Thermostatic purification system based on diphenyl monoisodecyl phosphite and control method thereof
By combining a BP neural network model with optimization algorithms, the problem of lagging quality control during the purification of diphenyl isodecyl phosphite was solved, enabling real-time dynamic optimization of product quality and automatic adjustment of globally optimal process parameters, thereby improving product consistency and production efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHANGHE CHEM NEW MATERIAL (JIANGSU) CO LTD
- Filing Date
- 2026-02-25
- Publication Date
- 2026-06-05
AI Technical Summary
In the existing technology, the purification process of diphenyl isodecanyl phosphite relies on the operator's experience, which leads to a lag in the adjustment of core product quality indicators, making it impossible to achieve real-time and forward-looking quality control, and making it difficult to find the globally optimal process parameters among multiple mutually restrictive indicators.
A isothermal purification system based on BP neural network was constructed. Through data acquisition and preprocessing, a training dataset was established to train a product quality prediction model. The optimization algorithm was used to adjust the process parameters in real time to achieve dynamic optimization control of product quality.
It achieves high batch-to-batch stability and consistency in product quality, reduces operator training costs, and improves the accuracy of real-time control and global optimization capabilities of product quality.
Smart Images
Figure CN122152008A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of purification technology for diphenyl isodecyl phosphite, specifically to a constant-temperature purification system and control method based on diphenyl isodecyl phosphite. Background Technology
[0002] Diphenyl isodecyl phosphite is an important asymmetric organic phosphite, widely used as an antioxidant and stabilizer intermediate for high-performance polymers, as well as a flame retardant plasticizer. Its product purity, acid value, color and other key indicators directly affect the performance and quality of downstream products.
[0003] Industrially, diphenyl isodecanyl phosphite is usually synthesized by transesterification or stepwise esterification. This results in the crude product containing impurities such as unreacted raw materials, symmetrical ester byproducts, catalyst residues, and trace amounts of moisture. Therefore, efficient purification of the crude product is a key step to ensure product quality.
[0004] Currently, the purification of diphenyl isodecyl phosphite mainly relies on vacuum distillation or molecular distillation techniques. However, current production control is mostly based on the experience of operators or limited data to pre-set a fixed set of process parameters. This results in the core indicators of product quality relying on offline laboratory analysis, with the analysis results lagging significantly behind the production process. Operators can only make adjustments after the fact based on the lagging and limited data, making it impossible to achieve real-time and forward-looking quality control. At the same time, there are various uncertainties in actual production. Even if operators adjust the process parameters, it is difficult to select the globally optimal indicator among multiple interdependent indicators such as purity, acid value, color, yield, and energy consumption with a single adjustment. Therefore, to address the above problems, a constant temperature purification system based on diphenyl isodecyl phosphite and its control method are proposed. Summary of the Invention
[0005] The purpose of this invention is to provide a constant-temperature purification system and control method based on diphenyl isodecyl phosphite to solve the problems mentioned in the background art.
[0006] To achieve the above objectives, the present invention provides the following technical solution:
[0007] The control method for isothermal purification of diphenyl isodecyl phosphite includes the following steps: Step 1: Data acquisition and preprocessing: During the isothermal purification production process of diphenyl isodecyl phosphite, multiple batches of historical production data are collected. The historical production data includes multiple process parameters affecting the purification process as input features, and multiple evaluation indicators characterizing the purification effect as output labels; the historical production data is preprocessed to construct a neural network training dataset.
[0008] Step 2: Neural Network Modeling and Training: Construct a BP neural network model, using the input features as the input layer and the output labels as the output layer; train the BP neural network model using the training dataset to obtain a trained product quality prediction model;
[0009] Step 3: Online optimization and control: For the current purification batch to be optimized, obtain its initial conditions; input the initial conditions and preset quality targets into the trained product quality prediction model, and solve for a set of optimal process parameter settings through optimization algorithms; send the optimal process parameter settings to the production control system to control the isothermal purification process of the current batch.
[0010] As a further optimization of the present invention, in step 1, the process parameters include at least four of the following: constant temperature, system vacuum, feed rate, reflux ratio, and protective gas flow rate; the evaluation indicators include at least two of the following: product main component purity, product acid value, and product color.
[0011] As a further optimization of the present invention, the evaluation index further includes at least one of product yield, unit energy consumption, and product moisture content.
[0012] As a further optimization of the present invention, in step 2, when training the BP neural network model, a weighted loss function is used, in which higher weights are given to the purity of the main components of the product and the acid value index of the product.
[0013] As a further optimization of the present invention, in step 3, the optimization algorithm is one of genetic algorithm, particle swarm optimization algorithm or gradient descent method; the objective function of the optimization algorithm is to minimize the comprehensive deviation between the predicted evaluation index and the preset quality target, and to satisfy the operational constraints and safety boundaries of the process parameters.
[0014] As a further optimization of the present invention, in step 3, the online optimization and control process is executed periodically using a model predictive control framework: during the purification process, real-time process data is re-collected at a set period as the current initial condition, the optimization steps are repeated, and new optimal process parameter settings are dynamically updated and issued until the purification process ends.
[0015] As a further optimization of the present invention, it also includes step 4: model update: the final actual process parameters and actual evaluation index data of the completed batch are added to the training dataset as new sample data; when the new sample data accumulates to a preset number or the model prediction error exceeds a threshold, the retraining and update of the product quality prediction model is triggered.
[0016] The isothermal purification system based on diphenyl isodecyl phosphite includes a data acquisition and processing module for acquiring and preprocessing historical and real-time production data; a neural network modeling and training module for constructing, training, and storing the product quality prediction model; an online optimization and control module for executing optimization algorithms, calculating optimal process parameter settings, and generating control commands; and a control execution module for receiving the control commands and driving the production equipment to perform corresponding parameter adjustments.
[0017] As a further optimization of the present invention, the system further includes a soft measurement module, which constructs a soft measurement model for real-time estimation of key evaluation indicators based on online sensor data and intermediate layer features of the product quality prediction model.
[0018] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.
[0019] Compared with the prior art, the beneficial effects of the present invention are:
[0020] 1. In this invention, a well-trained neural network model can accurately predict the optimal combination of process parameters under different initial conditions and disturbances, and make dynamic adjustments in real time. This effectively overcomes the shortcomings of the traditional fixed parameter method in terms of poor adaptability, ensures that the fluctuation range of the core indicators of products in different production batches is significantly narrowed, achieves high stability and consistency of product quality between batches, and meets the stringent requirements for indicators.
[0021] 2. In this invention, multiple mutually restrictive indicators such as purity, acid value, color, product yield, and unit energy consumption can be simultaneously incorporated into the output of the neural network and the optimization objective function. The optimization algorithm performs global optimization based on model prediction, and quickly and accurately finds the process operation mode that maximizes the yield or minimizes energy consumption while meeting the core quality threshold.
[0022] 3. In this invention, the implicit experience and knowledge of operators are transformed into a quantifiable, reproducible, and self-optimizable neural network model, which enables the system to automatically provide near-optimal process settings, reducing the training cost of operators and the risk of human error. Attached Figure Description
[0023] Figure 1 This is a system structure diagram of the present invention;
[0024] Figure 2 This is a flowchart illustrating the optimization process of the BP neural network in this invention. Detailed Implementation
[0025] Please see Figures 1-2 The present invention provides the following embodiments:
[0026] Example 1:
[0027] In the isothermal purification process of diphenyl isodecyl phosphite, multiple batches of historical production data were collected. The historical production data included multiple process parameters affecting the purification process as input features, and multiple evaluation indicators characterizing the purification effect as output labels. The historical production data were preprocessed to construct a neural network training dataset.
[0028] Where: input features It must cover measurable, controllable variables that have a significant impact on purification results;
[0029] Optional parameters include constant temperature, system vacuum, feed rate, reflux ratio, stirring intensity, protective gas flow rate, raw material batch number, initial main component content of raw material, initial acid value of raw material, raw material color, raw material moisture content, cooling medium inlet temperature, and continuous operating time of the equipment.
[0030] As a preferred option, input features It includes at least four of the following: constant temperature, system vacuum, feed rate, reflux ratio, and protective gas flow rate.
[0031] Furthermore, output tags It must cover the core indicators of final product quality and process economy;
[0032] Optional evaluation indicators include the purity of the main product components, the acid value of the product, the color of the product, and the moisture content of the product.
[0033] As a preferred option, the evaluation indicators should include at least two of the following: purity of the main product component, acid value, and color; the evaluation indicators should also include at least one of the following: product yield, unit energy consumption, and moisture content.
[0034] Furthermore, batches of data with obvious operational errors are removed, and all features of the remaining data are then analyzed. The dataset is standardized and randomly divided into training, validation, and test sets.
[0035] In a further implementation, a BP neural network model is constructed to take the input features as input. As an input layer, to output labels As the output layer, the BP neural network model is trained using the training dataset to obtain a trained product quality prediction model. When training the BP neural network model, a weighted loss function is used, in which the purity of the main components of the product and the acid value of the product are given higher weights.
[0036] As a preferred approach, a three-layer backpropagation neural network was constructed using the PyTorch framework in Python, employing the Adam optimizer with an initial learning rate of 0.001. The loss function was weighted mean squared error, with purity and acidity assigned a weight of 2.0, and color and moisture assigned a weight of 1.0. The network was trained on the training set for 500 epochs, and the loss was calculated on the validation set after every 10 epochs. When the validation set loss no longer decreased for 30 consecutive epochs, early stopping was triggered, and the optimal model was saved.
[0037] Example 2:
[0038] In a further implementation scheme, for the current purification batch to be optimized, its initial conditions are obtained; the initial conditions and preset quality objectives are input into the trained product quality prediction model, and a set of optimal process parameter settings are solved through an optimization algorithm; the optimal process parameter settings are sent to the production control system to control the isothermal purification process of the current batch, and the optimization algorithm is one of genetic algorithm, particle swarm optimization algorithm, or gradient descent method; the objective function of the optimization algorithm is to minimize the comprehensive deviation between the predicted evaluation index and the preset quality objective, and to satisfy the operational constraints and safety boundaries of the process parameters; the online optimization and control process is executed periodically using a model predictive control framework: during the purification process, real-time process data is re-collected at a set period as the current initial conditions, the optimization steps are repeated, and new optimal process parameter settings are dynamically updated and sent out until the purification process ends.
[0039] Specifically, the initial acid value of the crude raw material in this batch was measured, and the optimization target was set as purity ≥99.5% and acid value ≤0.05mg KOH / g, while maximizing the moisture content of the product under this constraint.
[0040] In the industrial control computer software, a trained neural network model is called as a predictor, and a particle swarm optimization algorithm is used as an optimizer to search for the optimal combination of parameters that satisfy the equipment safety constraints.
[0041] After iterative calculation using the particle swarm optimization algorithm, the first set of optimal settings was obtained: the evaporator constant temperature was 198.5℃, the system working pressure was 1.2Pa, the feed rate was 1.8mL / min, the stirring rate was 220rpm, and the protective gas flow rate was 0.3L / min.
[0042] Furthermore, the set values are sent to the PLC, which automatically adjusts each actuator, starts the isothermal purification, and initiates the MPC cycle with a cycle set to 30 minutes. After 30 minutes, the average process parameters of the previous cycle are collected as the new "current state", and the above optimization solution is executed again. After calculating the second set of set values, the isothermal purification operation is repeated.
[0043] Repeat this process until distillation is complete.
[0044] Furthermore, the final actual process parameters and actual evaluation index data of the completed batches are used as new sample data and added to the training dataset;
[0045] Preferably, the complete process data and final test results of this batch are stored as a new sample in the historical database; the system is set to trigger retraining when the cumulative number of new samples reaches 20 or the average prediction error of the model for the most recent 5 batches exceeds 10%.
[0046] Example 3:
[0047] The isothermal purification system for diphenyl isodecanyl phosphite includes a data acquisition and processing module for collecting and preprocessing historical and real-time production data; a neural network modeling and training module for building, training, and storing product quality prediction models; an online optimization and control module for executing optimization algorithms, calculating optimal process parameter setpoints, and generating control commands; a control execution module for receiving control commands and driving production equipment to perform corresponding parameter adjustments; and a soft measurement module, which constructs a soft measurement model for real-time estimation of key evaluation indicators based on online sensor data and intermediate layer features of the product quality prediction model.
[0048] The system is comprised of: a data acquisition and processing module consisting of a field sensor network, a data acquisition card, and a data processing unit in an industrial computer or DCS system; a neural network modeling and training module, which is a software service running on a high-performance server or cloud computing platform; an online optimization and control module, which is the core decision-making unit of the system; and a control execution module, which is the end-effector of the system.
[0049] Preferably, the sensor network is deployed at key locations in the purification equipment, including but not limited to high-precision thermocouples, capacitive thin-film vacuum gauges, mass flow meters, online turbidity / color analyzers, and pH meters;
[0050] The data interface and acquisition are conducted through industrial communication protocols to collect sensor data and equipment status data in the PLC in real time, and to import offline test evaluation index data from the laboratory information management system on a regular basis.
[0051] Preferably, the online optimization and control module is typically deployed on industrial control computers or edge computing devices with high real-time requirements.
[0052] Preferably, the control execution module typically receives instructions from the online optimization and control module through a dedicated function block in the PLC or DCS. Based on the instructions, the PLC or DCS drives the regulating valve to control the feeding, the entry of protective gas, the heater to control the temperature, and the vacuum pump frequency converter to control the vacuum level, so that the actual process parameters are close to the issued set values, thereby achieving precise control of the constant temperature purification process.
[0053] Example 4:
[0054] This embodiment provides a non-transitory computer-readable storage medium, such as a solid-state drive, USB flash drive, read-only memory, random access memory, magnetic disk, or optical disk. The storage medium stores a computer program.
[0055] When this computer program is loaded and executed by one or more processors, it performs the following functions:
[0056] This includes reading historical production data from storage media or received from external interfaces, performing data preprocessing, building or loading a BP neural network model architecture, using training datasets to complete model training and validation, periodically acquiring real-time data during the purification process, calling the trained model and optimization algorithm, calculating optimal process parameters, generating control commands, and sending them to the lower-level control system through the communication interface.
[0057] Optionally, perform model performance monitoring, new data archiving, and model-triggered update logic.
[0058] Optionally, the function of constructing and real-time estimating soft measurement models can be implemented.
[0059] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only for the purpose of helping to understand the method and core ideas of the present invention. The above descriptions are only preferred embodiments of the present invention. It should be noted that due to the limitations of textual expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of the present invention, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of the present invention.
Claims
1. A method for isothermal purification control based on diphenyl isodecyl phosphite, characterized in that, Includes the following steps: S1: Data Acquisition and Preprocessing: During the isothermal purification process of diphenyl isodecyl phosphite, multiple batches of historical production data are collected. The historical production data includes multiple process parameters that affect the purification process as input features, and multiple evaluation indicators that characterize the purification effect as output labels. The historical production data is preprocessed to construct a neural network training dataset; S2: Neural Network Modeling and Training: Construct a BP neural network model, using the input features as the input layer and the output labels as the output layer; The BP neural network model is trained using the training dataset to obtain a trained product quality prediction model; S3: Online optimization and control: Obtain the initial conditions for the current purification batch to be optimized; The initial conditions and preset quality targets are input into the trained product quality prediction model, and an optimization algorithm is used to solve for a set of optimal process parameter settings. The optimal process parameter settings are sent to the production control system to control the isothermal purification process of the current batch.
2. The isothermal purification control method based on diphenyl isodecanyl phosphite according to claim 1, characterized in that: In step S1, the process parameters include at least four of the following: constant temperature, system vacuum, feed rate, reflux ratio, and protective gas flow rate. The evaluation indicators include at least two of the following: purity of the main product component, acid value, and color.
3. The isothermal purification control method based on diphenyl isodecanyl phosphite according to claim 2, characterized in that: The evaluation indicators shall include at least one of the following: product yield, unit energy consumption, and product moisture content.
4. The isothermal purification control method based on diphenyl isodecanyl phosphite according to claim 1, characterized in that: In step S2, when training the BP neural network model, a weighted loss function is used, in which the purity of the main components of the product and the acid value of the product are given higher weights.
5. The isothermal purification control method based on diphenyl isodecanyl phosphite according to claim 1, characterized in that: In step S3, the optimization algorithm is one of the following: genetic algorithm, particle swarm optimization algorithm, or gradient descent method. The objective function of the optimization algorithm is to minimize the comprehensive deviation between the predicted evaluation index and the preset quality target, while satisfying the operational constraints and safety boundaries of the process parameters.
6. The isothermal purification control method based on diphenyl isodecanyl phosphite according to claim 1, characterized in that: In step S3, the online optimization and control process is executed periodically using a model predictive control framework: during the purification process, real-time process data is re-acquired at a set period as the current initial condition, the optimization steps are repeated, and new optimal process parameter settings are dynamically updated and issued until the purification process ends.
7. The isothermal purification and control method based on diphenyl isodecanyl phosphite according to any one of claims 1-6, characterized in that: It also includes step S4: Model update: The final actual process parameters and actual evaluation index data of the completed batch are added to the training dataset as new sample data; When new sample data accumulates to a preset quantity or the model prediction error exceeds a threshold, the product quality prediction model is retrained and updated.
8. A isothermal purification system for diphenyl isodecyl phosphite used to implement the method according to any one of claims 1-7, characterized in that: Includes a data acquisition and processing module, used to collect and preprocess historical and real-time production data; The neural network modeling and training module is used to build, train, and store the product quality prediction model. The online optimization and control module is used to execute optimization algorithms, calculate optimal process parameter setpoints, and generate control commands. The control execution module is used to receive the control commands and drive the production equipment to perform corresponding parameter adjustments.
9. The isothermal purification system for diphenyl isodecanyl phosphite according to claim 8, characterized in that: The system also includes a soft measurement module, which constructs a soft measurement model for real-time estimation of key evaluation indicators based on online sensor data and intermediate layer features of the product quality prediction model.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-7.