Crystallization quality control method and control system in carbon source production

Abstract

The invention relates to the technical field of crystallization quality control, in particular to a crystallization quality control method and control system in carbon source production, and the method comprises the following steps: monitoring various related parameter data in each observation time period in the crystallization process in real time during the carbon source production; based on principal component analysis and an entropy weight method, key variables in all related parameter data are extracted, and crystal impurity stability is calculated; a sliding window mean value-standard deviation method and dynamic time warping are adopted to quantify the process fluctuation strength and historical consistency, and a dynamic goodness of fit score is calculated; and identifying an abnormal state in the crystallization process before the current acquisition moment by using the dynamic goodness of fit score so as to optimize control parameters in real time through model prediction control, thereby realizing multivariable collaborative supervision control of the crystallization process. The invention aims to solve the problems of non-uniform crystal granularity, unstable crystal form and impurity residue in the traditional crystallization process, and multi-variable cooperative supervision and control of the crystallization process are realized.

Description

Technical Field

[0001] The present application relates to the technical field of crystallization quality control, and in particular to a crystallization quality control method and control system in carbon source production. Background Art

[0002] Crystallization quality control in carbon source production is a core link in the chemical industry. Traditional crystallization processes rely on experience to control parameters such as temperature and concentration, and are easily affected by kinetic and thermodynamic factors, leading to problems such as uneven crystal size, unstable crystal form or residual impurities. With the development of refined chemical industry and the upgrading of green manufacturing needs, crystallization process control technology has evolved from manual experience to intelligent supervision and control.

[0003] During the carbon source production process, the stability of crystal growth is interfered with by residual metal ions, organic by-products, and solvents that have not been completely removed in the raw materials. These impurities destroy the integrity of the crystal structure through various mechanisms. For example, impurity molecules are easily adsorbed on the active sites on the crystal surface, hindering the directional and orderly stacking of solute molecules, resulting in an imbalance in the crystal growth rate, thereby causing abnormal crystal morphology. In addition, solvent residues may induce the formation of non-target crystal forms by changing the crystallization environment, or compete with solute molecules for adsorption sites, forcing the crystals to exhibit abnormal morphologies such as needles and flakes. Conventional monitoring equipment can often only provide static data and cannot accurately predict and control the response to the dynamic interference of impurities. This limitation makes it difficult for existing technologies to fundamentally solve the quality problems of morphological abnormalities and batch inconsistencies caused by impurity interference during crystal growth. Summary of the Invention

[0004] In order to solve the above technical problems, the present application provides a crystallization quality control method and control system in carbon source production. The technical solutions adopted are as follows:

[0005] In a first aspect, an embodiment of the present application provides a method for controlling crystallization quality in carbon source production, the method comprising the following steps:

[0006] Real-time monitoring of various relevant parameter data during each observation period during the crystallization process during carbon source production;

[0007] The entropy weight method is used to calculate the weight assigned to each type of relevant parameter data, and the coefficient of variation of the corresponding relevant parameter data is weighted and summed to obtain the comprehensive weight. Principal component analysis is used to extract the principal component scores and variance contribution rates of the two principal components in all relevant parameter data within an observation period, and the stability of the crystalline impurities is calculated by combining the comprehensive weights.

[0008] All crystalline impurity stabilities calculated during all observation periods are grouped into a sequence, and a sliding time window is set. The ratio of the mean to the standard deviation within each time window, as well as the DTW distance between the elements in the corresponding time window and the elements in the sequence are fused, and the fusion result is used as the dynamic fit score.

[0009] Dynamic fit scoring is used to identify abnormal states in the crystallization process before the current acquisition moment, so that control parameters can be optimized in real time through model predictive control to achieve multivariable collaborative supervisory control of the crystallization process.

[0010] Preferably, the observation period is a period within a preset time period before a collection moment.

[0011] Preferably, the relevant parameter data include: temperature, solute concentration, metal ion concentration and solvent residual rate.

[0012] Preferably, the calculation method of the crystalline impurity stability is:

[0013] The normalized variance contribution rate of each principal component is weighted summed with the principal component score of the principal component;

[0014] Performing inverse proportional mapping on the comprehensive weight;

[0015] The result of the weighted sum of the principal component scores and the result of the inverse proportional mapping are forwardly fused to obtain the crystalline impurity stability.

[0016] Preferably, the method for obtaining the dynamic fit score is further determined as follows:

[0017]

[0018] Where B is the dynamic fit score, M is the number of time windows, μ j and σ j is the mean and standard deviation in the jth time window, DTW j is the DTW distance between the element in the jth time window and the element in the sequence, Norm() is the normalization function, and e is a natural constant.

[0019] Preferably, the crystallization state includes: a stable interval, a warning interval and an abnormal interval.

[0020] Preferably, the method for identifying abnormal states in the crystallization process before the current acquisition moment by using dynamic coincidence scoring is:

[0021] When the dynamic fit score is less than the preset warning threshold, the crystallization process before the current acquisition moment is in an abnormal state.

[0022] Preferably, when the crystallization process before the current acquisition moment is in an abnormal state, the method for optimizing the control parameters in real time through model predictive control is:

[0023] If the standard deviation calculated as a time window during the observation period at the current acquisition moment suddenly increases, the standard deviation calculated as a time window during the observation period at the current acquisition moment and the stirring rate data at the current acquisition moment are used as inputs for model predictive control, and a control signal is output to the crystallization kettle stirring device to control the stirring rate and the solution circulation flow rate;

[0024] If the DTW distance calculated using the observation period at the current acquisition moment as a time window suddenly increases, the DTW distance value calculated using the observation period at the current acquisition moment as a time window and the crystallization temperature gradient data at the current acquisition moment are used as inputs for model predictive control, and a control signal is output to the temperature control device to dynamically control the crystallization temperature gradient;

[0025] If the mean value calculated as a time window during the observation period at the current acquisition moment suddenly decreases, the mean value calculated as a time window during the observation period at the current acquisition moment and the solvent residual rate data at the current acquisition moment are used as the input of the model predictive control, and a control signal is output to the solvent recovery unit to control the evaporation pressure and degassing time.

[0026] Preferably, the judgment criteria for the sudden increase or sudden decrease are:

[0027] The standard deviation, DTW distance, and mean were recorded as the dominant factors;

[0028] The observation period of the current collection moment is regarded as a time window. The mean of historical data under various dominant factors at each collection moment in the time window is calculated, and a linear fit is performed on all the mean of historical data of various corresponding dominant factors calculated in the time window to obtain the linear regression slope of the corresponding dominant factor.

[0029] The mean of the historical data under various dominant factors at each collection moment in the time window is the average value of the corresponding dominant factors at the collection moment in the time window and all previous collection moments;

[0030] If the linear regression slope of various dominant factors is less than the preset decrease judgment threshold, it is determined that the corresponding dominant factor is suddenly decreased;

[0031] If the linear regression slope of various dominant factors is greater than the preset increase judgment threshold, it is determined that the corresponding dominant factor has increased suddenly.

[0032] In a second aspect, another embodiment of the present application further provides a crystallization quality control system in carbon source production, which implements the above-mentioned crystallization quality control method in carbon source production, and the system comprises:

[0033] 1. Initialization configuration module, used for system initialization and parameter configuration;

[0034] 2. Data flow and functional module chain, including three sub-modules: real-time process parameter monitoring and signal optimization module, multi-dimensional feature analysis and dominant factor identification module, and dynamic process stability diagnosis and root cause tracing module;

[0035] The process parameter real-time monitoring and signal optimization module is used to monitor various relevant parameter data in each observation period during the crystallization process during carbon source production;

[0036] The multi-dimensional feature analysis and dominant factor identification module is used to calculate the weight assigned to each type of relevant parameter data using the entropy weight method, and to perform weighted summation on the coefficient of variation of the corresponding relevant parameter data to obtain a comprehensive weight. Principal component analysis is used to extract the principal component scores and variance contribution rates of the two principal components in all relevant parameter data within an observation period, and combined with the comprehensive weight, the stability of the crystalline impurities is calculated.

[0037] The process stability dynamic diagnosis and root cause tracing module is used to sequence all crystalline impurity stabilities calculated during all observation periods and set a sliding time window. The ratio of the mean to the standard deviation within each time window, as well as the DTW distance between the elements in the corresponding time window and the elements in the sequence, are fused and the fusion result is used as the dynamic fit score.

[0038] 3. Supervisory control core module, which is used to identify abnormal conditions in the crystallization process before the current acquisition moment using dynamic fit scoring, so as to optimize control parameters in real time through model predictive control and realize multivariable collaborative supervisory control of the crystallization process.

[0039] This application has at least the following beneficial effects:

[0040] (1) In order to solve the quality problems of abnormal crystal form and uneven particle size caused by impurity adsorption and solvent residue during the crystallization process, this application uses principal component analysis combined with entropy weight method to dynamically quantify the stability characteristics of crystal quality. By integrating the principal component variance contribution rate with the real-time parameter state, the local disturbance of thermodynamic dominant factors and impurity interference is balanced to solve the problem of imbalance in crystal growth rate.

[0041] (2) In response to batch instability caused by process fluctuations and historical pattern deviations, the sliding window mean-standard deviation method and dynamic time warping are used to comprehensively reflect the process stability and historical consistency characteristics, eliminating the influence of random process fluctuations and non-target crystal formation;

[0042] (3) Through model predictive control under the supervisory control framework, parameters such as temperature, flow, and pressure are optimized in real time to achieve multivariable collaborative closed-loop supervisory control, improve process stability, and trigger differentiated control strategies based on dynamic fit scores to achieve multivariable collaborative closed-loop control of the carbon source crystallization process. Ultimately, this improves crystal size uniformity, crystal form stability, and solvent residue control accuracy, ensuring product batch consistency. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0044] Figure 1 A flow chart of a method for controlling crystal quality in carbon source production provided in one embodiment of the present application;

[0045] Figure 2 A flow chart of a multivariable collaborative supervisory control method for a crystallization process provided in one embodiment of the present application. DETAILED DESCRIPTION

[0046] One embodiment of the present application provides a method for controlling crystal quality in carbon source production. Figure 1 , the method comprises the following steps:

[0047] Step 1: During carbon source production, various relevant parameter data in each observation period during the crystallization process are monitored in real time.

[0048] In the carbon source production crystallization quality control system, a high-precision temperature sensor is installed on the top of the crystallization kettle to collect real-time temperature data during crystallization to monitor thermodynamic changes during the crystallization process; an online concentration detector is integrated in the solution circulation pipeline to obtain real-time solute concentration data to evaluate the impact of solute supersaturation on crystal growth; the raw material solution is continuously tested by an inductively coupled plasma (ICP) detector to collect metal ion concentration data during the crystallization process to identify the interference of impurity adsorption on crystal surface growth; a flow meter is installed in the solvent recovery unit in conjunction with a gas chromatograph to collect solvent residual rate (S) data to analyze the potential threat of solvent residual to crystal stability. The data acquisition frequency of all acquisition devices is set to 10Hz, and real-time data acquisition is achieved through high-precision sensors and online detectors, providing high-frequency input for supervisory control and ensuring comprehensive monitoring of process parameters. Among them, temperature, solute concentration, metal ion concentration and solvent residual rate are used as various relevant parameter data in the crystallization process during carbon source production.

[0049] For the raw data collected above, in the supervisory control process, the wavelet denoising algorithm is used to eliminate high-frequency noise, combined with the Z-score standardization to unify the dimensions, and the period within the preset time before each collection moment is used as an observation period. In this embodiment, the preset time is set to 1 hour. For collection moments with previous data of less than one hour, the mean filling method is used to fill the previous data to ensure the integrity of the observation period data, and obtain the pre-processed temperature series, solute concentration series, metal ion concentration series and solvent residual rate series in all observation periods, providing high signal-to-noise ratio input for the subsequent supervisory control model, thereby systematically solving the crystallization quality problems caused by impurity interference, crystal form abnormalities and solvent residues.

[0050] Step 2: Based on principal component analysis and entropy weight method, key variables in all relevant parameter data are extracted and the stability of crystal impurities is calculated.

[0051] During the carbon source production process, impurities such as metal ions and organic by-products remaining in the raw materials and solvents that have not been completely removed are adsorbed on the active sites on the crystal surface or change the crystallization environment, resulting in uneven crystal size, unstable crystal form and solvent residue.

[0052] Therefore, this application uses the entropy weight method to calculate the weight assigned to each type of relevant parameter data, and performs weighted summation on the coefficient of variation of the corresponding relevant parameter data to obtain a comprehensive weight; principal component analysis is used to extract the principal component scores and variance contribution rates of the two principal components in all relevant parameter data within an observation period, and combined with the comprehensive weight, the stability of the crystalline impurity is calculated.

[0053] Among them, the calculation method of the crystalline impurity stability is: weighted summing the normalized variance contribution rate of each principal component to the principal component score of the principal component; inversely mapping the comprehensive weight; forward fusion of the result of the weighted summation of the principal component scores and the result of the inverse proportional mapping to obtain the crystalline impurity stability.

[0054] Optionally, the inverse proportional mapping may be implemented by negative linear mapping, negative exponential mapping, or by setting adjustment parameters.

[0055] It can be understood that forward fusion is a fusion method such as addition and multiplication between data. The specific forward fusion method is determined by the implementer according to the actual situation, and this application does not impose any special restrictions.

[0056] In this embodiment, the temperature sequence, solute concentration sequence, metal ion concentration sequence, and solvent residual rate sequence pre-processed within an observation period are used as input. Based on the needs of supervisory control, the principal component analysis (PCA) algorithm is used to extract key variables, wherein the number of principal components is set to 2, and the principal component scores PC1 and PC2 of the two principal components are output, as well as their corresponding variance contribution rates V1 and V2. Subsequently, the four standardized sequences are used as input, and the entropy weight method is used to calculate the information entropy of each sequence and output the weight distribution result of each sequence. The weight value quantifies the significance of the impact of each parameter variation on quality. Among them, the principal component analysis algorithm is a well-known technology and will not be repeated here.

[0057] Based on the above analysis, the stability A of the crystalline impurities in each observation period is calculated. The specific calculation formula is:

[0058]

[0059] Among them, V1 and V2 are the variance contribution rates of the two principal components, which respectively explain the proportion of the total variation in the original data set. V1 corresponds to the parameter with the most significant influence in the crystallization process, and V2 corresponds to a minor but non-negligible factor. After normalization, V1 and V2 are used as the weights of PC1 and PC2. According to the proportion of data variation explained by the principal component, its contribution to the stability of crystallization impurities can be dynamically adjusted, so that the stability of crystallization impurities can be more focused on the dominant factors (such as thermodynamic state) while taking into account the influence of minor factors (such as impurity interference), thereby balancing the global trend and local disturbance of crystallization quality stability; PC1 and PC2 are the principal component scores of the two principal components, respectively. The larger the principal component score, the higher the matching degree between the current process parameter combination and the main variation mode extracted by PCA, which means the crystallization process is more stable, and vice versa.

[0060] in, L is the comprehensive weight, which is the weight ω assigned by the entropy weight method i and coefficient of variation Ki The result of weighted summation, ω i is the weight distribution result of the i-th sequence, K i is the coefficient of variation of the i-th sequence, ω i ×K i It represents the contribution of the i-th parameter to the comprehensive weight L. The larger the product of the two, the more likely the i-th parameter is to be the dominant factor affecting the quality. The sum of the two products represents the comprehensive influence of all parameters on the crystallization quality. The larger the sum, the greater the influence of the parameter change on the crystallization quality. i is the summation index, ranging from 1 to the number of sequences 4. The calculation of the coefficient of variation is a well-known technology and will not be elaborated here.

[0061] It should be understood that the crystalline impurity stability A dynamically quantifies the stability of the crystal quality by fusing the potential variation pattern extracted by PCA and the real-time parameter state weighted by the entropy weight method. The larger the value, the more stable the crystal quality. Otherwise, it indicates the risk of abnormal crystal form or solvent residue.

[0062] Step 3: Use the sliding window mean-standard deviation method and dynamic time warping to quantify the process fluctuation intensity and historical consistency, and calculate the dynamic fit score.

[0063] Due to factors such as impurity adsorption and solvent residue during the carbon source crystallization process, the crystal growth rate becomes unbalanced and the crystal form becomes abnormal. The monitoring parameters may mutate or deviate from the stable trend due to impurity adsorption, fluctuations in thermodynamic conditions or solvent residue, resulting in unstable crystal growth rate and the formation of non-target crystal forms. The increased volatility of the process and the deviation from the target state will reduce the principal component score and increase the comprehensive weight, resulting in a decrease in the crystal impurity stability A value, reflecting the deterioration of the stability of the crystal quality.

[0064] Therefore, this application sequences all the crystalline impurity stabilities calculated within all observation periods and sets a sliding time window; fuses the ratio of the mean to the standard deviation in each time window, as well as the DTW distance between the elements in the corresponding time window and the elements in the sequence, and uses the fusion result as a dynamic fit score.

[0065] It can be understood that fusion can be divided into forward fusion and reverse fusion. This embodiment adopts the reverse fusion method. Reverse fusion is a fusion method such as subtraction and division between data. The specific reverse fusion method is determined by the implementer according to the actual situation. The application does not impose any special restrictions.

[0066] In this embodiment, for the crystallization impurity stability A calculated in each observation period, the crystallization impurity stability A of all observation periods is composed into a sequence, and the sequence is used as input. The sliding window mean-standard deviation method is used to quantify the process fluctuation intensity in real time. Due to the characteristics of the short crystallization dynamics response cycle and the need for high-frequency monitoring, this embodiment sets the sliding window length to 1 minute, and outputs the mean μ and standard deviation σ in the window, where μ reflects the average level of the current crystallization quality, and σ quantifies the process fluctuation intensity. The larger the standard deviation, the worse the process stability.

[0067] Then, taking all the crystalline impurity stabilities in each time window and all the crystalline impurity stabilities in the sequence as input, the dynamic time warping distance is used to output the DTW distance between all the crystalline impurity stabilities in each time window and all the crystalline impurity stabilities in the sequence. The closer the DTW distance is to 0, the higher the match between the data in the time window and the historical crystallization quality.

[0068] Based on the above analysis, a dynamic fit score B is constructed. The specific calculation formula is:

[0069]

[0070] Where M is the number of time windows, μ j and σ j is the mean and standard deviation in the jth time window. The ratio between the two combines the quality level and process stability, and represents the quality score after fluctuation adjustment. The larger the value, the higher the crystallization quality in the time window and the smaller the fluctuation, and the more stable the crystallization process; DTW j It is the DTW distance between the elements in the j-th time window and the elements in the sequence, which measures the morphological matching degree between the crystalline impurity stability in the time window and the sequence composed of all crystalline impurity stabilities. The smaller the value, the more similar the morphology of the current window is to the historical data, that is, the current crystallization process is highly consistent with the historical crystallization process, and the more normal the crystal growth pattern is. Norm() is the normalization function, and e is the natural constant.

[0071] It should be understood that the dynamic fit score B comprehensively quantifies process stability and historical consistency, providing a dynamic decision-making basis for closed-loop control. It quantifies the crystallization quality through quality level and matching. The larger the value, the more stable the crystallization process, the thermodynamic state, impurity control, and solvent residue are all in line with expectations, and the crystal form is uniform.

[0072] Step 4: Use dynamic fit scoring to identify abnormal conditions in the crystallization process before the current acquisition moment, so as to optimize the control parameters in real time through model predictive control and realize multivariable collaborative supervisory control of the crystallization process.

[0073] Then, based on the historical stable production batch data, the benchmark range of the dynamic fit score B is determined. Specifically, a stability threshold and a warning threshold are preset, where the stability threshold is greater than the warning threshold. In this embodiment, the stability threshold is 0.8 and the warning threshold is 0.5.

[0074] That is, when the dynamic fit score is greater than or equal to the stable threshold, the crystallization process before the current acquisition moment is in a stable state; when the dynamic fit score is greater than or equal to the warning threshold and less than the stable threshold, the crystallization process before the current acquisition moment is in a warning state; when the dynamic fit score is less than the warning threshold, the crystallization process before the current acquisition moment is in an abnormal state.

[0075] Accordingly, the present application utilizes dynamic fit scoring to identify abnormal states in the crystallization process before the current acquisition moment.

[0076] Furthermore, this application is based on a supervisory control framework and adopts model predictive control (Model Predictive Control) to optimize control parameters in real time, forming a closed-loop supervisory control, solving the optimal control parameter combination in the future time window in real time, and realizing precise regulation through the actuator according to the dominant factors of crystallization quality deviation.

[0077] When the crystallization process before the current acquisition moment is abnormal, the dominant factor of the current crystallization quality deviation is located according to the components of the dynamic fit score B, and the control parameters are optimized in real time through model predictive control using the dominant factor. Specifically:

[0078] If the standard deviation calculated using the observation period at the current acquisition moment as a time window suddenly increases, it indicates that the process fluctuation has intensified. At this time, the standard deviation calculated using the observation period at the current acquisition moment as a time window and the stirring rate data at the current acquisition moment are used as the input of the model predictive control, and a control signal is output to the crystallization kettle stirring equipment to control the stirring rate and the solution circulation flow rate to improve the stirring uniformity and optimize the solution circulation flow rate.

[0079] If the DTW distance calculated using the observation period of the current acquisition moment as a time window suddenly increases, the DTW distance value calculated using the observation period of the current acquisition moment as a time window and the crystallization temperature gradient data of the current acquisition moment are used as the input of the model predictive control, and a control signal is output to the temperature control device to dynamically control the crystallization temperature gradient, that is, to reduce the cooling rate to suppress the imbalance of crystal plane growth and guide directional crystallization.

[0080] If the mean value calculated using the observation period at the current acquisition moment as a time window suddenly decreases, the mean value calculated using the observation period at the current acquisition moment as a time window and the solvent residual rate data at the current acquisition moment are used as inputs for model predictive control, and a control signal is output to the solvent recovery unit to control the evaporation pressure and degassing time, that is, to increase the evaporation pressure or extend the degassing time to reduce the impact of solvent residue on the crystal form.

[0081] It is worth noting that in the above process, the implementers can determine the judgment criteria on their own, such as using the threshold method, sliding window trend analysis method and other methods to make judgments.

[0082] In this application, the standard deviation, DTW distance and mean are all recorded as dominant factors; the observation period of the current acquisition moment is taken as a time window, and the linear regression slope of the corresponding dominant factor is obtained by calculating the mean value of the historical data under various dominant factors at each acquisition moment in the time window, and performing linear fitting on all the historical data means of various corresponding dominant factors calculated in the time window; wherein, the mean value of the historical data under various dominant factors at each acquisition moment in the time window is the average value of the corresponding dominant factor at the acquisition moment in the time window and all previous acquisition moments; if the linear regression slope of various dominant factors is less than the preset decrease judgment threshold, it is determined that the corresponding dominant factor is suddenly reduced; if the linear regression slope of various dominant factors is greater than the preset increase judgment threshold, it is determined that the corresponding dominant factor is suddenly increased.

[0083] Specifically, this embodiment uses the observation period of the current collection moment as an example to calculate the mean value within a time window. The linear regression slope is obtained by calculating the mean value of the historical data at each collection moment within the time window and performing a linear fit on all the historical data means calculated within the time window. The mean value of the historical data at each collection moment within the time window is the average of the means of the collection moment within the time window and all previous collection moments.

[0084] If the linear regression slope is less than a preset decrease judgment threshold, the mean is determined to have suddenly decreased. In this embodiment, the preset decrease judgment threshold is -0.1.

[0085] By precisely controlling various factors, such as stirring rate, temperature gradient, and evaporation pressure, we effectively suppress impurity adsorption interference, balance crystal growth rate, and reduce the risk of solvent residue. This significantly improves crystal size uniformity, crystal form stability, and batch consistency, providing reliable support for the large-scale production of high-purity carbon source products. Model Predictive Control (MPC), the core algorithm for supervisory control, approaches the optimal process state in real time through rolling optimization and feedback correction. Its specific implementation is well-known technology and will not be elaborated on here.

[0086] In this application, the multivariable collaborative supervisory control method flow chart of the crystallization process is as shown in the attached Figure 2 shown.

[0087] Another embodiment of the present application further provides a crystallization quality control system for carbon source production. The crystallization quality control system for carbon source production adopts a modular closed-loop design and is composed of the following core modules. Each module operates in coordination through a standardized interface. The system includes:

[0088] 1. Initialization configuration module, used for system initialization and parameter configuration;

[0089] 2. Data flow and functional module chain, including three sub-modules: real-time process parameter monitoring and signal optimization module, multi-dimensional feature analysis and dominant factor identification module, and dynamic process stability diagnosis and root cause tracing module;

[0090] The process parameter real-time monitoring and signal optimization module is used to monitor various relevant parameter data in each observation period during the crystallization process during carbon source production;

[0091] The multi-dimensional feature analysis and dominant factor identification module is used to calculate the weight assigned to each type of relevant parameter data using the entropy weight method, and to perform weighted summation on the coefficient of variation of the corresponding relevant parameter data to obtain a comprehensive weight. Principal component analysis is used to extract the principal component scores and variance contribution rates of the two principal components in all relevant parameter data within an observation period, and combined with the comprehensive weight, the stability of the crystalline impurities is calculated.

[0092] The process stability dynamic diagnosis and root cause tracing module is used to sequence all crystalline impurity stabilities calculated during all observation periods and set a sliding time window. The ratio of the mean to the standard deviation within each time window, as well as the DTW distance between the elements in the corresponding time window and the elements in the sequence, are fused and the fusion result is used as the dynamic fit score.

[0093] 3. Supervisory control core module, which is used to identify abnormal conditions in the crystallization process before the current acquisition moment using dynamic fit scoring, so as to optimize control parameters in real time through model predictive control and realize multivariable collaborative supervisory control of the crystallization process.

[0094] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention herein. This application is intended to cover any variations, uses, or adaptations of the present application that follow the general principles of the present application and include common knowledge or customary techniques in the art not invented herein.

[0095] It will be understood that the present application is not limited to the exact construction that has been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof.

Claims

1. A method for controlling crystal quality in carbon source production, characterized in that: The method comprises the following steps: Real-time monitoring of various relevant parameter data during each observation period during the crystallization process during carbon source production; The entropy weight method is used to calculate the weight assigned to each type of relevant parameter data, and the coefficient of variation of the corresponding relevant parameter data is weighted and summed to obtain the comprehensive weight. Principal component analysis is used to extract the principal component scores and variance contribution rates of the two principal components in all relevant parameter data within an observation period, and the stability of the crystalline impurities is calculated by combining the comprehensive weights. All crystalline impurity stabilities calculated during all observation periods are grouped into a sequence, and a sliding time window is set. The ratio of the mean to the standard deviation within each time window, as well as the DTW distance between the elements in the corresponding time window and the elements in the sequence are fused, and the fusion result is used as the dynamic fit score. Dynamic fit scoring is used to identify abnormal states in the crystallization process before the current acquisition moment, so that control parameters can be optimized in real time through model predictive control to achieve multivariable collaborative supervisory control of the crystallization process.

2. The method for controlling crystal quality in carbon source production according to claim 1, wherein: The observation period is a period of time within a preset time before a collection moment.

3. The method for controlling crystal quality in carbon source production according to claim 1, wherein: The relevant parameter data include: temperature, solute concentration, metal ion concentration and solvent residual rate.

4. The method for controlling crystal quality in carbon source production according to claim 1, wherein: The calculation method of the crystalline impurity stability is: The normalized variance contribution rate of each principal component is weighted summed with the principal component score of the principal component; Performing inverse proportional mapping on the comprehensive weight; The result of the weighted sum of the principal component scores and the result of the inverse proportional mapping are forwardly fused to obtain the crystalline impurity stability.

5. The method for controlling crystal quality in carbon source production according to claim 1, wherein: The method for obtaining the dynamic fit score is further determined as follows: Where B is the dynamic fit score, M is the number of time windows, μ j and σ j is the mean and standard deviation in the jth time window, DTW j is the DTW distance between the element in the jth time window and the element in the sequence, Norm() is the normalization function, and e is a natural constant.

6. The method for controlling crystal quality in carbon source production according to claim 1, wherein: The crystallization state includes: a stable range, a warning range and an abnormal range.

7. The method for controlling crystal quality in carbon source production according to claim 6, wherein: The method for identifying abnormal states in the crystallization process before the current acquisition moment by using dynamic coincidence scoring is as follows: When the dynamic fit score is less than the preset warning threshold, the crystallization process before the current acquisition moment is in an abnormal state.

8. The method for controlling crystal quality in carbon source production according to claim 1, wherein: When the crystallization process before the current acquisition moment is in an abnormal state, the method for real-time optimization of control parameters through model predictive control is as follows: If the standard deviation calculated as a time window during the observation period at the current acquisition moment suddenly increases, the standard deviation calculated as a time window during the observation period at the current acquisition moment and the stirring rate data at the current acquisition moment are used as inputs for model predictive control, and a control signal is output to the crystallization kettle stirring device to control the stirring rate and the solution circulation flow rate; If the DTW distance calculated using the observation period at the current acquisition moment as a time window suddenly increases, the DTW distance value calculated using the observation period at the current acquisition moment as a time window and the crystallization temperature gradient data at the current acquisition moment are used as inputs for model predictive control, and a control signal is output to the temperature control device to dynamically control the crystallization temperature gradient; If the mean value calculated as a time window during the observation period at the current acquisition moment suddenly decreases, the mean value calculated as a time window during the observation period at the current acquisition moment and the solvent residual rate data at the current acquisition moment are used as the input of the model predictive control, and a control signal is output to the solvent recovery unit to control the evaporation pressure and degassing time.

9. The method for controlling crystal quality in carbon source production according to claim 8, wherein: The judgment criteria for the sudden increase or decrease are: The standard deviation, DTW distance, and mean were recorded as the dominant factors; The observation period of the current collection moment is regarded as a time window. The mean of historical data under various dominant factors at each collection moment in the time window is calculated, and a linear fit is performed on all the mean of historical data of various corresponding dominant factors calculated in the time window to obtain the linear regression slope of the corresponding dominant factor. The mean of the historical data under various dominant factors at each collection moment in the time window is the average value of the corresponding dominant factors at the collection moment in the time window and all previous collection moments; If the linear regression slope of various dominant factors is less than the preset decrease judgment threshold, it is determined that the corresponding dominant factor is suddenly decreased; If the linear regression slope of various dominant factors is greater than the preset increase judgment threshold, it is determined that the corresponding dominant factor has increased suddenly.

10. A crystallization quality control system in carbon source production, implementing the crystallization quality control method in carbon source production according to any one of claims 1 to 9, the system comprising:

1. Initialization configuration module, used for system initialization and parameter configuration; 2. Data flow and functional module chain, including three sub-modules: real-time process parameter monitoring and signal optimization module, multi-dimensional feature analysis and dominant factor identification module, and dynamic process stability diagnosis and root cause tracing module; The process parameter real-time monitoring and signal optimization module is used to monitor various relevant parameter data in each observation period during the crystallization process during carbon source production; The multi-dimensional feature analysis and dominant factor identification module is used to calculate the weight assigned to each type of relevant parameter data using the entropy weight method, and to perform weighted summation on the coefficient of variation of the corresponding relevant parameter data to obtain the comprehensive weight; Principal component analysis was used to extract the principal component scores and variance contribution rates of two principal components from all relevant parameter data within an observation period, and the stability of crystalline impurities was calculated by combining the comprehensive weights. The process stability dynamic diagnosis and root cause tracing module is used to sequence all crystalline impurity stabilities calculated during all observation periods and set a sliding time window. The ratio of the mean to the standard deviation within each time window, as well as the DTW distance between the elements in the corresponding time window and the elements in the sequence, are fused and the fusion result is used as the dynamic fit score.

3. Supervisory control core module, which is used to identify abnormal conditions in the crystallization process before the current acquisition moment using dynamic fit scoring, so as to optimize control parameters in real time through model predictive control and realize multivariable collaborative supervisory control of the crystallization process.

Images