Soft-sensing method for silicon content in blast furnace molten iron based on copula function and spatio-temporal convolution network
By combining Copula functions and spatiotemporal convolutional networks, the inter-variable and temporal features of the blast furnace process are extracted, solving the problems of lag and high cost in measuring the silicon content of molten iron during blast furnace ironmaking, and achieving efficient and accurate soft measurement results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG UNIV
- Filing Date
- 2024-07-10
- Publication Date
- 2026-06-09
Smart Images
Figure CN119152980B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of predicting the quality of molten iron in blast furnace ironmaking, and in particular to a soft measurement method for the silicon content of molten iron in blast furnace based on Copula functions and spatiotemporal convolutional networks. Background Technology
[0002] In the blast furnace ironmaking process, many variables characterizing the process state are constantly changing, including state variables, quality variables, and operational variables. Predicting quality variables is crucial for process optimization, fault detection, diagnosis and prevention, and operational decision-making. Among these quality variables, silicon content, as a key indicator reflecting the internal chemical heat of molten iron, is one of the most important. With the continuous modernization of the ironmaking industry, the measurement problems of many state variables have been largely solved. However, due to the harsh industrial environment, some variables crucial to product quality cannot be directly measured or are too costly to measure. Therefore, soft measurement of variables that are difficult to measure directly becomes even more important.
[0003] Blast furnace ironmaking is a metallurgical process that transforms iron ore into molten iron. This process is a continuous production method, where raw materials interact from top to bottom, while gases move in the opposite direction from bottom to top. Blast furnace ironmaking involves the proportional mixing and grinding of solid raw materials such as iron ore, limestone, and coke. These materials are then fed into the upper part of the blast furnace. Preheated air and pulverized coal are injected into the blast furnace, where combustion produces high-temperature, high-pressure reducing gases. As these gases rise, they heat the furnace charge, leading to a series of complex physical and chemical transformations. These reactions reduce iron oxide in the iron ore to molten iron, which can be collected through the taphole. Additionally, limestone and other substances are injected into the blast furnace hearth to react with impurities in the charge, promoting slag separation. Since blast furnace exhaust gases and flue gases contain a large amount of heat energy, they are typically collected for power generation or to heat the furnace charge. Given the complexity of the blast furnace ironmaking process, its dynamic nature, complex nonlinearity, numerous variables, noise, and the presence of outliers pose significant challenges to the soft measurement of molten iron quality indicators.
[0004] In the blast furnace ironmaking process, relying primarily on manual testing to measure the silicon content of molten iron presents significant drawbacks. Firstly, testing procedures often exhibit a noticeable lag, negatively impacting subsequent blast furnace operation optimization and product quality management. However, with the continuous improvement of industrial informatization, data-driven soft measurement technologies for quality indicators have become crucial for improving quality and operational efficiency. Summary of the Invention
[0005] To overcome the shortcomings of existing methods, the present invention aims to provide a soft measurement method for silicon content in blast furnace molten iron based on Copula functions and spatiotemporal convolutional networks.
[0006] The technical solution for achieving the technical objective of this invention is as follows:
[0007] A soft measurement method for silicon content in blast furnace molten iron based on Copula functions and spatiotemporal convolutional networks is proposed. Combining R-Teng Copula functions and graph neural networks (GCNs), a feature extraction framework for process variables based on blast furnace data is proposed to obtain the relationships between variables, thereby extracting features between blast furnace process variables. A time-series feature extraction framework for process variables based on blast furnace data is also proposed, using a time-series convolutional neural network (TCN) with a gated mechanism to extract the time-series features of the blast furnace process variable sequence. The extracted features between process variables, time-series features, and the original process variable sequence are concatenated, and a regressor composed of multiple linear layers is used to achieve soft measurement of silicon content in blast furnace molten iron.
[0008] The framework for feature extraction between variables relying on blast furnace data is as follows: the vine copula fitting process includes constructing a bivariate copula using Bayes' theorem, calculating the conditional probability formula based on the assumed bivariate copula, and selecting and optimizing the vine copula model based on maximum likelihood; based on the tree structure matrix generated during the vine copula fitting process, constructing the adjacency matrix between variables, and then combining it with a graph convolutional neural network to extract features between variables.
[0009] The aforementioned framework for extracting time-series features of process variables that rely on blast furnace data uses two parallel TCN modules and different gating units, and uses sigmoid and tanh functions as activation functions to achieve reasonable time feature extraction.
[0010] The beneficial effects of this invention are as follows:
[0011] 1. A feature extraction framework for variables is proposed by combining R-copula function and graph neural network (GCN) to obtain more comprehensive relationships between variables and achieve efficient feature extraction, aiming to handle the nonlinear and dynamic characteristics of the blast furnace ironmaking process.
[0012] 2. Use a temporal convolutional neural network (TCN) with gating mechanism to extract time-series features of blast furnace process variable sequences, thereby enhancing the ability to capture long-term scale-dependent features.
[0013] 3. The extracted inter-variable features, time-series features, and original process variable sequences are concatenated to ensure the integrity of important information extracted from the features. Attached Figure Description
[0014] Figure 1 The diagram shows a framework of a soft measurement method for silicon content in blast furnace molten iron based on Copula functions and spatiotemporal convolutional networks according to the present invention.
[0015] Figure 2The results are experimental findings of soft measurement of silicon content in molten iron using VC-TGCN.
[0016] Figure 3 This is a test result of the soft measurement method of the present invention on a dataset of silicon content in molten iron. Detailed Implementation
[0017] To more clearly describe the technical content of the present invention, the specific implementation method will be further explained below in conjunction with the accompanying drawings.
[0018] like Figure 1 The diagram shown illustrates the framework of a soft measurement method for silicon content in blast furnace molten iron based on Copula functions and spatiotemporal convolutional networks according to the present invention. The steps are as follows:
[0019] A feature extraction framework based on blast furnace data was proposed, combining the R-Teng copula function and Graph Neural Network (GCN), to obtain more comprehensive inter-variable relationships and thus achieve efficient extraction of features among blast furnace process variables. In addition to extracting inter-variable features, a time-series feature extraction framework based on blast furnace data was used. This framework employs a Temporal Convolutional Neural Network (TCN) with a gated mechanism to extract time-series features from the blast furnace process variable sequence. By concatenating the extracted inter-variable features, time-series features, and the original process variable sequence, the integrity of important information extracted from the feature extraction was ensured. Then, a regressor composed of multiple linear layers was used to achieve soft measurement of the silicon content in the blast furnace molten iron.
[0020] The inter-variable feature extraction framework that relies on blast furnace data is as follows:
[0021] Copula functions are widely used in multivariate joint distribution modeling, connecting the cumulative probability distributions (CDFs) of each variable to be analyzed into their joint distribution. Traditional bivariate copula functions require pre-fixing the family of copula functions used when modeling multivariate joint probability distributions, which is ineffective for modeling highly dynamic blast furnace process variables. Vine copula decomposes a multivariate copula into a series of bivariate copulas by establishing a multi-tree structure. The actual vine copula fitting process includes constructing bivariate copulas using Bayes' theorem, calculating conditional probability formulas based on the assumed bivariate copulas, and selecting and optimizing the vine copula model based on maximum likelihood.
[0022] Let x1, x2, ..., x m Let there be m process variables in a blast furnace. Then, the joint probability distribution f(x) of these m process variables can be decomposed according to the following formula.
[0023]
[0024] After the above decomposition, the key term for fitting the copula function becomes the conditional probability of a certain variable under the condition of other variables. This conditional probability can be solved using the following formula.
[0025]
[0026] Where, when variable number i = t-1, in the formula
[0027] c m-i,m∣1:m-i-1 (F(x m-i |x1,…,x m-i-1 ),F(x m |x1,…,x m-i-1 ));
[0028] It can be simply represented as c 1,t (F(x1),F(x t This formula describes the steps for joint probability distribution decomposition using R-copula. After decomposition, the type of copula function can be selected based on the fitting effect. To better reflect the actual distribution characteristics of blast furnace process variables, this invention limits the copula function family to five types: Gaussian, Student, Clayton, Gumbel, and Frank. The specific data expressions are shown in the table below:
[0029]
[0030]
[0031] Since the bivariate copula equation fits the data by treating pairs of variables, the m process variables are decomposed into... Given two bivariate copula functions, the generalized modeling formula using R-copula can be expressed as follows:
[0032]
[0033] in This refers to removing x i and x j For the other variables besides those mentioned above, the CDF conditional distribution can be solved iteratively in a tree-like structure, based on the above formula. After calculating the CDF conditional distribution, the overall framework has been established; the next step is estimation and optimization. The parameters of each copula and its position in the tree structure are determined. The estimation and optimization process employs maximum likelihood estimation (MLE). Traditional optimization formulas parameterize the marginal distribution and optimize these parameters along with the parameters required for copula fitting. To simplify the process, the optimization of the marginal distribution and the optimization of the copula fitting parameters are performed separately. Kernel density estimation (KDE) is used to model the marginal distribution of the variables, and then the generalized copula parameter optimization formula is obtained based on maximum likelihood and the Akaike information criterion (AIC), as follows:
[0034]
[0035] in The parameter representing all copulas. The position within the tree structure formed by the copula function is called the tree order. Finally, after the fitting process, we obtain m-1 trees, each representing a set of dependencies. For example, the edges of the first tree represent direct relationships between variables, the edges of the second tree represent direct relationships between variables, the edges of the third tree represent direct relationships between variables, and the edges of the second tree represent conditional dependencies. The edges of the second tree represent conditional dependencies between variables given the first tree, and so on. The subsequent trees follow the same pattern. Starting from the first tree, the edges of each tree become nodes in the next tree. Therefore, the roles of nodes and edges continuously change to construct higher-level conditional dependencies. To briefly illustrate the tree structure construction process, Figure 2 This is the result of constructing a tree structure using R-vines for five blast furnace process variables. The tree structure numbers represent the order in which copulas are fitted to different variables. The bottom tree structure is a combination of trees 1, 2, 3, and 4, i.e., a vine structure. Each row in the diagram represents a pair of copulas connecting two variables. Once the overall structure is determined, the categories and parameters are also determined. In other words, using the tree structure, we can create the adjacency matrix required for the GCN in the network below.
[0036] The framework for extracting time-series features of process variables that rely on blast furnace data is as follows:
[0037] Compared to Recurrent Neural Networks (RNNs) and Long Short-Term Memory Neural Networks (LSTMs), Temporal Convolutional Neural Networks (TCNs) possess the unique advantage of parallel processing. Therefore, the time-series feature extraction process for silicon content information in blast furnace molten iron utilizes a TCN network. Furthermore, to enhance the TCN's ability to capture long-scale dependent features, crucial gating mechanisms from LSTMs and Gated Recurrent Units (GRUs) are introduced into the temporal convolutional network, such as forget gates, input gates, and output gates. Experiments revealed that temporal convolutional networks do not require input and forget gates; therefore, an output gating unit is added at the end of the temporal convolutional neural network. Experiments demonstrate that this LSTM-like output gating unit can more effectively analyze long-range spatiotemporal dependencies and extract spatiotemporal features. Therefore, in designing the TCN network, this invention uses two parallel-connected TCN modules and different gating units, employing sigmoid and tanh functions as activation functions to achieve reasonable temporal feature extraction.
[0038] The specific process of the method provided by this invention is as follows:
[0039] Step 1: Offline training;
[0040] Step 1.1: In the soft measurement of silicon content in molten iron based on VC-TGCN, the mean squared error (MSE) is used as the model training loss type, which can be given by the following formula:
[0041]
[0042] Among them, y i and This represents the soft-measured value and the true value of the silicon content in molten iron output by the model at time i. The model is updated based on the MSE loss.
[0043] Step 2: Real-time prediction;
[0044] First, real-time data is collected and standardized. Then, using the trained VC-TGCN model, the real-time soft measurement value of the silicon content in molten iron is output.
[0045] To verify this invention, we used actual industrial data, taking a steelmaking plant with a volume of 2650 m³. 3 The data sample includes the silicon content of molten iron from blast furnace No. 2 in 2020, containing 28 parameters. Training was performed using 600 normal samples, and testing was conducted using 200 points as the test set. (See attached...) Figure 3 The figure shows the test results of the soft measurement method of the present invention on the iron silicon content dataset. The soft measurement results show that the model performs well.
[0046] The technical features of the above embodiments can be further combined. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0047] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, all of which fall within the protection scope of the present invention. The protection scope of the present invention is defined by the appended claims and any equivalent technical solutions.
Claims
1. A soft measurement method for silicon content in blast furnace hot metal based on Copula functions and spatiotemporal convolutional networks, characterized in that, Combining the R-Teng copula function and graph neural network (GCN), a feature extraction framework for process variables based on blast furnace data is proposed to obtain the relationships between variables and thus extract the features between blast furnace process variables. The feature extraction framework for process variables based on blast furnace data uses a temporal convolutional neural network (TCN) with a gated mechanism to extract the temporal features of the blast furnace process variable sequence. The extracted features between process variables, temporal features, and the original process variable sequence are concatenated, and a regressor composed of multiple linear layers is used to achieve soft measurement of the silicon content of blast furnace molten iron. The inter-variable feature extraction framework that relies on blast furnace data is as follows: The vine copula fitting process includes constructing a bivariate copula using Bayes' theorem, calculating the conditional probability formula based on the assumed bivariate copula, and selecting and optimizing the vine copula model based on maximum likelihood. Based on the tree structure matrix generated during the vine copula fitting process, an adjacency matrix between variables is constructed, and then a graph convolutional neural network is used to extract features between variables. The aforementioned feature extraction framework for process variables dependent on blast furnace data uses two parallel TCN modules and different gating units, and uses sigmoid and tanh functions as activation functions to achieve reasonable time feature extraction.