higher weight to near-end data to obtain the trend direction and trend amplitude level. The amplitude level follows the aforementioned 3-level division to ensure consistency of dimensions. The enhanced prediction path incorporates temperature information as a correction to the baseline prediction. The correction direction is determined by the statistical direction of the strongly correlated index in the parameter mapping table. Specifically, it involves statistically analyzing the dominant change direction of the corresponding pressure interval number under the same temperature interval number, and using this dominant direction as the directional constraint of the pressure response to temperature disturbances. The correction magnitude is determined by the temperature change magnitude level, which is also divided into three levels. The threshold is determined based on the densely distributed boundary point of the temperature change magnitude distribution in the qualified batches, ensuring that the temperature correction is not affected by a few extreme points. Subsequently, the prediction errors of the two paths are calculated separately. The error calculation adopts the absolute deviation accumulation method: the absolute value of the difference between the predicted value and the smoothed measured pressure value is taken every second within the window and accumulated, then divided by the number of valid samples in the window to obtain the average error. If there are paired rejection counts or missing counts within the window, the number of valid samples is reduced accordingly, and the reduction record is retained for weight reliability correction. The contribution is the difference between the baseline average error and the enhanced average error; the larger the difference, the stronger the explanatory power of temperature information on pressure. To map contribution levels to feature weights, the module sorts the contribution levels of all windows within the same process segment and divides the weights into 5 levels. The 5 levels are determined by the combined constraints of online computational load and weight resolution; too few levels lead to insufficient differentiation, while too many levels result in sensitive fluctuations. The minimum number of levels that can stably distinguish the main contribution levels is chosen. Each level corresponds to a discrete weight value, with weight values ​​increasing from low to high according to the level. Subsequently, reliability correction is performed on each window based on the pairwise rejection count and missing count within that window. Higher counts result in a lower weight, with a reduction step size set to 0.05. This 0.05 is determined based on the statistical results of the sensitivity of missing and rejection counts in qualified batches to contribution stability, choosing the minimum step size that significantly suppresses the impact of unreliable windows without causing overall weight collapse. The completed output is a weighted feature dataset. Each record includes a temperature range number, pressure range number, alignment lag, synchronization level, contribution level, feature weight, process segment identifier, and the corresponding valid sample count.

[0034] When constructing the preliminary parameter correlation matrix based on the weighted feature dataset, the module first performs joint counting statistics and introduces a weight accumulation mechanism. The statistical process uses temperature interval numbers to locate matrix rows and pressure interval numbers to locate matrix columns. For each weighted feature record, its feature weight is accumulated into the corresponding two-dimensional cell, ensuring that high-contribution windows occupy a larger share in the statistics. Simultaneously, the effective sample count is accumulated for each two-dimensional cell to measure the data support strength of that cell. After completing the joint weight counting, the module calculates the temperature marginal weight count and the pressure marginal weight count separately: the temperature marginal weight count is obtained by summing the weights of all pressure interval number cells corresponding to the same temperature interval number, and the pressure marginal weight count is obtained by summing the weights of all temperature interval number cells corresponding to the same pressure interval number. Then, normalization processing is performed: the joint weight count of each two-dimensional cell is divided by the sum of the joint weight counts of all two-dimensional cells within the process section to obtain the joint weight percentage; the temperature marginal weight count is divided by the sum to obtain the temperature marginal percentage; and the pressure marginal weight count is divided by the sum to obtain the pressure marginal percentage. During the normalization process, if the joint weight count of a two-dimensional unit is 0, the joint weight percentage of that unit is recorded as 0 and the empty unit label is retained. The empty unit label is used for subsequent matrix sparsity evaluation to avoid instability caused by empty units participating in information calculation.

[0035] The calculation of mutual information value focuses on the deviation of the actual joint proportion from the expected joint proportion under the independence assumption. The module first calculates the expected independent joint proportion for each two-dimensional cell, which is obtained by multiplying the marginal proportion of temperature and the marginal proportion of pressure corresponding to that cell. This percentage represents the proportion that the cell should exhibit under the assumption of mutual independence between temperature and pressure. Next, the ratio of the actual joint proportion to the expected independent joint proportion is calculated. A ratio greater than 1 indicates enhanced dependence, while a ratio less than 1 indicates weakened dependence. Then, a logarithmic operation is performed on this ratio to obtain the information gain characterization value. The introduction of the logarithmic operation is based on the additivity requirement of information measurement for multiple changes, ensuring that the dependence contributions from multiple sources maintain a stable scale during accumulation. Finally, the actual joint proportion of the cell is used to weight the information gain characterization value to obtain the mutual information value of that two-dimensional cell, thus suppressing the impact of small, incidental cells on the overall correlation. If the actual joint proportion is 0, the mutual information value of the unit is directly recorded as 0 and the logarithmic operation is skipped to avoid numerical instability caused by zero proportion. If the independent expected joint proportion is 0 but the actual joint proportion is not 0, the unit is marked as marginal sparsity anomaly and transferred to the sparsity correction process. The sparsity correction process backtracks the source of the marginal proportion of the unit. If the backtracking finds that the marginal proportion of 0 is caused by the concentrated removal of long missing segments, the mutual information value of the unit in this round is replaced by the weighted average of the mutual information values ​​of the units in the adjacent temperature range and the adjacent pressure range. The weighting coefficient is proportional to the effective sample count of the adjacent units, so that the replacement value is dominated by the adjacent units with more sufficient data support.

[0036] After calculating the mutual information values, the module fills the mutual information values ​​of each two-dimensional cell into a preset structure to form a preliminary parameter correlation matrix. The preset structure adopts an interval numbering system consistent with the parameter mapping table, with matrix rows corresponding to temperature interval numbers and matrix columns corresponding to pressure interval numbers. The matrix stores the mutual information values ​​along with typical alignment lag and effective sample counts. The typical alignment lag is determined using a weighted counting method: the alignment lag of all records within the two-dimensional cell is counted, and the feature weight is used as the counting weight. The lag with the highest cumulative weight is taken as the typical alignment lag, making the influence of high-contribution windows on the typical lag more significant. To suppress abrupt changes in the matrix over time, the module performs stability screening on the mutual information value of each two-dimensional unit. The screening difference threshold is set to 0.15. The determination of 0.15 is based on the statistical results of the upper bound of the natural fluctuation of the mutual information value of the same process section in qualified batches between adjacent statistical periods. The smallest difference value above this upper bound is used to identify abnormal abrupt changes. When the difference between the mutual information values ​​of two adjacent statistical periods exceeds 0.15, the module backtracks the effective sample count, pair rejection count, and missing count in the corresponding statistical period of the unit. If the effective sample count is less than 1 / 2 of the average effective sample count of the same process section or the rejection and missing counts are significantly increased, the mutual information value of the unit in this period is replaced by the mutual information value of the previous period, and the replacement mark is retained for the subsequent temperature and pressure coupling adjustment module to identify that the unit is in a state of limited data quality, thereby obtaining a preliminary parameter correlation matrix reflecting the influence of temperature change on pressure response.

[0037] The temperature-pressure coupled regulation module uses temperature as the master perturbation parameter and pressure as the controlled regulation parameter in the initial parameter correlation matrix. When the temperature parameter fluctuation exceeds a preset threshold, a genetic algorithm iteratively adjusts the pressure parameter corresponding to the temperature parameter. The module sets the iteration convergence condition and population initialization method, and combines crossover and mutation probabilities with a fitness function to generate an adjusted pressure parameter set coupled with temperature changes. This includes extracting temperature and pressure parameter sequences from the initial parameter correlation matrix, identifying time windows where temperature parameter fluctuations exceed limits to determine the optimization space for the pressure parameter to be optimized, generating an initial population based on the optimization space, calculating the fitness value of the quantified response matching degree for the initial population, performing crossover and mutation operations based on the fitness value to generate the next generation population, and outputting the optimal individual when the next generation population meets the iteration convergence condition, and analyzing the optimal individual to generate the adjusted pressure parameter set coupled with temperature changes.

[0038] In this implementation, the temperature and pressure parameter sequences are first extracted and aligned based on a preliminary parameter correlation matrix. The module reads the pressure interval number, mutual information strength, typical lag time, effective sample count, and process segment identifier corresponding to each temperature interval number in the matrix, and simultaneously reads real-time temperature and pressure sample values ​​consistent with the source of the matrix unit. After extraction, the data is organized into segments based on the process segment identifier, ensuring that each sequence segment only covers the temperature and pressure interaction process within the same process stage. Subsequently, time alignment was performed on the temperature and pressure sequences. During alignment, the typical lag duration was used as the shift amount. The typical lag duration was determined based on the weighted mode statistical result of the alignment lag duration obtained by the parameter correlation analysis module from the records within the matrix cell. The statistical weight was the cumulative value of the feature weights, so that the data with higher contribution played a dominant role in the lag duration. The specific process of alignment was to shift the pressure sequence backward on the time axis by the number of seconds corresponding to the typical lag duration, so that the temperature change and pressure response at the same time point were in the same response stage. The head gap caused by the shift was removed, and the removal length was recorded. The removal length was used to correct the boundary of the subsequent optimization window to prevent the window from going out of bounds.

[0039] After alignment and organization, the module identifies time windows where temperature parameter fluctuations exceed limits, used to determine the optimization space for the pressure parameter to be optimized. The temperature fluctuation threshold is generated during triggering, constrained by both the target temperature band and the natural fluctuation band of historical qualified batches. The target temperature band is provided by the process formula, including the target center temperature, allowable upper deviation, and allowable lower deviation for that process segment. The natural fluctuation band of historical qualified batches is obtained statistically. The statistical process involves selecting batches consistent with the current formula version and meeting quality standards, extracting the temperature sequence by process segment, calculating the deviation of the temperature from the target center temperature at each time point, and then taking the maximum deviation within that process segment as the upper bound of the natural fluctuation. The temperature fluctuation threshold is determined by the stricter of the allowable deviation boundary and the upper bound of the natural fluctuation, forming the actual triggering threshold. The method is as follows: if the upper bound of the natural fluctuation is less than the allowable deviation boundary, the upper bound of the natural fluctuation is used as the threshold; if the upper bound of the natural fluctuation is greater than the allowable deviation boundary, the allowable deviation boundary is used as the threshold, ensuring that the triggering rule simultaneously satisfies process requirements and historical stability constraints. The process of determining the over-limit involves calculating the deviation of the real-time temperature from the target center temperature point by point. A deviation exceeding the temperature fluctuation threshold is considered an over-limit point. Consecutive over-limit points are merged into the same over-limit segment, and the duration of continuous non-over-limit operation until stabilization is used as the termination condition for the over-limit segment. The stabilization duration is set to 10 seconds. This 10-second duration is determined by the lower bound statistical result of the time required for the pressure regulating actuator to complete one effective action and bring the pressure into the stable fluctuation zone. The shortest duration from the action triggering the action to the pressure entering the stable zone is statistically analyzed in the qualified batch operation log, and the smallest integer second slightly higher than this lower bound is taken to ensure the window covers the complete response convergence process. Each over-limit segment extends forward by a 5-second preheating boundary and backward by a 10-second stabilization boundary. The 5-second and 10-second extensions are determined by the combined constraints of the upper bound of network delay jitter, the typical duration of the pressure hysteresis response, and the control calculation scheduling cycle: forward extension covers the leading changes in temperature disturbances, and backward extension covers the tail fluctuations of the pressure stabilization phase. If a sequence start or end boundary is encountered after extension, the segment is truncated at the boundary and a truncation mark is recorded.

[0040] After determining the temperature exceedance time window, the module uses this window as the center to define the optimization space for the pressure parameters to be optimized. The optimization space consists of the upper and lower boundaries of the pressure setpoint, the upper boundary of the pressure change rate, the lower boundary of the pressure change rate, and the upper boundary of the pressure adjustment step size. The upper and lower boundaries of the pressure setpoint are obtained by intersecting three types of constraints: the allowable pressure range of the equipment is given by the equipment safety parameters; the pressure range of the process formula is given by the target pressure band of the formula; and the historical stable pressure band of qualified batches is obtained statistically. The statistical process involves extracting pressure sequences within the same process segment, calculating the deviation of the pressure from the target pressure, and taking the maximum deviation to form the stable band boundary. The intersection of these three types of boundaries forms the upper and lower boundaries of the optimized pressure setpoint, ensuring that the optimization result simultaneously satisfies safety, process, and historical stability constraints. The pressure change rate boundary is determined jointly by the actuator's response capability and historical pressure change statistics: First, the maximum pressure change capability per unit time is obtained from the actuator specifications; then, the upper limit of the pressure change per unit time is statistically analyzed in qualified batches, and the smaller of the two is taken as the upper boundary of the pressure change rate. The lower boundary of the pressure change rate is taken as the symmetrical value in the opposite direction of the upper boundary, used to constrain the symmetry and feasibility of upward and downward adjustments. The upper boundary of the pressure adjustment step size is jointly determined by the minimum resolution of the pressure control valve, the resolution of the pressure sensor, and the online control cycle: First, the pressure change corresponding to the minimum effective change in valve opening is read; then, the minimum discernible pressure change corresponding to the sensor resolution is read, and the larger of the two is taken as the minimum effective step size; then, the upper limit of a single action is calculated by combining the maximum number of actions allowed within the control cycle, forming the upper boundary of the step size, making the single adjustment measurable while avoiding excessively frequent actions.

[0041] When generating the initial population based on the optimization space, the module first defines the individual expression structure, so that each individual corresponds to a pressure regulation trajectory within the over-limit time window. The trajectory consists of a segmented pressure target sequence, with the number of segments set to 6. The determination of 6 segments is based on the typical duration distribution of the over-limit window and the constraint of the segmented adjustment granularity that the controller can execute: the common range of over-limit window duration is statistically analyzed in historical qualified batches, and this range is divided into a segment granularity of no less than 4 segments and no more than 8 segments, taking the middle integer 6, which balances resolution and computational load. Each segment includes the segment start time, segment end time, segment target pressure, maximum upward adjustment rate within the segment, maximum downward adjustment rate within the segment, and action interval within the segment; all of the above parameters fall within the boundary of the optimization space. The initial population size was set at 40 individuals. The determination of 40 was based on the computational resource budget and population diversity requirements within the control period: first, the average time taken to assess the fitness of a single individual on the target hardware was measured; then, the maximum computation time allowed within the control period was combined to calculate the upper limit of the number of individuals that could be assessed per generation. Simultaneously, the population size was required to be no less than 30 to avoid insufficient sample coverage. Finally, the minimum value of 40, which satisfies the upper limit constraint and is an integer multiple of 10, was chosen to facilitate parallel evaluation and statistics. The initialization method used a combination of historical guidance and random coverage: the proportion of historically guided individuals was 0.6, determined based on the coverage density of historical qualified batches within the current temperature range and the reliability of the current matrix's effective sample count; a higher proportion was used for higher coverage density and larger effective sample counts. The proportion of randomly covered individuals was 0.4, used to expand the search range and maintain diversity. The generation process of historically guided individuals involves selecting records from qualified batches whose temperature range numbers match the current over-limit segment, extracting their pressure adjustment trajectories, and aligning them with the current over-limit window length using time scaling. The time scaling process involves mapping the segment boundaries of the historical trajectory proportionally to the start and end times of the current window and trimming the segment target pressure boundaries. The generation process of randomly covered individuals involves generating segment target pressure and intra-segment velocity within each segment based on the pressure setpoint boundary and velocity boundary, and applying a continuity constraint to the difference between target pressures of adjacent segments. The threshold for the continuity constraint is 5 times the upper boundary of the pressure adjustment step size. The determination of 5 times is based on the statistical results of typical pressure change amplitudes between adjacent effective actions in qualified batches, taking an integer multiple slightly higher than this typical amplitude to ensure that the trajectory has executable smooth transition characteristics.

[0042] Fitness values ​​are calculated for all individuals in each generation. Fitness is determined by three indicators: response matching, pressure stability, and adjustment cost. A weighted average is used to obtain the total score. The weights of the indicators are determined during the calculation. The weight of response matching is set to 0.5, based on the priority of restoring the temperature-pressure coupling relationship under temperature disturbance conditions, which is higher than stability and cost. The weight of pressure stability is set to 0.3, based on the fact that pressure fluctuations directly affect the subsequent microstructure stability of materials, but its effect is constrained by the prerequisite of response matching. The weight of adjustment cost is set to 0.2, based on the fact that cost is used to suppress excessive action and protect the actuator, and its importance is lower than the process stability target. The calculation process of response matching degree is as follows: First, based on the preliminary parameter correlation matrix, locate the set of high correlation intervals of pressure corresponding to the current temperature range. The screening threshold of the set of high correlation intervals is set to mutual information intensity of not less than 0.7. The basis for determining 0.7 is the statistical result of the lower bound of mutual information intensity corresponding to the time period when the pressure stabilizes after temperature disturbance and the material performance meets the standard in the qualified batch, and the minimum value slightly higher than the lower bound is taken to exclude weakly coupled units. Then, generate the predicted pressure sequence for individual trajectories within the over-limit window. The generation process of the predicted pressure sequence is to construct a step sequence according to the segmented target pressure and perform speed-limited interpolation on the transition between segments according to the rate limit within the segment. The interpolation process is to divide the pressure difference between segments into several small steps of adjustment according to the action interval, and the adjustment amplitude of each step does not exceed the upper boundary of the pressure adjustment step size. Then, align and compare the predicted pressure sequence with the measured pressure sequence. The alignment uses typical lag time, and the comparison content includes directional consistency, amplitude landing area consistency, and lag consistency. The calculation process for directional consistency involves comparing the direction of temperature change with the direction of pressure change at each 1-second time point to see if they conform to the dominant coupling direction recorded in the matrix. Discrepancies are counted as discrepancies. The calculation process for amplitude range consistency involves determining whether the predicted pressure falls within the pressure range defined by the set of highly correlated intervals. Failures to fall within the range are counted as failures. The calculation process for hysteresis consistency involves comparing the difference between the initial change time of the predicted pressure response and the initial change time of the measured pressure response. Differences exceeding the allowable hysteresis deviation are counted as deviations. The allowable hysteresis deviation is set to 2 seconds, determined based on the discernible time resolution limit under a 1-second sampling period and a 5-second network alignment window, combined with the smallest integer second obtained from the upper bound statistical results of hysteresis jitter in qualified batches. Discrepancies, failures, and deviations are converted into deduction items. The conversion method is to divide each number by the number of valid samples within the window to obtain a proportion, then convert it to a deduction value using a deduction coefficient of 1. The determination of 1 is to directly map the proportion to the deduction intensity, making the deduction monotonic and easy to interpret for thresholds.The calculation process for pressure stability is as follows: The difference between the maximum and minimum values ​​of the predicted pressure sequence within the out-of-limit window is calculated as the fluctuation amplitude; the smaller the fluctuation amplitude, the higher the score. The duration for which the predicted pressure sequence enters the target pressure zone and remains stable is calculated as the stabilization time; the shorter the stabilization time, the higher the score. The target pressure zone is given by the process formula and constrained by the stability zone of historical qualified batches. Its formation method is consistent with the temperature fluctuation threshold, and a stricter boundary is taken as the judgment zone. The calculation process for adjustment cost is as follows: The number of actions and the cumulative action amplitude of the predicted pressure sequence within the out-of-limit window are statistically analyzed. The number of actions is obtained by counting the number of non-zero steps in the sequence, and the cumulative action amplitude is obtained by accumulating the absolute change of each step. The higher the cost, the more points are deducted. The deduction slope is determined by the maintenance cycle constraint of the actuator life model. The maintenance cycle constraint is obtained by the upper limit of the number of actions per unit time allowed in the equipment maintenance plan. After converting the upper limit into a window scale, the deduction slope is determined to ensure that the cost constraint is consistent with the equipment maintenance requirements. After scoring the three categories of indicators, the individual fitness values ​​are obtained by summarizing them according to their weights. The fitness values ​​are then truncated to a range of 0 to 1. The criteria for determining the truncation boundary 0 and 1 are to unify the scores into a comparable range, which facilitates the selection of probability calculations and convergence determination.

[0043] Selection, crossover, and mutation are performed after each generation. The selection phase employs a preservation strategy, directly retaining the top two individuals with the highest fitness into the next generation. The selection of 2 is based on a retention ratio of 0.05 under a population size of 40, maintaining the continuation of high-quality solutions while avoiding premature population maturation. After retention, stratified sampling based on fitness is performed on the remaining individuals. The number of strata is set to 5, determined by the discriminative requirements of fitness distribution and implementation complexity constraints. Individuals are sorted by fitness from high to low and divided into 5 groups. Within each group, parents are extracted according to their group weights, decreasing by group number with a decreasing ratio of 0.7. This 0.7 is determined by the balance between high-fitness individuals dominating reproduction and maintaining diversity observed in qualified replay data, taking the minimum value that makes high-fitness groups more likely to be selected while low-fitness groups still have a chance to enter. The crossover phase is executed with a crossover probability of 0.8. The determination of 0.8 is based on the need to improve the efficiency of combinatorial search when the upper limit of the number of generations allowed by the control period is tight, while the instability caused by excessive crossover is offset by the preservation of excellence strategy. The specific process of crossover operation is to perform segment-level exchange for each pair of parents according to the segment structure. The exchange position is determined by a random segment index. The generation of the segment index is restricted by the continuity constraint to ensure that the target pressure difference between adjacent segments of the offspring after the exchange does not exceed the continuity constraint threshold. If it does, the resampling of the segment index is triggered. The upper limit of the number of resampling is set to 10. The determination of 10 is based on avoiding infinite retries under the online calculation delay constraint, and is sufficient to find the exchange position that meets the constraint in practice. The mutation phase is executed with a mutation probability of 0.1. The determination of 0.1 is based on the requirement to maintain the lower limit of exploration capability when the historical guidance ratio is 0.6, so as to avoid the population getting stuck in local optima within a few generations. The specific process of mutation operation is to adjust the target pressure and the rate within a randomly selected segment for the selected offspring. The adjustment range is constrained by the upper boundary of the pressure adjustment step size, and the adjustment direction is randomly selected between upward and downward adjustment. Boundary clipping is performed on the mutated trajectory. The clipping rule is that if the pressure exceeds the upper or lower boundary of the set value, the boundary value is taken; if the rate exceeds the rate boundary, the rate boundary is taken, so that the offspring always meets the equipment and process constraints.

[0044] The iterative convergence condition is determined before the next generation is generated after each generation. The convergence condition includes the improvement magnitude condition, the stability threshold condition, and the upper limit of the generation condition. The improvement magnitude condition is that the optimal fitness improvement is less than 0.01 for 5 consecutive generations. The 5 and 0.01 are explained as follows: the determination of 5 consecutive generations is based on the fact that short-cycle jitter of the optimal fitness improvement often occurs in generations 1 to 3 in the replay data. The smallest integer higher than the jitter period is taken to exclude random fluctuations. The determination of the improvement threshold of 0.01 is based on the statistical result of the upper limit of the natural fluctuation amplitude of fitness in qualified batch replays. The smallest value slightly higher than the upper limit is taken so that the threshold can distinguish between effective improvement and statistical jitter. The stability criteria are as follows: the predicted pressure fluctuation amplitude should not exceed 0.2 times the target pressure bandwidth, and the stabilization time should not exceed 10 seconds. The 0.2 value is determined based on the empirical boundary statistical results between pressure fluctuation amplitude and material performance compliance in qualified batches, taking the upper limit of relative fluctuations commonly seen in qualified batches while allowing a safety margin. The 10-second value is determined in the same way as the aforementioned stabilization time, ensuring that the pressure stabilization enters a controllable range. The upper limit of the number of generations is set to 30 generations. The 30 is determined by the product of the maximum allowable computation time of the control cycle, the average time consumed per generation evaluation, and the population size of 40. After measuring the time consumed per generation through online benchmark testing, the allowable time is divided by the time consumed per generation and rounded down to obtain the upper limit of the number of generations. Convergence stability within 30 generations is then verified in multiple batch replays. If any convergence condition is met, iteration stops and the current optimal individual is output. The optimal individual is determined by the highest fitness and priority given to those meeting the stability criteria. If the fitness is the same, the one with the lower adjustment cost is selected.

[0045] After outputting the optimal individual, the module analyzes the optimal individual to generate a set of adjusted pressure parameters coupled with temperature changes. The analysis process first converts the segmented pressure target sequence into a pressure setting sequence that the controller can execute. The conversion method is to expand each segment of target pressure at a 1-second time granularity and generate a gradually changing set value sequence in the transition zone between segments according to the rate limit. The magnitude of each step of the gradual change does not exceed the upper boundary of the pressure adjustment step size. Then, the setting sequence is bound to the temperature over-limit window. The binding information includes the window start and end timestamps, the direction of temperature over-limit, the temperature deviation magnitude level, the typical lag time, and the corresponding temperature interval number and pressure interval number, so that subsequent execution modules can call the corresponding setting sequence within the same window. After binding is completed, a consistency check is performed. The check includes verifying whether the pressure setpoint falls within its upper and lower boundaries, whether the setpoint change rate falls within its rate boundary, and whether the setpoint sequence has sufficient coverage within the set of highly correlated pressure intervals. Coverage is determined by a minimum percentage of time spent within the window that is not less than 0.7. This 0.7 is based on the statistical result of the minimum sustained percentage of pressure falling within the highly correlated interval in qualified batches, taking the minimum value slightly higher than the lower bound of this statistical result to ensure that the coupling relationship dominates within the window. If boundary conflicts occur during the check, boundary pruning is performed on the conflicting segment, and the stability and adjustment cost indices of that segment are recalculated to ensure that the output set meets safety and process constraints and remains executable. Finally, an adjusted pressure parameter set is formed, consisting of the pressure setpoint sequence, rate limits within the segment, action intervals, applicable temperature range numbers, applicable process segment identifiers, and typical lag times.

[0046] The parameter optimization processing module obtains the adjusted pressure parameter set, performs secondary processing on the data range mapping and standardization transformation formulas, and incorporates a missing value imputation method to complete incomplete data, resulting in an optimized parameter pairing list for temperature and pressure dual-parameter coordinated control. This includes obtaining the adjusted pressure parameter set and constructing a data range mapping model; projecting the pressure parameter set to obtain the mapped pressure data sequence; standardizing the mapped pressure data sequence to generate a standardized pressure data stream; performing breakpoint interpolation on the standardized pressure data stream to generate a completed pressure data sequence; associating the completed pressure data sequence with temperature data to construct a temperature-pressure dual-parameter coupling matrix; and parsing the temperature-pressure dual-parameter coupling matrix to output the optimized parameter pairing list for temperature-pressure dual-parameter coordinated control.

[0047] In this embodiment, the adjusted pressure parameter set output by the temperature-pressure coupling regulation module is first received, and this set is split into calculable data fields and traceable constraint fields. The data fields include a pressure setting sequence arranged in ascending order of timestamps, intra-segment rate limits, action intervals, typical lag times, applicable process segment identifiers, and applicable temperature range numbers; the constraint fields include upper and lower boundaries of pressure setting values, rate boundaries, step size boundaries, and boundary clipping marks. The module performs time consistency checks on the pressure setting sequence: it reads the interval between two adjacent timestamps one by one. If the interval deviates from the sampling period by 1 second, it is re-aligned to uniform time grids according to a fixed alignment window of 5 seconds. The 1-second sampling period is determined based on the unified value of the acquisition end configuration and the controller refresh cycle, taking the smallest integer second that meets the online control refresh requirements; the 5-second alignment window is determined based on the historical upper bound statistical results of production network transmission jitter and data arrival delay, taking the smallest integer second that covers the upper bound, thereby ensuring that subsequent mapping and standardization operate at a unified time granularity. If consecutive missing markers are found, the missing segments are marked according to their length. The missing length is obtained by accumulating the number of seconds from the start point to the end point of the missing segment. The accumulation method is to count every second according to the time grid and record the start and end timestamps of the missing segments for use in subsequent interpolation stages.

[0048] After completing the basic verification, the module constructs a data range mapping model to project the pressure parameter set into a mapped pressure data sequence. The boundary of the mapping model is constrained by three types of boundaries and their intersection: the first type of boundary is the lower limit and upper limit of the equipment's allowable pressure, derived from the equipment's safety settings; the second type of boundary is the lower limit and upper limit of the process formula's pressure target, derived from the current process section's formula; and the third type of boundary is the lower limit and upper limit of the historical qualified batch pressure stability zone, derived from qualified batch statistics. The statistical process for determining the stable pressure band of historical qualified batches is as follows: Batches with consistent formulation versions and qualified quality are selected. Pressure measurement sequences are extracted within the same process segment. The deviation from the target center pressure of the formulation is calculated second by second. The deviation is calculated by subtracting the target center pressure from the second-by-second pressure value and taking the absolute value. The maximum deviation within the process segment is then taken as the upper bound of natural fluctuation. Subsequently, the target center pressure is subtracted from the upper bound of natural fluctuation to obtain the lower limit of the stable band. The target center pressure is added to the upper bound of natural fluctuation to obtain the upper limit of the stable band. This upper limit is then intersected with the equipment allowable boundary and the formulation target boundary to obtain the final stable band boundary. The mapping interval is set to 0 to 1. The determination of 0 to 1 is based on unifying the pressure scale of different batches, different equipment, and different process segments into a dimensionless interval, facilitating subsequent standardization, interpolation, and threshold management for matrix statistics. The projection process is executed point-by-point: For the pressure setpoint every second, it is first determined whether it falls within the final stable zone boundary; if it does, the pressure value is linearly projected according to the correspondence between the lower limit of the stable zone (mapped to 0) and the upper limit of the stable zone (mapped to 1). The linear projection is calculated by first calculating the relative position of the pressure value from the lower limit of the stable zone, and then converting it to the 0-1 range according to the total width of the stable zone; if it is lower than the lower limit of the stable zone, the mapped value is 0 and marked as the lower boundary point; if it is higher than the upper limit of the stable zone, the mapped value is 1 and marked as the upper boundary point. The module synchronously saves the original pressure value, mapped value, boundary point mark, and process section identifier of each mapped point, forming a sequence of mapped pressure data.

[0049] After obtaining the mapped pressure data sequence, the module performs a standardization transformation on the sequence to generate a standardized pressure data stream. The standardization transformation is performed separately within each process segment to avoid interference from scale differences across process segments. Standardization employs a combination of sliding window centering and scale normalization. The sliding window length is set to 60 seconds, determined based on the typical duration covered by the pressure response caused by temperature disturbances in most qualified batches, combined with the delay constraint of the online decision refresh cycle, taking the smallest integer second that covers the main dynamic process without introducing excessive statistical lag. For each window, the module counts valid samples point by point and calculates the center value: the center value is calculated by summing all valid mapped values ​​within the window and then dividing by the number of valid samples to obtain the average value; missing points are not included in the summation and counting, while boundary points are included in the summation and counted separately as boundary samples for subsequent reliability assessment. The module then calculates the scale value, which describes the fluctuation intensity within the window. The scale value is calculated by taking the absolute difference between each valid mapped value and the center value within the window, summing them, and then dividing by the number of valid samples to obtain the average absolute deviation. To prevent excessively small scale values ​​from amplifying the standardization results, a lower limit of 0.01 is set. This 0.01 is determined by a comprehensive lower bound statistical result of three factors: sensor resolution within the mapping interval 0 to 1, the minimum discernible pressure change caused by the valve's minimum movement, and the control noise amplitude. The smallest value above the noise lower bound is selected to ensure numerical stability. When the scale value is below 0.01, it is set to 0.01, and the window is marked as a low-fluctuation window. Standardization values ​​are generated point-by-point: for each second of mapping value, the center value of its corresponding window is subtracted, and then divided by the scale value of that window to obtain the standardized pressure value for that second. The module simultaneously saves the window center value and window scale value for that second, facilitating subsequent back-translation to an executable pressure scale.

[0050] After generating a standardized pressure data stream, the module performs interpolation to fill in any missing points, forming a completed pressure data sequence. Breakpoint identification is based on timestamp continuity and missing markers: the module scans the standardized pressure data stream second by second, records the starting timestamp of the missing marker, and continues scanning until the first non-missing timestamp is encountered as the ending timestamp. The missing length is calculated by accumulating the number of seconds between the starting and ending timestamps. Segments with a missing length less than or equal to 10 seconds are classified as short missing segments. The 10-second threshold is determined by the historical upper bound of the duration of common instantaneous communication interruptions in production, taking the smallest integer second that covers this upper bound to ensure interpolation is effective for short-term data loss scenarios. Segments with a missing length greater than 10 seconds are classified as long missing segments. Long missing segments are not interpolated; only the missing marker is retained and its weight is reduced in subsequent matrix statistics to avoid generating unreliable completed values ​​when information is insufficient. For short missing segments, endpoint constraint interpolation is performed: the module locates the first valid point before the missing segment as the front point and the last valid point after the missing segment as the back endpoint. If both the front and back endpoints exist simultaneously, a transition value is generated per second within the missing segment. The transition process progresses second by second, gradually approaching the back endpoint value from the front point value. The progress per second is constrained by a rate limit. The rate limit is the upper bound of the intra-segment rate limit output by the temperature-pressure coupling adjustment module, converted to a standardized scale. The conversion process involves first projecting the rate limit onto the 0-1 interval using a mapping model, and then converting it to the maximum permissible change per second in standardized units, thus ensuring that the interpolation sequence is consistent with the change rhythm achievable by the actuator. If only a single endpoint exists, the endpoint value is extended into the missing segment and marked as a single-sided interpolation point. Single-sided interpolation points are assigned a lower reliability weight in subsequent statistics. After completion, the module generates an interpolation type label and a reliability label for each completion point: two-sided interpolation points are labeled as two-sided interpolation, one-sided interpolation points are labeled as one-sided interpolation, and true sampling points are labeled as true points; the reliability label is graded according to the missing length and the size of the endpoint difference. The longer the missing length and the larger the endpoint difference, the lower the reliability level, which is used as a deduction item for subsequent pairing scoring.

[0051] After obtaining the completed pressure data sequence, the module correlates this sequence with the temperature data to construct a temperature-pressure dual-parameter coupling matrix. The temperature data is taken from the clean production environment temperature data stream output by the data acquisition and fusion module. The module first performs alignment correction based on typical lag times: the alignment correction is performed by advancing the temperature sequence relative to the pressure sequence by the number of seconds corresponding to the typical lag time, so that the temperature value at the same time point corresponds to the pressure value at the same response stage. The typical lag time is determined based on the weighted mode statistical result of the alignment lag in the preliminary parameter correlation matrix, with the weight taken as a comprehensive weight of mutual information strength and effective sample count, ensuring that the data-supported and more dependent units play a dominant role in the lag time. After alignment, the module bins the temperature data according to temperature interval numbers, which are taken from the interval division results of the parameter mapping table, ensuring that the temperature binning is consistent with the interval system of the upstream module. The pressure interval number is obtained from the segmentation results of the completed pressure data sequence within the mapping interval 0 to 1. The number of segments is set to 12. The determination of 12 is a compromise between the sample size of historical qualified batches, interval statistical stability, and online computing load. Too few segments will result in insufficient differentiation, while too many segments will result in an increase in sparse units. The middle integer of 12 is chosen to satisfy the unit sample support and resolution capabilities. The coupling matrix adopts a two-dimensional structure, with the row index being the temperature interval number and the column index being the pressure interval number. Matrix filling is performed by point-by-point accumulation: for every 1-second time point, the row corresponding to its temperature interval number and the column corresponding to its pressure interval number are located. The standardized pressure value of that second is accumulated to the corresponding unit according to the reliability weight, and the reliability weight is simultaneously accumulated to the total weight of that unit. The reliability weighting rules are as follows: true point weight 1, two-sided interpolation point weight 0.7, and one-sided interpolation point weight 0.5. The determination of 1, 0.7, and 0.5 is based on the sensitivity playback results of interpolation error to coupled statistics. The true point is used as the benchmark and is assigned a weight of 1. Two-sided interpolation has lower uncertainty due to constraints at both ends and is assigned a weight of 0.7. One-sided interpolation has higher uncertainty and is assigned a weight of 0.5, thereby suppressing the dominance of unreliable completion points on the unit representative value. The unit representative value is calculated after matrix filling. The calculation method is to multiply the standardized pressure value within the unit by the sum of the reliability weights and divide by the total number of reliability weights in the unit to obtain the weighted average representative value. The module also calculates the unit coverage. The coverage method is to divide the number of effective points in the unit by the total number of effective points in the temperature range, which is used to measure the statistical support strength of the unit. The coverage threshold is set to 0.2. The determination of 0.2 is based on the statistical result of the minimum sample ratio required to maintain the stability of the representative value of the unit in the playback of qualified batches, and the minimum value slightly higher than this lower bound is taken. When the coverage is lower than 0.2, the unit is marked as a sparse unit and is reduced in weight during the parsing stage to reduce the impact of sparse statistics on the pairing output.

[0052] When parsing the temperature-pressure dual-parameter coupling matrix and outputting the optimized parameter pairing list, the module processes the parameters row by row, using the temperature interval number as the main line. For each temperature interval number, the module traverses all corresponding pressure interval numbering cells, first filtering candidate pairs based on the control objective. The control objective originates from the combined constraint of the optimal pressure trajectory output by the temperature-pressure coupling regulation module and the process formulation pressure target band: the module extracts the target level of the optimal pressure trajectory within the temperature over-limit window, projects it onto the 0-1 interval through a mapping model, and then converts it to the standardized scale using the window center value and window scale value saved by the standardized transformation to form the target center; then, it combines the formulation pressure target bandwidth to form the target range. The upper and lower boundaries of the target range are obtained by taking the stricter boundary between the formulation target band and the historical stability band, so that the target range simultaneously satisfies the formulation requirements and historical stability constraints. The screening rule for candidate pairs is: when the cell representative value falls into the target range and the cell coverage is not less than 0.2, it enters the candidate set, where 0.2 is the aforementioned coverage threshold. A pairing score is calculated for the candidate set, consisting of three weighted components: Part 1 is the deviation of the representative value, calculated as the absolute difference between the unit's representative value and the target center; the smaller the absolute value, the higher the score. Part 2 is the coverage; higher coverage results in a higher score. Part 3 is the interpolation percentage, calculated as the number of interpolation points within the unit divided by the number of valid points in the unit; a higher interpolation percentage results in a higher deduction. The weights for these three components are 0.5, 0.3, and 0.2, determined based on the priority ranking of the control targets. This prioritizes ensuring the representative value is close to the target center, ensuring sufficient statistical support, and mitigating the uncertainty introduced by the interpolation percentage. Each temperature range is numbered, and the module selects the top three pressure ranges with the highest scores as output pairs. The weight of 3 is chosen because online execution requires reserving alternatives to handle real-time fluctuations, while limiting the number of alternatives to reduce the complexity of execution decisions. Each paired entry output includes the temperature range number, pressure range number, the corresponding mapped pressure range, the translated executable pressure setting suggestion, typical hysteresis time, coverage, interpolation reliability level, and pairing score. The translation process is performed point-by-point: first, the standardized representative value is multiplied by the corresponding window scale value and then added back to the window center value to restore the mapping space; then, it is back-projected to the original pressure scale through the mapping model, and clipped according to the upper and lower boundaries of the pressure setpoint to ensure that the output pressure suggestion is within the safe and process-allowed range. The final result is an optimized parameter pairing list for temperature and pressure dual-parameter coordinated control, sorted from highest to lowest by process segment identifier, temperature range number, and pairing score.

[0053] The performance simulation and verification module performs virtual simulation tests on the crystallinity changes in the microstructure of hard carbon materials based on an optimized parameter pairing list. It then uses a selection and elimination mechanism to filter the simulation results, determining whether the material performance meets preset standards under corresponding temperature-pressure coupled control conditions. The resulting performance verification dataset includes: obtaining the optimized parameter pairing list; extracting the temperature and pressure variables from the optimized parameter pairing list to construct a hard carbon microstructure evolution model; placing the hard carbon microstructure evolution model into a virtual simulation test environment to generate a crystallinity change trajectory spectrum; analyzing the crystallinity change trajectory spectrum and eliminating samples exceeding the structural stability limit to retain the valid simulation result set; and mapping the valid simulation result set to material performance index values. If the material performance index values ​​fall within the preset standard threshold range, the performance verification dataset is output.

[0054] In this implementation, the module first receives the optimized parameter pairing list output by the parameter optimization processing module and parses the list into a simulation input set and a filtering constraint set. The simulation input set includes the process section identifier, temperature range number, pressure range number, back-translated pressure setting suggestion, corresponding temperature target center value and allowable temperature fluctuation range, typical hysteresis duration, pairing score, coverage, and interpolation reliability level. The filtering constraint set includes structural stability limit constraints, preset standard threshold ranges for material properties, and simulation resource allocation limits. The module groups all pairing items according to the process section identifier and sorts them in descending order of pairing score within each group. The sorting process involves reading the pairing scores one by one and performing a descending sort. If the scores are the same, the pairing with higher coverage is prioritized, based on the principle that pairing items with more statistical support have lower uncertainty in the simulation. If the coverage is still the same, the pairing with higher interpolation reliability level is prioritized, based on the principle that more items supported by real sampling have less error propagation. To control simulation overhead, the module sets a maximum of 50 simulation entries for each process segment. The number 50 is determined by the product of the allowable computing resource budget for a single control cycle and the average time taken to simulate a single trajectory. The maximum is obtained by measuring the average time taken to simulate a single trajectory on the deployed hardware, dividing the allowable computing time by the average time, and rounding down. The maximum is then set to 50 after reserving redundancy for peak loads under abnormal fluctuation scenarios, ensuring that the system remains schedulable under high loads. Entries exceeding the maximum are truncated according to the sorting results and marked as unsimulated entries. Unsimulated entries are retained for supplementary evaluation in subsequent cycles.

[0055] After completing the item screening, the module extracts temperature and pressure variables and constructs a hard carbon microstructure evolution model. The extraction of temperature and pressure variables is based on time series boundary conditions: the temperature boundary conditions are generated by the target temperature center value corresponding to the temperature interval number and the allowable temperature fluctuation band. The target temperature center value comes from the current process formula, and the allowable temperature fluctuation band is formed by the constraints of the formula allowable deviation and the natural fluctuation band of historical qualified batches. The formation process is as follows: first, data with the same formula version and qualified quality are screened from qualified batches; then, the temperature sequence is extracted by process segment; the deviation of the temperature from the target center value is calculated second by second. The deviation is calculated by subtracting the target center value from the temperature value of that second and taking the absolute value; then, the maximum deviation within that process segment is taken as the upper limit of the natural fluctuation; the stricter boundary is taken compared with the formula allowable deviation as the allowable temperature fluctuation band. The strict boundary means the side with the smaller value, so that the temperature boundary conditions reflect the historical stability level. The pressure-side boundary conditions are generated from the back-translated pressure setting suggestions, which are derived from the inverse conversion results of the mapping space and the standardized space in the optimized parameter pairing list. These suggestions have been trimmed by upper and lower boundaries of the pressure setting values ​​to meet the equipment safety range and the formulation target range. To reflect the timing of temperature-pressure coupling in the simulation, the module introduces a typical lag duration for alignment. The typical lag duration is determined based on the weighted mode statistics of the alignment lag amounts of the corresponding temperature and pressure interval cells in the preliminary parameter correlation matrix. The weights are a combination of mutual information strength and effective sample counts, ensuring that the lag amount with stronger dependence and more sufficient samples dominates the alignment. Specifically, the alignment is performed by delaying the pressure boundary conditions relative to the temperature boundary conditions by the number of seconds corresponding to the typical lag duration on the simulation time axis. This ensures that the temperature disturbance is applied first, and the pressure response is reflected later. The resulting initial gap is filled by the initial pressure value and marked as a filled segment. The length of the filled segment is used to deduct the effective duration during subsequent index calculations, avoiding the initial value filling affecting the crystallinity evolution evaluation.

[0056] The initial state of the hard carbon microstructure evolution model is determined by the material baseline state. The process of determining the material baseline state is as follows: select qualified batches that are consistent with the current raw material batch, equipment station, and process formula version; extract the crystallinity level and its fluctuation range in the initial stage from the offline characterization data or historical verification data corresponding to these batches; take the median of the crystallinity level as the initial crystallinity; the basis for selecting the median is that it is not sensitive to the residual influence of abnormal batches; take the upper bound of the fluctuation range as the initial uncertainty band, which is used to constrain the initial microstate perturbation in the simulation environment. To improve the model's ability to cover actual fluctuations, the module generates an initial number of 20 perturbation samples for each temperature and pressure pairing item. The determination of 20 is based on the statistical results of the convergence speed of the simulation output variance as the number of samples changes and the constraints of the computational resource budget. By replaying the data, it was evaluated that the output variance decreased significantly when the number of samples increased from 10 to 20, while the benefit of further increasing the number of samples decreased. Therefore, the middle point between benefit and cost, 20, was chosen. Each perturbation sample is obtained by stratifying and taking values ​​according to the initial uncertainty band near the initial crystallinity. The number of stratifications is set to 5. The determination of 5 layers is based on reducing the sample sparsity caused by excessively fine stratification while ensuring coverage of the upper and lower boundaries of the initial band. The minimum distinguishable number of layers, 5, is chosen so that the perturbation samples can both cover the boundary and keep the computational load controllable.

[0057] After model construction is completed, the module places the model into a virtual simulation test environment and generates a crystallinity change trajectory spectrum. The virtual simulation test environment advances using a time-stepping method, with a time step size set to 1 second. The determination of 1 second is based on the consistency of the time granularity of upstream temperature and pressure control and data sampling, so that the simulation output and control input can be directly aligned at the same time grid. For each temperature and pressure pairing item, the module uses the temperature boundary condition sequence and pressure boundary condition sequence of that item as the driving force to advance the simulation second by second: every second, the temperature boundary value and pressure boundary value of that second are read. If the temperature boundary value of that second exceeds the temperature fluctuation range, the excess is recorded as a temperature disturbance input, and this disturbance input is used for subsequent result interpretation; every second, the pressure boundary value of that second is read and checked whether it falls within the pressure safety range. If it exceeds the limit, it is trimmed according to the boundary and the number of trimmings is recorded. The number of trimmings is used for subsequent elimination weight deduction to avoid boundary conditions that do not meet the constraints affecting the crystallinity judgment. During the simulation, the model outputs a corresponding crystallinity value every second. The module combines this crystallinity value with a timestamp to form a sequence and saves it. The sequence also records the temperature perturbation input, pressure trimming mark, and initial perturbation sample number for that second, thus forming a single trajectory. The crystallinity sequences generated by 20 initial perturbation samples under the same temperature and pressure pairing are summarized into a crystallinity change trajectory group for that pairing. Then, the trajectory groups of all pairings within the same process segment are summarized to form a crystallinity change trajectory spectrum. The trajectory spectrum is stored in sorted order by process segment identifier, temperature range number, pressure range number, and sample number to ensure quick location by index during subsequent screening.

[0058] After the trajectory spectrum is generated, the module analyzes the crystallinity change trajectory spectrum and executes a selection and elimination mechanism to remove samples that exceed the structural stability limit, retaining the valid simulation result set. The structural stability limit is defined by the upper limit of the crystallinity change rate and the upper limit of the crystallinity fluctuation amplitude. The upper limit of the crystallinity change rate is determined based on the statistical upper bound of the crystallinity evolution rate in qualified batch calibration data or historical verification data. The statistical process involves calculating the absolute amount of the crystallinity difference between adjacent 2-second intervals for the crystallinity time series of qualified samples, forming a change rate sequence, and then taking the maximum value of the change rate sequence as the upper limit. To suppress the influence of occasional noise, a smoothing constraint is introduced. The smoothing constraint is implemented by calculating the moving average of the change rate sequence with a 3-second window and then taking the maximum value. The 3-second window is determined based on the statistical results of the upper limit of the common duration of simulation output numerical jitter in qualified calibration data, taking the smallest integer second that covers this upper limit. Finally, this maximum moving average is taken as the upper limit of the crystallinity change rate, so that the upper limit reflects continuous change rather than a single point spike. The upper limit of crystallinity fluctuation is determined based on the statistical upper limit of crystallinity fluctuation range in the stable phase of qualified batches. The statistical process involves locating the time period in the qualified samples when crystallinity enters the stable range. The stable range is located based on the duration of crystallinity change rate continuously below the upper limit of change rate for 30 seconds. The 30 seconds is determined based on the statistical results of the shortest common length of the stable segment duration in qualified samples, and the smallest integer seconds covering this length are taken. Within the stable range, the difference between the maximum and minimum crystallinity values ​​is calculated as the fluctuation range, and the maximum value of the fluctuation range of all qualified samples is taken as the upper limit of the fluctuation range, so that the upper limit is consistent with the actual qualified samples. After determining the upper limits for the two categories, a second-by-second screening is performed on each simulated crystallinity trajectory: the absolute value of the difference between the crystallinity of the current second and the crystallinity of the previous second is calculated second by second. If this absolute value exceeds the upper limit of the crystallinity change rate, it is marked as a rate over-limit point. At the same time, within a sliding window of 60 seconds, the difference between the maximum and minimum crystallinity values ​​in that window is calculated as the window fluctuation amplitude. The 60-second period is determined based on the statistical results of the effective observation time of the impact of crystallinity fluctuation on material properties within the process segment, and is consistent with the upstream standardized window length to ensure consistent interpretation across modules. If the window fluctuation amplitude exceeds the upper limit of the crystallinity fluctuation amplitude, it is marked as an amplitude over-limit window. To avoid trajectory elimination due to misjudgment of a single point or a single window, the module introduces a continuous judgment rule. The continuous judgment length is set to 3 consecutive time steps. The determination of 3 is based on the upper bound of the common continuous length of short-term jitter in the qualified calibration data of the simulation output. The rule is executed by marking the trajectory as an unstable sample and eliminating it only when the rate over-limit point occurs 3 times consecutively or the amplitude over-limit window is triggered 3 times consecutively. Trajectories that do not trigger continuous judgment are included in the valid simulation result set.The module synchronously records the reasons for elimination. The reasons for elimination are classified into three categories: rate exceeding the limit, amplitude exceeding the limit, and excessive boundary clipping. The threshold for judging excessive boundary clipping is set to more than 10 clipping times. The determination of 10 times is based on the upper limit statistical results of the frequency of boundary clipping in qualified playback. The smallest integer number slightly higher than the upper limit is taken to prevent long-term exceeding of the limit from being mistakenly regarded as effective control.

[0059] After the effective simulation result set is generated, the module maps the effective simulation results to material performance index values ​​and determines whether to output a performance verification dataset based on these values. The material performance indexes include at least four items: final-state crystallinity, average crystallinity in the stable segment, fluctuation amplitude in the stable segment, and time required to reach the stable segment. Final-state crystallinity is calculated by taking the crystallinity value of the last second of the trajectory. The location of the stable segment is still based on the crystallinity change rate being continuously lower than the upper limit for 30 seconds, with 30 seconds being the aforementioned stable segment location duration. The average crystallinity in the stable segment is calculated by summing the crystallinity values ​​per second within the stable segment and dividing by the effective number of seconds in the stable segment. The fluctuation amplitude in the stable segment is calculated by subtracting the minimum crystallinity value from the maximum crystallinity value within the stable segment. The time required to reach the stable segment is calculated by accumulating the difference in seconds between the timestamp from the trajectory start point to the first time the stable segment location conditions are met, deducting the length of the initial pressure fill segment mentioned above to avoid the fill segment affecting the evaluation. For the 20 perturbation sample trajectories of the same temperature-pressure pairing item, the module summarizes each indicator to obtain an item-level indicator. The summarization method uses the median as the representative value, and the upper and lower dispersion ranges of the representative value are used to describe the uncertainty. The median is selected based on its insensitivity to a small number of abnormal trajectories. The preset standard threshold range for material performance is jointly determined by the target performance requirements and historical qualified boundaries: the target performance requirements are derived from product specifications or process control targets, and the historical qualified boundaries are derived from the statistical intervals of qualified batch indicators. The formation method is to calculate the same type of indicator for qualified batches and take their minimum and maximum values ​​as historical boundaries, and then take the stricter boundary corresponding to the upper and lower limits of the target performance requirements as the preset standard threshold range, so that the threshold meets the specification target without deviating from the verified achievable range. The judgment process is performed item by item: if the final crystallinity falls within the preset range, the average crystallinity of the stable segment falls within the preset range, the fluctuation amplitude of the stable segment does not exceed the preset upper limit, and the time required to reach the stable segment does not exceed the preset upper limit, then the temperature-pressure pairing item is judged to be passed; if any indicator exceeds the limit, it is judged to be failed, and the exceeded item and the exceeded amplitude are recorded as the reason for failure. For items deemed passed, the module summarizes their temperature range number, pressure range number, translated pressure setting suggestion, typical lag time, representative value of item-level index, uncertainty description, pass mark, and corresponding effective simulation sample ratio. The effective simulation sample ratio is calculated by dividing the number of effective trajectories by 20. This ratio is used by the subsequent feedback update module to adjust the weights of historical constraints. For items deemed failed, the module also outputs a summary of temperature and pressure variables and the reason for failure, but these are not included in the main set of the performance verification dataset; they are only output as an adjunct to the constraint record. The final performance verification dataset is then generated and sorted by process section identifier and paired score.

[0060] The feedback update module, based on the performance verification dataset, uses a feedback loop mechanism to dynamically update the parameter correlation matrix, so that the adjustment result of the pressure parameter has a reverse effect on the temperature-pressure coupling relationship model. It processes historical data constraints and parameter adjustment step sizes in the verification data to determine the refined temperature-pressure dual-parameter coupled control scheme. This includes acquiring the performance verification dataset and extracting the dominant deviation features to generate a multidimensional feedback deviation dataset; updating the parameter correlation matrix using the multidimensional feedback deviation dataset and extracting the sensitivity factor of the pressure parameter to microstructure evolution; inputting the sensitivity factor into the temperature-pressure coupling relationship model and calculating the restricted parameter adjustment increment in conjunction with historical data constraints; and superimposing the restricted parameter adjustment increment onto the baseline control curve to determine the refined temperature-pressure dual-parameter coupled control scheme.

[0061] In this implementation, the performance verification dataset output by the performance simulation and verification module is first received, and the dataset undergoes field integrity verification and index reconstruction. Field integrity verification is performed record by record, and the verification content includes process section identifier, temperature range number, pressure range number, back-translated pressure setting suggestion, typical lag time, representative value of item-level crystallinity index, uncertainty range of item-level crystallinity index, pass mark, reason for failure, proportion of effective simulation samples, paired score, and the number of times records were pruned and the instability elimination count during the simulation phase. If a record lacks a key field, the record is marked as unusable and removed from the update queue for this round, and the reason for the missing field is recorded for upstream tracing. The index reconstruction uses the process segment identifier and temperature range number as the primary key and the pressure range number as the secondary key to build a multi-level index, so that each verification record can be located in a unique row and column position of the parameter association matrix. The location process is as follows: first, the process segment identifier is used to locate the matrix blocks, then the temperature range number is used to locate the row, and the pressure range number is used to locate the column. The effective sample count of the row and column in the matrix is ​​checked to see if it is greater than 0. If it is 0, the record is marked as a sparse update candidate and its update weight is reduced. The basis for reducing the update weight is that the historical statistics of the unit are insufficient.

[0062] After indexing, the module extracts dominant deviation features and generates a multidimensional feedback deviation dataset. The extraction of dominant deviation features revolves around two main lines: performance deviation and control deviation. Performance deviation is calculated using a preset standard threshold range as a reference. This preset standard threshold range is jointly determined by the target performance requirements and historical acceptable boundaries. The target performance requirements are derived from product specifications or process control targets, while the historical acceptable boundaries are derived from the statistical intervals of acceptable batch indicators. The final threshold is the more stringent boundary between the two, meaning the narrower range, making the threshold closer to the verified achievable and specification-compliant interval. For each verification record, the module calculates the final crystallinity deviation, the mean deviation of the stable segment, the fluctuation amplitude deviation of the stable segment, and the time deviation required to reach the stable segment. The final crystallinity deviation is calculated as follows: if the final crystallinity falls within the threshold range, the deviation is recorded as 0; if it is below the lower threshold limit, the deviation is recorded as the difference between the lower threshold limit and the final crystallinity and marked as low deviation; if it is above the upper threshold limit, the deviation is recorded as the difference between the final crystallinity and the upper threshold limit and marked as high deviation. The mean deviation of the stable segment uses the same calculation rules as the final deviation. The calculation method for the fluctuation amplitude deviation during the stable period is as follows: if the fluctuation amplitude is not higher than the upper limit threshold, the deviation is recorded as 0; if it is higher than the upper limit threshold, the deviation is recorded as the difference between the fluctuation amplitude and the upper limit threshold and marked as over-amplitude. The calculation method for the time deviation required to reach the stable period is as follows: if the stabilization time does not exceed the upper limit threshold of the stabilization time, the deviation is recorded as 0; if it exceeds the upper limit threshold, the deviation is recorded as the difference between the stabilization time and the upper limit threshold and marked as timeout. The upper limit thresholds for fluctuation amplitude and stabilization time are determined based on the joint constraints of the target performance requirements and the upper bound of historical qualified statistics, taking the stricter upper bound. Stricter means the smaller value, so that the constraints reflect a higher requirement for stability.

[0063] The control deviation is referenced to the baseline control curve, which is the pressure setting sequence output from the previous refining scheme. If the system is in the initial operation phase, the baseline control curve is the median trajectory of the pressure setting sequence of historical qualified batches under the same process section and temperature range number. The median trajectory is determined by taking the median of the pressure setting value at the same time point among qualified batches to form a sequence. The median is selected based on the principle of insensitivity to a small number of abnormal batches. The module calculates the pressure setting offset for each verification record. The pressure setting offset is calculated by subtracting the translated pressure setting suggestion of the record from the baseline pressure setting value under the same temperature range number point by point to obtain the difference sequence. The average absolute value of the difference sequence within the over-limit window is taken as the offset characterization value. The average absolute value is calculated by taking the absolute value of the difference per second within the window, summing them, and then dividing by the number of effective seconds within the window. The number of effective seconds within the window is reduced by the length of the filled and missing segments of the initial pressure value recorded in the simulation to prevent non-realistic adjustment ranges from affecting the offset statistics. The module simultaneously calculates the adjustment cost deviation, which is based on the number of actions and the cumulative action amplitude. The number of actions is obtained by counting the number of non-zero steps in the pressure setting sequence, and the cumulative action amplitude is obtained by accumulating the absolute change of each step. The recorded number of actions and cumulative action amplitude are compared with the corresponding statistical values ​​of the benchmark control curve to obtain the difference in the number of actions and the difference in amplitude, which are used for the cost constraints in subsequent limited increment calculations.

[0064] After calculating performance and control deviations, the module generates a multidimensional feedback deviation dataset and calculates and updates the confidence weight for each record. The updated confidence weight is determined by three factors: the proportion of effective simulation samples, the paired score, and the label. The proportion of effective simulation samples is the ratio of the number of effective trajectories to the initial number of disturbance samples in the simulation (20), where 20 is the fixed number of samples during the simulation phase. The paired score comes from the list of optimized parameter pairs. The label is the result of determining whether the preset standard threshold range is met. The weight calculation adopts a segmented assignment rule to ensure interpretability: when a sample passes and the proportion of valid simulation samples is not less than 0.8, the weight is 1. The determination of 0.8 is based on the statistical result of the lower bound of stability corresponding to the majority of valid samples under the elimination mechanism in the simulation stage, and the smallest value covering this lower bound is taken; when a sample passes and the proportion of valid simulation samples is less than 0.8, the weight is 0.7 to reflect insufficient sample stability; when a sample passes and the deviation is less than or equal to 0.2 of the threshold bandwidth, the weight is 0.5. The determination of 0.2 is based on the fact that small deviations near the qualified boundary have reference significance for the correction direction but should not dominate the update; when a sample passes and the deviation is greater than 0.2 of the threshold bandwidth, the weight is 0.3 to avoid severe deviation records from exerting too strong a pull on the matrix. Paired scores are used to refine and correct the weights mentioned above. The correction method is to normalize the paired scores to the range of 0 to 1 and then use them as multipliers to add to the weights. The normalization process is to subtract the minimum score within the current process segment from the score and then divide by the score range. The score range is the maximum value minus the minimum value. When the score range is 0, it is treated as 1 to avoid division by zero. Finally, a multidimensional feedback deviation dataset is obtained.

[0065] When updating the parameter correlation matrix using the multidimensional feedback deviation dataset, the module processes each process segment in blocks and updates row by row within each block according to the temperature range number. For each matrix unit, the module extracts the corresponding set of feedback records, which contains all verification records for the same temperature range number and pressure range number. The module first calculates the unit update direction, which is determined based on the sign of the performance deviation: if the majority of records are passed, the update direction is enhancement; if the majority of records are failed and the final crystallinity deviation is low, the update direction is suppression and marked as pressure direction sensitivity; if the majority of records are failed and the final crystallinity deviation is high, the update direction is suppression and marked as pressure direction sensitivity; if the main problem is excessive fluctuation amplitude or stabilization timeout, the update direction is suppression and marked as stability risk. The majority decision uses a weighted counting method, with the counting weight taken as the update confidence weight. The class with the higher accumulated weight is the majority class, so that high confidence records dominate the direction judgment. After determining the direction, the unit update amount is calculated. The update amount is increased or decreased based on the original matrix dependency strength, and the increase or decrease is determined by the weighted deviation strength. The weighted bias strength is calculated by summing the bias amplitudes of all records in the unit using weights and then dividing by the total weights to obtain the weighted average bias amplitude. The bias amplitude is taken as the dominant term among four items: final crystallinity bias, mean bias of stable segment, fluctuation amplitude bias, and stable time bias. The rule for determining the dominant term is to convert the four bias amplitudes to the relative scale of their respective threshold bandwidths and take the largest one. The conversion method is to divide the bias amplitude by the corresponding threshold bandwidth, where the threshold bandwidth is the upper limit minus the lower limit or the upper limit itself. The largest value after conversion is taken to avoid distortion in the selection of the dominant term due to unit differences. To prevent matrix mutations caused by excessively large single updates, the upper limit of the update step size is set to 0.05. The basis for determining 0.05 is the upper bound of the natural fluctuation of the matrix dependency strength in adjacent statistical periods in qualified replay data, and the smallest value slightly higher than this upper bound is taken to ensure that the update amplitude remains gradual even when it exceeds the natural fluctuation. When the absolute value of the update exceeds 0.05, it is pruned by 0.05, and the pruning mark is recorded for the next round of weighted decay. After the update is completed, the updated parameter correlation matrix is ​​obtained. At the same time, the reliability label of each unit is updated. The reliability label is determined by the sum of the weights of the units participating in the update and the effective sample count. The larger the sum of weights and the higher the effective sample count, the higher the reliability. This is used for subsequent sensitivity factor weighting.

[0066] While updating the matrix, the module extracts the sensitivity factor of pressure parameters to microstructure evolution. The sensitivity factor is calculated based on the response intensity between the pressure setting offset and crystallinity deviation, and is performed within the same temperature range number to avoid the inclusion of temperature differences. Specifically, under the same temperature range number, verification records from different pressure range numbers are collected and sorted by pressure setting offset from smallest to largest. For adjacent pairs of records, the offset difference and deviation difference are calculated. The offset difference is the absolute value of the difference between the pressure setting offsets of the two records, and the deviation difference is the absolute value of the difference between the magnitudes of the dominant performance deviations of the two records. The deviation difference is divided by the offset difference to obtain the response intensity value. A weighted average of the response intensity values ​​of all adjacent pairs is taken as the sensitivity factor for that temperature range. The weighting weight is the average of the updated confidence weights of the two records, ensuring that reliable records dominate the sensitivity. To avoid abnormally amplified response intensity due to excessively small offset differences, a lower limit for the offset difference is set to twice the pressure adjustment step size boundary. This twice-value is determined by the cumulative lower limit of the minimum discernible change in a single action. Adjacent pairs with offset differences below this lower limit are not included in the calculation. The sensitivity factor output is accompanied by a sensitivity reliability score. The sensitivity reliability score is determined by the sum of the number of adjacent pairs involved in the calculation and their weights. If the number of adjacent pairs is less than 3, the sensitivity reliability score is marked as low. The score of 3 is determined by the minimum number of adjacent pairs that can form a trend judgment.

[0067] When the sensitivity factor is input into the temperature-pressure coupling model, the module binds the sensitivity factor to the corresponding node in the model according to the process section and temperature range number, and uses the updated matrix dependency strength as the priority for pressure range selection. The binding process involves setting a sensitivity influence weight for each temperature range number in the model. The sensitivity influence weight varies with the sensitivity reliability: a weight of 1 for high sensitivity reliability, 0.7 for medium sensitivity reliability, and 0.5 for low sensitivity reliability. The determination of 1, 0.7, and 0.5 is based on the risk control principle of conservatively using uncertain sensitivity, ensuring that the model relies more on matrix statistical dependency when the sensitivity is unstable. Subsequently, the module calculates the limit parameter adjustment increment based on historical data constraints. The historical data constraints consist of three parts: the historical qualified batch pressure stability zone, the equipment safety boundary, and the parameter adjustment step size boundary. The historical qualified batch pressure stability zone and the equipment safety boundary are used to limit the allowable range of the target pressure, while the parameter adjustment step size boundary is used to limit the maximum magnitude of a single increment. The calculation of the adjustment increment of the restricted parameter is performed under each temperature range number: The module first determines the adjustment direction, which is determined based on the performance deviation direction and the sign of the sensitivity factor. If the final crystallinity deviation is low and the sensitivity factor indicates that increasing the pressure is beneficial to reducing the deviation, then the direction is to adjust upward; if the final crystallinity deviation is high and the sensitivity factor indicates that decreasing the pressure is beneficial to reducing the deviation, then the direction is to adjust downward; if the dominant problem is excessive fluctuation amplitude or timeout of stabilization, then the direction is to reduce the fluctuation and prioritize reducing the action amplitude. The adjustment range is then calculated. The initial increment is obtained by multiplying the ratio of the dominant deviation to the threshold bandwidth by a conservative amount equal to the reciprocal of the sensitivity factor. The conservative amount means that the sensitivity factor is reduced according to reliability, with a reduction rule of 0.7 for high-reliability sensitivity and 0.5 for low-reliability sensitivity, making the increment smaller when the sensitivity is unstable. The initial increment is then trimmed according to the parameter adjustment step size boundary. The upper limit of the trimming is taken as the step size boundary, which is derived from the pressure change corresponding to the minimum resolvable action of the actuator and the upper limit of the executable step size output by the upstream adjustment module. The stricter value means the smaller value, ensuring that the increment does not exceed the executable range. To suppress excessively dense actions caused by frequent fine-tuning, the module sets the increment trigger threshold to a dominant deviation ratio of not less than 0.05. The determination of 0.05 is based on the statistical results of the measurement and simulation error bandwidth lower bound of the performance index in qualified playback. Deviations below this ratio may come from error propagation. The trigger threshold is used to avoid adjusting deviations within the error band. When the deviation ratio is less than 0.05, the restricted parameter adjustment increment in this temperature range is 0 and recorded as an adjustment-free range.

[0068] After obtaining the incremental adjustment of the restricted parameters, the module superimposes this increment onto the baseline control curve to form a refined temperature and pressure dual-parameter coupled control scheme. Before superposition, the module first segments the baseline control curve, based on the process segment identifier and temperature range number; within each segment, the temperature disturbance window is aligned according to the typical lag time, so that the time position of the increment is consistent with the temperature disturbance stage. The superposition process is executed point by point with a 1-second time granularity: for each time point, the baseline pressure setpoint is read, the incremental adjustment of the restricted parameters in the temperature range to which the time point belongs is read, and the two are added together to obtain the updated pressure setpoint; after addition, a double boundary check is performed, first according to the equipment safety boundary, and then according to the process formula pressure target band. When the cut occurs, the cut direction and cut amplitude are recorded. The cut amplitude is used for the next round to reduce the update confidence weight of the range by 0.7. The basis for determining 0.7 is that the boundary restricted range has a high risk to update stability in historical playback, and it is necessary to reduce its pull on the update. After the overlay is completed, the module calculates the number of actions and the cumulative action amplitude for the updated pressure setting sequence, and compares it with the upper limit of the number of actions per unit time given in the maintenance plan. If it exceeds the upper limit, the increment is scaled by 0.8 and the number of actions is recalculated. The 0.8 is determined by a conservative scaling factor that rapidly reduces the action density without changing the direction. If it still exceeds the limit after scaling, the increment is further trimmed to meet the upper limit and recorded as action-limited. Finally, a refined temperature and pressure dual-parameter coupled control scheme is output. The scheme includes the updated pressure setting sequence, pressure interval selection priority by temperature interval number, typical lag time, sensitivity factor and its reliability, limited parameter adjustment increment, adjustment-free interval marker, boundary trimming marker, action-limited marker, and a summary of this round of updates.

[0069] The execution decision module, when the refined temperature-pressure dual-parameter coupled control scheme shows that it can make the pressure control index better than the historical average under temperature fluctuation conditions, integrates production efficiency indicators through data aggregation technology, and handles the real-time processing delay issue to generate a comprehensive benefit evaluation result. Based on this, the final optimized path of hard carbon material preparation process is determined, resulting in a precise temperature-pressure dual-parameter coupled control execution scheme for actual production. This includes acquiring pressure values ​​under temperature fluctuation conditions; if the pressure values ​​are better than the historical average, collecting hard carbon material production rate and energy consumption data; calculating the real-time processing delay of hard carbon material production rate and energy consumption data, and using the real-time processing delay to perform time compensation correction on the aggregated data; inputting the corrected data into a multi-dimensional benefit analysis model to obtain a comprehensive benefit evaluation result; determining the optimized path of hard carbon material preparation process based on the comprehensive benefit evaluation result; and analyzing the optimized path of hard carbon material preparation process to generate a precise temperature-pressure dual-parameter coupled control execution scheme.

[0070] In this implementation, the refined temperature and pressure dual-parameter coupled control scheme output by the feedback update module is first received, and the scheme is unpacked into an execution input set, a judgment threshold set, and a data alignment rule set. The execution input set includes process section identifier, temperature range number, pressure setting sequence, pressure range selection priority, typical lag time, sensitivity reliability identifier, boundary trimming mark, and action restriction mark; the judgment threshold set includes pressure control index comparison threshold, production efficiency index acquisition window length, and real-time processing delay compensation rules; the data alignment rule set includes sampling period, alignment window, and timestamp reliability level. The module first performs a consistency check on the pressure setting sequence. The check method is to check whether the interval between adjacent timestamps is a sampling period of 1 second. The determination of 1 second is based on the unified value of the acquisition end configuration and the controller refresh cycle. If the interval is not found to be 1 second, time grid resampling is performed according to the alignment window of 5 seconds. The determination of the alignment window of 5 seconds is based on the historical upper bound statistical results of production network transmission jitter and data arrival delay, and the smallest integer second covering the upper bound is taken, so that subsequent index calculations are performed on a unified time scale. After alignment, the module establishes an index according to the process section identifier and temperature range number, so that the pressure setpoint, actual pressure value, actual temperature value, output data and energy consumption data at each time point can be mapped to the same evaluation window.

[0071] The module acquires pressure values ​​in real time under temperature fluctuation conditions during production operation. The pressure values ​​consist of two sequences: measured pressure and set pressure. Temperature fluctuation conditions are determined by the deviation of the measured temperature from the target center temperature. The target center temperature is derived from the process formula. The temperature fluctuation threshold is determined jointly by the allowable deviation of the formula and the upper limit of natural fluctuations in historical qualified batches. The determination process involves extracting temperature sequences from qualified batches by process segment, calculating the deviation from the target center temperature second by second, and taking the maximum value as the upper limit of natural fluctuations. This is then compared with the allowable deviation of the formula, and the stricter boundary is taken as the temperature fluctuation threshold. The side with the smaller strict boundary index value makes the triggering condition closer to the stable operating level. The module calculates the deviation second by second. When the deviation exceeds the temperature fluctuation threshold, it is determined as the start of a temperature fluctuation, and the time period continuously exceeding the threshold is merged into a temperature fluctuation window. The window terminates when the deviation returns to the threshold range and remains stable for 10 seconds. The 10 seconds are determined based on the statistical results of the lower limit of the time required for pressure regulation to complete and enter the stable fluctuation zone. The smallest integer seconds covering this lower limit are taken to avoid short-term drops causing the window to be fragmented. To ensure complete coverage of the pressure response, the module extends the temperature fluctuation window based on the typical lag duration. The extension method is to extend the window termination time by the number of seconds corresponding to the typical lag duration, so that the window includes the tail end of the pressure response. The typical lag duration is derived from the weighted mode statistics of the aligned lag amount in the parameter correlation matrix. The weights are taken from the mutual information strength and the effective sample count, so that the lag amount with stronger dependence and more sufficient samples dominates the extension.

[0072] After determining the temperature fluctuation window, the module calculates pressure control indicators and compares them with historical averages, serving as the criteria for triggering subsequent efficiency data collection. Pressure control indicators include pressure deviation, pressure fluctuation, and pressure stabilization time. The pressure deviation is calculated second-by-second within the window. The calculation method involves subtracting the pressure setpoint from the measured pressure value per second, taking the absolute value to form a deviation sequence, and then averaging the deviation sequence to obtain the window average deviation. The averaging method involves summing the deviations per second within the window and dividing by the number of valid seconds within the window. Missing points and empty points introduced by alignment resampling are removed from the valid seconds within the window, and the number of removed points is recorded for reliability identification. The pressure fluctuation indicator is calculated by taking the difference between the maximum and minimum measured pressure values ​​within the window to obtain the fluctuation amplitude, and simultaneously calculating the duration of the fluctuation period. The duration is used to distinguish between short spikes and continuous fluctuations. The pressure stabilization time index is calculated by locating the point in time when the pressure returns from the deviation state to the target pressure zone within the window. The target pressure zone is determined by the combined constraints of the formula pressure target zone and the historical qualified batch pressure stability zone. The determination process involves statistically analyzing the upper limit of the pressure deviation relative to the target pressure in qualified batches according to process sections to form a stability zone, and then taking the intersection with the formula target zone to obtain the target pressure zone. The stabilization judgment requires that the measured pressure value remain stable for 10 seconds after entering the target pressure zone. The 10 seconds are determined based on the statistical results of the minimum duration of stability judgment, so that the stabilization time reflects the true stability rather than an instantaneous crossing. The stabilization time is the cumulative number of seconds from the start of the window to the moment when the stability judgment is first met.

[0073] Historical averages are constructed by grouping processes according to process segment identifiers and temperature range numbers. Historical samples are derived from batches with consistent formula versions and qualified quality assessments. For each group, the module first constructs the same temperature fluctuation window identification rules as the current one, locating the corresponding window set in the historical data. Then, for each window, the average deviation, fluctuation amplitude, and stabilization time are calculated using the same method, and the average of the window sets is calculated to obtain the historical average benchmark. To suppress the impact of occasional abnormal windows on the historical average, the module first removes windows whose average deviation exceeds twice the average value within the group before averaging. The determination of twice is based on the outlier upper bound statistical results of the pressure index distribution in qualified batches, taking the smallest multiple that can remove extreme anomalies without losing coverage of normal fluctuations. After completing the historical average benchmark, the module compares and judges the current window: if the current window's average deviation is less than the historical average, the current window's fluctuation amplitude is less than the historical average, and the current window's stabilization time is not greater than the historical average, then the pressure control index is judged to be better than the historical average level, and production efficiency index collection is triggered; if any condition is not met, the reason for not triggering is recorded and the efficiency evaluation process for this window ends.

[0074] After triggering the collection of production efficiency indicators, the module collects hard carbon material production rate and energy consumption data, and performs timestamp unification and window alignment on both types of data. The generation of production rate data is based on the metering cycle, which is set to 60 seconds. The determination of 60 seconds is based on the minimum stable metering cycle of the output metering system and the data noise suppression requirements. If the cycle is too short, the metering noise will be too large, and if the cycle is too long, the response will be lagging. The module reads the cumulative value of output mass or output volume every 60 seconds, and subtracts the previous cumulative value from the current cumulative value to obtain the output of the cycle. Then, it divides by 60 seconds to obtain the production rate. The division is implemented by converting the output to the output per unit time to obtain the output per second and keeping it to 3 decimal places for subsequent aggregation. Energy consumption data is also collected in 60-second metering cycles. The module reads the cumulative meter reading of electricity or gas, subtracts the previous reading from the current reading to obtain the energy consumption for that cycle, and then divides by 60 seconds to obtain the energy consumption per second. The reason for using the same cycle for output and energy consumption is to maintain consistency in time scale in multidimensional benefit analysis and reduce interpolation and alignment errors. For each output rate and energy consumption data, the module records the generation timestamp, arrival timestamp, and processing completion timestamp. The generation timestamp comes from the internal clock of the metering system. If there is a deviation between the internal clock and the control system clock, the time synchronization module is used to calibrate the clock. If the calibration deviation exceeds 1 second, the metering source is marked as a low-confidence source. The determination of 1 second is based on the control system time granularity and alignment window constraints. A deviation exceeding 1 second will significantly affect the window assignment judgment.

[0075] The module then calculates the real-time processing latency of the output rate and energy consumption data, and uses this latency to perform time compensation correction on the aggregated data. The acquisition latency is calculated by subtracting the generation timestamp from the arrival timestamp to obtain a difference in seconds; the processing latency is calculated by subtracting the arrival timestamp from the processing completion timestamp to obtain a difference in seconds; the total real-time processing latency is the sum of the acquisition latency and the processing latency. The module calculates the total latency for each data point and takes a representative latency value within the same temperature fluctuation window. The median is chosen as the representative latency value, based on the principle that a few extreme delays caused by network jitter should not dominate compensation. When there are fewer than 5 samples in the window, the average value is used instead. The determination of 5 samples is based on the lower bound of the minimum sample size required for stable median estimation. The time compensation correction is performed by shifting the effective timestamp of the data forward by the number of seconds corresponding to the total real-time processing delay, so that the data returns to its actual occurrence time on the timeline. If the shift crosses the temperature fluctuation window boundary, it is truncated according to the nearest boundary and marked as a boundary compensation point. The boundary compensation point is assigned a low weight of 0.7 in subsequent aggregation. The basis for determining 0.7 is that the boundary truncation causes the point to decrease its representativeness to the window, and it is necessary to reduce the weight to suppress its impact.

[0076] After time compensation, the module performs data aggregation on the corrected data and inputs the aggregation results into a multidimensional benefit analysis model to obtain a comprehensive benefit assessment result. The aggregation uses an assessment window of 300 seconds. The 300-second timeframe is determined based on the statistical results of the shortest response time that output and energy consumption have an observable impact on changes in control strategy after the improvement of pressure control indicators, while also considering decision-making response speed, taking the smallest integer second covering this response time. The assessment window is aligned with the starting point of the temperature fluctuation window and scrolls backward, with each scroll step being 60 seconds. The 60-second timeframe is determined based on the consistency of the measurement cycles for output and energy consumption. For each assessment window, the module calculates the pressure benefit component, output benefit component, and energy consumption benefit component respectively. The pressure benefit component consists of the improvement in pressure deviation, pressure fluctuation, and stabilization time. The improvement is calculated by subtracting the current window indicator from the historical average; a positive difference indicates improvement. The output benefit component is obtained by a weighted average of the output rates within the window. The weights of the weighted average are based on the data reliability level and boundary compensation point markings: true points have a weight of 1, and boundary compensation points have a weight of 0.7. The energy consumption benefit component is obtained by a weighted average of energy consumption within the window, and the overall evaluation adopts a directional constraint that lower energy consumption is better. To make different components comparable, the module performs a unified scaling transformation on the three types of components: maintaining a historical statistical range within the same process section, determined by the minimum and maximum values ​​of historical qualified batches for that component; mapping the current component to a 0-1 interval, where the mapping method is to subtract the historical minimum value from the current value and then divide by the historical range, where the historical range is the maximum value minus the minimum value; if the historical range is 0, it is treated as 1 to avoid division by zero. After scaling, the multidimensional benefit analysis model generates a comprehensive benefit assessment result by combining weights: pressure benefit 0.5, output benefit 0.3, and energy consumption benefit 0.2. The weights 0.5, 0.3, and 0.2 are determined based on the principle that stability and quality take precedence over efficiency in hard carbon material preparation, while also considering output and energy consumption. The model also sets a safety-priority constraint: if the pressure recovery time exceeds 1.2 times the historical average, the model is directly judged as failing and the comprehensive benefit result is invalidated. The 1.2 times threshold is determined based on the upper bound statistical result of recovery time fluctuations in qualified batches, taking the smallest multiple slightly higher than this upper bound to identify significant degradation. The comprehensive benefit assessment result is output as a window-level score, accompanied by sub-scores for the three components and validity markers.

[0077] The optimized path for hard carbon material preparation process was determined based on the comprehensive benefit assessment results. The optimized path was generated segment by segment: for each segment, the temperature and pressure interval numbers corresponding to the top three evaluation windows with the highest comprehensive benefit scores were selected. The number 3 was determined to retain both the primary and alternative schemes to address subsequent changes in operating conditions, while limiting the number of alternatives to reduce execution complexity. If the difference between the highest and second-highest score window was less than 0.05, the corresponding combination of two windows was retained, and switching conditions were set in the execution plan. The 0.05 threshold was determined based on the statistical results of the upper bound of the natural fluctuation of the comprehensive benefit score in the playback of qualified batches; a difference below this threshold was considered insignificant. The module further extracts pressure setting sequence segments from these windows, aligns them with temperature disturbance stages according to typical lag durations, and splices the segments to form a recommended control trajectory at the process segment level. During splicing, continuity checks are performed at the segment boundaries. The continuity threshold is set to 5 times the pressure adjustment step size boundary. This 5-fold threshold is determined by the statistical result of the upper limit of the cumulative change in adjacent actions allowed by the actuator without causing significant shock. If this is exceeded, a transition segment is inserted at the boundary, and the transition is performed step-by-step according to a rate limit. The transition segment length is 10 seconds, determined by the statistical result of the lower limit of the pressure regulating mechanism establishing a new stable state. Finally, a process optimization path is obtained, including the recommended temperature and pressure control parameter ranges, adjustment rhythms, and output and energy consumption target ranges for each process segment.

[0078] When analyzing the optimized path of hard carbon material preparation process and generating a precise control execution scheme coupled with temperature and pressure dual parameters, the module converts the optimized path into a set of executable instructions and monitoring sets. The instruction set is configured according to the output pressure setting sequence of the process segment, the rate limit within the segment, the action interval, the triggering condition, and the boundary protection strategy; the triggering condition includes temperature fluctuation threshold triggering and temperature range number matching, and the temperature fluctuation threshold is the aforementioned strict boundary threshold; the boundary protection strategy includes equipment safety boundary trimming and upper limit constraint on the number of actions, and the upper limit of the number of actions is derived from the unit time action limit of the maintenance plan and converted to the evaluation window scale of 300 seconds. The monitoring set includes six categories of indicators: pressure deviation, pressure fluctuation, stabilization time, output rate, energy consumption, and real-time processing delay. Alarm thresholds are set for each category, determined jointly by historical average benchmarks and safety priority constraints. Specifically, for pressure-related indicators, an alarm is triggered when the output exceeds the historical average by 1.2 times, where 1.2 times is the aforementioned degradation identification factor. For efficiency-related indicators, an alarm is triggered when output is 0.9 times lower than the historical average, and an alarm is triggered when energy consumption is 1.1 times higher than the historical average. The determination of 0.9 and 1.1 is based on the upper bound statistical results of output and energy consumption fluctuations in qualified batches, taking the smallest factor that can identify significant degradation and avoid frequent false alarms. The final output of the execution plan includes a version number, applicable formula version, applicable equipment identifier, generation timestamp, and a summary of differences from the historical benchmark plan.

[0079] like Figure 2 As shown, a method for preparing hard carbon anode materials is also provided. This method employs a temperature-pressure dual-parameter coupled control system for hard carbon material preparation. The method includes: real-time acquisition of temperature and pressure parameter data from the hard carbon material preparation equipment via a sensor network; generating an initial parameter interaction dataset characterized by the correlation between temperature and pressure parameters; obtaining a processed parameter mapping table; analyzing the interaction between temperature and pressure parameters based on the processed parameter mapping table; preprocessing fluctuating data; weighting key parameters; and constructing a preliminary parameter correlation matrix reflecting the influence of temperature changes on pressure response; in the preliminary parameter correlation matrix, temperature parameters are used as the main control disturbance parameter and pressure parameters as the controlled adjustment parameter; when the fluctuation of temperature parameters exceeds a preset threshold, iterative adjustment of the pressure parameters corresponding to the temperature parameters is performed using a genetic algorithm; setting iterative convergence conditions and population initialization methods; and combining crossover mutation probability and fitness function design to generate an adjusted pressure parameter set coupled with temperature changes; obtaining the adjusted pressure parameter set; and performing data range mapping and standardization transformation formulas within it. Secondary processing is performed, and a missing value imputation method is incorporated to complete incomplete data, resulting in an optimized parameter pairing list for temperature and pressure dual-parameter coordinated control. Based on the optimized parameter pairing list, virtual simulation tests are conducted on the crystallinity changes of hard carbon materials in their microstructure. The simulation results are then filtered using a selection and elimination mechanism to determine whether the material performance meets the preset standards under the corresponding temperature and pressure coupled control conditions, resulting in a performance verification dataset. Based on the performance verification dataset, a feedback loop mechanism is used to dynamically update the parameter correlation matrix, so that the adjustment results of the pressure parameters have a reverse effect on the temperature and pressure coupling relationship model. Historical data constraints and parameter adjustment step sizes in the verification data are processed to determine the refined temperature and pressure dual-parameter coupled control scheme. When the refined temperature and pressure dual-parameter coupled control scheme shows that it can make the pressure control index better than the historical average level under temperature fluctuation conditions, production efficiency indicators are integrated through data aggregation technology, and the real-time processing delay problem is addressed to generate a comprehensive benefit evaluation result. Based on this, the final optimized path of the hard carbon material preparation process is determined, resulting in a precise temperature and pressure dual-parameter coupled control execution scheme for actual production.

[0080] This method takes real-time temperature and pressure data as input. First, it establishes a temperature-pressure correspondence through data fusion and mapping. Then, through interactive analysis, it forms a correlation matrix that can be used for control, characterizing the influence path of temperature disturbances on the pressure response. When an excessive temperature deviation is detected, pressure regulation is transformed into a constrained optimization problem. A genetic algorithm iterates within the executable boundary to obtain a set of pressure parameters matching the temperature change. To ensure the comparability of parameters under different time periods and operating conditions, the pressure parameter set undergoes range mapping, standardization, and missing parameter completion, forming a temperature-pressure pairing list that can be directly used for collaborative control. Subsequently, the temperature-pressure pairing drives microstructure evolution simulation. Based on crystallinity stability and performance thresholds, a performance verification dataset is obtained, and the verification results are used to update the temperature-pressure coupling relationship model, making subsequent adjustments closer to the material performance targets. Under the premise that the refining scheme meets the requirement of pressure control being better than the historical average, output and energy consumption are introduced, and data delay bias is corrected to form a comprehensive benefit evaluation result. Finally, a feasible process optimization path and implementation scheme are output.

[0081] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A preparation system for a hard carbon anode material, characterized in that, include: The data acquisition and fusion module collects temperature and pressure parameter data in real time from the hard carbon material preparation equipment through a sensor network, generates an initial parameter interaction dataset characterized by the correlation between temperature and pressure parameters, and obtains a sorted parameter mapping table. The parameter correlation analysis module analyzes the interaction between temperature and pressure parameters based on the organized parameter mapping table, preprocesses fluctuating data, weights key parameters, and constructs a preliminary parameter correlation matrix that reflects the influence of temperature changes on pressure response. The temperature-pressure coupling regulation module uses temperature parameters as the main control disturbance parameters and pressure parameters as the controlled regulation parameters in the initial parameter correlation matrix. When the fluctuation of temperature parameters exceeds the preset threshold, the pressure parameters corresponding to the temperature parameters are iteratively adjusted through a genetic algorithm. The iterative convergence conditions and population initialization methods are set, and the adjusted pressure parameter set coupled with temperature changes is generated by combining the crossover mutation probability and fitness function design. The parameter optimization processing module obtains the adjusted pressure parameter set, performs secondary processing on the data range mapping and standardized transformation formula, and incorporates a missing value filling method to complete incomplete data, resulting in an optimized parameter pairing list for temperature and pressure dual-parameter coordinated control.

2. The preparation system for a hard carbon anode material according to claim 1, characterized in that: The data acquisition and fusion module collects temperature and pressure parameter data in real time from the hard carbon material preparation equipment through a sensor network, generates an initial parameter interaction dataset characterized by the correlation between temperature and pressure parameters, and obtains a sorted parameter mapping table including: The real-time temperature and pressure data streams of the hard carbon material preparation equipment were obtained, and the original multi-source mixed data sequence was constructed by combining them with historical production batch records. The original multi-source mixed data sequence is input into a multi-dimensional data fusion model for anomaly removal and noise reduction to obtain a clean production environment data stream. Feature extraction is performed on the data stream from the clean production environment to quantify the coupling degree between temperature and pressure, and a parameter dependency matrix is ​​generated. The strongly correlated features in the parameter dependency matrix are analyzed and indexed to obtain the sorted parameter mapping table.

3. The preparation system for a hard carbon anode material according to claim 1, characterized in that: The parameter correlation analysis module, based on the organized parameter mapping table, analyzes the interaction between temperature and pressure parameters, preprocesses fluctuating data, weights key parameters, and constructs a preliminary parameter correlation matrix reflecting the influence of temperature changes on pressure response, including: Temperature and pressure parameter sequences are extracted from the organized parameter mapping table. For fluctuating data segments in the temperature and pressure parameter sequences whose change gradient exceeds the preset gradient, exponential smoothing is performed to obtain smoothed interactive sequences. Time lag analysis is performed on the smoothed interaction sequence to align the parameters, and feature weights are assigned according to the contribution of temperature to pressure in the smoothed interaction sequence to generate a weighted feature dataset. The mutual information value is calculated based on the weighted feature dataset, and the mutual information value is filled into a preset structure to construct a preliminary parameter correlation matrix that reflects the influence of temperature change on pressure response.

4. The preparation system for a hard carbon anode material according to claim 1, characterized in that: The temperature-pressure coupled regulation module, in the initial parameter correlation matrix, uses the temperature parameter as the main control perturbation parameter and the pressure parameter as the controlled regulation parameter. When the fluctuation of the temperature parameter exceeds a preset threshold, it iteratively adjusts the pressure parameter corresponding to the temperature parameter using a genetic algorithm. It sets the iterative convergence condition and population initialization method, and combines the crossover mutation probability and fitness function design to generate a set of adjusted pressure parameters coupled with temperature changes, including: Temperature and pressure parameter sequences are extracted from the preliminary parameter correlation matrix, and time windows of temperature parameter sequence fluctuations exceeding limits are identified to determine the optimization space of the pressure parameter to be optimized. An initial population is generated based on the optimization space of the pressure parameters to be optimized, and the fitness value of the quantitative response matching degree is calculated for the initial population. The next generation population is generated by performing crossover and mutation operations based on the fitness value, and the optimal individual is output when the next generation population meets the iterative convergence condition. The optimal individual is analyzed to generate a set of adjusted pressure parameters coupled with temperature changes.

5. The preparation system for a hard carbon anode material according to claim 1, characterized in that: The parameter optimization processing module obtains the adjusted pressure parameter set, performs secondary processing on the data range mapping and standardized transformation formula, and incorporates a missing value filling method to complete incomplete data, resulting in an optimized parameter pairing list for temperature and pressure dual-parameter coordinated control, including: Obtain the adjusted pressure parameter set and construct a data range mapping model. Project the pressure parameter set to obtain the mapped pressure data sequence. The mapped pressure data sequence is standardized to generate a standardized pressure data stream. Breakpoint interpolation is performed on the standardized pressure data stream to generate a completed pressure data sequence. The completed pressure data sequence is associated with the temperature data to construct a temperature-pressure dual-parameter coupling matrix. The temperature-pressure dual-parameter coupling matrix is ​​parsed to output an optimized parameter pairing list for temperature-pressure dual-parameter coordinated control.

6. The preparation system for a hard carbon anode material according to claim 1, characterized in that, It also includes a performance simulation and verification module, which performs virtual simulation tests on the crystallinity changes of hard carbon materials in their microstructure based on an optimized parameter pairing list. The simulation results are then filtered using a selection and elimination mechanism to determine whether the material performance meets preset standards under corresponding temperature-pressure coupled control conditions, resulting in a performance verification dataset. Specifically, this dataset includes: Obtain the list of optimized parameter pairs, and extract the temperature and pressure variables from the list of optimized parameter pairs to construct a hard carbon microstructure evolution model; The microstructure evolution model of hard carbon was placed into a virtual simulation test environment to generate a trajectory spectrum of crystallinity change. Analyze the crystallinity change trajectory spectrum and remove samples that exceed the structural stability limit in the crystallinity change trajectory spectrum to retain the effective simulation result set; The effective simulation result set is mapped to material performance index values. If the material performance index values ​​fall within the preset standard threshold range, the performance verification dataset is output.

7. The preparation system for a hard carbon anode material according to claim 6, characterized in that, It also includes a feedback update module, which dynamically updates the parameter correlation matrix based on the performance verification dataset using a feedback loop mechanism. This allows the adjustment results of the pressure parameters to have a reverse effect on the temperature-pressure coupling relationship model. It processes historical data constraints and parameter adjustment step sizes in the verification data to determine the refined temperature-pressure dual-parameter coupling control scheme, specifically including: Obtain the performance verification dataset and extract the dominant bias features to generate a multidimensional feedback bias dataset; The parameter correlation matrix is ​​updated using a multidimensional feedback deviation dataset, and the sensitivity factor of the pressure parameter to the microstructure evolution is extracted. The sensitivity factor is input into the temperature-pressure coupling model, and the adjustment increment of the constrained parameters is calculated by combining historical data constraints. The incremental adjustment of the restricted parameters is superimposed on the baseline control curve to determine the temperature and pressure dual-parameter coupled control scheme after refining.

8. The preparation system for a hard carbon anode material according to claim 7, characterized in that, It also includes an execution decision module. When the refined temperature-pressure dual-parameter coupled control scheme shows that it can make the pressure control index better than the historical average under temperature fluctuation conditions, it integrates production efficiency indicators through data aggregation technology, and handles the real-time processing delay problem to generate a comprehensive benefit evaluation result. Based on this, it determines the final optimized path of hard carbon material preparation process and obtains a precise temperature-pressure dual-parameter coupled control execution scheme for actual production, specifically including: Obtain pressure values ​​under temperature fluctuation conditions. If the pressure values ​​are better than the historical average, collect data on the production rate and energy consumption of hard carbon materials. The real-time processing delay of hard carbon material production rate and energy consumption data is calculated, and the polymerization data is time-compensated and corrected using the real-time processing delay.

9. The preparation system for a hard carbon anode material according to claim 8, characterized in that: The execution decision module, when the refined temperature-pressure dual-parameter coupled control scheme shows that it can make the pressure control index better than the historical average level under temperature fluctuation conditions, integrates production efficiency indicators through data aggregation technology, and handles the real-time processing delay problem to generate a comprehensive benefit evaluation result. Based on this, it determines the final optimized path of hard carbon material preparation process, and obtains a temperature-pressure dual-parameter coupled precise control execution scheme for actual production, which also includes: The corrected data is input into a multidimensional benefit analysis model to obtain a comprehensive benefit assessment result. Based on the comprehensive benefit assessment result, the optimization path for the preparation process of hard carbon materials is determined. We analyzed the optimization path of hard carbon material preparation process and generated a precise control execution scheme based on temperature and pressure dual-parameter coupling.

10. A method for preparing a hard carbon anode material, using the hard carbon anode material preparation system according to any one of claims 1-9, characterized in that, The method includes: S1. Real-time acquisition of temperature and pressure parameters from the hard carbon material preparation equipment via sensor network, generating an initial parameter interaction dataset characterized by the correlation between temperature and pressure parameters, and obtaining a sorted parameter mapping table; S2. Based on the sorted parameter mapping table, analyze the interaction between temperature and pressure parameters, preprocess the fluctuating data, weight the key parameters, and construct a preliminary parameter correlation matrix that reflects the influence of temperature changes on pressure response. S3. In the preliminary parameter correlation matrix, the temperature parameter is used as the main control perturbation parameter and the pressure parameter is used as the controlled adjustment parameter. When the fluctuation of the temperature parameter exceeds the preset threshold, the pressure parameter corresponding to the temperature parameter is iteratively adjusted through a genetic algorithm. The iterative convergence condition and population initialization method are set, and the adjusted pressure parameter set coupled with the crossover mutation probability and fitness function is designed to generate the set of pressure parameters coupled with the temperature change. S4. Obtain the adjusted pressure parameter set, perform secondary processing on the data range mapping and standardized conversion formula, and incorporate the missing value filling method to complete the incomplete data, thereby obtaining an optimized parameter pairing list for temperature and pressure dual-parameter coordinated control. S5. Based on the optimized parameter pairing list, conduct virtual simulation tests on the crystallinity changes of hard carbon materials in the microstructure, and filter the simulation results using a selection and elimination mechanism to determine whether the material performance meets the preset standards under the corresponding temperature and pressure coupling control conditions, thereby obtaining the performance verification dataset. S6. Based on the performance verification dataset, a feedback loop mechanism is used to dynamically update the parameter correlation matrix, so that the adjustment result of the pressure parameter has a reverse effect on the temperature-pressure coupling relationship model. The historical data constraints and parameter adjustment step size in the verification data are processed to determine the refined temperature-pressure dual-parameter coupling control scheme. S7. When the refined temperature and pressure dual-parameter coupling control scheme shows that it can make the pressure control index better than the historical average level under temperature fluctuation conditions, the production efficiency index is integrated by data aggregation technology, and the real-time processing delay problem is handled to generate a comprehensive benefit evaluation result. Based on this, the final optimization path of hard carbon material preparation process is determined, and a temperature and pressure dual-parameter coupling precision control execution scheme for actual production is obtained.

Images