Steel manufacturing process energy consumption prediction method and system based on multi-scale channel hybrid convolutional neural network
By using a multi-scale channel hybrid convolutional neural network, the complex problem of processing multi-source heterogeneous features in steel production was solved, achieving higher accuracy and reliability in energy consumption prediction and improving the model's predictive performance.
Patent Information
- Application Number
- CN202510958312.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-11
- Publication Date
- 2025-11-11
AI Technical Summary
Existing technologies address the problems of complex multi-source heterogeneous feature processing and variable multi-feature coupling leading to reduced model prediction performance in steel production processes.
A multi-scale channel hybrid convolutional neural network is adopted, and an energy consumption prediction model is constructed through data embedding, channel mixing, multi-scale dilated convolution and gradient backpropagation algorithms. The model includes data preprocessing, embedding module, channel mixing module, multi-scale feature extraction module and regression prediction module.
It improves the accuracy and reliability of energy consumption prediction. By integrating channel mixing and attention mechanisms, it effectively extracts the potential features of process data, identifies long-distance dependencies, and establishes regression models, thereby improving prediction accuracy and efficiency.
Smart Images
Figure CN120930852A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of soft measurement technology, specifically relating to a method and system for predicting energy consumption in steel manufacturing processes based on a multi-scale channel hybrid convolutional neural network. Background Technology
[0002] The steel manufacturing industry is one of the world's most fundamental industries, consuming 10% of global energy and accounting for a significant portion of global carbon emissions. The goal of steelmaking is to produce high-quality steel products at the lowest possible cost. To achieve this, real-time measurement and control of energy consumption in steel production is essential. Traditionally, these real-time measurements have been conducted under highly complex conditions, including high temperature and high pressure, posing significant challenges to energy consumption monitoring. Therefore, the steel industry urgently needs technologies for timely and accurate energy consumption prediction. The steelmaking process itself contains multi-source heterogeneous characteristics. These characteristics exhibit different data properties, often reducing the accuracy of predictive models. Summary of the Invention
[0003] The technical problem to be solved by this invention is to provide a method and system for predicting energy consumption in steel manufacturing processes based on multi-scale channel hybrid convolutional neural networks, which solves the problems of complex processing of multi-source heterogeneous features in steel production processes and reduced model prediction performance caused by multi-feature coupling of variables in the prior art.
[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:
[0005] A method for predicting energy consumption in steel manufacturing processes based on multi-scale channel hybrid convolutional neural networks includes the following steps:
[0006] Step 1: Obtain production data from the steel manufacturing process, preprocess it to obtain a historical dataset, which includes input data and output data. The output data is energy consumption, and the input data is other parameters in the steelmaking process.
[0007] Step 2: Construct a data embedding module to perform size transformation on the input data and obtain the embedding features;
[0008] Step 3: Perform channel blending on the embedded features to achieve channel feature interaction, and construct a multi-scale dilated convolutional neural network with multiple convolutional kernels of different sizes to obtain multi-scale features;
[0009] Step 4: Construct a regressor based on a feedforward neural network, and input multi-scale features into the regressor to predict the output;
[0010] Step 5: Optimize the multi-scale channel hybrid convolutional neural network using the gradient backpropagation algorithm until the maximum number of iterations is reached;
[0011] Step 6: Obtain real-time data of input variables during the steel manufacturing process, input them into the optimized multi-scale channel hybrid convolutional neural network, and obtain the predicted energy consumption value.
[0012] In step 2, the input data is resized to obtain the embedded features, using the following formula:
[0013]
[0014] in, , representing embedded features, This indicates that input data will be provided. The elements in are rearranged to obtain a OK A matrix of columns , These represent the weight matrix and bias vector of the embedding layer, respectively. This represents the matrix transpose operation.
[0015] In step 3, the embedded features are channel-blended using the following formula:
[0016]
[0017]
[0018] in, They represent A fully connected network express The Line, i.e., the first One channel, This indicates a batch normalization operation. This indicates a merge operation.
[0019] In step 3, the multi-scale dilated convolutional neural network is constructed using the following formula:
[0020]
[0021] in, Representing multi-scale features, This represents different branches of a multi-scale convolutional network. Indicates channel mixing characteristics,
[0022]
[0023]
[0024] in, Let these represent the weight matrix and bias vector of the first and second linear transformations, respectively. Represents the linear rectified function. It is a three-layer fully connected network activated by GELU.
[0025] Multi-scale convolutional network branches include, the first branch consisting of a Ordinary convolution and one The second branch consists of two dilated convolutions. Ordinary convolution and The third branch consists of a dilated convolution; Ordinary convolution, one Ordinary convolution and one Composed of dilated convolutions; the fourth branch consists of one Ordinary convolution, one ,one Ordinary convolution, one Convolution, one Dilated convolution consists of convolution and a Composed of voided convolutions.
[0026] The regressor in step 4 is represented by the following formula:
[0027]
[0028] in, This represents a feedforward neural network consisting of three fully connected layers and a LeakyReLU activation function, with the last layer of the network having no activation function.
[0029] The specific process of step 5 is as follows:
[0030] Set the maximum number of iterations. ,when When optimizing, the following formula is used:
[0031]
[0032]
[0033]
[0034] in, Indicates the first The parameters of the multi-scale channel hybrid convolutional neural network during the next iteration. This represents the adaptive dynamic estimation algorithm, which yields an optimized multi-scale channel hybrid convolutional neural network after iteration.
[0035] The steel manufacturing process energy consumption prediction system includes a data embedding module, which is used to acquire steel production process data and perform data embedding to obtain embedding features;
[0036] The channel blending module performs channel blending on embedded features to achieve channel feature interaction;
[0037] The multi-scale feature extraction module constructs a multi-scale dilated convolutional neural network with multiple convolutional kernels of different sizes to obtain multi-scale features.
[0038] The regression prediction module constructs a regressor based on a feedforward neural network, inputting multi-scale features into the regressor to predict the output.
[0039] The network optimization module uses the gradient backpropagation algorithm to optimize the multi-scale channel hybrid convolutional neural network until the maximum number of iterations is reached.
[0040] An electronic device includes a memory and a processor; wherein the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, wherein when the computer-executable instructions are executed by the processor, the steps of the energy consumption prediction method for the steel manufacturing process are implemented.
[0041] A computer-readable storage medium, characterized in that: the computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement all or part of the steps of the energy consumption prediction method for the steel manufacturing process.
[0042] Compared with the prior art, the present invention has the following beneficial effects:
[0043] 1. A convolutional neural network based on multi-scale channel mixing is proposed. This network is designed based on channel mixing and attention mechanisms, and the proposed model can fully extract potentially useful representations from the original data. Furthermore, the channel mixing mechanism is used to capture channel dependencies to leverage the low-rank properties of process data, thereby achieving better prediction accuracy and efficiency. Dilated convolutional modules are used to identify long-distance dependencies between different features. By integrating channel mixing and attention mechanisms, multi-source process data from the entire steel production process can be meticulously combined at the data level.
[0044] 2. The extracted features are fed into a downstream fully connected layer to establish a regression model that maps process data to steel product quality. The proposed model effectively improves the accuracy and reliability of energy consumption prediction. Attached Figure Description
[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0046] Figure 1 This is a schematic diagram of the general process of the energy consumption forecasting method for steel manufacturing in this invention.
[0047] Figure 2 This is a structural diagram of a multi-scale dilated convolution module according to an embodiment of the present invention.
[0048] Figure 3 The graph shows the energy consumption prediction fitting curves obtained by different comparative algorithms of the energy consumption prediction method for the steel manufacturing process of this invention.
[0049] Figure 4 This is a graph showing the energy consumption prediction error curves obtained from different comparative algorithms of the energy consumption prediction method for the steel manufacturing process of this invention. Detailed Implementation
[0050] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.
[0051] A method for predicting energy consumption in steel manufacturing processes based on multi-scale channel hybrid convolutional neural networks includes the following steps:
[0052] Step 1: Obtain production data from the steel manufacturing process, preprocess it to obtain a historical dataset, which includes input data and output data. The output data is energy consumption, and the input data is other parameters in the steelmaking process.
[0053] Step 2: Construct a data embedding module to perform size transformation on the input data and obtain the embedding features;
[0054] Step 3: Perform channel blending on the embedded features to achieve channel feature interaction, and construct a multi-scale dilated convolutional neural network with multiple convolutional kernels of different sizes to obtain multi-scale features;
[0055] Step 4: Construct a regressor based on a feedforward neural network, and input multi-scale features into the regressor to predict the output;
[0056] Step 5: Optimize the multi-scale channel hybrid convolutional neural network using the gradient backpropagation algorithm until the maximum number of iterations is reached;
[0057] Step 6: Obtain real-time data of input variables during the steel manufacturing process, input them into the optimized multi-scale channel hybrid convolutional neural network, and obtain the predicted energy consumption value.
[0058] Specific embodiments, such as Figures 1 to 4 As shown,
[0059] A method for predicting energy consumption in steel manufacturing processes based on a multi-scale channel hybrid convolutional neural network is provided, including:
[0060] S1: Obtain operational data of the steel manufacturing process and preprocess it to obtain a dataset;
[0061] This invention predicts energy consumption in the steel manufacturing process, and the final output variable is steelmaking energy consumption;
[0062] Data processing for missing and outlier values in steel manufacturing processes, as well as data standardization;
[0063] Preferably, the preprocessed data is used as training samples, and the training samples are divided into two groups: 80% is used as the training dataset and the remaining 20% is used as the test dataset.
[0064] It should be noted that by handling missing and outlier values, and by standardizing the data, we can ensure the quality and reliability of the data, avoid the impact of these values on model training and prediction results, and improve model performance. In addition, dividing the data into training and testing sets can better evaluate the model's performance and prevent overfitting, which helps reduce energy consumption and improve production efficiency. Through data analysis and processing, we can provide more accurate data support for decision-making in the steelmaking process, thereby improving the accuracy of decision-making.
[0065] S2: Assume the obtained process input and output data are as follows: ,in The number of variables representing the input data
[0066] A data embedding module is constructed to perform size transformation on historical data and obtain embedded features;
[0067] Use formula (1) to process the input data Perform size transformation
[0068]
[0069] in, , representing embedded features, Indicates the vector The elements in are rearranged to obtain a OK A matrix of columns and The tables represent the weight matrix and bias vector of the embedding layer.
[0070] S3: Channel mixing is performed on the embedded features to achieve channel feature interaction, and a multi-scale dilated convolutional neural network with multiple convolutional kernels of different sizes is constructed to obtain multi-scale features;
[0071] Channel blending is performed on the embedded features using formulas (2)-(3):
[0072]
[0073]
[0074]
[0075]
[0076] in, They represent These are fully connected networks. Each of these fully connected networks typically consists of three layers, with each layer using Gaussian Error Linear Units (GELUs) as the activation function.
[0077] express The Line, i.e., the first One channel, This indicates batch normalization operation. This indicates a merge operation. It is a three-layer fully connected network activated by GELU. This represents the matrix transpose operation. This indicates the channel mixing characteristics. Let these represent the weight matrix and bias vector of the first and second linear transformations, respectively. This represents the rectified linear unit.
[0078] Construct a multi-scale dilated convolutional neural network using formula (6):
[0079]
[0080] in, It represents multi-scale features. This represents different branches of a multi-scale convolutional network, with the first branch consisting of a... Ordinary convolution and one It consists of dilated convolutions. The second branch consists of 2... Ordinary convolution and It consists of dilated convolutions. The third branch consists of a... Ordinary convolution and one Ordinary convolution and one It consists of dilated convolutions. The fourth branch consists of one... Ordinary convolution and one one Ordinary convolution and one Convolution and one Dilated convolution consists of convolution and a Composed of voided convolutions.
[0081] S4: Construct a regressor based on a feedforward neural network, and input multi-scale features into the regressor to predict the output;
[0082] Construct a regressor based on formula (7):
[0083]
[0084] in, This represents a feedforward neural network consisting of three fully connected layers and the LeakyReLU (Leaky Rectified Linear Unit) activation function. The last layer of the network has no activation function.
[0085] The gradient backpropagation algorithm is used to optimize the multi-scale channel hybrid convolutional neural network until the maximum number of iterations is reached;
[0086] Set the maximum number of iterations. ,when At that time, according to the target shown in formula (8)
[0087]
[0088]
[0089]
[0090] in, Indicates the first The parameters of the multi-scale channel hybrid convolutional neural network during the next iteration. This represents the Adaptive Moment Estimation algorithm. Once the iterations are complete, an optimized multi-scale channel hybrid convolutional neural network is obtained.
[0091] Real-time data of input variables during the steelmaking process are obtained and input into an optimized multi-scale channel hybrid convolutional neural network to obtain predicted energy consumption values.
[0092] Assume the real-time input data obtained during the steelmaking process is By using an optimized multi-scale channel hybrid convolutional neural network, the energy consumption prediction value is obtained according to the above steps.
[0093] For example,
[0094] To verify the beneficial effects of the energy consumption prediction method for steel manufacturing processes based on multi-scale channel hybrid convolutional neural networks, comparative results of several schemes are provided.
[0095] This embodiment uses real industrial data, sourced from DAE-WOO Steel Co., Ltd. of Gwangyang, South Korea.
[0096] The collected sample data were preprocessed, and missing values and outliers were handled and the data was standardized. Based on the process mechanism information and relevant expert experience, nine variables were selected as input variables, as shown in Table 2.
[0097] Table 1 Input variables for steel energy consumption prediction process
[0098]
[0099] 80% of the preprocessed data was used as the training dataset, and the remaining 20% was used as the test dataset to build a prediction model.
[0100] To analyze the performance of the proposed algorithm from multiple perspectives, this paper compares its performance with four soft measurement algorithms: CNN, DRDNet, MLP, and UNet. Furthermore, this invention uses the following three parameters as performance evaluation metrics for the prediction models in each prediction method, and their calculation formulas are as follows:
[0101] (1) Mean absolute error (MAPE): reflects the degree of deviation between the model's predicted value and the actual value. The smaller the value, the higher the accuracy of the model's prediction.
[0102]
[0103] in, and These represent the actual and predicted values of the output variable in the test set samples, respectively. for The average value;
[0104] (2) Root mean square error (RMSE): reflects the degree of deviation between the model's predicted value and the true value. The smaller the value, the higher the accuracy of the model's prediction.
[0105]
[0106] in, and These represent the actual and predicted values of the output variable in the test set samples, respectively. for The average value;
[0107] Table 2 Simulation results of different algorithms
[0108]
[0109] The simulation results of different algorithms are shown in Table 2. It can be seen that the method proposed in this invention achieves the best prediction performance, and it achieves the minimum values in both MAE and RMSE.
[0110] To more intuitively demonstrate the prediction performance of each algorithm, the energy consumption prediction fitting curves of different algorithms are shown below. Figure 3 As shown, the prediction error graphs for different algorithms are also shown below. Figure 4 As shown, it is clear that the prediction error curve of the MCNN algorithm fluctuates little around the zero error line, and the overall prediction error is significantly smaller than that of other algorithms, indicating that the model built by this algorithm can effectively improve the model's prediction performance.
[0111] The above is an illustrative scheme of an energy consumption forecasting method for steel manufacturing processes based on a multi-scale channel hybrid convolutional neural network, according to this embodiment. It should be noted that the technical solution of this system for forecasting energy consumption in steel manufacturing processes based on a multi-scale channel hybrid convolutional neural network belongs to the same concept as the aforementioned energy consumption forecasting method for steel manufacturing processes based on a multi-scale channel hybrid convolutional neural network. Details not described in detail in the technical solution of the system for forecasting energy consumption in steel manufacturing processes based on a multi-scale channel hybrid convolutional neural network in this embodiment can be found in the description of the aforementioned energy consumption forecasting method for steel manufacturing processes based on a multi-scale channel hybrid convolutional neural network.
[0112] The energy consumption prediction system for steel manufacturing processes based on a multi-scale channel hybrid convolutional neural network in this embodiment includes:
[0113] The data preprocessing module is used to acquire energy consumption data from the steel manufacturing process and preprocess it to obtain a dataset.
[0114] The data embedding module is used to acquire steel production process data and perform data embedding to obtain embedded features.
[0115] The channel blending module performs channel blending on embedded features to achieve channel feature interaction.
[0116] The multi-scale feature extraction module constructs a multi-scale dilated convolutional neural network with multiple convolutional kernels of different sizes to obtain multi-scale features.
[0117] The regression prediction module constructs a regressor based on a feedforward neural network, inputting multi-scale features into the regressor to predict the output.
[0118] The network optimization module uses the gradient backpropagation algorithm to optimize the multi-scale channel hybrid convolutional neural network until the maximum number of iterations is reached.
[0119] This embodiment also provides a computing device suitable for predicting energy consumption in steel manufacturing processes based on multi-scale channel hybrid convolutional neural networks, including:
[0120] The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement the energy consumption prediction method for the steel manufacturing process based on a multi-scale channel hybrid convolutional neural network, as proposed in the above embodiments.
[0121] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the energy consumption prediction method for steel manufacturing processes based on a multi-scale channel hybrid convolutional neural network as proposed in the above embodiments.
[0122] The storage medium proposed in this embodiment and the energy consumption prediction method for steel manufacturing process based on multi-scale channel hybrid convolutional neural network proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.
[0123] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.
[0124] It should be understood that this solution is not limited to the specific embodiments described above. Devices and structures not described in detail herein should be understood as being implemented in a manner common to the art. Any person skilled in the art can make many possible variations and modifications to this solution, or modify it into equivalent embodiments, without departing from the scope of this solution, using the methods and techniques disclosed above. This does not affect the substantive content of this solution. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of this solution, without departing from its scope, still fall within the protection scope of this solution.
Claims
1. A method for predicting energy consumption in steel manufacturing processes based on multi-scale channel hybrid convolutional neural networks, characterized by: Includes the following steps: Step 1: Obtain production data from the steel manufacturing process, preprocess it to obtain a historical dataset, which includes input data and output data. The output data is energy consumption, and the input data is other parameters in the steelmaking process. Step 2: Construct a data embedding module to perform size transformation on the input data and obtain the embedding features; Step 3: Perform channel blending on the embedded features to achieve channel feature interaction, and construct a multi-scale dilated convolutional neural network with multiple convolutional kernels of different sizes to obtain multi-scale features; Step 4: Construct a regressor based on a feedforward neural network, and input multi-scale features into the regressor to predict the output; Step 5: Optimize the multi-scale channel hybrid convolutional neural network using the gradient backpropagation algorithm until the maximum number of iterations is reached; Step 6: Obtain real-time data of input variables during the steel manufacturing process, input them into the optimized multi-scale channel hybrid convolutional neural network, and obtain the predicted energy consumption value.
2. The energy consumption prediction method for steel manufacturing process according to claim 1, characterized in that: In step 2, the input data is resized to obtain the embedded features, using the following formula: ， in, , representing embedded features, This indicates that input data will be provided. The elements in are rearranged to obtain a OK A matrix of columns , These represent the weight matrix and bias vector of the embedding layer, respectively. This represents the matrix transpose operation.
3. The energy consumption prediction method for steel manufacturing process according to claim 2, characterized in that: In step 3, the embedded features are channel-blended using the following formula: ， ， in, They represent A fully connected network express The Line, i.e., the first One channel, This indicates a batch normalization operation. This indicates a merge operation.
4. The energy consumption prediction method for steel manufacturing process according to claim 3, characterized in that: In step 3, the multi-scale dilated convolutional neural network is constructed using the following formula: ， in, Representing multi-scale features, This represents different branches of a multi-scale convolutional network. Indicates channel mixing characteristics, ， ， in, Let these represent the weight matrix and bias vector of the first and second linear transformations, respectively. Represents the linear rectified function. It is a three-layer fully connected network activated by GELU.
5. The energy consumption prediction method for steel manufacturing process according to claim 4, characterized in that: Multi-scale convolutional network branches include, the first branch consisting of a Ordinary convolution and one The second branch consists of two dilated convolutions. Ordinary convolution and The third branch consists of a dilated convolution; ,one Convolution and one Composed of dilated convolutions; the fourth branch consists of one ,one ,one ,one Convolution, one Dilated convolution consists of convolution and a Composed of voided convolutions.
6. The energy consumption prediction method for steel manufacturing process according to claim 4, characterized in that: The regressor in step 4 is represented by the following formula: ， in, This represents a feedforward neural network consisting of three fully connected layers and a LeakyReLU activation function, with the last layer of the network having no activation function.
7. The method for predicting energy consumption in the steel manufacturing process according to claim 1, characterized in that: The specific process of step 5 is as follows: Set the maximum number of iterations. ,when When optimizing, the following formula is used: ， ， ， in, Indicates the first The parameters of the multi-scale channel hybrid convolutional neural network during the iteration are given by adam, which represents the adaptive dynamic estimation algorithm. Once the iteration is complete, the optimized multi-scale channel hybrid convolutional neural network can be obtained.
8. An energy consumption prediction system for steel manufacturing processes, characterized in that: It includes a data embedding module, which is used to acquire steel production process data and perform data embedding to obtain embedded features; The channel blending module performs channel blending on embedded features to achieve channel feature interaction; The multi-scale feature extraction module constructs a multi-scale dilated convolutional neural network with multiple convolutional kernels of different sizes to obtain multi-scale features. The regression prediction module constructs a regressor based on a feedforward neural network, inputting multi-scale features into the regressor to predict the output. The network optimization module uses the gradient backpropagation algorithm to optimize the multi-scale channel hybrid convolutional neural network until the maximum number of iterations is reached.
9. An electronic device, characterized in that: It includes a memory and a processor; wherein the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the energy consumption prediction method for the steel manufacturing process according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions that, when executed by a processor, implement all or part of the steps of the energy consumption prediction method for the steel manufacturing process according to any one of claims 1 to 7.