CN117235599A - Deep learning-based smelting furnace temperature control optimization method - Google Patents
Deep learning-based smelting furnace temperature control optimization method Download PDFInfo
- Publication number
- CN117235599A CN117235599A CN202311210884.9A CN202311210884A CN117235599A CN 117235599 A CN117235599 A CN 117235599A CN 202311210884 A CN202311210884 A CN 202311210884A CN 117235599 A CN117235599 A CN 117235599A
- Authority
- CN
- China
- Prior art keywords
- workpiece
- model
- data
- size
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 66
- 238000005457 optimization Methods 0.000 title claims abstract description 34
- 238000013135 deep learning Methods 0.000 title claims abstract description 27
- 238000003723 Smelting Methods 0.000 title abstract description 4
- 238000012549 training Methods 0.000 claims abstract description 31
- 230000006870 function Effects 0.000 claims abstract description 25
- 238000010438 heat treatment Methods 0.000 claims abstract description 22
- 230000007246 mechanism Effects 0.000 claims abstract description 18
- 238000004088 simulation Methods 0.000 claims abstract description 18
- 238000013528 artificial neural network Methods 0.000 claims abstract description 12
- 238000007781 pre-processing Methods 0.000 claims abstract description 12
- 230000009466 transformation Effects 0.000 claims abstract description 7
- 230000008569 process Effects 0.000 claims description 29
- 230000003746 surface roughness Effects 0.000 claims description 18
- 238000004422 calculation algorithm Methods 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000000694 effects Effects 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 7
- 230000000630 rising effect Effects 0.000 claims description 7
- 125000004122 cyclic group Chemical group 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 5
- 238000010792 warming Methods 0.000 claims description 5
- 238000004140 cleaning Methods 0.000 claims description 4
- 238000009499 grossing Methods 0.000 claims description 4
- 238000010801 machine learning Methods 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 abstract description 7
- 230000000306 recurrent effect Effects 0.000 abstract description 4
- 229910001220 stainless steel Inorganic materials 0.000 abstract description 4
- 239000010935 stainless steel Substances 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 5
- 238000004519 manufacturing process Methods 0.000 description 5
- 239000000463 material Substances 0.000 description 5
- 230000006872 improvement Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000007405 data analysis Methods 0.000 description 2
- 238000013480 data collection Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 230000003647 oxidation Effects 0.000 description 2
- 238000007254 oxidation reaction Methods 0.000 description 2
- 230000008646 thermal stress Effects 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000001816 cooling Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000002844 melting Methods 0.000 description 1
- 230000008018 melting Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000004886 process control Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Landscapes
- Feedback Control In General (AREA)
Abstract
The application provides a furnace temperature control optimization method based on deep learning, which comprises the following three main steps: firstly, data related to the heating process of workpieces with different sizes are collected by constructing a simulation model and setting parameters. Secondly, the data preprocessing is performed by adopting the generation countermeasure network, and the discrete wavelet transformation is used for extracting the characteristics. Each data sequence is assigned a corresponding workpiece size label. Finally, training is performed using a model of the attention mechanism based on a recurrent neural network. In order to measure the difference between the generated data and the actual data, the application defines a cost function comprising the size of the workpiece and the influence of the heating process. In addition, a workpiece roughness regularization term was introduced to prevent model overfitting. The application can automatically identify and adapt to the specific requirements of workpieces with different sizes in the heating process, and is expected to obviously improve the efficiency and performance of a smelting furnace for processing high-strength non-magnetic stainless steel.
Description
Technical Field
The application relates to the field of big data, temperature sensors and deep learning, in particular to a furnace temperature control optimization method based on deep learning.
Background
In the manufacturing industry, the temperature rising process of a melting furnace for high-strength non-magnetic stainless steel workpieces is always a key link of process control. However, conventional optimization techniques have many drawbacks in handling this link. These techniques typically rely on empirical formulas, manual parameter settings, or simple numerical simulation methods, thereby facing limitations and challenges in many respects. First, processing workpieces of different sizes and shapes presents a significant challenge. Different heating curves and times are required for workpieces of different sizes, and conventional methods often cannot flexibly accommodate these changes, resulting in frequent manual adjustments and monitoring when processing irregular or small batches of workpieces. This not only increases the complexity of the operation, but may also lead to operational errors and inconsistencies. Second, most of the prior art does not consider energy efficiency, which is a non-negligible problem in the case of the current rise in energy costs and the enhancement of environmental protection consciousness. Inaccurate temperature control not only affects the quality of the workpiece, but also may result in a significant waste of energy, which further increases manufacturing costs. Third, conventional models often lack precision and adaptability. Even some advanced numerical simulation methods cannot accurately simulate complex physical processes and chemical reactions due to limitations in computational power or simplification of model assumptions. This means that the model output is likely to deviate greatly from the actual situation, thereby increasing production risk. Fourth, data preprocessing becomes another bottleneck due to the different data quality and dimension. Traditional data cleaning and normalization methods tend to be too simple to process multi-source heterogeneous data from different sensors and devices, which directly affects subsequent data analysis and model training. Fifth, many conventional machine learning and data analysis methods have presented considerable limitations in this area. For example, they may perform poorly in the face of uneven data distribution or high-dimensional feature space, which limits the applicability of these methods in complex manufacturing environments. Because of all of the above problems, there is an increasing need for a new approach that is more comprehensive, efficient, and adaptive. Such new methods should not only be able to more accurately simulate and control the temperature rise process, but also should be able to automatically accommodate workpieces of different sizes and types, as well as be able to better process and analyze large amounts of real-time data.
Disclosure of Invention
In order to solve the problems, the application discloses a furnace temperature control optimization method based on deep learning, which can heat according to different workpiece sizes.
In order to achieve the above purpose, the technical scheme adopted by the application is as follows:
the furnace temperature control optimization method based on deep learning comprises the following steps:
1) Collecting data of workpieces with different sizes;
the data set is ensured to contain relevant information of the size of the workpiece and the temperature rising process. The application can acquire data containing the heating process of workpieces with different sizes through a simulation method. Firstly, a simulation model is built, secondly, parameter setting, a simulation process and finally data generation.
2) Data preprocessing and feature extraction;
the data preprocessing step comprises data cleaning, data smoothing, normalization and the like, so that the consistency and quality of the processed data are ensured. The present application uses a GAN for data preprocessing, where the GAN includes two parts: a generator and a arbiter. The generator is configured to learn a mapping that maps random noise into a time series, and the arbiter is configured to learn a classifier that distinguishes between the generated data and the actual data. The purpose of feature extraction is to convert raw data into a more characterizable and interpretable feature representation so that the machine learning model can better understand and utilize the data. The application adopts discrete wavelet transformation to carry out wavelet transformation on the sequence to obtain the approximate coefficients and the detail coefficients with different sizes. And finally, assigning corresponding workpiece size labels for each sequence, and coding the workpiece size labels into integers according to the types of the workpiece sizes by adopting a classification label mode.
A cost function is defined in this application to measure the difference between the generated data and the real data. This cost function should take into account the effects of workpiece size and the warming process in order to better capture these features. Secondly, defining a workpiece roughness regularization term formula to punish the overfitting phenomenon in the model. To minimize this regularization term, a random gradient descent algorithm (SGD) is used to update the weight parameters.
3) Training a model;
the attention mechanism model training is one of the core steps of a furnace temperature control optimization method based on deep learning. This step aims at training a model which can automatically identify and detect the size of the workpiece to adapt to different heating processes. The application uses a cyclic neural network as a basis and introduces an attention mechanism on the cyclic neural network;
wherein step 3) model training comprises the sub-steps of:
constructing a model structure;
selecting a loss function and an optimization algorithm;
training a model;
model evaluation adjustment;
model retraining and application;
4) Improving the construction of an attention mechanism model network;
through the improved attention mechanism model network, the difference of the heating processes of the workpieces with different sizes can be better calculated. The application is based on a circulating neural network to introduce an attention mechanism, and the learning process of attention weights and the combination of loss functions are required to be optimized, so that the optimization thought of the application is to define an improved loss function, namely, a workpiece rough loss calculation formula is adopted.
As a further improvement of the present application, the workpiece roughness regularization term in step 2) is formulated as:
wherein, the workpiece roughness regularization term formula is expressed as:
wherein lambda is i Is the L2 regularization coefficient of the ith weight parameter, θ i Is the i-th weight parameter and,is the average of all weight parameters, delta is the surface roughness parameter of the workpiece. For parameters in the cost function, proper weights and bias terms are selected according to the size of the workpiece and the influence of surface roughness in the heating process. The workpiece size is assigned a greater weight in this application to better capture the impact of this feature on the generated data. The surface roughness may also be weighted more heavily to better capture the effect of this feature on the generated data.
As a further improvement of the present application, the random gradient descent algorithm (SGD) in step 2) is expressed as:
wherein the random gradient descent algorithm (SGD) is expressed as follows:
where α is the step size, N is the number of workpiece samples, f (θ t ;x i ) Is sample x i The loss function value, η, is the learning rate. The learning rate in the super-parameters is optimized according to the size and complexity of the data set. Generally, a smaller learning rate may allow the model to converge more slowly, but may achieve better generalization performance; a larger learning rate may allow the model to converge faster, but may result in the occurrence of overfitting. Likewise, a smaller number of iterations may make the model more stable, but require a longer training time; a larger number of iterations may make the training process faster, but may result in model overfitting.
As a further improvement of the present application, the loss calculation formula in the step 4) is expressed as:
wherein, the calculation formula of the rough loss of the workpiece is expressed as follows:
where y is the actual output, and where,is the model predictive output, N is the number of samples of the workpiece, α is the attention weight, and p is the workpiece surface roughness parameter. For samples with different workpiece sizes, the surface roughness of the workpiece is different, and different characteristics are also provided in the heating process. For samples of different workpiece sizes, the square of the residual helps us predict the magnitude of the prediction error for each sample, thereby better evaluating the performance of the model. By multiplying each residual square by a corresponding weight factor, the model can be better optimized to better accommodate data of different workpiece sizes. For important workpiece sizes, the weight of the important workpiece can be increased to improve the fitting capability of the model to the important workpiece; for other workpiece sizes, their weights may be reduced to reduce their impact on the overall error of the model.
The application has the following benefits:
1. the furnace temperature control optimization method based on deep learning provided by the application can quickly raise the temperature to the required temperature, and effectively reduce the thermal stress and oxidation damage of materials, thereby improving the processing efficiency and the product quality.
2. The furnace temperature control optimization method based on deep learning provided by the application adopts the characteristics of an advanced heating system, a temperature control system, an atmosphere control system, rapid heating and cooling and the like, and can reduce the processing time, the production cost and the energy consumption.
3. The furnace temperature control optimization method based on deep learning can effectively reduce thermal stress and deformation risk of materials, so that the utilization rate of the materials is improved, meanwhile, oxidation and surface damage of high-strength nonmagnetic stainless steel in the heating process can be prevented, and the environmental protection performance in the processing process is improved.
4. The furnace temperature control optimization method based on deep learning is suitable for processing various materials such as high-strength nonmagnetic stainless steel, has wide application prospect and market potential, and can meet the requirements of the industrial field on high-quality materials.
Drawings
FIG. 1 is a system flow diagram of a furnace temperature control optimization method based on deep learning provided in accordance with an embodiment of the present application;
FIG. 2 is a diagram for embodying a method for optimizing temperature control of a furnace based on deep learning according to an embodiment of the present application;
fig. 3 is an optimized architecture diagram of a deep learning-based furnace temperature control optimization method according to an embodiment of the present application.
Detailed Description
The present application is further illustrated in the following drawings and detailed description, which are to be understood as being merely illustrative of the application and not limiting the scope of the application. It should be noted that the words "front", "rear", "left", "right", "upper" and "lower" used in the following description refer to directions in the drawings, and the words "inner" and "outer" refer to directions toward or away from, respectively, the geometric center of a particular component.
FIG. 1 is a system flow chart of the furnace temperature control optimization method based on deep learning.
Step S1: data collection of workpieces of different sizes.
In the step, data collection of the temperature rising process of different workpiece sizes is one of core steps of a deep learning-based smelting furnace temperature control optimization method, and the fact that the data set contains relevant information of the workpiece sizes and the temperature rising process is ensured. The application can acquire data containing the heating process of workpieces with different sizes through a simulation method. Firstly, constructing a simulation model: a physical model is used to simulate the heating process of the workpiece. The model is built based on known physical principles and parameters and takes into account the effect of workpiece size on the warming process. And the following parameters are set: according to actual conditions, setting parameters such as initial temperature, ambient temperature, heating rate and the like in the simulation process. Different parameter values are set according to different workpiece sizes so as to simulate different conditions. The simulation process follows: and carrying out iterative computation on the simulation model by a numerical simulation method to obtain temperature values of the workpiece at different time points. A numerical solution method was used to simulate the warming process. And finally, data generation: and generating a large amount of workpiece temperature rising process data by using the simulation model, recording the temperature value at each time point, and generating a plurality of groups of data of workpieces with different sizes so as to expand the diversity of the data set. In the simulation, the model parameters and boundary conditions are adjusted according to the actual situation so as to ensure that the simulation result accords with the actual situation.
Step S2: data preprocessing and feature extraction.
The data preprocessing and the feature extraction in the step are key links of a furnace temperature control optimization method based on deep learning. The collected data is preprocessed to fit the training of the model. The preprocessing steps comprise data cleaning, data smoothing, normalization and the like, and the consistency and quality of the processed data are ensured. The present application uses a GAN for data preprocessing, where the GAN includes two parts: a generator and a arbiter. The generator is configured to learn a mapping that maps random noise into a time series, and the arbiter is configured to learn a classifier that distinguishes between the generated data and the actual data. The data cleansing operation is used to check whether there are missing values, outliers or noise in the data. Incomplete or unreliable data samples are removed, and the smooth position of the quality initial data of the data set is ensured. Data smoothing techniques enable the principle of pairing to reduce noise or outliers due to bad measurements. Data normalization is to make it have a similar numerical range, which is beneficial for better model learning and convergence. And then serializing the data, expressing the temperature rise process of the workpiece as time series data, ensuring the consistency of the length of each sequence, and intercepting a data window with continuous time steps by using a sliding window mode as one sequence. The purpose of feature extraction is to convert raw data into a more characterizable and interpretable feature representation so that the machine learning model can better understand and utilize the data. The application adopts discrete wavelet transformation to carry out wavelet transformation on the sequence to obtain the approximate coefficients and the detail coefficients with different sizes. And finally, assigning corresponding workpiece size labels for each sequence, and coding the workpiece size labels into integers according to the types of the workpiece sizes by adopting a classification label mode.
The characteristic consideration of the workpiece in the application comprises the size, weight, effective surface area, surface roughness and thickness similarity of the workpiece. Wherein the four types of size, weight, effective surface area and surface roughness are directly used as characteristic values, and a cost function is defined in the application to measure the difference between the generated data and the real data. This cost function should take into account the variations in workpiece size, surface roughness during the warming process to better capture these features. Second, a regularization term is defined to penalize the overfitting phenomenon in the model. In this problem, L2 regularization is used as a regularization term because it can force the weight parameters of the model to remain small, thereby reducing the likelihood of overfitting.
Wherein, when L2 regularization is used, the workpiece roughness regularization term can be expressed as:
wherein lambda is i Is the L2 regularization coefficient of the ith weight parameter, θ i Is the i-th weight parameter and,is the average of all weight parameters, delta is the surface roughness parameter of the workpiece. For parameters in the cost function, proper weights and bias terms are selected according to the size of the workpiece and the influence of surface roughness in the heating process. The workpiece size is assigned a greater weight in this application to better capture the impact of this feature on the generated data. The surface roughness may also be weighted more heavily to better capture the effect of this feature on the generated data. In order to minimize this regularization term,the weight parameters are updated using a random gradient descent algorithm (SGD).
Wherein, the formula of the random gradient descent algorithm is as follows:
where α is the step size, N is the number of workpiece samples, f (θ t ;x i ) Is sample x i The loss function value, η, is the learning rate. The learning rate in the super-parameters is optimized according to the size and complexity of the data set. Generally, a smaller learning rate may allow the model to converge more slowly, but may achieve better generalization performance; a larger learning rate may allow the model to converge faster, but may result in the occurrence of overfitting. Likewise, a smaller number of iterations may make the model more stable, but require a longer training time; a larger number of iterations may make the training process faster, but may result in model overfitting.
Step S3: and (5) model training.
Fig. 2 is a diagram showing a specific implementation of the furnace temperature control optimization method based on deep learning.
The attention mechanism model training is one of the core steps of a furnace temperature control optimization method based on deep learning. This step aims at training a model which can automatically identify and detect the size of the workpiece to adapt to different heating processes. This application uses a recurrent neural network as a basis and introduces an attention mechanism thereon.
Wherein step S3 of attention mechanism model training comprises the sub-steps of:
and A1, constructing a model structure.
And designing the structure of the attention mechanism model according to the problem requirement. First, a basic recurrent neural network is constructed, where LSTM (long short time memory network) is used to define the input layer, hidden layer and output layer structures of the recurrent neural network. The input layer is a sequence feature, the hidden layer is the hidden state of the cyclic neural network, and the output layer is the target of the prediction task. Attention mechanisms are introduced on this basis to enhance the importance of the model to different time steps.
Step A2: loss function and optimization algorithm selection.
The loss function is the degree of difference or error between the predicted and actual results of the metrology model. The model is an important component in a deep learning task, and aims to measure the performance of the model, guide parameter optimization and adjust the complexity of the model, so that the generalization capability and the prediction accuracy of the model are improved. The current application selection is based on a mean square error loss function. And simultaneously, a random gradient descent optimization algorithm is selected to update the weight and bias of the model, so that the purpose of minimizing the loss function is achieved.
Step A3: and (5) model training.
The dataset was divided into training, validation and test sets, with the proportion of 80% training set for training the model, 10% validation set for adjusting the hyper-parameters, and the remaining 10% for testing the performance of the model. Inputting training data into the model, and iterating the training process by looking at the forward propagation and direction propagation update parameters, and iterating the whole training data set for a plurality of times by using the Epoch until the model is proficient or a preset training round number is reached
Step A4: model evaluation adjustment.
And evaluating the performance of the model by using the verification set, and calculating an evaluation index to measure the quasi-determination and effect of the model. Based on the evaluation result, the problems and bottlenecks of the model are analyzed, and further, adjustment is performed according to the performance of the model to improve the performance.
Step A5: model retraining and application.
And retraining the model according to the adjusted parameters, performing model evaluation by using a verification set and a test set, comparing with the previous version, and selecting the model with the best performance as a final model. And finally, applying the final model to new data for prediction.
Step S4: improving the attention mechanism model network construction.
Fig. 3 shows an optimization architecture diagram of the furnace temperature control optimization method based on deep learning.
The application is based on a circulating neural network to introduce an attention mechanism, and the learning process of attention weights and the combination of loss functions are required to be optimized, so that the optimization thought of the application is to define an improved loss function, namely, a workpiece rough loss calculation formula is adopted.
The specific calculation formula of the workpiece roughness loss is as follows:
where y is the actual output, and where,is the model predictive output, N is the number of samples of the workpiece, α is the attention weight, and δ is the workpiece surface roughness parameter. For samples with different workpiece sizes, the surface roughness of the workpiece is different, and different characteristics are also provided in the heating process. For samples of different workpiece sizes, the square of the residual helps us predict the magnitude of the prediction error for each sample, thereby better evaluating the performance of the model. By multiplying each residual square by a corresponding weight factor, the model can be better optimized to better accommodate data of different workpiece sizes. For important workpiece sizes, the weight of the important workpiece can be increased to improve the fitting capability of the model to the important workpiece; for other workpiece sizes, their weights may be reduced to reduce their impact on the overall error of the model.
The technical means disclosed by the scheme of the application is not limited to the technical means disclosed by the embodiment, and also comprises the technical scheme formed by any combination of the technical features.
Claims (4)
1. The furnace temperature control optimization method based on deep learning is characterized by comprising the following steps of:
1) Collecting data of workpieces with different sizes;
the data set is ensured to contain relevant information of the size of the workpiece and the temperature rising process. The application can acquire data containing the heating process of workpieces with different sizes through a simulation method. Firstly, a simulation model is built, secondly, parameter setting, a simulation process and finally data generation.
2) Data preprocessing and feature extraction;
the data preprocessing step comprises data cleaning, data smoothing, normalization and the like, so that the consistency and quality of the processed data are ensured;
3) Training a model;
the attention mechanism model training is one of the core steps of a furnace temperature control optimization method based on deep learning; this step aims at training a model which can automatically identify and detect the size of the workpiece to adapt to different heating processes. The application uses a cyclic neural network as a basis and introduces an attention mechanism on the cyclic neural network;
wherein step 3) model training comprises the sub-steps of:
constructing a model structure;
selecting a loss function and an optimization algorithm;
training a model;
model evaluation adjustment;
model retraining and application;
4) Improving the construction of an attention mechanism model network;
the temperature rising process of workpieces with different sizes can be better calculated through an improved attention mechanism model network; based on the attention mechanism introduced by the cyclic neural network, the combination of the learning process of attention weight and the loss function needs to be optimized, so that the optimization thinking is to define an improved loss function, namely, a workpiece rough loss calculation formula is adopted.
2. The deep learning based furnace temperature control optimization method according to claim 1, wherein:
the step 2: data preprocessing was performed using a GAN comprising two parts: a generator and a arbiter; the generator is used for learning a mapping which maps random noise into a time sequence, and the discriminator is used for learning a classifier which can distinguish generated data from real data; the purpose of feature extraction is to convert raw data into a more characterizable and interpretable feature representation so that the machine learning model can better understand and utilize the data; performing wavelet transformation on the sequence by adopting discrete wavelet transformation to obtain approximation coefficients and detail coefficients with different sizes; finally, corresponding workpiece size labels are allocated to each sequence, and the workpiece size labels are encoded into integers according to the types of the workpiece sizes in a classification label mode; a cost function is defined to measure the difference between the generated data and the real data. This cost function should take into account the effects of workpiece size and the warming process in order to better capture these features. Secondly, defining a workpiece roughness regularization term formula to punish the overfitting phenomenon in the model; to minimize this regularization term, a random gradient descent algorithm (SGD) is used to update the weight parameters.
3. The deep learning based furnace temperature control optimization method according to claim 2, characterized in that:
the workpiece roughness regularization term in the step 2) is expressed as follows:
wherein, the workpiece roughness regularization term formula is expressed as:
wherein lambda is i Is the L2 regularization coefficient of the ith weight parameter, θ i Is the i-th weight parameter and,is the average value of all weight parameters, delta is the surface roughness parameter of the workpiece; for parameters in the cost function, proper weights and bias terms are selected according to the size of the workpiece and the influence of surface roughness in the heating process. In the application, larger weight is distributed for the size of the workpiece so as to better capture the influence of the feature on the generated data; the surface roughness may also be given a greater weight to better capture the effect of this feature on the generated data;
the random gradient descent algorithm (SGD) in step 2) is expressed as:
wherein the random gradient descent algorithm (SGD) is expressed as follows:
where α is the step size, N is the number of workpiece samples, f (θ t ;x i ) Is sample x i The loss function value, η, is the learning rate. The learning rate in the super-parameters is optimized according to the size and complexity of the data set. Generally, a smaller learning rate may allow the model to converge more slowly, but may achieve better generalization performance; a larger learning rate may allow the model to converge faster, but may result in the occurrence of overfitting. Likewise, a smaller number of iterations may make the model more stable, but require a longer training time; a larger number of iterations may make the training process faster, but may result in model overfitting.
4. The deep learning based furnace temperature control optimization method according to claim 1, wherein:
the loss calculation formula in the step 4) is expressed as follows:
wherein, the calculation formula of the rough loss of the workpiece is expressed as follows:
where y is the actual output, and where,is the model predictive output, N is the number of samples of the workpiece, α is the attention weight, and p is the workpiece surface roughness parameter. For samples with different workpiece sizes, the surface roughness of the workpiece is different, and different characteristics are also provided in the heating process. Residual squaring helps for samples of different workpiece sizesWe predict the magnitude of the prediction error for each sample, thereby better evaluating the performance of the model. By multiplying each residual square by a corresponding weight factor, the model can be better optimized to better accommodate data of different workpiece sizes. For important workpiece sizes, the weight of the important workpiece can be increased to improve the fitting capability of the model to the important workpiece; for other workpiece sizes, their weights may be reduced to reduce their impact on the overall error of the model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311210884.9A CN117235599A (en) | 2023-09-20 | 2023-09-20 | Deep learning-based smelting furnace temperature control optimization method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311210884.9A CN117235599A (en) | 2023-09-20 | 2023-09-20 | Deep learning-based smelting furnace temperature control optimization method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117235599A true CN117235599A (en) | 2023-12-15 |
Family
ID=89097997
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311210884.9A Pending CN117235599A (en) | 2023-09-20 | 2023-09-20 | Deep learning-based smelting furnace temperature control optimization method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117235599A (en) |
-
2023
- 2023-09-20 CN CN202311210884.9A patent/CN117235599A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Labach et al. | Survey of dropout methods for deep neural networks | |
CN109060001B (en) | Multi-working-condition process soft measurement modeling method based on feature transfer learning | |
CN110414788B (en) | Electric energy quality prediction method based on similar days and improved LSTM | |
CN111241657B (en) | Cold-rolled strip steel plate shape prediction method based on deep learning | |
CN110287583B (en) | Industrial equipment residual life prediction method based on cyclic neural network | |
Li et al. | Dynamic time features expanding and extracting method for prediction model of sintering process quality index | |
CN108732931A (en) | A kind of multi-modal batch process modeling method based on JIT-RVM | |
CN113591374B (en) | Cement finished product specific surface area prediction method based on convolution simple circulation unit network | |
CN114282443B (en) | Residual service life prediction method based on MLP-LSTM supervised joint model | |
CN113449919B (en) | Power consumption prediction method and system based on feature and trend perception | |
Liu et al. | Correntropy long short term memory soft sensor for quality prediction in industrial polyethylene process | |
CN114015825A (en) | Method for monitoring abnormal state of blast furnace heat load based on attention mechanism | |
CN115828754A (en) | Cutter wear state monitoring method based on multi-scale space-time fusion network model | |
CN116520772A (en) | Sheet part numerical control machining quality prediction method based on multitasking transfer learning | |
CN110987436A (en) | Bearing fault diagnosis method based on excitation mechanism | |
CN114298134A (en) | Wind power prediction method and device and electronic equipment | |
CN114596726B (en) | Parking berth prediction method based on interpretable space-time attention mechanism | |
CN116975645A (en) | Industrial process soft measurement modeling method based on VAE-MRCNN | |
CN114118401A (en) | Neural network-based power distribution network flow prediction method, system, device and storage medium | |
CN113762591A (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM counterstudy | |
CN117289652A (en) | Numerical control machine tool spindle thermal error modeling method based on multi-universe optimization | |
CN112149896A (en) | Attention mechanism-based mechanical equipment multi-working-condition fault prediction method | |
CN117235599A (en) | Deep learning-based smelting furnace temperature control optimization method | |
CN115936062A (en) | Photovoltaic power day-ahead interval prediction method based on scene generation and deep learning | |
CN115438309A (en) | Unmanned aerial vehicle engine health assessment method based on LSTM-SVR |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |