A method and system for predicting ferrous oxide content during sintering.

Abstract

This invention relates to the field of automated detection technology, and more particularly to a method and system for predicting ferrous oxide content during sintering. The invention first acquires original images, ferrous oxide content data, and environmental parameters during the fracture process of the sintered product. The best frame, best representing the cross-sectional characteristics, is selected as the original image. The original image is preprocessed, including grayscale processing and contrast enhancement. An SVM model and an improved BP neural network model are constructed and trained respectively to obtain a hierarchical model and multiple ferrous oxide content prediction models. In the testing phase, images from one cycle are acquired, and the best frame image is obtained through optimal frame processing. The hierarchical model is used to classify the images, and the corresponding ferrous oxide content prediction model is selected based on the classification results to predict the ferrous oxide content in the sintered product. Through multi-level training and model construction, and by fully utilizing image features and environmental parameters, the prediction accuracy and precision are improved.

Description

Technical Field

[0001] This invention relates to the field of automated detection technology, and in particular to a method and system for predicting the ferrous oxide content during sintering. Background Technology

[0002] Sintering is a preliminary step in blast furnace ironmaking. Iron ore is a crucial raw material for blast furnace ironmaking, and the process of turning lean iron ore powder into high-iron ore lumps is called sintering. On one hand, FeO content is an important indicator for evaluating the quality of sinter. Excessive FeO content reduces the reducibility of sinter, while insufficient FeO content affects its mechanical strength. On the other hand, the FeO content of sinter is influenced by many parameters, including the raw material ratio, ignition temperature, bed thickness, fuel moisture content, and fuel ratio. The FeO content of sinter reflects the dynamic control of sintering production; maintaining a stable FeO content indicates good control during the sintering process.

[0003] For the detection of ferrous oxide content in sintered ore, the commonly used methods in China are potassium dichromate chemical analysis and manual visual estimation at the tail end of the sintering machine. Potassium dichromate chemical analysis is a laboratory-based chemical detection method and is currently the most accurate. Sintered product samples from the steel plant are sent to the laboratory for chemical analysis every few hours to obtain the ferrous oxide content. However, this method also has its drawbacks. First, the chemical analysis is performed after the sintered product is obtained, by which time the entire sintering process is complete, and the obtained FeO content lags behind the sintering process, failing to provide timely feedback to guide the control of the sintering state. Second, the sample analyzed by the chemical analysis method is only a small part of the entire sintered product and cannot fully reflect the characteristics of the entire sintered product. Manual visual estimation at the tail end of the sintering machine refers to experienced fire watchers judging the FeO content of the sintered product through close observation of the machine tail section. This method was once widely used in domestic sintering plants, but its drawbacks include the harsh working environment, which is detrimental to the personal safety of the fire watchers, and its heavy reliance on the personal experience of the fire watchers.

[0004] Currently, some solutions have been proposed to assist in the detection of ferrous oxide content using machine learning algorithms. These solutions analyze various parameters affecting the sintering process and establish mathematical models for prediction. However, due to the numerous influencing parameters and complex mechanisms of the sintering process, general mathematical modeling methods are often insufficient for accurate prediction of FeO content. Currently, clustering and backpropagation (BP) neural networks are commonly used for modeling. Clustering is an unsupervised learning method that only models and predicts the FeO content level, not the numerical value. BP neural networks, on the other hand, have the ability to approximate complex functions, and are currently widely used for modeling and prediction. However, BP neural networks suffer from drawbacks such as high sensitivity to noise interference, heavy dependence on initial parameters, the need for multiple trials, slow learning speed, and susceptibility to local minima. Therefore, these methods cannot be directly applied to the accurate prediction of ferrous oxide content during the sintering process.

[0005] In summary, given the current problems in the detection of ferrous oxide content, there is an urgent need for a method that utilizes machine vision and neural network technology to achieve real-time prediction of ferrous oxide content during the sintering process. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method and system for predicting the ferrous oxide content during sintering, thereby achieving accurate prediction of the ferrous oxide content during sintering.

[0007] This invention provides a method for predicting the ferrous oxide content during sintering, comprising the following steps: S100: Obtain the original image of the sintering product fracture process and the corresponding ferrous oxide content and environmental parameters; wherein, the best frame that best reflects the cross-sectional characteristics during the fracture process of the sintering product is taken as the original image. S200. The original image is preprocessed, and the preprocessed original image is used as the first training sample. Feature parameters are calculated on the first training sample to obtain the feature vector corresponding to the first training sample, and the first training sample is graded and labeled to obtain the grade label corresponding to the first training sample; wherein, the preprocessing includes grayscale processing and contrast enhancement processing. S300: Calculate the correlation coefficient between the environmental parameters and the ferrous oxide content, and use the environmental parameters with the correlation coefficient greater than the preset value as the second training sample. S400. Construct an SVM model and train the SVM model based on the first training sample to obtain a hierarchical model; S500. Construct an improved BP neural network model. Based on the level label corresponding to the first training sample, divide the second training sample into multiple training groups of different levels according to the level division of the first training sample. Train the improved BP neural network model using the training groups of different levels respectively to obtain multiple ferrous oxide content prediction models of different levels. The environmental parameters of each training group are used as inputs, and the corresponding ferrous oxide content is used as output. S600. Acquire one cycle of the image to be tested and perform optimal frame processing to obtain the optimal frame image to be tested. Calculate the feature parameters of the optimal frame image to be tested and input them into the grading model for grading. Select the corresponding ferrous oxide content prediction model based on the grading results. Input the environmental parameters corresponding to the optimal frame image to be tested into the ferrous oxide content prediction model to predict the ferrous oxide content in the sintered product. Wherein, one cycle includes: the process of the sintered ore fracturing due to the pushing of the trolley and its own gravity.

[0008] Further, in step S200, the original image is preprocessed, and the preprocessed original image is used as a first training sample. Feature parameters are calculated on the first training sample to obtain the feature vector corresponding to the first training sample, and the first training sample is graded and labeled to obtain the grade label corresponding to the first training sample; wherein, the preprocessing includes grayscale processing and contrast enhancement processing. S210. Perform grayscale processing on the original image: perform grayscale conversion on the original image based on the R component of the RGB components; S220. Perform contrast enhancement processing on the original image after grayscale processing: Perform nonlinear transformation on the original image after grayscale processing through gamma filtering to increase image contrast and obtain the first training sample. S230. Calculate the feature parameters of the first training sample to obtain the feature vector corresponding to the first training sample; S240. The first training sample is graded and labeled according to the ferrous oxide content to obtain the ferrous oxide content grade label corresponding to the first training sample.

[0009] Further, in step S230, feature parameters are calculated on the first training sample to obtain the feature vector corresponding to the first training sample, including: S231. Perform Otsu's threshold segmentation on all images in the first training sample to obtain a binary image set. S232. Smooth all images in the binarized image set by performing morphological opening and closing operations to extract the red flame layer region and pore region of each image. S233. Perform pixel-based area calculations on each image from which the red flame layer region and pore region have been extracted to obtain the area ratio of the red flame layer region and pore region in each image, and construct the feature vector of the first training sample set.

[0010] Further, in step S300, the correlation coefficients are calculated by correlating the environmental parameters with the ferrous oxide content, and the environmental parameters with correlation coefficients greater than a preset value are used as the second training samples, including: The Pearson correlation coefficient is used to calculate the correlation. The formula for calculating the correlation coefficient is as follows:

[0011] in, The correlation coefficient is... This represents the total amount of data for environmental parameters and their corresponding ferrous oxide content. These are environmental parameters acquired through sensors. This represents the average value of environmental parameters. The content of ferrous oxide, This represents the average ferrous oxide content.

[0012] Furthermore, in S500, constructing the improved BP neural network model includes: S510. The BP neural network model is a three-layer network structure; S520. The additional momentum method is used to perform smooth weight updates, which causes the weight adjustment to change in the average direction of the bottom of the error surface; S530: Automatically adjust the learning rate of the BP neural network using the adaptive learning rate method; S540. Optimize the parameters of the BP neural network using a genetic algorithm.

[0013] Furthermore, in S540, the optimization of the BP neural network by the genetic algorithm includes: S541. Determine the lengths of the weights and thresholds based on the topology of the BP neural network; S542. Randomly assign values ​​to the weights and thresholds of the BP neural network, and encode the randomly assigned weights and thresholds with real numbers to form the first generation chromosome in the genetic algorithm; S543. Determine the fitness function, wherein the fitness function aims to minimize the error between the ferrous oxide content predicted by the BP neural network and the true value of the ferrous oxide content; S544. Set the parameters of the genetic algorithm, and iteratively optimize the parameters of the genetic algorithm through selection, crossover and mutation operations until the set stopping condition is met to obtain the final optimization result. Decode the optimal result to obtain the optimal weights and thresholds, and apply the optimal weights and thresholds to the training of the BP neural network to improve the network performance.

[0014] Furthermore, in S600, the image to be tested for one cycle is acquired and the best frame is processed to obtain the best frame image to be tested. The best frame processing includes: acquiring the brightness value of each frame image within one cycle, calculating the brightness difference between each frame image and the previous frame image, determining whether the brightness difference between frames is greater than a preset threshold, and when the brightness difference between frames is greater than the preset threshold, acquiring the current frame image and the next four consecutive frames at that moment, and selecting the frame image with the highest brightness value among these five frames as the best frame image.

[0015] The present invention also provides a prediction system for ferrous oxide content during sintering, used to implement any of the above-mentioned methods for predicting ferrous oxide content during sintering. The prediction system includes: a data acquisition module, a data processing module, a model building module, and a data analysis module. The data acquisition module is used to acquire the original images of the fracture process of the sintering product and the corresponding ferrous oxide content and environmental parameters. The data processing module is connected to the data acquisition module and is used to preprocess the original image. The preprocessing includes grayscale processing and contrast enhancement processing to obtain a first training sample. The first training sample is then subjected to feature parameter calculation and hierarchical labeling to obtain the feature vector and level label corresponding to the first training sample. The correlation coefficient between environmental parameters and ferrous oxide content is calculated, and environmental parameters with a correlation coefficient greater than a preset value are used as the second training sample. The model building module is connected to the data processing module to build an SVM model, which is trained using the first training sample to obtain a hierarchical model; and to build multiple improved BP neural network models corresponding to the hierarchical model level, which are trained by dividing the second training sample into multiple training groups to obtain multiple ferrous oxide content prediction models corresponding to the hierarchical model level. The data analysis module and model building module are used to collect the image to be tested for one cycle, process the best frame to obtain the best frame image to be tested, calculate the feature parameters of the best frame image to be tested and input it into the grading model for grading, and select the corresponding ferrous oxide content prediction model according to the grading results. The environmental parameters corresponding to the best frame image to be tested are input into the ferrous oxide content prediction model to realize the prediction of ferrous oxide content in sintering products.

[0016] Furthermore, the feature parameter calculation of the data processing module includes: performing Otsu's threshold segmentation on all images in the first training sample to obtain a binary image set; performing smoothing processing on all images in the binary image set through morphological opening and closing operations to extract the red flame layer region and pore region of each image; performing pixel-based area calculation on each image of the extracted red flame layer region and pore region to obtain the red flame layer area and pore ratio of each image, and constructing the feature vector of the first training sample set.

[0017] The embodiments of the present invention have the following technical effects: 1. Grading before quantification: The strategy of grading before quantification is adopted. First, the image to be tested is acquired and the best frame is processed. The feature parameters of the best frame image are calculated and then input into the grading model for grading. Based on the grading results, the corresponding ferrous oxide content prediction model is selected. The environmental parameters corresponding to the best frame image are input into the ferrous oxide content prediction model to predict the ferrous oxide content in the sintering product. Compared with the existing technology of directly predicting the ferrous oxide content, the grading before quantification strategy has the advantages of high prediction accuracy and fast prediction speed in predicting ferrous oxide content.

[0018] 2. Optimization of BP Neural Network: The BP neural network is optimized by using the additional momentum method, adaptive learning rate method and genetic algorithm, which comprehensively improves the performance of the BP neural network and gets rid of the problems of poor global search ability, easy to get trapped in local optima and slow convergence speed of the standard BP algorithm. This comprehensive optimization scheme makes the neural network more suitable for predicting the ferrous oxide content in complex sintering processes.

[0019] 3. Optimal Frame Selection: By analyzing the brightness value and the brightness difference between frames, the frame that best reflects the cross-sectional characteristics during the fracture process of the sintered product is selected as the optimal frame image. This ensures that the selected image has the highest information quality and helps to improve the accuracy of ferrous oxide content prediction. Attached Figure Description

[0020] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is a flowchart of a method for predicting ferrous oxide content during sintering, provided by an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a ferrous oxide content prediction system provided in an embodiment of the present invention during the sintering process. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0023] Figure 1 This is a flowchart illustrating a method for predicting ferrous oxide content during sintering, provided in an embodiment of the present invention. See also... Figure 1 It includes the following steps: S100: Obtain the original image of the sintering product fracture process and the corresponding ferrous oxide content and environmental parameters; wherein, the best frame that best reflects the cross-sectional characteristics during the fracture process of the sintering product is taken as the original image. S200. The original image is preprocessed, and the preprocessed original image is used as the first training sample. Feature parameters are calculated on the first training sample to obtain the feature vector corresponding to the first training sample, and the first training sample is graded and labeled to obtain the grade label corresponding to the first training sample; wherein, the preprocessing includes grayscale processing and contrast enhancement processing. Specifically, including: S210. Perform grayscale processing on the original image: perform grayscale conversion on the original image based on the R component of the RGB components to reduce the interference of color changes on image features in the original image.

[0024] S220. Perform contrast enhancement processing on the original image after grayscale processing: Apply a gamma filter to the original image after grayscale processing using a non-linear transformation to increase image contrast, obtaining the first training sample. Gamma filtering is an image enhancement algorithm that is effective for images with low contrast and high overall brightness. The mathematical formula is as follows:

[0025] in, This indicates the adjusted output image; This is the grayscale scaling factor, usually set to 1; This represents the grayscale value of a pixel in the original image; The gamma factor controls the scaling of the entire transformation. When this happens, the grayscale value of the original image will be reduced. When this happens, the grayscale value of the original image will be increased.

[0026] S230. Calculate the feature parameters of the first training sample to obtain the feature vector corresponding to the first training sample; Specifically, the calculation of characteristic parameters includes: S231. Perform Otsu's threshold segmentation on all images in the first training sample to obtain a binary image set. The purpose of this step is to transform the original image into an image containing only black and white pixel values, providing a basis for subsequent processing.

[0027] S232. Morphological opening and closing operations are used to smooth all images in the binarized image set, extracting the red flame layer region and pore region of each image. Morphological opening operations are used to remove small holes and connected regions, extracting the red flame layer region. Morphological closing operations are used to connect smaller broken regions, extracting the pore region. This accurately segments the red flame layer and pores, providing accurate regions for subsequent feature calculations.

[0028] S233. Perform pixel-based area calculations on each image from which the red flame layer region and pore region have been extracted to obtain the area ratio of the red flame layer region and pore region in each image, and construct the feature vector of the first training sample set.

[0029] S240. The first training sample is graded and labeled according to the ferrous oxide content to obtain the ferrous oxide content grade label corresponding to the first training sample.

[0030] S300: Calculate the correlation coefficient between the environmental parameters and the ferrous oxide content, and use the environmental parameters with the correlation coefficient greater than the preset value as the second training sample. Specifically, the correlation calculation includes: The Pearson correlation coefficient is used to calculate the correlation. The formula for calculating the correlation coefficient is as follows:

[0031] in, The correlation coefficient is... This represents the total amount of data for environmental parameters and their corresponding ferrous oxide content. These are environmental parameters acquired through sensors. This represents the average value of environmental parameters. The content of ferrous oxide, The average ferrous oxide content is used. By using environmental parameters with correlation coefficients greater than a preset value as the second training sample, the influence of these parameters can be given more attention during model training, thereby improving the model's prediction accuracy and reliability. The data obtained after calculating the correlation coefficients still needs to undergo certain processing, including outlier removal and normalization, to further optimize the correlation calculation results.

[0032] S400. Construct an SVM model and train the SVM model based on the first training sample to obtain a hierarchical model; S500. Construct an improved BP neural network model. Based on the level label corresponding to the first training sample, divide the second training sample into multiple training groups of different levels according to the level division of the first training sample. Train the improved BP neural network model using the training groups of different levels respectively to obtain multiple ferrous oxide content prediction models of different levels. The environmental parameters of each training group are used as inputs, and the corresponding ferrous oxide content is used as output. Specifically, constructing an improved BP neural network model includes: S510. The BP neural network model described is a three-layer network structure. A BP neural network generally consists of an input layer, an output layer, and several hidden layers, each layer composed of several unconnected neurons. The number of neurons in the input layer is equal to the number of independent variables in the input model to be constructed. The number of neurons in the output layer is the number of dependent variables to be predicted; in this embodiment, the output layer is 1. The number of neurons in the hidden layers is not strictly defined and is generally adjusted through empirical formulas and experimental results. Kolmogorov's theorem has proven that a three-layer neural network can simulate arbitrarily complex nonlinear models; this embodiment uses a three-layer BP neural network.

[0033] S520. A smooth weight update is performed using the additional momentum method, causing the weight adjustment to change towards the average direction of the bottom of the error surface. Specifically, the additional momentum method involves adding a portion of the previous weight adjustment to the weight adjustment calculated based on the current error, using this as the actual weight adjustment for the current operation, thus causing the weight adjustment to change towards the average direction of the bottom of the error surface. The mathematical formula is as follows:

[0034] in, Indicates the first The adjustment amount of the weights in the next iteration; Momentum factor , No. -1 iteration output error, which is taken as 0.9 in this invention; This represents the learning rate, with a value range of (0,1). Indicates the first The output error of the next iteration. By employing the additional momentum method, during gradient descent of the search network, if the current gradient descent direction is the same as the previous gradient descent direction, the search will accelerate; otherwise, it will slow down. This improvement helps the network find a better global solution during computation, reduces the network's sensitivity to local details of the error surface, effectively avoids the network getting trapped in local extrema, and reduces network oscillations.

[0035] S530. Use the adaptive learning rate method to automatically adjust the learning rate of the BP neural network. The specific method of the adaptive learning rate method is that if the error function decreases after the weights are corrected, it indicates that the adjustment direction of the weights and thresholds is correct, and the training step size can be increased by multiplying the learning rate by the incremental learning factor; otherwise, a decreasing factor is used to multiply the learning rate.

[0036] The formula is adjusted as follows:

[0037] In the formula, This represents the learning rate of the BP neural network, with a value range of (0,1). This is the learning rate increment factor, typically 1.05; This is the learning rate reduction factor, typically 0.7; Indicates the first The error between the expected value and the actual value. Generally, a larger learning rate leads to faster network convergence, but an excessively large learning rate can cause network instability; conversely, an excessively small learning rate slows down network convergence. Adaptive learning rate methods can automatically adjust the learning rate based on changes in network error, increasing the learning rate in flat regions and decreasing it as the slope increases, thereby accelerating network convergence.

[0038] S540. Optimize the parameters of the BP neural network using a genetic algorithm.

[0039] Specifically, the optimization of BP neural networks by genetic algorithms includes: S541. Determine the lengths of the weights and thresholds based on the topology of the BP neural network; the solution required by the genetic algorithm is the weights and thresholds of the BP neural network. A chromosome is expressed as a set of floating-point vectors, and the length of the chromosome is the sum of the number of weights and thresholds in the BP neural network. The calculation formula is:

[0040] in, Chromosome length; This represents the number of neurons in the input layer of the BP neural network. This represents the number of neurons in the hidden layer. This represents the number of neurons in the output layer.

[0041] S542. Randomly assign values ​​to the weights and thresholds of the BP neural network, and encode the randomly assigned weights and thresholds with real numbers to form the first generation chromosome in the genetic algorithm; S543. Determine the fitness function, wherein the fitness function aims to minimize the error between the ferrous oxide content predicted by the BP neural network and the actual value of the ferrous oxide content; the fitness function of the genetic algorithm in this embodiment should be:

[0042] Where P represents an individual in the population; This represents the actual output of the BP neural network configured with P individuals; This represents the corresponding expected output. This represents the fitness of individual P. Since the output error of a BP neural network is always non-negative, the fitness function does not require scaling transformation.

[0043] S544. Set the parameters of the genetic algorithm, and iteratively optimize the parameters of the genetic algorithm through selection, crossover and mutation operations until the set stopping condition is met to obtain the final optimization result. Decode the optimal result to obtain the optimal weights and thresholds, and apply the optimal weights and thresholds to the training of the BP neural network to improve the network performance.

[0044] Selection operation: Calculate the fitness value of each individual, calculate the probability of each individual based on the fitness value, and select the individual with the higher probability value for crossover and mutation operations. The selection operation uses a roulette wheel algorithm, assuming there are a total of [number missing] individuals in the population. Individuals, among which individuals The probability of being selected is Its calculation formula is:

[0045] In the formula, Represents an individual The probability of being selected; Indicates the chromosome to be calculated; Represents an individual The fitness of; This represents the sum of the fitness of all individuals in the population.

[0046] Crossover operation: Using real-number crossover, two chromosomes are linearly combined according to the crossover probability to generate a new chromosome. The new chromosome replaces the original chromosome, while chromosomes that have not undergone crossover are directly replicated. and In the vector The cross operation is performed on each position, and the calculation formula is:

[0047] In the formula, and They represent chromosomes respectively. and The Genes at each location; It is a random number between [0,1].

[0048] Mutation operation: Non-uniform mutation is used. According to the mutation probability, a certain gene is randomly perturbed, and the result of the perturbation is used as the new gene value. In position The mutations that occur are calculated using the following formula:

[0049]

[0050] In the formula, Chromosomes In the Genes at each location; For genes The upper bound; For genes The lower bound; A random number between [0, 1]; The formula for the variation factor related to evolutionary generations used in the calculation is given by: Let be the algebra of the current iteration; The maximum number of generations is set. It is a random number between [0,1].

[0051] Backpropagation (BP) neural networks solve for the minimum error function using gradient descent. However, they are highly sensitive to network parameters such as initial weights and thresholds, and suffer from drawbacks including poor global search capability, susceptibility to local optima, and slow convergence. Genetic algorithms, on the other hand, can adaptively optimize the search space and adjust the search direction, continuously optimizing the connection weights and thresholds of each layer in the BP neural network, thus reducing the standard BP algorithm's over-reliance on gradient information.

[0052] S600. Acquire one cycle of the image to be tested and perform optimal frame processing to obtain the optimal frame image to be tested. Calculate the feature parameters of the optimal frame image to be tested and input them into the grading model for grading. Select the corresponding ferrous oxide content prediction model based on the grading results. Input the environmental parameters corresponding to the optimal frame image to be tested into the ferrous oxide content prediction model to predict the ferrous oxide content in the sintered product. One cycle includes the process of the sintered ore fracturing due to the pushing of the trolley and its own gravity.

[0053] Specifically, the optimal frame processing includes: acquiring the brightness value of each frame image within a period, calculating the brightness difference between each frame image and the previous frame image, determining whether the inter-frame brightness difference is greater than a preset threshold, and when the inter-frame brightness difference is greater than the preset threshold, acquiring the current frame image and the next four consecutive frames at that moment, and selecting the frame image with the highest brightness value from these five frames as the optimal frame image.

[0054] Figure 2 This is a schematic diagram of a system for predicting ferrous oxide content during sintering, provided in an embodiment of the present invention. (See also...) Figure 2 The system includes: a data acquisition module, a data processing module, a model building module, and a data analysis module; The data acquisition module is used to acquire the original images of the fracture process of the sintering product and the corresponding ferrous oxide content and environmental parameters. The data processing module is connected to the data acquisition module and is used to preprocess the original image. The preprocessing includes grayscale processing and contrast enhancement processing to obtain a first training sample. The first training sample is then subjected to feature parameter calculation and hierarchical labeling to obtain the feature vector and level label corresponding to the first training sample. The correlation coefficient between environmental parameters and ferrous oxide content is calculated, and environmental parameters with a correlation coefficient greater than a preset value are used as the second training sample. The model building module is connected to the data processing module to build an SVM model, which is trained using the first training sample to obtain a hierarchical model; and to build multiple improved BP neural network models corresponding to the hierarchical model level, which are trained by dividing the second training sample into multiple training groups to obtain multiple ferrous oxide content prediction models corresponding to the hierarchical model level. The data analysis module is connected to the model building module. It is used to collect the image to be tested for one cycle, process the best frame to obtain the best frame image to be tested, calculate the feature parameters of the best frame image to be tested, and input it into the grading model for grading. Based on the grading results, the corresponding ferrous oxide content prediction model is selected. The environmental parameters corresponding to the best frame image to be tested are input into the ferrous oxide content prediction model to realize the prediction of ferrous oxide content in sintering products.

[0055] Furthermore, the feature parameter calculation of the data processing module includes: performing Otsu's threshold segmentation on all images in the first training sample to obtain a binary image set; performing smoothing processing on all images in the binary image set through morphological opening and closing operations to extract the red flame layer region and pore region of each image; performing pixel-based area calculation on each image of the extracted red flame layer region and pore region to obtain the red flame layer area and pore ratio of each image, and constructing the feature vector of the first training sample set.

[0056] It should be noted that the terminology used in this invention is for describing specific embodiments only and is not intended to limit the scope of this application. As shown in this specification, unless the context clearly indicates otherwise, words such as "a," "an," "an," and / or "the" do not specifically refer to the singular and may include the plural. The terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, or apparatus that includes said element.

[0057] It should also be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Unless otherwise expressly specified and limited, the terms "installed," "connected," "linked," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components. For those skilled in the art, the specific meaning of the above terms in the present invention can be understood according to the specific circumstances.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting ferrous oxide content during sintering, characterized in that, Includes the following steps: S100: Obtain the original image of the sintering product fracture process and the corresponding ferrous oxide content and environmental parameters of the original image; wherein, the best frame that best reflects the cross-sectional characteristics during the fracture process of the sintering product is taken as the original image. S200. The original image is preprocessed, and the preprocessed original image is used as the first training sample. Feature parameters are calculated on the first training sample to obtain the feature vector corresponding to the first training sample, and the first training sample is graded and labeled to obtain the grade label corresponding to the first training sample; wherein, the preprocessing includes grayscale processing and contrast enhancement processing. S300: Calculate the correlation coefficient between the environmental parameters and the ferrous oxide content, and use the environmental parameters with the correlation coefficient greater than a preset value as the second training sample. S400. Construct an SVM model and train the SVM model based on the first training sample to obtain a hierarchical model; S500. Construct an improved BP neural network model. Based on the level label corresponding to the first training sample, divide the second training sample into multiple training groups of different levels according to the level division of the first training sample. Train the improved BP neural network model using the training groups of different levels respectively to obtain multiple ferrous oxide content prediction models of different levels. The environmental parameters of each training group are used as inputs, and the corresponding ferrous oxide content is used as output. S600: Acquire one cycle of the image to be tested, perform optimal frame processing to obtain the optimal frame image to be tested, calculate the feature parameters of the optimal frame image to be tested and input them into the grading model for grading, and select the corresponding ferrous oxide content prediction model according to the grading results. Input the environmental parameters corresponding to the optimal frame image to be tested into the ferrous oxide content prediction model to predict the ferrous oxide content in the sintering product; wherein, one cycle includes: the process of the sintered ore fracturing due to the pushing of the trolley and its own gravity; The S500 method for constructing an improved BP neural network model includes: S510. The BP neural network model is a three-layer network structure; S520. The additional momentum method is used to perform smooth weight updates, which causes the weight adjustment to change in the average direction of the bottom of the error surface; S530: Automatically adjust the learning rate of the BP neural network using the adaptive learning rate method; S540. Optimize the parameters of the BP neural network using a genetic algorithm; In S540, the optimization of the BP neural network by the genetic algorithm includes: S541. Determine the lengths of the weights and thresholds based on the topology of the BP neural network; S542. Randomly assign values ​​to the weights and thresholds of the BP neural network, and encode the randomly assigned weights and thresholds with real numbers to form the first generation chromosome in the genetic algorithm; S543. Determine the fitness function, wherein the fitness function aims to minimize the error between the ferrous oxide content predicted by the BP neural network and the true value of the ferrous oxide content; S544. Set the parameters of the genetic algorithm, and iteratively optimize the parameters of the genetic algorithm through selection, crossover and mutation operations until the set stopping condition is met to obtain the final optimization result. Decode the optimal result to obtain the optimal weight and optimal threshold, and apply the optimal weight and optimal threshold to the training of the BP neural network to improve the network performance.

2. The method for predicting ferrous oxide content during sintering according to claim 1, characterized in that, In step S200, the original image is preprocessed, and the preprocessed original image is used as a first training sample. Feature parameters are calculated on the first training sample to obtain the feature vector corresponding to the first training sample, and the first training sample is graded and labeled to obtain the grade label corresponding to the first training sample. The feature vector and grade label corresponding to the first training sample include: S210. Perform grayscale processing on the original image: perform grayscale conversion on the original image based on the R component of the RGB components; S220. Perform contrast enhancement processing on the original image after grayscale processing: Perform nonlinear transformation on the original image after grayscale processing through gamma filtering to increase image contrast and obtain the first training sample. S230. Calculate the feature parameters of the first training sample to obtain the feature vector corresponding to the first training sample; S240. The first training sample is graded and labeled according to the ferrous oxide content to obtain the ferrous oxide content grade label corresponding to the first training sample.

3. The method for predicting ferrous oxide content during sintering according to claim 2, characterized in that, In step S230, feature parameters are calculated on the first training sample to obtain the feature vector corresponding to the first training sample, including: S231. Perform Otsu's threshold segmentation on all images in the first training sample to obtain a binary image set. S232. Smooth all images in the binarized image set by performing morphological opening and closing operations to extract the red flame layer region and pore region of each image. S233. Perform pixel-based area calculations on each image from which the red flame layer region and pore region have been extracted to obtain the area ratio of the red flame layer region and pore region in each image, and construct the feature vector of the first training sample set.

4. The method for predicting ferrous oxide content during sintering according to claim 1, characterized in that, In step S300, the correlation coefficients are calculated by relating the environmental parameters to the ferrous oxide content, and the environmental parameters with correlation coefficients greater than a preset value are used as the second training samples, including: The Pearson correlation coefficient is used to calculate the correlation. The formula for calculating the correlation coefficient is as follows: in, The correlation coefficient is... This represents the total amount of data for environmental parameters and their corresponding ferrous oxide content. These are environmental parameters acquired through sensors. This represents the average value of environmental parameters. The content of ferrous oxide, This represents the average ferrous oxide content.

5. The method for predicting ferrous oxide content during sintering according to claim 1, characterized in that, In S600, the image to be tested for one cycle is acquired and the best frame is processed to obtain the best frame image to be tested. The best frame processing includes: acquiring the brightness value of each frame image within one cycle, calculating the brightness difference between each frame image and the previous frame image, determining whether the brightness difference between frames is greater than a preset threshold, and when the brightness difference between frames is greater than the preset threshold, acquiring the current frame image and the next four consecutive frames at that moment, and selecting the frame image with the highest brightness value among these five frames as the best frame image.

6. A system for predicting ferrous oxide content during sintering, used to implement the method for predicting ferrous oxide content during sintering as described in any one of claims 1-5, characterized in that, include: Data acquisition module, data processing module, model building module, and data analysis module; The data acquisition module is used to acquire the original images of the fracture process of the sintering product and the corresponding ferrous oxide content and environmental parameters. The data processing module is connected to the data acquisition module and is used to preprocess the original image. The preprocessing includes grayscale processing and contrast enhancement processing to obtain a first training sample. The first training sample is used to calculate feature parameters and perform hierarchical labeling to obtain the feature vector and level label corresponding to the first training sample. The correlation coefficient between environmental parameters and ferrous oxide content is calculated, and environmental parameters with a correlation coefficient greater than a preset value are used as second training samples. The model building module is connected to the data processing module and is used to build an SVM model. It is trained with the first training sample to obtain a hierarchical model. Multiple improved BP neural network models corresponding to the hierarchical model level are built. By dividing the second training sample into multiple training groups for training, multiple ferrous oxide content prediction models corresponding to the hierarchical model level are obtained. The data analysis module is connected to the model building module. It is used to collect the image to be tested for one cycle, process the best frame to obtain the best frame image to be tested, calculate the feature parameters of the best frame image to be tested, and input it into the grading model for grading. Based on the grading results, the corresponding ferrous oxide content prediction model is selected. The environmental parameters corresponding to the best frame image to be tested are input into the ferrous oxide content prediction model to realize the prediction of ferrous oxide content in sintering products.

7. The system for predicting ferrous oxide content during sintering according to claim 6, characterized in that, The feature parameter calculation of the data processing module includes: performing Otsu's threshold segmentation on all images in the first training sample to obtain a binary image set; performing smoothing processing on all images in the binary image set through morphological opening and closing operations to extract the red flame layer region and pore region of each image; performing pixel-based area calculation on each image of the extracted red flame layer region and pore region to obtain the red flame layer area and pore ratio of each image, and constructing the feature vector of the first training sample set.

Images