A Neural Network-Based Method and System for Optimizing Plastic Manufacturing Process Parameters

Abstract

This invention provides a method and system for optimizing plastic manufacturing process parameters based on neural networks, belonging to the field of plastic manufacturing process parameter optimization technology. The method includes the following steps: selecting a reference plastic and obtaining its reference performance parameters; obtaining reference process parameters based on the reference performance parameters using a first optimization model; obtaining target performance parameters; when the target performance parameters match the reference performance parameters, the reference process parameters are directly used as the optimized process parameters; otherwise, the optimized process parameters are obtained using a second optimization model based on the reference performance parameters, the reference process parameters, and the target performance parameters. This invention has the advantages of achieving efficient, controllable, accurate, and wide-coverage process parameter optimization.

Description

Technical Field

[0001] This invention relates to the field of plastic manufacturing process parameter optimization technology, and more specifically, to a method and system for optimizing plastic manufacturing process parameters based on neural networks. Background Technology

[0002] Plastic products are widely used in various industrial and consumer fields such as automobiles, electronics, electrical appliances, packaging, and medical devices. With the continuous improvement of manufacturing precision and performance requirements, how to intelligently set manufacturing process parameters according to different plastic performance targets has become one of the key issues in improving plastic quality and production efficiency.

[0003] In existing technologies, the plastic manufacturing process involves the control of various complex process parameters, such as temperature, pressure, cooling time, and injection speed. These parameters have a significant impact on the properties of the final plastic product (such as strength, toughness, hardness, and heat resistance). Traditional process parameter setting relies on rules of thumb or repeated trials, making it difficult to adapt to customized parameters for different target performance requirements. This approach is inefficient, costly, and lacks generalization ability. With the development of artificial intelligence, introducing neural network-based process parameter optimization methods into the plastic manufacturing field has become a feasible direction. However, in practical applications, predicting process parameters for arbitrary target performance conditions remains challenging. This is mainly due to the large variety of controllable parameters in the manufacturing process and the diverse changes in performance targets. Directly establishing an end-to-end mapping model from "target performance to optimal process" is easily limited by sample coverage and has poor robustness.

[0004] Therefore, there is an urgent need for a process parameter reasoning method that combines baseline process knowledge with neural network optimization capabilities, which can adapt to the target performance requirements of various plastic types and achieve efficient, controllable, accurate and wide-coverage process parameter optimization. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for optimizing plastic manufacturing process parameters based on neural networks, which achieves efficient, controllable, accurate and wide-coverage optimization of process parameters.

[0006] This invention is achieved through the following technical solution:

[0007] A neural network-based method for optimizing plastic manufacturing process parameters includes the following steps:

[0008] Select a reference plastic and obtain the plastic reference performance parameters of the reference plastic. The plastic reference performance parameters are used as the reference values ​​for the performance parameters of the plastic. The performance parameters of the plastic include D types of performance parameters used to characterize the performance of the plastic.

[0009] Based on the plastic baseline performance parameters, the plastic baseline process parameters are obtained through the first optimization model. The plastic baseline process parameters are a set of M types of process parameters that enable the finished plastic to meet the plastic baseline performance parameters.

[0010] Obtain the target performance parameters of the plastic. The target performance parameters of the plastic are the target values ​​of the performance parameters of the plastic to be produced, which are a set of D types of performance parameters.

[0011] When the target performance parameters of the plastic are consistent with the reference performance parameters of the plastic, the reference process parameters of the plastic are directly used as the optimized process parameters of the plastic. Otherwise, based on the reference performance parameters of the plastic, the reference process parameters of the plastic, and the target performance parameters of the plastic, the optimized process parameters of the plastic are obtained through the second optimization model. The optimized process parameters of the plastic are a set of M types of process parameters that make the finished plastic meet the target performance parameters of the plastic.

[0012] Preferably, the method for obtaining the reference plastic and the reference performance parameters of the plastic is as follows:

[0013] Obtain the planned plastic production data within time T at the target plant. The planned plastic production data includes the types of plastics to be produced, the output of each type of plastic, and the target performance parameters of each type of plastic.

[0014] Based on the planned plastic production data, benchmark evaluation parameters are obtained for each type of plastic. These benchmark evaluation parameters are used to assess the suitability of the plastic's performance parameters as benchmark performance parameters.

[0015] The plastic with the highest suitability is selected as the benchmark plastic based on the benchmark evaluation parameters of all types of plastics, and its target performance parameters are used as the benchmark performance parameters.

[0016] Preferably, the method for obtaining the benchmark evaluation parameters for each type of plastic is as follows:

[0017] ;

[0018] ;

[0019] in, Let be the benchmark evaluation parameter for the k-th type of plastic. The production volume of the k-th type of plastic in the planned plastic production data. Let e ​​be the total output of all plastics in the planned plastic production data, and let A be an intermediate parameter. Let d be the target performance parameter of the k-th plastic. d is the average value of the d-th parameter among the target performance parameters of all types of plastics.

[0020] Preferably, the first optimization model includes:

[0021] The input layer is used to receive the plastic reference performance parameters and obtain the input vector. , ;

[0022] The first hidden layer, consisting of 128 neurons, uses the ReLU activation function and outputs... :

[0023] ;

[0024] The second hidden layer, consisting of 64 neurons, uses the ReLU activation function and outputs... :

[0025] ;

[0026] Output layer, used to output a vector of the plastic reference process parameters. :

[0027] ;

[0028] in, , and As weight, , and Represents bias. It represents the set of real numbers.

[0029] Preferably, the loss function used when training the first optimization model is... for:

[0030] ;

[0031] ;

[0032] ;

[0033] ;

[0034] in, and These are the first loss function and the second loss function, respectively. and As weight, This refers to the d-th parameter in the benchmark performance parameters. This refers to the d-th parameter in the simulation performance parameters of the plastic obtained through simulation manufacturing based on the output of the output layer. The controllable threshold of the d-th parameter in the benchmark performance parameters. For smoothing terms, It is a constant. The variance of the d-th parameter among the actual performance parameters of the L historically manufactured benchmark plastics.

[0035] Preferably, the method for obtaining optimized plastic process parameters through the second optimization model is as follows:

[0036] Based on the plastic baseline performance parameters, the plastic baseline process parameters, and the plastic target performance parameters, an optimization objective function and constraints are established through a fitting model.

[0037] Multiple candidate solutions are obtained by solving the objective function and constraints.

[0038] An evaluation index is established to comprehensively evaluate multiple candidate solutions from multiple dimensions, and the optimal solution is selected from the multiple candidate solutions.

[0039] Preferably, the method for establishing the optimization objective function and the constraints is as follows:

[0040] Establish fitting models for the changes in each of the aforementioned performance parameters and the changes in multiple of the aforementioned process parameters:

[0041] ;

[0042] in, Let be the dependent variable and represent the change in the d-th performance parameter. , Let m be the independent variable and represent the change in the m-th process parameter. The coefficient of the b-th term of the polynomial is used to calculate the change of the d-th performance parameter, where B is the number of polynomial terms.

[0043] The method for establishing the optimization objective function is as follows:

[0044] ;

[0045] ;

[0046] in, As an optimization target, , This refers to the d-th parameter in the target performance parameter X of the plastic. Let d be the d-th parameter in the baseline performance parameters, and P be the penalty term. The penalty term weights are defined by `if`, which is a truth function. This is the error tolerance value for the d-th parameter among the target performance parameters of the plastic;

[0047] The method for establishing the constraints is to set a threshold range for each of the process parameters.

[0048] Preferably, the method for obtaining multiple candidate solutions based on the optimization objective function and constraints is as follows:

[0049] The first type of candidate process parameters Y are obtained by solving the objective function and constraints using the gradient descent method, and then added to the set of candidate solutions.

[0050] The m-th process parameter in the first type of candidate process parameters Y Centered on, Establish normal distributions respectively :

[0051] ;

[0052] Based on normal distribution Single-point sampling yields a result based on Disturbance process value , all of Combine the parameters to obtain the second type of candidate process parameters. Repeat this operation to obtain multiple second type of candidate process parameters. Add all the second type of candidate process parameters to the set of candidate solutions.

[0053] Preferably, the method for establishing evaluation indicators to comprehensively evaluate multiple candidate solutions from multiple dimensions is as follows:

[0054] Let the i-th candidate solution in the set of candidate solutions be... , , The total number of candidate solutions;

[0055] The evaluation parameters of the i-th candidate solution in the set of candidate solutions are obtained based on the deviation of the candidate solutions from the target performance parameters of the plastic and the cost consumption. :

[0056] ;

[0057] ;

[0058] ;

[0059] in, and As weight, It is a natural constant. To achieve the fitting model based on The calculated performance parameters, The target performance parameters of the plastic are... Representative process parameters are The energy cost consumed during the process, To find the Euclidean norm;

[0060] The candidate solution with the smallest evaluation parameter is selected as the plastic optimization process parameter.

[0061] This invention also provides a neural network-based plastic manufacturing process parameter optimization system, applied to the aforementioned neural network-based plastic manufacturing process parameter optimization method, comprising:

[0062] The reference plastic selection module is used to select a reference plastic and obtain the plastic reference performance parameters of the reference plastic. The plastic reference performance parameters are used as reference values ​​for the performance parameters of the plastic. The performance parameters of the plastic include D types of performance parameters used to characterize the performance of the plastic.

[0063] The benchmark process parameter calculation module is used to obtain the benchmark process parameters of plastic based on the benchmark performance parameters of plastic through the first optimization model. The benchmark process parameters of plastic are a set of M types of process parameters that make the finished plastic meet the benchmark performance parameters of plastic.

[0064] The plastic target performance parameter acquisition module is used to acquire plastic target performance parameters, which are the target values ​​of the performance parameters of the plastic to be produced, and are a set of D types of performance parameters;

[0065] The plastic optimization process parameter calculation module directly uses the plastic baseline process parameters as the plastic optimization process parameters when the plastic target performance parameters are consistent with the plastic baseline performance parameters. Otherwise, it obtains the plastic optimization process parameters through the second optimization model based on the plastic baseline performance parameters, plastic baseline process parameters, and plastic target performance parameters. The plastic optimization process parameters are a set of M types of process parameters that make the finished plastic meet the plastic target performance parameters.

[0066] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0067] This invention obtains a benchmark plastic and benchmark performance parameters of the plastic, and uses them as input and prior basis for the optimization model. It uses a stable reference value as the benchmark for the process parameters of other plastics, so that the optimization of process parameters has good transferability and stability, and can quickly generate usable solutions in new target scenarios, thereby improving the practicality and robustness of the overall system.

[0068] This invention selects the most representative plastic for future production as the benchmark plastic. When the target performance parameters are consistent with the benchmark performance parameters, the benchmark process parameters can be used directly to avoid unnecessary consumption of computing resources. When they are inconsistent, the optimization model is automatically called to obtain adjustment parameters, thereby improving the system's operating efficiency and intelligence level.

[0069] This invention combines baseline process knowledge and target performance parameters with a second optimization model to achieve adaptive adjustment of process parameters, eliminating the need for extensive re-testing and verification, effectively shortening the parameter debugging cycle and reducing development costs;

[0070] When selecting a benchmark plastic, this invention is based on the target factory's planned production data within a set time period, so that the evaluation results can be adapted to the current production reality and recent process capabilities of the factory. It also comprehensively considers the proportion of the plastic's output in the production plan and the concentration of its target performance parameters relative to the overall plastic performance distribution, thereby more scientifically reflecting the target factory's current process capabilities and mainstream product characteristics. Attached Figure Description

[0071] Figure 1 This is a flowchart illustrating the method for optimizing plastic manufacturing process parameters based on neural networks provided in Embodiment 1 of the present invention.

[0072] Figure 2 This is a schematic diagram of the principle of the neural network-based plastic manufacturing process parameter optimization system provided in Embodiment 2 of the present invention. Detailed Implementation

[0073] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0074] Example 1

[0075] This embodiment provides a method for optimizing plastic manufacturing process parameters based on neural networks. (See attached document.) Figure 1 This includes the following steps:

[0076] Step S1: Select a reference plastic and obtain the plastic reference performance parameters of the reference plastic. The plastic reference performance parameters are used as reference values ​​for the performance parameters of the plastic. The performance parameters of the plastic include D types of performance parameters used to characterize the performance of the plastic.

[0077] In this embodiment, the method for obtaining the reference plastic and the reference performance parameters of the plastic is as follows:

[0078] Obtain the planned plastic production data within time T at the target plant. The planned plastic production data includes the types of plastics to be produced, the output of each type of plastic, and the target performance parameters of each type of plastic.

[0079] Based on the planned plastic production data, benchmark evaluation parameters are obtained for each type of plastic. These benchmark evaluation parameters are used to assess the suitability of the plastic's performance parameters as benchmark performance parameters.

[0080] The plastic with the highest suitability is selected as the benchmark plastic based on the benchmark evaluation parameters of all types of plastics, and its target performance parameters are used as the benchmark performance parameters.

[0081] Based on this, the method for obtaining the benchmark evaluation parameters for each type of plastic is as follows:

[0082] ;

[0083] ;

[0084] in, Let be the benchmark evaluation parameter for the k-th type of plastic. The production volume of the k-th type of plastic in the planned plastic production data. Let e ​​be the total output of all plastics in the planned plastic production data, and let A be an intermediate parameter. Let d be the target performance parameter of the k-th plastic. d is the average value of the d-th parameter among the target performance parameters of all types of plastics.

[0085] In this step, by using planned data from the target factory's current production cycle as the basis for analysis, the selected benchmark plastic accurately reflects the factory's current product structure and manufacturing priorities. The benchmark evaluation parameters used in this embodiment comprehensively consider production volume percentages. and performance concentration While ensuring sufficient sample support for the benchmark plastics, it also ensures that their performance parameters are at a general level, meaning that the selection process balances quantitative representativeness and performance stability. The intermediate parameter A represents performance deviation, and a normalized form is used to measure the relative differences in multidimensional performance indicators. It is evaluated using the Sigmoid function, effectively reducing the impact of individual extreme values ​​or a few outliers on the evaluation results, thus enhancing the robustness and universality of the results.

[0086] Step S2: Based on the plastic baseline performance parameters, obtain the plastic baseline process parameters through the first optimization model. The plastic baseline process parameters are a set of M types of process parameters that make the finished plastic meet the plastic baseline performance parameters.

[0087] As a preferred embodiment, the first optimization model includes:

[0088] The input layer is used to receive the plastic reference performance parameters and obtain the input vector. , ;

[0089] The first hidden layer, consisting of 128 neurons, uses the ReLU activation function and outputs... :

[0090] ;

[0091] The second hidden layer, consisting of 64 neurons, uses the ReLU activation function and outputs... :

[0092] ;

[0093] Output layer, used to output a vector of the plastic reference process parameters. :

[0094] ;

[0095] in, , and As weight, , and Represents bias. It represents the set of real numbers.

[0096] Preferably, the loss function used when training the first optimization model is... for:

[0097] ;

[0098] ;

[0099] ;

[0100] ;

[0101] in, and These are the first loss function and the second loss function, respectively. and As weight, This refers to the d-th parameter in the benchmark performance parameters. This refers to the d-th parameter in the simulation performance parameters of the plastic obtained through simulation manufacturing based on the output of the output layer. The controllable threshold of the d-th parameter in the benchmark performance parameters. For example, a smoothing term can be taken as... , It is a constant. The variance of the d-th parameter among the actual performance parameters of the L historically manufactured benchmark plastics.

[0102] The first optimization model in this embodiment uses two ReLU hidden layers, which has strong nonlinear modeling capabilities. At the same time, the number of network parameters is moderate, which can improve training efficiency and generalization ability while ensuring the expressive power of the model. It is suitable for modeling complex mapping relationships between various plastic properties and process parameters.

[0103] During model training, the mean square error between the target plastic properties and the simulated manufacturing properties is incorporated. And improved Huber losses Optimize network performance. This includes mitigating the losses of traditional Huber networks. The design represents a threshold in the Huber loss, which determines the level of error at which a larger penalty is triggered. Here, it is set to a value of [value missing]. , The larger the value, the more forgiving it is; that is, only large deviations are considered abnormal. Simply put... This refers to the penalty threshold of the d-th parameter in the baseline performance parameters, enabling the model to automatically adapt to indicators with different physical quantities and unit differences, thereby improving the reliability of the loss function. A coefficient is set here. Make its size in It is controllable on the basis of [the above], generally speaking It can be set to 1-3.

[0104] Step S3: Obtain the target performance parameters of the plastic. The target performance parameters of the plastic are the target values ​​of the performance parameters of the plastic to be produced, which are a set of D types of performance parameters.

[0105] Step S4: When the target performance parameters of the plastic are consistent with the reference performance parameters of the plastic, the reference process parameters of the plastic are directly used as the optimized process parameters of the plastic. Otherwise, based on the reference performance parameters of the plastic, the reference process parameters of the plastic, and the target performance parameters of the plastic, the optimized process parameters of the plastic are obtained through the second optimization model. The optimized process parameters of the plastic are a set of M types of process parameters that make the finished plastic meet the target performance parameters of the plastic.

[0106] The method for obtaining optimized plastic process parameters through the second optimization model is as follows:

[0107] Based on the plastic baseline performance parameters, the plastic baseline process parameters, and the plastic target performance parameters, an optimization objective function and constraints are established through a fitting model.

[0108] Multiple candidate solutions are obtained by solving the objective function and constraints.

[0109] An evaluation index is established to comprehensively evaluate multiple candidate solutions from multiple dimensions, and the optimal solution is selected from the multiple candidate solutions.

[0110] Based on this, the method for establishing the optimization objective function and the constraints is as follows:

[0111] Establish fitting models for the changes in each of the aforementioned performance parameters and the changes in multiple of the aforementioned process parameters:

[0112] ;

[0113] in, Let be the dependent variable and represent the change in the d-th performance parameter. , Let m be the independent variable and represent the change in the m-th process parameter. The coefficient of the b-th term of the polynomial is used to calculate the change of the d-th performance parameter, where B is the number of polynomial terms.

[0114] The method for establishing the optimization objective function is as follows:

[0115] ;

[0116] ;

[0117] in, As an optimization target, , This refers to the d-th parameter in the target performance parameters of the plastic. Let d be the d-th parameter in the baseline performance parameters, and P be the penalty term. The penalty term weights are defined by `if`, which is a truth function. This is the error tolerance value of the d-th parameter in the target performance parameter X of the plastic, that is, X here is a set of multiple parameters;

[0118] The method for establishing the constraints is to set a threshold range for each of the process parameters.

[0119] Based on the above scheme, the method for obtaining multiple candidate solutions by solving the objective function and constraints is as follows:

[0120] The first type of candidate process parameters Y are obtained by solving the objective function and constraints using the gradient descent method, and then added to the set of candidate solutions.

[0121] The m-th process parameter in the first type of candidate process parameters Y Centered on, Establish normal distributions respectively :

[0122] ;

[0123] Based on normal distribution Single-point sampling yields a result based on Disturbance process value , all of Combine the parameters to obtain the second type of candidate process parameters. Repeat this operation to obtain multiple second type of candidate process parameters. Add all the second type of candidate process parameters to the set of candidate solutions.

[0124] Finally, the method for establishing evaluation indicators to comprehensively evaluate multiple candidate solutions from multiple dimensions is as follows:

[0125] Let the i-th candidate solution in the set of candidate solutions be... , , The total number of candidate solutions;

[0126] The evaluation parameters of the i-th candidate solution in the set of candidate solutions are obtained based on the deviation of the candidate solutions from the target performance parameters of the plastic and the cost consumption. :

[0127] ;

[0128] ;

[0129] ;

[0130] in, and As weight, It is a natural constant. To achieve the fitting model based on The calculated performance parameters, The target performance parameters of the plastic are... Representative process parameters are The energy cost consumed during the process, To find the Euclidean norm;

[0131] The candidate solution with the smallest evaluation parameter is selected as the plastic optimization process parameter.

[0132] In this embodiment, a function relating performance and process is constructed based on a polynomial fitting model. Specifically, the second optimization model is used to obtain optimized plastic process parameters that meet the performance target and have cost advantages by fitting modeling and target optimization, based on known plastic baseline performance parameters, plastic baseline process parameters, and plastic target performance parameters. Specifically, during optimization through fitting, the difference between the baseline performance and the target performance is calculated, and the process offset is used as the optimization variable. The optimization objective is clear, the path is well-defined, and it helps to quickly converge to a reasonable process solution.

[0133] In establishing the objective function, besides introducing a term to minimize the error... It also introduces the product of penalty weights and penalty terms. As a penalty, a larger penalty is imposed on unstable components with excessive deviations to improve optimization accuracy. After obtaining an optimal option, namely the first type of candidate process parameter Y, it is subjected to Gaussian perturbation to obtain multiple second type of candidate process parameters, enriching the diversity of the solution space. This approach can avoid getting trapped in local optima and guide the model to explore more potential global better solutions within the feasible region.

[0134] Based on this, for the set of candidate solutions consisting of the first type of candidate process parameters and the second type of candidate process parameters, the evaluation function comprehensively considers performance deviation and energy cost, and adopts an exponential weighting mechanism to automatically adjust the priority according to their respective values, which has strong adaptability and practical value.

[0135] In summary, the solution in this embodiment first selects a representative benchmark plastic and collects its performance and process parameters. Based on this, a first optimization model is established using a neural network to predict the benchmark process. If the target performance parameters are inconsistent with the benchmark, a second optimization model is used to construct a mapping relationship between performance changes and process changes based on a fitting method, establish an optimization objective function and constraints, obtain an initial solution using the gradient method, and then generate multiple candidate solutions through Gaussian perturbation. Finally, a multi-dimensional evaluation index is constructed by combining performance error and cost to comprehensively evaluate the candidate solutions, and the optimal solution is selected as the final process parameter output. This achieves transfer analysis based on a highly accurate reference benchmark, avoiding the need to call complex neural networks for extensive training and optimal output for each production run, thus improving the efficiency, stability, and adaptability of plastic performance optimization to different plastic types.

[0136] Example 2

[0137] This embodiment uses a neural network-based plastic manufacturing process parameter optimization system, which is applied to the neural network-based plastic manufacturing process parameter optimization method described in the above embodiment. See [link / reference]. Figure 2 ,include:

[0138] The reference plastic selection module is used to select a reference plastic and obtain the plastic reference performance parameters of the reference plastic. The plastic reference performance parameters are used as reference values ​​for the performance parameters of the plastic. The performance parameters of the plastic include D types of performance parameters used to characterize the performance of the plastic.

[0139] The benchmark process parameter calculation module is used to obtain the benchmark process parameters of plastic based on the benchmark performance parameters of plastic through the first optimization model. The benchmark process parameters of plastic are a set of M types of process parameters that make the finished plastic meet the benchmark performance parameters of plastic.

[0140] The plastic target performance parameter acquisition module is used to acquire plastic target performance parameters, which are the target values ​​of the performance parameters of the plastic to be produced, and are a set of D types of performance parameters;

[0141] The plastic optimization process parameter calculation module directly uses the plastic baseline process parameters as the plastic optimization process parameters when the plastic target performance parameters are consistent with the plastic baseline performance parameters. Otherwise, it obtains the plastic optimization process parameters through the second optimization model based on the plastic baseline performance parameters, plastic baseline process parameters, and plastic target performance parameters. The plastic optimization process parameters are a set of M types of process parameters that make the finished plastic meet the plastic target performance parameters.

[0142] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for optimizing plastic manufacturing process parameters based on neural networks, characterized in that, Includes the following steps: Select a reference plastic and obtain the plastic reference performance parameters of the reference plastic. The plastic reference performance parameters are used as the reference values ​​for the performance parameters of the plastic. The performance parameters of the plastic include D types of performance parameters used to characterize the performance of the plastic. Based on the plastic baseline performance parameters, the plastic baseline process parameters are obtained through the first optimization model. The plastic baseline process parameters are a set of M types of process parameters that enable the finished plastic to meet the plastic baseline performance parameters. Obtain the target performance parameters of the plastic. The target performance parameters of the plastic are the target values ​​of the performance parameters of the plastic to be produced, which are a set of D types of performance parameters. When the target performance parameters of plastic are consistent with the reference performance parameters of plastic, the reference process parameters of plastic are directly used as the optimized process parameters of plastic. Otherwise, based on the reference performance parameters of plastic, the reference process parameters of plastic, and the target performance parameters of plastic, the optimized process parameters of plastic are obtained through the second optimization model. The optimized process parameters of plastic are a set of M types of process parameters that make the finished plastic meet the target performance parameters of plastic. The method for obtaining the reference plastic and the reference performance parameters of the plastic is as follows: Obtain the planned plastic production data within time T at the target plant. The planned plastic production data includes the types of plastics to be produced, the output of each type of plastic, and the target performance parameters of each type of plastic. Based on the planned plastic production data, benchmark evaluation parameters are obtained for each type of plastic. These benchmark evaluation parameters are used to assess the suitability of the plastic's performance parameters as benchmark performance parameters. The plastic with the highest suitability is selected as the benchmark plastic based on the benchmark evaluation parameters of all types of plastics, and its target performance parameters are used as the benchmark performance parameters. The first optimization model includes: The input layer is used to receive the plastic reference performance parameters and obtain the input vector. , ; The first hidden layer, consisting of 128 neurons, uses the ReLU activation function and outputs... : ； The second hidden layer, consisting of 64 neurons, uses the ReLU activation function and outputs... : ； Output layer, used to output a vector of the plastic reference process parameters. : ； in, , and As weight, , and Represents bias. Represents the set of real numbers; The method for obtaining optimized plastic process parameters through the second optimization model is as follows: Based on the plastic baseline performance parameters, the plastic baseline process parameters, and the plastic target performance parameters, an optimization objective function and constraints are established through a fitting model. Multiple candidate solutions are obtained by solving the objective function and constraints. An evaluation index is established to comprehensively evaluate multiple candidate solutions from multiple dimensions, and the optimal solution is selected from the multiple candidate solutions. The method for establishing the optimization objective function and the constraints is as follows: Establish fitting models for the changes in each of the aforementioned performance parameters and the changes in multiple of the aforementioned process parameters: ； in, Let be the dependent variable and represent the change in the d-th performance parameter. , Let m be the independent variable and represent the change in the m-th process parameter. The coefficient of the b-th term of the polynomial is used to calculate the change of the d-th performance parameter, where B is the number of polynomial terms. The method for establishing the optimization objective function is as follows: ； ； in, As an optimization target, , This refers to the d-th parameter in the target performance parameter X of the plastic. Let d be the d-th parameter in the baseline performance parameters, and P be the penalty term. The penalty term weights are defined by `if`, which is a truth function. This is the error tolerance value for the d-th parameter among the target performance parameters of the plastic; The method for establishing the constraints is as follows: set a threshold range for each of the process parameters; The method for obtaining multiple candidate solutions based on solving the objective function and constraints is as follows: The first type of candidate process parameters Y are obtained by solving the objective function and constraints using the gradient descent method, and then added to the set of candidate solutions. The m-th process parameter in the first type of candidate process parameters Y Centered on, Establish normal distributions respectively : ； Based on normal distribution Single-point sampling yields a result based on Disturbance process value , all of Combine the parameters to obtain the second type of candidate process parameters. Repeat this operation to obtain multiple second type of candidate process parameters. Add all the second type of candidate process parameters to the set of candidate solutions.

2. The method for optimizing plastic manufacturing process parameters based on neural networks according to claim 1, characterized in that, The method for obtaining the benchmark evaluation parameters for each type of plastic is as follows: ； ； in, Let be the benchmark evaluation parameter for the k-th type of plastic. The production volume of the k-th type of plastic in the planned plastic production data. Let e ​​be the total output of all plastics in the planned plastic production data, and let A be an intermediate parameter. Let d be the target performance parameter of the k-th plastic. d is the average value of the d-th parameter among the target performance parameters of all types of plastics.

3. The method for optimizing plastic manufacturing process parameters based on neural networks according to claim 1, characterized in that, The loss function used when training the first optimized model for: ； ； ； ； in, and These are the first loss function and the second loss function, respectively. and As weight, This refers to the d-th parameter in the benchmark performance parameters. This refers to the d-th parameter in the simulation performance parameters of the plastic obtained through simulation manufacturing based on the output of the output layer. The controllable threshold of the d-th parameter in the benchmark performance parameters. For smoothing terms, It is a constant. The variance of the d-th parameter among the actual performance parameters of the L historically manufactured benchmark plastics.

4. The method for optimizing plastic manufacturing process parameters based on neural networks according to claim 1, characterized in that, The method for establishing evaluation indicators to comprehensively evaluate multiple candidate solutions from multiple dimensions is as follows: Let the i-th candidate solution in the set of candidate solutions be... , , The total number of candidate solutions; The evaluation parameters of the i-th candidate solution in the set of candidate solutions are obtained based on the deviation of the candidate solutions from the target performance parameters of the plastic and the cost consumption. : ； ； ； in, and As weight, It is a natural constant. To achieve the fitting model based on The calculated performance parameters, The target performance parameters of the plastic are... Representative process parameters are The energy cost consumed during the process, To find the Euclidean norm; The candidate solution with the smallest evaluation parameter is selected as the plastic optimization process parameter.

5. A neural network-based plastic manufacturing process parameter optimization system, applied to the neural network-based plastic manufacturing process parameter optimization method according to any one of claims 1-4, characterized in that, include: The reference plastic selection module is used to select a reference plastic and obtain the plastic reference performance parameters of the reference plastic. The plastic reference performance parameters are used as reference values ​​for the performance parameters of the plastic. The performance parameters of the plastic include D types of performance parameters used to characterize the performance of the plastic. The benchmark process parameter calculation module is used to obtain the benchmark process parameters of plastic based on the benchmark performance parameters of plastic through the first optimization model. The benchmark process parameters of plastic are a set of M types of process parameters that make the finished plastic meet the benchmark performance parameters of plastic. The plastic target performance parameter acquisition module is used to acquire plastic target performance parameters, which are the target values ​​of the performance parameters of the plastic to be produced, and are a set of D types of performance parameters; The plastic optimization process parameter calculation module directly uses the plastic baseline process parameters as the plastic optimization process parameters when the plastic target performance parameters are consistent with the plastic baseline performance parameters. Otherwise, it obtains the plastic optimization process parameters through the second optimization model based on the plastic baseline performance parameters, plastic baseline process parameters, and plastic target performance parameters. The plastic optimization process parameters are a set of M types of process parameters that make the finished plastic meet the plastic target performance parameters. The method for obtaining the reference plastic and the reference performance parameters of the plastic is as follows: Obtain the planned plastic production data within time T at the target plant. The planned plastic production data includes the types of plastics to be produced, the output of each type of plastic, and the target performance parameters of each type of plastic. Based on the planned plastic production data, benchmark evaluation parameters are obtained for each type of plastic. These benchmark evaluation parameters are used to assess the suitability of the plastic's performance parameters as benchmark performance parameters. The plastic with the highest suitability is selected as the benchmark plastic based on the benchmark evaluation parameters of all types of plastics, and its target performance parameters are used as the benchmark performance parameters. The first optimization model includes: The input layer is used to receive the plastic reference performance parameters and obtain the input vector. , ; The first hidden layer, consisting of 128 neurons, uses the ReLU activation function and outputs... : ； The second hidden layer, consisting of 64 neurons, uses the ReLU activation function and outputs... : ； Output layer, used to output a vector of the plastic reference process parameters. : ； in, , and As weight, , and Represents bias. Represents the set of real numbers; The method for obtaining optimized plastic process parameters through the second optimization model is as follows: Based on the plastic baseline performance parameters, the plastic baseline process parameters, and the plastic target performance parameters, an optimization objective function and constraints are established through a fitting model. Multiple candidate solutions are obtained by solving the objective function and constraints. An evaluation index is established to comprehensively evaluate multiple candidate solutions from multiple dimensions, and the optimal solution is selected from the multiple candidate solutions. The method for establishing the optimization objective function and the constraints is as follows: Establish fitting models for the changes in each of the aforementioned performance parameters and the changes in multiple of the aforementioned process parameters: ； in, Let be the dependent variable and represent the change in the d-th performance parameter. , Let m be the independent variable and represent the change in the m-th process parameter. The coefficient of the b-th term of the polynomial is used to calculate the change of the d-th performance parameter, where B is the number of polynomial terms. The method for establishing the optimization objective function is as follows: ； ； in, As an optimization target, , This refers to the d-th parameter in the target performance parameter X of the plastic. Let d be the d-th parameter in the baseline performance parameters, and P be the penalty term. The penalty term weights are defined by `if`, which is a truth function. This is the error tolerance value for the d-th parameter among the target performance parameters of the plastic; The method for establishing the constraints is as follows: set a threshold range for each of the process parameters; The method for obtaining multiple candidate solutions based on solving the objective function and constraints is as follows: The first type of candidate process parameters Y are obtained by solving the objective function and constraints using the gradient descent method, and then added to the set of candidate solutions. The m-th process parameter in the first type of candidate process parameters Y Centered on, Establish normal distributions respectively : ； Based on normal distribution Single-point sampling yields a result based on Disturbance process value , all of Combine the parameters to obtain the second type of candidate process parameters. Repeat this operation to obtain multiple second type of candidate process parameters. Add all the second type of candidate process parameters to the set of candidate solutions.

Images