Near-infrared coal quality detection method based on multi-dimensional data fusion

Abstract

The invention relates to a near-infrared coal quality detection method based on multi-dimensional data fusion so as to improve the coal detection precision. According to the method, a modeling sample set containing multi-dimensional data of the producing area of a coal sample, granularity characteristics, the distance between the coal sample and a probe, air humidity and environment temperature is constructed, near infrared spectrum data of the coal sample is collected, spectrum preprocessing, feature extraction and data fusion are carried out, and finally a quantitative analysis model is constructed to predict the coal quality characteristics in a high-precision mode. According to the method, a one-dimensional convolutional neural network (1d-CNN) technology is adopted, automation and intelligence of coal quality detection are realized, and the detection efficiency is improved. In addition, accurate data of moisture, ash content and volatile components in the coal sample are obtained strictly according to a national standard experimental method, and the representativeness of the sample and the reliability of the data are improved. Noise interference is effectively reduced through the preprocessing step, and scattering influence is corrected. The method provides powerful technical support for quality control, reasonable utilization of resources and the like in the coal industry.

Description

Technical Field

[0001] The present invention relates to the technical field of coal detection, and in particular to a near-infrared coal quality detection method based on multi-dimensional data fusion. Background Art

[0002] In the field of coal detection technology, traditional near-infrared spectroscopy is widely used due to its high efficiency and non-destructive characteristics. However, this technology is often affected by many factors in practical applications, including the physical state of the coal sample (particle size composition), spectral acquisition conditions (the distance between the detector and the coal sample), and environmental conditions (air humidity, temperature). Changes in these variables may cause fluctuations in spectral data, thereby affecting the accuracy and stability of coal quality detection results.

[0003] The particle size composition of coal samples has a significant impact on the spectral absorption characteristics. Coal samples of different particle sizes have different spectral responses. The distance between the detector and the coal sample will also affect the intensity and quality of the spectral signal. Changes in air humidity and temperature may affect changes in spectral signals. The combined effect of these factors makes it difficult for traditional near-infrared spectroscopy analysis technology to provide stable and reliable coal quality detection results.

[0004] In order to solve the above problems and improve the accuracy of coal quality detection, the present invention proposes a new near-infrared coal quality detection method. This method is based on multi-dimensional data fusion technology. By constructing multiple data sets such as coal sample particle size composition, the distance between the probe and the coal sample, and air humidity, a multi-dimensional data model is established. The dimensions of the multi-dimensional data include: particle size, the distance between the detector and the coal sample, air humidity, temperature, and near-infrared spectral characteristic wavelength. The model comprehensively considers the above influencing factors, reduces their influence on spectral data, enhances the robustness of the model, and can perform more accurate detection of coal quality to meet the strict requirements of the modern coal industry for coal quality detection. Summary of the invention

[0005] The technical problem solved by the present invention is to eliminate or reduce the influence of coal sample state and spectrum acquisition conditions on the spectrum, and improve the detection accuracy. In order to solve the above technical problem, the present invention provides the following technical solutions:

[0006] 1. Construct a modeling sample set: Determine the type of data that needs to be collected, including coal quality indicators, indicator ranges, particle size characteristics, distance between coal samples and probes, air humidity, temperature, etc., to ensure that the sample set can fully reflect the key factors affecting the modeling results.

[0007] 2. Near-infrared spectroscopy detection: Obtain spectral data of coal samples under different conditions for subsequent modeling and analysis.

[0008] 3. Spectral preprocessing: De-noising, normalization, smoothing and baseline correction are performed on the collected near-infrared spectral data to improve data quality and stability.

[0009] 4. Feature extraction: Principal component analysis is used to extract key features such as absorption peak, wavelength position, peak width and peak area from the preprocessed spectral data. These features are closely related to the characteristics of the coal samples.

[0010] 5. Data fusion: Fuse multi-dimensional data (particle size, humidity and spectral characteristics) to enhance the robustness of the model.

[0011] 6. Construct a quantitative analysis model: Use one-dimensional convolutional neural network (1d-CNN) to build a model to predict coal quality characteristics.

[0012] Beneficial Effects

[0013] 1. Compared with specific commonly used technologies, the near-infrared coal quality detection method based on multi-dimensional data fusion can detect coal samples quickly, in real time and non-destructively.

[0014] 2. Based on multi-dimensional data fusion, the representativeness of the samples and the reliability of the data are improved. The preprocessing step effectively reduces noise interference, corrects scattering effects and enhances data comparability.

[0015] 3. Model construction and optimization The quantitative analysis model constructed by one-dimensional convolutional neural network (1d-CNN) realizes the automation and intelligence of coal quality detection, reduces manual intervention and analysis time, and improves detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The process framework diagram of the present invention is

[0017] Figure 2 Quantitative analysis model network architecture diagram DETAILED DESCRIPTION

[0018] In order to have a further understanding and recognition of the structural features and the effects achieved by the present invention, a preferred embodiment and accompanying drawings are used for detailed description as follows:

[0019] like Figure 1 As shown, the near-infrared coal quality detection method based on multi-dimensional data fusion described in the present invention comprises the following steps:

[0020] 1. Construct a modeling sample set:

[0021] Determine the type of data that needs to be collected, including the origin of the coal sample, particle size characteristics, distance between the coal sample and the probe, air humidity, etc., to ensure that the sample set can fully reflect the key factors that affect the modeling results. For the above-determined data types, use standardized methods and equipment to collect data. Integrate the collected multi-dimensional data to build a comprehensive modeling sample set that can cover the characteristics of coal samples under different conditions.

[0022] 2. Specifically, in order to construct the coal sample origin data set, the present invention adopts the following approach:

[0023] In order to ensure that the model can identify the differences in the characteristics of coal samples from different origins, the present invention constructs a coal sample origin data set. To this end, the present invention collects coal samples from multiple regions and coal mines under different mining conditions to ensure the diversity of this data set. These samples cover from lignite in the north to anthracite in the south, as well as bituminous coal from different regions, thereby ensuring that the data set can fully reflect the coal quality characteristics of major coal producing areas in China and even the world. For the collection of each type of coal sample, the present invention adopts a standardized sampling and analysis method, uses a unified sampling device, follows the same sampling frequency and depth, and uses the same analysis process in the laboratory to determine the moisture, ash, volatile matter and fixed carbon content of the coal sample. Through these strict sampling and analysis processes, the present invention constructs a coal sample origin data set containing several samples. This data set contains spectral data of coal samples from different regions across the country, as well as the ash, volatile matter, moisture and fixed carbon content of the coal samples, providing rich data support for subsequent modeling and analysis.

[0024] 3. Specifically, in order to construct a coal sample particle size characteristic data set, the present invention adopts the following approach:

[0025] Particle size is an important factor affecting coal quality characteristics and spectral response. In order to capture the impact of coal sample particle size on spectral characteristics, the present invention constructs a coal sample particle size characteristic data set. The present invention collects coal samples with a specific particle size range from coal samples of multiple different origins, including but not limited to coal samples of 0.15 to 0.075 mm and 0.075 to 0.045 mm, to ensure that the data set can cover the key intervals of coal sample particle size characteristics. This data set not only contains the spectral data of coal samples of each particle size level, but also includes the ash, volatile matter, moisture and fixed carbon content of the coal samples. This data set can be used to analyze and model the relationship between coal sample particle size characteristics and spectral detection results.

[0026] 4. Specifically, the present invention adopts the following approach to construct a data set of distances between coal samples and probes:

[0027] The present invention determines the distance between the coal sample and the probe as a key parameter, which has a direct impact on the intensity and quality of the spectral signal. The present invention systematically changes the distance between the coal sample and the probe, and collects corresponding spectral data at each distance point. These distance points cover a range of values ​​from 1.3m to 3.0m to ensure that the data set can fully reflect the impact of different distances on the spectral detection results. This data set not only contains spectral data collected at different distances between the coal sample and the probe, but also includes the ash, volatile matter, moisture and fixed carbon content of the coal sample. This data set is used to analyze and model the impact of the distance between the coal sample and the probe on the spectral detection results.

[0028] 5. Specifically, the present invention adopts the following approach to construct the coal sample air humidity data set:

[0029] In order to ensure that the model can adapt to coal quality detection under different humidity conditions, the present invention records the air humidity of each coal sample. This process involves spectral detection of coal samples under different humidity environments to collect spectral data related to humidity. This dataset not only contains spectral data collected under different humidity conditions, but also includes ash, volatile matter, moisture and fixed carbon content of coal samples. This dataset is used to analyze and model the impact of air humidity on spectral detection results.

[0030] 6. Specifically, the present invention adopts the following approach to constructing the ambient temperature data set:

[0031] In order to ensure that the model can adapt to coal quality detection under different temperature conditions, the present invention records the ambient temperature when measuring the near-infrared spectrum of each coal sample. This process involves spectral detection of coal samples under different temperature environments to collect temperature-related spectral data. This dataset not only contains spectral data collected under different temperature conditions, but also includes ash, volatile matter, moisture, and fixed carbon content of the coal samples. This dataset is used to analyze and model the impact of ambient temperature on spectral detection results.

[0032] 7. Near infrared spectroscopy detection:

[0033] The near infrared spectroscopy detection process aims to obtain spectral data of coal samples under different environmental and coal quality conditions, which will be used for subsequent modeling and analysis. In order to ensure the comprehensiveness and accuracy of the spectral data, the present invention implements near infrared spectroscopy detection on the sample set under the above control conditions, which include but are not limited to different coal sample origins, particle size characteristics, distance between coal samples and probes, and air humidity, forming a comprehensive near infrared spectroscopy database. This database is used to analyze and model the relationship between coal sample characteristics and spectral data, thereby optimizing the design and operation of the detection system.

[0034] 8. Specifically, for the spectral data collection of near infrared spectroscopy, the present invention adopts the following approach:

[0035] The spectral data collection and averaging of the present invention uses the NIR25S fiber optic spectrometer produced by Shanghai Fuxiang Optics Co., Ltd. to collect the near-infrared spectral data of the above-mentioned coal samples that have been processed and measured for moisture, ash, and volatile content. Each coal sample is collected 30 times and the average value is taken. The corresponding moisture, ash, and volatile content values ​​are used as markers for spectral data, and the corresponding relationship between spectral data and coal quality component content is established to provide an accurate basis for subsequent model training and prediction. During the collection process, it is ensured that the wavelength range of the spectrometer covers the near-infrared characteristic absorption band of coal, the spectral resolution meets the accuracy requirements for coal quality component analysis, and the stability and repeatability of the instrument are maintained to reduce data errors caused by instrument factors.

[0036] 9. After the near infrared spectrum data is collected, the present invention performs spectrum preprocessing and adopts the following methods:

[0037] In the process of building a modeling sample set, near-infrared spectroscopy is an important means of obtaining coal sample characteristic data. For spectral data collected under different conditions, this technical solution proposes a detailed preprocessing process to ensure the quality and availability of the data, thereby providing accurate and reliable spectral data support for modeling and analysis.

[0038] 10. Specifically, in spectral preprocessing, spectral denoising:

[0039] For the spectral denoising, continuous wavelet transform, Savitzky-Golay filtering and other methods are proposed for comparative analysis to retain the key information in the spectral data.

[0040] The following is the mathematical expression of continuous wavelet transform (CWT):

[0041]

[0042] Where x(t) is the input signal, ψ* is the complex conjugate of the wavelet function, a is the scale parameter, and b is the translation parameter.

[0043] The following are the specific steps of Savitzky-Golay filtering:

[0044] Sliding window: Slide a fixed window of size 11 over the data sequence.

[0045] Polynomial fitting: In each window, a polynomial with an order of 3 is fitted using the least squares method.

[0046] Smoothing value calculation: The original data point is replaced by the value of the polynomial at the center point of the window to achieve a smoothing effect. This process is repeated for each point in the data, effectively smoothing and reducing noise.

[0047] 11. Specifically, in spectral preprocessing, the normalization of the spectrum:

[0048] The spectral data is normalized to eliminate the influence of different samples due to different measurement conditions. The normalization method is to use multiple algorithms such as maximum-minimum normalization and Z-score normalization to make the data have a unified scale for easy comparison and analysis.

[0049] The following is the specific algorithm for maximum-minimum normalization:

[0050]

[0051] Where X is the original data point. Xmin is the minimum value in the data set. Xmax is the maximum value in the data set. Xnorm is the normalized data point.

[0052] The following is the specific algorithm for Z-score standardization:

[0053]

[0054] Where z is the Z-score value, which is the normalized score. x is the original data point. μ is the mean of the data set. σ is the standard deviation of the data set.

[0055] 12. Specifically, in spectral preprocessing, spectrum smoothing:

[0056] The spectral data is smoothed to reduce random noise and fluctuation and improve the stability of the spectral data. The smoothing method is to use moving average technology.

[0057] The following is the specific implementation of moving average:

[0058] Select the window size: The window size for the moving average is 11, which means 11 consecutive data points are used to calculate the average.

[0059] Calculate a moving average: For each point in the time series, calculate the average of the values ​​at that point and the 10 points before it. The window is then moved to the right by one data point and the calculation process is repeated.

[0060] 13. Specifically, in spectral preprocessing, the baseline of the spectrum is corrected:

[0061] Baseline correction is performed on the spectral data to eliminate baseline drift caused by improper instrument or sample preparation. Baseline correction helps to highlight the characteristic absorption peaks in the spectrum and improve the recognition ability of the model. Key features such as peak value, wavelength, absorption coefficient, etc. are extracted from the preprocessed spectral data, which are of great significance for modeling and analysis.

[0062] 14. After the spectrum preprocessing, the present invention performs feature extraction on the spectrum:

[0063] In the process of building a comprehensive modeling sample set, feature extraction is an important step to convert preprocessed spectral data into key information that can be used for modeling. The following is a detailed description of the feature extraction process to ensure that key features that can represent the characteristics of coal samples are extracted from the spectral data. Identify and select spectral features that are closely related to the characteristics of coal samples, and extract absorption peaks, wavelength positions, peak widths, and peak areas at different positions, which can significantly affect the modeling results. Then use principal component analysis (PCA) to automatically extract key features from spectral data, reduce human bias, and improve extraction efficiency.

[0064] Finally, the spectral features are combined with multiple features of the physical and chemical properties of coal samples to enhance the predictive ability and generalization of the model.

[0065] 15. After feature extraction, the extracted features are fused:

[0066] After feature extraction, the extracted features are fused to integrate the coal sample particle size characteristics, the distance between the coal sample and the probe, the air humidity, and the characteristic wavelengths in the spectral data to form a unified data set. During the fusion process, the correlation between the features is analyzed, and the redundant features are identified and eliminated through principal component analysis (PCA), retaining the most representative and informative features. Then, the recursive feature elimination (RFE) algorithm is used to identify the features that contribute most to the model prediction ability. In order to enhance the spectral absorption characteristics and eliminate the influence of the detection conditions on the diffuse reflectance spectrum of coal rock, feature enhancement techniques such as first-order differential (FD), Savitzky-Golay convolution smoothing (SG convolution smoothing) and standard normal entropy transformation (SNV) are used. Multi-dimensional data are fused into a comprehensive feature vector through feature-level fusion and decision-level fusion methods, in which feature-level fusion directly combines different features, while decision-level fusion combines the prediction results of different features in the model prediction stage.

[0067] 16. Finally, the present invention constructs a quantitative analysis model for coal quality detection:

[0068] After completing the collection, preprocessing, feature extraction and data fusion of multi-dimensional data, the present invention further proposes a method for constructing a quantitative analysis model and Figure 2The network architecture diagram of the quantitative analysis model is shown in Figure 1. The model aims to achieve high-precision prediction of coal quality characteristics through a one-dimensional convolutional neural network (1d-CNN).

[0069] The present invention designs a simple 1d-CNN architecture, which includes 3 convolutional layers and 3 pooling layers, to automatically learn deep features in the spectral data of coal samples. The convolutional layer can extract local features, while the pooling layer is used to reduce the spatial dimension of the features and increase the generalization ability of the model. At the end of the 1d-CNN, the present invention adds 3 fully connected layers with 64, 16, and 4 nodes to map the learned features to the specific characteristics of coal quality: ash, moisture, volatile matter, and fixed carbon content.

[0070] The present invention uses the above-mentioned modeling sample set containing multi-dimensional data to train the 1d-CNN model, which contains the spectral data of coal samples under different conditions and their coal quality characteristics. Through the back propagation algorithm and the gradient descent algorithm, the model can establish the corresponding relationship between the spectral data and the coal quality characteristics, so that the coal quality characteristics can be predicted through the spectral data. During the training process, the present invention uses cross-validation and other techniques to prevent overfitting and ensure that the model can be widely used.

[0071] After the model training is completed, it can be used in the detection system. By analyzing the spectral data of the coal sample in real time, the input model can quickly predict its coal quality characteristics. This quantitative analysis model based on 1d-CNN can improve the accuracy of coal quality detection, and due to its automatic feature learning ability, it greatly reduces manual intervention and analysis time, greatly improves the efficiency of coal quality detection, and realizes the automation and intelligence of coal quality detection.

[0072] Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and the specification only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention to be protected. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A near-infrared coal quality detection method based on multi-dimensional data fusion, characterized in that: The method comprises the following steps: (1) Construct a modeling sample set and determine the type of data that needs to be collected, including the origin of the coal sample, particle size characteristics, distance between the coal sample and the probe, and air humidity; (2) Near infrared spectroscopy detection, to obtain spectral data of coal samples under different environmental and coal quality conditions; (3) Spectral preprocessing: denoising, normalization, smoothing and baseline correction of the collected near-infrared spectral data; (4) Feature extraction: extracting key features such as absorption peak, wavelength position, peak width, and peak area from the preprocessed spectral data; (5) Data fusion, integrating multi-dimensional data of particle size, humidity and spectral characteristics; (6) Construct a quantitative analysis model and use a one-dimensional convolutional neural network to build a model to predict coal quality characteristics with high accuracy.

2. The method according to claim 1, characterized in that Constructing the modeling sample set involves collecting coal samples from multiple regions and mines under different mining conditions to ensure the diversity of the dataset.

3. The method according to claim 1, characterized in that: The spectral preprocessing step includes the following sub-steps: (1) Continuous wavelet transform denoising, where the mathematical expression of continuous wavelet transform is: Where x(t) is the input signal, ψ* is the complex conjugate of the wavelet function, a is the scale parameter, and b is the translation parameter; (2) Savitzky-Golay filtering denoising, including sliding window, polynomial fitting and smoothing value calculation.

4. The method according to claim 1, characterized in that In the spectral preprocessing step, normalization processing including maximum-minimum value normalization and Z-score standardization is performed. The specific algorithm is as follows: (1) Maximum-minimum value normalization: Where X is the original data point. Xmin is the minimum value in the data set. Xmax is the maximum value in the data set. Xnorm is the normalized data point. (2) Z-score standardization: Where z is the Z-score value, which is the normalized score. x is the original data point. μ is the mean of the data set. σ is the standard deviation of the data set.

5. The method according to claim 1, characterized in that In the spectral preprocessing step, a moving average technique was included for smoothing, and the window size was 11, i.e., 11 consecutive data points to calculate the average value.

6. The method according to claim 1, characterized in that In the spectral preprocessing step, baseline correction is included to eliminate baseline drift caused by improper instrumentation or sample preparation.

7. The method according to claim 1, characterized in that In the feature extraction step, principal component analysis is used to automatically extract key features, reduce human bias, and improve extraction efficiency.

8. The method according to claim 1, characterized in that In the data fusion step, a recursive feature elimination algorithm is used to identify and retain the features that contribute most to the model's predictive ability.

9. The method according to claim 1, characterized in that In the data fusion step, first-order differentiation, Savitzky-Golay convolution smoothing and standard normal entropy transformation feature enhancement techniques are used.

10. The method according to claim 1, characterized in that In the step of building a quantitative analysis model, the designed 1d-CNN architecture contains 3 convolutional layers, 3 pooling layers, and 3 fully connected layers, with 64, 16, and 4 nodes respectively, mapping the learned features to the specific characteristics of coal quality: ash, moisture, volatile matter, and fixed carbon content.

11. The method according to claim 1, characterized in that During the model training process, cross-validation techniques are used to prevent overfitting and ensure that the model can be widely applied.

12. A coal quality detection system based on the method according to any one of claims 1 to 12, characterized in that: The system includes software and hardware for executing the method, including: (1) Spectral data acquisition equipment, used to collect near-infrared spectral data of coal samples; (2) a data processing unit, used to perform spectral preprocessing, feature extraction and data fusion; (3) a model training unit, used to train a one-dimensional convolutional neural network model; (4) A prediction unit is used to analyze the spectral data of coal samples in real time and input the data into the model to quickly predict the coal quality characteristics.

Images