Reaction kettle operation control method and system for resin production

Abstract

The invention relates to the field of control, in particular to a reaction kettle operation control method and system for resin production, real-time operation parameters of a reaction kettle are obtained, a fuzzy neural network model is iteratively trained by adopting a hierarchical collaborative hybrid optimization strategy, a preceding member membership function of the fuzzy neural network model is composed of a Gaussian mixture model, and the preceding member membership function of the fuzzy neural network model is obtained. According to the optimization strategy, an improved quantum particle swarm optimization algorithm is used for carrying out global search to determine Gaussian mixture model parameters, a recursive least square algorithm is used for carrying out local search to determine consequent coefficients after each time of iteration, and in the training process, the parameters of the Gaussian mixture model are subjected to global search to determine the parameters of the Gaussian mixture model. And calculating an importance index according to the average activation degree of the fuzzy rule and the contribution of the fuzzy rule to the prediction error, removing the rule of which the importance is continuously lower than a preset threshold value, and after training is completed, generating and executing a control instruction for controlling the heating system power and the material feeding rate of the reaction kettle at the next moment according to the real-time parameters.

Description

Technical Field

[0001] This application belongs to the field of operation control, and in particular relates to a method and system for operation control of a reaction vessel used in resin production. Background Technology

[0002] Resin polymerization is a typical complex chemical process characterized by multiple variables, strong nonlinearity, and large time delay. Key parameters such as temperature, pressure, stirring rate, and material viscosity within the reactor are tightly coupled, and even small fluctuations in any parameter can significantly impact the reaction process and product performance. Traditional control methods, such as proportional-integral-derivative (PID) control, while simple in structure and widely used, are difficult to tune for complex, nonlinear, time-varying systems like resin reactors. They struggle to adapt to changing operating conditions, often resulting in low control accuracy and large overshoot. Model predictive control relies on precise mathematical models, but the mechanism of resin polymerization is complex, making it extremely difficult to establish a first-principles model that accurately describes its reaction kinetics and heat and mass transfer processes. Furthermore, overly complex models are not conducive to online real-time solutions.

[0003] Fuzzy Neural Networks (FNNs) combine the ability of fuzzy logic to handle uncertain information with the self-learning and nonlinear approximation capabilities of neural networks. They can construct the input-output relationships of a process by learning from historical operational data without requiring a precise mechanistic model. However, most FNN models prioritize data fitting accuracy during training, making them black-box models. Their predictions may contradict fundamental physical and chemical laws, resulting in poor generalization ability and insufficient reliability under unseen conditions. Traditional FNN training algorithms, such as gradient descent, are prone to getting trapped in local optima, and network structures, such as the number of fuzzy rules, often rely on manual experience, making it difficult to balance model accuracy and computational complexity. Redundant rules not only increase the computational burden but may also negatively impact model performance. Summary of the Invention

[0004] To improve the accuracy of the operation control of a reactor used in resin production, this application proposes a method for controlling the operation of a reactor used in resin production, comprising the following steps: The real-time operating parameters of the reactor used for resin production are obtained, including the reactor temperature, reactor pressure, stirring rate, and material viscosity. A hierarchical and collaborative hybrid optimization strategy is employed to iteratively optimize and train a fuzzy neural network model. The antecedent membership function of the fuzzy neural network model is composed of a Gaussian mixture model, and the consequent is a nonlinear polynomial function of the input variables. A weighted prediction error is calculated based on an asymmetric penalty cost function, which imposes a large penalty on errors where the predicted output exceeds the target output. The physical consistency deviation is obtained by comparing the output of the fuzzy neural network model with the predicted value of a simplified mechanism model. The iterative optimization training aims to minimize the comprehensive objective function, which is the weighted sum of the weighted prediction error and the physical consistency deviation of the fuzzy neural network model. The hybrid optimization strategy involves using an improved quantum particle swarm optimization algorithm to perform a global search based on the comprehensive objective function to determine the parameters in the Gaussian mixture model. After each global iteration, a recursive least squares algorithm is used to perform a local search to determine the coefficients in the consequent nonlinear polynomial. During the iterative optimization training process, the importance index of each rule is calculated based on the average activation degree of each fuzzy rule and its contribution to the weighted prediction error, and rules whose importance index is consistently lower than the pruning threshold are removed. The acquired real-time operating parameters are input into the trained fuzzy neural network model to obtain control commands for controlling the heating system power and material feed rate of the reactor at the next moment.

[0005] Preferably, the formula for calculating the comprehensive objective function J is: ,in The weighted prediction error is... The physical consistency deviation is α, which is a weighting coefficient with a value between 0 and 1.

[0006] Preferably, the weighted prediction error is calculated based on an asymmetric penalty cost function, wherein the asymmetric penalty cost function imposes a large penalty on errors where the predicted output is higher than the target output, including: Prediction error at each sampling time ; When e≥0, the error cost at the sampling time is ; When e < 0, the error cost at the sampling time is , where λ is a penalty factor greater than 1; The weighted prediction error is the average of the error costs at all sampling times.

[0007] Preferably, the step of obtaining the physical consistency deviation by comparing the output of the fuzzy neural network model with the predicted value of the simplified mechanism model includes: A reaction rate model based on the Arrhenius equation and a heat transfer model of the reactor based on the law of energy conservation are established as the simplified mechanism model. During the training process, the real-time operating parameters and the corresponding control quantities are used as inputs to both the fuzzy neural network model and the simplified mechanism model to obtain their respective predicted outputs of the reactor state at the next moment. The root mean square error between the predicted output of the fuzzy neural network model and the predicted output of the simplified mechanism model within a preset time window is calculated as the physical consistency deviation.

[0008] Preferably, the step of using an improved quantum particle swarm optimization algorithm to perform a global search based on the comprehensive objective function to determine the parameters in the Gaussian mixture model, and after each global iteration, using a recursive least squares algorithm to perform a local search to determine the coefficients in the consequent nonlinear polynomial, includes: In a single iteration, the quantum particle swarm optimization algorithm is executed, using the center and width parameters in the Gaussian mixture model as particle positions, and the comprehensive objective function as the fitness function to update the particle positions, thereby obtaining a set of optimized antecedent parameters; With the antecedent parameters fixed, the recursive least squares algorithm is used to solve and update all coefficients in the consequent nonlinear polynomial in one operation.

[0009] Preferably, the step of calculating the rule importance index based on the average activation degree of each fuzzy rule and its contribution to the weighted prediction error includes: The importance index for rule j is: ,in The average activation of the rule across all training samples. The sum of weighted prediction errors belonging to the rules is ε, which is a preset small positive number used to prevent the denominator from being zero.

[0010] Preferably, the rule for removing an importance metric that is consistently below a pruning threshold includes: Obtain the rule importance pruning threshold and persistence judgment threshold; After each iteration of optimization, if the number of iterations in which the importance index of a rule is lower than the pruning threshold reaches the persistence judgment threshold, then the rule is deleted from the fuzzy neural network model.

[0011] On the other hand, this application also proposes a reactor operation control system for resin production, including the following modules: The parameter acquisition module acquires the real-time operating parameters of the reactor used for resin production, including the reactor temperature, reactor pressure, stirring rate, and material viscosity. The iterative optimization training module employs a hierarchical collaborative hybrid optimization strategy to iteratively optimize and train a fuzzy neural network model. The antecedent membership function of the fuzzy neural network model is composed of a Gaussian mixture model, and the consequent is a nonlinear polynomial function of the input variables. A weighted prediction error is calculated based on an asymmetric penalty cost function, which imposes a large penalty on errors where the predicted output exceeds the target output. The physical consistency deviation is obtained by comparing the output of the fuzzy neural network model with the predicted value of a simplified mechanism model. The iterative optimization training aims to minimize the comprehensive objective function, which is the weighted sum of the weighted prediction error and the physical consistency deviation of the fuzzy neural network model. The hybrid optimization strategy involves using an improved quantum particle swarm optimization algorithm to perform a global search based on the comprehensive objective function to determine the parameters in the Gaussian mixture model. After each global iteration, a recursive least squares algorithm is used to perform a local search to determine the coefficients in the consequent nonlinear polynomial. The rule removal module calculates the rule importance index based on the average activation degree of each fuzzy rule and its contribution to the weighted prediction error, and removes rules whose importance index is consistently below the pruning threshold. The instruction acquisition and execution module inputs the acquired real-time operating parameters into the trained fuzzy neural network model to obtain and execute control instructions for controlling the heating system power and material feed rate of the reactor at the next moment.

[0012] Preferably, the formula for calculating the comprehensive objective function J is: ,in The weighted prediction error is... The physical consistency deviation is α, which is a weighting coefficient with a value between 0 and 1.

[0013] Preferably, the weighted prediction error is calculated based on an asymmetric penalty cost function, wherein the asymmetric penalty cost function imposes a large penalty on errors where the predicted output is higher than the target output, including: Prediction error at each sampling time ; When e≥0, the error cost at the sampling time is ; When e < 0, the error cost at the sampling time is , where λ is a penalty factor greater than 1; The weighted prediction error is the average of the error costs at all sampling times.

[0014] Preferably, the step of obtaining the physical consistency deviation by comparing the output of the fuzzy neural network model with the predicted value of the simplified mechanism model includes: A reaction rate model based on the Arrhenius equation and a heat transfer model of the reactor based on the law of energy conservation are established as the simplified mechanism model. During the training process, the real-time operating parameters and the corresponding control quantities are used as inputs to both the fuzzy neural network model and the simplified mechanism model to obtain their respective predicted outputs of the reactor state at the next moment. The root mean square error between the predicted output of the fuzzy neural network model and the predicted output of the simplified mechanism model within a preset time window is calculated as the physical consistency deviation.

[0015] Preferably, the step of using an improved quantum particle swarm optimization algorithm to perform a global search based on the comprehensive objective function to determine the parameters in the Gaussian mixture model, and after each global iteration, using a recursive least squares algorithm to perform a local search to determine the coefficients in the consequent nonlinear polynomial, includes: In a single iteration, the quantum particle swarm optimization algorithm is executed, using the center and width parameters in the Gaussian mixture model as particle positions, and the comprehensive objective function as the fitness function to update the particle positions, thereby obtaining a set of optimized antecedent parameters; With the antecedent parameters fixed, the recursive least squares algorithm is used to solve and update all coefficients in the consequent nonlinear polynomial in one operation.

[0016] Preferably, the step of calculating the rule importance index based on the average activation degree of each fuzzy rule and its contribution to the weighted prediction error includes: The importance index for rule j is: ,in The average activation of the rule across all training samples. The sum of weighted prediction errors belonging to the rules is ε, which is a preset small positive number used to prevent the denominator from being zero.

[0017] Preferably, the rule for removing an importance metric that is consistently below a pruning threshold includes: Obtain the rule importance pruning threshold and persistence judgment threshold; After each iteration of optimization, if the number of iterations in which the importance index of a rule is lower than the pruning threshold reaches the persistence judgment threshold, then the rule is deleted from the fuzzy neural network model.

[0018] This invention employs a hierarchical collaborative hybrid optimization strategy, combining the global search capability of the quantum particle swarm optimization algorithm with the local fast convergence characteristic of the recursive least squares algorithm, to find the optimal parameters of the fuzzy neural network, thereby improving the model's prediction accuracy. By introducing the physical consistency bias of the mechanism-based model into the training objective function, the model is constrained by physical laws during the learning process, ensuring the rationality and generalization ability of the model's output. This avoids the control decisions that violate physical common sense that pure black-box models may produce, enhancing the model's reliability under complex operating conditions. Furthermore, by introducing an asymmetric penalty cost function, overshoot of key process parameters can be specifically suppressed, making the control strategy more aligned with safety risk considerations in actual production and improving system safety. Attached Figure Description

[0019] Figure 1 A flowchart of a specific embodiment; Figure 2 This is a schematic diagram of the reactor control system; Figure 3 This is a diagram illustrating the pruning rules. Detailed Implementation

[0020] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0021] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0022] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0023] In one embodiment, this application proposes a method for controlling the operation of a reactor used in resin production, such as... Figure 1 As shown, it includes the following steps: S1, Obtain the real-time operating parameters of the reactor used for resin production, including the reactor temperature, reactor pressure, stirring rate and material viscosity; Real-time data on reactor temperature, pressure, stirring rate, and material viscosity are collected via thermocouple temperature sensors, pressure transmitters, motor speed sensors, and online viscometers installed on the reactor. These data are then transmitted to a host computer via an industrial Ethernet network at a fixed sampling period, e.g., once per second, through a distributed control system (DCS) or programmable logic controller (PLC). Figure 2 As shown.

[0024] S2, a hierarchical collaborative hybrid optimization strategy is used to iteratively optimize and train the fuzzy neural network model. The antecedent membership function of the fuzzy neural network model is composed of a Gaussian mixture model, and the consequent is a nonlinear polynomial function of the input variables. The weighted prediction error is calculated based on an asymmetric penalty cost function, which imposes a large penalty on the error where the predicted output is higher than the target output. The physical consistency deviation is obtained by comparing the output of the fuzzy neural network model with the predicted value of the simplified mechanism model. The iterative optimization training aims to minimize the comprehensive objective function, which is the weighted sum of the weighted prediction error and the physical consistency deviation of the fuzzy neural network model. The hybrid optimization strategy is as follows: using an improved quantum particle swarm optimization algorithm, a global search is performed based on the comprehensive objective function to determine the parameters in the Gaussian mixture model. After each global iteration, a recursive least squares algorithm is used to perform a local search to determine the coefficients in the consequent nonlinear polynomial. A fuzzy neural network model is constructed, where the antecedent of each fuzzy rule adopts a Gaussian mixture model, with parameters such as center and width to be determined; the consequent adopts a second-order nonlinear polynomial function of the input variables, for example... Where T is temperature, P is pressure, V is viscosity, and coefficient to To be determined. Define the comprehensive objective function. ,in Weight the prediction error of the model. For physical consistency deviation, w1 and w2 are weighting coefficients. Calculation When calculating, an asymmetric penalty cost function is preferred. For example, when the model predicts a temperature higher than the target temperature, the error penalty coefficient is set to 1.5, and when it predicts a temperature lower than the target temperature, it is set to 1.0. At this point, the predicted temperature from the fuzzy neural network is compared with the predicted temperature from a simplified heat transfer reaction model based on the Arrhenius equation and the law of conservation of energy. The square of the difference between the two is then calculated. During training, the outer loop employs an improved quantum particle swarm optimization algorithm, using all center and width parameters of the Gaussian mixture model as particle positions for global optimization. In each iteration of the quantum particle swarm algorithm, for each particle's represented antecedent parameter, the inner loop calls a recursive least squares algorithm, minimizing... Following the principle of [missing information], all optimal coefficients of the consequent nonlinear polynomial are quickly analyzed. The entire process is guided by minimizing the comprehensive objective function J, iterating repeatedly until convergence.

[0025] S3, during the iterative optimization training process, the rule importance index is calculated based on the average activation degree of each fuzzy rule and its contribution to the weighted prediction error, and rules whose importance index is consistently lower than the pruning threshold are removed. After a fixed number of iterations during model training, such as 20 iterations, an importance metric is calculated for each fuzzy rule in the network. This importance metric is a weighted average of two parts: the average activation strength of the rule, representing its average triggering degree across the entire training dataset and its output energy; and the weighted prediction error of the model when the rule is activated. The reciprocal of the average contribution. The two are weighted and summed to obtain the final importance index. A pruning threshold is set, for example, 0.05, and an observation counter is set. If the importance index of a rule is below the pruning threshold for five consecutive evaluations, the rule is considered redundant or harmful and is permanently removed from the model structure, including its corresponding antecedent Gaussian mixture model parameters and consequent polynomial coefficients.

[0026] S4. The acquired real-time operating parameters are input into the trained fuzzy neural network model to obtain control commands for controlling the heating system power and material feed rate of the reactor at the next moment.

[0027] A trained and simplified fuzzy neural network model is deployed on an industrial control computer. At each control moment, for example every 10 seconds, the real-time collected values ​​of four parameters—vessel temperature, vessel pressure, stirring rate, and material viscosity—are fed into the fuzzy neural network model as an input vector. After forward computation, the model outputs two specific values: one representing the target power of the heating system for the next control cycle (in kilowatts), and the other representing the target material feed rate for the next control cycle (in kilograms per hour). These two control command values ​​are then sent to the field actuators via the DCS system, such as the thyristor power regulator controlling the heating jacket and the frequency converter controlling the feed pump, achieving closed-loop control of the reactor.

[0028] In an optional embodiment, the formula for calculating the comprehensive objective function J is: ,in The weighted prediction error is... The physical consistency deviation is α, which is a weighting coefficient with a value between 0 and 1.

[0029] For example, in the task of predicting the temperature of a polymerization reactor, This represents the difference between the temperature predicted by the model and the actual measured temperature. This represents the degree to which the model's predictions violate fundamental physical principles such as the law of conservation of energy. If α is set to a high value, such as 0.9, the optimization process will primarily focus on reducing prediction errors. The model more closely resembles historical measurement data. This setup is suitable for scenarios where the process mechanism is relatively unclear, but a large amount of high-quality historical data is available.

[0030] Conversely, if α is set to a lower value, such as 0.2, the focus of the optimization process will shift to reducing physical consistency deviations. This ensures that the model's output is physically reasonable and interpretable, even if it slightly sacrifices the fitting accuracy to some historical data points. Taking a practical optimization step as an example, assume that the model's weighted prediction error is calculated under the current model parameters. The physical consistency deviation is 0.05. The value is 0.12. If α is set to 0.6, the comprehensive objective function value J is 0.078. The next iteration of the optimization algorithm will adjust the model parameters to seek ways to further reduce this comprehensive objective value, thereby obtaining a prediction model that is both accurate and reliable.

[0031] In an optional embodiment, the weighted prediction error is calculated based on an asymmetric penalty cost function, wherein the asymmetric penalty cost function imposes a large penalty on errors where the predicted output is higher than the target output, including: Prediction error at each sampling time ; When e≥0, the error cost at the sampling time is ; When e < 0, the error cost at the sampling time is , where λ is a penalty factor greater than 1; The weighted prediction error is the average of the error costs at all sampling times.

[0032] In industrial settings, the risks or costs associated with overestimating and underestimating forecasts are disproportionate. Take, for example, the forecasting of key component content in production. It is the target content. This refers to the content predicted by the model. If the model underestimates the content (e > 0), it may lead to a longer production process and increased energy consumption, but product quality will still be guaranteed. However, if the model overestimates the content (e < 0), it may lead to premature termination of production, resulting in batches of substandard products, causing huge economic losses and safety risks.

[0033] To address this situation, a penalty factor λ greater than 1 can be set, for example, λ equals 4. Suppose at a certain moment, the target content is 95%, and the model predicts 93%, then the error e is 2, and the error cost is 4. At another moment, the target content is 95%, and the model predicts 97%, then the error e is -2, and the error cost is 16. It is clear that for the same 2% prediction bias, the cost of overestimation is four times that of underestimation. By calculating the average error cost across all moments on the entire training dataset, training will tend to produce more conservative predictions, avoiding the serious consequences of overestimation.

[0034] In an optional embodiment, obtaining the physical consistency deviation by comparing the output of the fuzzy neural network model with the predictions of the simplified mechanistic model includes: A reaction rate model based on the Arrhenius equation and a heat transfer model of the reactor based on the law of energy conservation are established as the simplified mechanism model. During the training process, the real-time operating parameters and the corresponding control quantities are used as inputs to both the fuzzy neural network model and the simplified mechanism model to obtain their respective predicted outputs of the reactor state at the next moment. The root mean square error between the predicted output of the fuzzy neural network model and the predicted output of the simplified mechanism model within a preset time window is calculated as the physical consistency deviation.

[0035] When simulating a chemical reaction process, a simplified model is constructed based on generally accepted chemical and physical principles. This simplified model may include the Arrhenius equation describing the relationship between reaction rate and temperature, and the energy conservation equation describing heat exchange in the reactor. While this mechanistic model may be imprecise due to inaccurate parameters or failure to consider all factors, it accurately reflects the fundamental dynamic trends of the system.

[0036] In each step of model training, a set of identical inputs, such as the current temperature of 350 Kelvin and the feed rate of 0.5 kg / s, are provided to both the fuzzy neural network model and the simplified mechanistic model. The fuzzy neural network model might predict the temperature in the next minute to be 351.2 Kelvin, while the simplified mechanistic model, based on physical laws, calculates that the temperature in the next minute should be 350.8 Kelvin. The difference between these two predictions, 0.4 Kelvin, constitutes the physical consistency discrepancy at the current moment. The physical consistency deviation is obtained by calculating the root mean square error of these discrepancies over a time window, for example, 10 consecutive sampling points. This deviation value is added as a penalty term to the overall objective function, forcing the prediction results of the fuzzy neural network to avoid serious conflicts with basic physical laws.

[0037] In an optional embodiment, the step of using an improved quantum particle swarm optimization algorithm to perform a global search based on the comprehensive objective function to determine the parameters in the Gaussian mixture model, and after each global iteration, using a recursive least squares algorithm to perform a local search to determine the coefficients in the consequent nonlinear polynomial, includes: In a single iteration, the quantum particle swarm optimization algorithm is executed, using the center and width parameters in the Gaussian mixture model as particle positions, and the comprehensive objective function as the fitness function to update the particle positions, thereby obtaining a set of optimized antecedent parameters; With the antecedent parameters fixed, the recursive least squares algorithm is used to solve and update all coefficients in the consequent nonlinear polynomial in one operation.

[0038] The antecedent parameters of the fuzzy rules, namely the center and width of the Gaussian membership function, determine the partitioning of the input space. These parameters have a highly nonlinear relationship with the final output, making their optimization a complex, multi-peaked global search problem. Therefore, the quantum particle swarm optimization algorithm, with its powerful global exploration capabilities, is employed to handle these parameters. For example, for a model with 10 rules, each corresponding to 2 antecedent parameters, the quantum particle swarm algorithm treats a vector containing these 20 parameters as a particle and searches in 20-dimensional space for the particle position that minimizes the comprehensive objective function J.

[0039] Once the quantum particle swarm optimization algorithm completes one iteration and finds a set of currently optimal antecedent parameters, these parameters are temporarily fixed. At this point, the consequent parameters of the model, i.e., the polynomial coefficients of each rule output, form a linear relationship with the total output of the model. For this type of linear parameter solving problem, the recursive least squares algorithm is a computationally extremely fast and efficient local search method. Using the recursive least squares algorithm, based on the fixed antecedent parameters, all consequent polynomial coefficients are solved precisely in one go, quickly fine-tuning the model's output. This iterative and collaborative approach of first globally searching the nonlinear part and then quickly analytically solving the linear part greatly improves the convergence speed and solution quality compared to using a single algorithm to optimize all parameters.

[0040] In an optional embodiment, the step of calculating the rule importance index based on the average activation degree of each fuzzy rule and its contribution to the weighted prediction error includes: The importance index for rule j is: ,in The average activation of the rule across all training samples. The sum of weighted prediction errors belonging to the rules is ε, which is a preset small positive number used to prevent the denominator from being zero.

[0041] An ideal rule should be frequently activated to cover a sufficient number of operating conditions, while also ensuring minimal prediction error upon activation. The numerator in the formula... This indicates the generality or coverage of the rule. For example, after processing 1000 training samples, if the average activation of rule 5... The calculated average activation value is 0.7, which corresponds to rule number 8. With a value of only 0.05, this indicates that rule 5 applies to a wider range of input scenarios than rule 8.

[0042] denominator in the formula This represents the accuracy of the rule, that is, the total weighted prediction error caused by the rule. A rule may have high activation, but if it consistently introduces large errors into the predictions, then its contribution is negative. For example, suppose rule 5 has high activation, but its associated total error... It also reached as high as 25.3; although rule number 8 has a low activation level, it is very accurate, with a total correlation error of 25.3. It is only 0.1. Let the small positive number ε be 0.001, then the importance index of rule 5 is... Approximately 0.028, the importance index of Rule 8. The value is approximately 0.5. This metric clearly demonstrates that although rule number 8 is less frequently used, its value far exceeds that of rule number 5, which is prone to errors.

[0043] In an optional embodiment, the rule for removing an importance metric that is consistently below a pruning threshold includes: Obtain the rule importance pruning threshold and persistence judgment threshold; After each iteration of optimization, if the number of iterations in which the importance index of a rule is lower than the pruning threshold reaches the persistence judgment threshold, then the rule is deleted from the fuzzy neural network model.

[0044] The rule importance pruning threshold is set to 0.1, and the persistence threshold is set to 10 iterations. Any rule with an importance index below 0.1 will be marked as a potentially useless rule, but it will only be removed if it is marked this way for 10 consecutive iterations. Figure 3 As shown.

[0045] Suppose that after the 50th iteration of model training, the importance index of the third rule is calculated to be 0.08, which is lower than 0.1, so a counter is started for the third rule, set to 1. After the 51st iteration, its importance index becomes 0.12, which is higher than the threshold, and the counter is reset to zero. In the 10 consecutive iterations from the 60th to the 69th iteration, the importance index of the third rule remains below 0.1, and its counter increments from 1 to 10. When the counter reaches the persistence threshold of 10, it is confirmed that the third rule performs poorly under various model parameter configurations and is a stable redundant or harmful rule. At this point, a deletion operation will be automatically performed to completely remove the third rule from the fuzzy neural network model structure.

[0046] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, compact disc read-only memory (CD-ROM), optical storage, etc.) containing computer-usable program code.

[0047] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0048] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

Claims

1. A method for controlling the operation of a reaction vessel used in resin production, characterized in that, Includes the following steps: The real-time operating parameters of the reactor used for resin production are obtained, including the reactor temperature, reactor pressure, stirring rate, and material viscosity. A hierarchical and collaborative hybrid optimization strategy is employed to iteratively optimize and train a fuzzy neural network model. The antecedent membership function of the fuzzy neural network model is composed of a Gaussian mixture model, and the consequent is a nonlinear polynomial function of the input variables. A weighted prediction error is calculated based on an asymmetric penalty cost function, which imposes a large penalty on errors where the predicted output exceeds the target output. The physical consistency deviation is obtained by comparing the output of the fuzzy neural network model with the predicted value of a simplified mechanism model. The iterative optimization training aims to minimize the comprehensive objective function, which is the weighted sum of the weighted prediction error and the physical consistency deviation of the fuzzy neural network model. The hybrid optimization strategy involves using an improved quantum particle swarm optimization algorithm to perform a global search based on the comprehensive objective function to determine the parameters in the Gaussian mixture model. After each global iteration, a recursive least squares algorithm is used to perform a local search to determine the coefficients in the consequent nonlinear polynomial. During the iterative optimization training process, the importance index of each rule is calculated based on the average activation degree of each fuzzy rule and its contribution to the weighted prediction error, and rules whose importance index is consistently lower than the pruning threshold are removed. The acquired real-time operating parameters are input into the trained fuzzy neural network model to obtain control commands for controlling the heating system power and material feed rate of the reactor at the next moment, and then executed.

2. The method according to claim 1, characterized in that, The formula for calculating the comprehensive objective function J is as follows: ,in The weighted prediction error is... The physical consistency deviation is α, which is a weighting coefficient with a value between 0 and 1.

3. The method according to claim 1, characterized in that, The weighted prediction error is calculated based on the asymmetric penalty cost function, which imposes a large penalty on errors where the predicted output is higher than the target output, including: Prediction error at each sampling time ; When e≥0, the error cost at the sampling time is ; When e < 0, the error cost at the sampling time is , where λ is a penalty factor greater than 1; The weighted prediction error is the average of the error costs at all sampling times.

4. The method according to claim 1, characterized in that, The method of obtaining physical consistency deviation by comparing the output of the fuzzy neural network model with the predicted value of the simplified mechanism model includes: A reaction rate model based on the Arrhenius equation and a heat transfer model of the reactor based on the law of energy conservation are established as the simplified mechanism model. During the training process, the real-time operating parameters and the corresponding control quantities are used as inputs to both the fuzzy neural network model and the simplified mechanism model to obtain their respective predicted outputs of the reactor state at the next moment. The root mean square error between the predicted output of the fuzzy neural network model and the predicted output of the simplified mechanism model within a preset time window is calculated as the physical consistency deviation.

5. The method according to claim 1, characterized in that, The improved quantum particle swarm optimization algorithm performs a global search based on the comprehensive objective function to determine the parameters in the Gaussian mixture model. After each global iteration, a recursive least squares algorithm is used to perform a local search to determine the coefficients in the consequent nonlinear polynomial, including: In a single iteration, the quantum particle swarm optimization algorithm is executed, using the center and width parameters in the Gaussian mixture model as particle positions, and the comprehensive objective function as the fitness function to update the particle positions, thereby obtaining a set of optimized antecedent parameters; With the antecedent parameters fixed, the recursive least squares algorithm is used to solve and update all coefficients in the consequent nonlinear polynomial in one operation.

6. The method according to claim 1, characterized in that, The calculation of rule importance indicators based on the average activation degree of each fuzzy rule and its contribution to the weighted prediction error includes: The importance index for rule j is: ,in The average activation of the rule across all training samples. The sum of weighted prediction errors belonging to the rules is ε, which is a preset small positive number used to prevent the denominator from being zero.

7. The method according to claim 1, characterized in that, The rules for removing importance metrics that are consistently below the pruning threshold include: Obtain the rule importance pruning threshold and persistence judgment threshold; After each iteration of optimization, if the number of iterations in which the importance index of a rule is lower than the pruning threshold reaches the persistence judgment threshold, then the rule is deleted from the fuzzy neural network model.

8. A control system for a reaction vessel used in resin production, characterized in that, Includes the following modules: The parameter acquisition module acquires the real-time operating parameters of the reactor used for resin production, including the reactor temperature, reactor pressure, stirring rate, and material viscosity. The iterative optimization training module employs a hierarchical collaborative hybrid optimization strategy to iteratively optimize and train a fuzzy neural network model. The antecedent membership function of the fuzzy neural network model is composed of a Gaussian mixture model, and the consequent is a nonlinear polynomial function of the input variables. A weighted prediction error is calculated based on an asymmetric penalty cost function, which imposes a large penalty on errors where the predicted output exceeds the target output. The physical consistency deviation is obtained by comparing the output of the fuzzy neural network model with the predicted value of a simplified mechanism model. The iterative optimization training aims to minimize the comprehensive objective function, which is the weighted sum of the weighted prediction error and the physical consistency deviation of the fuzzy neural network model. The hybrid optimization strategy involves using an improved quantum particle swarm optimization algorithm to perform a global search based on the comprehensive objective function to determine the parameters in the Gaussian mixture model. After each global iteration, a recursive least squares algorithm is used to perform a local search to determine the coefficients in the consequent nonlinear polynomial. The rule removal module calculates the rule importance index based on the average activation degree of each fuzzy rule and its contribution to the weighted prediction error, and removes rules whose importance index is consistently below the pruning threshold. The instruction acquisition and execution module inputs the acquired real-time operating parameters into the trained fuzzy neural network model to obtain and execute control instructions for controlling the heating system power and material feed rate of the reactor at the next moment.

9. The system according to claim 8, characterized in that, The formula for calculating the comprehensive objective function J is as follows: ,in The weighted prediction error is... The physical consistency deviation is α, which is a weighting coefficient with a value between 0 and 1.

10. The system according to claim 8, characterized in that, The weighted prediction error is calculated based on the asymmetric penalty cost function, which imposes a large penalty on errors where the predicted output is higher than the target output, including: Prediction error at each sampling time ; When e≥0, the error cost at the sampling time is ; When e < 0, the error cost at the sampling time is , where λ is a penalty factor greater than 1; The weighted prediction error is the average of the error costs at all sampling times.

Images