CN118070040A - Steel mill data acquisition method and device, electronic equipment and storage medium - Google Patents

Abstract

The invention relates to a steel mill data acquisition method, a device, electronic equipment and a storage medium, wherein the method is characterized in that time sequence data of the current production process of a steel mill are acquired, the time sequence data are input into a feature extraction model to obtain nonlinear data features, the feature extraction model is obtained by training a pre-constructed extraction model through sample data, a dynamic threshold value is determined based on a dynamic density estimated value of the nonlinear data features, and data acquisition frequency is adjusted according to a comparison result of the dynamic density estimated value and the dynamic threshold value so as to acquire data of the steel mill through the adjusted data acquisition frequency; the invention can adjust the data acquisition frequency according to the comparison result of the dynamic density estimation value and the dynamic threshold value, and solves the technical problems of excessive acquisition or insufficient acquisition quantity caused by adopting fixed frequency to acquire data of a steel mill.

Description

A steel plant data collection method, device, electronic equipment and storage medium

Technical Field

The present invention relates to the technical field of data analysis, and in particular to a steel mill data collection method, device, electronic equipment and storage medium.

Background technique

With the rapid development of Industry 4.0 and intelligent manufacturing, steel mills, as an important basic industrial production link, have an increasing demand for real-time data collection and accurate analysis. The steelmaking production process involves multiple complex steps, such as smelting, continuous casting, rolling, etc., and each step is accompanied by a large amount of data. These data not only reflect the status of the production process, but are also closely related to key indicators such as equipment health, production efficiency, and product quality.

Related data collection and analysis methods are often based on fixed collection frequencies and predefined models, which to some extent limits the depth and breadth of data analysis and presents the following technical problems: (1) Since the production environment of a steel plant is highly dynamic and uncertain, collecting steel plant data at a fixed collection frequency results in insufficient collection when the data changes dramatically and excessive collection when the data changes slowly, which wastes resources and may miss key information; (2) It is difficult to capture deep patterns and features in steel plant data, especially features at different time scales; (3) Due to the complexity of the steelmaking production process, the data often contains a large amount of noise and interference. There may be a lack of effective identification and processing methods for the interference part of the data, resulting in interference in the data analysis results and low accuracy.

Summary of the invention

In view of the above-mentioned shortcomings of the prior art, the present application provides a steel plant data collection method, device, electronic equipment and storage medium to solve the above-mentioned technical problems.

The present invention provides a steel mill data collection method, which comprises: acquiring time series data of a current production process of the steel mill; inputting the time series data into a feature extraction model to obtain nonlinear data features, wherein the feature extraction model is obtained by training a pre-constructed extraction model with sample data, wherein the pre-constructed extraction model comprises a time series analysis layer for capturing structural features, a self-attention mechanism layer for identifying abnormal structural features, and a nonlinear transformation layer for extracting nonlinear data features, wherein the sample data comprises all or part of the time series data of the historical production process of the steel mill; determining a dynamic threshold based on a dynamic density estimation value of the non-linear data feature; and adjusting a data collection frequency according to a comparison result of the dynamic density estimation value with the dynamic threshold value, so as to collect data from the steel mill at the adjusted data collection frequency.

In one embodiment of the present invention, after collecting data from the steel mill through the adjusted data collection frequency, the steel mill data collection method includes: constructing a multi-level data vector field based on the collected steel mill data, and the multi-level data vector field is used to extract data features at different levels; inputting the multi-level data vector field into a preset interference identification function to obtain interference data; and removing the interference data from the steel mill data to obtain pure steel mill data.

In one embodiment of the present invention, the process of adjusting the data acquisition frequency according to the comparison result of the dynamic density estimation value and the dynamic threshold value includes: obtaining the current data acquisition frequency of the time series data;

If the dynamic density estimation value is less than or equal to the dynamic threshold, the data feature corresponding to the dynamic density estimation value is determined to be a normal data feature, and the current data collection frequency is adjusted to a first data collection frequency, which is less than the current data collection frequency; if the dynamic density estimation value is greater than the dynamic threshold, the data feature corresponding to the dynamic density estimation value is determined to be an abnormal data feature, and the current data collection frequency is adjusted to a second data collection frequency, which is greater than the current data collection frequency.

In one embodiment of the present invention, if the pre-constructed extraction model also includes a multi-scale analysis layer for extracting data features of different scales, the pre-constructed extraction model is trained by sample data to obtain a feature extraction model, including: inputting the sample data into the time series analysis layer to obtain structural features; inputting the structural features into the self-attention mechanism layer to obtain abnormal structural features; inputting the abnormal structural features into the nonlinear transformation layer to obtain nonlinear data features; inputting the nonlinear data features into the multi-scale analysis layer to obtain data features of different scales; constructing a loss function based on the data features of different scales and the time series data, and updating the parameters of the time series analysis layer, the parameters of the self-attention mechanism layer, the parameters of the nonlinear transformation layer, and the parameters of the multi-scale analysis layer with the goal of minimizing the loss function; combining the time series analysis layer after parameter update, the self-attention mechanism layer after parameter update, the nonlinear transformation layer after parameter update, and the multi-scale analysis layer after parameter update to obtain the feature extraction model.

In one embodiment of the present invention, the expression of the time series analysis layer is: Among them, ts(x) represents the structural characteristics of time series data, x represents time series data, sigm represents activation function, W _t represents weight, b _t represents bias, β represents weight coefficient of Gaussian integral, and T represents transposition; the expression of the self-attention mechanism layer is:/> Among them, Q, K, V represent the query matrix, key matrix and value matrix in the self-attention mechanism layer respectively, a(ts) represents the abnormal structural feature, T represents the transposition, d _k represents the dimension of the key, and ts represents the structural feature; the expression of the nonlinear transformation layer is: Wherein, f(a) represents nonlinear data features, _Wf represents the weight matrix of the nonlinear transformation layer, _bf represents the bias vector of the nonlinear transformation layer, μ represents the regularization coefficient, a represents the weight coefficient of the sine function, _Wr represents the regularized weight matrix, and a represents the abnormal structure feature; the expression of the multi-scale analysis layer is:/> Wherein, m(f) represents data features of different scales, _Wi represents the weight matrix of the i-th scale, _σi represents the standard deviation of the i-th scale, and f represents nonlinear data features; the expression of the loss function is:/> Wherein, L represents the loss function, μ ₁ and μ ₂ are both regularization coefficients, _hi represents the state of the neuron at the i-th scale, F represents the Frobenius norm, _Wi represents the weight matrix of the i-th scale, and x represents the time series data; the calculation formula of the dynamic density estimate is: Wherein, _dt (f) represents the dynamic density estimation value of the nonlinear data feature f at time t, N represents the number of nonlinear data features, K represents the space-time kernel function, which is used to measure the similarity between nonlinear data features in space and time, u represents the spatial distance between nonlinear data features, ω represents the time attenuation function, Δt represents the time difference between any time _t0 of the nonlinear data feature and the current time t; the calculation formula of the dynamic threshold is: Where Φ represents the dynamic threshold,/> represents the mean of the dynamic density estimate _dt of the nonlinear data feature, θ represents a constant, and σ represents the standard deviation of the Gaussian kernel function.

In one embodiment of the present invention, the process of constructing a multi-level data vector field based on the collected steel mill data includes: dimensional transformation of the steel mill data to obtain a data point set; based on the data point set, calculating a data persistence graph; based on the data persistence graph, calculating the multi-level data vector field.

In one embodiment of the present invention, the expression of the data point set is: P(t)=M(U)=exp(-k ₁ t ² )U·sin(k ₂ t)+k ₃ tU, wherein k ₁ represents the time attenuation factor, k ₂ represents the periodic adjustment factor, t represents time, k ₃ represents the linear time influence factor, U represents the steel plant data, M(U) represents the mapping function for dimensional transformation of the steel plant data, and P(t) represents the data point set; the expression of the data persistence graph is: Wherein, λ represents the attenuation factor, P(t) represents the data point set, D(t) represents the data persistence diagram, and t represents time; the expression of the multi-level data vector field is:/> Where V _k (t) represents the k-th layer data vector field at time t, /> represents the gradient operator, γ represents the weight parameter, D _k (t′) represents the data persistence diagram of the kth layer at time t′, t′∈[0, t]; the expression of the preset interference recognition function is:/> Where, Idf represents the preset interference identification function, θ represents the threshold function, V _k (t) represents the k-th layer data vector field at time t, /> represents the average vector of V _k (t), I _k (t) represents the interference data identified in the k-th layer data vector field at time t; the expression of the pure steel mill data is:/> Among them, y(t) represents the pure steel plant data at time t, _wkj represents the weight of the jth interference vector of the kth layer, _Ikj represents the jth interference vector identified at the kth layer at time t, _nk represents the number of interference vectors in the kth layer, K represents the total number of layers of the data vector field, δ represents the adjustment factor, U(t) represents the steel plant data at time t, and _Ik (t) represents the combination of _Ikj identified at the kth layer at time t.

According to one aspect of an embodiment of the present invention, a data acquisition device for a steel mill is provided, the data acquisition device for the steel mill comprising: a data acquisition module for acquiring time series data of a current production process of the steel mill; a feature extraction module for inputting the time series data into a feature extraction model to obtain nonlinear data features, the feature extraction model is obtained by training a pre-constructed extraction model with sample data, the pre-constructed extraction model comprises a time series analysis layer for capturing structural features, a self-attention mechanism layer for identifying abnormal structural features, and a nonlinear transformation layer for extracting nonlinear data features, the sample data comprises all or part of the time series data of the historical production process of the steel mill; a threshold determination module for determining a dynamic threshold based on a dynamic density estimation value of the non-linear data feature; a frequency determination module for adjusting the data acquisition frequency according to a comparison result of the dynamic density estimation value with the dynamic threshold value, so as to perform data acquisition on the steel mill at the adjusted data acquisition frequency.

According to one aspect of an embodiment of the present invention, there is provided an electronic device, comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the electronic device implements the steel plant data collection method as described above.

According to one aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor of a computer, the computer executes the steel mill data collection method described above.

The beneficial effects of the present invention are as follows: the present invention obtains the time series data of the current production process of the steel plant, inputs the time series data into a feature extraction model, obtains nonlinear data features, the feature extraction model is obtained by training a pre-constructed extraction model with sample data, and determines a dynamic threshold based on a dynamic density estimate of the nonlinear data feature. According to the comparison result of the dynamic density estimate with the dynamic threshold, the data collection frequency is adjusted to collect data from the steel plant at the adjusted data collection frequency. The above process can adjust the data collection frequency according to the comparison result of the dynamic density estimate with the dynamic threshold, thereby solving the technical problem of excessive collection or insufficient collection caused by using a fixed frequency to collect data from the steel plant.

In addition, the feature extraction model that combines the time series analysis layer, self-attention mechanism layer, nonlinear transformation layer, orthogonal constraint and multi-scale analysis layer can deeply capture the complex patterns and characteristics of time series data, especially capturing features at different time scales, and can consider the time dependence of data. It is more suitable for data environments such as steel mills that have a high degree of time dependence, and can analyze and process time series data more accurately; based on the data topological structure and multi-level data vector field, the interference part in the steel plant data can be accurately identified, thereby greatly improving the accuracy of data acquisition and improving the efficiency of data acquisition.

It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application.

BRIEF DESCRIPTION OF THE DRAWINGS

The drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present application, and together with the specification, are used to explain the principles of the present application. Obviously, the drawings described below are only some embodiments of the present application, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

FIG1 is a schematic diagram of an exemplary system architecture shown in an exemplary embodiment of the present application;

FIG2 is a flow chart of a steel plant data collection method shown in an exemplary embodiment of the present application;

FIG3 is a flow chart of a steel plant data collection method shown in another exemplary embodiment of the present application;

FIG4 is a structural diagram of a feature extraction model shown in another exemplary embodiment of the present application;

FIG5 is a block diagram of a data acquisition device for a steel plant shown in an exemplary embodiment of the present application;

FIG6 shows a schematic diagram of the structure of a computer system of an electronic device suitable for implementing an embodiment of the present application.

Detailed ways

The following will describe the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention, not for limiting the scope of protection of the present invention.

It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

FIG. 1 is a schematic diagram of an exemplary system architecture shown in an exemplary embodiment of the present application.

As shown in FIG1 , the system architecture may include an acquisition device 101 and a computer device 102. The computer device 102 may be at least one of a desktop graphics processing unit (GPU) computer, a GPU computing cluster, a neural network computer, etc. Relevant technicians may use the computer device 102 to obtain the time series data of the current production process of the steel mill, input the time series data into a feature extraction model, and obtain nonlinear data features. The feature extraction model is obtained by training a pre-built extraction model with sample data. Based on the dynamic density estimation value of the nonlinear data feature, a dynamic threshold is determined. According to the comparison result of the dynamic density estimation value and the dynamic threshold, the data acquisition frequency is adjusted to collect data from the steel mill through the adjusted data acquisition frequency. The acquisition device 101 is used to collect time series data. In this embodiment, the data acquisition device 101 uses a sensor and other devices to collect time series data and provides it to the computer device 102 for processing.

Illustratively, after acquiring the time series data from the acquisition device 101, the computer device 102 inputs the time series data into a feature extraction model to obtain nonlinear data features. The feature extraction model is obtained by training a pre-constructed extraction model with sample data. A dynamic threshold is determined based on a dynamic density estimate of the nonlinear data feature. According to a comparison result between the dynamic density estimate and the dynamic threshold, the data acquisition frequency is adjusted to collect data from the steel plant at the adjusted data acquisition frequency. The above process can adjust the data acquisition frequency according to the comparison result between the dynamic density estimate and the dynamic threshold, thereby solving the technical problem of excessive or insufficient data collection from the steel plant caused by using a fixed frequency for data collection.

It should be noted that the steel mill data collection method provided in the embodiment of the present application is generally executed by the computer device 102 , and accordingly, the steel mill data collection device is generally set in the computer device 102 .

The implementation details of the technical solution of the embodiment of the present application are described in detail below:

FIG2 is a flow chart of a steel mill data collection method shown in an exemplary embodiment of the present application. The steel mill data collection method can be executed by a computing and processing device, and the computing and processing device can be the computer device 102 shown in FIG1. Referring to FIG2, the steel mill data collection method at least includes steps S210 to S240, which are described in detail as follows:

In step S210, the time series data of the current production process of the steel plant is obtained.

In one embodiment of the present application, the time series data includes parameter change data in the current production process of the steel plant, such as temperature change data over time, chemical composition change data over time, pressure change data over time, and material flow change data over time.

In step S220, the time series data is input into a feature extraction model to obtain nonlinear data features.

In this embodiment, the feature extraction model is obtained by training a pre-built extraction model with sample data. The pre-built extraction model includes a time series analysis layer for capturing structural features, a self-attention mechanism layer for identifying abnormal structural features, and a non-linear transformation layer for extracting non-linear data features. The sample data includes all or part of the time series data of the historical production process of the steel plant.

In this embodiment, the structural features of the time series data include stationary structural features and non-stationary structural features, wherein the abnormal structural features (or key structural features) include non-stationary structural features and the like.

In this embodiment, the nonlinear data features of the time series data are extracted through the feature extraction module, thereby capturing the inherent structural features of the steel mill data, which is conducive to reducing the dimension or complexity of the steel mill data.

In step S230, a dynamic threshold is determined based on the dynamic density estimate of the nonlinear data feature.

In this embodiment, the calculation formula of the dynamic density estimation value is:

Wherein, _dt (f) represents the dynamic density estimate of the nonlinear data feature f at time t, N represents the number of nonlinear data features, K represents the space-time kernel function, which is used to measure the similarity between nonlinear data features in space and time, u represents the spatial distance between nonlinear data features, ω represents the time attenuation function, and Δt represents the time difference between any time _t0 and the current time t of the nonlinear data feature.

In this embodiment, a weight is assigned to each extracted nonlinear data feature by calculating a dynamic density estimation value. The weight is based on the time attribute of the nonlinear data feature, which means that the data feature closest to the abnormal data feature will be assigned a higher weight.

In this embodiment, the expression of the time decay function is:

ω(Δt)＝e ^-βΔt Formula (2)

Among them, β represents a constant, which determines the decay speed of the weight, Δt represents the time difference between any time t ₀ and the current time t of the nonlinear data feature, and ω represents the time decay function. Selecting the exponential decay function as the time decay function can ensure that the data points closest to the abnormal data feature have a larger weight.

In this embodiment, the expression of the space-time kernel function is:

Among them, σ represents the standard deviation of the Gaussian kernel function, which determines the width of the kernel function, s represents the dimension of the time series data, u represents the spatial distance between nonlinear data features, ω represents the time decay function, Δt represents the time difference between any time _t0 and the current time t of the nonlinear data feature, and K represents the space-time kernel function.

In this embodiment, the space-time kernel function is used to measure the similarity between nonlinear data features in space and time. The space-time kernel function takes into account both the spatial and temporal attributes of the data, and combines the Gaussian kernel function and the time attenuation function. The closer two nonlinear data features are in space and time, the higher the similarity between the two nonlinear data features.

In this embodiment, the calculation formula of the dynamic threshold is:

Where Φ represents the dynamic threshold, represents the mean of the dynamic density estimate _dt of the nonlinear data feature, θ represents a constant, and σ represents the standard deviation of the Gaussian kernel function.

In this embodiment, the dynamic threshold is calculated based on the mean of the dynamic density estimation values _dt of all nonlinear data features and the standard deviation of the Gaussian kernel function.

In step S240, the data collection frequency is adjusted according to the comparison result between the dynamic density estimation value and the dynamic threshold value, so as to collect data from the steel plant through the adjusted data collection frequency.

In this embodiment, for each nonlinear data feature, its dynamic density estimation value _dt is compared with the dynamic threshold value Φ. If the dynamic density estimation value of a nonlinear data feature is greater than the dynamic threshold value, then the nonlinear data feature is considered to be an abnormal data feature (or a key data feature). Once the abnormal data feature is identified, the data sampling frequency is increased to obtain more detailed steel mill data; on the contrary, if the dynamic density estimation value of the nonlinear data feature is less than or equal to the dynamic threshold value, the data sampling frequency is reduced to reduce the burden of steel mill data collection and processing. Therefore, the dynamic data density estimation method solves the technical problem of excessive or insufficient data collection caused by using a fixed frequency to collect data from a steel mill, and can not only more accurately identify abnormal data features in time series data, but also improve the collection efficiency of steel mill data, which can help steel mills better monitor and optimize production processes.

In one embodiment of the present application, after data is collected from a steel mill through the adjusted data collection frequency, the data collection method for the steel mill includes:

Based on the collected steel mill data, a multi-level data vector field is constructed, and the multi-level data vector field is used to extract data features at different levels.

In this embodiment, the multi-level data vector field can reveal the flow direction and pattern in the steel mill data. Each layer of the data vector field is constructed based on different resolutions to capture different levels of data features.

The multi-level data vector field is input into a preset interference identification function to obtain interference data.

In this embodiment, the expression of the preset interference identification function is:

Where Idf represents the preset interference identification function, θ represents the threshold function, which is used to filter out V _k (t) and The difference vector, V _k (t) represents the k-th layer data vector field at time t, /> represents the average vector of V _k (t), and I _k (t) represents the interference data identified in the k-th layer data vector field at time t.

In this embodiment, vectors that are inconsistent with the flow direction of normal data characteristics are identified through a preset interference identification function, and these vectors correspond to the interference parts in the steel mill data.

The interference data is removed from the steel plant data to obtain pure steel plant data.

In this embodiment, the expression of the pure steel plant data is:

Among them, y(t) represents the pure steel plant data at time t, _wkj represents the weight of the jth interference vector of the kth layer, _Ikj represents the jth interference vector identified at the kth layer at time t, _nk represents the number of interference vectors in the kth layer, K represents the total number of layers of the data vector field, δ represents the adjustment factor, U(t) represents the steel plant data at time t, and _Ik (t) represents the combination of _Ikj identified at the kth layer at time t.

In this embodiment, by removing interference data from the steel mill data, the interference data in the collected steel mill data is effectively reduced, thereby ensuring the accuracy of the collected steel mill data.

In one embodiment of the present application, the process of adjusting the data acquisition frequency according to the comparison result of the dynamic density estimation value and the dynamic threshold value includes:

Get the current data collection frequency of time series data.

In this embodiment, the current data collection frequency can be set according to the process stage, production conditions, etc. of the steel plant production, which will not be elaborated here.

If the dynamic density estimation value is less than or equal to the dynamic threshold value, the data feature corresponding to the dynamic density estimation value is determined to be a normal data feature, and the current data collection frequency is adjusted to the first data collection frequency.

In this embodiment, the first data collection frequency is less than the current data collection frequency. The first data collection frequency can be set according to actual conditions, which will not be described in detail herein.

In this embodiment, data is collected from the steel plant at a first data collection frequency to reduce the burden of data collection and processing on the steel plant.

If the dynamic density estimation value is greater than the dynamic threshold, the data feature corresponding to the dynamic density estimation value is determined to be an abnormal data feature, and the current data collection frequency is adjusted to a second data collection frequency.

In this embodiment, the second data collection frequency is greater than the current data collection frequency. The second data collection frequency can be set according to actual conditions, which will not be described in detail herein.

In this embodiment, data is collected from the steel plant through the second data collection frequency to obtain more detailed abnormal data, so as not to miss abnormal data, improve the efficiency and accuracy of data collection in the steel plant, and also help the steel plant better monitor and optimize the production process.

In one embodiment of the present application, if the pre-constructed extraction model further includes a multi-scale analysis layer for extracting features of data at different scales, the process of training the pre-constructed extraction model with sample data to obtain a feature extraction model includes:

The sample data is input into the time series analysis layer to obtain structural features.

In this embodiment, the expression of the time series analysis layer is:

Among them, tx(x) represents the structural characteristics of time series data, x represents time series data, sigm represents the activation function, _Wt represents the weight, _bt represents the bias, β represents the weight coefficient of Gaussian integral, and T represents transpose.

In this embodiment, by processing the time series data through the time series analysis layer, local patterns in the time series data can be captured and the inherent structural characteristics of the time series data can be better understood. For example, by analyzing the changes in temperature and chemical composition over time, the furnace temperature and material input can be predicted and adjusted to optimize the steelmaking process.

The structural features are input into the self-attention mechanism layer to obtain abnormal structural features.

In this embodiment, the expression of the self-attention mechanism layer is:

Among them, Q, K, V represent the query matrix, key matrix and value matrix in the self-attention mechanism layer respectively, a(ts) represents the abnormal structural feature, T represents transpose, _dk represents the dimension of the key, and ts represents the structural feature.

In this embodiment, the expression of the query matrix Q is:

Q＝W _q ts Formula (9)

Among them, ts represents the structural features, _Wq represents the weight matrix in the query matrix, and Q represents the query matrix in the self-attention mechanism layer.

In this embodiment, the key matrix K is expressed as:

K＝W _k ts Formula (10)

Among them, ts represents the structural features, _Wk represents the weight matrix in the key matrix, and K represents the key matrix in the self-attention mechanism layer.

In this embodiment, the expression of the value matrix V is:

V＝W _v ts Formula (11)

Among them, ts represents the structural features, W _v represents the weight matrix in the value matrix, and V represents the value matrix in the self-attention mechanism layer.

In this embodiment, the self-attention mechanism layer can capture and focus on abnormal time points or abnormal structural features in the time series data, and after capturing the abnormal time points or abnormal structural features, analyze the important influencing factors in the steelmaking production process to obtain the causes of the abnormality. For example, when the temperature fluctuates abnormally, potential production problems can be identified and handled in a timely manner.

The abnormal structure features are input into the nonlinear transformation layer to obtain nonlinear data features.

In this embodiment, the expression of the nonlinear transformation layer is:

Among them, f(a) represents the nonlinear data feature, _Wf represents the weight matrix of the nonlinear transformation layer, _bf represents the bias vector of the nonlinear transformation layer, μ represents the regularization coefficient, α represents the weight coefficient of the sine function, _Wr represents the regularized weight matrix, and a represents the abnormal structure feature.

In this embodiment, adding a nonlinear transformation layer enables the feature extraction model to process complex and interrelated production data, such as temperature, pressure, and material proportion, which can not only ensure the independence of the data but also avoid redundancy and error accumulation. The orthogonal constraint is introduced through the regularized weight matrix to ensure that the obtained data features are orthogonal.

The nonlinear data features are input into the multi-scale analysis layer to obtain data features of different scales.

In this embodiment, the expression of the multi-scale analysis layer is:

Among them, m(f) represents data features of different scales, _Wi represents the weight matrix of the i-th scale, _σi represents the standard deviation of the i-th scale, and f represents nonlinear data features.

In this embodiment, the multi-scale analysis layer can capture the nonlinear data characteristics of time series data at different time scales, thereby obtaining the characteristic changes of time series data at different time and space scales, such as rapid changes in the short term and long-term trends, to comprehensively monitor the production process.

Based on the data features of different scales and time series data, a loss function is constructed. With the goal of minimizing the loss function, the parameters of the time series analysis layer, the parameters of the self-attention mechanism layer, the parameters of the nonlinear transformation layer, and the parameters of the multi-scale analysis layer are updated.

In this embodiment, the loss function is expressed as:

Among them, L represents the loss function, _μ1 and _μ2 are regularization coefficients, _hi represents the state of the neuron at the i-th scale, F represents the Frobenius norm, _Wi represents the weight matrix of the i-th scale, and x represents the time series data.

In this embodiment, the loss function uses the gradient descent method to update and optimize the parameters of the time series analysis layer, the parameters of the self-attention mechanism layer, the parameters of the nonlinear transformation layer, and the parameters of the multi-scale analysis layer. By minimizing the loss function value, it can be ensured that the feature extraction model can accurately capture the structural features in the steel mill data and reflect the key information in the steel mill data through the structural features. At the same time, the dimension and complexity of the steel mill data are reduced, making subsequent data processing and analysis more efficient and accurate.

The time series analysis layer after parameter update, the self-attention mechanism layer after parameter update, the nonlinear transformation layer after parameter update and the multi-scale analysis layer after parameter update are combined to obtain a feature extraction model.

In this embodiment, the feature extraction model that combines the time series analysis layer, the self-attention mechanism layer, the nonlinear transformation layer, the orthogonal constraint and the multi-scale analysis layer can deeply capture the complex patterns and characteristics of the time series data, especially capturing features at different time scales, and can consider the time dependency of the data. It is more suitable for data environments such as steel mills that have a high degree of time dependency, and can analyze and process time series data more accurately.

In one embodiment of the present application, the process of constructing a multi-level data vector field based on the collected steel mill data includes:

The steel plant data is transformed into a dimension to obtain a set of data points.

In this embodiment, the expression of the data point set is:

P(t)＝M(U)＝exp(-k ₁ t ² )U·sin(k ₂ t)+k ₂ tU Formula (15)

Among them, _k1 represents the time attenuation factor, _k2 represents the periodic adjustment factor, t represents time, _k3 represents the linear time influence factor, U represents the steel plant data, M(U) represents the mapping function for dimensional transformation of the steel plant data, and P(t) represents the data point set.

In this embodiment, k ₁ is used to control the decay rate of the exp(-k ₁ t ² ) term, k ₂ is used to affect the frequency of data changes, k ₃ is used to control the linear change rate of data over time, exp(-k ₁ t ² ) represents the influence of the steel mill data decaying over time, and more directly reflects the influence of time changes, sin(k ₂ t) is used to keep capturing periodic changes, k ₃ tU is a linear term, which directly associates the time factor with the steel mill data U, and represents the trend of the steel mill data U linearly increasing or decreasing over time.

Based on the set of data points, a data persistence graph is calculated.

In this embodiment, the expression of the data persistence graph is:

Among them, λ represents the decay factor, P(t) represents the data point set, D(t) represents the data persistence diagram, and t represents time.

In this embodiment, λ is used to balance the first-order derivative of the data point set. and the second-order derivative of the data point set/> The method of calculating the data persistence graph based on the data point set is the topological data analysis method. Through the topological data analysis method, the topological structure and data characteristics of the steel plant data are further revealed.

Based on the data persistence graph, a multi-level data vector field is calculated.

In this embodiment, the expression of the multi-level data vector field is:

Where V _k (t) represents the k-th layer data vector field at time t, represents the gradient operator, γ represents the weight parameter, D _k (t′) represents the data persistence graph of the kth layer at time t′, t′∈[0,t].

In this embodiment, the gradient operator Represents the directional characteristics of steel mill data, and the weight parameter γ is used to balance the gradient term/> and quadratic terms/>

In this embodiment, based on the data topology structure and the multi-level data vector field, the interference part in the steel plant data is accurately identified, thereby greatly improving the accuracy of data collection and improving the efficiency of data collection.

FIG3 is a flow chart of a steel mill data collection method shown in another exemplary embodiment of the present application. As shown in FIG3 , the steel mill data collection method includes: (1) constructing a feature extraction model to capture the intrinsic structural features in the time series data through the feature extraction model; (2) using a data density estimation method based on time series characteristics and spatial distribution to determine a dynamic threshold, and based on the comparison result between the dynamic density estimation value and the dynamic threshold, identifying abnormal data features, and performing targeted data collection; (3) collecting steel mill data according to a dynamically adjusted collection frequency, and eliminating interference data in the steel mill data based on a data topology structure and a multi-level vector field processing method.

In this embodiment, a feature extraction model is constructed to capture the intrinsic structural features in the time series data, which is helpful to reduce the dimension or complexity of the steel plant data.

In this embodiment, a data density estimation method based on time series characteristics and spatial distribution is used to determine the dynamic threshold, and based on the comparison result between the dynamic density estimation value and the dynamic threshold, the abnormal data features are identified, and targeted data collection can be performed on the abnormal data features to avoid missing abnormal data.

In this embodiment, based on the data topology structure and the multi-level vector field processing method, interference data in the steel mill data is eliminated, which is beneficial to ensure the accuracy of data collection.

In this embodiment, the data acquisition device collects data through sensors, and dynamically adjusts the data acquisition frequency of the steel plant according to the nonlinear data characteristics of the time series data, thereby realizing adaptive acquisition. For example, during the smelting process, when the temperature and chemical composition are changing rapidly, the acquisition frequency is increased to obtain more accurate data; while in a relatively stable stage, the acquisition frequency is reduced to reduce the data processing burden.

In this embodiment, the edge computing node is connected to the collection device through an interface to monitor parameter change data in the steelmaking process in real time, such as temperature change data, pressure change data, and material flow change data. The nonlinear data characteristics of the change data are calculated in real time. When the nonlinear data characteristics exceed the dynamic threshold, the data collection frequency is increased; when the nonlinear data characteristics are lower than the dynamic threshold, the data collection frequency is reduced.

FIG4 is a structural diagram of a feature extraction model shown in another exemplary embodiment of the present application. As shown in FIG4 , the feature extraction model includes a time series analysis layer, a self-attention mechanism layer, a nonlinear transformation layer, and a multi-scale analysis layer.

In this embodiment, the time series analysis layer is used to capture the structural features in the time series data through the convolution kernel, which is conducive to better understanding the intrinsic structure of the data.

In this embodiment, the self-attention mechanism layer allows the data extraction model to focus on different parts of the time series data. By focusing on abnormal time points or abnormal structural features in the production process, potential production problems can be identified and handled in a timely manner.

In this embodiment, the nonlinear transformation layer is used to enable the feature extraction model to process complex and interrelated production data, such as temperature, pressure, and material proportion, which can not only ensure the independence of the data but also avoid redundancy and error accumulation. The orthogonal constraint is introduced through the regularized weight matrix to ensure that the obtained data features are orthogonal.

In this embodiment, the multi-scale analysis layer is used to capture the nonlinear characteristics of time series data at different time scales, so as to obtain the characteristic changes of time series data at different time and space scales, such as rapid changes in the short term and long-term trends, so as to comprehensively monitor the production process.

The following describes an apparatus embodiment of the present application, which can be used to execute the steel mill data collection method in the above-mentioned embodiment of the present application. For details not disclosed in the apparatus embodiment of the present application, please refer to the above-mentioned embodiment of the steel mill data collection method of the present application.

FIG5 is a block diagram of a steel plant data acquisition device shown in an exemplary embodiment of the present application. The device can be applied to the implementation environment shown in FIG1 and is specifically configured in the computer device 102. The device can also be applied to other exemplary implementation environments and specifically configured in other devices. This embodiment does not limit the implementation environment to which the device is applicable.

As shown in FIG5 , the exemplary steel plant data collection device includes:

The data acquisition module 501 is used to acquire the time series data of the current production process of the steel plant.

The feature extraction module 502 is used to input the time series data into the feature extraction model to obtain nonlinear data features.

The threshold determination module 503 is used to determine a dynamic threshold based on a dynamic density estimation value of a nonlinear data feature.

The frequency determination module 504 is used to adjust the data collection frequency according to the comparison result between the dynamic density estimation value and the dynamic threshold value, so as to collect data from the steel mill through the adjusted data collection frequency.

In one embodiment of the present application, the time series data includes parameter change data in the current production process of the steel plant, such as temperature change data over time, chemical composition change data over time, pressure change data over time, and material flow change data over time.

In this embodiment, the feature extraction model is obtained by training a pre-built extraction model with sample data. The pre-built extraction model includes a time series analysis layer for capturing structural features, a self-attention mechanism layer for identifying abnormal structural features, and a non-linear transformation layer for extracting non-linear data features. The sample data includes all or part of the time series data of the historical production process of the steel plant.

In this embodiment, the structural features of the time series data include stable structural features and non-stationary structural features, wherein the abnormal structural features include non-stationary structural features and the like.

In this embodiment, the nonlinear data features of the time series data are extracted through the feature extraction module, thereby capturing the inherent structural features of the steel mill data, which is conducive to reducing the dimension or complexity of the steel mill data.

In this embodiment, the calculation formula of the dynamic density estimation value is shown in formula (1), which is not repeated here. Through the calculation of the dynamic density estimation value, a weight is assigned to each extracted nonlinear data feature, and this weight is based on the time attribute of the nonlinear data feature, which means that the data feature closest to the abnormal data feature will be assigned a higher weight.

In this embodiment, the expression of the time decay function is shown in formula (2), which is not described in detail here. The expression of the space-time kernel function is shown in formula (3), which is not described in detail here.

In this embodiment, the space-time kernel function is used to measure the similarity between nonlinear data features in space and time. The space-time kernel function takes into account both the spatial and temporal attributes of the data, and combines the Gaussian kernel function and the time attenuation function. The closer two nonlinear data features are in space and time, the higher the similarity between the two nonlinear data features.

In this embodiment, the calculation formula of the dynamic threshold is shown in formula (4), which is not described here. The dynamic threshold is calculated based on the mean of the dynamic density estimation value _dt of all nonlinear data features and the standard deviation of the Gaussian kernel function.

In this embodiment, for each nonlinear data feature, its dynamic density estimation value _dt is compared with the dynamic threshold Φ. If the dynamic density estimation value of a nonlinear data feature is greater than the dynamic threshold, then the nonlinear data feature is considered to be an abnormal data feature (key data feature). Once the abnormal data feature is identified, the data sampling frequency is increased to obtain more detailed steel mill data; on the contrary, if the nonlinear data feature is less than or equal to the dynamic threshold, the data sampling frequency is reduced to reduce the burden of steel mill data collection and processing. Therefore, the dynamic data density estimation method solves the technical problem of excessive or insufficient data collection caused by using a fixed frequency to collect data from a steel mill. It can not only more accurately identify abnormal data features in time series data, but also improve the collection efficiency of steel mill data, which can help steel mills better monitor and optimize production processes.

It should be noted that the steel mill data acquisition device provided in the above embodiment and the steel mill data acquisition method provided in the above embodiment belong to the same concept, wherein the specific manner in which each module and unit performs the operation has been described in detail in the method embodiment, and will not be repeated here. In actual application, the steel mill data acquisition device provided in the above embodiment can allocate the above functions to different functional modules as needed, that is, divide the internal structure of the device into different functional modules to complete all or part of the functions described above, and this is not limited here.

An embodiment of the present application also provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by one or more processors, the electronic device implements the steel plant data collection method provided in the above-mentioned embodiments.

Fig. 6 shows a schematic diagram of a computer system of an electronic device suitable for implementing an embodiment of the present application. It should be noted that the computer system 600 of the electronic device shown in Fig. 6 is only an example and should not bring any limitation to the functions and scope of use of the embodiment of the present application.

As shown in Figure 6, computer system 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 602 or the program loaded from the storage part 608 to the random access memory (RAM) 603, such as executing the method in the above embodiment. In RAM 603, various programs and data required for system operation are also stored. CPU 601, ROM 602 and RAM 603 are connected to each other through bus 604. Input/output (I/O) interface 605 is also connected to bus 604.

The following components are connected to the I/O interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I/O interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as needed so that a computer program read therefrom is installed into the storage section 608 as needed.

In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication section 609, and/or installed from a removable medium 611. When the computer program is executed by a central processing unit (CPU) 601, various functions defined in the system of the present application are executed.

It should be noted that the computer-readable medium shown in the embodiment of the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a computer-readable computer program is carried. This propagated data signal can take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. A computer program contained on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

The flowchart and block diagram in the accompanying drawings illustrate the possible architecture, functions and operations of the system, method and computer program product according to various embodiments of the present application. Wherein, each box in the flowchart or block diagram can represent a module, a program segment, or a part of the code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of the boxes in the block diagram or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

The units involved in the embodiments described in this application may be implemented by software or hardware, and the units described may also be set in a processor. The names of these units do not constitute limitations on the units themselves in some cases.

Another aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor of a computer, causes the computer to execute the steel mill data collection method as described above. The computer-readable storage medium may be included in the electronic device described in the above embodiment, or may exist independently without being assembled into the electronic device.

Another aspect of the present application also provides a computer program product or a computer program, the computer program product or the computer program includes computer instructions, the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steel mill data collection method provided in each of the above embodiments.

The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by a person with ordinary knowledge in the technical field without departing from the spirit and technical ideas disclosed by the present invention should still be covered by the claims of the present invention.

Claims (10)

1. The steel mill data acquisition method is characterized by comprising the following steps of:

acquiring time sequence data of the current production process of a steel mill;

Inputting the time sequence data into a feature extraction model to obtain nonlinear data features, wherein the feature extraction model is obtained by training a pre-built extraction model by sample data, the pre-built extraction model comprises a time sequence analysis layer for capturing structural features, a self-attention mechanism layer for identifying abnormal structural features and a nonlinear transformation layer for extracting nonlinear data features, and the sample data comprises all or part of time sequence data of a steel mill historical production process;

determining a dynamic threshold based on the dynamic density estimation value of the nonlinear data characteristic;

And adjusting the data acquisition frequency according to the comparison result of the dynamic density estimation value and the dynamic threshold value, so as to acquire the data of the steel plant through the adjusted data acquisition frequency.

2. The steel mill data collection method according to claim 1, wherein after data collection of the steel mill by the adjusted data collection frequency, the steel mill data collection method comprises:

constructing a multi-level data vector field based on the acquired steel mill data, wherein the multi-level data vector field is used for extracting data characteristics of different levels;

Inputting the multi-level data vector field into a preset interference recognition function to obtain interference data;

And eliminating the interference data from the steel mill data to obtain pure steel mill data.

3. The steel mill data acquisition method according to claim 1 or 2, wherein the process of adjusting the data acquisition frequency according to the comparison result of the dynamic density estimation value and the dynamic threshold value comprises:

Acquiring the current data acquisition frequency of the time sequence data;

If the dynamic density estimation value is smaller than or equal to the dynamic threshold value, judging that the data characteristic corresponding to the dynamic density estimation value is a normal data characteristic, and adjusting the current data acquisition frequency to be a first data acquisition frequency, wherein the first data acquisition frequency is smaller than the current data acquisition frequency;

And if the dynamic density estimation value is larger than the dynamic threshold value, judging that the data characteristic corresponding to the dynamic density estimation value is an abnormal data characteristic, and adjusting the current data acquisition frequency to a second data acquisition frequency, wherein the second data acquisition frequency is larger than the current data acquisition frequency.

4. The steel mill data collection method according to claim 1 or 2, wherein if the pre-built extraction model further comprises a multi-scale analysis layer for extracting features of different scale data, training the pre-built extraction model by sample data, and obtaining the feature extraction model comprises:

Inputting the sample data into the time sequence analysis layer to obtain structural characteristics;

Inputting the structural features into the self-attention mechanism layer to obtain abnormal structural features;

inputting the abnormal structural characteristics into the nonlinear transformation layer to obtain nonlinear data characteristics;

Inputting the nonlinear data features into the multi-scale analysis layer to obtain data features with different scales;

constructing a loss function based on the different scale data features and the time sequence data, and updating parameters of the time sequence analysis layer, the self-attention mechanism layer, the nonlinear transformation layer and the multiscale analysis layer with the aim of minimizing the loss function;

and combining the time sequence analysis layer after parameter updating, the self-attention mechanism layer after parameter updating, the nonlinear transformation layer after parameter updating and the multi-scale analysis layer after parameter updating to obtain the feature extraction model.

5. The steel mill data collection method according to claim 4, wherein the expression of the time series analysis layer is:

Wherein ts (x) represents a structural feature of time series data, x represents time series data, sigm represents an activation function, W _t represents a weight, b _t represents a bias, β represents a weight coefficient of gaussian integration, and T represents a transpose;

The expression of the self-attention mechanism layer is as follows:

Wherein Q, K, V represents a query matrix, a key matrix, and a value matrix in the self-attention mechanism layer, respectively, a (ts) represents an abnormal structural feature, T represents a transpose, d _k represents a dimension of a key, and ts represents a structural feature;

the expression of the nonlinear transformation layer is as follows:

Wherein f (a) represents a nonlinear data characteristic, W _f represents a weight matrix of a nonlinear transformation layer, b _f represents a bias vector of the nonlinear transformation layer, μ represents a regularization coefficient, α represents a weight coefficient of a sine function, W _r represents a regularized weight matrix, and a represents an abnormal structural characteristic;

the expression of the multi-scale analysis layer is as follows:

Wherein m (f) represents data features of different scales, W _i represents a weight matrix of an ith scale, sigma _i represents a standard deviation of the ith scale, and f represents a nonlinear data feature;

The expression of the loss function is:

Wherein L represents a loss function, μ ₁ and μ ₂ are regularization coefficients, h _i represents a state of a neuron where the ith scale is located, F represents a Frobenius norm, W _i represents a weight matrix of the ith scale, and x represents time series data;

the calculation formula of the dynamic density estimation value is as follows:

Wherein d _t (f) represents a dynamic density estimate of the nonlinear data feature f at time t, N represents the number of nonlinear data features, K represents a space-time kernel function for measuring the similarity in space and time between the nonlinear data features, u represents the spatial distance between the nonlinear data features, ω represents a time decay function, Δt represents the time difference between any time t ₀ of the nonlinear data features and the current time t;

The calculation formula of the dynamic threshold value is as follows:

wherein phi represents the dynamic threshold value, The mean value of the dynamic density estimation d _t representing the nonlinear data feature, θ represents a constant, and σ represents the standard deviation of the gaussian kernel function.

6. The steel mill data acquisition method according to claim 2, wherein the process of constructing a multi-level data vector field based on the acquired steel mill data comprises:

Performing dimension conversion on the steel mill data to obtain a data point set;

calculating to obtain a data persistence graph based on the data point set;

and calculating the multi-level data vector field based on the data persistence graph.

7. The steel mill data collection method of claim 6, wherein the expression of the data point set is:

P(t)＝M(U)＝exp(-k₁t²)U·sin(k₂t)+k₃tU

Wherein k ₁ represents a time attenuation factor, k ₂ represents a periodicity adjustment factor, t represents time, k ₃ represents a linear time influence factor, U represents steelworks data, M (U) represents a mapping function for dimensional transformation of the steelworks data, and P (t) represents a set of data points;

The expression of the data persistence map is as follows:

Wherein λ represents an attenuation factor, P (t) represents a data point set, D (t) represents a data persistence map, and t represents time;

the expression of the multi-level data vector field is as follows:

wherein V _k (t) represents the kth layer data vector field at time t, Representing a gradient operator, gamma representing a weight parameter, D _k (t ') representing a data persistence map of a kth layer at time t ', t ' e [0, t ];

the expression of the preset interference recognition function is as follows:

Where Idf represents a preset interference recognition function, θ represents a threshold function, V _k (t) represents the kth layer data vector field at time t, Representing the average vector of V _k (t), I _k (t) representing the interference data identified in the k-th layer data vector field at time t;

The expression of the pure steel mill data is as follows:

Where y (t) represents clean steelworks data at time t, w _kj represents the weight of the j-th interference vector of the K-th layer, I _kj represents the j-th interference vector identified by the K-th layer at time t, n _k represents the number of interference vectors of the K-th layer, K represents the total number of layers of the data vector field, δ represents the adjustment factor, U (t) represents steelworks data at time t, I _k (t) represents the combination of I _kj identified by the K-th layer at time t.

8. A steel mill data acquisition device, characterized in that the steel mill data acquisition device comprises:

the data acquisition module is used for acquiring time sequence data of the current production process of the steel mill;

The feature extraction module is used for inputting the time sequence data into a feature extraction model to obtain nonlinear data features, the feature extraction model is obtained by training a pre-built extraction model through sample data, the pre-built extraction model comprises a time sequence analysis layer for capturing structural features, a self-attention mechanism layer for identifying abnormal structural features and a nonlinear transformation layer for extracting the nonlinear data features, and the sample data comprises all or part of time sequence data of a steel mill historical production process;

The threshold determining module is used for determining a dynamic threshold based on the dynamic density estimated value of the nonlinear data characteristic;

And the frequency determining module is used for adjusting the data acquisition frequency according to the comparison result of the dynamic density estimated value and the dynamic threshold value so as to acquire the data of the steel mill through the adjusted data acquisition frequency.

9. An electronic device, comprising:

One or more processors;

Storage means for storing one or more programs which when executed by the one or more processors cause the electronic device to implement the steel mill data collection method of any one of claims 1 to 7.

10. A computer readable storage medium, having stored thereon computer readable instructions, which when executed by a processor of a computer, cause the computer to perform the steel mill data collection method of any one of claims 1 to 7.