Blast furnace air permeability index prediction method based on multi-scale time-delay characteristic mining and application

Abstract

The invention discloses a blast furnace air permeability index prediction method based on multi-scale time-delay characteristic mining and application, and belongs to the field of soft measurement modeling in the blast furnace ironmaking process. The invention provides a blast furnace air permeability index prediction framework combining wavelet decomposition, wavelet coherence analysis and a time-frequency fusion model aiming at the characteristics of nonlinearity, unsteady state and large time delay of blast furnace operation data. The method comprises the following steps: firstly, carrying out peak separation and multi-scale decomposition on data, and extracting multi-scale time delay information among variables by utilizing wavelet coherence analysis; and then modeling time domain global features and frequency domain local features by adopting orthogonal subspace analysis and a wavelet neural network, and performing model fusion through Gauss-Markov estimation to realize advanced multi-step prediction of the air permeability index. The dynamic time lag characteristic of the blast furnace process can be effectively captured, the prediction precision and robustness are improved, and technical support is provided for stable operation of the blast furnace.

Description

Technical Field

[0001] This invention belongs to the field of soft measurement modeling technology for blast furnace ironmaking process, specifically involving a method and application for predicting blast furnace permeability index based on multi-scale time delay characteristic mining. Background Technology

[0002] Blast furnace ironmaking is a crucial link in the modern steel manufacturing process, undertaking the core task of smelting iron ore into pig iron. This process involves extremely complex heat, momentum, and mass transfers, as well as various chemical reactions, making it a typical complex industrial system involving multiphase and multi-field coupling. Its energy consumption accounts for more than 70% of the total energy consumption in steel production. Therefore, achieving stable, smooth, and efficient optimized operation of blast furnaces is of great strategic significance for energy conservation, emission reduction, cost reduction, and efficiency improvement throughout the entire steel industry.

[0003] Among the numerous operating parameters of a blast furnace, the permeability index is widely recognized as a crucial comprehensive condition indicator. It directly reflects the smoothness of the gas flow through the solid burden bed within the furnace. The permeability index is the ratio of blast volume to total differential pressure. A stable permeability index is a fundamental prerequisite for smooth blast furnace operation; abnormal fluctuations are often precursors to furnace malfunctions such as burden suspension, collapse, and pipeline travel. Therefore, accurate and timely monitoring and prediction of the permeability index can provide operators with critical decision support, allowing them to adjust operating parameters (such as blast volume, blast temperature, and oxygen enrichment) before serious failures occur, thereby ensuring the long-term stable operation of the blast furnace and avoiding significant economic losses.

[0004] However, achieving accurate prediction of the permeability index, especially in real industrial environments with multi-step advance prediction, faces severe technical challenges. This is mainly due to the inherent complexity of blast furnace data, including strong nonlinearity, non-stationarity, and multi-scale large time delay characteristics. In recent years, with the development of industrial big data technology, data-driven methods have become the mainstream research approach. In most application scenarios, data-driven methods have demonstrated a powerful ability to handle strongly nonlinear and non-stationary process data. However, for blast furnace ironmaking scenarios, existing methods still have significant limitations: First, most studies lack specific processing for non-stationary spike signals caused by hot blast stove switching, making the model susceptible to interference; second, and more critically, they generally ignore or fail to deeply explore the scale-dependent inherent time delay characteristics between different process variables and the permeability index, resulting in a misalignment between model input and output in the time dimension, limiting prediction accuracy, especially the performance of multi-step advance prediction; finally, many models suffer from singularity in feature extraction, failing to comprehensively characterize the spatial characteristics of the data from multiple perspectives, including time and frequency domains, global and local data.

[0005] Therefore, developing a method for predicting the permeability index of blast furnaces that can effectively handle non-stationarity and accurately mine multi-scale time delay characteristics has become a technical bottleneck that urgently needs to be solved in this field. This has significant theoretical value and engineering application prospects for improving the intelligent operation level of blast furnaces. Summary of the Invention

[0006] To overcome the shortcomings of existing modeling techniques, this invention provides a method and application for predicting blast furnace permeability index based on multi-scale time delay characteristic mining. It can achieve accurate modeling of blast furnace permeability index and has high prediction accuracy for blast furnace permeability index with different lead times.

[0007] To achieve the above-mentioned technical objectives, the technical solution of the present invention is as follows:

[0008] A method for predicting blast furnace permeability index based on multi-scale time delay characteristics includes the following steps:

[0009] Determine the operating parameters related to the blast furnace permeability index, read the relevant blast furnace operating data from the database, and perform data noise reduction.

[0010] The operating parameters and air permeability index data are simultaneously subjected to peak separation, decomposed into stationary sequences and peak components;

[0011] Multi-scale wavelet decomposition of stationary sequences yields detail components and approximate components at different time scales.

[0012] At each time scale, wavelet coherence analysis is used to extract the time delay information between each operating parameter and the air permeability index, and the decomposition sequence at the corresponding scale is time-aligned based on the time delay information to construct training samples.

[0013] A time-frequency fusion prediction model based on orthogonal subspace analysis and wavelet neural network was constructed and trained using the training samples. The outputs of orthogonal subspace analysis and wavelet neural network at the same time scale were fused using Gauss-Markov estimation. The fusion results at different time scales were summed to obtain the final predicted value of the blast furnace permeability index.

[0014] The operating parameters include cold air flow rate, oxygen enrichment rate, blast energy, furnace gas index, total differential pressure, blast pressure, hot air temperature, and top temperature.

[0015] The peak separation includes: Gaussian smoothing of the original signal; detection of peak extrema based on second-order difference and threshold; determination of the peak start and end intervals by combining first-order difference; and fitting of the peak region with a quadratic spline curve to preserve the trend and suppress the amplitude.

[0016] The aforementioned multi-scale wavelet decomposition of stationary sequences refers to forming a series of wavelet functions by scaling and shifting the basic wavelet function in different ways, and then performing an inner product with the original signal to obtain the detail components and approximate components of the original signal at different scaling scales and decomposition levels.

[0017] The wavelet coherence analysis includes: calculating the cross wavelet transform between the input variable and the permeability index; determining the time delay relationship based on the peak value of the wavelet coherence coefficient; and adjusting the time series to achieve alignment.

[0018] The aforementioned construction of a time-frequency fusion prediction model based on orthogonal subspace analysis and wavelet neural networks specifically involves:

[0019] (1) The orthogonal subspace analysis algorithm is used to analyze the global time-domain information of the data. It decomposes the process data time series X and the key performance indicator permeability index, and the time series Y data into three orthogonal subspaces, and solves the following projection problem to build a prediction model:

[0020]

[0021] in, Θ is the estimate of Y obtained through orthogonal subspace analysis. X→V and Θ Y→V V are the score matrices V of X and Y projected onto the common subspace. com The projection transformation matrix;

[0022] (2) The wavelet neural network is used to capture local frequency domain features more effectively. The neural network is constructed by replacing the traditional activation function used by the hidden layer neurons in the backpropagation neural network with the Morlet wavelet function as shown in the following formula:

[0023]

[0024] The Gauss-Markov estimation method described above fuses the outputs of different sub-models by solving the following equation to obtain the overall prediction result of the hybrid model:

[0025]

[0026] in, Let V be the output error variance of the i-th sub-model. This is the output of the i-th sub-model.

[0027] A computer system for implementing the method includes:

[0028] The data acquisition module is used to obtain operating parameters and permeability index data from the blast furnace operating system;

[0029] The preprocessing module is used to perform transient spike separation and wavelet denoising;

[0030] The wavelet decomposition module is used to perform multi-scale decomposition.

[0031] The time delay analysis and alignment module is used to calculate wavelet coherence and perform time alignment;

[0032] The time-frequency fusion module is used to run the orthogonal subspace analysis model and the wavelet neural network model respectively;

[0033] The fusion decision module is used to perform Gaussian-Markov estimation and scaling summation;

[0034] The output module is used to output the multi-step predicted value of the blast furnace permeability index.

[0035] A computer-readable storage medium having program instructions stored thereon, which, when executed by a processor, implement the blast furnace permeability index prediction method.

[0036] The beneficial effects of this invention are as follows:

[0037] 1. The proposed method explores for the first time from a multi-timescale perspective the reliable multi-scale time delay knowledge between different process variables and permeability index, and realizes effective time delay extraction for the complex noise environment inside the blast furnace;

[0038] 2. The fusion prediction method based on OSA and WNN extracts the spatial characteristics between variables from different perspectives, including time domain and frequency domain, global and local, thus expanding the characterization ability of the air permeability index.

[0039] 3. The proposed prediction framework integrates the spatiotemporal dimensions using the Gauss-Markov estimation method, demonstrating accurate prediction capabilities for the blast furnace permeability index. Attached Figure Description

[0040] Figure 1 This is a diagram illustrating the construction of the blast furnace permeability index prediction model based on multi-scale time delay characteristic mining according to the present invention. Detailed Implementation

[0041] To more clearly describe the technical content of the present invention, the following description, in conjunction with the accompanying drawings, further illustrates the specific implementation methods of the present invention.

[0042] like Figure 1 The diagram shown illustrates the construction of the blast furnace permeability index prediction model based on multi-scale time delay characteristic mining according to the present invention.

[0043] The method includes the following steps:

[0044] 1. First, it is necessary to determine the input variables related to the blast furnace permeability index. Relevant blast furnace operation data is read from the database and denoised. The input variables are those closely related to the blast furnace permeability index, including cold blast flow rate, oxygen enrichment rate, blast kinetic energy, furnace gas index, total differential pressure, blast pressure, hot blast temperature, and top temperature. These variables are used as input variables, and the permeability index is used as the output variable.

[0045] 2. Perform peak separation on the air permeability index data, decomposing it into a stationary sequence and individual peak sequences. The specific steps are as follows:

[0046] (1) By using the Gaussian function The original signal x(t) is convolved with the original signal x(t) to smooth the original signal and obtain the convolved sequence x. conv (t);

[0047] (2) For x conv (t) Perform second-order difference and set an appropriate threshold. The peak extreme point p is selected. Then x is... conv (t) Perform first-order difference and set an appropriate time interval range to obtain the nearest extreme point q after each peak extreme point p;

[0048] (3) with x conv The first difference of (t) is the search sequence, and the corresponding threshold is set. Find the first point less than the threshold before the peak extreme point p as the peak start point, and the first point less than the threshold after the extreme point q as the peak end point;

[0049] (4) Connect the starting and ending points of the peaks with a smooth quadratic spline curve that retains the original peak change trend while significantly reducing its change amplitude. The resulting sequence is the stationary sequence after peak separation, and the remaining part is the peak sequence obtained by separation, thereby realizing peak separation of the original signal.

[0050] 3. Perform multi-scale wavelet decomposition on stationary sequences to obtain detail and approximate components at different time scales. Multi-scale wavelet decomposition involves scaling and shifting a basic wavelet function (such as the Daubechies 4 wavelet) to create a series of wavelet functions, and then performing an inner product with the original signal to obtain the detail and approximate components of the original signal at different decomposition levels at different scaling scales. The decomposition levels can range from three to more than ten, depending on the modeling requirements.

[0051] 4. At different time scales, extract the decomposition sequences of all variables belonging to that decomposition level. Using wavelet coherence analysis, extract the time delay information of each input variable and the air permeability index at the corresponding scale. Based on the time delay information, align the multi-scale sequences to construct training samples. The specific steps are as follows:

[0052] (1) After performing multi-scale wavelet decomposition on different time series x (input variables) and time series y (permeability index), calculate the detail coefficients at the scaling scale j and displacement scale k:

[0053]

[0054] in, and Let ψ be the approximate components of time series x and y at the (j-1)th decomposition level. j,k (t) represents the wavelet function at scaling scale j and displacement scale k:

[0055]

[0056] ψ(t) is the basic wavelet function; subsequently, the cross wavelet transforms of the time series x and y are calculated:

[0057]

[0058] (2) Wavelet coherence is calculated by the following formula:

[0059]

[0060] Where S is the smoothing operator in both time and scale:

[0061] S(w)=S scale (S tume (w)) (9)S scale It is a smoothing operation on the wavelet scale axis, S time A smoothing operation on the timeline:

[0062]

[0063] S scale (w j,k )| k =(w j,k *c2Π(0.6j))| k (11) Where c1 and c2 are normalization constants, and Π is a rectangular function.

[0064] 5. Construct a time-frequency fusion prediction model based on orthogonal subspace analysis and wavelet neural network, and train the model using training samples; use Gauss-Markov estimation to fuse the prediction results of the sub-models to obtain prediction results at different scales; sum the prediction results at different scales to obtain the final predicted value of the air permeability index, specifically:

[0065] (1) The global time-domain information of the data is analyzed using the orthogonal subspace analysis algorithm. By decomposing the process data (time series X) and key performance indicators (air permeability index, time series Y) into three orthogonal subspaces, the key features are extracted:

[0066]

[0067] Where T com T is a common latent variable of X and Y, therefore T com =XΞ X =YΞ Y Ξ X and Ξ Y For the corresponding load matrix, and Let E be a subspace in X and Y that has common components. OSA and F OSA Let X and Y be the residual spaces excluding the common component subspaces, respectively. Therefore, and It can be represented as

[0068]

[0069] in, and From X to and Y to The projection transformation matrix.

[0070] Due to the orthogonality of subspaces, we can obtain

[0071]

[0072] when When there is a high linear correlation, principal component extraction is required:

[0073]

[0074] Among them, V com It is the score matrix, U com It is the load matrix, E res It is the residual matrix. Correspondingly, X to V cim and Y to V com The projection transformation matrix is:

[0075]

[0076] Therefore, the estimated value of Y obtained through orthogonal subspace analysis is

[0077]

[0078] (2) Wavelet neural networks are used to capture local frequency domain features more effectively. Wavelet neural networks are a variant of backpropagation neural networks, replacing the traditional activation functions used by the hidden layer neurons with the Morlet wavelet function as shown in the following equation.

[0079]

[0080] Therefore, the output formula of the hidden layer neurons in a wavelet neural network is as follows:

[0081]

[0082] Where h(j) is the output value of the j-th node in the hidden layer, ω ij x represents the connection weights between the input layer and the hidden layer; i Let a be the output value of the i-th node in the input layer. j and b j These are the scaling factor and shift factor of the j-th node in the hidden layer, respectively. Then, the output of the wavelet neural network is obtained through the connection weights between the hidden and output layers.

[0083] (3) Use Gauss-Markov estimation to fuse the prediction results of the sub-models, and combine the output results of different sub-models. The model prediction results obtained from Gauss-Markov estimation

[0084]

[0085] in, Let V be the output error variance of the i-th sub-model. This represents the output of the i-th sub-model. Gauss-Markov estimation can reduce the prediction error of individual sub-models, thereby improving the prediction accuracy of the overall model.

[0086] This invention proposes a method for predicting blast furnace permeability index based on multi-scale time-delay characteristic mining, aiming to improve the modeling and prediction accuracy of blast furnace permeability index under complex production environments. First, peak separation and multi-scale decomposition are performed on the data, and wavelet coherence analysis is used to extract multi-scale time-delay information between variables. Then, orthogonal subspace analysis and wavelet neural networks are used to model global features in the time domain and local features in the frequency domain, respectively. Gauss-Markov estimation is then used for model fusion to achieve multi-step advance prediction of the permeability index. This invention can effectively capture the dynamic time-delay characteristics of the blast furnace process, improve prediction accuracy and robustness, and provide technical support for stable blast furnace operation.

[0087] The above description is not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the permeability index of a blast furnace based on multi-scale time delay characteristics, characterized by the following steps: include: Determine the operating parameters related to the blast furnace permeability index, read the relevant blast furnace operating data from the database, and perform data noise reduction. The operating parameters and air permeability index data are simultaneously subjected to peak separation, decomposed into stationary sequences and peak components; Multi-scale wavelet decomposition of stationary sequences yields detail components and approximate components at different time scales. At each time scale, wavelet coherence analysis is used to extract the time delay information between each operating parameter and the air permeability index, and the decomposition sequence at the corresponding scale is time-aligned based on the time delay information to construct training samples. A time-frequency fusion prediction model based on orthogonal subspace analysis and wavelet neural network was constructed and trained using the training samples. The outputs of orthogonal subspace analysis and wavelet neural network at the same time scale were fused using Gauss-Markov estimation. The fusion results at different time scales were summed to obtain the final predicted value of the blast furnace permeability index.

2. The method for predicting blast furnace permeability index based on multi-scale time delay characteristic mining according to claim 1, characterized in that, The operating parameters include cold air flow rate, oxygen enrichment rate, blast energy, furnace gas index, total differential pressure, blast pressure, hot air temperature, and top temperature.

3. The method for predicting blast furnace permeability index based on multi-scale time delay characteristic mining according to claim 1, characterized in that, The peak separation includes: Gaussian smoothing of the original signal; detection of peak extrema based on second-order difference and threshold; determination of the peak start and end intervals by combining first-order difference; and fitting of the peak region with a quadratic spline curve to preserve the trend and suppress the amplitude.

4. The method for predicting blast furnace permeability index based on multi-scale time delay characteristic mining according to claim 1, characterized in that, The aforementioned multi-scale wavelet decomposition of stationary sequences refers to forming a series of wavelet functions by scaling and shifting the basic wavelet function in different ways, and then performing an inner product with the original signal to obtain the detail components and approximate components of the original signal at different scaling scales and decomposition levels.

5. The method for predicting blast furnace permeability index based on multi-scale time delay characteristic mining according to claim 1, characterized in that, The wavelet coherence analysis includes: calculating the cross wavelet transform between the input variable and the permeability index; determining the time delay relationship based on the peak value of the wavelet coherence coefficient; and adjusting the time series to achieve alignment.

6. The method for predicting blast furnace permeability index based on multi-scale time delay characteristic mining according to claim 1, characterized in that, The aforementioned construction of a time-frequency fusion prediction model based on orthogonal subspace analysis and wavelet neural networks specifically involves: (1) The orthogonal subspace analysis algorithm is used to analyze the global time domain information of the data. It decomposes the process data time series X and the key performance index permeability index and time series Y into three orthogonal subspaces, and solves the following projection problem to build a prediction model: （1） in, This is an estimate of Y obtained through orthogonal subspace analysis. and V are the score matrices V of X and Y projected onto the common subspace. com The projection transformation matrix; (2) The wavelet neural network is used to capture local frequency domain features more effectively. The neural network is constructed by replacing the traditional activation function used by the hidden layer neurons in the backpropagation neural network with the Morlet wavelet function as shown in the following formula: （2）。 7. The method for predicting blast furnace permeability index based on multi-scale time delay characteristic mining according to claim 1, characterized in that, The Gauss-Markov estimation method described above fuses the outputs of different sub-models by solving the following equation to obtain the overall prediction result of the hybrid model: （3）； in, For the first i The output error variance of each sub-model For the first i The output results of each sub-model.

8. A computer system for implementing the method according to any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to obtain operating parameters and permeability index data from the blast furnace operating system; The preprocessing module is used to perform transient spike separation and wavelet denoising; The wavelet decomposition module is used to perform multi-scale decomposition. The time delay analysis and alignment module is used to calculate wavelet coherence and perform time alignment; The time-frequency fusion module is used to run the orthogonal subspace analysis model and the wavelet neural network model respectively; The fusion decision module is used to perform Gaussian-Markov estimation and scaling summation; The output module is used to output the multi-step predicted value of the blast furnace permeability index.

9. A computer-readable storage medium having stored thereon program instructions that, when executed by a processor, implement the blast furnace permeability index prediction method as described in any one of claims 1 to 8.

Images