Blast furnace throat temperature estimation method and device based on online fusion, and electronic equipment

Abstract

The application discloses a blast furnace throat temperature estimation method and device based on online fusion, and electronic equipment, comprising: taking blast furnace equipment parameters and distribution matrix as input, calculating the shape of each layer of the blast furnace and the parameters of each layer of ore coke ratio; according to the multi-time scale characteristics of the blast furnace, the blast furnace main parameters, the throat temperature measuring point data and the ore coke ratio parameters are divided to obtain high and low frequency data sets, and the feature importance screening is carried out respectively; the low frequency data set after feature screening is used for training the support vector regressor to obtain a low frequency throat temperature estimation model, and the high frequency data set after feature screening is used for training the integrated gate recurrent unit network to obtain a high frequency throat temperature estimation model; the to-be-measured low frequency data set and the high frequency data set are taken as the input of different throat temperature estimation models respectively to obtain the low frequency throat temperature estimation value and the high frequency throat temperature estimation value, and the final high frequency temperature estimation value is obtained through the online fusion network, and high frequency real-time estimation is realized.

Description

Technical Field

[0001] This application relates to the field of energy and power engineering technology, and in particular to a method and apparatus for estimating blast furnace throat temperature based on online fusion, and electronic equipment. Background Technology

[0002] As a large-scale countercurrent moving bed chemical reactor, the blast furnace involves complex mass transfer phenomena and chemical reactions, and is also a typical "black box" container. In addition, the internal environment of the blast furnace is extremely harsh. High temperature, high pressure, and high dust levels make it difficult to directly detect many parameters inside the blast furnace, thus making it difficult to form a closed-loop control for blast furnace operation.

[0003] Accurate detection and proper control of the gas flow distribution and development are crucial for the control of the blast furnace ironmaking process. The gas flow distribution is closely related to important indicators such as blast furnace operating conditions, energy and raw material consumption, iron production and quality, and blast furnace lifespan. Different gas flow distributions will also lead to different operating conditions inside the blast furnace. Therefore, proper control of the gas flow distribution is of great significance for improving the quality and output of blast furnace products.

[0004] The temperature measurement at the blast furnace throat directly reflects the gas flow distribution and is a crucial parameter characterizing the gas flow state at the furnace top. Monitoring the throat temperature is an indispensable part of the blast furnace ironmaking process. In production practice, the throat cross-shaped temperature measuring device based on thermocouple sensors is the most classic and widely used method, offering advantages such as high accuracy, good real-time performance, and excellent dynamic performance. However, on the one hand, the throat location presents challenges due to the direct contact between high-temperature gas and the temperature sensor, significant dust levels, and harsh high-pressure environment, making the collected data difficult to use directly. On the other hand, sensor malfunctions are frequent, leading to difficulties in detecting accurate temperature data. Furthermore, the long shutdown and maintenance cycles of the blast furnace affect operators' judgment of the actual operating status, thus posing risks to blast furnace operation monitoring and seriously impacting the stable and efficient operation of the blast furnace. Summary of the Invention

[0005] The purpose of this application is to provide a method, apparatus, and electronic device for estimating blast furnace throat temperature based on online fusion, so as to solve the problem that the throat temperature is difficult to measure accurately in the prior art.

[0006] According to a first aspect of the embodiments of this application, a method for estimating the throat temperature of a blast furnace based on online fusion is provided, comprising:

[0007] Acquire blast furnace equipment parameters, charging matrix, blast furnace main parameters, and blast furnace throat temperature measurement point data;

[0008] Based on the material distribution pattern, the principle of equal volume, and the descent pattern, the blast furnace equipment parameters and the material distribution matrix are used as inputs to calculate the material surface shape of each layer inside the blast furnace, and the ore-coke ratio parameter of each layer is calculated based on the material surface shape.

[0009] Considering the multi-timescale characteristics of the internal reaction of the blast furnace, the main parameters of the blast furnace, the temperature measurement point data of the blast furnace throat, and the ore-coke ratio parameters are divided into high-frequency datasets and low-frequency datasets.

[0010] Feature importance filtering is performed on the two datasets separately to obtain the high-frequency dataset and the low-frequency dataset after feature filtering;

[0011] A support vector regressor is trained using the low-frequency dataset after feature selection to obtain a low-frequency furnace throat temperature estimation model. An integrated gated recurrent unit network is trained using the high-frequency dataset after feature selection to obtain a high-frequency furnace throat temperature estimation model.

[0012] Input the low-frequency dataset to be tested into the low-frequency furnace throat temperature estimation model to obtain the low-frequency furnace throat temperature estimate; input the high-frequency dataset to be tested into the high-frequency furnace throat temperature estimation model to obtain the high-frequency furnace throat temperature estimate.

[0013] The low-frequency and high-frequency throat temperature estimates are input into an online fusion network to obtain the final high-frequency temperature estimate. The online fusion network is used to calculate the temperature correction value for the two low-frequency times by obtaining the current high-frequency temperature estimate and the low-frequency temperature estimates of the two closest low-frequency times. The temperature correction value is measured by the update gate weight and the forget gate weight, which are summed to 1. Finally, the correction value is multiplied by the over-selection weight and summed with the current high-frequency temperature estimate to obtain the final high-frequency throat temperature estimate.

[0014] Optionally, based on the material distribution pattern, the principle of equal volume, and the descent pattern, the blast furnace equipment parameters and the material distribution matrix are used as input to calculate the material surface shape of each layer inside the blast furnace, including:

[0015] Based on the blast furnace equipment parameters, the coordinate position of the tip of the pile formed by the furnace charge inside the blast furnace can be calculated to obtain the shape of the material surface. The coordinate position consists of an abscissa and a ordinate, which are calculated based on the principle that the volume of the single ring furnace charge in the charge distribution matrix is ​​equal to the volume between the two material surface shapes.

[0016] Using the material surface formed by the previous inclination angle as the initial material surface shape, and following the order of the fabric matrix, calculate the material surface shape from the second chute inclination angle to the last chute inclination angle in turn. By completing a cycle of the fabric matrix, repeat this step to obtain the material surface shape of each layer.

[0017] Optionally, the coke ratio parameters of each layer are calculated based on the shape of the material surface, including:

[0018] Based on the material surface shape, the calculation formulas for the coke ratio parameters of each layer are as follows:

[0019]

[0020] Where b represents the distance between a point and the centerline of the blast furnace, χ(b) η Let represent the shape function of the material surface of the ηth layer, M represent the selected layer number, η and M are positive integers, the subscript o represents the ore layer, and c represents the coke layer.

[0021] Optionally, considering the multi-timescale characteristics of the reaction inside the blast furnace, the blast furnace master parameters, blast furnace throat temperature measurement point data, and ore-coke ratio parameters are divided into high-frequency datasets and low-frequency datasets, including:

[0022] Based on the multi-timescale characteristics of the internal reaction of the blast furnace, a high-frequency dataset and a low-frequency dataset were constructed with two different sampling frequencies. The high-frequency dataset contains the main parameters of the blast furnace and the temperature measurement point data of the furnace throat, while the low-frequency dataset contains the main parameters of the blast furnace and the temperature measurement point data of the furnace throat and the ore-coke ratio parameter.

[0023] Optionally, feature importance filtering is performed on the two datasets separately to obtain a high-frequency dataset and a low-frequency dataset after feature filtering, including:

[0024] Feature importance screening was performed on high-frequency and low-frequency datasets separately. For high-frequency datasets, feature importance scores based on random forests were used for feature screening; for low-frequency datasets, feature screening was performed based on expert experience and literature.

[0025] Optionally, a support vector regressor is trained using the feature-selected low-frequency dataset to obtain a low-frequency furnace throat temperature estimation model, and an ensemble gated recurrent unit network is trained using the feature-selected high-frequency dataset to obtain a high-frequency furnace throat temperature estimation model, including:

[0026] A support vector regressor is trained using the low-frequency dataset after feature filtering to obtain a low-frequency furnace throat temperature estimation model.

[0027] An integrated gated recurrent unit network was trained using the high-frequency dataset after feature filtering to obtain a high-frequency furnace throat temperature estimation model.

[0028] Optionally, the low-frequency dataset to be tested is used as the input of the low-frequency furnace throat temperature estimation model to obtain the low-frequency furnace throat temperature estimate, and the high-frequency dataset to be tested is used as the input of the high-frequency furnace throat temperature estimation model to obtain the high-frequency furnace throat temperature estimate.

[0029] Alternatively, the calculation formula for the online converged network is as follows:

[0030] Based on update and forget gates, an online fusion network is proposed, and its calculation formula is as follows:

[0031]

[0032]

[0033]

[0034] λ u =tanh(tt) l )

[0035] λ u +λ f =1

[0036] Among them, t l t represents the lowest frequency moment closest to the current time. l-1 Then it means t l In the preceding low-frequency moment, x represents the input blast furnace sample features, p represents the high-frequency blast furnace throat temperature estimation model, s represents the low-frequency blast furnace throat temperature estimation model, and F(x) represents the previous low-frequency moment. t The final high-frequency temperature estimate is the output of the online fusion network. and They represent the values ​​at t l-1 and t l The correction value at time λ u To update the door, indicating The importance of λ u The value of changes from 0 to 1 as time increases, when t = t l The value is 0 when the next low-frequency moment arrives, and 1 when the next low-frequency moment arrives; λ f The Gate of Oblivion signifies The importance of λ u The value of changes from 1 to 0 as time increases, when t = t l The value is 1 when the next low-frequency moment arrives, and 0 when the next low-frequency moment arrives; w o To avoid overweighting, since all correction values ​​are derived from information from past time steps, w is set... o Information is re-filtered for options.

[0037] According to a second aspect of the embodiments of this application, a blast furnace throat temperature estimation device based on online fusion is provided, comprising:

[0038] The acquisition module is used to acquire blast furnace equipment parameters, charging matrix, blast furnace main parameters, and blast furnace throat temperature measurement point data;

[0039] The first calculation module is used to calculate the material surface shape of each layer inside the blast furnace based on the material distribution pattern, the principle of equal volume and the descent pattern, taking the blast furnace equipment parameters and the material distribution matrix as input, and calculating the ore-coke ratio parameter of each layer based on the material surface shape.

[0040] The partitioning module is used to divide the blast furnace main parameters, blast furnace throat temperature measurement point data and ore-coke ratio parameters into high-frequency datasets and low-frequency datasets based on the multi-timescale characteristics of the reaction inside the blast furnace.

[0041] The filtering module is used to filter the two datasets by feature importance, resulting in a high-frequency dataset and a low-frequency dataset after feature filtering.

[0042] The training module is used to train a support vector regressor using a low-frequency dataset after feature selection to obtain a low-frequency furnace throat temperature estimation model, and to train an integrated gated recurrent unit network using a high-frequency dataset after feature selection to obtain a high-frequency furnace throat temperature estimation model.

[0043] The second calculation module is used to input the low-frequency dataset to be measured into the low-frequency furnace throat temperature estimation model to obtain the low-frequency furnace throat temperature estimate, and to input the high-frequency dataset to be measured into the high-frequency furnace throat temperature estimation model to obtain the high-frequency furnace throat temperature estimate.

[0044] The estimation module is used to input the low-frequency and high-frequency furnace throat temperature estimates into the online fusion network to obtain the final high-frequency temperature estimate. The online fusion network is used to calculate the temperature correction value for the two low-frequency times by obtaining the high-frequency temperature estimate at the current time and the low-frequency temperature estimates at the two closest low-frequency times. The temperature correction value is measured by the update gate weight and the forget gate weight, which are summed to 1 respectively. Finally, the correction value is multiplied by the overdue weight and summed with the high-frequency temperature estimate at the current time to obtain the final high-frequency furnace throat temperature estimate.

[0045] According to a third aspect of the embodiments of this application, an electronic device is provided, the electronic device comprising:

[0046] One or more processors;

[0047] Storage device for storing one or more programs.

[0048] When the one or more programs are executed by the one or more processors, the one or more processors perform the method as described in the first aspect.

[0049] According to a fourth aspect of the embodiments of this application, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that the program, when executed by a processor, implements the method described in the first aspect.

[0050] The technical solutions provided by the embodiments of this application may include the following beneficial effects:

[0051] As can be seen from the above embodiments, this application establishes a mathematical model of the blast furnace burden distribution by combining the movement law of the burden layer inside the blast furnace, calculates the ore-coke ratio parameters of each layer inside the furnace, and obtains high-frequency and low-frequency datasets based on the multi-timescale characteristics of the reaction inside the blast furnace. High-frequency estimation models and low-frequency estimation models are trained respectively. The low-frequency dataset to be measured is used as the input of the low-frequency throat temperature estimation model to obtain the low-frequency throat temperature estimate. The high-frequency dataset to be measured is used as the input of the high-frequency throat temperature estimation model to obtain the high-frequency throat temperature estimate. Finally, the low-frequency throat temperature estimate and the high-frequency throat temperature estimate are input into an online fusion network to obtain the final high-frequency temperature estimate, which is used to perform high-frequency real-time estimation of the temperature at the corresponding position of the throat.

[0052] When the cross temperature measuring device in the blast furnace throat malfunctions, this invention can estimate the blast furnace throat temperature in real time using modern methods. This can effectively help on-site personnel to more accurately judge the distribution of blast furnace gas flow, which is of great practical significance for ensuring smooth flow inside the blast furnace.

[0053] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0054] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0055] Figure 1 This is a flowchart illustrating an online fusion-based method for estimating the throat temperature of a blast furnace, according to an exemplary embodiment.

[0056] Figure 2 This is a flowchart illustrating an online fusion-based blast furnace throat temperature estimation according to an exemplary embodiment.

[0057] Figure 3 This is a graph showing the results of blast furnace throat temperature estimation using four methods according to an exemplary embodiment.

[0058] Figure 4 This is a graph showing the blast furnace throat temperature estimation error results of four methods according to an exemplary embodiment.

[0059] Figure 5 This is a block diagram illustrating an online fusion-based blast furnace throat temperature estimation device according to an exemplary embodiment.

[0060] Figure 6This is a block diagram illustrating an electronic device according to an exemplary embodiment. Detailed Implementation

[0061] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0062] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0063] Figure 1 This is a flowchart illustrating an online fusion-based method for estimating blast furnace throat temperature according to an exemplary embodiment, such as... Figure 1 As shown, the method may include the following steps:

[0064] S1: Obtain blast furnace equipment parameters, charging matrix, blast furnace main parameters, and blast furnace throat temperature measurement point data;

[0065] S2: Based on the material distribution pattern, the principle of equal volume and the descent pattern, the blast furnace equipment parameters and the material distribution matrix are used as inputs to calculate the material surface shape of each layer inside the blast furnace, and the ore-coke ratio parameter of each layer is calculated based on the material surface shape.

[0066] S3: Considering the multi-timescale characteristics of the reaction inside the blast furnace, the main parameters of the blast furnace, the temperature measurement point data of the blast furnace throat, and the ore-coke ratio parameters are divided into high-frequency datasets and low-frequency datasets.

[0067] S4: Perform feature importance filtering on the two datasets to obtain the high-frequency dataset and low-frequency dataset after feature filtering;

[0068] S5: Train a support vector regressor using the low-frequency dataset after feature selection to obtain a low-frequency furnace throat temperature estimation model; train an integrated gated recurrent unit network using the high-frequency dataset after feature selection to obtain a high-frequency furnace throat temperature estimation model.

[0069] S6: Use the low-frequency dataset to be tested as the input of the low-frequency furnace throat temperature estimation model to obtain the low-frequency furnace throat temperature estimate; use the high-frequency dataset to be tested as the input of the high-frequency furnace throat temperature estimation model to obtain the high-frequency furnace throat temperature estimate.

[0070] S7: Input the low-frequency and high-frequency throat temperature estimates into the online fusion network to obtain the final high-frequency temperature estimate, which is then used to perform high-frequency real-time estimation of the temperature at the corresponding position of the throat. The online fusion network obtains the current high-frequency temperature estimate and the low-frequency temperature estimates of the two closest low-frequency times, calculates the temperature correction values ​​for the two low-frequency times, measures the temperature correction values ​​by using update gate weights and forget gate weights that are summed to 1, and finally multiplies the correction values ​​by the overdue weights and sums them with the current high-frequency temperature estimate to obtain the final high-frequency throat temperature estimate.

[0071] As can be seen from the above technical solutions, this application establishes a mathematical model of the blast furnace burden distribution by combining the movement law of the internal burden layer, calculates the ore-coke ratio parameters of each layer in the furnace, and obtains high-frequency and low-frequency datasets based on the multi-timescale characteristics of the internal reaction of the blast furnace. High-frequency estimation models and low-frequency estimation models are trained respectively. The low-frequency dataset to be measured is used as the input of the low-frequency throat temperature estimation model to obtain the low-frequency throat temperature estimate. The high-frequency dataset to be measured is used as the input of the high-frequency throat temperature estimation model to obtain the high-frequency throat temperature estimate. Finally, the low-frequency throat temperature estimate and the high-frequency throat temperature estimate are input into an online fusion network to obtain the final high-frequency temperature estimate, which is used to perform high-frequency real-time estimation of the temperature at the corresponding position of the throat.

[0072] In the specific implementation of S2: based on the material distribution pattern, the principle of equal volume, and the descent pattern, the blast furnace equipment parameters and the material distribution matrix are used as inputs to calculate the material surface shape of each layer inside the blast furnace, and the ore-coke ratio parameter of each layer is calculated based on the material surface shape. This step may include the following sub-steps:

[0073] S21: Based on the blast furnace equipment parameters, calculate the coordinate position of the tip of the material pile formed inside the blast furnace to obtain the shape of the material surface layer;

[0074] Specifically, based on the blast furnace equipment parameters, through force analysis, the abscissa of the pile tip formed by the furnace charge reaching the charge surface in the blast furnace radius direction is calculated, and the ordinate is calculated based on the principle that the volume of the single ring charge in the charge distribution matrix is ​​equal to the volume between the two successive charge surface shapes.

[0075] S22: Using the material surface formed by the previous inclination angle as the initial material surface shape, and following the order of the fabric matrix, calculate the material surface shape from the second chute inclination angle to the last chute inclination angle sequentially. Repeat this step by completing a loop of the fabric matrix to obtain the material surface shape of each layer.

[0076] S23: Based on the shape of the material surface, the calculation formula for the coke ratio parameter of each layer is as follows:

[0077]

[0078] Where b represents the distance between a point and the centerline of the blast furnace, χ(b) η Let represent the shape function of the material surface of the ηth layer, M represent the selected layer number, η and M are positive integers, the subscript o represents the ore layer, and c represents the coke layer.

[0079] In the specific implementation of S3: Considering the multi-timescale characteristics of the reaction inside the blast furnace, the blast furnace master parameters, blast furnace throat temperature measurement point data, and ore-coke ratio parameters are divided into high-frequency datasets and low-frequency datasets. This step may include the following sub-steps:

[0080] S31: Based on the multi-timescale characteristics of the internal reaction of the blast furnace, a high-frequency dataset and a low-frequency dataset are constructed with two different sampling frequencies. The high-frequency dataset contains the main parameters of the blast furnace and the temperature measurement point data of the furnace throat, while the low-frequency dataset contains the main parameters of the blast furnace and the temperature measurement point data of the furnace throat and the ore-coke ratio parameter.

[0081] Specifically, the high-frequency dataset is sampled at a high time frequency and has rich time-scale information; the low-frequency dataset not only includes blast furnace main parameters and throat temperature measurement point data, but also ore-coke ratio parameters, and has rich feature information.

[0082] In the specific implementation of S4: feature importance filtering is performed on the two datasets separately to obtain the high-frequency dataset and low-frequency dataset after feature filtering. This step may include the following sub-steps:

[0083] S41: Feature importance screening is performed on high-frequency and low-frequency datasets respectively. For high-frequency datasets, feature importance scores based on random forests are used for feature screening. For low-frequency datasets, feature screening is performed based on expert experience and literature.

[0084] Specifically, based on the feature importance scores of random forests, the specific steps include:

[0085] Suppose there are p trees in the forest. For the i-th decision tree, a random sample is drawn from all sample sets. The tree is built using the randomly drawn sample, and the mean squared error between the out-of-bag data and the true value is calculated, denoted as .

[0086] random permutation variable X j The observed values ​​are used to rebuild the tree and calculate the data error of the out-of-bag data, denoted as .

[0087] Calculate the difference between the two data errors, and after standardization, obtain the variable X. j The importance score of a variable in a random forest is obtained by taking its contribution in the i-th decision tree, i.e., its importance, and then averaging the contribution across all trees. The formula is as follows:

[0088]

[0089] VIM j Represents variable X j Importance score.

[0090] Based on the importance scores, the top L feature variables with the highest importance scores are selected as the filtered features, in descending order.

[0091] In the specific implementation of S5: a support vector regressor is trained using the low-frequency dataset after feature selection to obtain a low-frequency furnace throat temperature estimation model; an ensemble gated recurrent unit network is trained using the high-frequency dataset after feature selection to obtain a high-frequency furnace throat temperature estimation model. This step may include the following sub-steps:

[0092] S51: A support vector regressor is trained using the low-frequency dataset after feature selection to obtain a low-frequency furnace throat temperature estimation model. The calculation process of support vector regression is expressed as follows:

[0093] Given a blast furnace training sample D = {(x1,y1),(x2,y2),…,(x m ,y m )},y i ∈R, the support vector regression problem is represented as:

[0094]

[0095] Where w is the weight vector, b is the bias, and w and b together represent a hyperplane, ξ i and Let C be the slack variable, C be the regularization constant, and ∈ be the tolerance bias.

[0096] The high-frequency furnace throat temperature estimation model trained on the high-frequency dataset after feature filtering contains richer feature information. Therefore, the model has higher accuracy and smaller error in temperature estimation at low frequencies.

[0097] S52: Train an integrated gated recurrent unit network using the high-frequency dataset after feature filtering to obtain a high-frequency furnace throat temperature estimation model. This includes: dividing the high-frequency dataset into different training and validation sets multiple times; training multiple gated recurrent unit network base models with different parameters using different training and validation sets; and obtaining a high-frequency furnace throat temperature estimation model based on all the gated recurrent unit network base models.

[0098] Specifically, the high-frequency dataset is divided into k parts, with one part selected as the validation set and the remaining k-1 parts used as the training set. This yields k different training and validation sets. Using these k different training and validation sets, k gated recurrent unit network (GRU) base models with different parameters are trained. Based on these k GRU base models, a high-frequency furnace throat temperature estimation model is obtained, expressed as follows:

[0099]

[0100] Where t represents the estimated time, x t u represents the feature at time t of the high-frequency dataset. k This represents the k-th gated recurrent unit network base model after training. Let represent the high-frequency throat temperature estimate calculated by the k-th gated recurrent unit network base model after training at time t. u represents the final blast furnace throat temperature estimation model. The final high-frequency throat temperature estimate is obtained by calculating the mean of the throat temperature estimates from the k base models. This represents the estimated temperature of the blast furnace throat at time t.

[0101] Specifically, the computation process of the gated recurrent unit network is expressed as follows:

[0102] Gated cyclic unit networks control the way information is updated by introducing a gating mechanism, including updating the gating gate. t and reset door r t The update gate controls how much information to retain from historical states for the current state, and how much new information to accept from candidate states as the current state h. t ,Right now

[0103] h t =z t ⊙h t-1 +(1-z t )⊙g(x t ,h t-1 ;θ)

[0104] Among them, z t For updating the gate:

[0105] z t =σ(W z xt +U z h t-1 +b z )

[0106] Among them, the function g(x) t ,h t-1 ;θ) is defined as

[0107]

[0108] in, Let r represent the candidate state at the current moment. t This is a reset gate used to control candidate states. Does the calculation depend on the state h from the previous time step? t-1 Its expression is as follows:

[0109] r t =σ(W r x t +U r h t-1 +b r )

[0110] Therefore, the state update method for gated recurrent networks is as follows:

[0111]

[0112] Specifically, the output of the high-frequency furnace throat temperature estimation model is the average of the furnace throat temperature estimates output by all base models.

[0113] The high-frequency throat temperature estimation model trained on the high-frequency dataset after feature filtering contains richer time-scale information. Therefore, compared with the low-frequency throat temperature estimation model, this model can estimate the temperature at more times and can more accurately capture the trend of throat temperature change over time.

[0114] In the specific implementation of S6: the low-frequency dataset to be tested is input into the low-frequency furnace throat temperature estimation model to obtain the low-frequency furnace throat temperature estimate; the high-frequency dataset to be tested is input into the high-frequency furnace throat temperature estimation model to obtain the high-frequency furnace throat temperature estimate.

[0115] This step allows us to obtain estimated values ​​for both low-frequency and high-frequency furnace throat temperatures.

[0116] In the specific implementation of S7: the low-frequency and high-frequency furnace throat temperature estimates are input into the online fusion network to obtain the final high-frequency temperature estimate, which is then used to perform high-frequency real-time estimation of the temperature at the corresponding position of the furnace throat. The online fusion network obtains the current high-frequency temperature estimate and the low-frequency temperature estimates of the two closest low-frequency moments, calculates the temperature correction values ​​for the two low-frequency moments, and measures the temperature correction values ​​using update gate weights and forget gate weights that are summed to 1. Finally, the correction values ​​are multiplied by the excess weight and summed with the current high-frequency temperature estimate to obtain the final high-frequency furnace throat temperature estimate.

[0117] Specifically, based on the update gate and the forget gate, an online fusion network is proposed, and its calculation formula is as follows:

[0118]

[0119]

[0120]

[0121] λ u =tanh(tt) l )

[0122] λ u +λ f =1

[0123] Among them, t l t represents the lowest frequency moment closest to the current time. l-1 Then it means t l In the preceding low-frequency moment, x represents the input blast furnace sample features, p represents the high-frequency blast furnace throat temperature estimation model, s represents the low-frequency blast furnace throat temperature estimation model, and F(x) represents the previous low-frequency moment. t The final high-frequency temperature estimate is the output of the online fusion network. and They represent the values ​​at t l-1 and t l The correction value at time λ u To update the door, indicating The importance of λ u The value of changes from 0 to 1 as time increases, when t = t l The value is 0 when the next low-frequency moment arrives, and 1 when the next low-frequency moment arrives; λ f The Gate of Oblivion signifies The importance of λ u The value of changes from 1 to 0 as time increases, when t = t l The value is 1 when the next low-frequency moment arrives, and 0 when the next low-frequency moment arrives; w oTo avoid overweighting, since all correction values ​​are derived from information from past time steps, w is set... o Information is re-filtered for options.

[0124] The final high-frequency temperature estimate obtained through the online fusion network combines the advantages of both the high-frequency and low-frequency throat temperature estimation models. It includes the advantage of the low-frequency throat temperature model having higher accuracy in temperature estimation at low frequencies, as well as the advantage of the high-frequency throat temperature estimation model being able to estimate temperatures at more times and more accurately capture the trend of throat temperature changes over time.

[0125] Figure 2 A flowchart of the blast furnace throat temperature estimation based on online fusion is provided in this embodiment of the invention. The high-frequency and low-frequency datasets are obtained after feature filtering. A support vector regressor is trained using the low-frequency dataset to obtain a low-frequency temperature estimation model, and an ensemble gated recurrent unit network is trained using the high-frequency dataset to obtain a high-frequency temperature estimation model. The low-frequency dataset to be tested is used as input to the low-frequency throat temperature estimation model to obtain a low-frequency throat temperature estimate. The high-frequency dataset to be tested is used as input to the high-frequency throat temperature estimation model to obtain a high-frequency throat temperature estimate. Finally, the low-frequency and high-frequency throat temperature estimates are input to the online fusion network to obtain the final high-frequency temperature estimate, which is used for real-time high-frequency estimation of the temperature at the corresponding location in the throat.

[0126] In this embodiment, blast furnace master parameter data, throat temperature measurement point data, and blast furnace charging matrix data for a specific month were selected from a blast furnace database in South China. A total of 4320 sets of blast furnace master parameter data and throat temperature measurement point data were used. The first 3840 sets of samples were used for training, and the last 480 sets were used for testing. Based on the sampling frequency, 3840 sets of high-frequency training datasets, 480 sets of high-frequency test datasets, 160 sets of low-frequency training datasets, and 20 sets of low-frequency test datasets were obtained.

[0127] To conduct a more comprehensive analysis and discussion, four methods were used to estimate the furnace throat temperature:

[0128] (1) Furnace throat temperature estimation method based on support vector regression: After feature selection, the high-frequency dataset is trained using support vector regression, and the furnace throat temperature is estimated by the trained support vector regression model.

[0129] (2) Furnace throat temperature estimation method based on gated recurrent unit network: After feature selection, the high-frequency dataset is trained using a gated recurrent unit network, and the furnace throat temperature is estimated by the trained gated recurrent unit network model.

[0130] (3) Furnace throat temperature estimation method based on integrated gated recurrent unit network: After feature selection, the high-frequency dataset is trained using an integrated gated recurrent unit network, and the furnace throat temperature is estimated by the trained integrated gated recurrent unit network model.

[0131] (4) The method proposed in this embodiment of the invention: After feature selection, the low-frequency dataset is trained using support vector regression to obtain a low-frequency temperature estimation model. After feature selection, the high-frequency dataset is trained using an integrated gated recurrent unit network to obtain a high-frequency temperature estimation model. The low-frequency dataset to be tested is used as the input of the low-frequency furnace throat temperature estimation model to obtain a low-frequency furnace throat temperature estimate. The high-frequency dataset to be tested is used as the input of the high-frequency furnace throat temperature estimation model to obtain a high-frequency furnace throat temperature estimate. Finally, the low-frequency furnace throat temperature estimate and the high-frequency furnace throat temperature estimate are input into an online fusion network to obtain the final high-frequency temperature estimate.

[0132] The estimation performance of the model is evaluated using statistical indicators such as MAPE (Mean Absolute Percentage Error), MAE (Mean Absolute Error), and RMSE (Root Mean Square Error). The calculation methods for these three indicators are shown below:

[0133]

[0134]

[0135]

[0136] Where y(i) is the measured value of the furnace throat temperature. T represents the estimated value of the furnace throat temperature measurement, and T is the sample size.

[0137] The temperature estimation results of the four methods are as follows Figure 3 As shown, the error results between the temperature estimation and measurement values ​​obtained by the four methods are as follows: Figure 4 As shown. SVR is a furnace temperature estimation method based on support vector regression, GRU is a furnace throat temperature estimation method based on gated cyclic unit network, I-GRU is a furnace throat temperature estimation method based on integrated gated cyclic unit network, and I-GRU_MTSF is the method proposed in this invention. The evaluation index results are as follows:

[0138]

[0139] like Figure 3 and Figure 4As shown, the SVR method can roughly reflect the trend of the true temperature, but the difference from the true temperature value is relatively large. The GRU method captures the changes in the true temperature faster than the SVR method, and the error is further reduced. The I-GRU method is closer to the actual measurement value and more stable, proving the effectiveness of the improvement to the GRU algorithm. I-GRU_MTSF is closer to the actual measurement value than other methods, proving the effectiveness of the method proposed in this invention.

[0140] As can be seen from the table, the proposed method has smaller errors compared to the other three control methods, with MAPE of 2.728%, MAE of 1.519℃, and RMSE of 1.850℃, all reaching the best among the four methods, thus providing evidence that... Figure 3 , Figure 4 The conclusions demonstrate that the temperature estimation method proposed in this invention is superior to other methods.

[0141] Corresponding to the aforementioned embodiment of a blast furnace throat temperature estimation method based on online fusion, this application also provides an embodiment of a blast furnace throat temperature estimation device based on online fusion.

[0142] Figure 5 This is a block diagram illustrating an online fusion-based blast furnace throat temperature estimation device according to an exemplary embodiment. (Refer to...) Figure 5 The device includes:

[0143] Module 1 is used to acquire blast furnace equipment parameters, charging matrix, blast furnace main parameters, and blast furnace throat temperature measurement point data;

[0144] The first calculation module 2 is used to calculate the material surface shape of each layer inside the blast furnace based on the material distribution pattern, the principle of equal volume and the descent pattern, taking the blast furnace equipment parameters and the material distribution matrix as input, and to calculate the ore-coke ratio parameter of each layer based on the material surface shape.

[0145] The partitioning module 3 is used to partition the blast furnace main parameters, blast furnace throat temperature measurement point data and ore-coke ratio parameters into high-frequency datasets and low-frequency datasets based on the multi-timescale characteristics of the reaction inside the blast furnace.

[0146] The filtering module 4 is used to filter the feature importance of the two datasets separately, and obtain the high-frequency dataset and the low-frequency dataset after feature filtering.

[0147] Training module 5 is used to train a support vector regressor using the low-frequency dataset after feature selection to obtain a low-frequency furnace throat temperature estimation model, and to train an integrated gated recurrent unit network using the high-frequency dataset after feature selection to obtain a high-frequency furnace throat temperature estimation model.

[0148] The second calculation module 6 is used to input the low-frequency dataset to be measured into the low-frequency furnace throat temperature estimation model to obtain the low-frequency furnace throat temperature estimate, and to input the high-frequency dataset to be measured into the high-frequency furnace throat temperature estimation model to obtain the high-frequency furnace throat temperature estimate.

[0149] The estimation module 7 is used to input the low-frequency and high-frequency furnace throat temperature estimates into the online fusion network to obtain the final high-frequency temperature estimate. The online fusion network is used to calculate the temperature correction value of the two low-frequency times by obtaining the high-frequency temperature estimate at the current time and the low-frequency temperature estimates at the two closest low-frequency times. The temperature correction value is measured by the update gate weight and the forget gate weight, which are summed to 1 respectively. Finally, the correction value is multiplied by the overdue weight and summed with the high-frequency temperature estimate at the current time to obtain the final high-frequency furnace throat temperature estimate.

[0150] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0151] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0152] Accordingly, this application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; and, when the one or more programs are executed by the one or more processors, causing the one or more processors to implement the above-described online fusion-based blast furnace throat temperature estimation method. Figure 6 The diagram shown is a hardware structure diagram of any device with data processing capabilities, including a blast furnace throat temperature estimation device based on online fusion provided in an embodiment of the present invention. (Except for...) Figure 6 In addition to the processor and memory shown, any data processing device in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0153] Accordingly, this application also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the above-described online fusion-based blast furnace throat temperature estimation method. The computer-readable storage medium can be an internal storage unit of any data-processing device as described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device for a wind turbine, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., mounted on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data-processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data-processing device, and can also be used to temporarily store data that has been output or will be output.

[0154] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

[0155] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for estimating blast furnace throat temperature based on online fusion, characterized in that, include: Acquire blast furnace equipment parameters, charging matrix, blast furnace main parameters, and blast furnace throat temperature measurement point data; Based on the material distribution pattern, the principle of equal volume, and the descent pattern, the blast furnace equipment parameters and the material distribution matrix are used as inputs to calculate the material surface shape of each layer inside the blast furnace, and the ore-coke ratio parameter of each layer is calculated based on the material surface shape. Considering the multi-timescale characteristics of the internal reaction of the blast furnace, the main parameters of the blast furnace, the temperature measurement point data of the blast furnace throat, and the ore-coke ratio parameters are divided into high-frequency datasets and low-frequency datasets. Feature importance filtering is performed on the two datasets separately to obtain the high-frequency dataset and the low-frequency dataset after feature filtering; A support vector regressor is trained using the low-frequency dataset after feature selection to obtain a low-frequency furnace throat temperature estimation model. An integrated gated recurrent unit network is trained using the high-frequency dataset after feature selection to obtain a high-frequency furnace throat temperature estimation model. Input the low-frequency dataset to be tested into the low-frequency furnace throat temperature estimation model to obtain the low-frequency furnace throat temperature estimate; input the high-frequency dataset to be tested into the high-frequency furnace throat temperature estimation model to obtain the high-frequency furnace throat temperature estimate. The low-frequency and high-frequency throat temperature estimates are input into an online fusion network to obtain the final high-frequency temperature estimate. The online fusion network is used to calculate the temperature correction value for the two low-frequency times by obtaining the current high-frequency temperature estimate and the low-frequency temperature estimates for the two closest low-frequency times. The temperature correction value is measured by the update gate weight and the forget gate weight, which are summed to 1. Finally, the correction value is multiplied by the overdue weight and summed with the current high-frequency temperature estimate to obtain the final high-frequency throat temperature estimate. Considering the multi-timescale characteristics of the internal reaction of the blast furnace, the blast furnace master parameters, blast furnace throat temperature measurement point data, and ore-coke ratio parameters are divided into high-frequency datasets and low-frequency datasets, including: Based on the multi-timescale characteristics of the internal reaction of the blast furnace, a high-frequency dataset and a low-frequency dataset were constructed with two different sampling frequencies. The high-frequency dataset contains the main parameters of the blast furnace and the temperature measurement point data of the furnace throat, while the low-frequency dataset contains the main parameters of the blast furnace and the temperature measurement point data of the furnace throat and the ore-coke ratio parameter.

2. The method according to claim 1, characterized in that, Based on the material distribution pattern, the principle of equal volume, and the descent pattern, the blast furnace equipment parameters and the material distribution matrix are used as inputs to calculate the material surface shape of each layer inside the blast furnace, including: Based on the blast furnace equipment parameters, the coordinate position of the tip of the pile formed by the furnace charge inside the blast furnace can be calculated to obtain the shape of one layer of charge surface. The coordinate position consists of the horizontal and vertical coordinates, which are calculated based on the principle that the volume of the single ring charge in the charge distribution matrix is ​​equal to the volume between the two successive charge surface shapes. Using the material surface formed by the previous inclination angle as the initial material surface shape, and following the order of the fabric matrix, calculate the material surface shape from the second chute inclination angle to the last chute inclination angle in turn to complete a cycle of the fabric matrix. Repeat the steps to obtain the material surface shape of each layer.

3. The method according to claim 1, characterized in that, Based on the shape of the material surface, calculate the coke ratio parameters for each layer, including: Based on the material surface shape, the calculation formulas for the coke ratio parameters of each layer are as follows: ； in, This represents the distance between a point and the centerline of the blast furnace. Indicates the first The shape function of the material surface of the layer, Indicates the selected number of material layers. and A positive integer, with index Indicates the ore layer. This indicates the coke layer.

4. The method according to claim 1, characterized in that, Feature importance filtering was performed on the two datasets separately to obtain the high-frequency and low-frequency datasets after feature filtering, including: Feature importance screening was performed on high-frequency and low-frequency datasets separately. For high-frequency datasets, feature importance scores based on random forests were used for feature screening; for low-frequency datasets, feature screening was performed based on expert experience and literature.

5. The method according to claim 1, characterized in that, The calculation formula for online converged networks is as follows: ； ； ； ； ； in, This represents the lowest frequency moment closest to the current time. Then it means The previous low-frequency moment, For the input blast furnace sample features, This is the high-frequency furnace throat temperature estimation model. For the low-frequency furnace throat temperature estimation model, The final high-frequency temperature estimate is the output of the online fusion network. and They represent in and Correction value at time, To update the door, indicating The importance of The value of changes from 0 to 1 as time increases. The value is 0 when the next low-frequency moment arrives, and 1 when the next low-frequency moment arrives. The Gate of Oblivion signifies The importance of The value of changes from 1 to 0 as time increases. The value is 1 when the next low-frequency moment arrives, and 0 when the next low-frequency moment arrives. To avoid overweighting, since all correction values ​​are derived from information from past time steps, a setting is made. Information is re-filtered for options.

6. A blast furnace throat temperature estimation device based on online fusion, characterized in that, include: The acquisition module is used to acquire blast furnace equipment parameters, charging matrix, blast furnace main parameters, and blast furnace throat temperature measurement point data; The first calculation module is used to calculate the material surface shape of each layer inside the blast furnace based on the material distribution pattern, the principle of equal volume and the descent pattern, taking the blast furnace equipment parameters and the material distribution matrix as input, and calculating the ore-coke ratio parameter of each layer based on the material surface shape. The partitioning module is used to divide the blast furnace main parameters, blast furnace throat temperature measurement point data and ore-coke ratio parameters into high-frequency datasets and low-frequency datasets based on the multi-timescale characteristics of the reaction inside the blast furnace. The filtering module is used to filter the two datasets by feature importance, resulting in a high-frequency dataset and a low-frequency dataset after feature filtering. The training module is used to train a support vector regressor using a low-frequency dataset after feature selection to obtain a low-frequency furnace throat temperature estimation model, and to train an integrated gated recurrent unit network using a high-frequency dataset after feature selection to obtain a high-frequency furnace throat temperature estimation model. The second calculation module is used to input the low-frequency dataset to be measured into the low-frequency furnace throat temperature estimation model to obtain the low-frequency furnace throat temperature estimate, and to input the high-frequency dataset to be measured into the high-frequency furnace throat temperature estimation model to obtain the high-frequency furnace throat temperature estimate. The estimation module is used to input the low-frequency and high-frequency furnace throat temperature estimates into the online fusion network to obtain the final high-frequency temperature estimate. The online fusion network is used to obtain the high-frequency temperature estimate at the current moment and the low-frequency temperature estimates at the two closest low-frequency moments, calculate the temperature correction values ​​at the two low-frequency moments, measure the temperature correction values ​​by the update gate weight and the forget gate weight that are summed to 1, and finally multiply by the over-weight and sum with the high-frequency temperature estimate at the current moment to obtain the final high-frequency furnace throat temperature estimate. Considering the multi-timescale characteristics of the internal reaction of the blast furnace, the blast furnace master parameters, blast furnace throat temperature measurement point data, and ore-coke ratio parameters are divided into high-frequency datasets and low-frequency datasets, including: Based on the multi-timescale characteristics of the internal reaction of the blast furnace, a high-frequency dataset and a low-frequency dataset were constructed with two different sampling frequencies. The high-frequency dataset contains the main parameters of the blast furnace and the temperature measurement point data of the furnace throat, while the low-frequency dataset contains the main parameters of the blast furnace and the temperature measurement point data of the furnace throat and the ore-coke ratio parameter.

7. An electronic device, characterized in that, The electronic device includes: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-5.

Images