Coal coke residue sample classification detection method and system

Abstract

The invention belongs to the technical field of spectral analysis, and particularly relates to a coal coke residue sample classification detection method and system, and the method comprises the steps: obtaining the noise reduction spectrum of each coke residue sample; according to the difference between the spectrums of the coke residue samples and the corresponding noise reduction spectrums and the fluctuation frequency of the spectrums of the coke residue samples, noise indexes of all wavelengths of the spectrums of all the coke residue samples are obtained; according to the law of the difference of the noise indexes of each wavelength in different coke slag samples of the same kind of coke slag, obtaining the structural performance of each wavelength; according to the absorbance difference of the same wavelength of different coke residues and the absorbance difference of the same wavelength of coke residue samples of the same type of coke residues with different crushing degrees, the component expression of each wavelength is obtained; and in combination with the structural expression and component expression of the wavelength, obtaining a characteristic wavelength, and establishing a classification detection model of the coke residue sample. According to the method, the classification detection model of the coke residue samples is constructed, and the accuracy and the detection efficiency of classification detection of the coke residue samples are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of spectral analysis. More specifically, the present invention relates to a method and system for classifying and detecting coal cinder samples. Background Art

[0002] Coal is a complex mixture, mainly containing moisture, ash, fixed carbon, etc. Coals with different compositions have different uses in the industrial field. For example, anthracite is used to manufacture gas or directly as fuel. Different types of coal will produce different types of cinders after combustion. The types of cinders are divided into eight types, such as powdery, weakly caking, non-swelling fused caking, strongly colliding fused caking, etc. In addition, the combustion performance of coal also affects the composition and morphology of cinders. Therefore, by analyzing the composition and category of cinders, the combustion performance, coal quality and type of coal can be further understood.

[0003] In order to classify and detect coal cinders, near-infrared spectroscopy is currently often used to analyze the composition of cinders. Group characteristic information in cinders, such as C-H, O-H, S-H, etc., will absorb near-infrared light, and then produce characteristic peaks in the spectrum to achieve the purpose of composition detection. Then, a neural network is used to train the composition of different types of cinders to complete cinder classification. In related technologies, for example, the Chinese patent document with the authorization announcement number CN106990066B discloses a method and device for identifying coal types, which discloses obtaining the type information of coal based on visible near-infrared hyperspectral data, using a multi-layer perceptron classification model to improve the classification accuracy, and realizing non-destructive type identification of coal samples.

[0004] However, there are certain differences between coal samples and cinder samples. Cinders have powdery types and massive types containing bubbles. These structures will affect the reflection of near-infrared light, causing the attenuation of absorbed light, increasing spectral noise, and affecting composition detection and cinder classification. If all samples are crushed, the morphological characteristics of cinders will be lost and complex pre-processing will be required for cinder classification, reducing the efficiency of classification and detection. Summary of the Invention

[0005] To solve the above technical problems of insufficient accuracy and low efficiency in classifying and detecting cinder samples, the present invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for classifying and detecting coal cinder samples, including: Obtain several kinds of cinder, crush various cinders and take samples to obtain cinder samples with several degrees of crushing of various cinders. Use a near-infrared spectrometer to obtain the spectra of each cinder sample, where the spectra are sequences of absorbances corresponding to several wavelengths; based on a filtering algorithm, obtain the noise-reduced spectra of each cinder sample; according to the differences between the spectra of the cinder samples and the corresponding noise-reduced spectra, and the fluctuation frequencies of the spectra of the cinder samples, obtain the noise indices of each wavelength of the spectra of each cinder sample; according to the laws of the differences in the noise indices of each wavelength among different cinder samples of the same kind of cinder, obtain the structural manifestations of each wavelength; according to the differences in absorbances of the same wavelength of different cinders and the differences in absorbances of the same wavelength of cinder samples with different degrees of crushing of the same kind of cinder, obtain the compositional manifestations of each wavelength; combine the structural manifestations and compositional manifestations of the wavelengths to obtain characteristic wavelengths; based on the characteristic wavelengths and a neural network model, establish a classification detection model for cinder samples.

[0007] In the present invention, by analyzing the spectra of cinder samples with several degrees of crushing, the influence of the degree of crushing on the spectra of cinder samples is obtained, and then the structural manifestations of the spectral wavelengths are obtained. Using spectral noise as a feature for the classification detection of cinder samples. The classification detection model for cinder samples established in the present invention enables the coal cinder samples to be directly subjected to automated classification detection without complex pre-processing, improving the efficiency of the classification detection of cinder samples.

[0008] Preferably, the obtaining of the noise indices of each wavelength of the spectra of each cinder sample includes: Denote the spectrum of any cinder sample as the target spectrum, and denote any wavelength of the target spectrum as the target wavelength; obtain the neighborhood of the target wavelength; obtain the error sequence of the target spectrum and the error sign difference sequence of the target spectrum; ; wherein, represents the noise index of the target wavelength; represents the number of wavelengths in the neighborhood of the target wavelength; represents the i-th value at the corresponding position in the neighborhood of the target wavelength in the error sequence of the target spectrum; represents the c-th value at the corresponding position in the neighborhood of the target wavelength in the error sign difference sequence of the target spectrum; represents the absolute value function.

[0009] In the present invention, by obtaining the noise indices of each wavelength of the spectra of cinder samples, the noise generated by cinder samples due to factors such as porous structures can be quantified, improving the accuracy of the classification detection of cinder samples.

[0010] Preferably, the obtaining of the error sequence of the target spectrum and the error sign difference sequence of the target spectrum includes: Subtract the absorbance sequence of the target spectrum from the absorbance sequence of the corresponding denoised spectrum to obtain the error sequence of the target spectrum; use the sign function to symbolize the error sequence of the target spectrum to obtain the error sign sequence of the target spectrum; perform forward difference on the error sign sequence of the target spectrum to obtain the error sign difference sequence of the target spectrum.

[0011] Preferably, the obtaining of the structural performance of each wavelength includes: taking any coke breeze as the target coke breeze, obtaining the noise index of the a-th wavelength of the spectra of all coke breeze samples of the target coke breeze, using the least squares method for linear fitting to obtain the noise index fitting line of the a-th wavelength of the target coke breeze; based on the fitting effect of the noise index fitting line of the a-th wavelength of the target coke breeze and the distribution law of the noise index of the a-th wavelength of the spectra of all coke breeze samples of the target coke breeze according to the pulverization degree, obtaining the structural characteristics of the a-th wavelength relative to the target coke breeze; taking the mean value of the structural characteristics of the a-th wavelength relative to all types of coke breezes as the structural performance of the a-th wavelength.

[0012] By fitting the noise indices of the same wavelength of all coke breeze samples of the same type of coke breeze in the present invention, the regular changes of the same wavelength of the same type of coke breeze with the change of the pulverization degree of the coke breeze samples can be obtained, so as to obtain the structural performance of the wavelength, making the characteristic dimensions of the classification detection of coke breeze samples more, and improving the accuracy of the classification detection of coke breeze samples.

[0013] Preferably, the structural characteristics of the a-th wavelength relative to the target coke breeze satisfy the expression: ; In the formula, represents the structural characteristics of the a-th wavelength relative to the target coke breeze; represents the number of coke breeze samples of the target coke breeze; represents the distance between the h-th coke breeze sample of the target coke breeze and the noise index fitting line of the a-th wavelength of the target coke breeze; represents the type of pulverization degree; , , represent the average noise indices of the a-th wavelength of the spectra of the coke breeze samples of the (q + 1)-th, q-th, and (q - 1)-th pulverization degrees of the target coke breeze; represents the exponential function with the natural constant as the base; represents the normalization function.

[0014] Preferably, the obtaining of the compositional performance of each wavelength includes: Obtaining the absorbance differences of each wavelength relative to all coke breezes; obtaining the regularity of the absorbance change of each wavelength relative to all coke breezes; The compositional performance of any wavelength satisfies the expression: ; In the formula, represents the component performance of the a-th wavelength; represents the absorbance difference of the a-th wavelength relative to all cokes; represents the regularity of the absorbance change of the a-th wavelength relative to all cokes; represents the number of coke types; represents the variance of the absorbance of the a-th wavelength of all coke samples of the u-th coke; represents the exponential function with the natural constant as the base.

[0015] The present invention obtains the component performance of the wavelength by the difference in absorbance of the same wavelength of different types of cokes, the change law of the absorbance of the same wavelength of different types of cokes, and the discrete situation of the absorbance of the same wavelength of coke samples of the same type of coke, and obtains the wavelength that can characterize the difference in component content of different types of cokes, ensuring the accuracy of the classification detection of coke samples.

[0016] Preferably, the absorbance difference of each wavelength relative to all cokes satisfies the expression: ; In the formula, represents the absorbance difference of the a-th wavelength relative to all cokes; represents the number of coke types; , represent the average absorbance of the a-th wavelength of the u-th and v-th cokes; represents the absolute value function; represents the normalization function.

[0017] Preferably, the obtaining of the regularity of the absorbance change of each wavelength relative to all cokes includes: Obtain the average absorbance sequence of the a-th wavelength of all types of cokes in ascending order of the type number, perform forward difference on the average absorbance sequence of the a-th wavelength of all types of cokes, and obtain the difference sequence of the average absorbance sequence of the a-th wavelength of all types of cokes; Compare the number of values greater than 0 with the number of values less than 0 in the difference sequence of the average absorbance sequence of the a-th wavelength of all types of cokes, and record it as the regularity of the absorbance change of the a-th wavelength relative to all cokes.

[0018] Preferably, the obtaining of the characteristic wavelength includes: comparing the structural performance and the compositional performance of the a-th wavelength as the first ratio of the a-th wavelength; comparing the compositional performance and the structural performance of the a-th wavelength as the second ratio of the a-th wavelength; taking the maximum value of the first ratio and the second ratio of the a-th wavelength as the characteristic goodness of the a-th wavelength; and obtaining Z wavelengths with the maximum characteristic goodness among all wavelengths as the characteristic wavelengths.

[0019] The present invention comprehensively considers the structural performance and the compositional performance of the wavelength to obtain several characteristic wavelengths with the strongest characteristic performance, further improving the accuracy of the classification detection of the coke residue samples.

[0020] In a second aspect, the present invention provides a classification detection system for coal coke residue samples, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned classification detection method for coal coke residue samples is implemented.

[0021] By adopting the above technical solution, the above-mentioned classification detection method for coal coke residue samples is generated into a computer program and stored in the memory to be loaded and executed by the processor, so as to manufacture a terminal device according to the memory and the processor, which is convenient to use.

[0022] The beneficial effects of the present invention are as follows: (1) The present invention extracts several effective wavelengths from the wavelength range of near-infrared light as characteristic wavelengths, improving the classification accuracy of the classification detection model of the coke residue samples affected by invalid wavelengths; (2) The present invention classifies and detects the coke residue samples by taking the noise of the spectrum as a kind of feature, increasing the richness of the features for the classification detection of the coke residue samples; (3) The present invention establishes a classification detection model of the coke residue samples by extracting several characteristic wavelengths representing the types of coal coke residues, enabling the coke residue samples to be classified and detected automatically more accurately and efficiently. Description of the Drawings

[0023] Figure 1 is a flowchart schematically showing a classification detection method for coal coke residue samples in the present invention; Figure 2 is a schematic diagram schematically showing eight kinds of coke residues. Detailed Embodiments

[0024] The embodiments of the present invention disclose a classification detection method for coal coke residue samples. Referring to Figure 1 , it includes steps S1 - S4: S1: Obtain several kinds of coke residues, perform crushing treatment and sampling on various coke residues to obtain several coke residue samples with various crushing degrees of each coke residue, and use a near-infrared spectrometer to obtain the spectra of each coke residue sample.

[0025] It should be noted that cinder is divided into eight types according to physical properties such as viscosity, degree of melting, and expansion height. For example, Figure 2 is a schematic diagram of eight types of cinders. As the type serial number increases, the viscosity, degree of melting, and expansion height of the cinder gradually increase. These physical properties will affect the intensity transmission of near-infrared light, and further affect the spectral results and component analysis. Therefore, the present invention analyzes the influence of the cinder morphology on the spectrum, removes the complicated pretreatment before cinder spectrum detection, and improves the accuracy of cinder sample classification detection.

[0026] Specifically, a number of cinders are obtained, and the cinders include all types of cinders; a pulverizing device is used to perform pulverizing treatment on various cinders in several batches and sample them to obtain cinder samples of several pulverizing degrees of various cinders; a near-infrared spectrometer is used to obtain the spectra of each cinder sample, and the spectra are absorbance sequences corresponding to several wavelengths.

[0027] The pulverizing device is a device that can pulverize cinder into a specified size, such as a small pulverizer. The pulverizing degree includes several finenesses such as 60 mesh and 120 mesh, and also includes the original cinder fineness with a pulverizing degree of 0, that is, without pulverizing. The volumes of the cinder samples are the same, and the volume size is set by the implementer according to the actual implementation situation. For example, the volume can be set to 10 mL. The wavelength range of the spectra is 780 - 2526 nm.

[0028] It should be noted that to ensure the robustness of the detection, the number of cinder samples with the same cinder and the same pulverizing degree is greater than 1.

[0029] So far, the spectra of cinder samples of several pulverizing degrees of several cinders have been obtained.

[0030] S2: Based on the filtering algorithm, obtain the noise-reduced spectra of each cinder sample; according to the difference between the spectrum of the cinder sample and the corresponding noise-reduced spectrum, and the fluctuation frequency of the spectrum of the cinder sample, obtain the noise index of each wavelength of the spectrum of each cinder sample; according to the law of the difference of the noise index of each wavelength in all the same type of cinder and different cinder samples, obtain the structural performance of each wavelength.

[0031] It should be noted that cinder with a higher expansion height has a porous structure, which will affect the reflection of light and cause a decrease in the signal-to-noise ratio of the spectrum. Therefore, there are certain noise differences in the spectra of cinder samples with different pulverizing degrees of cinder with a higher expansion height. This difference reflects the structural characteristics of the cinder. Therefore, the present invention obtains the structural characteristics of each wavelength relative to the cinder according to the spectral noise difference of the same type of cinder and different cinder samples.

[0032] It should be noted that filtering algorithms are often used for denoising spectral data. For example, the Savitzky-Golay (S-G) filter realizes the smoothing and denoising of spectra by performing a least squares fitting of a k-th order polynomial on the data within a sliding window. Then, considering it reversely, based on the denoising effect of the filtering algorithm, the noise performance of the spectrum can be obtained. Therefore, in the present invention, the spectra of coke residue samples are filtered and denoised, and according to the differences in absorbance at each wavelength between the denoised spectrum and the original spectrum, the noise index at each wavelength of the spectra of each coke residue sample is obtained.

[0033] Specifically, the spectra of all coke residue samples are filtered using the S-G filtering algorithm to obtain the denoised spectra of the spectra of each coke residue sample. It should be noted that the parameters of the S-G filtering algorithm include the polynomial order and the window length, and the parameters are set by the implementer according to the actual implementation situation. For example, the polynomial order can be set to 3 and the window length can be set to 5.

[0034] It should be noted that for the spectrum of any coke residue sample and the corresponding denoised spectrum, the greater the difference in absorbance at each wavelength within the neighborhood of the same wavelength, the higher the degree of denoising, and the higher the noise index of the wavelength. In addition, noise also has the characteristic of fluctuation. The spectrum of the coke residue sample fluctuates up and down around the denoised spectrum. The higher the fluctuation frequency of the spectrum of the coke residue sample, the higher the noise index of the wavelength.

[0035] Preferably, according to the difference between the spectrum of the coke residue sample and the corresponding denoised spectrum, and the fluctuation frequency of the spectrum of the coke residue sample, the noise index at each wavelength of the spectra of each coke residue sample is obtained: Denote the spectrum of any coke residue sample as the target spectrum, and any wavelength of the target spectrum as the target wavelength. Denote the target wavelength and A wavelengths in the left neighborhood and A wavelengths in the right neighborhood of the target wavelength as the neighborhood of the target wavelength. It should be noted that A is set by the implementer according to the actual implementation situation. For example, A can be set to 5.

[0036] Subtract the absorbance sequence of the target spectrum from the absorbance sequence of the corresponding denoised spectrum to obtain the error sequence of the target spectrum; use the sign function to symbolize the error sequence of the target spectrum to obtain the error sign sequence of the target spectrum; perform a forward difference on the error sign sequence of the target spectrum to obtain the error sign difference sequence of the target spectrum. It should be noted that the error sign sequence of the target spectrum only contains -1, 0, 1, removing the physical meaning of the specific numerical values, indicating the error direction of each wavelength of the target spectrum. The error sign difference sequence of the target spectrum only contains -1, 0, 1. The number of -1 and 1 reflects the fluctuation frequency of the target spectrum. If the error sign difference sequence of the target spectrum contains more 0s, it means the sequence is more stable and the fluctuation frequency is lower, while if there are more -1 and 1, it means the sequence fluctuates up and down more frequently.

[0037] The noise index of the target wavelength satisfies the expression: ; In the formula, represents the noise index of the target wavelength; represents the number of wavelengths in the neighborhood of the target wavelength; represents the i-th value at the corresponding position in the neighborhood of the target wavelength in the error sequence of the target spectrum; represents the c-th value at the corresponding position in the neighborhood of the target wavelength in the error symbol difference sequence of the target spectrum; represents the absolute value function.

[0038] In the formula, represents the accumulation of the values at the corresponding positions of the target wavelength and the neighborhood wavelengths in the error sequence of the target spectrum, representing the total error within the neighborhood of the target wavelength. The larger this value, the larger the noise index of the target wavelength; represents the accumulation of the values at the corresponding positions of the target wavelength and the neighborhood wavelengths in the error symbol difference sequence of the target spectrum, representing the fluctuation frequency within the neighborhood of the target wavelength. The larger this value, the higher the fluctuation frequency within the neighborhood of the target wavelength, thus indicating a larger noise index of the target wavelength.

[0039] Thus far, the noise indices of each wavelength of the spectra of each coke residue sample have been obtained.

[0040] It should be noted that for coke residue samples with different degrees of crushing of the same type of coke residue, if the noise index of the wavelength continuously decreases as the degree of crushing deepens, it indicates that the wavelength can strongly reflect the structural characteristics of the coke residue; if the noise index of the wavelength can reflect the structural characteristics of the coke residue in all coke residues, it means that the degree of crushing affects the absorbance data of the target wavelength. The absorbance of the wavelength can inversely characterize the degree of crushing of the coke residue. Since the degree of crushing of the coke residue is relatively intuitive and can also be measured conveniently, the present invention extracts the wavelengths that reflect the structural characteristics of the coke residue. On the one hand, it can be used as a feature to characterize the type of coke residue, and on the other hand, it can be used as a feature to characterize the volume of the coke residue sample, improving the accuracy of the classification and detection of the coke residue sample.

[0041] Preferably, according to the law of the differences in the noise indices of each wavelength among different coke residue samples of the same type of coke residue, the structural performance of each wavelength is obtained: Taking any coke residue as the target coke residue, obtaining the noise index of the a-th wavelength of the spectra of all coke residue samples of the target coke residue, and using the least squares method for linear fitting to obtain the noise index fitting line of the a-th wavelength of the target coke residue.

[0042] It should be noted that the better the fitting effect of the noise index fitting line of the a-th wavelength of the target coke residue and the more regularly distributed according to the degree of crushing, the stronger the structural characteristics of the a-th wavelength relative to the target coke residue.

[0043] The structural characteristics of each wavelength relative to the target coke satisfy the expression: ; In the formula, represents the structural characteristics of the a-th wavelength relative to the target coke; represents the number of coke samples of the target coke; represents the distance between the h-th coke sample of the target coke and the noise index fitting line of the a-th wavelength of the target coke; represents the type of crushing degree; , , represent the average noise index of the a-th wavelength of the coke sample spectra of the (q + 1)-th, q-th, and (q - 1)-th crushing degrees of the target coke; represents the exponential function with the natural constant as the base; represents the normalization function.

[0044] In the formula, compared with represents the average distance between all coke samples of the target coke and the noise index fitting line of the a-th wavelength of the target coke. This value reflects the fitting effect of the noise index fitting line of the a-th wavelength of the target coke. The larger this value is, the worse the fitting effect of the noise index fitting line of the a-th wavelength of the target coke, and then the worse the structural characteristics of the a-th wavelength relative to the target coke; represents the change regularity of the noise index of the coke samples with adjacent crushing degrees of the target coke. The larger this value is, the more obvious the performance of the noise index of the coke samples of the target coke changing with the crushing degree, thus reflecting that the structural characteristics of the a-th wavelength relative to the target coke are stronger.

[0045] It should be noted that if the noise index of the a-th wavelength can strongly reflect the structural characteristics of the coke in all coke samples, the structural performance of the a-th wavelength is higher.

[0046] Take the mean value of the structural characteristics of the a-th wavelength relative to all types of coke as the structural performance of the a-th wavelength.

[0047] So far, the structural performance of any wavelength has been obtained.

[0048] S3: According to the consistency of the absorbance differences of the same wavelength and the coke type differences of different cokes, obtain the component characteristics of each wavelength relative to any type of coke.

[0049] It should be noted that the essential reason for the morphological differences in different types of coke residues is the difference in their components. As the ash content and volatile matter decrease and the content of the plasticizer increases, the coke residues gradually exhibit adhesiveness and expansibility. Therefore, among all wavelengths, the wavelength at which the absorbance varies due to the type of coke residue but not due to the degree of comminution of the coke residue is more capable of reflecting the components of the coke residue.

[0050] Specifically, the absorbance difference of any wavelength relative to all coke residues satisfies the expression: ; In the formula, represents the absorbance difference of the a-th wavelength relative to all coke residues; represents the number of types of coke residues; , represent the average absorbances of the a-th wavelength of the u-th and v-th types of coke residues; represents the absolute value function; represents the normalization function.

[0051] In the formula, represents the average absorbance difference of the a-th wavelength of all different types of coke residues. The larger this value is, the greater the difference in the absorbances of the a-th wavelength of different types of coke residues, and the greater the absorbance difference of the a-th wavelength relative to all coke residues.

[0052] It should be noted that the greater the absorbance difference of the a-th wavelength relative to all coke residues, and the closer the absorbances of the a-th wavelength of the same type of coke residue are, the more capable the a-th wavelength is for component detection; in addition, if the absorbance of the a-th wavelength changes regularly with the increase of the type number, it means that the reason for the absorbance difference of the a-th wavelength due to the type of coke residue lies in the gradual change of the components, thus indicating that the a-th wavelength can be used for component detection.

[0053] Preferably, the average absorbance sequence of the a-th wavelength of all types of coke residues is obtained in ascending order of the type number, and the forward difference is performed on the average absorbance sequence of the a-th wavelength of all types of coke residues to obtain the difference sequence of the average absorbance sequence of the a-th wavelength of all types of coke residues; the ratio of the number of values greater than 0 to the number of values less than 0 in the difference sequence of the average absorbance sequence of the a-th wavelength of all types of coke residues is recorded as the absorbance change regularity of the a-th wavelength relative to all coke residues.

[0054] Preferably, the component performance of any wavelength satisfies the expression: ; In the formula, represents the component performance of the a-th wavelength; represents the absorbance difference of the a-th wavelength relative to all coke residues; Indicates the regularity of the absorbance change of the a-th wavelength relative to all coke residues; Indicates the number of coke residue types; Indicates the variance of the absorbance of the a-th wavelength of all coke residue samples of the u-th coke residue; Indicates the exponential function with the natural constant as the base.

[0055] In the formula, the greater the regularity of the absorbance change of the a-th wavelength relative to all coke residues, the more the absorbance change of the a-th wavelength conforms to the change of the components; Indicates the sum of the variances of the absorbances of the a-th wavelength of all coke residue samples of all coke residues. The larger this value is, the more discrete the absorbances of the a-th wavelength of all coke residues are. Therefore, the stability of the absorbance change of the a-th wavelength in the coke residue samples of the same type of coke is low and it is difficult to be used as a characteristic detection component.

[0056] So far, the component performance of any wavelength has been obtained.

[0057] S4: Combine the structural performance and component performance of the wavelengths to obtain characteristic wavelengths; based on the characteristic wavelengths and the neural network model, establish a classification detection model for coke residue samples.

[0058] It should be noted that since the greater the structural performance of a wavelength, the greater the difference in absorbance among the coke residue samples of the same type of coke with different pulverization degrees, and the smaller the difference in absorbance among the coke residue samples of the same type of coke with different pulverization degrees for the wavelength with greater component performance, for a wavelength, it is impossible to have both a high structural performance and a high component performance at the same time. In order to make the characteristic meaning of the wavelength stronger, it is necessary to screen the wavelengths so that the wavelength only has a strong structural performance meaning or a strong component performance meaning.

[0059] It should be noted that since the coke residue types and pulverization degrees corresponding to the coke residue samples are known, a supervised machine learning model can be used to learn the absorbances of the characteristic wavelengths of all coke residue samples, and then it is possible to perform automated sample detection on subsequent coal coke residues. Random forest has strong generalization ability and can handle more features, which conforms to the characteristics of spectral data. Therefore, the present invention uses random forest to construct a classification detection model for coke residue samples.

[0060] Specifically, compare the structural performance and component performance of the a-th wavelength as the first ratio of the a-th wavelength; compare the component performance and structural performance of the a-th wavelength as the second ratio of the a-th wavelength; take the maximum value of the first ratio and the second ratio of the a-th wavelength as the characteristic goodness of the a-th wavelength; obtain the Z wavelengths with the largest characteristic goodness among all wavelengths as the characteristic wavelengths. It should be noted that the value of Z is set by the implementer according to the actual implementation situation. For example, the value of Z can be set to 20.

[0061] Construct a classification and detection model for coke breeze samples: Use the absorbance of the characteristic wavelengths of all coke breeze samples as the training set. Perform Bootstrap sampling on the training set to generate several sub-datasets. Randomly select features for each sub-dataset to construct decision trees. All decision trees form a random forest, which serves as the classification and detection model for coke breeze samples. The input is the absorbance of the characteristic wavelengths of coke breeze samples, and the output is the type of coke breeze samples. The number of decision trees is set by the implementer according to the actual implementation situation. For example, the number of decision trees can be set to 100.

[0062] Input the absorbance of the characteristic wavelengths of the coal coke breeze sample to be measured into the classification and detection model of coke breeze samples to obtain the type of the coal coke breeze sample to be measured.

[0063] Thus, the classification and detection of coal coke breeze samples are completed.

[0064] An embodiment of the present invention also discloses a classification and detection system for coal coke breeze samples, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, a classification and detection method for coal coke breeze samples according to the present invention is implemented.

[0065] The above system also includes other components well-known to those skilled in the art, such as a communication bus and a communication interface. Their settings and functions are known in the art, so they will not be elaborated here.

[0066] Although this specification has shown and described multiple embodiments of the present invention, it is obvious to those skilled in the art that such embodiments are provided only by way of example. Those skilled in the art will think of many changes, alterations, and alternative ways without departing from the spirit and idea of the present invention.

Claims

1. A method for classifying and detecting coal cinder samples, characterized in that, Including: Obtain several kinds of cinder, crush various cinders and take samples to obtain cinder samples with several degrees of crushing of various cinders, and use a near-infrared spectrometer to obtain the spectra of each cinder sample, where the spectra are sequences of absorbances corresponding to several wavelengths; Based on a filtering algorithm, obtain the noise-reduced spectra of each cinder sample; according to the differences between the spectra of the cinder samples and the corresponding noise-reduced spectra, and the fluctuation frequencies of the spectra of the cinder samples, obtain the noise indices of each wavelength of the spectra of each cinder sample; according to the laws of the differences in the noise indices of each wavelength among different cinder samples of the same kind of cinder, obtain the structural manifestations of each wavelength; According to the absorbance differences of the same wavelength of different cinders and the absorbance differences of the same wavelength of cinder samples with different degrees of crushing of the same kind of cinder, obtain the compositional manifestations of each wavelength; Combine the structural manifestations and compositional manifestations of the wavelengths to obtain characteristic wavelengths; Based on the characteristic wavelengths and a neural network model, establish a classification and detection model for cinder samples.

2. The method for classifying and detecting coal cinder samples according to claim 1, wherein The obtaining of the noise indices of each wavelength of the spectra of each cinder sample includes: Denote the spectrum of any cinder sample as the target spectrum, and denote any wavelength of the target spectrum as the target wavelength; obtain the neighborhood of the target wavelength; obtain the error sequence of the target spectrum and the error sign difference sequence of the target spectrum; ； Wherein, represents the noise index of the target wavelength; represents the number of wavelengths in the neighborhood of the target wavelength; represents the i-th value at the corresponding position in the neighborhood of the target wavelength in the error sequence of the target spectrum; represents the c-th value at the corresponding position in the neighborhood of the target wavelength in the error sign difference sequence of the target spectrum; represents the absolute value function.

3. A method for classifying and detecting coal cinder samples according to claim 2, characterized in that, The obtaining of the error sequence of the target spectrum and the error sign difference sequence of the target spectrum includes: Subtract the absorbance sequence of the target spectrum from the absorbance sequence of the corresponding noise-reduced spectrum to obtain the error sequence of the target spectrum; symbolize the error sequence of the target spectrum using the sign function to obtain the error sign sequence of the target spectrum; perform forward difference on the error sign sequence of the target spectrum to obtain the error sign difference sequence of the target spectrum.

4. A method for classifying and detecting coal cinder samples according to claim 1, characterized in that, The obtaining of the structural manifestations of each wavelength includes: Take any cinder as the target cinder, obtain the noise indices of the a-th wavelength of the spectra of all cinder samples of the target cinder, and use the least squares method for linear fitting to obtain the noise index fitting line of the a-th wavelength of the target cinder; Based on the fitting effect of the noise index fitting line of the a-th wavelength of the target cinder and the distribution law of the noise indices of the a-th wavelength of the spectra of all cinder samples of the target cinder according to the degree of crushing, obtain the structural characteristics of the a-th wavelength relative to the target cinder; Take the mean of the structural characteristics of the a-th wavelength relative to all kinds of cinders as the structural manifestation of the a-th wavelength.

5. A method for classifying and detecting coal cinder samples according to claim 4, characterized in that, The structural characteristics of the a-th wavelength relative to the target cinder satisfy the expression: ； Wherein, represents the structural characteristics of the a-th wavelength relative to the target coke breeze; represents the number of coke breeze samples of the target coke breeze; represents the distance between the h-th coke breeze sample of the target coke breeze and the noise index fitting line of the a-th wavelength of the target coke breeze; represents the type of crushing degree; 、 、 represent the average noise index of the a-th wavelength of the coke breeze spectra of the (q + 1)-th, q-th, and (q - 1)-th crushing degrees of the target coke breeze; represents the exponential function with the natural constant as the base; represents the normalization function.

6. A method for classifying and detecting coal cinder samples according to claim 1, characterized in that, The obtaining of the compositional manifestations of each wavelength includes: Obtain the absorbance differences of each wavelength relative to all cinders; obtain the regularity of the absorbance changes of each wavelength relative to all cinders; The compositional manifestation of any wavelength satisfies the expression: ； In the formula, represents the component expression of the a-th wavelength; represents the absorbance difference of the a-th wavelength relative to all coke residues; represents the absorbance change regularity of the a-th wavelength relative to all coke residues; represents the number of coke residue types; represents the absorbance variance of the a-th wavelength of all coke residue samples of the u-th coke residue; represents the exponential function with the natural constant as the base.

7. A method for classifying and detecting coal cinder samples according to claim 6, characterized in that, The absorbance differences of each wavelength relative to all cinders satisfy the expression: ； In the formula, represents the absorbance difference of the a-th wavelength relative to all coke residues; represents the number of coke residue types; , represent the average absorbance of the a-th wavelength of the u-th and v-th coke residues; represents the absolute value function; represents the normalization function.

8. A method for classifying and detecting coal cinder samples according to claim 6, characterized in that, The obtaining of the regularity of the absorbance changes of each wavelength relative to all cinders includes: Obtain the average absorbance sequence of the a-th wavelength of all types of cinders in ascending order of the type number, and perform forward difference on the average absorbance sequence of the a-th wavelength of all types of cinders to obtain the difference sequence of the average absorbance sequence of the a-th wavelength of all types of cinders; Compare the number of values greater than 0 with the number of values less than 0 in the difference sequence of the average absorbance sequence at the a-th wavelength of all types of coke residues, and denote it as the absorbance change regularity of the a-th wavelength relative to all coke residues.

9. A method for classifying and detecting coal cinder samples according to claim 1, characterized in that, The obtaining of the characteristic wavelength includes: Compare the structural performance and compositional performance at the a-th wavelength as the first ratio at the a-th wavelength; compare the compositional performance and structural performance at the a-th wavelength as the second ratio at the a-th wavelength; take the maximum value of the first ratio and the second ratio at the a-th wavelength as the characteristic goodness of the a-th wavelength; obtain the Z wavelengths with the largest characteristic goodness among all wavelengths as the characteristic wavelengths.

10. A classification and detection system for coal cinder samples, characterized in that, including: A processor and a memory, where the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a method for classifying and detecting coal coke residue samples according to any one of claims 1-9 is implemented.

Images