Coil temperature prediction method based on multi-level feature fusion under cloud edge cooperation

Abstract

The application discloses a coiling temperature prediction method based on multi-level feature fusion under cloud edge end cooperation and belongs to the technical field of industrial control. The method comprises the following steps: respectively deploying a manufacturing execution layer, a process control layer and a real-time control layer at a cloud side, an edge side and an end side; collecting current strip process data of the process control layer and steel coil setting data of the manufacturing execution layer, and pre-processing the collected data at the end side; extracting features from the pre-processed current strip process data at the edge side to obtain bottom working condition information of a current slab; extracting features from the pre-processed steel coil setting data at the cloud side to obtain upper production scheduling setting information of a next slab; fusing the bottom working condition information of the current slab with the upper production scheduling setting information of the next slab to obtain multi-level fusion features, and realizing coiling temperature prediction based on the multi-level fusion features. The application scheme can effectively improve the coiling temperature prediction precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of industrial control technology, in particular to a coiling temperature prediction method based on multi-level feature fusion under cloud edge-end cooperation. BACKGROUND

[0002] In a complex industrial process, key process parameters have an irreplaceable influence on product quality. For example, in the hot strip rolling (HSRM) process, the coiling temperature (CT) has a significant impact on the mechanical properties of the coil. Existing CT prediction methods mostly only consider single-level information, without fully considering the influence of multi-level information in the actual process, resulting in poor prediction results.

[0003] The hot strip rolling process is complex and variable, and the continuous change of temperature is influenced by multiple factors, such as strip quality, speed, thickness, cooling water volume, water pressure, finishing temperature, heat conduction, convection, radiation conditions, etc. The interaction and influence between these factors make the prediction of coiling temperature extremely complex. In order to meet the requirements of product strip performance in the production process, the coiling temperature needs to have high prediction accuracy. Therefore, the accuracy of the prediction model becomes a problem to be solved. Coiling temperature control is one of the core links of hot strip rolling production, not only directly affecting the final performance of the strip, but also being a decisive factor for the smooth coiling and safe production of the strip. High-precision control of the coiling temperature can optimize the microstructure of the strip and improve the mechanical properties and dimensional accuracy of the product. Therefore, accurate control of the coiling temperature is a key link to ensure the quality of hot strip rolling production.

[0004] In modern steel industry, laminar cooling is a process of improving the organizational performance of the strip and improving the quality and yield of the strip through post-rolling forced water cooling, which has a direct and significant impact on the control accuracy of the coiling temperature. At the same time, the laminar cooling process has complex industrial characteristics such as drastic change of working conditions, strong nonlinearity, time-varying parameters, and difficulty in accurately describing mathematical models, so it is difficult to accurately describe based on mechanism model, which brings certain difficulties to process modeling. Data-driven methods directly learn and extract features from data and are suitable for complex and variable industrial fields, and have been successfully applied in suspension magnetization roasting process, blast furnace ironmaking process, hydrocracking process, petroleum refining, wastewater treatment, etc. and achieved good results.

[0005] With the continuous development of machine learning algorithms and the continuous accumulation of process history data, machine learning-based methods have been widely developed. Methods such as Support Vector Machine (SVM) and Random Forest (RF) are difficult to fully extract the time sequence characteristics of the complex hot continuous rolling process, and the prediction accuracy is limited. In addition to traditional machine learning-based methods, many improved neural network methods based on intelligent optimization algorithms have also been developed for temperature prediction, which do not fully consider the characteristics of the multi-level information of the process, and cannot guarantee high accuracy of temperature prediction modeling.

[0006] In order to extract the characteristics of continuous time series data, Long Short Term Memory (LSTM) networks and Gated Recurrent Unit (GRU) networks developed for time series data modeling are usually used. The GRU network is a variant of the LSTM network, which has the advantages of simpler architecture, fewer parameters and higher computational efficiency, and attention mechanisms have been incorporated into various GRU models to extract local features, but the above methods do not fully consider the influence of the target variable on the variable, and the multi-scale feature extraction of the variable is not sufficient. In addition, in modern industrial processes, traditional prediction methods only use single-scale information of the variable in the data preprocessing process, which may lose some useful information. SUMMARY

[0007] The present application provides a coiling temperature prediction method based on multi-level feature fusion under cloud edge end cooperation to solve the technical problem of low prediction accuracy of traditional coiling temperature prediction methods.

[0008] To solve the above technical problems, the present application provides the following technical solutions:

[0009] On the one hand, the present application provides a coiling temperature prediction method based on multi-level feature fusion under cloud edge end cooperation, which is used for coiling temperature prediction in hot continuous rolling industrial process, wherein the hot continuous rolling industrial process is divided into manufacturing execution layer, process control layer and real-time control layer from top to bottom; the method comprises:

[0010] The manufacturing execution layer is deployed on the cloud side, the process control layer is deployed on the edge side, and the real-time control layer is deployed on the end side;

[0011] The current strip process data of the process control layer and the coil set data of the manufacturing execution layer are collected, and the collected current strip process data and coil set data are preprocessed on the end side;

[0012] The feature extraction is performed on the preprocessed current strip process data on the edge side to obtain the bottom working condition information of the current slab.

[0013] The cloud side extracts features from the preprocessed steel coil setting data to obtain upper layer production scheduling setting information of the next slab;

[0014] The cloud side fuses features of a bottom layer working condition information of the current slab and the upper layer production scheduling setting information of the next slab to obtain multi-level fusion features, and realizes the coiling temperature prediction based on the multi-level fusion features.

[0015] Further, the current strip process data includes: actual value and set value of the finishing mill outlet thickness, actual value and deviation value of the finishing mill outlet width, finishing mill outlet speed, first intermediate speed, second intermediate speed, finishing mill outlet temperature, laminar cooling intermediate temperature, water tank water temperature, fast cooling section actuator temperature, fast cooling first section, fast cooling second section, and fast cooling eighteenth section upper and lower actuator water temperature, valve position, fast cooling section actuator pressure, and flow increment.

[0016] Further, the steel coil setting data includes: finishing mill temperature set value, thickness set value, width set value, coiling temperature set value, and actual coiling temperature of the previous steel coil.

[0017] Further, the collected current strip process data and steel coil setting data are preprocessed, including:

[0018] The current strip process data and the steel coil setting data are subjected to data interception and alignment operation;

[0019] The discrete variables in the current strip process data are subjected to time window interception and down-sampling processing to obtain multi-scale discrete data; wherein the discrete variables include the header switch state of the top and bottom of each zone.

[0020] Further, the data interception and alignment operation includes:

[0021] The collected data are subjected to slab tracking, i.e. interception and processing according to different slab numbers, and the processed data are stored according to slab numbers, specifically: first, the time interval of the same slab is intercepted according to the finishing mill outlet and coiling flag bit, then the position and time of the header passing through different sections are calculated according to the speed and time of the slab, and finally the data alignment in the length direction with an interval of 1 meter is realized based on the total length of the slab.

[0022] Further, the time window interception and down-sampling processing includes:

[0023] The discrete variables are subjected to time window interception, wherein the step length of the time window is set to 1, the width of the time window is w, and the number of the time window is n, and w and n are both preset values.

[0024] After the time window is intercepted, the time window is down-sampled to obtain multi-scale discrete data.

[0025] Further, the pre-processed current strip steel process data is subjected to feature extraction, including:

[0026] The GRU layer with the fusion attention mechanism is adopted to extract features from the current strip steel process data after the alignment operation, so as to extract a continuous feature vector of the previous coil at the process control layer; and the SlowFast Networks is adopted to extract features from the multi-scale discrete data, so as to extract a discrete feature vector of the previous coil at the process control layer.

[0027] Further, when the SlowFast Networks is adopted to extract features from the multi-scale discrete data, the transverse connection result of the second residual network in the SlowFast Networks is taken as the extracted features.

[0028] Further, the pre-processed coil setting data is subjected to feature extraction, including:

[0029] The SENet is adopted to extract features from the coil setting data after the alignment operation, so as to obtain a feature vector of the next coil at the manufacturing execution layer.

[0030] Further, the bottom layer working condition information of the current slab is fused with the upper layer production scheduling setting information of the next slab to obtain multi-level fusion features, and the coiling temperature prediction is realized based on the multi-level fusion features, including:

[0031] After the discrete feature vector of the previous coil at the process control layer is averaged in the channel and flattened, the discrete feature vector is spliced with the continuous feature vector of the previous coil at the process control layer to obtain a feature vector of the previous coil at the process control layer;

[0032] After the feature vector of the previous coil at the process control layer is averaged in the time step, the feature vector is spliced with a feature vector of the next coil at the manufacturing execution layer to obtain multi-level fusion features.

[0033] The multi-level fusion features pass through two fully connected layers to obtain a coiling temperature prediction value.

[0034] In still another aspect, the present application also provides an electronic device, which comprises a processor and a memory; wherein the memory stores at least one instruction, which is loaded and executed by the processor to realize the above method.

[0035] In still another aspect, the present application also provides a computer readable storage medium, which stores at least one instruction, which is loaded and executed by the processor to realize the above method.

[0036] The technical scheme provided by the application brings at least the following beneficial effects:

[0037] 1. The application provides an industrial process coiling temperature method fusing multi-level features. The method uses a GRU layer fusing an attention mechanism and a SlowFast Network to extract features of a hot continuous rolling process control layer, extract multi-scale features of process variables, and obtain state information such as slab bottom layer operation conditions. Then, higher level feature extraction is performed on the manufacturing execution layer, features of set data are extracted through an SENet to obtain slab top layer production scheduling information. Then, the multi-level features extracted by the manufacturing execution layer and the process control layer are fused through a full connection layer, and finally the prediction value of the coiling temperature is obtained.

[0038] 2. The application deploys the proposed prediction model in a cloud-edge-end collaborative framework for application verification. The manufacturing execution layer, the process control layer and the real-time control layer are respectively deployed on the cloud side, the edge side and the end side. The end side real-time control layer performs preprocessing work such as process data alignment and sliding time window; the edge side process control layer extracts features of strip steel process data to extract bottom layer working condition information and obtain process control layer features; the cloud side manufacturing execution layer extracts features of steel coil set data to extract top layer scheduling and production, and also receives features extracted by the edge side process control layer, and then fuses the multi-system level features of the manufacturing execution layer and the process control layer to finally realize real-time prediction of the coiling temperature. The effectiveness and real-time performance of the method are verified by using actual industrial data. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0040] Figure 1 is a schematic diagram of a hot continuous rolling process;

[0041] Figure 2 is a schematic diagram of a laminar cooling process;

[0042] Figure 3 is a framework diagram of a coiling temperature prediction method based on multi-level feature fusion provided by the embodiment of the application;

[0043] Figure 4 is a principle diagram of end side preprocessing provided by the embodiment of the application;

[0044] Figure 5 is a schematic diagram of data alignment process provided by the embodiment of the application;

[0045] Figure 6A coiling temperature prediction result schematic diagram provided by an embodiment of the present application is shown in FIG. 1.

[0046] Figure 7 A model deployment schematic diagram under a cloud-edge-end collaborative framework provided by an embodiment of the present application is shown in FIG. 2.

[0047] Figure 8 A system block diagram of an electronic device provided by an embodiment of the present application is shown in FIG. 3. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions and advantages of the present application clearer, the following will further describe the embodiments of the present application in detail with reference to the accompanying drawings.

[0049] First of all, it should be noted that in the embodiments of the present application, the words such as "exemplarily", "for example" and the like are used to represent as an example, illustration or explanation. Any embodiment or design scheme described as "exemplary" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "exemplarily" is intended to present the concept in a specific manner. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be either of the two.

[0050] First embodiment

[0051] The present embodiment provides a coiling temperature prediction method based on multi-level feature fusion under cloud-edge-end collaboration, which is used for coiling temperature prediction in hot continuous rolling industrial process. First, a complex industrial process (such as a hot continuous rolling industrial process) can be divided into a manufacturing execution layer, a process control layer and a real-time control layer from top to bottom. The process control layer receives the job plan of the manufacturing execution layer and the operation data of the process control layer, and respectively feeds back the job actual and operation instruction to them. Secondly, considering the efficiency of cloud-side calculation and the flexibility and real-time performance of edge-side calculation, the manufacturing execution layer, the process control layer and the real-time control layer are respectively deployed on the cloud side, the edge side and the end side. The end-side real-time control layer implements preprocessing work such as process data alignment and sliding time window; the edge-side process control layer extracts features from process data to extract bottom-layer working condition information and obtain process control layer features; the cloud-side manufacturing execution layer extracts features from steel coil setting data to extract top-layer scheduling production information, and also receives the process control layer features extracted by the edge-side process control layer, and then fuses the features of the multi-system level of the manufacturing execution layer and the process control layer, to finally realize coiling temperature prediction. The application of this method in the coiling temperature prediction of an actual hot continuous rolling production line has been verified: the average absolute error of the method in this paper is reduced by 53.01% compared with other traditional methods; the effectiveness and real-time performance of the method are verified by using actual industrial data.

[0052] The method can be implemented by an electronic device, specifically, the method comprises the following steps:

[0053] S1, respectively deploying the manufacturing execution layer, the process control layer and the real-time control layer at the cloud side, the edge side and the end side;

[0054] S2, collecting current strip process data of the process control layer and coil set data of the manufacturing execution layer, and preprocessing the collected current strip process data and coil set data at the end side;

[0055] S3, performing feature extraction on the preprocessed current strip process data at the edge side to obtain bottom layer working condition information of the current slab;

[0056] S4, performing feature extraction on the preprocessed coil set data at the cloud side to obtain upper layer production scheduling set information of the next slab;

[0057] S5, performing feature fusion on the bottom layer working condition information of the current slab and the upper layer production scheduling set information of the next slab at the cloud side to obtain multi-level fusion features, and realizing coil temperature prediction based on the multi-level fusion features.

[0058] It should be noted that the hot continuous rolling process is as shown in Figure 1 The process generally consists of the following units: heating furnace, rough rolling reversible mill, heat preservation cover, flying shear, descaling machine, finishing rolling mill train, laminar cooling, coiling machine train. The strip steel undergoes complex water cooling, air cooling heat exchange and internal heat conduction process during laminar cooling, has complex industrial characteristics such as drastic change of working condition, strong nonlinearity, time-varying parameters and difficult to accurately describe mathematical model, therefore a data-driven deep learning method is adopted to fully extract the features of variables of multiple time scales. As shown in Figure 2 The laminar cooling mainly controls the cooling speed and coiling temperature of the steel plate by changing the roller speed and the switching state and flow of the water curtain header. Therefore, the switching state of the header in the laminar cooling process has an irreplaceable influence on the coiling temperature. These strip process data reflect the bottom layer working condition information of the equipment end, and feature extraction on these data can reflect the equipment working state.

[0059] The multi-level framework of the industrial process is as shown in Figure 3 The main part is the real-time control layer, the process control layer and the manufacturing execution layer. The operation data of the bottom layer real-time control layer is uploaded to the process control layer, and the job performance is further uploaded to the manufacturing execution layer in this layer. The manufacturing execution layer formulates and issues a job plan to the process control layer according to the production control requirements, and then provides operation instructions for the bottom layer real-time control layer.

[0060] The framework of the coil temperature prediction method based on multi-level feature fusion is as shown in Figure 3As shown, the method makes full use of bottom layer working condition information and upper layer production scheduling relationship and other multi-level information. First, the method obtains data at different levels, including current strip process data obtained from the process control layer and next coil setting data obtained from the manufacturing execution layer. Then, feature extraction is performed on the current strip at the process control layer to obtain bottom layer working condition information. Then, manufacturing execution layer feature extraction is performed on the setting data of the top manufacturing execution layer to obtain top layer scheduling information. Then, the features extracted by the manufacturing execution layer are fused with the features extracted by the process control layer, and finally the coiling temperature prediction of the next coil is realized.

[0061] The above modules and techniques are integrated to obtain a coiling temperature prediction model that fuses multi-level information. The multi-level coiling temperature is beneficial for timely discovering potential problems and providing scheduling suggestions for the manufacturing execution layer, and is also beneficial for process parameter optimization of the process control layer, improving the safety and reliability of the process.

[0062] The data preprocessing and feature extraction operations are described in detail below.

[0063] 1. End side data preprocessing

[0064] The real-time control layer is deployed on the end side, mainly performing data alignment and sliding time window preprocessing operations, and the implementation principle is as shown in Figure 4 The data of the same slab in the hot continuous rolling process is continuous, and the data of different slabs is discontinuous. The coiling temperature prediction model established in this paper is mainly based on effective rolling data, so it is necessary to intercept the effective rolling data of different slabs and perform corresponding preprocessing to obtain part of the slab data. After collecting the field data, the data needs to be tracked according to the slab number, that is, the data is intercepted and processed according to different slab numbers, and the preprocessed data is stored according to the slab number. The number of slab tracking is m. Because the effective data of different cooling sections of the same slab is not aligned in the time dimension, it is necessary to align the data according to the length percentage. Data alignment first intercepts the time interval of the same slab according to the finish rolling outlet and the coiling flag, then calculates the position and time of the manifold passing through different sections according to the speed and time of the slab, and finally realizes the data alignment in the length direction with an interval of 1 meter based on the total length L of the slab, to obtain the data of the slab passing through each section of the manifold.

[0065] In order to fully extract the features of multi-time scale discrete data, time window operation is also needed. Discrete variables with width w and variable number d are intercepted, and n time windows are intercepted as a group, corresponding to continuous variables with variable number c at the last time of the time window, to obtain the final input form of the model.

[0066] 2. Side process data feature extraction

[0067] The GRU effectively alleviates the gradient vanishing and gradient explosion problems that the traditional RNN is prone to encounter when processing long-term dependencies. Meanwhile, in the complex task of processing discrete data feature extraction, the GRU has a more concise structure and faster training speed than the LSTM while maintaining performance. The GRU layer input is the multi-plate information tracking obtained by the The GRU unit can be represented by the following mathematical expression:

[0068] V t =σ(W v ·[H t-1 ,X t ]+b v ) (1)

[0069] R t =σ(W r ·[H t-1 ,X t ]+b r ) (2)

[0070]

[0071] where R t , V t and H t represent the reset gate, the update gate and the output gate, and H t-1 represent the current memory information and the previous hidden state, W v represents the weight matrix, b v represents the bias vector of the update gate, W r represents the weight matrix, b r represents the bias vector of the reset gate, W h represents the corresponding weight matrix, b h represents the bias vector, I represents a matrix with all elements being 1, represents the Hadamard product of two vectors, σ represents the sigmoid activation function, and tanh represents the hyperbolic tangent activation function.

[0072] After that, the attention mechanism is used to assign weights to different parts of the GRU layer output, highlighting more relevant information and helping the model focus on parts of the input sequence that are considered more important for the current task. K, Q, and V represent the key, query, and value in the attention mechanism, respectively. The input to the attention mechanism is composed of key-value pairs <K, V>. The calculation of the attention mechanism consists of three steps: first, the correlation between K and Q is calculated, then the weight coefficient of each K corresponding to V is obtained, and finally the weighted sum of the V value obtained by the weight coefficient is obtained to get the output of the attention. The attention mechanism weight coefficient calculation method is as follows:

[0073] θ=Q·KT (5)

[0074]

[0075] where θ is the correlation coefficient of Q and K; θ j is the attention weight coefficient of the jth sample; n represents the number of input key-value pairs. The output of the GRU layer is weighted and summed under the attention mechanism to obtain the feature vector z C , which is input as part of the feature fusion layer C The calculation is as follows:

[0076]

[0077] Further, the discrete data of the hot continuous rolling laminar cooling process has the characteristics of multiple time scales, and it is necessary to capture the characteristics of fast and slow changes. To this end, the SlowFast network is used to extract features of the discrete variable, different sampling frequency discrete data is obtained through the sliding time window, and the SlowFast network is used for multi-scale feature extraction, and the transverse connection is used for fusion. Specifically: the discrete data is taken and down-sampled as the input of the fast path and the slow path. The SlowFast Networks used in this paper are composed of two paths with different sampling frequencies for feature extraction and fusion. First, two high-frequency and low-frequency sequences are obtained through different sampling intervals and input into the network. Both paths use 3D Resnet convolutional neural network for spatio-temporal feature extraction. The slow path processes the input data at a low sampling rate to capture spatial features that change slowly over time, and the fast path processes the input data at a higher sampling rate to capture features that change faster. The extracted feature information in the two paths is fused through transverse connection.

[0078] Considering the complexity of the network and the characteristics of the data, the transverse connection result of the second residual network in the SlowFast network structure is used as the final feature extracted from the discrete variable. The discrete data is first input and sampled through the data layer. The data of the fast path taken by the time window in the slow path is down-sampled to obtain the fast path and the slow path input with a sampling frequency ratio of α, that is, the sampling density ratio of the two is α. Thus, an n x w x d sequence framework is generated. Then the generated sequence is input to the 3D convolution layer for feature extraction.

[0079] The multi-scale feature fusion after each stage is completed by a cross-connection operation. The information extracted by the fast path is fused into the slow path through cross-connection, which enables the slow path to perceive the information extracted by the fast path, and realizes the extraction and fusion of multi-scale features. Since the two paths in this paper have different time dimensions, the information obtained from the fast path needs to be transformed before being fused into the slow path. The output feature map of the slow path is represented as {n, w x d, c}, and the output feature map of the fast path is represented as {a x n, w x d, b x c}. In this paper, a time-step-based convolution is used for cross-connection: set the step length as a and control the size of the output feature map, perform 3D convolution on the information extracted from the fast path, and then perform a splicing operation with the information of the slow path.

[0080] First, in the two paths, the variable x F of the fast path and the variable x S of the slow path pass through convolution and pooling operations respectively to obtain F1 and S1, where:

[0081] S1 = maxpool (relu (conv (x S ))) (8)

[0082] F1 = maxpool (relu (conv (x F ))) (9)

[0083] Then, F1 obtained from the fast path is fused into S1 to obtain S1' through a cross-operation composed of conv3D, three-dimensional BatchNorm3D, and relu activation function, which satisfies:

[0084] S1' = [S1; relu (batchnorm (conv (F1))) ] (10)

[0085] In the second stage, S1' and F1 pass through a residual network composed of 3 groups of convolution operations to obtain S2 and F2, where:

[0086] S2 = res2 (S1') (11)

[0087] F2 = res2 (F1) (12)

[0088] Then, F2 is similarly fused into S2 to obtain S2' through cross-connection, where:

[0089] S'2 = [S2; relu (batchnorm (conv (F2))) ] (13)

[0090] Finally, S'2 is channel-averaged and flattened to obtain the final features extracted by the SlowFast network.

[0091] 3. Cloud-side setting data feature extraction

[0092] There is abundant multi-level data in industrial processes, including setting data of the manufacturing execution layer, process parameters of the process control layer, and process data of the real-time control layer, etc. These data are coupled with each other in a multi-level manner, and therefore, it is necessary to fully integrate multi-level information to improve the prediction accuracy. The squeeze-and-excitation network (SENet) enhances the ability of convolution. The neural network perceives different features by adaptively adjusting the weight of each channel, and this mechanism enables the network to pay more attention to the feature channels with rich information, which is beneficial to improve the accuracy and efficiency of feature extraction of the manufacturing execution layer.

[0093] Based on this, the embodiment uses a model based on SENet to extract features of the top manufacturing execution layer data. The model can automatically learn by learning a weight coefficient on each channel and weighting it respectively, enhancing the model's attention to important channels and weakening the attention to unimportant channels, so as to more effectively utilize the input features. The slab setting information is obtained by convolution and batch normalization to satisfy:

[0094] E = batchnorm(conv(x E )) (14)

[0095] SENet mainly includes three processes of squeezing, excitation, and re-scaling. In the squeezing process, the feature map in the input channel is compressed into a feature vector through global average pooling. In the excitation process, the weight of each channel is learned by using a fully connected layer and a nonlinear activation function to capture the relationship between channels. In the re-scaling, the weighted output features are calculated by multiplying the weight of each channel with the features on the original channel. E' is obtained by the squeezing and excitation block of E, satisfying:

[0096] E' = E · sigmoid(W1 · relu(W · GAP(E))) (15)

[0097] In the formula, GAP represents global average pooling, W and W1 represent two fully connected layers, and relu and sigmoid represent activation functions. Then, two groups of convolution, batch normalization, relu activation function, and maximum pooling are performed to obtain the feature z E :

[0098] z E = maxpool(relu(batchnorm(conv(maxpool(relu(batchnorm(conv(E')))))))(16)

[0099] 4. Cloud-side multi-level feature fusion and parameter learning

[0100] The multi-level feature fusion process uses a continuous feature vector of a coil on the process control layer Discrete feature vector and a feature vector z of a next coil on the manufacturing execution layer E are fused. The discrete features are channel-averaged and flattened and then concatenated with the continuous features to obtain Z L , and then Z L is time-step averaged and concatenated with z E to obtain the input to the fully connected layer, denoted as Z S , which satisfies:

[0101]

[0102] Z S =[Z L ,z E ] (18)

[0103] The input Z S is passed through two fully connected layers to obtain the model's prediction of the coiling temperature. Assuming that the output dimension of the first fully connected layer is λ3, the prediction of the coiling temperature is denoted as , which is calculated as follows:

[0104]

[0105] where relu is the relu activation function; and are weights; and ε1 and ε2 are biases.

[0106] The model parameters are trained using backpropagation, and the model is trained using the optimization algorithm of Adaptive Moment Estimation (Adam), with the mean absolute error used as the loss function, which is calculated as:

[0107]

[0108] Based on the above, the coiling temperature prediction method based on multi-level feature fusion of the present embodiment can be summarized as follows:

[0109] Fully extract multi-level features. Existing methods mostly only consider the information of the manufacturing execution layer when predicting the coiling temperature, while the method of the present embodiment not only considers the information of the manufacturing execution, but also considers the information of the process control layer. Specifically, the method of the present embodiment extracts the bottom-level working condition information of the current slab by performing process control layer feature extraction on the current strip process data of the process control layer, and extracts the upper-level production scheduling setting information of the next slab by performing feature extraction on the coil setting data of the manufacturing execution layer. Then, the multi-level features extracted in the above feature extraction process are fused, and the multi-level information is used for CT prediction, which more comprehensively considers the influencing factors of the slab temperature, and is conducive to achieving higher precision prediction.

[0110] Cloud-edge-end distributed deployment. Considering that there is a certain time interval between adjacent slabs in actual industrial processes, the feature extraction of the current strip of the process control layer and the feature extraction of the next coil of the manufacturing execution layer can be deployed on the edge side and the cloud side respectively, fully utilizing the advantages of cloud-edge computing. On the one hand, since the edge side computing has flexibility and real-time performance, the process data feature extraction with high real-time requirement is deployed to the edge side for extracting the bottom-level working condition information. On the other hand, since the cloud side computing is efficient, the coiling temperature requiring multi-level feature fusion is deployed on the cloud side. The multi-level structure of the method is conducive to distributed deployment, thereby ensuring the effectiveness and real-time performance of the prediction. In addition, deploying data preprocessing on the real-time control layer of the end side is conducive to reducing the burden and cost of edge side computing, and speeding up the speed of multi-level feature extraction.

[0111] Consider multi-scale information of process data. The method of the present embodiment extracts information of continuous variables and discrete variables through GRU and SlowFast respectively in view of the existence of continuous variables and discrete variables in actual processes. The GRU in the method combined with attention mechanism can capture long-term dependencies and focus on key information at the same time. GRU reduces the number of parameters while capturing long-term dependencies through a simplified gating mechanism, and the attention mechanism is conducive to focusing on key information at specific time points or time periods, thereby improving the accuracy of feature extraction. The SlowFast network in the method adopts a double-path structure to capture field time scale static information and short time scale dynamic changes in data through fast and slow paths. The multi-scale features are fused through horizontal connection, and the fusion mechanism is conducive to more comprehensively capturing multi-scale information in discrete data.

[0112] Next, the implementation process of the method and its effect are described with actual application examples.

[0113] 1. Data Description

[0114] In this paper, the actual process data of 50 steel coils in a steel plant are taken as an example, including the data of 6 steel grades for training and testing, of which 40 steel coils are used for training and 10 steel coils are used for testing. Before that, effective data interception, data alignment and sliding time window operation need to be performed.

[0115] The data derived from each steel coil has time sequence, and the rolling of different slabs is not continuous in time, so it is necessary to intercept effective rolling data and pre-process. The data interception and alignment process is shown in Figure 5 According to the time interval of the two key signals "FDT_ON" and "CT_ON" being 1, the data of the slab in the time period of the laminar cooling process is intercepted, and the data of different slabs is stored according to the slab number variable "COIL". Secondly, the effective rolling data of different header sections of each steel coil is not aligned in time dimension, so it is necessary to align the data according to the length position. Specifically, the data processing process includes: first, the data of a piece of steel needs to be intercepted according to the finishing mill outlet and coiling flag, that is, the earliest time and the latest time of the corresponding variable of a piece of steel are included in the time interval of the intercepted data. Second, according to the different characteristics of different header sections of different variables, the variables of each header section are grouped, the position of the slab in the laminar cooling process is calculated through the variables "FDT Strip Transfer Length" and "CT Strip Transfer Length", and the effective data of different header sections is intercepted according to the actual physical distribution of the laminar cooling header. Finally, according to the total length of the slab and the length of different header sections, the data from different header sections is aligned according to the length of 1 meter between each piece of steel. Repeat the above operation until all the data of the slab in the time range is processed.

[0116] In order to extract multi-scale features, the discrete variables also need to be intercepted by time window, the step of time window is set to 1, the width of time window is w, and the number of time window is n, then the time window is down-sampled to get multi-scale discrete data as the input of SlowFast network.

[0117] In order to establish a data-driven CT prediction model, the input variables of the model need to be selected. Combined with the knowledge of heat transfer principle related to CT mechanism and the temperature drop model of laminar cooling, the formulas are shown in (21) ~ (22).

[0118] T(t) = T h +(T FDT -T h )e -λt (21)

[0119]

[0120] In the formula, T(t) is the real-time temperature of the strip surface, ℃; TFDT is the finishing mill exit temperature, ℃; λ is the model factor; t is the strip cooling time, s; T e is the coiling temperature, ℃; T W is the cooling water temperature, ℃; F h is the finishing mill exit thickness, mm; is the empirical parameter given by the field equipment.

[0121] According to the existing process variables of the actual production process, the input variables are divided into process control level variables and manufacturing execution level variables. The continuous input variables of the process control layer are selected from the actual value and the set value of the finishing mill exit thickness, the actual value and the deviation value of the finishing mill exit width, the finishing mill exit speed, the intermediate speed 1 and the intermediate speed 2, the finishing mill exit temperature, the laminar cooling intermediate temperature, the water tank water temperature, the fast cooling section actuator temperature, the fast cooling 1, 2 and 18 section upper and lower actuator water temperature, the valve position, the fast cooling section actuator pressure and the flow increment. The discrete input variables of the process control layer are selected from the top and bottom header switch states of the 1-20 area. The set variables of the manufacturing execution level include the finishing mill temperature setting, the thickness setting, the width setting, the coiling temperature setting and the actual coiling temperature of the previous coil. Therefore, the input of the final CT model includes 110 continuous variables, denoted as 98 discrete variables, denoted as set variables The input variables of the multi-level prediction model are shown in Table 1.

[0122] Table 1 Input variables of the multi-level prediction model

[0123]

[0124]

[0125] 2. Model training and parameter determination

[0126] 2.1 Evaluation index

[0127] In this paper, the mean absolute error (MAE), the root mean square error (RMSE) and the mean absolute percentage error (MAPE) are used as the evaluation index of the prediction performance of the model:

[0128]

[0129] In the formula, is the predicted value of the coiling temperature, y i is the actual value of the coiling temperature, n is the number of prediction samples.

[0130] 2.2 Model parameter determination

[0131] The main parameters of the method include the learning rate γ1 of the edge-side process control layer feature extraction part, the output dimension λ1 of the GRU layer, the width w of the time window, the number n of the time window, the learning rate γ2 of the cloud-side manufacturing execution layer feature extraction part, and the output dimension λ3 of the multi-level feature fusion part. These parameters all have an impact on the performance of the model. The values of the parameters are determined through experiments, as shown in Tables 2 to 6.

[0132] Table 2 Prediction results of different γ1

[0133]

[0134]

[0135] After multiple experiments, the learning rate γ1 of the edge-side process control layer feature extraction part is taken as [0.01, 0.005, 0.001, 0.0005, 0.0001], and the prediction results of the coiling temperature prediction model are shown in Table 2. When γ1 is taken as 0.0005, the performance indicators are the smallest, and the prediction effect is better. Finally, γ1 = 0.0005 is determined. The output dimension λ1 of the GRU layer is taken from [32, 48, 64, 80, 96, 112, 128], and the performance indicators of the prediction results are shown in Table 3. When λ1 is taken as 96, the prediction performance is optimal, and finally λ1 = 96 is determined.

[0136] Table 3 Prediction results of different λ1

[0137]

[0138] The width w and the number n of the time window determine the input shape of the edge-side process control layer multi-scale feature extraction, and have a great influence on the effect of feature extraction. Table 4 shows the model performance indicators under different w and n, and finally n = 16 and w = 16 are determined.

[0139] Table 4 Prediction results of different n and w

[0140]

[0141] The learning rate γ2 of the cloud-side manufacturing execution layer feature extraction part is taken from [0.01, 0.005, 0.001, 0.0005, 0.0001], and the results of the coiling temperature prediction model are shown in Table 5. When γ2 is taken as 0.0001, the performance indicators are the smallest, the prediction effect is better, and finally γ2 = 0.0001 is determined. The output dimension λ3 of the multi-level feature fusion full connection layer is taken from [32, 48, 64, 80, 96, 112, 128], and the performance indicators of the prediction results are shown in Table 6. Finally, λ3 = 80 is determined.

[0142] Table 5 Prediction results of different γ2

[0143]

[0144] Table 6 Prediction results of different λ3

[0145]

[0146] 2.3 Experimental results and analysis

[0147] Using the above model setting parameters for training, this paper selects MAE as the loss function for back propagation training of the model. The coiling temperature prediction results are shown in Figure 6 As can be seen from the figure, under the current model parameter setting, the fitting degree of the coiling temperature prediction value based on multi-level feature fusion and the actual value of the coiling temperature is good, the deviation is small, and the prediction value conforms to the change trend of the actual value to a certain extent. At the same time, the prediction effect based on multi-level feature fusion is better than that without fusing the process control layer features, which shows that fusing multi-level information is beneficial to improving the prediction accuracy of the coiling temperature, and verifies the effectiveness of the method described in this paper.

[0148] The prediction results of the method based on multi-level feature fusion proposed in this paper are compared with the traditional method, and the comparison is shown in Table 7. In the table, each comparison method does not fuse multi-level features. The prediction results of the method proposed in this paper have higher prediction accuracy than the prediction results without fusing multi-level features, with lower MAE, RMSE and MAPE. The MAE of the prediction model proposed in this paper is 2.6560 lower than that of the comparison model on average. This reflects that feature extraction and fusion of process control layer features can more fully extract multi-level information of process control layer and manufacturing execution layer, and improve the prediction accuracy of CT.

[0149] Table 7 Analysis of coiling temperature comparison experiment results

[0150]

[0151] 3. Real-time application of cloud-edge-end cooperation

[0152] 3.1 Description of cloud-edge-end framework

[0153] The information flow of real data from the hot rolling process within the cloud edge framework is as follows: Figure 7 As shown, cloud-side servers have higher resource allocation than edge-side servers. Edge-side servers handle tasks with high real-time requirements, while cloud-side servers handle tasks with high memory and computing performance requirements. Preprocessing operations such as data alignment are performed on the edge side, after which real-time process data is uploaded to the edge's real-time database. The edge side retrieves the uploaded strip steel real-time process data from the real-time database for feature extraction and uploads the extracted features to the cloud side. The cloud side retrieves manufacturing execution layer data from a relational database and receives the features uploaded from the edge side for coiling temperature prediction.

[0154] Leveraging the flexibility and real-time performance of edge computing and the efficiency of cloud computing, the winding temperature prediction model used in this paper is deployed on platforms such as... Figure 7 In the cloud-edge-device collaborative framework shown, multi-level feature extraction is performed on the prediction model at both the cloud and edge sides. At the edge, pre-processed data is first acquired and uploaded in real-time. Then, a pre-trained model is loaded to perform multi-scale feature extraction on the current steel coil, obtaining features containing underlying operational status information, which are then uploaded to the cloud. The cloud retrieves historical steel coil data from a relational historical database, performs feature extraction to obtain features containing top-level scheduling and production information, and then performs multi-level feature fusion with the features uploaded from the edge to finally obtain the predicted coiling temperature for the next steel coil.

[0155] 3.2 Real-time prediction of winding temperature

[0156] The winding temperature prediction method based on multi-level feature fusion is deployed in a cloud-edge-device collaborative framework. This framework utilizes a PLC to upload data at the edge, and a server at the edge receives data from the PLC and uploads it to a real-time database. The edge server then uploads the real-time data to the real-time database of the edge server via TCP communication. Leveraging the efficiency of cloud computing and the flexibility and real-time performance of edge computing, feature extraction and multi-level feature fusion prediction for the manufacturing execution layer are performed at the cloud, while feature extraction for the process control layer is performed at the edge.

[0157] Table 8 lists the sampling interval and average time cost of the cloud-edge-device collaborative framework. On the cloud side, the average time from the end of the previous coil to the completion of the next coil winding temperature prediction is 11.9 s, which is less than the 90 s time interval between adjacent coils, reflecting the feasibility of real-time winding temperature prediction. On the edge side, the feature extraction time for one sample is 7.2 ms, which is less than the 500 ms sampling interval, ensuring real-time prediction capability. On the device side, the average data preprocessing time is 58.4 ms. Based on the above, the feasibility and real-time performance of the proposed method deployed under the cloud-edge-device collaborative framework have been verified.

[0158] Table 8 Sampling interval and average time length under cloud-edge-end collaborative framework

[0159]

[0160] 4. Conclusion

[0161] In order to improve the prediction accuracy of coiling temperature in hot rolling process, this paper proposes a coiling temperature prediction method based on multi-level feature fusion. In the process control layer feature extraction part of the method, GRU and SlowFast Networks are used for multi-scale feature extraction to capture the long-term dependence of continuous variables and the multi-scale information of discrete variables, and to obtain the current running condition of the bottom equipment end steel coil. Then in the manufacturing execution layer feature extraction part of the method. Through SEnet, the scheduling and production information of the next steel coil in the top manufacturing execution layer is extracted, and then the multi-level feature fusion prediction is performed with the features extracted in the process control layer. Finally, the high-precision coiling temperature prediction result of the next steel coil is realized.

[0162] This paper uses 50 steel coil data for training and testing. After data alignment and sliding time window preprocessing, the samples required for model input are obtained. Then the hyperparameters required by the model are determined using grid search algorithm. The prediction accuracy of the method described in this paper is compared with that of SVM, RF, XGB, GBRT, TCN, LSTM and GRU prediction methods. The performance of the model prediction is revealed by calculating the performance indicators such as MAE, RMSE and MAPE, and the prediction error results of each method are analyzed and compared. The experimental results show that the prediction model based on multi-level feature fusion proposed in this paper has good prediction performance. In addition, the proposed prediction model is deployed in the cloud-edge-end collaborative framework for application verification. The manufacturing execution layer, process control layer and real-time control layer are deployed in the cloud side, edge side and end side respectively. The end side realizes the preprocessing work such as data alignment and sliding time window; the edge side extracts the bottom working condition information features; the cloud side extracts the top scheduling and production information, and also receives the process control layer features extracted by the edge side process control layer, and then fuses the multi-system level features of the manufacturing execution layer and the process control layer, finally realizes the real-time prediction of the coiling temperature. The actual hot rolling process data is used to verify the effectiveness and real-time performance of the method.

[0163] Second embodiment

[0164] The present embodiment provides an electronic device, such as Figure 8As shown, the electronic device includes a processor and a memory; wherein the processor and the memory can be connected through a communication bus; the memory stores at least one instruction, which is loaded and executed by the processor to implement the method of the first embodiment. In addition, the electronic device can also include a transceiver, and the processor and the transceiver can be connected through a communication bus, and the transceiver is used to communicate with other devices.

[0165] Next, the method will be described in detail in combination with Figure 8 The various components of the electronic device will be described in detail:

[0166] The processor is the control center of the electronic device, and the electronic device can include multiple processors, each of which can be a single-CPU or a multi-CPU. The processor here can be a processor or a general term for multiple processing elements. For example, the processor is one or more central processing units (CPU), which can also be other general-purpose processors, application specific integrated circuits (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application, such as one or more microprocessors (digital signal processors, DSP), or one or more field programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor can perform various functions of the electronic device by running or executing software programs stored in the memory and calling data stored in the memory.

[0167] In a specific implementation, as an embodiment, the processor can include one or more CPUs, such as the CPU0 and CPU1 shown in Figure 8 of course, this is only an exemplary description.

[0168] The memory is used to store software programs for implementing the solutions of the present application, and is controlled by the processor to execute, and the specific implementation can refer to the above-mentioned method embodiments, which will not be described here.

[0169] Optionally, the memory may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory may be integrated with the processor or may exist independently, and may be accessed through the interface circuit of the electronic device (…). Figure 8 (Not shown in the image) is coupled to the processor; however, this embodiment of the invention does not impose specific limitations on this.

[0170] The transceiver may include a receiver and a transmitter. Figure 8 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function. The transceiver can be integrated with the processor or exist independently, and is connected through the interface circuit of the electronic device (…). Figure 8 (Not shown in the image) is coupled to the processor, and this embodiment of the invention does not specifically limit this.

[0171] In addition, it should be noted that, Figure 8 The structure of the electronic device shown is not intended to limit the device. Actual devices may include more or fewer components than shown, or combine certain components, or have different component arrangements. Furthermore, the technical effects achieved by this electronic device when performing the method of the first embodiment described above can be referenced to the technical effects described in the first embodiment; therefore, they will not be repeated here.

[0172] Third Embodiment

[0173] This embodiment provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the method of the first embodiment described above. The computer-readable storage medium may be a ROM, random access memory, CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc. The instruction stored therein can be loaded and executed by a processor in a terminal.

[0174] Moreover, it should be noted that the present application can be provided as a method, an apparatus, or a computer program product. Therefore, the embodiments of the present application can take the form of an entirely or partially hardware embodiment, an entirely or partially software embodiment, or an embodiment combining software and hardware aspects. Furthermore, when implemented in software, the embodiments of the present application can take the form of a computer program product including one or more computer program instructions embodied in a computer-usable storage medium. The computer program instructions can be stored in a computer-usable storage medium, or transmitted from one computer-usable storage medium to another, for example, from a website site, a computer, a server, or a data center to another website site, a computer, a server, or a data center through a wired (for example, infrared, wireless, microwave, or the like) manner. The computer-usable storage medium can be any available medium or a collection of medium(s) accessible by a computer or a data storage device such as a server, a data center, or the like, including one or more of a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid state disk.

[0175] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, an embedded processor, or other programmable data processing terminal devices to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal devices generate a device that implements the flow Figure 1 The flow or flows and / or blocks in the flowcharts and / or block diagrams Figure 1 The device that implements the function specified in the flow or flows and / or blocks.

[0176] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing terminal devices to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including instruction devices that implement the flow Figure 1 The flow or flows and / or blocks in the flowcharts and / or block diagrams Figure 1the functions specified in the individual block or blocks. Such computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable devices to generate a computer-implemented process, thus the instructions executed on the computer or other programmable devices provide a process for implementing the functions specified in the flowchart block(s) or block(s). Figure 1 the functions specified in the individual block or blocks. Such computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable devices to generate a computer-implemented process, thus the instructions executed on the computer or other programmable devices provide a process for implementing the functions specified in the flowchart block(s) or block(s). Figure 1 the functions specified in the individual block or blocks. Such computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable devices to generate a computer-implemented process, thus the instructions executed on the computer or other programmable devices provide a process for implementing the functions specified in the flowchart block(s) or block(s).

[0177] It should also be noted that, in the present document, the terms such as first and second, and the like, are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between such entities or operations. The terms "comprising", "containing", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements does not only include those elements, but can also include other elements not expressly listed, or other elements inherent to such process, method, article, or terminal device. Without further limitation, an element defined by the phrase "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or terminal device that includes the element. In addition, the term "and / or" is merely a descriptive association relationship of associated objects, which means that there can be three relationships, for example, A and / or B, which means that there can be three cases, namely, A alone, A and B together, and B alone, where A and B can be singular or plural. In addition, the character " / " in the present document generally represents an "or" relationship between the front and rear associated objects, but can also represent an "and / or" relationship, which can be understood in the context before and after. "At least one" means one or more, and "multiple" means two or more. "At least one of the following" or similar expressions means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c can mean a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, and c can be single or multiple.

[0178] In addition, it can be understood that in various embodiments of the present application, the size of the sequence number of the above processes does not mean the order of execution, and the execution order of the processes should be determined by their functions and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0179] Those skilled in the art can appreciate that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or in a combination of computer software and electronic hardware. Whether the functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0180] In several embodiments provided by the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely schematic, for example, the division of functional modules / units is merely a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another device, or some features can be omitted or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed units can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms. The units described as separate components can be or can not be physically separated, and the components displayed as units can be or can not be physical units, that is, can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment. In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present, or two or more units can be integrated in one unit.

[0181] If the method is implemented in the form of a software function unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the embodiments of the present application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0182] Finally, it should be noted that the above description is only the preferred embodiment of the application, it should be pointed out that although the preferred embodiment of the application has been described, for those skilled in the art, once the basic creative concept of the application is known, several improvements and refinements can be made without departing from the principles of the application, and these improvements and refinements should also be considered as the protection scope of the application. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the application.

Claims

1. A cloud edge collaboration based multi-level feature fusion based coiling temperature prediction method for coiling temperature prediction in hot rolling industrial processes, wherein, The hot continuous rolling industrial process is divided into a manufacturing execution layer, a process control layer and a real-time control layer from top to bottom; and the method comprises: The manufacturing execution layer is deployed on the cloud side, the process control layer is deployed on the edge side, and the real-time control layer is deployed on the end side; Current strip process data of the process control layer and coil set data of the manufacturing execution layer are collected, and the collected current strip process data and coil set data are preprocessed on the end side, including: data interception and alignment operation on the current strip process data and the coil set data; time window interception and downsampling processing are performed on the discrete variables in the current strip process data to obtain multi-scale discrete data; wherein the discrete variables include the state of the header switch at the top and bottom of each zone; Feature extraction is performed on the preprocessed current strip process data on the edge side to obtain the bottom layer working condition information of the current slab, including: a GRU layer with a fusion attention mechanism is used to extract features from the current strip process data after the alignment operation to extract a continuous feature vector of the previous coil in the process control layer; and SlowFast Networks are used to extract features from the multi-scale discrete data to extract a discrete feature vector of the previous coil in the process control layer; Feature extraction is performed on the preprocessed coil set data on the cloud side to obtain the upper layer production scheduling setting information of the next slab, including: feature extraction is performed on the coil set data after the alignment operation by using SE-Net to obtain a feature vector of the next coil in the manufacturing execution layer; The bottom layer working condition information of the current slab and the upper layer production scheduling setting information of the next slab are fused on the cloud side to obtain a multi-level fusion feature, and based on the multi-level fusion feature, the coiling temperature prediction is realized, including: the discrete feature vector of the previous coil in the process control layer is averaged in the channel and flattened, and then spliced with the continuous feature vector of the previous coil in the process control layer to obtain a feature vector of the previous coil in the process control layer; the feature vector of the previous coil in the process control layer is averaged in the time step and then spliced with the feature vector of the next coil in the manufacturing execution layer to obtain a multi-level fusion feature; the multi-level fusion feature passes through two fully connected layers to obtain a coiling temperature prediction value. 2.The cloud-edge collaboration based coiled temperature prediction method according to claim 1, wherein, The current strip process data includes: actual value and set value of the exit thickness of the finishing mill, actual value and deviation value of the exit width of the finishing mill, exit speed of the finishing mill, first intermediate speed, second intermediate speed, exit temperature of the finishing mill, intermediate temperature of the laminar cooling, water temperature of the water tank, temperature of the fast cooling actuator, water temperature of the first, second and eighteenth sections of the fast cooling, valve position, pressure of the fast cooling actuator, and flow increment. 3.The cloud-edge collaboration based coiled temperature prediction method according to claim 1, wherein, The coil set data includes: set value of the finishing mill temperature, set value of the thickness, set value of the width, set value of the coiling temperature, and actual coiling temperature of the previous coil. 4.The cloud-edge collaboration based coiled temperature prediction method according to claim 1, wherein, The data interception and alignment operation includes: The collected data is tracked by slabs, that is, the data is intercepted and processed according to different slab numbers, and the processed data is stored according to the slab numbers, specifically: first, the time interval of the same slab is intercepted according to the finishing outlet and the coiling flag bit, then the position and time of the manifold passing through different sections are calculated according to the speed and time of the slab, and finally the data alignment in the length direction with an interval of 1 meter is realized based on the total length of the slab. 5.The cloud-edge collaboration based coiled temperature prediction method according to claim 1, wherein, The time window interception and down-sampling processing comprises: intercepting the discrete variable by a time window, wherein the step length of the time window is set to 1, the width of the time window is w, and the number of the time window is n, w and n are both preset values; after completing the time window interception, the time window is subjected to down-sampling processing to obtain multi-scale discrete data. 6.The cloud-edge collaboration based coiled temperature prediction method according to claim 1, wherein, When SlowFast Networks are used to extract features from the multi-scale discrete data, the transverse connection result of the second residual network in the SlowFast Networks is taken as the extracted features.

Images