Cement production quality control method based on AI technology

Abstract

The present application relates to the technical field of cement production quality control, and specifically to a cement production quality control method based on AI technology, the method comprising: obtaining a near-infrared spectrum of cement raw materials, and obtaining a standard near-infrared spectrum of each main component of the cement raw materials respectively; determining the spectral fluctuation of each band in the near-infrared spectrum; determining the fluctuation characteristic value and inter-segment difference of each main component corresponding to each band in the near-infrared spectrum; obtaining multiple clusters of each main component; determining the adjustment coefficient, filter window size and filter order of each cluster of each main component; obtaining a characteristic enhanced spectrum of each main component, and controlling the content of each main component in the cement raw materials. The present application can enhance the protection of the characteristic spectral information of the main components in the near-infrared spectrum when denoising the near-infrared spectrum, and improve the detection accuracy of the main component content of the cement raw materials.

Description

Technical Field

[0001] The present application relates to the technical field of cement production quality control, and specifically to a cement production quality control method based on AI technology. Background Art

[0002] The cement industry is one of the important basic raw material industries for economic development, including raw material preparation, clinker calcination and cement production. Among them, cement raw material batching is the primary link in cement production. The quality of cement raw material has an important impact on the subsequent clinker burning, and largely determines the quality of cement finished products. The composition of cement raw material is the main indicator for evaluating the quality of raw material. Rapid and accurate detection of cement raw material composition is the prerequisite for achieving real-time control of cement production quality.

[0003] Near-infrared spectroscopy detection technology uses near-infrared spectral sensors to collect near-infrared spectra of substances and conduct quantitative analysis of the composition of substances. As an efficient, fast, safe and pollution-free non-destructive analysis technology, it is currently a common method for detecting the content of cement raw materials in the industry. The near-infrared spectrum of cement raw materials obtained by near-infrared spectral sensors contains spectral information of different components of cement raw materials, as well as a large amount of interference noise. When a filtering algorithm is used to remove the interference noise of the near-infrared spectrum of cement raw materials, its fixed filter window size and filter order will lose the spectral information of the main components in the cement raw materials while denoising, which ultimately affects the accuracy of near-infrared spectroscopy detection of the content of the main components of cement raw materials, resulting in a decrease in the stability of cement raw material production quality. Summary of the invention

[0004] In order to solve the above technical problems, a cement production quality control method based on AI technology is provided to solve the existing problems.

[0005] The solution to the technical problem of this application is to provide a cement production quality control method based on AI technology, including the following steps:

[0006] Obtaining the near infrared spectrum of cement raw meal and obtaining the standard near infrared spectrum of each main component in cement raw meal;

[0007] The wavenumber ranges in the near infrared spectrum and the standard near infrared spectrum of each main component are divided into multiple bands respectively; the spectral fluctuation degree of each band in the near infrared spectrum is determined according to the variation trend of the absorbance in each band in the near infrared spectrum and the correlation degree between the absorbance and the wavenumber;

[0008] Determine the fluctuation characteristic value of each main component in each band of the near infrared spectrum according to the similarity of the local changes of the spectral fluctuations in the same band of the near infrared spectrum and the standard near infrared spectrum of different main components, combined with the spectral fluctuations;

[0009] Analyze the distance relationship between different bands in the near-infrared spectrum, as well as the difference in the fluctuation characteristic value of each main component between different bands in the near-infrared spectrum, and determine the inter-segment difference between each band in the near-infrared spectrum and each main component;

[0010] Based on the fluctuation characteristic value and the inter-segment difference of each main component in each band of the near-infrared spectrum, an iterative formula of the membership matrix is ​​determined, and all bands in the near-infrared spectrum are clustered in combination with a clustering algorithm to obtain multiple clusters of each main component;

[0011] Determine the adjustment coefficient of each cluster of each main component based on the fluctuation characteristic values ​​of all bands in each cluster and the inter-segment difference; determine the filter window size and filter order of each cluster of each main component based on the adjustment coefficient;

[0012] Combined with the filtering algorithm, all bands in the near-infrared spectrum are filtered separately to obtain the characteristic enhanced spectrum of each main component, and the content of each main component in the cement raw material is controlled.

[0013] Preferably, the determining of the spectral fluctuation of each band in the near-infrared spectrum comprises:

[0014] The wave numbers in each band of the near-infrared spectrum and their corresponding absorbances form a two-dimensional array; linear fitting is performed on all two-dimensional arrays in each band to obtain the slope of the fitting line;

[0015] Calculate the absolute value of the correlation between all absorbances and all wave numbers in each band of the near-infrared spectrum as the correlation coefficient of each band in the near-infrared spectrum;

[0016] Calculate the mean of all absorbances in each band of the near-infrared spectrum;

[0017] The ratio of the square result of the slope of each band in the near-infrared spectrum to the correlation coefficient is calculated, and the product of the ratio and the mean is used as the spectral fluctuation of each band in the near-infrared spectrum.

[0018] Preferably, determining the fluctuation characteristic value of each main component corresponding to each band in the near-infrared spectrum includes:

[0019] A local window of a preset size is set with each band in the near infrared spectrum and the standard near infrared spectrum of each main component as the center;

[0020] The normalized results of the spectral fluctuations of all bands in the local window are used to form the local fluctuation vectors of each band in the near-infrared spectrum and the standard near-infrared spectrum of each main component;

[0021] The similarity of the local fluctuation vectors in the same band between the near-infrared spectrum and the standard near-infrared spectrum of each main component is calculated respectively as the fluctuation similarity of each band in the near-infrared spectrum corresponding to each main component;

[0022] The sum of the fluctuation similarity and a preset initial threshold is calculated, and the product of the sum and the spectral fluctuation is used as the fluctuation characteristic value of each main component corresponding to each band in the near-infrared spectrum.

[0023] Preferably, determining the inter-segment difference between each band in the near-infrared spectrum corresponding to each main component includes:

[0024] Obtaining the segmentation thresholds of the fluctuation characteristic values ​​of each main component in all bands of the near-infrared spectrum respectively, recording the bands of each main component whose fluctuation characteristic values ​​are greater than the segmentation thresholds as the characteristic bands of each main component; numbering each band in the near-infrared spectrum in ascending order of wave number;

[0025] The calculation formula for the inter-segment difference of each main component in each band of the near-infrared spectrum is: ,in, In near infrared spectroscopy The band corresponds to The inter-segment differences of the main components, In near infrared spectroscopy The serial number of the band, is the near-infrared spectrum corresponding to The main ingredient The serial number of the characteristic band, is the near-infrared spectrum corresponding to The main ingredient The fluctuation characteristic value of the characteristic band, In the near infrared spectrum The band corresponds to The fluctuation eigenvalues ​​of the main components, To preset a value greater than 0, is the near-infrared spectrum corresponding to The number of all characteristic bands of the principal component.

[0026] Preferably, the iterative formula for determining the membership matrix includes:

[0027] The fluctuation characteristic value and inter-segment difference of each main component in each band of the near-infrared spectrum are used to form the characteristic vector of each main component in each band of the near-infrared spectrum;

[0028] Calculate the quartile of the modulus of the eigenvector of each main component corresponding to all bands in the near-infrared spectrum, and use the eigenvector with a modulus value equal to the quartile as the initial cluster center of each main component;

[0029] Based on the initial cluster centers, an iterative formula for the membership matrix of each main component is determined.

[0030] Preferably, the iterative formula of the membership matrix is ​​specifically: ,in, In near infrared spectroscopy The band belongs to The main ingredient The membership degree of each cluster center to the corresponding category, In near infrared spectroscopy The serial number of the band, For the The corresponding serial number of each cluster center in the near infrared spectrum, In near infrared spectroscopy The band corresponds to The eigenvectors of the principal components, For the The main ingredient Cluster centers, For the The number of all cluster centers of the principal components, is an exponential function with a natural constant as base, is the distance calculation function, To preset a value greater than 0, is the normalization function.

[0031] Preferably, the adjustment coefficient of each cluster of each main component is a normalized result of the mean value of the modulus of the eigenvectors of all bands in each cluster corresponding to each main component.

[0032] Preferably, the filter window size of each cluster of each main component is determined by the calculation formula: ,in, For the The main ingredient The filter window size for clusters, To preset the minimum window size, For the The maximum value of the adjustment coefficient of all clusters of the principal components, For the The main ingredient The adjustment coefficient of the clusters, For the The minimum value of the adjustment coefficient of all clusters of the principal components.

[0033] Preferably, the calculation formula of the filter order is: ,in, For the The main ingredient The filter order of clusters, To preset the minimum filter order, For the The main ingredient The adjustment coefficient of the clusters, For the The maximum value of the adjustment coefficients of all clusters of the principal components.

[0034] Preferably, the controlling the content of each main component in the cement raw material comprises:

[0035] The characteristic enhanced spectrum of each main component in the cement raw material is input into the neural network model to obtain the predicted content of each main component in the cement raw material;

[0036] The predicted content of each main component in the cement raw material is input into the PLC system to control and adjust the proportion and feed amount of the cement raw material.

[0037] This application has at least the following beneficial effects:

[0038] The present application determines the spectral fluctuation of each band in the near-infrared spectrum according to the variation trend of the absorbance in each band in the near-infrared spectrum and the correlation between the absorbance and the wavenumber. The beneficial effect of the application is that it reflects the spectral information contained in each band. The fluctuation characteristic value of each main component in each band of the near-infrared spectrum is determined according to the similarity of the local variation of the spectral fluctuation of the near-infrared spectrum and the same band of the standard near-infrared spectrum of different main components. The beneficial effect is that the similarity of spectral fluctuations in each band between the near-infrared spectrum of cement raw materials and the standard near-infrared spectrum of each main component is taken into account to reflect the spectral information of the corresponding band containing the corresponding main component, so as to retain its characteristic spectral information during subsequent filtering; the distance relationship between different bands in the near-infrared spectrum and the difference in the fluctuation characteristic value of each main component between different bands in the near-infrared spectrum are analyzed to determine the inter-segment difference between each band in the near-infrared spectrum corresponding to each main component. The beneficial effect is that the absorption peak characteristics of the near-infrared spectrum of cement raw materials are used to distinguish the spectra of different bands, so as to facilitate the subsequent filtering of bands containing different spectral information to different degrees and enhance the subsequent noise filtering ability; multiple clustering clusters of each main component are obtained. ; Determine the adjustment coefficient of each cluster of each main component; determine the filter window size and filter order of each cluster of each main component, which has the beneficial effect of improving the clustering effect of each band in the near-infrared spectrum, avoiding clustering the bands containing noise into the clusters containing more spectral feature information, and enhancing the ability to remove abnormal band spectra; filter all bands in the near-infrared spectrum to obtain the characteristic enhanced spectrum of each main component, and control the content of each main component in the cement raw material. The beneficial effect is that while improving the noise removal ability, it takes into account the enhancement of the protection of the characteristic spectral information of each main component in the near-infrared spectrum of the cement raw material, and improves the accuracy of near-infrared spectroscopy detection of the main component content of cement raw material, thereby ensuring the stability of the production quality of cement raw material. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The cement production quality control method based on AI technology of the present application is further described in detail below in conjunction with the accompanying drawings.

[0040] Figure 1 A flowchart of the cement production quality control method based on AI technology provided in an embodiment of the present application;

[0041] Figure 2 A schematic diagram of the near infrared spectrum of cement raw material provided in the embodiment of the present application;

[0042] Figure 3 A flowchart of the steps of a method for obtaining the spectral fluctuation degree of each band in a near-infrared spectrum provided in an embodiment of the present application;

[0043] Figure 4 A flowchart of the steps of a method for obtaining the fluctuation characteristic value of each main component corresponding to each band in the near-infrared spectrum provided in an embodiment of the present application;

[0044] Figure 5 The near-infrared spectrum of silicon dioxide provided by the original SG filtering algorithm in the embodiment of the present application;

[0045] Figure 6 Near-infrared spectrum of silicon dioxide with improved SG filtering algorithm provided in an embodiment of the present application. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of this application more clear, the cement production quality control method based on AI technology proposed in this application is further described in detail below in conjunction with the accompanying drawings and implementation examples. It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0047] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0048] See also Figure 1 , which shows a flowchart of a cement production quality control method based on AI technology provided by an embodiment of the present application, the method comprising the following steps:

[0049] Step 1, obtaining the near infrared spectrum of cement raw meal, and respectively obtaining the standard near infrared spectrum of each main component in cement raw meal.

[0050] A near-infrared spectral sensor is installed above the cement raw meal chute to collect and obtain the near-infrared spectrum of the cement raw meal in the chute. Secondly, since the main components of cement raw meal are silica, alumina, iron oxide and calcium oxide, pure silica, alumina, iron oxide and calcium oxide standard samples are collected, and the near-infrared spectral sensor is used to measure the standard near-infrared spectra of silica, alumina, iron oxide and calcium oxide, respectively, and obtain the standard near-infrared spectrum of each main component in cement raw meal.

[0051] Further, the near infrared spectrum diagram of cement raw material in the embodiment of the present application is as follows Figure 2 As shown, the horizontal axis in the near-infrared spectrum is the wave number, which is expressed as the inverse of the wavelength, and the vertical axis in the near-infrared spectrum is the absorbance, which reflects the ability of cement raw material to absorb infrared radiation. The near-infrared spectrum determines the molecular structure and the existence of chemical bonds by measuring the absorption of substances in cement raw material at different wave numbers.

[0052] At this point, the near infrared spectrum of the cement raw meal is obtained, and the standard near infrared spectrum of each main component in the cement raw meal is obtained respectively.

[0053] Step 2, dividing the wavenumber range in the near infrared spectrum and the standard near infrared spectrum of each main component into multiple bands respectively; determining the spectral fluctuation of each band in the near infrared spectrum according to the change trend of the absorbance in each band in the near infrared spectrum and the correlation between the absorbance and the wavenumber.

[0054] The near-infrared spectrum of cement raw meal contains the material composition information of the four main components in cement raw meal. Since the detected cement raw meal is a complex system, the diversity of material composition and the inhomogeneity of raw meal particles will cause the near-infrared spectrum collected by the near-infrared spectral sensor to contain more noise and interference information in addition to the spectral information of the main components.

[0055] In the near-infrared spectrum, the peak value of absorbance represents the absorption peak. Therefore, considering that in cement raw material, the spectral characteristics such as the absorption peaks of different main components are distributed in different wavenumber ranges of the near-infrared spectrum, in order to remove the interference information in the near-infrared spectrum of cement raw material and retain the spectral characteristics of the main components in cement raw material, the near-infrared spectrum of cement raw material is denoised and enhanced.

[0056] Based on the above analysis, the wavenumber in the near-infrared spectrum is divided to analyze the fluctuation of absorbance in the near-infrared spectrum and determine the spectral fluctuation to reflect the spectral information contained in each band, specifically:

[0057] The wavenumber range in the near infrared spectrum and the standard near infrared spectrum of each main component is divided into a plurality of bands respectively;

[0058] Preferably, in this embodiment, the wave number range in the near infrared spectrum and the standard near infrared spectrum of each main component is divided into 50 bands respectively. As other implementation modes, the implementers can set them according to the actual situation.

[0059] It should be noted that the wave number range of the near infrared spectrum and the standard near infrared spectrum of each main component is the same, which is [4000, 10000]. Therefore, the wave number range of each band divided by the near infrared spectrum and the standard near infrared spectrum of each main component is the same.

[0060] The wave numbers in each band of the near-infrared spectrum and their corresponding absorbances form a two-dimensional array; linear fitting is performed on all two-dimensional arrays in each band to obtain the slope of the fitting line;

[0061] Preferably, in this embodiment, the least squares method is used for linear fitting, wherein the least squares method is a well-known technology and will not be described in detail here. As other implementation methods, the implementer may adopt other methods of the prior art, such as the maximum likelihood estimation method, etc. This embodiment does not impose any special restrictions on this.

[0062] Calculate the absolute value of the correlation between all absorbances and all wave numbers in each band of the near-infrared spectrum, and record it as the correlation coefficient of each band in the near-infrared spectrum;

[0063] Preferably, in this embodiment, the Pearson correlation coefficient of all absorbances and wavenumbers in each band in the near-infrared spectrum is calculated, wherein the Pearson correlation coefficient is a well-known technology and will not be described in detail here. As other implementation methods, the implementer can adopt other methods of the prior art, such as the Spearman correlation coefficient, etc. This embodiment does not impose any special restrictions on this.

[0064] Calculate the mean of all absorbances in each band of the near-infrared spectrum;

[0065] Calculating the ratio of the square result of the slope of each band in the near-infrared spectrum to the correlation coefficient, and taking the product of the ratio and the mean as the spectral fluctuation of each band in the near-infrared spectrum;

[0066] In this embodiment, the calculation formula for the spectral fluctuation of each band in the near-infrared spectrum is: ,in, In near infrared spectroscopy The spectral fluctuation of each band, In near infrared spectroscopy The mean value of all absorbances in the band, In near infrared spectroscopy The slope of the band, In near infrared spectroscopy The correlation coefficient of the bands, To preset a value greater than 0 to avoid the denominator being 0, in this embodiment, The value is 0.001.

[0067] It should be noted that near the absorption peak in the near-infrared spectrum of cement raw materials, the absorbance shows a more obvious upward or downward trend with the increase of wave number. The larger the slope of the fitting straight line in the corresponding band, and the closer each band is to the absorption peak of the spectrum, the larger the mean value of the absorbance in the corresponding band, the greater the fluctuation of the spectrum curve, and the greater the fluctuation of the obtained spectrum. Secondly, at the absorption peak of the spectrum, the fluctuation of the spectrum curve is large, and the distribution of the absorbance of the corresponding band and its fitting straight line is more discrete, then the absolute value of the correlation coefficient is small, and the spectral fluctuation of the corresponding band is large, indicating that the corresponding band contains more spectral information.

[0068] Further, the step flow chart of the method for obtaining the spectral fluctuation degree of each band in the near-infrared spectrum provided in the embodiment of the present application is as follows: Figure 3 shown.

[0069] At this point, the spectral fluctuation of each band in the near-infrared spectrum is obtained.

[0070] Step 3, based on the similarity of the local changes of the spectral fluctuations in the same band of the near-infrared spectrum and the standard near-infrared spectrum of different main components, combined with the spectral fluctuations, determine the fluctuation characteristic value of each main component corresponding to each band in the near-infrared spectrum.

[0071] Furthermore, the distribution trend of the fluctuation degree of different bands in the near-infrared spectrum can reflect the richness of the spectral feature information contained in the spectral curve. Therefore, the similarity between the fluctuation trend of the bands in the near-infrared spectrum and the corresponding bands in the standard near-infrared spectra of different components is analyzed to construct the fluctuation characteristic value, which is specifically:

[0072] A local window of preset size is set centered on each band in the near-infrared spectrum;

[0073] Preferably, in this embodiment, a local window with a size of 5 band lengths is set. As other implementation methods, the implementer can set it according to the actual situation.

[0074] The normalized results of the spectral fluctuations of all bands in the local window are used to form a local fluctuation vector of each band in the near-infrared spectrum;

[0075] It should be noted that, for the bands whose window size length in the local window of each band is less than 5, the mean filling method is used to obtain the local fluctuation vector.

[0076] For the standard near-infrared spectrum of each main component, the same method as the local fluctuation vector of each band in the near-infrared spectrum is adopted to obtain the local fluctuation vector of each band in the standard near-infrared spectrum of each main component.

[0077] Calculate the similarity of the local fluctuation vectors in the same band between the near-infrared spectrum and the standard near-infrared spectrum of each main component as the fluctuation similarity of each band in the near-infrared spectrum corresponding to each main component;

[0078] It should be noted that it is assumed that the similarity of the local fluctuation vectors in the same band between the near-infrared spectrum and the standard near-infrared spectrum of silicon dioxide is calculated as the fluctuation similarity of the silicon dioxide component corresponding to each band in the near-infrared spectrum.

[0079] Preferably, in this embodiment, the cosine similarity of the local fluctuation vector in the same band between the near-infrared spectrum and the standard near-infrared spectrum of each main component is calculated as the fluctuation similarity of each band in the near-infrared spectrum corresponding to each main component, wherein the cosine similarity is a well-known technology and will not be described in detail here. As other implementation methods, the implementer may adopt other methods of the prior art, such as the Pearson correlation coefficient, the Jaccard similarity coefficient, etc., and this embodiment does not impose any special restrictions on this.

[0080] Calculate the sum of the fluctuation similarity of each main component in each band of the near-infrared spectrum and a preset initial threshold, and use the product of the sum and the spectral fluctuation as the fluctuation characteristic value of each main component in each band of the near-infrared spectrum;

[0081] Preferably, in this embodiment, the preset initial threshold value is 2, so that the calculated fluctuation characteristic value is a positive number. As other implementation methods, the implementer can set it according to the actual situation.

[0082] In this embodiment, the calculation formula for the fluctuation characteristic value of each main component corresponding to each band in the near-infrared spectrum is: ,in, In near infrared spectroscopy The band corresponds to The fluctuation eigenvalues ​​of the main components, In near infrared spectroscopy The spectral fluctuation of each band, To preset the initial threshold, in this embodiment The value is 2. In near infrared spectroscopy The local wave vector of the band, For the The standard infrared spectrum of the main components The local wave vector of the band, Indicates calculating similarity.

[0083] It should be noted that the greater the spectral fluctuation of a band in the near-infrared spectrum, the greater its fluctuation degree, the closer the band is to the absorption peak on the spectral curve, the richer the spectral information it contains, and the larger the fluctuation characteristic value of the corresponding band in the near-infrared spectrum of the cement raw meal; secondly, the closer the fluctuation degree of the near-infrared spectrum of the cement raw meal and the standard near-infrared spectrum corresponding to each main component in the same band are, the more spectral information of the corresponding main components is contained in the near-infrared spectrum of the cement raw meal, the greater the role of the subsequent prediction of the content of each main component in the cement raw meal, the greater the obtained fluctuation similarity, and the larger the fluctuation characteristic value of the corresponding band in the near-infrared spectrum of the cement raw meal.

[0084] Further, the step flow chart of the method for obtaining the fluctuation characteristic value of each main component corresponding to each band in the near infrared spectrum provided in the embodiment of the present application is as follows: Figure 4 shown.

[0085] At this point, the fluctuation characteristic value of each main component corresponding to each band in the near-infrared spectrum is obtained.

[0086] Step 4, analyzing the distance relationship between different bands in the near-infrared spectrum and the difference in the fluctuation characteristic value of each main component between different bands in the near-infrared spectrum, and determining the inter-segment difference between each band in the near-infrared spectrum and each main component.

[0087] Furthermore, the absorption peak in the near-infrared spectrum of cement raw materials usually has a certain width. The distance between the band and the absorption peak indicates the richness of the spectral characteristic information it contains, reflecting the role of the band in predicting the content of the main components of cement raw materials. Therefore, the distance relationship between each band and the band with the absorption peak is analyzed to construct the inter-segment fluctuation difference, which is:

[0088] A threshold segmentation algorithm is used to obtain the segmentation thresholds of the fluctuation characteristic values ​​of each main component in all bands of the near-infrared spectrum, and the bands whose fluctuation characteristic values ​​of each main component are greater than the segmentation threshold are recorded as the characteristic bands of each main component;

[0089] Preferably, in this embodiment, the maximum inter-class variance algorithm is used for threshold segmentation, wherein the maximum inter-class variance algorithm is a well-known technology and will not be described in detail here.

[0090] Each band in the near-infrared spectrum is numbered in ascending order of wave number, so each band in the near-infrared spectrum corresponds to a serial number;

[0091] The calculation formula for the inter-segment difference of each main component in each band of the near-infrared spectrum is: ,in, In near infrared spectroscopy The band corresponds to The inter-segment differences of the main components, In near infrared spectroscopy The serial number of the band, is the near-infrared spectrum corresponding to The main ingredient The serial number of the characteristic band, is the near-infrared spectrum corresponding to The main ingredient The fluctuation characteristic value of the characteristic band, In near infrared spectroscopy The band corresponds to The fluctuation eigenvalues ​​of the main components, To preset a value greater than 0 to avoid the denominator being 0, in this embodiment The value is 0.01. is the near-infrared spectrum corresponding to The number of all characteristic bands of the principal component.

[0092] It should be noted that the closer the distance between each band and the characteristic band in the near-infrared spectrum of cement raw material is, the greater the characteristic information contained in its spectral band is. In addition, when the absorption peak in the near-infrared spectrum is wide, the bands farther away from the absorption peak also contain more spectral information, and the difference in the fluctuation characteristic value with the characteristic band is small, then the corresponding bands have a larger fluctuation difference between the segments.

[0093] At this point, the inter-segment difference between each band and each main component in the near-infrared spectrum is obtained.

[0094] Step 5, based on the fluctuation characteristic value and the inter-segment difference of each main component corresponding to each band in the near-infrared spectrum, determine the iterative formula of the membership matrix, cluster all bands in the near-infrared spectrum in combination with the clustering algorithm, and obtain multiple clusters for each main component.

[0095] Furthermore, different bands in the near-infrared spectrum of cement raw materials contain different amounts of spectral feature information, and the role of predicting the content of each main component in cement raw materials will also be different. In order to retain the spectral information of each main component in the near-infrared spectrum of cement raw materials in the process of removing spectral interference information, cluster analysis is performed on different bands in the near-infrared spectrum, specifically:

[0096] The fluctuation characteristic value and inter-segment difference of each main component in each band of the near-infrared spectrum are used to form the characteristic vector of each main component in each band of the near-infrared spectrum;

[0097] Calculate the quartile of the modulus of the eigenvector of each main component corresponding to all bands in the near-infrared spectrum, and use the eigenvector with a modulus value equal to the quartile as the initial clustering center of each main component;

[0098] Based on the initial cluster center of each main component, an iterative formula of the membership matrix of each main component is determined, and a fuzzy C-means clustering algorithm is used for iteration to cluster the characteristic vectors of each main component corresponding to all bands in the near-infrared spectrum to obtain multiple clusters of each main component;

[0099] It should be noted that the fuzzy C-means clustering algorithm is a well-known technology and will not be described in detail here.

[0100] Among them, the iterative formula of the membership matrix of the fuzzy C-means clustering algorithm is: ,in, In the near infrared spectrum The band belongs to The main ingredient The membership degree of each cluster center to the corresponding category, In the near infrared spectrum The serial number of the band, For the The corresponding serial number of each cluster center in the near infrared spectrum, In the near infrared spectrum The band corresponds to The eigenvectors of the principal components, For the The main ingredient Cluster centers, For the The number of all cluster centers of the principal components, is an exponential function with a natural constant as base, is the distance calculation function, To preset a value greater than 0 to avoid the denominator being 0, in this embodiment The value is 0.01. is the normalization function.

[0101] Preferably, in this embodiment, is a Euclidean distance calculation function. As other implementations, the implementer may adopt other methods in the prior art, such as DTW distance, Manhattan distance, etc. This embodiment does not impose any special limitation on this.

[0102] It should be noted that after normalization, .

[0103] It should be noted that the smaller the distance between the characteristic vector of each band and the cluster center in the near-infrared spectrum, the greater the similarity between the band and the band corresponding to the cluster center, and the greater the probability that the band belongs to the category to which the cluster center belongs, that is, the closer it belongs to the cluster corresponding to the cluster center, the greater the membership degree obtained; however, if there is abnormal spectral data caused by noise in a certain band, the fluctuation characteristic value of a certain band will be larger, and the Euclidean distance between the characteristic vector of a certain band and the characteristic vector of the band with absorption peak will be smaller, which may erroneously determine that a certain band is more similar to the band with absorption peak; therefore, in order to enhance the ability to remove abnormal spectral data, the bands in the near-infrared spectrum with a large difference in serial number from the cluster center are assigned a smaller membership degree to avoid clustering the noisy bands into clusters with more spectral feature information.

[0104] At this point, multiple clusters are obtained for each main component.

[0105] Step 6, based on the fluctuation characteristic values ​​of all bands in each cluster and the inter-segment difference, determine the adjustment coefficient of each cluster of each main component; determine the filter window size and filter order of each cluster of each main component based on the adjustment coefficient; filter all bands in the near-infrared spectrum separately in combination with the filtering algorithm to obtain the characteristic enhanced spectrum of each main component, and control the content of each main component in the cement raw material.

[0106] Furthermore, based on the clustering results of the feature vectors of each band, the filter window size and filter order of each cluster cluster are determined, specifically:

[0107] The normalized result of the mean value of the modulus of the eigenvectors of all bands corresponding to each main component in each cluster is used as the adjustment coefficient of each cluster of each main component;

[0108] Preferably, in this embodiment, a maximum-minimum normalization method is used for normalization processing, wherein the maximum-minimum normalization method is a well-known technology and will not be described in detail here.

[0109] The filter window size for each cluster of each principal component is: ,in, For the The main ingredient The filter window size for clusters, To preset the minimum window size, For the The maximum value of the adjustment coefficient of all clusters of the principal components, For the The main ingredient The adjustment coefficient of the clusters, For the The minimum value of the adjustment coefficient of all clusters of the principal components.

[0110] The filter order for each cluster of each principal component is: ,in, For the The main ingredient The filter order of clusters, To preset the minimum filter order, For the The main ingredient The adjustment coefficient of the clusters, For the The maximum value of the adjustment coefficients of all clusters of the principal components.

[0111] Preferably, in this embodiment, the minimum window size is preset The value is 25, which presets the minimum filter order. The value is 2. As other implementation methods, the implementer can set it according to the actual situation.

[0112] It should be noted that when the modulus of the characteristic vector of each band in the cluster is larger, it means that the amount of spectral feature information contained is larger, and the adjustment coefficient of the corresponding cluster is larger. At this time, a smaller filter window size is set to better retain the effective information in the near-infrared spectrum of the cement raw material. At the same time, for clusters with less spectral information, the adjustment coefficient is smaller. At this time, a smaller filter order is set to enhance the ability to remove spectral noise and avoid overfitting of the spectrum line.

[0113] Furthermore, based on the filter window size and filter order of each cluster of each main component, each band in the near-infrared spectrum is filtered as follows:

[0114] Based on the filter window size and filter order of the cluster to which each band in the near-infrared spectrum of each main component belongs, the SG (Savitzky-Golay) smoothing algorithm is used to filter each band in the near-infrared spectrum respectively, and the near-infrared spectrum of each main component after filtering is recorded as the characteristic enhanced spectrum of each main component;

[0115] Further, the filtering result diagram of the original SG filtering algorithm provided in the embodiment of the present application is as follows: Figure 5 As shown, the schematic diagram of the filtering result of the improved SG filtering algorithm is as follows Figure 6 As shown, Figure 5 The original SG filtering algorithm for the near-infrared spectrum of silicon dioxide, Figure 6 is the near-infrared spectrum of silica after filtering, i.e., the characteristic enhanced spectrum of silica, compared to Figure 5 , Figure 6 While filtering the near-infrared spectrum, more spectral information is retained.

[0116] Furthermore, based on the characteristic enhanced spectrum of each main component, the contents of the four main components in the cement raw meal are predicted to control the quality of the cement raw meal, specifically:

[0117] Collect N historical cement raw material samples as training sets, use X-ray fluorescence spectrometry to determine the content of silicon dioxide, aluminum oxide, iron oxide and calcium oxide in each historical cement raw material sample as training labels. It should be noted that X-ray fluorescence spectrometry is a well-known technology and will not be described in detail here. Obtain a near-infrared spectrum of each historical cement raw material, and use the above-mentioned filtering method to obtain a characteristic enhanced spectrum of each main component in each historical cement raw material, that is, a characteristic enhanced spectrum of silicon dioxide, aluminum oxide, iron oxide and calcium oxide in each historical cement raw material. The characteristic enhanced spectrum of each main component in all historical cement raw materials and the training labels of the corresponding main components are respectively input into a BP neural network (back propagation neural network) for training, and a trained BP neural network model corresponding to each main component is obtained;

[0118] Preferably, in this embodiment, 100 cement raw material samples are collected as training sets. BP neural network is a well-known technology and will not be described in detail here. The cross entropy loss function is used as the loss function of the BP neural network, and the Adam optimizer is selected as the optimization algorithm.

[0119] The characteristic enhanced spectrum of each main component in the cement raw material is input into the corresponding trained BP neural network model to obtain the predicted content of each main component in the cement raw material;

[0120] The predicted content of each main component in the cement raw material is input into the PLC (Programmable Logic Controller) system, and the PLC system is used to adjust the proportion of cement raw materials and the feeding amount, so as to ensure the stability of the cement raw material production quality and control the cement raw material production quality.

[0121] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0122] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0123] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the present application. It should be pointed out that, for those of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application. Therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present application without departing from the content of the technical solution of the present application, shall fall within the protection scope of the technical solution of the present application.

Claims

1. A cement production quality control method based on AI technology, characterized in that: The method comprises the following steps: Obtaining the near infrared spectrum of cement raw meal and obtaining the standard near infrared spectrum of each main component in cement raw meal; The wavenumber ranges in the near infrared spectrum and the standard near infrared spectrum of each main component are divided into multiple bands respectively; the spectral fluctuation degree of each band in the near infrared spectrum is determined according to the variation trend of the absorbance in each band in the near infrared spectrum and the correlation degree between the absorbance and the wavenumber; Determine the fluctuation characteristic value of each main component in each band of the near infrared spectrum according to the similarity of the local changes of the spectral fluctuations in the same band of the near infrared spectrum and the standard near infrared spectrum of different main components, combined with the spectral fluctuations; Analyze the distance relationship between different bands in the near-infrared spectrum, as well as the difference in the fluctuation characteristic value of each main component between different bands in the near-infrared spectrum, and determine the inter-segment difference between each band in the near-infrared spectrum and each main component; Based on the fluctuation characteristic value and the inter-segment difference of each main component in each band of the near-infrared spectrum, an iterative formula of the membership matrix is ​​determined, and all bands in the near-infrared spectrum are clustered in combination with a clustering algorithm to obtain multiple clusters of each main component; Determine the adjustment coefficient of each cluster of each main component based on the fluctuation characteristic values ​​of all bands in each cluster and the inter-segment difference; determine the filter window size and filter order of each cluster of each main component based on the adjustment coefficient; Combined with the filtering algorithm, all bands in the near-infrared spectrum are filtered respectively to obtain the characteristic enhanced spectrum of each main component, and the content of each main component in the cement raw material is controlled; Determining the spectral fluctuation of each band in the near-infrared spectrum includes: The wave numbers in each band of the near-infrared spectrum and their corresponding absorbances form a two-dimensional array; linear fitting is performed on all two-dimensional arrays in each band to obtain the slope of the fitting line; Calculate the absolute value of the correlation between all absorbances and all wave numbers in each band of the near-infrared spectrum as the correlation coefficient of each band in the near-infrared spectrum; Calculate the mean of all absorbances in each band of the near-infrared spectrum; Calculating the ratio of the square result of the slope of each band in the near-infrared spectrum to the correlation coefficient, and taking the product of the ratio and the mean as the spectral fluctuation of each band in the near-infrared spectrum; The step of determining the fluctuation characteristic value of each main component corresponding to each band in the near infrared spectrum includes: A local window of a preset size is set with each band in the near infrared spectrum and the standard near infrared spectrum of each main component as the center; The normalized results of the spectral fluctuations of all bands in the local window are used to form the local fluctuation vectors of each band in the near-infrared spectrum and the standard near-infrared spectrum of each main component; The similarity of the local fluctuation vectors in the same band between the near-infrared spectrum and the standard near-infrared spectrum of each main component is calculated respectively as the fluctuation similarity of each band in the near-infrared spectrum corresponding to each main component; The sum of the fluctuation similarity and a preset initial threshold is calculated, and the product of the sum and the spectral fluctuation is used as the fluctuation characteristic value of each main component corresponding to each band in the near-infrared spectrum.

2. The cement production quality control method based on AI technology as claimed in claim 1, characterized in that: The step of determining the inter-segment difference between each band in the near infrared spectrum and each main component thereof includes: Obtaining the segmentation thresholds of the fluctuation characteristic values ​​of each main component in all bands of the near-infrared spectrum respectively, recording the bands of each main component whose fluctuation characteristic values ​​are greater than the segmentation thresholds as the characteristic bands of each main component; numbering each band in the near-infrared spectrum in ascending order of wave number; The calculation formula for the inter-segment difference of each main component in each band of the near-infrared spectrum is: Among them, P x,i is the inter-segment difference between the i-th band and the x-th main component in the near-infrared spectrum, D x,i is the serial number of the i-th band in the near-infrared spectrum, D x,j is the serial number of the jth characteristic band corresponding to the xth main component in the near-infrared spectrum, W x,j is the fluctuation characteristic value of the jth characteristic band corresponding to the xth main component in the near-infrared spectrum, W x,i is the fluctuation characteristic value of the xth main component corresponding to the ith band in the near-infrared spectrum, 𝜏 is a value preset to be greater than 0, T x is the number of all characteristic bands corresponding to the xth main component in the near-infrared spectrum.

3. The cement production quality control method based on AI technology as claimed in claim 2, characterized in that: The iterative formula for determining the membership matrix includes: The fluctuation characteristic value and inter-segment difference of each main component in each band of the near-infrared spectrum are used to form the characteristic vector of each main component in each band of the near-infrared spectrum; Calculate the quartile of the modulus of the eigenvector of each main component corresponding to all bands in the near-infrared spectrum, and use the eigenvector with a modulus value equal to the quartile as the initial cluster center of each main component; Based on the initial cluster centers, an iterative formula for the membership matrix of each main component is determined.

4. The cement production quality control method based on AI technology as claimed in claim 3, characterized in that: The iterative formula of the membership matrix is ​​specifically: in, is the degree of membership of the category corresponding to the kth cluster center of the xth main component in the i-th band of the near-infrared spectrum, q i is the serial number of the i-th band in the near-infrared spectrum, q k is the serial number corresponding to the kth cluster center in the near-infrared spectrum, T x,i is the eigenvector of the xth main component corresponding to the ith band in the near-infrared spectrum, T x,k is the kth cluster center of the xth principal component, K x is the number of all cluster centers of the x-th main component, exp() is an exponential function with a natural constant as the base, d() is a distance calculation function, ε is a value preset to be greater than 0, and norm[ ] is a normalization function.

5. The cement production quality control method based on AI technology as claimed in claim 3, characterized in that: The adjustment coefficient of each cluster of each main component is a normalized result of the mean value of the modulus of the eigenvectors of all bands in each cluster corresponding to each main component.

6. The cement production quality control method based on AI technology as claimed in claim 1, characterized in that: The filter window size of each cluster of each main component is determined by the calculation formula: Among them, L x,k is the filter window size of the kth cluster of the xth main component, L0 is the preset minimum window size, h x,max is the maximum value of the adjustment coefficient of all clusters of the xth principal component, h x,k is the adjustment coefficient of the kth cluster of the xth principal component, h x,min is the minimum value of the adjustment coefficient of all clusters of the x-th principal component.

7. The cement production quality control method based on AI technology as claimed in claim 1, characterized in that: The calculation formula of the filtering order is: Among them, α x,k is the filter order of the kth cluster of the xth main component, α0 is the preset minimum filter order, and h x,k is the adjustment coefficient of the kth cluster of the xth principal component, h x,max is the maximum value of the adjustment coefficients of all clusters of the xth principal component.

8. The cement production quality control method based on AI technology as claimed in claim 1, characterized in that: The control of the content of each main component in the cement raw material includes: The characteristic enhanced spectrum of each main component in the cement raw material is input into the neural network model to obtain the predicted content of each main component in the cement raw material; The predicted content of each main component in the cement raw material is input into the PLC system to control and adjust the proportion and feed amount of the cement raw material.

Images