Chemical production prediction and parameter optimization system based on GRU neural network and multi-target Bayesian optimization

Abstract

The invention relates to a chemical production prediction and parameter optimization system based on a GRU neural network and multi-target Bayesian optimization, and the system comprises a data collection module, a data preprocessing module, a GRU neural network prediction module, a multi-target Bayesian optimization module, a visualization module, a big data module, a data comparison module, and a recording module. According to the method, the prediction precision is improved, the GRU neural network is adopted to process the time sequence data, the production result can be predicted more accurately, the optimization efficiency is high, the Bayesian optimization can find out the optimal parameter combination within a small number of experiments, the production efficiency is improved, a large number of training samples are provided for the GRU neural network model through the big data module, and the prediction accuracy is improved. By analyzing the samples, the model can learn and understand modes, relationships and rules in input data, more data generally means a better training effect, and the GRU neural network model can predict the result of unknown data more accurately.

Description

Technical Field

[0001] The present invention relates to the field of chemical production prediction and parameter optimization, and in particular to a chemical production prediction and parameter optimization system based on GRU neural network and multi-objective Bayesian optimization. Background Art

[0002] Traditional empirical methods for chemical production prediction and parameter optimization often rely on the experience of operators and lack systematicity and precision. With the development of artificial intelligence and machine learning technologies, the use of neural networks for production prediction and optimization has gradually become possible.

[0003] GRU (Gated Recurrent Unit) neural network is an improved recurrent neural network with good ability to process sequence data and is suitable for time series prediction tasks. Bayesian optimization is a problem used for global optimization. It estimates the objective function by building a surrogate model and selects the next query point by maximizing the expected improvement (EI) of the surrogate model to find the optimal solution with the least number of experiments. Combining GRU neural network and Bayesian optimization can effectively improve the accuracy of chemical production prediction and the efficiency of parameter optimization.

[0004] Therefore, we propose a chemical production prediction and parameter optimization system based on GRU neural network and multi-objective Bayesian optimization. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art, adapt to actual needs, and provide a chemical production prediction and parameter optimization system based on GRU neural network and multi-objective Bayesian optimization to solve the technical problems that the current traditional chemical production prediction and parameter optimization empirical methods often rely on the experience of operators and lack systematicity and accuracy.

[0006] In order to achieve the purpose of the present invention, the technical solution adopted by the present invention is: designing a chemical production prediction and parameter optimization system based on GRU neural network and multi-objective Bayesian optimization, including a data acquisition module, a data preprocessing module, a GRU neural network prediction module, a multi-objective Bayesian optimization module, a visualization module, a big data module, a data comparison module, and a recording module.

[0007] The data acquisition module is used to collect various parameter data in the chemical production process, including but not limited to temperature, pressure, flow rate, particle size, etc.

[0008] The data preprocessing module cleans, normalizes and extracts features from the collected data, and the processed data is used for model training and prediction;

[0009] The GRU neural network prediction module: based on the GRU neural network model, trains the preprocessed data to generate a production prediction model, and predicts the production results based on the model;

[0010] The target Bayesian optimization module: based on the production prediction model, uses a multi-objective Bayesian optimization algorithm to optimize the production parameters and generate the optimal production parameter combination;

[0011] The visualization module is used to display the prediction results and optimization results, and provide a human-computer interaction interface to facilitate operators to view and adjust production parameters;

[0012] The big data module: Big data provides a large number of training samples for the GRU neural network model. By analyzing these samples, the model can learn and understand the patterns, relationships and rules in the input data. More data usually means better training results, enabling the model to more accurately predict the results of unknown data;

[0013] The data comparison module: after creating the comparison task, it will query the production result prediction database obtained by the production prediction model and the database in the big data at the same time, obtain the query results sorted by the primary key, and compare and verify the consistency of the production result prediction database obtained by the production prediction model and the database in the big data;

[0014] The recording module: adopts Single trigger mode: In this mode, all data within the trigger condition time period will be recorded. The user can set the Logging conditions in the Time area, such as start trigger Toggle on, end trigger Toggle off and trigger times. This mode is suitable for recording error parameter data within a specific time period.

[0015] Preferably, it also includes a chemical production prediction process, a parameter optimization process and a result display process;

[0016] The chemical production prediction process includes the following steps:

[0017] S1. Data collection: collect various parameter data from chemical production equipment in real time and store the data in the database;

[0018] S2, data preprocessing: clean the collected data, remove noise and outliers, perform normalization, and extract feature parameters;

[0019] S3. Model training: Based on the GRU neural network model, the big data model is established using the big data module, and the GRU neural network is connected to the big data model through the fit method, providing a large number of training samples for the GRU neural network model. By analyzing these samples, the model can learn and understand the patterns, relationships and laws in the input data, train the pre-processed data, generate a production prediction model, and predict the production results based on the model. The trained neural network model can not only be used for the prediction of current production, but also for the analysis of future production trends and anomaly detection, and has strong practical value and prediction ability;

[0020] S4. Data comparison: After creating the comparison task, the production result prediction database obtained by the production prediction model and the production result prediction database obtained in the big data will be queried at the same time to obtain the query results sorted by the primary key, and the production result prediction database obtained by the production prediction model and the production result prediction database obtained in the big data will be compared and verified for consistency. When the comparison fails, the data will be corrected. The staff will determine the errors in the data in the production result prediction database and the production result prediction database obtained in the big data, and replace the wrong data parameters with the correct data parameters;

[0021] S5. Data recording: When the comparison is completed, the recording module is used to record the erroneous comparison. When the next production prediction model is used to obtain the same production result prediction, the error record of the recording module will be triggered to prompt the staff.

[0022] Preferably, the parameter optimization process comprises the following steps:

[0023] S1. Parameter optimization: Based on the trained production prediction model, the multi-objective Bayesian optimization algorithm is used to optimize the production parameters. Bayesian optimization constructs an agent model to evaluate the effects of different parameter combinations and select the optimal parameter combination.

[0024] Preferably, the specific steps of parameter optimization are as follows:

[0025] S1. Initial sample point selection: Randomly select several groups of production parameters as initial sample points, and add characteristic variables to the initial sample points. By adding characteristic variables, the data can be analyzed more carefully, so as to discover the potential patterns and relationships in the data, which helps the model to better understand the data and improve the accuracy of prediction. After the characteristic variables are added, the corresponding production results are calculated;

[0026] S2. Constructing a proxy model: Based on the initial sample points, a Gaussian process is used to construct a proxy model, which can approximate the objective function.

[0027] S3. Select acquisition function: Select an acquisition function, which is used to evaluate the potential benefit of selecting the next sample point in the surrogate model;

[0028] S4, optimize the acquisition function: maximize the acquisition function in the agent model and determine the next acquisition point, that is, the next set of production parameters;

[0029] S5, update model: collect new sample points and their corresponding production results, and update the proxy model;

[0030] S6. Iterative optimization: Repeat the above optimization process until convergence or reaching the preset number of iterations.

[0031] Preferably, in S1 of the specific step of parameter optimization, various characteristic variables, such as product output, number of machines and equipment, raw material consumption, energy consumption, and time series, need to be added to the sample points. These characteristics can be qualitative or quantitative, depending on the nature of the data and the requirements of the model.

[0032] Preferably, the result display process includes the following steps:

[0033] S1. Through the visualization module, the prediction results and optimization results are displayed to the operator, and an interface for adjusting production parameters is provided.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] 1. The present invention improves prediction accuracy: the GRU neural network is used to process time series data, which can more accurately predict production results.

[0036] 2. The present invention has high optimization efficiency: Bayesian optimization can find the optimal parameter combination within a relatively small number of experiments, thereby improving production efficiency.

[0037] 3. The present invention is user-friendly: it provides a visual interface to facilitate operators to view and adjust production parameters.

[0038] 4. The present invention provides a large number of training samples for the GRU neural network model through the big data module. By analyzing these samples, the model can learn and understand the patterns, relationships and laws in the input data. More data usually means better training effects, so that the GRU neural network model can more accurately predict the results of unknown data.

[0039] 5. The present invention uses a data comparison module: after creating a comparison task, it will simultaneously query the production result prediction database obtained by the production prediction model and the database in the big data, obtain the query results sorted by the primary key, and perform a consistency comparison and verification on the production result prediction database obtained by the production prediction model and the database in the big data, thereby increasing the accuracy of data parameters.

[0040] 6. The present invention uses a data recording module: when the comparison is completed, the recording module can be used to record the erroneous comparison. When the next production prediction model is used to obtain the same production result prediction, the error record of the recording module will be triggered to prompt the staff, so that they can avoid the same error in time when encountering it next time.

[0041] 7. By adding characteristic variables, the present invention can perform a more detailed analysis of the data, thereby discovering the potential patterns and relationships in the data, which helps the model better understand the data and improve the accuracy of the prediction.

[0042] Through the above methods, the present invention realizes accurate prediction and efficient optimization of chemical production processes and has high application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of the chemical production prediction process of the present invention;

[0044] Figure 2 It is a schematic diagram of the parameter optimization process of the present invention;

[0045] Figure 3 A schematic diagram is shown for the results of the present invention. DETAILED DESCRIPTION

[0046] The present invention is further described below in conjunction with the accompanying drawings and embodiments:

[0047] Chemical production prediction and parameter optimization system based on GRU neural network and multi-objective Bayesian optimization, see Figures 1 to 3 , including data acquisition module, data preprocessing module, GRU neural network prediction module, multi-objective Bayesian optimization module, visualization module, big data module, data comparison module, and recording module.

[0048] Data acquisition module: used to collect various parameter data in the chemical production process, including but not limited to temperature, pressure, flow rate, particle size, etc.;

[0049] Data preprocessing module: cleans, normalizes and extracts features from the collected data. The processed data is used for model training and prediction.

[0050] GRU neural network prediction module: Based on the GRU neural network model, the preprocessed data is trained to generate a production prediction model, and the production results are predicted based on the model;

[0051] Objective Bayesian optimization module: Based on the production prediction model, the multi-objective Bayesian optimization algorithm is used to optimize the production parameters and generate the optimal combination of production parameters;

[0052] Visualization module: used to display prediction results and optimization results, and provide a human-computer interaction interface to facilitate operators to view and adjust production parameters;

[0053] Big data module: Big data provides a large number of training samples for the GRU neural network model. By analyzing these samples, the model can learn and understand the patterns, relationships, and rules in the input data. More data usually means better training results, allowing the model to more accurately predict the results of unknown data.

[0054] Data comparison module: after creating the comparison task, the production result prediction database obtained by the production prediction model and the database in the big data will be queried at the same time, and the query results sorted by the primary key will be obtained. The production result prediction database obtained by the production prediction model and the database in the big data will be compared and verified for consistency;

[0055] Recording module: Single trigger mode: In this mode, all data within the trigger condition time period will be recorded. The user can set the Logging conditions in the Time area, such as start trigger Toggle on, end trigger Toggle off and trigger times. This mode is suitable for recording error parameter data within a specific time period.

[0056] For details, see Figure 1 , also includes chemical production prediction process, parameter optimization process and result display process;

[0057] The chemical production forecasting process includes the following steps:

[0058] S1. Data collection: collect various parameter data from chemical production equipment in real time and store the data in the database;

[0059] S2, data preprocessing: clean the collected data, remove noise and outliers, perform normalization, and extract feature parameters;

[0060] S3. Model training: Based on the GRU neural network model, the big data model is established using the big data module, and the GRU neural network is connected to the big data model through the fit method, providing a large number of training samples for the GRU neural network model. By analyzing these samples, the model can learn and understand the patterns, relationships and laws in the input data, train the pre-processed data, generate a production prediction model, and predict the production results based on the model. The trained neural network model can not only be used for the prediction of current production, but also for the analysis of future production trends and anomaly detection, and has strong practical value and prediction ability;

[0061] S4. Data comparison: After creating the comparison task, the production result prediction database obtained by the production prediction model and the production result prediction database obtained in the big data will be queried at the same time to obtain the query results sorted by the primary key, and the production result prediction database obtained by the production prediction model and the production result prediction database obtained in the big data will be compared and verified for consistency. When the comparison fails, the data will be corrected. The staff will determine the errors in the data in the production result prediction database and the production result prediction database obtained in the big data, and replace the wrong data parameters with the correct data parameters;

[0062] S5. Data recording: When the comparison is completed, the recording module is used to record the erroneous comparison. When the next production prediction model is used to obtain the same production result prediction, the error record of the recording module will be triggered to prompt the staff.

[0063] For further information, see Figure 2 , the parameter optimization process includes the following steps:

[0064] S1. Parameter optimization: Based on the trained production prediction model, the multi-objective Bayesian optimization algorithm is used to optimize the production parameters. Bayesian optimization constructs an agent model to evaluate the effects of different parameter combinations and select the optimal parameter combination.

[0065] It is worth noting that see Figure 2 , the specific steps of parameter optimization are as follows:

[0066] S1. Initial sample point selection: Randomly select several groups of production parameters as initial sample points, and add characteristic variables to the initial sample points. By adding characteristic variables, the data can be analyzed more carefully, so as to discover the potential patterns and relationships in the data, which helps the model to better understand the data and improve the accuracy of prediction. After the characteristic variables are added, the corresponding production results are calculated;

[0067] S2. Constructing a proxy model: Based on the initial sample points, a Gaussian process is used to construct a proxy model, which can approximate the objective function.

[0068] S3. Select acquisition function: Select an acquisition function, which is used to evaluate the potential benefit of selecting the next sample point in the surrogate model;

[0069] S4, optimize the acquisition function: maximize the acquisition function in the agent model and determine the next acquisition point, that is, the next set of production parameters;

[0070] S5, update model: collect new sample points and their corresponding production results, and update the proxy model;

[0071] S6. Iterative optimization: Repeat the above optimization process until convergence or reaching the preset number of iterations.

[0072] It is worth noting that see Figure 2 In S1 of the specific step of parameter optimization, various characteristic variables, such as product output, number of machines and equipment, raw material consumption, energy consumption, and time series, need to be added to the sample points. These characteristics can be qualitative or quantitative, depending on the nature of the data and the requirements of the model.

[0073] It is worth mentioning that see Figure 3 , the result display process includes the following steps:

[0074] S1. Through the visualization module, the prediction results and optimization results are displayed to the operator, and an interface for adjusting production parameters is provided.

[0075] Embodiment 1

[0076] The specific steps for parameter optimization are as follows:

[0077] S1. Initial sample point selection: Randomly select several groups of production parameters as initial sample points, and add characteristic variables to the initial sample points. By adding characteristic variables, the data can be analyzed more carefully, so as to discover the potential patterns and relationships in the data, which helps the model to better understand the data and improve the accuracy of prediction. After the characteristic variables are added, the corresponding production results are calculated;

[0078] S2. Constructing a proxy model: Based on the initial sample points, a Gaussian process is used to construct a proxy model, which can approximate the objective function.

[0079] S3. Select acquisition function: Select an acquisition function (such as expected improvement EI) that is used to evaluate the potential benefit of selecting the next sample point in the surrogate model;

[0080] S4, optimize the acquisition function: maximize the acquisition function in the agent model and determine the next acquisition point, that is, the next set of production parameters;

[0081] S5, update model: collect new sample points and their corresponding production results, and update the proxy model;

[0082] S6. Iterative optimization: Repeat the above optimization process until convergence or reaching the preset number of iterations.

[0083] The result presentation process includes the following steps:

[0084] S1. Through the visualization module, the prediction results and optimization results are displayed to the operator, and an interface for adjusting production parameters is provided.

[0085] Embodiment 2

[0086] Compared with the first embodiment, in the Bayesian optimization, multiple objectives are considered, such as the balance between production efficiency and product quality.

[0087] The specific steps for parameter optimization are as follows:

[0088] S1. Initial sample point selection: Randomly select several groups of production parameters as initial sample points, and add characteristic variables to the initial sample points. By adding characteristic variables, the data can be analyzed more carefully, so as to discover the potential patterns and relationships in the data, which helps the model to better understand the data and improve the accuracy of prediction. After the characteristic variables are added, the corresponding multiple objective function values ​​are calculated;

[0089] S2. Constructing a proxy model: Based on the initial sample points, a multi-objective proxy model is constructed using a Gaussian process, which can approximate each objective function;

[0090] S3. Select acquisition function: Select a multi-objective acquisition function (such as the Pareto frontier of expected improvement, EI) that is used to evaluate the potential benefit of selecting the next sample point in the surrogate model;

[0091] S4, optimize the acquisition function: maximize the multi-objective acquisition function in the agent model and determine the next acquisition point, that is, the next set of production parameters;

[0092] S5, update model: collect new sample points and their corresponding multiple objective function values, and update the multi-objective proxy model;

[0093] S6. Iterative optimization: Repeat the above optimization process until convergence or reaching the preset number of iterations.

[0094] The result presentation process includes the following steps:

[0095] S1. Through the visualization module, not only the optimization results of a single objective are displayed, but also the trade-off results of multiple objectives are displayed. Operators can choose different optimization strategies according to actual needs.

[0096] In addition, the components designed in the present invention are all universal standard parts or components known to technical personnel in this field. Their structures and principles can be known to technical personnel through technical manuals or through conventional experimental methods. They can be fully implemented by technical personnel in this field. Needless to say, the content protected by the present invention does not involve improvements to internal structures and methods.

[0097] The embodiments of the present invention disclose preferred embodiments, but are not limited thereto. A person skilled in the art can easily understand the spirit of the present invention based on the above embodiments and make different extensions and changes. However, as long as they do not deviate from the spirit of the present invention, they are all within the protection scope of the present invention.

Claims

1. A chemical production prediction and parameter optimization system based on GRU neural network and multi-objective Bayesian optimization, including a data acquisition module, a data preprocessing module, a GRU neural network prediction module, a multi-objective Bayesian optimization module, a visualization module, a big data module, a data comparison module, and a recording module, characterized in that: The data acquisition module is used to collect various parameter data in the chemical production process, including but not limited to temperature, pressure, flow rate, particle size, etc. The data preprocessing module cleans, normalizes and extracts features from the collected data, and the processed data is used for model training and prediction; The GRU neural network prediction module: based on the GRU neural network model, trains the preprocessed data to generate a production prediction model, and predicts the production results based on the model; The target Bayesian optimization module: based on the production prediction model, uses a multi-objective Bayesian optimization algorithm to optimize the production parameters and generate the optimal production parameter combination; The visualization module is used to display the prediction results and optimization results, and provide a human-computer interaction interface to facilitate operators to view and adjust production parameters; The big data module: Big data provides a large number of training samples for the GRU neural network model. By analyzing these samples, the model can learn and understand the patterns, relationships and rules in the input data. More data usually means better training results, enabling the model to more accurately predict the results of unknown data; The data comparison module: after creating the comparison task, it will query the production result prediction database obtained by the production prediction model and the database in the big data at the same time, obtain the query results sorted by the primary key, and compare and verify the consistency of the production result prediction database obtained by the production prediction model and the database in the big data; The recording module adopts the Single trigger mode: in this mode, all data within the trigger condition time period will be recorded. The user can set the Logging conditions in the Time area, such as start trigger Toggle on, end trigger Toggle off and trigger times. This mode is suitable for recording error parameter data within a specific time period.

2. The chemical production prediction and parameter optimization system based on GRU neural network and multi-objective Bayesian optimization as claimed in claim 1, characterized in that: It also includes chemical production prediction process, parameter optimization process and result display process; The chemical production prediction process includes the following steps: S1. Data collection: collect various parameter data from chemical production equipment in real time and store the data in the database; S2, data preprocessing: clean the collected data, remove noise and outliers, perform normalization, and extract feature parameters; S3. Model training: Based on the GRU neural network model, the big data model is established using the big data module, and the GRU neural network is connected to the big data model through the fit method, providing a large number of training samples for the GRU neural network model. By analyzing these samples, the model can learn and understand the patterns, relationships and laws in the input data, train the pre-processed data, generate a production prediction model, and predict the production results based on the model. The trained neural network model can not only be used for the prediction of current production, but also for the analysis of future production trends and anomaly detection, and has strong practical value and prediction ability; S4. Data comparison: After creating the comparison task, the production result prediction database obtained by the production prediction model and the production result prediction database obtained in the big data will be queried at the same time to obtain the query results sorted by the primary key, and the production result prediction database obtained by the production prediction model and the production result prediction database obtained in the big data will be compared and verified for consistency. When the comparison fails, the data will be corrected. The staff will determine the errors in the data in the production result prediction database and the production result prediction database obtained in the big data, and replace the wrong data parameters with the correct data parameters; S5. Data recording: When the comparison is completed, the recording module is used to record the erroneous comparison. When the next production prediction model is used to obtain the same production result prediction, the error record of the recording module will be triggered to prompt the staff.

3. The chemical production prediction and parameter optimization system based on GRU neural network and multi-objective Bayesian optimization as claimed in claim 2, characterized in that: The parameter optimization process includes the following steps: S1. Parameter optimization: Based on the trained production prediction model, the multi-objective Bayesian optimization algorithm is used to optimize the production parameters. Bayesian optimization constructs an agent model to evaluate the effects of different parameter combinations and select the optimal parameter combination.

4. The chemical production prediction and parameter optimization system based on GRU neural network and multi-objective Bayesian optimization as claimed in claim 3, characterized in that: The specific steps of parameter optimization are as follows: S1. Initial sample point selection: Randomly select several groups of production parameters as initial sample points, and add characteristic variables to the initial sample points. By adding characteristic variables, the data can be analyzed more carefully, so as to discover the potential patterns and relationships in the data, which helps the model to better understand the data and improve the accuracy of prediction. After the characteristic variables are added, the corresponding production results are calculated; S2. Constructing a proxy model: Based on the initial sample points, a Gaussian process is used to construct a proxy model, which can approximate the objective function. S3. Select acquisition function: Select an acquisition function, which is used to evaluate the potential benefit of selecting the next sample point in the surrogate model; S4, optimize the acquisition function: maximize the acquisition function in the agent model and determine the next acquisition point, that is, the next set of production parameters; S5, update model: collect new sample points and their corresponding production results, and update the proxy model; S6. Iterative optimization: Repeat the above optimization process until convergence or reaching the preset number of iterations.

5. The chemical production prediction and parameter optimization system based on GRU neural network and multi-objective Bayesian optimization as claimed in claim 4, characterized in that: In S1 of the specific step of parameter optimization, various characteristic variables, such as product output, number of machines and equipment, raw material consumption, energy consumption, and time series, need to be added to the sample points. These characteristics can be qualitative or quantitative, depending on the nature of the data and the requirements of the model.

6. The chemical production prediction and parameter optimization system based on GRU neural network and multi-objective Bayesian optimization as claimed in claim 2, characterized in that: The result display process includes the following steps: S1: Through the visualization module, the prediction results and optimization results are displayed to the operator, and an interface for adjusting production parameters is provided.

Images