Method for improving prediction accuracy of finish rolling temperature based on deep learning algorithm

Abstract

The invention belongs to the technical field of steel production, and relates to a method for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm. According to the method, multiple key data, such as finish rolling inlet temperature, rolling force, cooling water flow and rolling speed, influencing the finish rolling temperature of hot rolled strip steel are selected through correlation analysis, and a generalized regression neural network (COA-GRNN) optimized by a raccoon optimization algorithm is utilized to establish a finish rolling temperature prediction model. The model realizes remote monitoring of the finish rolling temperature, so that production personnel can predict the trend of the finish rolling temperature in advance and adjust the finish rolling temperature in time, thereby ensuring the quality stability of strip steel and the yield of production.

Description

Technical Field

[0001] The present invention belongs to the technical field of steel production, and relates to a method for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm, which is used for final rolling temperature prediction in a hot-rolled strip production process. Background Art

[0002] In the steel production process, hot rolled strip is a key material widely used in multiple industrial fields. One of the core goals of the hot rolling process is to ensure the accuracy of the final rolling temperature (i.e. the temperature of the strip at the end of the last rolling process) because it has a direct impact on the microstructure, mechanical properties and surface quality of the strip. Accurate control of the final rolling temperature is essential to improving product quality, reducing energy consumption and increasing production efficiency. However, due to the complexity of the hot rolling process, the influence of multiple factors such as the finishing inlet temperature, rolling force, cooling water flow, rolling speed, etc., makes it a challenge to accurately predict and control the final rolling temperature.

[0003] Traditional methods usually rely on operator experience to make manual adjustments or use physical model-based control systems to try to maintain the required final rolling temperature. These methods often fail to fully consider all influencing factors and are difficult to adapt to changes in the production process, resulting in unstable yields and quality fluctuations.

[0004] In order to solve these problems, a method is needed that can comprehensively consider multiple key data and accurately predict the final rolling temperature, so as to realize real-time monitoring and control of the final rolling temperature, thereby improving the quality and efficiency of strip production. Summary of the invention

[0005] Improper control of the final rolling temperature may lead to problems such as poor strip surface quality, mechanical property deviation, and reduced product dimensional accuracy. The present invention proposes a method for improving the accuracy of final rolling temperature prediction based on deep learning, which can effectively predict the final rolling temperature.

[0006] The present invention is achieved through the following technical solutions:

[0007] A method for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm, the method comprising the following steps:

[0008] (1) Collect various key data during the hot-rolled strip production process;

[0009] (2) performing correlation analysis on the multiple key data collected in step (1) to obtain strongly correlated data affecting the final rolling temperature of the hot-rolled strip;

[0010] (3) performing outlier detection and normalization processing on the strongly correlated data to obtain pre-processed strongly correlated data;

[0011] (4) using the pre-processed strongly correlated data as input, using a deep learning algorithm to establish a final rolling temperature prediction model, and training the final rolling temperature prediction model to obtain a trained final rolling temperature prediction model;

[0012] (5) Use the trained final rolling temperature prediction model to predict the final rolling temperature.

[0013] Furthermore, in step (1), the multiple key data include: finishing rolling inlet temperature, rolling force of each rolling mill, roll gap setting value, AGC locking thickness, rolling mill speed, cooling water flow, strip steel pressure and target thickness.

[0014] Furthermore, in step (2), after correlation analysis, data with a correlation with the final rolling temperature greater than 45% are determined as strongly correlated data affecting the final rolling temperature of the hot-rolled strip.

[0015] Furthermore, in step (3), specifically:

[0016] The isolation forest algorithm is used to detect outliers in the strongly correlated data. The parameter selection range is: the number of trees is 50-300, the proportion of outliers is 2%-15%, the seed of the random number generator is 25-80, and the outliers are deleted; then the Z-Score normalization method is used for normalization processing to obtain the pre-processed strongly correlated data.

[0017] Furthermore, in step (4), the raccoon optimization algorithm is used to optimize the smoothness factor of the generalized regression neural network to establish a COA-GRNN final rolling temperature prediction model; and through the optimized smoothness factor, a trained GRNN final rolling temperature prediction model is obtained.

[0018] Furthermore, the initialization parameter range of the COA-GRNN final rolling temperature prediction model is set to: 10-500 iterations, the number of wolf packs is set to 1-100, the number of leaders is set to 1-10, the update factor is 0.1-0.9, the exploration factor is 0.001-0.9 and the convergence factor is set to 0.0001-0.85.

[0019] Furthermore, step (5) is specifically as follows: inputting the real-time production data into the trained COA-GRNN final rolling temperature prediction model to predict the final rolling temperature of the current strip.

[0020] Furthermore, the method further comprises the following steps:

[0021] After predicting the final rolling temperature of the current strip, the predicted result is compared with the preset normal temperature range. If the predicted result is within the normal range, the production line continues as planned; if the predicted temperature is not within the normal range, an early warning signal is issued, and the production personnel need to intervene in time to adjust the parameters of each rolling mill. The adjusted parameters will be input into the trained COA-GRNN final rolling temperature prediction model for prediction again, and the adjustment will be repeated until the predicted temperature returns to the set normal temperature range and the adjustment is stopped.

[0022] Further, the normal temperature range of the setting is 750°C to 950°C.

[0023] A system for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm, the system comprising:

[0024] Big data collection and analysis module: used to collect various key data that affect the final rolling temperature of hot-rolled strip during the production process of hot-rolled strip, and to perform correlation analysis on the various key data collected, and select the strongly correlated data that affect the final rolling temperature of hot-rolled strip;

[0025] Data processing module: used for performing outlier detection and normalization processing on the strongly correlated data to obtain the strongly correlated data after preprocessing;

[0026] Deep learning algorithm module: The raccoon optimization algorithm is used to optimize the generalized regression neural network and establish the COA-GRNN final rolling temperature prediction model; and the trained COA-GRNN final rolling temperature prediction model is obtained by optimizing the parameters of the neural network;

[0027] Remote monitoring module: used to compare the prediction results of the trained COA-GRNN final rolling temperature prediction model with the preset normal temperature range. If the prediction results are within the normal range, the production line will continue as planned; if the predicted temperature is not within the normal range, an early warning signal will be issued, and production personnel will intervene in time to adjust the parameters of each rolling mill.

[0028] Beneficial technical effects of the present invention:

[0029] (1) The method provided by the present invention adopts a generalized regression neural network (COA-GRNN) optimized by a raccoon optimization algorithm. The trained COA-GRNN final rolling temperature prediction model can more accurately predict the final rolling temperature; the high-precision prediction enables production personnel to better control the production process, thereby ensuring product quality.

[0030] (2) The method provided by the present invention realizes remote monitoring of the final rolling temperature, so that production personnel can understand the temperature change trend in real time and make timely adjustments, which greatly improves the flexibility and response speed of production.

[0031] (3) The method provided by the present invention can reduce the scrap rate in the production process through accurate temperature prediction and effective temperature control, avoid product defects caused by improper temperature, and thus improve the overall production efficiency.

[0032] (4) The method provided by the present invention can avoid unnecessary energy waste, such as excessive heating or cooling, by accurately controlling the final rolling temperature, which not only reduces production costs but also meets the environmental protection requirements of energy conservation and emission reduction.

[0033] (5) The method provided by the present invention can enhance the stability of product quality: a stable final rolling temperature helps to maintain the performance consistency of the strip, which is particularly important for high-end applications such as the automotive and aviation fields that have strict requirements on material properties.

[0034] (6) The method provided by the present invention can promote intelligent production: it helps to promote the development of hot-rolled strip production towards intelligence and automation, reduce the interference of human factors, and improve the overall intelligence level of production.

[0035] In summary, the present invention not only improves the prediction accuracy of the final rolling temperature, but also promotes the improvement of production efficiency and product quality through real-time monitoring and adjustment, while saving energy consumption and promoting the intelligence of the production process, which has significant economic and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of a method for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm in an embodiment of the present invention;

[0037] Figure 2 A thermal map showing the correlation between the collected data and the final rolling temperature in an embodiment of the present invention;

[0038] Figure 3 This is a flow chart of optimizing the smoothness factor of a generalized regression neural network using the raccoon optimization algorithm in an embodiment of the present invention;

[0039] Figure 4 It is the MSE convergence curve diagram of the COA-GRNN final rolling temperature prediction model in the embodiment of the present invention;

[0040] Figure 5 It is a line chart of the training results of the COA-GRNN final rolling temperature prediction model in an embodiment of the present invention;

[0041] Figure 6 It is a line chart of the test set results of the COA-GRNN final rolling temperature prediction model in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The embodiments of the present invention are described in detail below in conjunction with the accompanying drawings so that the implementation methods of the present invention can be thoroughly understood by those skilled in the art and the protection scope of the present invention can be better defined.

[0043] The present invention provides a method for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm, aiming to achieve high-precision prediction of the final rolling temperature by comprehensively analyzing a variety of process parameters and combining advanced optimization algorithms and neural network models. The method of the present invention includes: selecting a variety of key data affecting the final rolling temperature of hot-rolled strip by correlation analysis, such as finishing inlet temperature, rolling force, cooling water flow rate, and rolling speed, etc., removing abnormal points by the isolation forest method and normalizing the data, and using data training to optimize the relevant parameters using a generalized regression neural network (COA-GRNN) optimized by the raccoon optimization algorithm to establish a final rolling temperature prediction model. This model realizes remote monitoring of the final rolling temperature, allowing production personnel to predict the trend of the final rolling temperature in advance and make adjustments in time, thereby ensuring the quality stability of the strip and the yield of production. The specific implementation process of the above content will be described below:

[0044] Embodiment 1: A method for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm, such as Figure 1 As shown, the method comprises the following steps:

[0045] (1) Collect various key data during the hot-rolled strip production process;

[0046] (2) performing correlation analysis on the multiple key data collected in step (1) to obtain strongly correlated data affecting the final rolling temperature of the hot-rolled strip;

[0047] (3) performing outlier detection and normalization processing on the strongly correlated data to obtain pre-processed strongly correlated data;

[0048] (4) using the pre-processed strongly correlated data as input, using a deep learning algorithm to establish a final rolling temperature prediction model, and training the final rolling temperature prediction model to obtain a trained final rolling temperature prediction model;

[0049] (5) Use the trained final rolling temperature prediction model to predict the final rolling temperature.

[0050] In step (1) of this embodiment, the various key data include: finishing rolling inlet temperature, rolling force of each rolling mill, roll gap setting value, AGC locking thickness, rolling mill speed, cooling water flow, strip steel pressure and target thickness.

[0051] In step (2) of this embodiment, after correlation analysis, data with a correlation with the final rolling temperature greater than 45% are determined as strongly correlated data affecting the final rolling temperature of the hot-rolled strip. Specifically, Figure 2 In order to collect the correlation heat map between the data and the final rolling temperature, in this embodiment, the rolling mills are numbered in sequence according to the distance from the finishing rolling exit, and the selected data with strong correlation include: the main machine speed of the No. 1 rolling mill; the rolling force of the No. 1 to No. 6 rolling mills, the roll gap setting and the AGC locked thickness of the No. 2 to No. 5 rolling mills, and 16 items of target thickness. The data with strong correlation can be selected by re-implementing step (3) according to different production line conditions.

[0052] In step (3) of this embodiment, specifically:

[0053] The isolation forest algorithm is used to detect outliers in the strongly correlated data, and the parameters are set as follows: the number of trees is 100, the proportion of outliers is 10%, the seed of the random number generator is 42, and the outliers are deleted; then the Z-Score normalization method is used to perform normalization processing to obtain the pre-processed strongly correlated data.

[0054] The corresponding function of this step is: since data collection in the industrial production process is affected by many factors, the strongly correlated data affecting the final rolling temperature of the hot-rolled strip selected in step (2) may have abnormal points (such as sensor failure, noise interference, etc.). Detecting and processing the abnormal points of the data can ensure the quality of the model training data, and normalizing the data to eliminate the dimensional differences between different data.

[0055] In step (4) of this embodiment, the raccoon optimization algorithm is used to optimize the smoothness factor of the generalized regression neural network to establish a COA-GRNN final rolling temperature prediction model; and through the optimized smoothness factor, the trained optimal GRNN final rolling temperature prediction model is obtained.

[0056] The raccoon optimization algorithm simulates the feeding and exploration behavior of raccoons and adopts a population search strategy, showing strong robustness and optimization capabilities in multi-peak, nonlinear, discrete and other problems. It is simple and easy to implement, and is widely used in function optimization, path planning, machine learning and other fields. It can quickly converge to the optimal solution, while balancing global exploration and local development to avoid falling into the local optimal solution too early; Figure 3 As shown, the main steps are as follows:

[0057] A certain number of individuals are randomly initialized in the optimization space, each of which represents a set of initial solutions, and these individuals are selected. Once an individual is determined, it is used as a search sub-population, and the set corresponding to the optimal solution in all populations is found through a sorting algorithm. The initialization process can be expressed as:

[0058]

[0059] Among them, xi，j is the population particle; is the lower boundary of the optimization range; is the upper boundary of the optimization range; r is a random number ranging from 0 to 1.

[0060] The possible optimal solution is found by the movement of individuals in the search space; this process is similar to the behavior of a raccoon digging on the ground, observing and exploring the surrounding environment. By constantly moving and trying in the search space, areas that may contain the optimal solution are found. This is often used to simulate the behavior of raccoons attacking iguanas.

[0061] If we assume that the optimal particle position in the population is the position of the iguana, when half of the raccoons climb the tree and the other half wait on the ground for the iguana to fall, the position model is expressed as:

[0062]

[0063] In the formula, is the position of the i-th raccoon at time j; G j is the optimal position of the iguana at time j; r and i are both random numbers; N is the total number of raccoons; I is a number randomly selected from the set {1, 2}; m is the number of decision variables.

[0064] The iguana is considered to be in a random position when it falls, while the ground raccoon moves in the search space waiting to be killed; this process is represented as:

[0065]

[0066] In the formula, is the ground position of the iguana at time j;

[0067]

[0068] In the formula, is the objective function value of the iguana falling to the ground at time j; F i，j is the objective function value of the ith raccoon at time j. If the selected individual is better after the update, it is updated, otherwise it remains the same.

[0069]

[0070] In the formula, and F i is the objective function value of the i-th raccoon at the new and old positions respectively.

[0071] Perform a detailed search for the areas of possible optimal solutions that have been discovered; in this stage, the algorithm will pay more attention to those individuals that have shown good performance and conduct a more detailed search around them; this process is similar to the behavior of a raccoon carefully digging and observing the area around the food after finding it; the quality of the solution is gradually improved, and a solution closer to the global optimal solution is found. To simulate this behavior, we can get:

[0072]

[0073] In the formula, is the lower bound of the j-th decision quantity, is the upper bound of the j-th decision quantity; t is the number of iterations; T is the maximum number of iterations;

[0074]

[0075]

[0076] In the formula, is the new position of the i-th raccoon at time j; if the updated individual is better, the current position is replaced, otherwise it remains the same until the optimal position is found and the optimal smoothness factor is output.

[0077] Figure 4 The training result line graph of the COA-GRNN model in the embodiment of the present invention shows that the accuracy on the training set gradually increases with the increase of training rounds. This shows that the model is constantly learning and adapting to the training data, and can more accurately predict the sample labels in the training set; Figure 5 This is a line graph of the training set error of the COA-GRNN model in the embodiment of the present invention. As the number of training rounds increases, the error on the training set gradually decreases, which is consistent with the improvement of the accuracy of the training set. The reduction in error means that the model's prediction on the training data is becoming more and more accurate.

[0078] The initialization parameter range of the COA-GRNN final rolling temperature prediction model is set to: 10-500 iterations, the number of wolf packs is set to 1-100, the number of leaders is set to 1-10, the update factor is 0.1-0.9, the exploration factor is 0.001-0.9 and the convergence factor is set to 0.0001-0.85. In this embodiment, preferably, the initialization parameters of the COA-GRNN final rolling temperature prediction model are set to: 100 iterations, the number of wolf packs is set to 20, the number of leaders is set to 4, the update factor, the exploration factor and the convergence factor are all set to 0.5; the model's optimal smoothing factor value is 0.029401; the above parameters can be adjusted according to actual conditions.

[0079] In this embodiment, step (5) is specifically as follows: inputting the real-time production data into the trained COA-GRNN final rolling temperature prediction model to predict the final rolling temperature of the current strip. Figure 6 The test set result line chart of the COA-GRNN model in the embodiment of the present invention shows that the R^2 of the prediction model is 0.99509, which is significantly improved compared with the R^2 of GRNN: 0.58984.

[0080] In this embodiment, the method further comprises the following steps:

[0081] After predicting the final rolling temperature of the current strip, the predicted result is compared with the preset normal temperature range. If the predicted result is within the normal range, the production line continues to proceed as planned; if the predicted temperature is not within the normal range, an early warning signal is issued, and the production personnel need to intervene in time to adjust the parameters of each rolling mill. The adjusted parameters will be input into the trained COA-GRNN final rolling temperature prediction model for prediction again, and the adjustment is repeated until the predicted temperature returns to the set normal temperature range (the normal temperature range in this embodiment is 750°C to 950°C), and the adjustment is stopped.

[0082] The method for improving the prediction accuracy of the final rolling temperature provided by the present invention can realize real-time prediction and remote monitoring of the final rolling temperature of the hot-rolled strip, and output prediction results and trend analysis; by optimizing and updating the deep learning model, the prediction accuracy and stability of the final rolling temperature of the hot-rolled strip are continuously improved; it can provide production personnel with instant final rolling temperature prediction information, and support decision-making and adjustment of the production process to maximize the production efficiency and quality stability of the strip; the method effectively reduces the defective rate in the production process and improves the final quality of the strip through remote monitoring and real-time adjustment; the method improves production efficiency and reduces resource waste by real-time monitoring and prediction of the final rolling temperature of the hot-rolled strip; and promotes the intelligence of the production process, with significant economic and social benefits.

[0083] Embodiment 2: A system for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm, using the method described in Embodiment 1, the system comprising:

[0084] Big data collection and analysis module: used to collect various key data that affect the final rolling temperature of hot-rolled strip during the production process of hot-rolled strip, and perform correlation analysis on the various key data collected to select strongly correlated data that affect the final rolling temperature of hot-rolled strip; specifically, data with a correlation with the final rolling temperature greater than 45% is determined as strongly correlated data that affects the final rolling temperature of hot-rolled strip;

[0085] Data processing module: used for outlier detection and normalization of the strongly correlated data to obtain pre-processed strongly correlated data; this module detects and processes outliers on the data to ensure the quality of the model training data; and normalizes the data to eliminate the dimensional differences between different data. The specific operations in the data processing module are: using the isolation forest algorithm for outlier detection, setting the parameters to: the number of trees is 100, the proportion of outliers is 10%, and the seed of the random number generator is 42. And delete the outliers. Finally, use the Z-Score normalization method to normalize the data.

[0086] Deep learning algorithm module: The raccoon optimization algorithm is used to optimize the generalized regression neural network and establish the COA-GRNN final rolling temperature prediction model; and the trained COA-GRNN final rolling temperature prediction model is obtained by optimizing the parameters of the neural network;

[0087] Remote monitoring module: used to compare the prediction results of the trained COA-GRNN final rolling temperature prediction model with the preset normal temperature range. If the prediction results are within the normal range, the production line will continue as planned; if the predicted temperature is not within the normal range, an early warning signal will be issued, and production personnel will intervene in time to adjust the parameters of each rolling mill.

[0088] The above is a specific implementation process of the present invention, which is intended to facilitate the understanding of those skilled in the art, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process change made according to the description and the drawings within the technical scope disclosed by the present invention is included in the patent protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm, characterized in that: The method comprises the following steps: (1) Collect various key data during the hot-rolled strip production process; (2) performing correlation analysis on the multiple key data collected in step (1) to obtain strongly correlated data affecting the final rolling temperature of the hot-rolled strip; (3) performing outlier detection and normalization processing on the strongly correlated data to obtain pre-processed strongly correlated data; (4) using the pre-processed strongly correlated data as input, using a deep learning algorithm to establish a final rolling temperature prediction model, and training the final rolling temperature prediction model to obtain a trained final rolling temperature prediction model; (5) Use the trained final rolling temperature prediction model to predict the final rolling temperature.

2. A method for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm according to claim 1, characterized in that: In step (1), the multiple key data include: finishing rolling inlet temperature, rolling force of each rolling mill, roll gap setting value, AGC locking thickness, rolling mill speed, cooling water flow, strip steel pressure and target thickness.

3. The method for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm according to claim 1, characterized in that: In step (2), after correlation analysis, data with a correlation with the final rolling temperature greater than 45% are determined as strongly correlated data affecting the final rolling temperature of the hot-rolled strip.

4. The method for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm according to claim 1, characterized in that: In step (3), specifically: The isolation forest algorithm is used to detect outliers in the strongly correlated data. The parameter selection range is: the number of trees is 50-300, the proportion of outliers is 2%-15%, the seed of the random number generator is 25-80, and the outliers are deleted; then the Z-Score normalization method is used for normalization processing to obtain the pre-processed strongly correlated data.

5. The method for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm according to claim 1, characterized in that: In step (4), the raccoon optimization algorithm is used to optimize the smoothness factor of the generalized regression neural network, and a COA-GRNN final rolling temperature prediction model is established; and through the optimized smoothness factor, a trained GRNN final rolling temperature prediction model is obtained.

6. A method for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm according to claim 5, characterized in that: The initialization parameter range of the COA-GRNN final rolling temperature prediction model is set to: 10-500 iterations, the number of wolf packs is set to 1-100, the number of leaders is set to 1-10, the update factor is 0.1-0.9, the exploration factor is 0.001-0.9 and the convergence factor is set to 0.0001-0.

85.

7. The method for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm according to claim 5, characterized in that: Step (5) is specifically as follows: inputting the real-time production data into the trained COA-GRNN final rolling temperature prediction model to predict the final rolling temperature of the current strip.

8. The method for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm according to claim 1, characterized in that: The method further comprises the following steps: After predicting the final rolling temperature of the current strip, the predicted result is compared with the preset normal temperature range. If the predicted result is within the normal range, the production line continues as planned; if the predicted temperature is not within the normal range, an early warning signal is issued, and the production personnel need to intervene in time to adjust the parameters of each rolling mill. The adjusted parameters will be input into the trained COA-GRNN final rolling temperature prediction model for prediction again, and the adjustment will be repeated until the predicted temperature returns to the set normal temperature range and the adjustment is stopped.

9. A method for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm according to claim 8, characterized in that: The normal temperature range for the arrangement is 750°C to 950°C.

10. A system for improving the accuracy of final rolling temperature prediction based on a deep learning algorithm, characterized in that: The system comprises: Big data collection and analysis module: used to collect various key data that affect the final rolling temperature of hot-rolled strip during the production process of hot-rolled strip, and to perform correlation analysis on the various key data collected, and select the strongly correlated data that affect the final rolling temperature of hot-rolled strip; Data processing module: used for performing outlier detection and normalization processing on the strongly correlated data to obtain the strongly correlated data after preprocessing; Deep learning algorithm module: The raccoon optimization algorithm is used to optimize the generalized regression neural network and establish the COA-GRNN final rolling temperature prediction model; and the trained COA-GRNN final rolling temperature prediction model is obtained by optimizing the parameters of the neural network; Remote monitoring module: used to compare the prediction results of the trained COA-GRNN final rolling temperature prediction model with the preset normal temperature range. If the prediction results are within the normal range, the production line will continue as planned; if the predicted temperature is not within the normal range, an early warning signal will be issued, and production personnel will intervene in time to adjust the parameters of each rolling mill.

Images