A temperature prediction method for a batch melt crystallization process

Abstract

The application discloses a temperature prediction method for a batch melt crystallization process, belongs to the field of artificial intelligence and chemical process modeling, and is characterized in that: through collecting temperature and flow instrument data in an industrial field, the data are cut and respectively used for modeling of cooling, metastable and crystallization stages of the batch melt crystallization process, parameter estimation of a mechanism model and data model training are completed, a segmented mechanism and data hybrid driving model is established, and a cooling-to-crystallization whole-cycle temperature curve and crystal growth amount prediction are given.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and chemical process modeling, specifically involving a mechanism-data hybrid driven modeling method for industrial batch melting and crystallization processes, which is applicable to the dynamic prediction of temperature curves and phase changes during crystallization processes under limited instrument measurement conditions. Background Technology

[0002] Melt crystallization, as an important separation and purification technology, is widely used in the production of high-value-added organic intermediates and polymer materials. This process achieves the separation of target components from impurities by controlling temperature changes to transfer components from the liquid phase to the solid phase, exhibiting strong coupling of heat and mass transfer. Existing melt crystallization mechanism models can describe crystal growth, simulating crystal growth through particle size measurement. However, these models rely on measured information about crystal growth behavior, making them difficult to apply directly to industrial settings with only temperature measurement points. Furthermore, they cannot autonomously identify melt phase transitions during the relatively random metastable initial crystal growth stage, and cannot continuously predict the temperature throughout the entire crystallization cycle from cooling to completion. Therefore, a technology is needed that can predict the temperature throughout the entire batch melt crystallization process under limited temperature measurement points. Summary of the Invention

[0003] To overcome the shortcomings of existing technologies, this invention provides a mechanism-data hybrid driven modeling temperature prediction method for industrial batch melting and crystallization processes. This method can predict the temperature and crystallization amount throughout the entire batch melting and crystallization process using only limited measurement point information from the industrial site, without relying on particle size measurements that describe crystal growth.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A method for predicting the temperature of a batch melt crystallization process includes the following steps: S1. Collect historical data on the batch melting and crystallization process; S2. Based on historical data from S1, a cooling phase is established through regression analysis. With temperature Linear correlation, crystallization stage With crystal layer thickness Piecewise correlation and With crystal layer thickness Average heat transfer temperature difference with crystallizer By analyzing the correlation, extracting features of the cooling stage, and training the data-driven model, the starting temperature of the metastable stage is predicted. Final temperature and temperature rebound rate ; S3, based on the starting temperature of the cooling stage and the flow rate of the utility system. Public works inlet temperature Feed quality As input, a cooling phase is adopted. The correlation formula calculates the temperature recursively based on the temperature change rate of the cooling stage, determines the termination condition based on the starting temperature of the metastable stage, and outputs the temperature curve of the cooling stage. S4. Extract the characteristic input data of the cooling stage to drive the model and obtain the starting temperature of the metastable stage. Final temperature and temperature rebound rate The temperature curve is generated according to the temperature change rate of the metastable stage, and the crystallization rate is combined with the initial crystal layer thickness of the crystallization stage to output the temperature curve of the metastable stage. S5. Using the crystallization stage initiation temperature and the initial crystal layer thickness as conditions, adopt the crystallization stage... Piecewise correlation and The correlation formula is used to recursively calculate and update the temperature step by step based on the rate of temperature change during the crystallization stage. Crystallization quality Crystal layer thickness Output the temperature curve of the crystallization stage; S6. Full-cycle curve synthesis: Combines the temperature curves of the cooling, metastable, and crystallization stages to output the full-cycle temperature curve. .

[0005] Furthermore, the process of S1 is as follows: S1.1 Collect physical property data of raw materials and utilities, including the isobaric specific heat capacity of utilities. Liquid phase heat capacity Latent heat of crystallization Solid density Crystallization area ; S1.2 Collect field instrument data from multiple batches of industrial melting and crystallization equipment during operation. Measurement points include: melt temperature curve. Public works inlet temperature curve Public works outlet temperature curve Public works flow and the quality of incoming materials ; S1.3, Sampling data is divided according to the feed and discharge of materials during the crystallization process. Batch, the j-th batch ( Each variable's running curve has I j The data, of which the first... i The data points are consistent ; S1.4 Based on the melt characteristics, the crystallization process is divided into three stages: cooling, metastable, and crystallization. S1.5, Define the starting temperature of the metastable stage. Final temperature Temperature rebound rate ; S1.6, Define the instantaneous sensible heat distribution coefficient Heat transfer capacity is defined as the ratio of the sensible heat required to cool the melt to the heat transferred from the utility. .

[0006] Furthermore, the process of S2 is as follows: S2.1 Based on the historical data of each batch cooling stage in step S1, establish the cooling stage through regression analysis. With temperature The linear correlation, in the first... Within each historical batch, the moment of any cooling phase ( Melt temperature With heat transfer capacity Can be written as The parameters are estimated using the least squares method. and , with predicted value Compared with actual value Taking the sum of squared errors as the objective, and setting the objective function for parameter estimation, we can solve for the parameters to be estimated. , ; S2.2 Based on the historical data of each batch crystallization stage in step S1, establish the crystallization stage. With crystal layer thickness The relational expression; S2.3. Based on the historical data of each batch crystallization stage in step S1, establish a regression analysis system. With crystal layer thickness Average heat transfer temperature difference with crystallizer The correlation is used to calculate the average heat transfer temperature difference in the crystallizer; S2.4. Using the sliding window method, extract the features of the XGBoost (eXtreme Gradient Boosting) model from the historical data of the cooling stage in step S1, with the melt temperature decrease as the step size. The inputs are as follows: temperature. Interval time, corresponding utility inlet temperature change, corresponding utility inlet and outlet average temperature difference, corresponding crystallizer average heat transfer temperature difference, corresponding equivalent heat storage coefficient and corresponding apparent thermal resistance; Three variables in the metastable phase: onset temperature Temperature rebound rate and endpoint temperature As the output of the metastable stage temperature prediction data model; Historical datasets are randomly split into training and test sets according to a preset ratio (e.g., 8:2 or 7:3). K-fold cross-validation (k ranges from 5 to 10) is used to adjust model hyperparameters to optimize generalization performance. Hyperparameters include: number of decision trees (range: 1-1000) and decision tree depth (range: 3-10). The model training termination condition simultaneously satisfies the correlation coefficient R on the training set. 2 ≥0.95, R0.95 for both training and test sets 2 The difference is ≤0.05, and the root mean square error (RMSE) on the test set is ≤0.1℃.

[0007] The trained metastable phase data-driven model is used to determine the metastable phase initiation temperature of a new batch. Temperature rebound rate and endpoint temperature predict.

[0008] Preferably, in S2.2, the crystallization stage With crystal layer thickness Establishing a relational relationship involves the following three steps: a. Data on the variation of crystal layer thickness over time during the crystallization stage in each batch. Historical melt temperature data for each batch can be used. Public works inlet and outlet temperature and Public works flow Feed quality The crystallization mass was calculated. ; b. Solving for the crystal layer thickness ; c. Establishing heat transfer capacity during the crystallization stage With crystal layer thickness The correlation formula, which adopts a piecewise function form, has a first threshold. The range of values ​​depends on the physical properties, the second threshold. The range of values ​​depends on the physical properties; Parameters estimated using the least squares method , with predicted value Compared with actual value Taking the sum of squared errors as the objective, we define the objective function for parameter estimation and solve for the parameters to be estimated. , , , , , , , , , , , .

[0009] In S2.3, the objective function for parameter estimation is the predicted value. Compared with actual value The sum of squared errors is used to define the objective function for parameter estimation. The parameters to be estimated are then solved. , , , , .

[0010] Furthermore, the process of S3 is as follows: S3.1 The online measurable input variables for the predicted batch include: the starting temperature of the cooling stage. Public works inlet temperature Public works flow ; S3.2 Solve for the melt temperature change rate during the cooling stage and the outlet temperature of the utility; Starting from the start time and start temperature of the cooling phase, perform the following operations at each time step: Based on the current melt temperature First, solve for the cooling stage. Relational; Substitute the value into the melt temperature change rate, perform forward integration with a fixed step size, and gradually update the melt temperature for the next time step. Repeat the above process to obtain the temperature change curve of the cooling stage; the termination temperature and time are determined by the metastable stage starting point predicted in step S4.

[0011] The process of S4 is as follows: S4.1. According to the feature input requirements of the metastable stage temperature prediction model, extract the feature input required for the metastable stage from the online measurable data of the predicted batch cooling stage, and input it into the trained metastable stage temperature prediction model to obtain the starting temperature of the metastable stage. Temperature rebound rate and endpoint temperature and will The cooling stage predicted in step S3 is taken as the endpoint; S4.2. Define the temperature change equation and crystallization rate of the metastable stage, and use the metastable stage temperature prediction model to obtain the temperature curve and accumulated crystal mass of the metastable stage. The thickness of the final crystal layer in the metastable stage is converted into the final crystal layer thickness, and the crystallization stage is assigned as the initial crystal layer thickness, with the end time of the metastable stage being taken as the start time of the crystallization stage.

[0012] The process of S5 is as follows: S5.1 The predicted on-site measurable variables for the new batch include: utility inlet temperature. Public works flow The starting temperature and initial crystal layer thickness of the crystallization stage are determined by the prediction results of the metastable stage in step S4. S5.2 Solve for the melt temperature change rate and crystallization rate during the crystallization stage. Starting from the initial temperature and initial crystal layer thickness of the crystallization stage, perform the following operations at each time step: Based on the current crystal layer thickness ,use correlation and Associative computation crystallization stage and instantaneous sensible heat distribution coefficient .Will Substituting the melt temperature change rate formula and the crystallization rate formula, and performing forward integration with a fixed step size, the cumulative crystallization mass at the next time step is calculated. Melt temperature and crystal layer thickness Repeat the above process until the melt temperature reaches the batch final temperature specified by the device.

[0013] In step S6, the process of synthesizing the full-cycle temperature curve is as follows: the three temperature curves predicted by S3, S4, and S5 are merged to obtain a continuous full-cycle melt temperature trajectory. .

[0014] The beneficial effects of this invention are as follows: This method can provide a temperature curve and crystal growth prediction for the entire cooling to crystallization cycle using only temperature measurement information that is easily obtained in industrial settings, without the need to add crystal growth measuring instruments. The prediction can achieve engineering accuracy and is simple to implement. It can be used for the control and optimization of batch crystallization processes, has strong industrial applicability, and has good application prospects. Attached Figure Description

[0015] Figure 1 This is a flowchart of a temperature prediction method for a batch melting and crystallization process according to the present invention; Figure 2 This is a schematic diagram of the measuring points of the crystallization apparatus; Figure 3 A schematic diagram showing the stage division of the melt temperature-time curve; Figure 4 This is a temperature curve for the inlet of a public utility.

[0016] Figure 5 This is a comparison chart of the measurement and prediction of the melt temperature trajectory throughout the entire cycle. Detailed Implementation

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Reference Figures 1-5 A method for predicting the temperature of a batch melting and crystallization process, the flowchart of which is shown below. Figure 1 This includes the following steps: S1. Collect historical data on the batch melting and crystallization process. The process is as follows: S1.1 Collect physical property data of raw materials and utilities, including the isobaric specific heat capacity of utilities. Liquid phase heat capacity Latent heat of crystallization Solid density Crystallization area ; S1.2 Collect field instrument data from multiple batches of industrial melting and crystallization equipment during operation. Measurement points include: melt temperature curve. Public works inlet temperature curve Public works outlet temperature curve Public works flow and the quality of incoming materials ; S1.3, Sampling data is divided according to the feed and discharge of materials during the crystallization process. Batch, the j-th batch ( Each variable's running curve has I j The data, of which the first... i The data points are consistent ; S1.4 Based on the melt characteristics, the crystallization process is divided into three stages: cooling, metastable, and crystallization. S1.5, Define the starting temperature of the metastable stage. Final temperature Temperature rebound rate ; S1.6, Define the instantaneous sensible heat distribution coefficient The ratio of the sensible heat required to cool the melt to the heat transferred from the utility works is calculated using the following formula: ; Define heat transfer capacity Calculate using the following formula: .

[0019] This embodiment uses data collected from on-site instruments during the industrial crystallization process of a chemical enterprise over two months as a specific implementation method. Physical property data is obtained using actual raw materials and equipment on-site, including: heat capacity of utilities. Liquid phase heat capacity Latent heat of crystallization Solid density As attached Figure 2As shown, online measurable variables are obtained, including: melt temperature profile. Public works inlet temperature curve Public works outlet temperature curve Public works flow and the quality of incoming materials The obtained sampling data was batch-divided, resulting in a total of 100 batches of operational data. Based on the characteristics of the melt temperature profile, the process was divided into three stages: cooling, metastable, and crystallization. Figure 3 As shown.

[0020] The sensible heat distribution coefficient is calculated using the following formula. : ; The heat transfer capacity is calculated using the following formula. : ; S2. Based on historical data from S1, a cooling phase is established through regression analysis. With temperature Linear correlation, crystallization stage With crystal layer thickness Piecewise correlation and With crystal layer thickness Average heat transfer temperature difference with crystallizer By analyzing the correlation, extracting features of the cooling stage, and training the data-driven model, the starting temperature of the metastable stage is predicted. Final temperature and temperature rebound rate The process is as follows: S2.1 Based on the historical data of each batch cooling stage in step S1, establish the cooling stage through regression analysis. With temperature The linear correlation is as follows: ; In the formula , These are parameters to be estimated.

[0021] In the Within each historical batch, the moment of any cooling phase ( Melt temperature With heat transfer capacity Can be written as The parameters are estimated using the least squares method. and , with predicted value Compared with actual value The objective function for parameter estimation is as follows: (The sum of squared errors is the target.) ; The parameters to be estimated can be obtained by solving the problem. , ; S2.2 Based on the historical data of each batch crystallization stage in step S1, establish the crystallization stage. With crystal layer thickness The relational expression; Crystallization stage With crystal layer thickness Establishing a relational relationship involves the following three steps: a. Data on the variation of crystal layer thickness over time during the crystallization stage in each batch. Historical melt temperature data for each batch can be used. Public works inlet and outlet temperature and Public works flow Feed quality The crystallization mass is calculated using the following formula. : ; In the formula This is the start time of the crystallization stage. This is the end time of the crystallization stage.

[0022] b. Solving for the crystal layer thickness : ; c. Establishing heat transfer capacity during the crystallization stage With crystal layer thickness The correlation expression. The correlation expression takes the form of a piecewise function as shown below: Section I When the crystal layer thickness is less than hour: ; Section II When the crystal layer thickness is to In between: ; Section III When the crystal layer thickness is greater than hour: ; In the formula , , , , , , , , , , , The parameter to be estimated; The first threshold The range of values ​​depends on the physical properties, the second threshold. The range of values ​​depends on the physical properties; Parameters estimated using the least squares method , with predicted value Compared with actual value The objective function for parameter estimation is as follows: (The sum of squared errors is the target.) ; The parameters to be estimated can be obtained by solving the problem. , , , , , , , , , , , .

[0023] S2.3. Based on the historical data of each batch crystallization stage in step S1, establish a regression analysis system. With crystal layer thickness Average heat transfer temperature difference with crystallizer The correlation is as follows: ; Average heat transfer temperature difference in crystallizer: ; In the formula , , , , The parameter to be estimated; The objective function for parameter estimation is the predicted value. Compared with actual value The sum of squared errors and the objective function for parameter estimation are as follows: ; Solving for the parameters to be estimated , , , , .

[0024] S2.4. Using the sliding window method with the melt temperature drop as the step size, extract the features of the XGBoost (eXtreme Gradient Boosting) model from the historical data of the cooling stage in step S1. The input is as follows: temperature Interval duration: ; Corresponding changes in inlet temperature of public works: ; Average temperature difference between the inlet and outlet of the corresponding public works: ; The average heat transfer temperature difference of the corresponding crystallizer: ; Corresponding equivalent heat storage coefficient: ; Corresponding apparent thermal resistance: ; Represents temperature The relative time within the interval, Represents temperature The maximum relative time within the interval; Three variables in the metastable phase: onset temperature Temperature rebound rate and endpoint temperature As the output of the metastable stage temperature prediction data model; Historical datasets are randomly split into training and test sets according to a preset ratio (e.g., 8:2 or 7:3). K-fold cross-validation (k ranges from 5 to 10) is used to adjust model hyperparameters to optimize generalization performance. Hyperparameters include: number of decision trees (range: 1-1000) and decision tree depth (range: 3-10). The model training termination condition simultaneously satisfies the correlation coefficient R on the training set. 2 ≥0.95, R0.95 for both training and test sets 2 The difference is ≤0.05, and the root mean square error (RMSE) on the test set is ≤0.1℃.

[0025] The trained metastable phase data-driven model is used to determine the metastable phase initiation temperature of a new batch. Temperature rebound rate and endpoint temperature predict.

[0026] In this embodiment, parameters are obtained through estimation during the cooling stage. Relational: ; Obtained through parameter estimation during the crystallization stage. as well as Relational: ; ; ; ; The cooling stage data were discretized using a sliding window method at melt temperatures of 141℃ to... The interval is divided into zones with a step size of 1℃. For each interval, a set of dynamic physical features is extracted, including the interval duration. ), temperature change at the inlet of public works ( ), average temperature difference between the inlet and outlet of public works ( ), average heat transfer temperature difference of crystallizer ( ), equivalent heat storage coefficient ( ), apparent thermal resistance ( ).

[0027] The dataset was divided into training and validation sets in an 8:2 ratio, and trained using an XGBoost model. Hyperparameters were tuned using grid search and four-fold cross-validation. The final hyperparameters were determined as follows: 350 decision trees, a maximum depth of 5 layers, and a learning rate of 0.01. The trained model outputs three key variables for the metastable phase: initial temperature. Temperature rebound rate Final temperature .

[0028] S3, based on the starting temperature of the cooling stage and the flow rate of the utility system. Public works inlet temperature Feed quality As input, a cooling phase is adopted. The correlation algorithm recursively calculates the temperature based on the rate of temperature change during the cooling phase, determines the termination condition based on the starting temperature of the metastable phase, and outputs the temperature curve for the cooling phase. The process is as follows: S3.1 The online measurable input variables for the predicted batch include: the starting temperature of the cooling stage. Public works inlet temperature Public works flow ; S3.2 Calculate the melt temperature change rate during the cooling stage using the following formula: ; And the temperature at the outlet of the public works: ; Starting from the start time and start temperature of the cooling phase, perform the following operations at each time step: Based on the current melt temperature First, solve for the cooling stage. Relational; Substitute the value into the melt temperature change rate, perform forward integration with a fixed step size, and gradually update the melt temperature for the next time step. Repeat the above process to obtain the temperature change curve of the cooling stage; the termination temperature and time are determined by the metastable stage starting point predicted in step S4.

[0029] In this embodiment, the on-site measurable variables for the predicted new batch include: the starting temperature of the cooling stage. Public works inlet temperature Public works flow Solve for the melt temperature change rate during the crystallization stage and the outlet temperature of the utility, combined with the cooling stage. The correlation is used to gradually update the melt temperature until the metastable stage starting temperature is reached.

[0030] S4. Extract the characteristic input data of the cooling stage to drive the model and obtain the starting temperature of the metastable stage. Final temperature and temperature rebound rate Temperature curves are generated based on the rate of temperature change during the metastable stage. These curves are then converted to the initial crystal layer thickness during the crystallization stage using the crystallization rate, and the metastable stage temperature curves are output. The process is as follows: S4.1. According to the feature input requirements of the metastable stage temperature prediction model, extract the feature input required for the metastable stage from the online measurable data of the predicted batch cooling stage, and input it into the trained metastable stage temperature prediction model to obtain the starting temperature of the metastable stage. Temperature rebound rate and endpoint temperature and will The cooling stage predicted in step S3 is taken as the endpoint; S4.2, The temperature change equation during the metastable stage is obtained as follows: ; And the crystallization rate in the metastable stage: ; Integrating the above equation allows us to obtain the temperature curve of the metastable stage and the accumulated crystal mass during the metastable stage using a metastable stage temperature prediction model. The thickness of the final crystal layer in the metastable stage is converted into the final crystal layer thickness, and the crystallization stage is assigned as the initial crystal layer thickness, with the end time of the metastable stage being taken as the start time of the crystallization stage.

[0031] S5. Using the crystallization stage initiation temperature and the initial crystal layer thickness as conditions, adopt the crystallization stage... Piecewise correlation and The correlation formula is used to recursively calculate and update the temperature step by step based on the rate of temperature change during the crystallization stage. Crystallization quality Crystal layer thickness Output the temperature curve of the crystallization stage; the process is as follows: S5.1 The predicted on-site measurable variables for the new batch include: utility inlet temperature. Public works flow The starting temperature and initial crystal layer thickness of the crystallization stage are determined by the prediction results of the metastable stage in step S4. 5.2. Calculate the melt temperature change rate during the crystallization stage using the following formula: ; And the crystallization rate during the crystallization stage: ; Starting from the initial temperature and initial crystal layer thickness of the crystallization stage, perform the following operations at each time step: Based on the current crystal layer thickness ,use correlation and Associative computation crystallization stage and instantaneous sensible heat distribution coefficient .Will Substituting the melt temperature change rate formula and the crystallization rate formula, and performing forward integration with a fixed step size, the cumulative crystallization mass at the next time step is calculated. Melt temperature and crystal layer thickness Repeat the above process until the melt temperature reaches the batch final temperature specified by the device.

[0032] In this embodiment, known measuring points during the crystallization stage include the inlet temperature of the utility works. Public works flow Using the end temperature of the metastable stage and the initial crystal layer thickness as initial conditions, combined with the crystallization stage... as well as Correlational, progressively updating the melt temperature at the next time step. Crystallization quality and crystal layer thickness Continue until the melt temperature reaches the batch final temperature specified by the device.

[0033] S6. Full-cycle curve synthesis: Combines the temperature curves of the cooling, metastable, and crystallization stages to output the full-cycle temperature curve. .

[0034] In this embodiment, the temperature curves of the cooling stage, metastable stage, and crystallization stage are combined to obtain a complete melt temperature prediction curve. Thus, as... Figure 4 and Figure 5 As shown, by Figure 4 Predictable inlet temperature profiles for public works projects Figure 5Full-cycle melt temperature trajectory, Figure 5 The high degree of agreement between the predicted and measured values ​​indicates that the model has good predictive ability and is sufficient to support engineering decisions.

[0035] The embodiments described in this specification are merely examples of implementations of the inventive concept and are for illustrative purposes only. The scope of protection of this invention should not be considered limited to the specific forms described in these embodiments; rather, it extends to equivalent technical means conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for predicting the temperature of a batch melting and crystallization process, characterized in that, The method includes the following steps: S1. Collect historical data on the batch melting and crystallization process; S2. Based on historical data from S1, a cooling phase is established through regression analysis. With temperature Linear correlation, crystallization stage With crystal layer thickness Piecewise correlation and With crystal layer thickness Average heat transfer temperature difference between the crystallizer and the crystallizer By analyzing the correlation, extracting features of the cooling stage, and training the data-driven model, the starting temperature of the metastable stage is predicted. Final temperature and temperature rebound rate ; S3, based on the starting temperature of the cooling stage and the flow rate of the utility system. Public works inlet temperature Feed quality As input, a cooling phase is adopted. The correlation formula calculates the temperature recursively based on the temperature change rate of the cooling stage, determines the termination condition based on the starting temperature of the metastable stage, and outputs the temperature curve of the cooling stage. S4. Extract the characteristic input data of the cooling stage to drive the model and obtain the starting temperature of the metastable stage. Final temperature and temperature rebound rate The temperature curve is generated according to the temperature change rate of the metastable stage, and the crystallization rate is combined with the initial crystal layer thickness of the crystallization stage to output the temperature curve of the metastable stage. S5. Using the crystallization stage initiation temperature and the initial crystal layer thickness as conditions, adopt the crystallization stage... Piecewise correlation and The correlation formula is used to recursively calculate and update the temperature step by step based on the rate of temperature change during the crystallization stage. Crystallization quality Crystal layer thickness Output the temperature curve of the crystallization stage; S6. Full-cycle curve synthesis: Combines the temperature curves of the cooling, metastable, and crystallization stages to output the full-cycle temperature curve. .

2. The temperature prediction method for a batch melting and crystallization process as described in claim 1, characterized in that, The process of S1 is as follows: S1.1 Collect physical property data of raw materials and utilities, including the isobaric specific heat capacity of utilities. Liquid phase heat capacity Latent heat of crystallization Solid density Crystallization area ; S1.2 Collect field instrument data from multiple batches of industrial melting and crystallization equipment during operation. Measurement points include: melt temperature curve. Public works inlet temperature curve Public works outlet temperature curve Public works flow and the quality of incoming materials ; S1.3, Sampling data is divided according to the feed and discharge of materials during the crystallization process. In each batch, the running curve for each variable in the j-th batch has I. j One data point, , of which i The data points are consistent ; S1.4 Based on the melt characteristics, the crystallization process is divided into three stages: cooling, metastable, and crystallization. S1.5, Define the starting temperature of the metastable stage. Final temperature Temperature rebound rate ; S1.6, Define the instantaneous sensible heat distribution coefficient Heat transfer capacity is defined as the ratio of the sensible heat required to cool the melt to the heat transferred from the utility. .

3. The temperature prediction method for a batch melting and crystallization process as described in claim 1 or 2, characterized in that, The process of S2 is as follows: S2.1 Based on the historical data of each batch cooling stage in step S1, establish the cooling stage through regression analysis. With temperature The linear correlation; In the Within each historical batch, the moment of any cooling phase ( Melt temperature With heat transfer capacity Can be written as The parameters are estimated using the least squares method. and , with predicted value Compared with actual value Given the sum of squared errors as the objective, and providing the objective function for parameter estimation, the parameters to be estimated can be obtained by solving it. , ; S2.2 Based on the historical data of each batch crystallization stage in step S1, establish the crystallization stage. With crystal layer thickness The correlation; S2.

3. Based on the historical data of each batch crystallization stage in step S1, establish a regression analysis system. With crystal layer thickness Average heat transfer temperature difference between the crystallizer and the crystallizer The correlation is used to calculate the average heat transfer temperature difference in the crystallizer; S2.

4. Using the sliding window method, extract the features of the XGBoost model from the historical data of the cooling stage in step S1, with the melt temperature decrease as the step size. The inputs are as follows: temperature. Interval time, corresponding utility inlet temperature change, corresponding utility inlet and outlet average temperature difference, corresponding crystallizer average heat transfer temperature difference, corresponding equivalent heat storage coefficient and corresponding apparent thermal resistance; Three variables in the metastable phase: onset temperature Temperature rebound rate and endpoint temperature As the output of the metastable stage temperature prediction data model; The historical dataset is randomly split into training and test sets according to a preset ratio. K-fold cross-validation is used to adjust the model's hyperparameters to optimize its generalization performance. Hyperparameters include the number of decision trees and their depth. The model training terminates when the correlation coefficient R on the training set is simultaneously satisfied. 2 ≥0.95, R0.95 for both training and test sets 2 The difference ≤ 0.05, and the root mean square error (RMSE) on the test set ≤ 0.1℃; The trained metastable phase data-driven model is used to determine the metastable phase initiation temperature for a new batch. Temperature rebound rate and endpoint temperature predict.

4. The temperature prediction method for a batch melting and crystallization process as described in claim 3, characterized in that, In S2.2, the crystallization stage With crystal layer thickness Establishing a relational relationship involves the following three steps: a. Data on the variation of crystal layer thickness over time during the crystallization stage in each batch. Historical melt temperature data for each batch can be used. Public works inlet and outlet temperatures and Public works flow Feed quality The crystallization mass was calculated. ; b. Solving for the crystal layer thickness ; c. Establishing heat transfer capacity during the crystallization stage With crystal layer thickness The correlation formula, which adopts a piecewise function form, has a first threshold. The range of values ​​depends on the physical properties, the second threshold. The range of values ​​depends on the physical properties; the least squares method is used to estimate the parameters. , with predicted value Compared with actual value The objective function for parameter estimation is given by taking the sum of squared errors as the objective; the parameters to be estimated can be obtained by solving it.

5. The temperature prediction method for a batch melting and crystallization process as described in claim 3, characterized in that, In S2.3, the objective function for parameter estimation is the predicted value. Compared with actual value The sum of squared errors is used to define the objective function for parameter estimation, and the parameters to be estimated are obtained by solving the problem.

6. The temperature prediction method for a batch melting and crystallization process as described in claim 1 or 2, characterized in that, The process of S3 is as follows: S3.1 The online measurable input variables for the predicted batch include: the starting temperature of the cooling stage. Public works inlet temperature Public works flow ; S3.2 Solve for the melt temperature change rate during the cooling stage and the outlet temperature of the utility; Starting from the beginning time and temperature of the cooling phase, perform the following operations at each time step: based on the current melt temperature First, solve for the cooling stage. Relational; Substitute the value into the melt temperature change rate, perform forward integration with a fixed step size, and gradually update the melt temperature for the next time step. Repeat the above process to obtain the temperature change curve of the cooling stage; the termination temperature and time are determined by the metastable stage starting point predicted in step S4.

7. A method for predicting the temperature of a batch melting and crystallization process as described in claim 1 or 2, characterized in that, The process of S4 is as follows: S4.

1. According to the feature input requirements of the metastable stage temperature prediction model, extract the feature input required for the metastable stage from the online measurable data of the predicted batch cooling stage, and input it into the trained metastable stage temperature prediction model to obtain the starting temperature of the metastable stage. Temperature rebound rate and endpoint temperature and will The cooling stage predicted in step S3 is taken as the endpoint. S4.2 Calculate the temperature change and crystallization rate during the metastable stage, and use the metastable stage temperature prediction model to obtain the temperature curve and accumulated crystal mass during the metastable stage. The thickness of the final crystal layer in the metastable stage is converted into the final crystal layer thickness, and the crystallization stage is assigned as the initial crystal layer thickness, with the end time of the metastable stage being taken as the start time of the crystallization stage.

8. A method for predicting the temperature of a batch melting and crystallization process as described in claim 1 or 2, characterized in that, The process of S5 is as follows: S5.1 The predicted on-site measurable variables for the new batch include: utility inlet temperature. Public works flow The starting temperature and initial crystal layer thickness of the crystallization stage are determined by the prediction results of the metastable stage in step S4. S5.2 Solve for the melt temperature change rate and the crystallization rate during the crystallization stage; Starting from the initial temperature and initial crystal layer thickness of the crystallization stage, perform the following operations at each time step: Based on the current crystal layer thickness ,use correlation and Associative computation crystallization stage and instantaneous sensible heat distribution coefficient ;Will Substituting the melt temperature change rate formula and the crystallization rate formula, and performing forward integration with a fixed step size, the cumulative crystallization mass at the next time step is calculated. Melt temperature and crystal layer thickness Repeat the above process until the melt temperature reaches the batch final temperature specified by the device.

9. A method for predicting the temperature of a batch melting and crystallization process as described in claim 1 or 2, characterized in that, In step S6, the process of synthesizing the full-cycle temperature curve is as follows: the three temperature curves predicted by S3, S4, and S5 are merged to obtain a continuous full-cycle melt temperature trajectory. .

Images