A sintering mixing water content on-line soft measurement method and system based on multi-source information fusion

Abstract

The application discloses a kind of based on sintering mixing material moisture on-line soft measurement method and system of multi-source information fusion, by obtaining sintering mixing material image and process variable data, based on the hue feature distribution and texture feature distribution of sintering mixing material image, establish fog concentration estimation linear model, obtain sintering mixing material image with fog concentration less than preset value, and based on the color, texture, morphological characteristics of sintering mixing material image with fog concentration less than preset value, establish sintering mixing material moisture soft measurement submodel based on image feature, according to process variable data, establish sintering mixing material moisture soft measurement submodel based on GRU, to establish adaptive sintering mixing material moisture soft measurement model, sintering mixing material moisture is measured, solve the technical problem that existing sintering mixing material moisture detection precision is low, can realize continuous high-precision sintering mixing material moisture detection, and have strong anti-interference ability, can effectively guide two mixing moisture control operation.

Description

Technical Field

[0001] This invention mainly relates to the field of steel industry technology, specifically to an online soft measurement method and system for moisture content of sintering mixtures based on multi-source information fusion. Background Technology

[0002] With the rapid development of the steel industry, the quantity and quality of natural rich ore are far from meeting the requirements of blast furnace smelting. In my country's steel industry, artificial rich ore is mainly produced by sintering, accounting for more than 80% of the iron-containing furnace charge used in blast furnaces. Sintering is a key link in conveying refined feed to the blast furnace, and accurate online mixing moisture information is of great significance for controlling the permeability of materials, improving the quality of sinter, and reducing the emission of harmful gases. Due to the complex and variable environment around the sintering mixing belt and the frequent fluctuations in sintering raw materials, existing moisture detection methods are difficult to accurately obtain the mixing moisture online.

[0003] Currently, the main method for detecting moisture content in sintered materials is to install a moisture detector between the primary and secondary mixing stages. This real-time moisture level is then used to guide moisture control in the secondary mixing stage. However, due to the frequent fluctuations in raw material composition and the presence of water vapor on the surface of the sintered material, existing moisture detection devices such as near-infrared and microwave moisture detectors have relatively poor accuracy. Currently, on-site moisture assessment still relies on workers to judge the moisture content based on the surface condition and their experience. This manual judgment is subjective, and the reaction of quicklime with water in the sintered material releases a significant amount of heat. When the ambient temperature is low, an uneven layer of water vapor forms on the surface of the material, leading to misjudgments by the workers. Furthermore, the prediction results of intelligent sensing models that use the image features of the material as input are easily affected by water vapor, resulting in significant fluctuations in the accuracy of the detected moisture content.

[0004] Due to the interference of water mist, the method of extracting image features and building an intelligent model by analyzing the water monitoring experience of sintering workers results in significant fluctuations. Considering that the sintering site database stores massive amounts of process variable data, and that process variables have the advantage of being unaffected by water mist, this patent proposes an online soft measurement method for sintering mixture moisture based on multi-source information fusion, combining the water monitoring experience of sintering operators with sintering process variable data. By analyzing the advantages and disadvantages of soft measurement models for mixture moisture constructed from two different modal data from the site, a moisture value fusion model is established. This fully leverages the advantages of different modal data, achieving continuous and accurate measurement of sintering mixture moisture, overcoming the difficulties of non-contact moisture detection methods being greatly affected by the environment and having large time delays in direct measurement.

[0005] The invention patent with publication number CN113447392A proposes a calibration and verification method for a sintering mixture moisture measurement device. The patent uses the drying and weighing method to obtain the relative true value of the material moisture content at the sampling point at the sampling time, providing a reference value for the measurement benchmark for device verification. Then, it combines the expert experience method to express the current mixture moisture percentage with linguistic variables or fuzzy mathematics to verify the consistency of the device's dynamic measurement of large / small or slightly large / slightly small values. Finally, based on the drying and weighing method and the worker / expert experience method, the continuous measurement value of the device is calibrated and verified using the big data analysis relative correction method.

[0006] The big data correction part of this invention has a small data volume and does not take into account various external factors, making it difficult to cope with changes in working conditions.

[0007] Patent publication number CN108133106B proposes a method for predicting the moisture content of sintering machine mixtures based on the amount of water added. This method allows operators to estimate the moisture content of the mixture in the sintering machine based on the amount of water added during the sintering mixing process. This enables operators to predict the moisture content of the mixture in the later stages of sintering based on the water added in the preceding mixing stages, facilitating adjustments to the water addition amount based on the process conditions of the sintering mixture's moisture content, thus optimizing the water addition during the sintering mixing process. While this invention establishes a formula for calculating moisture content, accurate calculation is difficult due to fluctuations in material proportions and numerous external interference conditions at the sintering site. Summary of the Invention

[0008] The present invention provides an online soft measurement method and system for moisture content of sintering mixtures based on multi-source information fusion, which solves the technical problem of low accuracy in existing sintering mixture moisture detection.

[0009] To address the aforementioned technical problems, the present invention proposes an online soft measurement method for the moisture content of sintering mixes based on multi-source information fusion, comprising:

[0010] Acquire images of the sintering mixture and process variable data;

[0011] A linear model for fog concentration estimation is established based on the hue and texture feature distributions of sintered mixture images.

[0012] Based on the linear model for fog concentration estimation, obtain sintering mixture images where the fog concentration is less than the preset value;

[0013] Based on the color, texture, and morphological features of the sintered mixture image with a fog concentration less than a preset value, a soft measurement sub-model of sintered mixture moisture based on image features is established.

[0014] Based on process variable data, a soft measurement sub-model for moisture content in sintering mixes based on GRU is established;

[0015] Based on the image feature-based soft measurement sub-model of sintered mixture moisture and the GRU-based soft measurement sub-model of sintered mixture moisture, an adaptive soft measurement model of sintered mixture moisture is established, and the moisture of sintered mixture is measured based on the adaptive soft measurement model of sintered mixture moisture.

[0016] Furthermore, acquiring sintering mixing images and process variable data includes:

[0017] Collect sintering mixing images and process variable data;

[0018] Preprocess the sintering mixing images and process variable data;

[0019] Time-series matching was performed on the pre-processed sintering mixture image and process variable data.

[0020] Furthermore, based on the hue and texture feature distributions of the sintered mixture image, a linear model for estimating fog concentration is established, including:

[0021] The tonal and texture feature distributions of the sintered mixture image are obtained, and the specific formulas for calculating the tonal and texture feature distributions of the sintered mixture image are as follows:

[0022]

[0023] Among them, Q H and Q T The tonal and texture feature distributions of the sintered mixture image are Ω(p) i ) indicates that the pixel index is p i Let |Ω| be the neighborhood of the center, and |Ω| represent the number of pixels in the neighborhood. H and H represent the gradient map and tone map of the foggy image, respectively;

[0024] Based on the hue and texture feature distributions of the sintered mixture image, a linear model for fog concentration estimation is established, and the specific formula for the linear model for fog concentration estimation is as follows:

[0025] D = Q H -αQ T ,

[0026] Q H and Q T These represent the hue feature distribution and texture feature distribution of the sintered mixture image, respectively. α is the weighting coefficient, and D is the coarse fog concentration quantization map.

[0027] Furthermore, after establishing a linear model for fog concentration estimation, the following steps are also included:

[0028] Bilateral filtering is applied to the coarse fog concentration quantization map.

[0029] Furthermore, based on the linear model for estimating mist concentration, obtaining sintering mixture images where the mist concentration is less than a preset value includes:

[0030] The sintering mixture image is partitioned to obtain the sintering mixture image partitions;

[0031] Based on the linear model for fog concentration estimation, the mean and standard deviation of the two-dimensional fog concentration field corresponding to the sintering mixture image partition are calculated, and the two-dimensional fog concentration field is normalized.

[0032] Calculate the mean fog concentration in the normalized two-dimensional fog concentration field;

[0033] Based on the average fog concentration, obtain sintering mixture images where the fog concentration is less than a preset value.

[0034] Furthermore, based on the color, texture, and morphological features of the sintered mixture image (where the mist concentration is less than a preset value), a soft measurement sub-model for the moisture content of the sintered mixture based on image features is established, including:

[0035] Calculate the gray-level co-occurrence matrix of the sintered mixture image, and calculate the first feature based on the gray-level co-occurrence matrix. The first feature is specifically a texture feature.

[0036] The second feature of the sintered mixture image in the HSV color space is extracted. The second feature includes the mean, standard deviation, and slope features.

[0037] The OTSU algorithm is used to separate the shadows of material blocks from the sintering mixture image, and the maximum shadow area, average shadow area and number of shadows are calculated to obtain the third feature;

[0038] The first, second, and third features are input into the XGBoost model for training to obtain a soft measurement sub-model of sintering mixture moisture based on image features.

[0039] Furthermore, based on the image feature-based soft measurement sub-model of sintered mixture moisture content and the GRU-based soft measurement sub-model of sintered mixture moisture content, an adaptive soft measurement model of sintered mixture moisture content is established, including:

[0040] The average fog concentration of the sintering mixture image is calculated based on the linear model for fog concentration estimation.

[0041] Based on the average mist concentration, the image feature-based soft measurement sub-model of sintered mixture moisture content, and the GRU-based soft measurement sub-model of sintered mixture moisture content, an adaptive soft measurement model of sintered mixture moisture content is established. The calculation formula of the adaptive soft measurement model of sintered mixture moisture content is as follows:

[0042]

[0043] Where y is the output of the adaptive sintering mix moisture soft measurement model, y1 and y2 are the outputs of the image feature-based sintering mix moisture soft measurement sub-model and the GRU-based sintering mix moisture soft measurement sub-model, respectively, and μ average The average fog concentration is given by 0 ≤ μ. average ≤1.

[0044] The online soft measurement system for moisture content of sintering mixes based on multi-source information fusion provided by this invention includes:

[0045] The present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the online soft measurement method for moisture content of sintering mixtures based on multi-source information fusion provided by the present invention.

[0046] Key aspects of this invention include:

[0047] (1) To address the shortcomings of existing fog concentration estimation methods when applied to sintered images with fog, an improved fog concentration estimation method is proposed. The hue feature distribution is used as the color term in the linear fog concentration estimation model, which improves the fog concentration estimation accuracy under low saturation and low brightness conditions.

[0048] (2) In view of the uneven distribution of fog on the surface of sintered mixture, a method is proposed to divide the two-dimensional fog concentration field and visible light image into regions. Based on the fog concentration of the fog concentration region, the visible light image is divided into a dense fog region and an effective information region. The dense fog region is discarded, which improves the effectiveness of the mixture image features.

[0049] (3) Based on the prior knowledge and experience accumulated by the water monitoring workers, color, texture and morphological features were extracted and a soft measurement model of sintering mixture moisture based on image features was established; considering that the process variables in the sintering process have nonlinearity, strong coupling and time continuity, a soft measurement model of sintering mixture moisture was constructed using a GRU network.

[0050] (4) Considering that image and process variable data have their own advantages and disadvantages, in order to give full play to the advantages of the two types of data, an adaptive weighted mixing moisture value fusion method is proposed. The fusion weight of the moisture detection values ​​of the two models is calculated based on the fog concentration in the effective information area of ​​the visible light image of the mixture, and the weighted fusion is performed to improve the detection accuracy and anti-interference ability of the proposed mixing moisture soft measurement method.

[0051] (5) The soft measurement method for moisture content of sintering mixtures of the present invention enables continuous and accurate measurement of moisture content of sintering mixtures, and still has good detection accuracy when the water vapor concentration is large and the material type fluctuates.

[0052] This invention focuses on the mixing of materials on the conveyor belt between the first and second mixing sections of a sintering plant, proposing an online soft measurement method for the moisture content of sintering materials based on multi-source information fusion. Firstly, addressing the time-series mismatch between images and process variables, the images and process variable data are matched according to certain rules. Since uneven water mist reduces the effective information content in the sintering material images, making it difficult to extract effective features, dense fog areas need to be removed. To address the shortcomings of existing fog concentration estimation methods for sintering material images, this invention proposes an improved fog concentration method. This method incorporates hue feature distribution into the linear fog concentration estimation model, effectively calculating the two-dimensional fog concentration field of the visible light image. The two-dimensional fog concentration field and the visible light image are then divided into three regions, and different visible light image regions are defined as dense fog areas and effective information areas based on the different fog concentrations in each region of the fog concentration field. Considering the different advantages and disadvantages of different modal data, this invention proposes a fog concentration estimation method with adaptive weights, adaptively adjusting the fusion weights of the detection values ​​from the two moisture soft measurement models according to changes in water mist concentration. The final moisture content of the sintering mixture is calculated using a fusion model, which then guides the moisture control operation of the secondary mixing process.

[0053] Furthermore, the embodiments of the present invention solve the problem of poor accuracy in detecting moisture content in sintering mixes, and realize real-time online high-precision detection of moisture content in sintering mixes, which has the advantages of real-time, continuous, high precision, and minimal impact from external interference. Attached Figure Description

[0054] Figure 1 This is a diagram of the sintering mixture moisture detection system according to Embodiment 2 of the present invention;

[0055] Figure 2 This is a flowchart of the online soft measurement method for moisture content of sintering mixtures based on multi-source information fusion according to Embodiment 2 of the present invention;

[0056] Figure 3 This is a structural block diagram of the online soft measurement system for moisture content of sintering mixtures based on multi-source information fusion, according to an embodiment of the present invention.

[0057] Figure label:

[0058] 1. Belt support; 2. Feeding belt; 3. Near-infrared moisture detector support; 4. Near-infrared moisture detector; 5. Sintering mixing; 6. Visible light camera; 7. Camera support; 8. Fiber optic cable; 9. Sintering database; 10. LCD display; 11. Main unit; 20. Memory; 30. Processor. Detailed Implementation

[0059] To facilitate understanding of the present invention, the present invention will be described more fully and in detail below with reference to the accompanying drawings and preferred embodiments, but the scope of protection of the present invention is not limited to the following specific embodiments.

[0060] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0061] Example 1

[0062] The online soft measurement method for moisture content of sintering mixes based on multi-source information fusion provided in Embodiment 1 of this invention includes:

[0063] Step S101: Obtain sintering mixing images and process variable data;

[0064] Step S102: Based on the hue feature distribution and texture feature distribution of the sintering mixture image, establish a linear model for fog concentration estimation;

[0065] Step S103: Based on the linear model for estimating fog concentration, obtain sintering mixture images where the fog concentration is less than a preset value;

[0066] Step S104: Based on the color, texture, and morphological features of the sintered mixture image with a mist concentration less than a preset value, establish a soft measurement sub-model of sintered mixture moisture based on image features.

[0067] Step S105: Based on the process variable data, establish a soft measurement sub-model for moisture content of sintering mixtures based on GRU.

[0068] Step S106: Based on the soft measurement sub-model of sintered mixture moisture based on image features and the soft measurement sub-model of sintered mixture moisture based on GRU, an adaptive soft measurement model of sintered mixture moisture is established, and the moisture of the sintered mixture is measured based on the adaptive soft measurement model of sintered mixture moisture.

[0069] The online soft measurement method for sintered material moisture based on multi-source information fusion provided in this invention acquires sintered material images and process variable data. Based on the hue and texture feature distributions of the sintered material images, a linear model for fog concentration estimation is established. According to the fog concentration estimation linear model, sintered material images with fog concentrations below a preset value are acquired. Based on the color, texture, and morphological features of the sintered material images with fog concentrations below the preset value, a soft measurement sub-model for sintered material moisture based on image features is established. According to the process variable data, a soft measurement sub-model for sintered material moisture based on GRU is established. Based on the soft measurement sub-model for sintered material moisture based on image features and the soft measurement sub-model for sintered material moisture based on GRU, an adaptive soft measurement model for sintered material moisture is established. The moisture content of the sintered material is measured based on the adaptive soft measurement model for sintered material moisture. This method solves the technical problem of low detection accuracy of existing sintered material moisture, enables continuous and high-precision detection of sintered material moisture, has strong anti-interference ability, and can effectively guide the moisture control operation of the secondary mixing process. Furthermore, the moisture soft measurement method proposed in this embodiment of the invention can detect the moisture content of the mixture after sintering and mixing in real time, and has the advantages of continuity, high precision, and strong robustness.

[0070] Specifically, the online soft measurement method for moisture content of sintered mixtures based on multi-source information fusion provided in this invention involves acquiring visible light image data of the sintered mixture surface using a visible light camera installed after sintering and mixing, and preprocessing the images acquired on-site to obtain the region of interest. Since uneven water mist reduces the effective information content in the sintered mixture image, making it difficult to extract effective features, dense fog areas need to be removed. First, the fog on the surface of the mixture image is estimated. Addressing the shortcomings of existing fog concentration estimation methods applied to sintered mixture images, this patent introduces hue feature distribution into a linear fog concentration estimation model, thereby obtaining a two-dimensional fog concentration field of the sintered mixture. The fog concentration field and the visible light image are divided into three equal partitions, and the average fog concentration of the fog field in each partition is calculated. Based on the fog concentration, the visible light image is divided into dense fog areas and effective information areas. The dense fog areas are discarded, and the effective information areas are reserved for feature extraction in the next step. This patent analyzes the surface agglomeration, color, and texture characteristics of the mixed material, extracting features from the effective information area and establishing a moisture soft measurement model using XGBoost. However, the effective information area obtained through partitioning still contains some water mist. Therefore, the moisture soft measurement model established using only a single image modality has limited anti-interference capability. Since the sintering database contains massive amounts of process variable data, and there are close relationships between various parameters in the sintering process, this patent considers mining deeper moisture information from the sintering process variables to establish a moisture soft measurement model. Because the process variables in the sintering process are generally continuous in time, this patent uses a GRU network, which can mine the temporal relationships between variables, to model the sintering process data and predict the moisture content of the mixed material. Image-based moisture soft measurement models have high accuracy and conform to workers' prior knowledge but are greatly affected by mist, while GRU network-based moisture soft measurement models generally have lower accuracy but better stability. To fully utilize the advantages of two different modal data, this patent proposes a moisture value fusion method with adaptive weights. By detecting the water mist concentration on the surface of the sintering mix in real time, the fusion weights of the moisture detection values ​​of the two models are adaptively adjusted. The fusion model constructed by this method can achieve continuous and high-precision detection of moisture in the sintering mix and has strong anti-interference ability, which can effectively guide the moisture control operation of the two-mixing process.

[0071] Example 2

[0072] This invention proposes an online soft measurement method for the moisture content of sintering mixes based on multi-source information fusion. Figure 1 This is a schematic diagram of an online moisture detection system for sintering materials installed at the mixing belt after sintering. It includes a belt support 1, a feeding belt 2, a near-infrared moisture detector support 3, a near-infrared moisture detector 4, a sintering material mixer 5, a visible light camera 6, a camera support 7, an optical fiber 8, a sintering database 9, an LCD display 10, and a main unit 11. Figure 2This is a flowchart illustrating the implementation steps of the online soft measurement method for moisture content in mixed materials proposed in this invention, including the following steps:

[0073] (1) After the first mixing of sintering, a visible light camera is installed to acquire visible light image data of the sintering mixture. At the same time, the near-infrared moisture detection value, mixing temperature, belt speed, water addition amount of the first mixing and other ten process variable data are acquired from the sintering database. The acquired sintering mixture image and the process variables of the sintering site are preprocessed. The preprocessed data is time-matched, and the sintering mixture at the corresponding time is sampled and dried as the true value.

[0074] (2) Considering that the existing fog concentration estimation methods are difficult to apply well to the water fog concentration estimation of the sintered mixture surface, this embodiment of the invention proposes an improved fog concentration estimation method for the sintered mixture surface. A linear model for fog concentration estimation is defined by the hue feature distribution and texture feature distribution of the fogged image. First, the fog concentration is coarsely estimated, and then the two-dimensional fog concentration field is calculated using edge-preserving filtering.

[0075] (3) Calculate the mean and standard deviation of the two-dimensional fog concentration field and normalize the image using Z-score. Divide the image into three regions evenly, calculate the mean fog concentration in each region, discard the visible light image region corresponding to the region with higher fog concentration, and wait for the next step of feature extraction for the visible light image region corresponding to the region with lower fog concentration.

[0076] (4) Extract the color, texture, and morphological features of the image respectively, and calculate the maximum information coefficient of the extracted image features. Select the features with strong correlation and input them into XGBoost for training to obtain a soft measurement sub-model of sintering mixture moisture based on image features. Input the preprocessed process variables in (1) into the GRU network to obtain a soft measurement sub-model of sintering mixture moisture based on GRU.

[0077] (5) Areas with low fog concentration are often covered by a thin layer of water mist. As the water mist concentration increases, the effective information content in the visible light image gradually decreases, and the effectiveness of the extracted image features decreases. Therefore, this embodiment of the invention proposes a water value fusion method with adaptive weights to calculate the fog concentration in the effective information area, and then adjust the weights of the water content predicted by different models in real time according to the water mist concentration. The data fusion at the model level is performed by the formula proposed in this embodiment of the invention.

[0078] The specific operational plan is as follows:

[0079] (1) Acquire visible light images and process variable data and perform data preprocessing.

[0080] a) Acquire visible light images and process variable data

[0081] Visible light cameras were used to capture images of the material between the first and second mixing stages. The video stream was then transmitted via a dedicated fiber optic network to a computer in the monitoring room for storage. Image data was acquired at a sampling frequency of half an hour. The required images were also obtained via a remote connection to the monitoring computer, and ten process variables from the sintering data information system were extracted at the same frequency. Samples of the sintering mixture on the sintering conveyor belt after the first mixing stage were taken at a sampling frequency of half an hour. The mixture was sealed, and ten samples were taken as a batch and sent to a hot drying oven for drying to obtain the actual moisture content of the sintering mixture at the corresponding time points.

[0082] b) Preprocessing of image and process variable data

[0083] First, rotate the image counterclockwise by α degrees. Then, select a rectangle formed by (x1, y1) and (x2, y2) to crop the image, removing as many irrelevant areas as possible outside the mixture to obtain the region of interest. Select one month's process variables to handle missing and outlier values. Normalize the processed data, calculate the maximum information coefficient of the normalized process variables, and filter variables with high correlation.

[0084] Step 1: The size of the acquired visible light image is 1920×1080. Since the shooting angle is from the side towards the mixed material, the image needs to be rotated 28 degrees counterclockwise first. Due to the interference of debris at the edge of the image, two diagonal points (40,350) and (1600,650) are selected in sequence to form a rectangular area to crop the image. The obtained rectangular area is taken as the region of interest of the image.

[0085] Step 2: Obtain process variable data for half a month from the database at the same frequency as image sampling. Due to data anomalies and missing data during storage and transmission, these anomalies can affect the model's prediction accuracy; therefore, anomaly data processing is necessary. This embodiment of the invention uses the quartile method to clean the anomaly data, arranging all data in ascending order and calculating the upper quartile Q3 and lower quartile Q1 as the upper and lower bounds of the normal data. The calculation formula is as follows:

[0086] [da l ,da h ]=[Q1-1.5I QR Q3+1.5I QR (1)

[0087] In the formula:da l da h I represents the upper and lower bounds of normal data. QR This represents the interquartile range. When the process variable data is less than da at a certain moment... l Or greater than da hIf a data point is found to be missing from the list of missing and cleaned-up data, that data is considered outlier and needs to be removed. After cleaning the outlier data, interpolation is used to fill in the missing data and the cleaned-up outlier data. The calculation formula is as follows:

[0088]

[0089] In the formula: da t For the process variable data at time t, da t-1 da t+1 The data are process variable data at time t-1 and t+1, respectively.

[0090] Step 3: Normalization: Since the process variable data have different dimensions, to prevent these different dimensions from affecting model learning and to improve the training speed and convergence of the model, the input needs to be normalized. In this embodiment of the invention, the Z-score normalization method is used to normalize all process variables within half a month to a distribution with a mean of 0 and a variance of 1. The calculation formula is as follows:

[0091]

[0092] x is the value of the process variable at a certain moment; x mean σ is the mean of the process variable over one month; σ is the standard deviation of the process variable over one month; x scale These are the normalized process variables.

[0093] Step 4: Correlation Analysis: The selected process variables are based solely on experience, suggesting a strong correlation with the moisture content of the sintering mix, which is highly subjective. Considering the difficulty in obtaining labels and the limited data volume, it is necessary to quantitatively analyze the correlation coefficients between process variables and the moisture content of the sintering mix, thereby selecting process variables with strong correlations. Since the relationship between process variables and the moisture content of the sintering mix is ​​highly nonlinear, conventional correlation coefficient calculation methods are insufficient to characterize this relationship. Therefore, this embodiment of the invention uses the maximum information coefficient to calculate the correlation between variables. The maximum information coefficient can capture the complex relationships between variables and has the advantages of universality, fairness, and symmetry. Its calculation formula is as follows:

[0094]

[0095]

[0096] In the formula, I(x,y) represents the mutual information between the process variable x and the moisture content y of the sintered mixture; a and b are the number of grids in the x and y directions, respectively; and B is a fixed value, typically set to approximately 0.6 times the data volume. p(x,y) is the joint probability distribution of x and y, p(x) is the marginal probability distribution of x, and p(y) is the marginal probability distribution of y.

[0097] c) Timing matching of images and process variables

[0098] Considering that sintering and water addition is a continuous process with relatively small fluctuations over a short period, the image data sampling frequency is one image per hour. Furthermore, due to the inherent latency in remote image data transmission, the acquired image data sampling time is often non-integer minutes. Since the database records a stable set of data every minute, it is often difficult to synchronize the image data with the process variable time. Therefore, time-series matching of the data is required, with the matching rule being |t... p -t d |<30s.

[0099] Where t p t represents the image acquisition time. d For process variables obtained from the database, the time corresponding to each variable is considered as a pair when the absolute value of the difference between the two is less than 30 seconds.

[0100] (2) Image-based two-dimensional fog concentration field estimation

[0101] In most foggy scenes, suspended particles in the fog scatter atmospheric light, causing images acquired by optical imaging devices to exhibit low saturation, high local brightness, and blurred details. For any given scene, the denser the fog, the greater the scattering effect of suspended particles on light reflected from objects, resulting in more blurred details, higher brightness, and lower saturation in that area.

[0102] At low temperatures, the surface of sintered mixtures is often covered with a layer of uneven water mist. Due to the obstruction caused by the water mist, it is difficult to extract effective information from the water mist portion of the acquired visible light images. In order to partition the image and establish an adaptive weighted soft measurement model for the moisture content of the sintered mixture, it is necessary to calculate the two-dimensional mist concentration field on the surface of the sintered mixture. Existing mist concentration estimation methods generally select brightness feature distribution or saturation feature distribution as color distribution features to make a coarse estimate of the mist concentration field on the surface of the sintered mixture. However, since the sintered mixture images received by the visible light imaging equipment at the sintering site are indirect light that has been bounced multiple times, there is a large energy loss. Therefore, the overall brightness of the acquired images is low. Using brightness feature distribution as a mist concentration feature is difficult to effectively distinguish between the mixture and the misty area. At the same time, the sintered mixture portion of the sintered mixture image itself has low saturation, and the saturation decrease caused by the presence of water mist on the surface is small. Therefore, the saturation feature distribution also has significant limitations. Considering the shortcomings of existing fog concentration estimation methods in estimating fog concentration in mixed materials, this invention proposes an improved fog concentration estimation method for the surface of sintered mixed materials based on the imaging characteristics of sintered mixed materials. By introducing the hue feature distribution into the linear model for fog concentration estimation, a coarse estimate of fog concentration is made, and edge-preserving filtering is used to calculate the two-dimensional fog concentration field.

[0103] This invention uses the tonal feature distribution Q of an image. H and texture feature distribution Q T Define a linear model based on fog concentration estimation:

[0104] D = Q H -αQ T (6)

[0105] Where α is the weighting coefficient, a larger α indicates a greater influence of texture feature distribution on fog concentration, and vice versa. D is a coarse fog concentration quantization map. In the formula, the hue feature distribution Q... H and texture feature distribution Q T It can be characterized by the local brightness mean and the "adjusted" local gradient mean, and its calculation formula is as follows:

[0106]

[0107] In the formula, Ω(p) i ) indicates that the pixel index is p i Let |Ω| be the neighborhood of the center, and |Ω| represent the number of pixels in the neighborhood. H and H represent the gradient map and tone map of the foggy image, respectively.

[0108] Because the quantized values ​​of the coarse quantization map D are discontinuous in the depth approximation region, containing significant texture noise, while fog concentration remains essentially constant at the same scene depth, quantization map D cannot accurately reflect the true fog concentration distribution. Therefore, it is necessary to perform edge-preserving smoothing on quantization map D, aiming to eliminate texture details as much as possible while retaining its original depth information. Bilateral filtering has good noise reduction and edge-preserving capabilities; the calculation formula is as follows:

[0109]

[0110] Where ω is a scalar, a normalization function, expressed as:

[0111]

[0112] In the formula, i represents the pixel to be determined, N represents the square region centered on pixel i in image I, and j represents any pixel in the neighborhood. The parameter σ... d and σ r Based on the standard deviation of the Gaussian function, they measure the amount of filtering out in image I, where It is a spatial proximity function used to reduce the influence of distant pixels. This is a gray-level similarity function used to reduce the influence of pixel j, which has a large gray-level difference for pixel i. The image is then filtered to obtain a two-dimensional fog concentration field for further processing.

[0113] (3) Divide the visible light image into zones according to fog concentration.

[0114] The surface mist distribution of sintered mixtures is uneven. In areas with excessively high water mist concentration, the effective information content in visible light images is extremely low, making it difficult for subsequently extracted image features to reflect moisture change trends. Therefore, it is necessary to partition the misty images and remove densely misted areas. The specific partitioning method is as follows:

[0115] The mean μ and standard deviation σ of each two-dimensional fog concentration field are calculated, and the image is normalized using Z-score. The calculation formula is as follows:

[0116]

[0117] Where src(i,j) is any element of the two-dimensional fog concentration matrix before normalization, and dst(i,j) is the corresponding element of the two-dimensional fog concentration matrix after normalization.

[0118] Let the two-dimensional fog concentration field be a matrix of size M×N. Divide the matrix into three equal columns to obtain three... The matrix is ​​divided into three equal parts using the same method for the visible light image. The matrix is ​​used to calculate the mean fog concentrations μ1, μ2, and μ3 of the three fog concentration matrices. When the mean fog concentration of any matrix is ​​greater than β, it is denoted as a dense fog region; when the mean fog concentration is less than β, it is denoted as a region with effective information. Since the effective information content in the visible light image is extremely low when the water mist on the surface of the mixture is too large, and the extracted image features are difficult to reflect the trend of moisture changes, the visible light image corresponding to the dense fog region is discarded and not included in subsequent feature extraction. The visible light image of the region with effective information is retained for the next step of feature extraction.

[0119] (4) Establish a soft measurement sub-model for moisture content in sintering mixtures

[0120] The moisture content of the sintering mix varies, resulting in different textures, shapes, and colors on its surface. When the moisture content is high, the top of the mix pile forms sharp angles, indicating high agglomeration, large agglomerates, and a generally messy and rough surface texture. Conversely, when the moisture content is low, the top of the mix pile forms rounded arcs, indicating low agglomeration, small agglomerates, and a generally regular and smooth surface texture, indicating lower moisture content. On-site moisture monitors determine the moisture content of the sintering mix by observing its color, texture, and shape. This invention proposes a soft measurement method for sintering mix moisture content based on color, texture, and shape characteristics, based on the experience of on-site moisture monitors. The specific steps are as follows:

[0121] Step 1: To address the different roughness and irregularity exhibited by the mixture under different moisture contents, the texture features were calculated using the gray-level co-occurrence matrix. First, the image was converted into a grayscale image, and its gray-level co-occurrence matrix was calculated. Based on the gray-level co-occurrence matrix, five statistical quantities—inverse difference, energy, entropy, moment of inertia, and correlation—were calculated as subsequent modeling variables.

[0122] Step 2: Since sintering mixtures with different moisture contents have significant color differences, the HSV color model, which is more in line with the visual characteristics of the human eye, is used to extract the mean, standard deviation, and slope of the mixture image in the HSV color space as features.

[0123] Step 3: Because the surface of the mixture with higher moisture content produces more and larger clumps, while the surface with lower moisture content produces smaller and smoother clumps, and wrinkles form around the clumps, with coarser shadows in the wrinkled areas, the OTSU algorithm is used to separate the shadows of the material clumps from the mixture image. The maximum shadow area, average shadow area, and number of shadows are then calculated to indirectly characterize the agglomeration rate and agglomeration size of the sintered mixture surface.

[0124] Step 4: Calculate the maximum information coefficient for the selected features, filter out the features with high correlation, and input them into XGBoost for training to obtain the required model.

[0125] Because quicklime releases a lot of heat when it reacts with water, when the outside temperature is low, the surface of the sintering mixture will be covered with a lot of water mist. When the water mist concentration on the surface of the sintering mixture is too high, it is difficult to extract moisture-related image features from visible light images. Therefore, it is difficult to achieve the desired effect by predicting the moisture content from the image perspective alone. It is necessary to establish a time-series prediction network from the data perspective. Since different networks have their own characteristics under different conditions, the models are finally fused according to the network characteristics to obtain the final moisture value.

[0126] In the sintering process, process variables exhibit characteristics such as nonlinearity, strong coupling, and temporal continuity. This invention employs a GRU-based temporal prediction network to identify the temporal relationships between process variables, and then performs regression to obtain the moisture content of the sintered mixture. The GRU network is an improvement upon the LSTM network, replacing the forget gate and input gate with update gates. Structurally, it only has two gates: an update gate and a reset gate. Compared to the LSTM network, the simplified GRU network reduces training parameters, lowers learning time requirements, and outperforms the LSTM network in most predictions.

[0127] The preprocessed process variables obtained in (1) are input into the GRU network for training to obtain the soft measurement model of moisture content in sintering mixture.

[0128] (5) Establish a soft measurement model for moisture content of sintering mixture based on multi-source information fusion.

[0129] The soft measurement model of sintering mixture moisture based on color, texture, and morphology obtained in step (4) and the soft measurement model of sintering mixture moisture based on GRU network each have their own advantages and disadvantages. Therefore, it is necessary to fuse the two soft measurement models of sintering mixture moisture to make up for their respective problems and achieve accurate and stable detection of sintering mixture moisture. At present, the most effective and stable method for detecting the moisture of the mixture at the sintering site is for the water inspector to make a comprehensive judgment on the moisture of the mixture based on the prior knowledge and experience accumulated over a long period of time. The soft measurement model of sintering mixture moisture based on color, texture, and morphology can effectively utilize the water inspector's experience, but when there is water mist on the surface of the mixture, the effective information content of the image decreases. The soft measurement model of sintering mixture moisture based on GRU network predicts the moisture content based on the detection value of the near-infrared moisture detector at the sintering site, the temperature of the mixture, the belt speed, and other process variables. It has the advantage of not being affected by water mist. However, due to the frequent changes in the imported raw materials of the sintering plant, its accuracy is not enough. Therefore, in order to improve the detection accuracy and anti-interference ability of the model, it is necessary to fuse the detection values ​​of the two models.

[0130] This invention proposes a model detection value fusion method with adaptive weights to address the respective advantages and disadvantages of two models. It establishes a water mist concentration classification model to determine the water mist concentration in real time and adjusts the weights of different model detection values ​​according to the water mist concentration, thus performing model-level data fusion. The fusion method steps are as follows:

[0131] Let y1 be the output value of the sintering mixture moisture prediction model obtained from the soft measurement model based on color, texture, and morphology of sintering mixture moisture, and y2 be the output value of the soft measurement model based on GRU network of sintering mixture moisture. The average fog concentration μ of the entire image is obtained by averaging the average fog concentration of the effective information area of ​​the image. average The final formula for calculating the moisture content y of the sintered mixture in the fusion model is as follows:

[0132]

[0133] When the water mist concentration is zero, i.e., μ average When the result is 0, the fusion method is to add the detection values ​​obtained from the two models and take the average, as shown in the following formula:

[0134]

[0135] When the water mist concentration is dense fog, i.e. μ average When the value is 1, it is difficult to extract effective information from the sintering mixture image. Therefore, the detection value of the soft measurement model of sintering mixture moisture based on image features has no reference value at this time, and the fusion method degenerates into:

[0136] y = y² (13)

[0137] At this moment, the final moisture content of the sintering mixture is the moisture content y2 based on the GRU network.

[0138] The test set was input into the trained multi-source information fusion model to obtain the moisture content of the sintered mixture. The RMSE, MSE, MAE, MAPE, and hit rate J of the fusion model and the near-infrared moisture detector were calculated respectively. The four evaluation indicators of the fusion model and the near-infrared moisture detector were compared and analyzed. The calculation formulas of RMSE, MSE, MAE, MAPE, and hit rate J are as follows:

[0139] Mean square error:

[0140]

[0141] Root mean square error:

[0142]

[0143] Mean absolute error:

[0144]

[0145] Where y (i) For the true value of the sample, These are predicted values.

[0146] Consider the model's hit rate J, which is defined as:

[0147]

[0148] Where l represents the length of the predicted sample. H(·) represents the predicted value of the sample, and H(·) is the Heaviside function, defined as:

[0149]

[0150] This invention, in its embodiments, acquires visible light image data of the sintered mixture surface by installing a visible light camera after sintering and mixing, and preprocesses the images acquired on-site to obtain the region of interest. Due to uneven water mist reducing the effective information content in the sintered mixture image, it is difficult to extract effective features; therefore, dense fog areas need to be discarded. First, the fog on the surface of the mixture image is estimated. Addressing the shortcomings of existing fog concentration estimation methods applied to sintered mixture images, this invention introduces hue feature distribution into a linear fog concentration estimation model, thereby obtaining a two-dimensional fog concentration field of the sintered mixture. The fog concentration field and the visible light image are divided into three equal partitions, and the average fog concentration of the fog field in each partition is calculated. Based on the fog concentration, the visible light image is divided into dense fog areas and effective information areas. The dense fog areas are discarded, and the effective information areas are reserved for feature extraction in the next step. This patent analyzes the surface agglomeration, color, and texture characteristics of the mixed material, extracting features from the effective information area and establishing a moisture soft measurement model using XGBoost. However, the effective information area obtained through partitioning still contains some water mist. Therefore, the moisture soft measurement model established using only a single image modality has limited anti-interference capability. Since the sintering database contains massive amounts of process variable data, and there are close relationships between various parameters in the sintering process, this patent considers mining deeper moisture information from the sintering process variables to establish a moisture soft measurement model. Because the process variables in the sintering process are generally continuous in time, this patent uses a GRU network, which can mine the temporal relationships between variables, to model the sintering process data and predict the moisture content of the mixed material. Image-based moisture soft measurement models have high accuracy and conform to workers' prior knowledge but are greatly affected by mist, while GRU network-based moisture soft measurement models generally have lower accuracy but better stability. To fully utilize the advantages of two different modal data, this invention proposes a moisture value fusion method with adaptive weights. By real-time detection of water mist concentration on the surface of the sintering mix, the fusion weights of the moisture detection values ​​from the two models are adaptively adjusted. The fusion model constructed by this method can achieve continuous and high-precision detection of moisture in the sintering mix and has strong anti-interference ability, effectively guiding the moisture control operation of the second mixing process.

[0151] Furthermore, the embodiments of the present invention solve the problem of poor accuracy in detecting moisture content in sintering mixes, and realize real-time online high-precision detection of moisture content in sintering mixes, which has the advantages of real-time, continuous, high precision, and minimal impact from external interference.

[0152] Example 3

[0153] To verify the accuracy of the detection values ​​of the algorithm of this invention and the stability of the detection under water mist interference, this patent uses a sintering plant as the implementation platform and builds the system as shown in the attached figure. Figure 1The sintering mixture moisture detection system shown deploys the algorithm of this invention on the sintering expert system in the sintering control room. It obtains the required data from the detection equipment in real time and transmits it to the computer in the sintering main control room. The moisture content of the sintering mixture detected by this algorithm is calculated, and the accuracy and stability of the moisture detection results are verified as follows.

[0154] Accuracy verification:

[0155] Obtaining the true moisture content of the mixture using the hot drying method:

[0156] Samples were taken at the sintering site and sealed. The sealed material was then placed in a drying oven and heated at a certain temperature to remove excess moisture. The change in weight of the mixture before and after drying was calculated to determine the true moisture content. The accuracy of the algorithm of this invention was verified by comparing the mean square error, root mean square error, mean absolute error, and hit rate of the detection results with those of near-infrared moisture detection.

[0157] The effect of moisture on sintering ore yield and ore quality:

[0158] During sintering, moisture has the functions of heat absorption and heat transfer, effectively improving or increasing the heat exchange conditions of the material layer. Secondly, it reduces the surface roughness of the material, supplies trace amounts of oxygen during the sintering combustion process, reduces airflow resistance, and strengthens the sintering process. When the moisture content is too low, flames will spray outwards during ignition, the mixed material will not form lumps, and undercooked material will appear at the tail end of the machine, resulting in a decreased sinter yield. Excessive moisture will increase the thickness of the over-wet layer, increase airflow resistance, worsen sinter permeability, and reduce sinter production. Therefore, the accuracy of moisture detection is closely related to the yield and quality of sinter. To verify the detection accuracy of the method proposed in this invention, the ore formation rate of sinter at the sintering site was compared and verified statistically before and after the application of this algorithm over a period of time.

[0159] Stability verification:

[0160] When water mist is present on the surface of the mixture, the detection accuracy of the method proposed in this embodiment of the invention is compared with other methods:

[0161] The image-feature-based soft measurement model for moisture content in sintered mixtures struggles to extract effective information from images when water mist interference occurs on the mixture surface, resulting in significant fluctuations in detected values. To verify whether the moisture detection algorithm proposed in this invention can maintain good detection accuracy under water mist interference, its stability is verified by comparing the detection accuracy of the proposed method with that of the image-feature-based moisture detection method under different water mist concentrations.

[0162] Reference Figure 3 The online soft measurement system for moisture content of sintering mixes based on multi-source information fusion proposed in this embodiment of the invention includes:

[0163] The system includes a memory 20, a processor 30, and a computer program stored in the memory 20 and executable on the processor 30. When the processor 30 executes the computer program, it implements the steps of the online soft measurement method for moisture content of sintering mixtures based on multi-source information fusion proposed in this embodiment.

[0164] The specific working process and working principle of the online soft measurement system for sintering mixture moisture based on multi-source information fusion in this embodiment can be referred to the working process and working principle of the online soft measurement method for sintering mixture moisture based on multi-source information fusion in this embodiment.

[0165] The above are merely preferred embodiments of the present invention and are 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 online soft measurement of moisture content in sintering mixes based on multi-source information fusion, characterized in that, The method includes: Acquire images of the sintering mixture and process variable data; A linear model for estimating fog concentration is established based on the hue and texture feature distributions of the sintering mixture image. This model includes: The tonal and texture feature distributions of the sintered mixture image are obtained, and the specific formulas for calculating the tonal and texture feature distributions of the sintered mixture image are as follows: Among them, Q H and Q T The tonal and texture feature distributions of the sintered mixture image are Ω(p) i ) indicates that the pixel index is p i Ω represents the neighborhood centered at the center, |Ω| represents the number of pixels in the neighborhood, and ▽I and H are the gradient map and tone map of the foggy image, respectively; Based on the hue and texture feature distributions of the sintered mixture image, a linear model for fog concentration estimation is established, and the specific formula for the linear model for fog concentration estimation is as follows: D=Q H -αQ T , Q H and Q T These are the tonal and texture feature distributions of the sintered mixture image, respectively, where α is the weighting coefficient and D is the coarse fog concentration quantization map. Based on the linear model for fog concentration estimation, obtain sintering mixture images where the fog concentration is less than the preset value; Based on the color, texture, and morphological features of the sintered mixture image with a fog concentration less than a preset value, a soft measurement sub-model of sintered mixture moisture based on image features is established. Based on process variable data, a soft measurement sub-model for moisture content in sintering mixes based on GRU is established; An adaptive soft measurement model for sintered material moisture content is established based on an image feature-based soft measurement sub-model for sintered material moisture content and a GRU-based soft measurement sub-model for sintered material moisture content. The moisture content of the sintered material is then measured based on the adaptive soft measurement model for sintered material moisture content.

2. The online soft measurement method for moisture content of sintering mixtures based on multi-source information fusion according to claim 1, characterized in that, Acquiring sintering mixing images and process variable data includes: Collect sintering mixing images and process variable data; Preprocess the sintering mixing images and process variable data; Time-series matching was performed on the pre-processed sintering mixture image and process variable data.

3. The online soft measurement method for moisture content of sintering mixtures based on multi-source information fusion according to claim 2, characterized in that, After establishing the linear model for fog concentration estimation, the following is also included: Bilateral filtering is applied to the coarse fog concentration quantization map.

4. The online soft measurement method for moisture content of sintering mixtures based on multi-source information fusion according to claim 1 or 3, characterized in that, Based on the linear model for estimating fog concentration, images of sintered mixtures with fog concentrations lower than preset values ​​are obtained, including: The sintering mixture image is partitioned to obtain the sintering mixture image partitions; Based on the linear model for fog concentration estimation, the mean and standard deviation of the two-dimensional fog concentration field corresponding to the sintering mixture image partition are calculated, and the two-dimensional fog concentration field is normalized. Calculate the mean fog concentration in the normalized two-dimensional fog concentration field; Based on the average fog concentration, obtain sintering mixture images where the fog concentration is less than a preset value.

5. The online soft measurement method for moisture content of sintering mixtures based on multi-source information fusion according to claim 4, characterized in that, Based on the color, texture, and morphological features of the sintered mixture image (where the fog concentration is less than a preset value), a soft measurement sub-model for the moisture content of the sintered mixture based on image features is established, including: Calculate the gray-level co-occurrence matrix of the sintered mixture image, and calculate the first feature based on the gray-level co-occurrence matrix, wherein the first feature is specifically a texture feature; The second feature of the sintered mixture image in the HSV color space is extracted. The second feature includes the mean, standard deviation, and slope features. The OTSU algorithm is used to separate the shadows of material blocks from the sintering mixture image, and the maximum shadow area, average shadow area and number of shadows are calculated to obtain the third feature; The first feature, the second feature, and the third feature are input into the XGBoost model for training to obtain a soft measurement sub-model of sintering mixture moisture based on image features.

6. The online soft measurement method for moisture content of sintering mixtures based on multi-source information fusion according to claim 1, characterized in that, Based on the image feature-based soft measurement sub-model of sintered mixture moisture content and the GRU-based soft measurement sub-model of sintered mixture moisture content, an adaptive soft measurement model of sintered mixture moisture content is established, including: The average fog concentration of the sintering mixture image is calculated based on the linear model for fog concentration estimation. Based on the average fog concentration, the image feature-based soft measurement sub-model of sintered mixture moisture content, and the GRU-based soft measurement sub-model of sintered mixture moisture content, an adaptive soft measurement model of sintered mixture moisture content is established, and the calculation formula of the adaptive soft measurement model of sintered mixture moisture content is as follows: Where y is the output of the adaptive sintering mix moisture soft measurement model, y1 and y2 are the outputs of the image feature-based sintering mix moisture soft measurement sub-model and the GRU-based sintering mix moisture soft measurement sub-model, respectively, and μ average The average fog concentration is given by 0 ≤ μ. average ≤1.

7. An online soft measurement system for moisture content of sintering mixes based on multi-source information fusion, the system comprising: The memory (20), the processor (30), and the computer program stored in the memory (20) and executable on the processor (30) are characterized in that the processor (30) implements the steps of the method according to any one of claims 1 to 6 when executing the computer program.

Images