Improved generative network-based double-hearth furnace optimization method

Abstract

The application discloses a kind of based on improved generative network's double hearth optimization method, including steps: S1, double hearth design parameter and result output data representation;S2, improved double hearth generative network dataset architecture construction;S3, improved double hearth output performance and environmental protection cost prediction network architecture construction;S4, based on improved generative network's double hearth optimization network architecture deployment and application.The based on improved generative network's double hearth optimization method of the application, the generative adversarial network used can learn and simulate these nonlinear relationships, providing more accurate and stable basis for parameter adjustment and optimization, and can more accurately predict and control the operating state of the double hearth in actual operation, improving production efficiency.

Description

Technical Field

[0001] This invention belongs to the field of double-hearth furnace technology, specifically relating to a double-hearth furnace optimization method based on an improved generative network. Background Technology

[0002] Double-hearth furnaces are widely used in lime production. Their working principle is mainly based on the high temperature of two main furnace chambers to calcine limestone and produce active lime. In this process, optimizing operating conditions and equipment parameters is key to increasing output and reducing energy consumption and emissions.

[0003] In existing technical solutions, the optimization of double-hearth furnaces mainly focuses on adjusting their internal physical parameters and operating conditions. These solutions optimize the lime production process by improving furnace design parameters, such as furnace dimensions, the length of the preheating and calcination zones, and operating conditions such as calcination temperature and internal airflow. While these traditional optimization methods have improved the activity of lime and reduced energy consumption to some extent, they often overlook some potential, non-linear influencing factors, resulting in limited optimization effects, hindering the improvement of double-hearth furnace operating efficiency, and increasing energy consumption.

[0004] On the other hand, with the development of data science and machine learning technologies, some researchers have begun to explore data-driven optimization methods for double-hearth furnaces. These methods focus on collecting and analyzing large amounts of actual operational data, using statistical and machine learning models to reveal factors affecting lime activity and environmental emissions, thereby optimizing the double-hearth furnace. The advantage of this approach is its ability to consider the combined effects of multiple factors, but it also faces challenges related to data quality and model complexity.

[0005] However, few existing technologies address the optimization of double-hearth furnaces using generative adversarial networks (GANs). Specifically, no literature describes methods for generating subnetworks with different adjustable parameters using generative networks and then further filtering the data to improve its reliability. Furthermore, while some existing technologies consider environmental factors, few solutions integrate environmental costs with output performance and combine complex network architectures such as BP and RNNs for comprehensive prediction and analysis.

[0006] Existing technologies for optimizing double-hearth furnaces have several significant drawbacks. First, traditional methods for adjusting physical parameters and optimizing operating conditions often focus only on linear and intuitive influencing factors, neglecting the complex nonlinear relationships that may exist between multiple parameters, resulting in unsatisfactory optimization results. For example, these methods typically ignore the interaction between external conditions such as ambient temperature and humidity and the gas flow and calcination temperature inside the furnace, causing the optimization results to be unstable under different environmental conditions.

[0007] Secondly, while some data-driven optimization methods can take into account the combined effects of multiple factors, these methods often face issues of data quality and model complexity. On the one hand, real-world operating data often contains noise and outliers, the quality of which directly affects the accuracy and reliability of the model. On the other hand, building complex machine learning models requires substantial computing resources and technical expertise, which to some extent limits the widespread adoption and application of these methods.

[0008] Furthermore, most existing optimization methods do not adequately consider environmental factors and lack a comprehensive analysis and trade-off between environmental costs and output performance. This leads to a situation where, in practical applications, the pursuit of output and efficiency often sacrifices environmental protection, creating difficulties for sustainable development. Summary of the Invention

[0009] This invention aims to at least solve one of the technical problems existing in the prior art. To this end, this invention provides a double-hearth furnace optimization method based on an improved generative network, with the goal of improving the operating efficiency of the double-hearth furnace.

[0010] To achieve the above objectives, the technical solution adopted by this invention is: a dual-hearth furnace optimization method based on improved generative networks, comprising the following steps:

[0011] S1. Characterization of dual-furnace design parameters and output data;

[0012] S2. Improved architecture for constructing generative network datasets using a dual-hearth furnace;

[0013] S3. Improved network architecture for predicting the output performance and environmental costs of double-hearth furnaces;

[0014] S4. Deployment and application of optimized network architecture for dual-hearth furnace based on improved generative networks.

[0015] In step S1, the operating data of the double-hearth furnace in actual production is collected, relevant data of the double-hearth furnace is obtained, and the collected data is preprocessed.

[0016] In step S1, the operating data of the double-hearth furnace to be collected includes the design parameters of the double-hearth furnace and the related output data, including the waste flue temperature, waste flue dust content, lime activity and residual carbon dioxide content.

[0017] Step S2 includes:

[0018] S201, FPG operator double-hearth furnace design parameter classification;

[0019] S202, generation architecture of each stage parameter of double-hearth furnace;

[0020] S203, Determination of generation parameters for the classification characteristics of double-hearth furnace.

[0021] In step S201, an FPG operator is proposed, and FPG is represented as follows: t1, t2, t3 = class(x i ), i=1, 2, 3,..., 9, θ j =Concat(t j x 10 x 11 x 12 x 13 ), j = 1, 2, 3; where x i Design features include furnace diameter, preheating zone length, calcination zone length, cooling zone length, calcination temperature, annular channel diameter, flue gas pipe diameter, fuel calorific value, and limestone size. These parameters correspond to x1, x2, x3, x4, x5, x6, x7, x8, and x9, respectively. i This indicates that design features are categorized according to their modifiability, where t1, t2, and t3 represent the corresponding values ​​for the three categories based on the modifiability dimension, and x... 10 x 11 x 12 x 13 To correspond to the target values ​​of the furnace design parameters, namely, flue gas temperature, flue gas dust content, lime activity, and residual carbon dioxide content, Concat(t) j x 10 x 11 x 12 x 13 ) represents the categorized design parameters, which are then concatenated with the corresponding design target values ​​of the double-hearth furnace to form an information vector, θ. j This is the generated vector.

[0022] In step S202, branch subnetworks are generated using GAN1, GAN2, and GAN3 for fixed features, modifiable but difficult features, and features that can be modified in real time, respectively. A cost-effective dual-chamber furnace subnetwork generation formula is proposed, which can be expressed as follows:

[0023]

[0024] Num(GAN2) = 2*Num(GAN3); where Num(GAN1), Num(GAN2), and Num(GAN3) represent the number of fixed features, modifiable but difficult features, and real-time modifiable features generated by GAN1, GAN2, and GAN3, respectively, and Cost(change(t1)), Cost(change(t2)), and Cost(change(t3)) represent the average economic cost incurred when modifying fixed features, modifiable but difficult features, and real-time modifiable features after the double-hearth furnace is built.

[0025] In step S203, a general generator network branch architecture is proposed. The general generator GAN4 is used to generate data. All data generated by the general generator is discarded, but its general decision-maker function is retained. The general decision-maker is used to judge the design vectors generated by the sub-generator networks in step S202. Only the design parameters with a confidence level of more than 0.9 are retained as dataset parameters that can be used for subsequent network analysis.

[0026] In step S3, two parallel networks are used to predict the output performance for each design parameter, including lime activity and residual carbon dioxide content, as well as environmental costs such as flue gas temperature and flue gas dust content. A BP network is used to predict the flue gas temperature and flue gas dust content.

[0027] In step S3, the network architecture can be specifically represented as a fully connected layer of 9*512, 512*1024, and 1024*2. The final output values ​​are the waste smoke temperature and the waste smoke dust content, which are simultaneously retained with the specific generated feature parameters. In addition, the loss value of this branch network architecture is the MSE loss.

[0028] The double-hearth furnace optimization method based on improved generative networks of the present invention uses a generative adversarial network that can learn and simulate these nonlinear relationships, providing a more accurate and stable basis for parameter adjustment and optimization. In actual operation, it can more accurately predict and control the operating status of the double-hearth furnace, thereby improving production efficiency. Attached Figure Description

[0029] This manual includes the following figures, which illustrate the following:

[0030] Figure 1 This is a flowchart of the double-hearth furnace optimization method based on improved generative networks of the present invention;

[0031] Figure 2 This is a network architecture diagram of the dual-hearth furnace optimization method based on improved generative networks of this invention. Detailed Implementation

[0032] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings, in order to help those skilled in the art to have a more complete, accurate and in-depth understanding of the concept and technical solutions of the present invention, and to facilitate its implementation.

[0033] like Figure 1 As shown, this invention provides a dual-hearth furnace optimization method based on an improved generative network, comprising the following steps:

[0034] S1. Characterization of dual-furnace design parameters and output data;

[0035] S2. Improved architecture for constructing generative network datasets using a dual-hearth furnace;

[0036] S3. Improved network architecture for predicting the output performance and environmental costs of double-hearth furnaces;

[0037] S4. Deployment and application of optimized network architecture for dual-hearth furnace based on improved generative networks.

[0038] Specifically, in step S1 above, this application first plans to collect operational data of the double-hearth furnace in actual production to obtain relevant double-hearth furnace data to ensure the stability and accuracy of model training. This data will cover double-hearth design parameters, including furnace diameter, preheating zone length, calcination zone length, cooling zone length, calcination temperature, annular channel diameter, flue gas duct diameter, fuel calorific value, and limestone block size. Related output data will also be recorded, including flue gas temperature, flue gas dust content, lime activity, and residual carbon dioxide content. Furthermore, a series of preprocessing steps will be implemented on the collected data, including missing value handling, outlier detection and removal, data normalization, and one-heat encoding of the annular channel diameter for categorical data. The dataset will also be divided: 70% for model training, 20% for model validation, and 10% for model testing, thereby ensuring the model's generalization performance. All feature labels will be encoded into a 1*13 vector.

[0039] like Figure 2 The diagram shown is a network architecture diagram of a dual-hearth furnace optimization method based on an improved generative network proposed in this application.

[0040] This application proposes an improved generative network architecture for dual-hearth furnace optimization, based on an improved generative network. The aim is to guide the generative network on the dispersion intensity of various design parameters by classifying the design parameters according to their modifiability.

[0041] Step S2 above includes:

[0042] S201, FPG operator double-hearth furnace design parameter classification;

[0043] S202, generation architecture of each stage parameter of double-hearth furnace;

[0044] S203, Determination of generation parameters for the classification characteristics of double-hearth furnace.

[0045] In step S201 above, an FPG operator is proposed, and FPG is represented as follows: t1, t2, t3 = class(x i ), i=1, 2, 3,..., 9, θ j =Concat(t j x 10 x 11 x 12 x 13 ), j = 1, 2, 3; where x i Design features include furnace diameter, preheating zone length, calcination zone length, cooling zone length, calcination temperature, annular channel diameter, flue gas pipe diameter, fuel calorific value, and limestone size. These parameters correspond to x1, x2, x3, x4, x5, x6, x7, x8, and x9, respectively. i This indicates that design features are categorized according to their modifiability, where t1, t2, and t3 represent the corresponding values ​​for the three categories based on the modifiability dimension, and x... 10 x 11 x 12 x 13 To correspond to the target values ​​of the furnace design parameters, namely, flue gas temperature, flue gas dust content, lime activity, and residual carbon dioxide content, Concat(t) j x 10 x 11 x 12 x 13 ) represents the categorized design parameters, which are then concatenated with the corresponding design target values ​​of the double-hearth furnace to form an information vector, θ. j This is the generated vector.

[0046] This application proposes for the first time an FPG operator that categorizes the modifiability of the dual-furnace design parameters after they are finalized into a specific actual device. Specifically, this is reflected in the eight design characteristic parameters in this application: furnace diameter, preheating zone length, calcination zone length, cooling zone length, calcination temperature, annular channel diameter, flue gas pipe diameter, fuel calorific value, and limestone size. Among these, the furnace diameter, preheating zone length, calcination zone length, and cooling zone length are almost impossible to modify after the design and construction are determined, as modifications would involve large-scale equipment upgrades. Therefore, these are classified as fixed parameter characteristics. The annular channel diameter and flue gas pipe diameter are modifiable but difficult characteristics. Subsequent modifications to these usually involve certain equipment and structural adjustments, but the economic cost is relatively small compared to modifying the kiln diameter, etc. The last category, which can be modified in real time, includes calcination temperature, fuel calorific value, and limestone size. For example, by selecting fuels with different calorific values ​​or adjusting the fuel ratio, operators can adjust the fuel calorific value in real time or relatively easily to adapt to different needs and conditions in the production process.

[0047] Through the FPG operator in step S201, this application generates three types of vectors with different modifiability. Compared to traditional GAN ​​networks, such as those that directly generate vectors based on all design parameters (furnace diameter, preheating zone length, calcination zone length, cooling zone length, calcination temperature, annular channel diameter, flue gas pipe diameter, fuel calorific value, and limestone block size), the sub-vectors classified by the FPA operator in this application have a smaller parameter space and computational complexity, thus enabling each sub-generator to have higher and more stable computational efficiency. In step S202, GAN1, GAN2, and GAN3 are used to generate branch sub-networks for fixed features, modifiable but difficult features, and real-time modifiable features, respectively. A cost-effective dual-furnace sub-network generation formula is proposed, which can be expressed as follows:

[0048]

[0049] Num(GAN2) = 2 * Num(GAN3);

[0050] Wherein, Num(GAN1), Num(GAN2), and Num(GAN3) represent the number of fixed features, modifiable but difficult features, and real-time modifiable features generated by GAN1, GAN2, and GAN3, respectively, and Cost(change(t1)), Cost(change(t2)), and Cost(change(t3)) represent the average economic cost incurred when modifying fixed features, modifiable but difficult features, and real-time modifiable features after the double-hearth furnace is built.

[0051] The above formula can be used to generate different numbers of different types of double-hearth furnace design parameters based on different data feature types. The number of fixed features, modifiable but difficult features, and features that can be changed in real time needs to meet the above formula.

[0052] The generated vectors are then processed by a determiner to filter out data deemed false by the sub-determiner. Finally, the dimensions of all generated fixed features, modifiable but difficult features, and real-time modifiable features from the sub-generator network are concatenated.

[0053] A specific example can be represented as follows: GAN1 generates 1000 fixed feature parameters for a double-hearth furnace, GAN2 generates 200 modifiable but difficult feature parameters for a double-hearth furnace, and GAN3 generates 80 real-time modifiable feature parameters for a double-hearth furnace. The 1000 fixed feature parameters for a double-hearth furnace are then sequentially arranged and combined with the parameters of GAN2 and GAN3 to form 1000*200*80 1*13-dimensional data features. The predicted values, namely the waste flue gas temperature, waste flue gas dust content, lime activity, and residual carbon dioxide content, can be obtained by simply averaging the parameters generated by the same GAN1, GAN2, and GAN3 networks.

[0054] In step S203 above, a new overall generator network validator is proposed for the first time to further filter the features generated by the sub-generators in step S202 to increase their credibility. Specifically, an overall generator network branch architecture is proposed, using the overall generator GAN4 to generate data. Unlike previous generator approaches, all data generated by the overall generator is discarded, that is, the 1*13 dimensional overall double-hearth furnace design parameters are directly discarded. Since the design parameters have a high dimension and poor stability, their overall decision-maker function is retained. The overall decision-maker is used to judge the spliced ​​design vectors generated by the sub-generator networks in the above steps, and only the design parameters with a confidence level of more than 0.9 are retained as dataset parameters that can be used for subsequent network analysis in this application.

[0055] Since the double-hearth furnace dataset is for a specific industrial device, the available operational data may be limited compared to general applications. Furthermore, the industrial processes involved in double-hearth furnaces involve core technologies and trade secrets, and the data collection process involves multiple parameters and processes, requiring data collection from multiple sources. Therefore, obtaining sufficient double-hearth furnace datasets is quite difficult. In step S2, this application proposes for the first time an improved generative network dataset architecture for double-hearth furnaces to generate sufficient design parameters. In step S3, the output values—waste flue gas temperature, waste flue gas dust content, lime activity, and residual carbon dioxide content—are determined based on this data.

[0056] In step S3 above, a network architecture for predicting the output performance and environmental costs of a double-hearth furnace is proposed. This involves using two parallel networks to predict the output performance, including lime activity and residual carbon dioxide content, as well as the environmental costs, such as flue gas temperature and flue gas dust content, for each design parameter. A BP network is used to predict the flue gas temperature and flue gas dust content. The network architecture can be specifically represented as a fully connected layer of 9*512, 512*1024, and 1024*2. The final output values ​​are the flue gas temperature and flue gas dust content, which are simultaneously retained along with the specific generated feature parameters. Furthermore, the loss value of this branch network architecture is the MSE loss.

[0057] Regarding the output performance, namely lime activity and residual carbon dioxide content, since the product output of the double-hearth furnace exhibits a strong time-series correlation with the entire process, this application uses an RNN network for prediction. The specific network architecture is a three-layer unidirectional recurrent RNN, BatchNormal, and ReLU. All output values ​​are labeled and retained, and the network loss value is also the mean squared error (MSE) loss value.

[0058] Once all the data values ​​of the generative networks have been determined, that is, once all datasets have been trained, the loss values ​​of the two score prediction networks have decreased to below 0.05, indicating that the dual-hearth furnace optimization network architecture based on the improved generative network proposed in this application has been completed.

[0059] In step S4 above, the network architecture was constructed. This application then deploys and applies it. In practical application, the length of the preheating zone, the length of the calcination zone, the length of the cooling zone, the calcination temperature, the fuel calorific value, and the limestone size are relatively easy to design and relatively easy to limit according to production requirements and industrial environment. However, the diameter of the furnace, the diameter of the annular channel, and the diameter of the flue gas pipe are indeed difficult to select. This is mainly because when the length of the preheating zone, the length of the calcination zone, the length of the cooling zone, the calcination temperature, the fuel calorific value, and the limestone size change, the fuel flow rate changes, the exhaust gas volume also changes, and the airflow velocity within the original dimensions also changes drastically, leading to changes in production conditions. This makes it difficult to produce qualified active lime in a standard kiln. Therefore, in this application, after determining the preheating zone length, calcination zone length, cooling zone length, calcination temperature, fuel calorific value, and limestone block size characteristic parameters, a large amount of data on furnace diameter, annular channel diameter, and flue gas pipe diameter are first generated using GAN1 and GAN2. This data is then combined with the previously determined preheating zone length, calcination zone length, cooling zone length, calcination temperature, fuel calorific value, and limestone block size characteristic parameters to form the complete design parameters. For each set of generated parameters, they are input into a pre-trained dual-hearth furnace network to predict the corresponding environmental costs (waste flue gas temperature, waste flue gas dust content) and product performance (lime activity and residual carbon dioxide content). The dual-hearth furnace optimization objective function is then used to analyze and balance the output performance and environmental costs.

[0060] The objective function can be expressed as:

[0061] F(x)=δ*P(x)+ε*E(x) (1)

[0062]

[0063]

[0064] Where F(x) is the overall objective function, x is the final generated design parameter, P(x) is the output performance evaluation function, E(x) is the environmental cost evaluation function, δ is the product performance factor, ε is the environmental cost factor, and A t and C t These are the target values ​​for lime activity and residual carbon dioxide content, C t C p(x) These are the lime activity and residual carbon dioxide content predicted by the model, T a D a T represents the maximum allowable temperature and dust content of the waste smoke. p(x) D p(x) These are the waste smoke temperature and waste smoke dust content predicted by the model. These are the relative percentage parameters of lime activity, residual carbon dioxide content, waste smoke temperature, and waste smoke dust content.

[0065] Once all parameters have been predicted by the aforementioned network, the minimum overall objective function can be selected to complete the optimal parameter design for the double-hearth furnace under the existing environment. This means that given a subset of parameters, the optimal design parameters can be generated and selected to achieve a balance between output performance and environmental costs, thereby enabling the double-hearth furnace to operate efficiently, environmentally friendly, and economically, bringing tangible benefits to the enterprise.

[0066] The technical solution of this application, by combining generative adversarial networks, complex network architectures, and multi-objective optimization strategies, provides an innovative solution to several problems existing in the prior art, and has the following advantages:

[0067] (1) Handling Nonlinear Relationships: Existing technologies often neglect the complex nonlinear relationships between parameters, focusing only on intuitive factors. In contrast, the generative adversarial network used in this application can learn and simulate these nonlinear relationships, providing a more accurate and stable basis for parameter adjustment and optimization. This means that in actual operation, this technical solution can more accurately predict and control the operating status of the double-hearth furnace, improving production efficiency;

[0068] (2) Solving data quality problems: This application employs a complex network architecture to handle noise and outliers in actual operating data, thereby ensuring the accuracy and reliability of the model. This solves the data quality problems often faced by data-driven methods in the prior art, making this optimization method more practical and reliable;

[0069] (3) Reduced model complexity: Existing data-driven optimization methods require a large amount of computational resources and expertise, resulting in high model complexity. This application reduces model complexity and computational resource requirements by adopting innovative network structures and algorithms, making the method easier to promote and apply;

[0070] (4) Achieving a balance between environmental protection and output: Existing methods often sacrifice environmental protection in pursuit of output and efficiency, making sustainable development difficult. This application, through a multi-objective optimization strategy, fully considers environmental factors and achieves a balance between environmental protection and output. This will help improve the sustainable development capability of the double-hearth furnace, while also representing a trade-off between environmental protection and corporate interests;

[0071] (5) Improved stability and reliability of optimization: By comprehensively analyzing and exploring the nonlinear relationships between multiple influencing factors, the technical solution of this application improves the stability and reliability of optimization. This means that the solution can achieve stable and reliable optimization results under different environmental conditions, meeting the needs of actual production.

[0072] The above-mentioned optimization method for a double-hearth furnace based on improved generative networks has the following characteristics:

[0073] 1. Hierarchical generation and filtering;

[0074] FPG operator: By using the FPG operator, the scheme achieves classification and hierarchical generation of feature parameters with different modifiability, reducing parameter space and computational complexity, and improving computational efficiency and model stability.

[0075] The combination of sub-generators and the overall generator: the sub-generators focus on generating feature parameters for specific categories, while the overall generator performs comprehensive judgment and filtering to ensure the credibility of the generated data.

[0076] 2. Multi-network architecture and parallel prediction;

[0077] Parallel Prediction Network: The scheme uses BP network and RNN network to predict different categories of output values ​​(environmental cost and output performance) in parallel, which meets the needs of both time-series and non-time-series data and improves prediction accuracy.

[0078] Multi-level network validation: By combining sub-generators, a total generator, and multiple prediction networks, the scheme achieves multi-level network validation and data filtering, improving data credibility and model robustness.

[0079] 3. Real-time adjustments and optimizations;

[0080] Real-time feature adjustment: By distinguishing between fixed features, modifiable but difficult features, and features that can be changed in real time, the solution can adjust feature parameters in real time according to production needs and conditions, enhancing the practicality and flexibility of the model.

[0081] The objective function for optimizing the double-hearth furnace is to analyze and balance output performance and environmental costs, thereby achieving real-time optimization of the double-hearth furnace and improving output quality and environmental friendliness.

[0082] 4. Large-scale data generation and application;

[0083] Large-scale data generation: GANs generate a large amount of design parameter data, which enriches the dataset and improves the stability and accuracy of model training.

[0084] Practical application and verification: The solution is not limited to the theoretical and model building level, but is further deployed and applied to the actual double-hearth furnace production, which verifies the practicality and effectiveness of the model.

[0085] 5. Comprehensive feature analysis and coding;

[0086] Feature analysis and coding: By conducting detailed analysis and coding of various characteristic parameters of the double-hearth furnace, the scheme achieves effective description and modeling of complex systems.

[0087] Dimensional splicing and combination: The feature parameters generated by different generators are spliced ​​and combined in dimensions, realizing the comprehensive analysis and application of multi-dimensional features.

[0088] In summary, this solution optimizes and improves the performance of the double-hearth furnace through a series of innovative technologies and methods, demonstrating high practical value and innovation.

[0089] The present invention has been described above by way of example with reference to the accompanying drawings. Obviously, the specific implementation of the present invention is not limited to the above-described manner. Any non-substantial improvements made using the inventive concept and technical solution; or the direct application of the inventive concept and technical solution to other situations without modification, are all within the protection scope of the present invention.

Claims

1. A method for optimizing a double-hearth furnace based on an improved generative network, characterized in that, include: Step S1: Characterization of dual-furnace design parameters and output data; Step S2: Improve the architecture construction of the generative network dataset for the dual-hearth furnace; Step S3: Improve the network architecture for predicting the output performance and environmental costs of double-hearth furnaces; Step S4: Deployment and application of optimized network architecture for dual-hearth furnace based on improved generative network; Step S2 includes: Step S201: FPG operator double-hearth furnace design parameter classification; Step S202: Generation architecture of stage parameters for the double-hearth furnace; Step S203: Determine the generation parameters for the classification characteristics of the double-hearth furnace; In step S201, an FPG operator is proposed, and FPG is represented as follows: , ;in, Design features include furnace diameter, preheating zone length, calcination zone length, cooling zone length, calcination temperature, annular channel diameter, flue gas duct diameter, fuel calorific value, and limestone size. These parameters correspond to... ; This indicates that design features are categorized according to their modifiability. This indicates that the values ​​are divided into three categories based on the modifiability dimension. To correspond to the target values ​​of the furnace design parameters, namely, flue gas temperature, flue gas dust content, lime activity, and residual carbon dioxide content, The categorized design parameters are concatenated with the corresponding design target values ​​of the double-hearth furnace to form an information vector. To generate vectors; In step S202, through , and Branch subnetworks are generated for fixed features, modifiable but difficult features, and features that can be changed in real time, respectively. A cost-based formula for generating subnetworks for a double-hearth furnace is proposed, which can be expressed as follows: , ;in, , , They represent passing through , and The number of features generated is calculated for fixed features, modifiable but difficult features, and features that can be changed in real time. This represents the average economic cost incurred when modifying fixed features, modifiable but difficult features, and features that can be modified in real time after the double-hearth furnace has been built. In step S203, a general generator network branch architecture is proposed, utilizing the general generator. Data generation is performed, and all data generated by the overall generator is discarded, while its overall decision-maker function is retained. The overall decision-maker is used to determine the design vectors generated and spliced ​​by the sub-generator network in step S202, and only the design parameters with a confidence level of more than 0.9 are retained as dataset parameters that can be used for subsequent network analysis.

2. The dual-hearth furnace optimization method based on improved generative networks according to claim 1, characterized in that, In step S1, the operating data of the double-hearth furnace in actual production is collected, relevant data of the double-hearth furnace is obtained, and the collected data is preprocessed.

3. The dual-hearth furnace optimization method based on improved generative networks according to claim 2, characterized in that, In step S1, the operating data of the double-hearth furnace to be collected includes the design parameters of the double-hearth furnace and the related output data, including the waste flue temperature, waste flue dust content, lime activity and residual carbon dioxide content.

4. The dual-hearth furnace optimization method based on improved generative networks according to any one of claims 1 to 3, characterized in that, In step S3, two parallel networks are used to predict the output performance for each design parameter, including lime activity and residual carbon dioxide content, as well as environmental costs such as flue gas temperature and flue gas dust content. A BP network is used to predict the flue gas temperature and flue gas dust content.

Images