CN114354654B - DW-KNN-based rapid non-destructive detection method for coal moisture content - Google Patents

Abstract

The invention relates to a DW‑KNN-based rapid nondestructive detection method for moisture content of coal. It comprises the following steps: Step 1, build the coal moisture content detection model based on DW-KNN; When building the coal moisture content detection model based on DW-KNN, the training data set that makes comprises the coal training sample moisture content label of coal training sample and The coal training sample microwave S parameter spectrum information of the coal training sample under the coal training sample moisture content label; Step 2, providing the coal sample with the moisture content to be detected, obtaining the coal sample consistent with the coal training sample microwave S parameter spectrum information The microwave S-parameter spectrum information of the coal sample is processed by using the coal moisture content detection model based on DW-KNN to process the microwave S-parameter spectrum information of the coal sample to obtain and output the moisture content of the coal sample. The invention realizes the detection of the moisture content of the coal under the condition of no damage, and improves the detection accuracy and robust performance.

Description

DW-KNN-based rapid non-destructive detection method for coal moisture content

technical field

The invention relates to a rapid non-destructive detection method for moisture content of coal, in particular to a rapid non-destructive detection method for moisture content of coal based on DW-KNN.

Background technique

Moisture is one of the four basic indicators (moisture, ash, volatile matter, fixed carbon) to measure the economic value of coal. In the process of coal coking and combustion, too high or too low coal moisture content will cause problems such as reduced coal utilization efficiency, environmental pollution and energy waste. Therefore, it is of great significance to quickly and accurately detect coal moisture content.

The measurement methods of coal moisture content are mainly divided into direct and indirect measurement methods. The method of direct measurement is the standard gravimetric method, commonly known as the weight loss method, which is mainly divided into nitrogen drying method and air drying method. The standard gravimetric method is a laboratory measurement method that can obtain high measurement accuracy; however, this method takes a long time and needs to destroy the original properties of the sample.

Conventional indirect measurement methods mainly include neutron method, conductivity method, capacitance method, infrared reflection method, microwave method and so on. The principle of the neutron method to determine hydrogen in coal is based on the determination of the hydrogen content in the sample. However, besides water, there are many chemical impurities in coal that also contain hydrogen. In addition, the equipment is expensive, and there are risks in using radioactive neutron sources. Conductometric and capacitive methods are affected by various factors such as temperature, density, and electrolyte content, and have poor accuracy without compensation strategies. The infrared reflectance method has a good effect on the detection of the water content of the mixed liquid sample. For the detection of solids, only the surface moisture of small particle samples can be detected (the penetration depth is within microns). The microwave method has the advantages of safety, stability, non-destructive, non-contact, etc., and the most commonly used method is the free space transmission measurement method. This method usually only selects the single-point frequency or discrete frequency band of the microwave as the calibration signal, which will lead to the loss of the signal closely related to the moisture content; and the fitting method is simple, the detection accuracy is low, and the robustness is not high.

Contents of the invention

The purpose of the present invention is to overcome the deficiencies in the prior art and provide a DW-KNN-based rapid non-destructive detection method for coal moisture content, which can realize the detection of coal moisture content under non-destructive conditions, improve detection accuracy and robust performance.

According to the technical solution provided by the present invention, a DW-KNN-based rapid nondestructive detection method for moisture content of coal, the rapid nondestructive detection method for moisture content of coal comprises the following steps:

Step 1. Construct a coal moisture content detection model based on DW-KNN, wherein, when constructing a coal moisture content detection model based on DW-KNN, the training data set produced includes the coal training sample moisture content label of the coal training sample and the coal training sample In the coal training sample microwave S parameter spectrum information under the coal training sample moisture content label, utilize the training data set made to DW-KNN model training, to construct the coal moisture content detection model based on DW-KNN;

Step 2, provide the coal sample with the moisture content to be detected, obtain the microwave S parameter spectrum information of the coal sample consistent with the microwave S parameter spectrum information of the coal training sample, and use the above-mentioned coal moisture content detection model based on DW-KNN to The obtained coal sample microwave S-parameter spectrum information is processed to obtain and output the moisture content of the coal sample.

In step 1, when making the training data set, the following steps are specifically included:

Step 1.1. Collect raw coal samples from different areas, and screen all the collected raw coal samples, spread the screened raw coal samples evenly in the corresponding carrier, and dry the raw coal samples in the carrier, To obtain experimental coal samples in different regions;

Step 1.2. For the above-mentioned experimental coal samples obtained in different regions, they are all sequentially processed by sealing and cooling, spraying water and stirring, and sealing and standing to obtain the coal training samples under the moisture content label of the preset coal training samples;

Step 1.3: Perform a microwave test on the coal training samples obtained above to obtain the coal sample microwave S-parameter spectrum information of the coal training samples in the corresponding area under the moisture content label of the coal training samples.

In step 1.2, when water is sprayed and stirred, for any experimental coal sample, the water spray quality determined according to the moisture content label of the preset coal training sample is:

Among them, M _add is the mass of water injection, M _c&m is the mass of experimental coal sample mixed with water, MC _before is the moisture content of experimental coal sample, and MC _after is the coal sample moisture content label of coal training sample.

In step 1.3, during the microwave test of the coal training sample, for any coal training sample, collect and obtain the spectrum data of the corresponding frequency point at a frequency of 8.05GHz to 12.01GHz, so as to obtain the coal sample microwave S parameter spectrum of the coal training sample information, wherein all coal sample microwave S-parameter spectrum data of a coal training sample form an initial curve of microwave S-parameter spectrum.

The microwave S-parameter spectrum initial curves of all coal training samples from the same area are preprocessed to obtain the coal sample microwave S-parameter spectrum information of the coal training sample under the moisture content label of the coal training sample after preprocessing, wherein, Preprocessing includes the following steps:

Step 1.3.1, carry out linear fitting to any microwave S-parameter spectrum initial curve, and calculate the spectrum data distance corresponding to the linear fitting straight line between the microwave S-parameter spectrum initial curve and the microwave S-parameter spectrum initial curve, obtain the spectrum data A linear fit to the distance matrix D,

Among them, A[m-1][i] is the spectrum data of the i-th frequency point of the initial curve of the m-1 microwave S-parameter spectrum, and A*[m-1][i] is the m-1 microwave S The fitting spectrum data of the i-th frequency point of the linear fitting line corresponding to the initial curve of the parametric spectrum, n is the number of corresponding frequency points of each microwave S-parameter spectrum initial curve obtained by testing, and m is the bar of the microwave S-parameter spectrum initial curve number;

Step 1.3.2, compare the spectral data linear fitting distance in the spectral data linear fitting distance matrix D with the preset distance threshold, and eliminate the initial curve of the microwave S-parameter spectrum greater than the distance threshold to obtain the current area Corresponding coal sample microwave S-parameter spectrum curve;

The microwave S-parameter spectrum curves of coal samples in all regions constitute the microwave S-parameter spectrum information of coal training samples in the training data set.

The preset distance threshold is 1.5×dis_m, wherein dis_m is the median distance after sorting the spectral data linear fitting distance of the area where the current raw coal sample is located.

Divide the prepared training data set according to the collection area of coal training samples to obtain multiple sets of training data subsets;

Input the training data subset into the DW-KNN model with k-nearest neighbors according to the leave-one-out cross-validation method, and calculate the cross-validation mean value of the evaluation index at different k-nearest neighbor numbers, and select the k-nearest neighbor when the cross-validation mean value of the evaluation index is optimal Number, and according to the k-nearest neighbor number when the mean value of the selected evaluation index is cross-validated, the coal moisture content detection model based on DW-KNN is constructed.

When the coal moisture content detection model based on DW-KNN processes the microwave S-parameter spectrum information of coal samples to obtain the moisture content of coal samples, the following steps are included:

Step 2.1. Calculate the Euclidean distance between the microwave S-parameter spectrum information of the coal sample and the qth coal training sample δ _q

为煤炭标本微波S参数波谱信息第p个频率点相应的波谱数据，δ_qp为第q个煤炭训练样本δ_q的煤炭训练样本微波S参数波谱信息第p个频率点相应的波谱数据；in,

is the spectral data corresponding to the pth frequency point of the coal sample microwave S-parameter spectral information, δ _qp is the corresponding spectral data of the p-th frequency point of the coal training sample microwave S-parameter spectral information of the qth coal training sample δ _q ;

按升序排序，确定与前k个欧式距离分别对应的煤炭训练样本；Step 2.2, all the calculated Euclidean distances

Sort in ascending order to determine the coal training samples corresponding to the first k Euclidean distances;

以及相应的煤炭训练样本，计算得到并输出所述煤炭标本的水分含量为：Step 2.3, the first k Euclidean distances determined according to the selection

And the corresponding coal training sample, calculate and output the moisture content of the described coal sample as:

为煤炭标本的水分含量，ψ(δ_j)为确定前k个煤炭训练样本中第j个煤炭训练样本δ_j的煤炭训练样本水分含量标签，σ_j为确定前k个煤炭训练样本中第j个煤炭训练样本δ_j的距离权重，

为确定的前k个欧式距离内相应的第j个欧氏距离。in,

is the moisture content of the coal sample, ψ(δ _j ) is the moisture content label of the jth coal training sample δ _j in the first k coal training samples, and σ _j is the jth coal training sample j in the first k coal training samples. The distance weight of coal training samples δ _j ,

is the corresponding j-th Euclidean distance within the determined first k Euclidean distances.

In step 1.1, use a sieve with 13mm sieve holes to sieve the raw coal sample, and for the coal samples to be tested for moisture content, you also need to use a 13mm sieve sieve to sieve the coal samples, and obtain the sieved coal samples The microwave S-parameter spectrum information of the coal sample is consistent with the microwave S-parameter spectrum information of the coal training sample.

It also includes a microwave test system for microwave testing, the microwave test system includes a vector network analyzer and two horn antennas adapted to the vector network analyzer, and is placed between the two horn antennas that are placed correspondingly. A sample container for accommodating coal samples or coal samples, and the vector network analyzer is electrically connected to the main testing controller.

Advantages of the present invention: build a coal moisture content detection model based on DW-KNN, obtain the coal sample microwave S parameter spectrum information consistent with the coal training sample microwave S parameter spectrum information, use the above-mentioned DW-KNN based coal The moisture content detection model processes the acquired microwave S-parameter spectrum information of the coal sample to obtain and output the moisture content of the coal sample, that is, to realize the detection of the coal moisture content without damage, and to improve the detection accuracy and robust performance.

Description of drawings

Fig. 1 is a flowchart of the present invention.

Fig. 2 is a schematic diagram of the microwave test system of the present invention.

Explanation of reference numerals: 1—master test controller, 2—vector network analyzer, 3—horn antenna, and 4—sample container.

Detailed ways

The present invention will be further described below in conjunction with specific drawings and embodiments.

As shown in Figure 1: in order to realize the detection of coal moisture content under the nondestructive situation, improve detection accuracy and robust performance, the coal moisture content rapid nondestructive detection method of the present invention comprises the following steps:

Step 1. Construct a coal moisture content detection model based on DW-KNN, wherein, when constructing a coal moisture content detection model based on DW-KNN, the training data set produced includes the coal training sample moisture content label of the coal training sample and the coal training sample In the coal training sample microwave S parameter spectrum information under the coal training sample moisture content label, utilize the training data set made to DW-KNN model training, to construct the coal moisture content detection model based on DW-KNN;

Specifically, the DW-KNN (Distance Weighted K-NearestNeighbor) model is a distance-weighted K-nearest neighbor algorithm, which is an existing commonly used model. familiar. The DW-KNN model is used to construct the coal moisture content detection model, that is, the coal moisture content detection model based on DW-KNN is obtained. When using the DW-KNN model to build a DW-KNN-based coal moisture content detection model, it is necessary to make a training data set, and use the prepared training data set to train the DW-KNN model, and obtain the coal moisture content based on DW-KNN during training. Moisture content detection model.

For the training data set, the coal training sample moisture content label of the coal training sample and the coal training sample microwave S parameter spectrum information of the coal training sample under the coal training sample moisture content label, the coal training sample moisture content of the coal training sample The label is the current moisture content value of the coal training sample. The coal training sample microwave S parameter spectrum information of the coal training sample under the coal training sample moisture content label specifically refers to the coal training sample microwave S parameter spectrum information of the coal training sample and the coal training sample moisture content of the coal training sample Labels are in one-to-one correspondence.

For the production process of the training data set and the specific process of constructing the coal moisture content detection model based on DW-KNN, the following expressions are used for specific description.

During the specific implementation, when making the training data set, it specifically includes the following steps:

Step 1.1. Collect raw coal samples from different areas, and screen all the collected raw coal samples, spread the screened raw coal samples evenly in the corresponding carrier, and dry the raw coal samples in the carrier, To obtain experimental coal samples in different regions;

In the embodiment of the present invention, the area of the raw coal sample can be selected according to actual needs, and will not be repeated here. Generally, raw coal samples can be collected manually. Raw coal samples collected from different regions are sieved through 13mm sieves, and then spread evenly on high-temperature resistant aluminum pans after sieving; put the aluminum pans into a blast drying oven to dry at 108°C After 2 hours, the experimental coal sample was obtained; that is, the above-mentioned carrier can be a high-temperature-resistant aluminum plate. Of course, the carrier can also use other forms, which can be selected according to needs, and will not be listed here.

During specific implementation, take 500±10g of the raw coal sample after screening, and evenly spread it on the high-temperature-resistant aluminum pan, and the thickness of the raw coal sample should not exceed 2 times the maximum particle diameter.

Step 1.2. For the above-mentioned experimental coal samples obtained in different regions, they are all sequentially processed by sealing and cooling, spraying water and stirring, and sealing and standing to obtain the coal training samples under the moisture content label of the preset coal training samples;

In the embodiment of the present invention, after the experimental coal sample is taken out from the blast drying oven, it is immediately put into a bucket with a sealed cover for sealing and cooling treatment. At the same time, the mass of the dry coal sample is weighed at this time for subsequent calculation. After the coal sample is cooled to normal temperature, open the sealing cover and spray a certain amount of water on the coal sample, and at the same time slowly stir the coal sample to fully mix the coal and water; seal the coal sample again and place it in room temperature for 6 hours Above, make the moisture fully diffuse in the coal, and finally get the coal training samples.

During specific implementation, all experimental processes were carried out at room temperature; the quality of the dry coal sample was obtained by weighing the total weight of the dry coal and the barrel minus the net weight of the barrel; During the procedure, minimize the contact time between the coal sample and the air, and seal it in time to prevent the moisture in the air from interfering with the experimental results; the technical means commonly used in this technical field can be used to perform the process of spraying and stirring the experimental coal sample , which can be selected according to actual needs.

In the embodiment of the present invention, when water is sprayed and stirred, for any experimental coal sample, the water spray quality determined according to the moisture content label of the preset coal training sample is:

Among them, M _add is the mass of water injection, M _c&m is the mass of experimental coal sample mixed with water, MC _before is the moisture content of experimental coal sample, and MC _after is the coal sample moisture content label of coal training sample.

During specific implementation, a coal training sample moisture content label can be preset according to actual needs, and the preset coal training sample moisture content label can be reasonably set according to the situation of the raw coal sample. After setting the coal training sample moisture content label, it can be determined. Water quality, so that the required coal training samples and the coal training sample moisture content labels of the produced coal training samples can be prepared finally. Since it is difficult to control the quality of the sprayed water in the actual experiment, the moisture content of the new coal sample can be recalculated according to the quality difference before and after the water spraying to correct the error.

When spraying water, the moisture content label of the preset coal training sample can be finally reached through multiple water spraying operations. When the water spraying operation is performed multiple times, the moisture content MC _before of the experimental coal sample is the moisture content after the last water spraying. For example, after sealing and cooling, when the first water spraying is stirred, the moisture content MC _before of the experimental coal sample is A value that is 0 or close to 0. After the first water spraying and before the second water spraying, it is possible to obtain the moisture content MC _before of the corresponding coal inspection sample, and the rest of the cases can be deduced by analogy, which will not be repeated here.

Step 1.3: Perform a microwave test on the coal training samples obtained above to obtain the coal sample microwave S-parameter spectrum information of the coal training samples in the corresponding area under the moisture content label of the coal training samples.

Specifically, it also includes a microwave testing system for microwave testing, and utilizes the microwave testing system to realize the microwave testing of coal training samples. A specific implementation of the microwave testing system is shown in Figure 2, specifically: the microwave testing The system includes a vector network analyzer 2 and two horn antennas 3 adapted to the vector network analyzer 2, and a sample container 4 for accommodating coal samples or coal samples is placed between the two horn antennas 3 that are placed correspondingly, The vector network analyzer 2 is electrically connected to the test master controller 1 .

Specifically, the main test controller 1 can use an existing commonly used computer, which is mainly used to control the vector network analyzer 2 to send and receive microwave signals and to visualize microwave S-parameter spectrum information. The main test controller 1 communicates with the vector network through a data transmission line. Analyzer 2 is connected, and described vector network analyzer 2 can adopt existing commonly used portable analyzer, and the specific situation of vector network analyzer 2 is consistent with existing, is well known to those skilled in the art, repeats no more here. The two signal ports of the vector network analyzer 2 are respectively connected to the rectangular horn antenna 3 through a coaxial cable. The two horn antennas 3 are placed opposite to each other and installed on a horizontal linear slide rail. The mouth spacing is 10cm, the horn antenna 3 can adopt the existing commonly used form, and the specific connection and cooperation form of the horn antenna 3 and the vector network analyzer 2 is consistent with the existing ones.

Coal samples can be accommodated through the sample container 4, and the sample container 4 is placed in the center area of the central axis of the horn antenna 3. In the embodiment of the present invention, the microwave signals of 8.05-12.01 GHz frequency transmitted and received by the vector network analyzer 2 include 133 frequencies in total Points, through the 133 frequency points of the vector network analyzer 2, 133 spectrum data can be obtained. The horn antenna 3 chooses the EIA standard WR90 waveguide horn antenna with a gain of 20 dB, and the sample container 4 chooses a rectangular container made of PMMA material with a thickness of 3 mm.

During specific implementation, during the microwave test of the coal training sample, for any coal training sample, the spectrum data of the corresponding frequency point at the frequency of 8.05GHz to 12.01GHz is collected to obtain the coal sample microwave S-parameter spectrum of the coal training sample information, wherein all coal training sample microwave S-parameter spectrum data of a coal training sample constitute an initial microwave S-parameter spectrum data.

It can be known from the above description that for any coal training sample, under the action of 133 frequency points, 133 spectral data can be obtained, that is, microwave S-parameter spectral data of 133 coal training samples are included. Therefore, using 133 coal training samples The microwave S-parameter spectrum data of coal training samples can constitute an initial curve of microwave S-parameter spectrum. During specific implementation, the spectrum data at a frequency point specifically refers to the insertion loss parameter S21, wherein the specific situation of the insertion loss parameter S21 is consistent with the existing ones, and will not be repeated here.

During specific implementation, for coal training samples collected in different regions or under the moisture content label of different coal training samples, after microwave testing, an initial microwave S-parameter spectrum curve can be generated. The total number of microwave S-parameter spectrum initial curves can be determined according to actual needs. Choose OK.

Further, the microwave S-parameter spectrum initial curves of all coal training samples from the same area are preprocessed to obtain the coal sample microwave S-parameter spectrum information of the coal training sample under the moisture content label of the coal training sample after preprocessing , where the preprocessing includes the following steps:

Step 1.3.1, carry out linear fitting to any microwave S-parameter spectrum initial curve, and calculate the spectrum data distance corresponding to the linear fitting straight line between the microwave S-parameter spectrum initial curve and the microwave S-parameter spectrum initial curve, obtain the spectrum data A linear fit to the distance matrix D,

Among them, A[m-1][i] is the spectrum data of the i-th frequency point of the initial curve of the m-1 microwave S-parameter spectrum, and A*[m-1][i] is the m-1 microwave S The fitting spectrum data of the i-th frequency point of the linear fitting line corresponding to the initial curve of the parametric spectrum, n is the number of corresponding frequency points of each microwave S-parameter spectrum initial curve obtained by testing, and m is the bar of the microwave S-parameter spectrum initial curve number;

Step 1.3.2, compare the spectral data linear fitting distance in the spectral data linear fitting distance matrix D with the preset distance threshold, and eliminate the initial curve of the microwave S-parameter spectrum greater than the distance threshold to obtain the current area Corresponding coal sample microwave S-parameter spectrum curve;

The microwave S-parameter spectrum curves of coal samples in all regions constitute the microwave S-parameter spectrum information of coal training samples in the training data set.

Specifically, any initial curve of the microwave S-parameter spectrum can be linearly fitted using commonly used technical means in this technical field, and the specific linear fitting method can be selected according to needs, and will not be repeated here. After the linear fitting, a linear fitting straight line corresponding to the initial curve of a microwave S-parameter spectrum can be obtained. It can be seen from the above description that each frequency point corresponds to a spectral data, therefore, after obtaining the linear fitting line, any frequency point can determine the spectral data on the initial curve of the microwave S-parameter spectrum and the corresponding spectral data on the linear fitting straight line , through all the spectral data on the initial curve of the microwave S-parameter spectrum and the corresponding spectral data on the linear fitting line, the linear fitting distance matrix D of the spectral data can be obtained.

In the embodiment of the present invention, according to the above description, it can be seen that the number n of frequency points on each initial microwave S-parameter spectrum curve is 133, and the number m of the initial microwave S-parameter spectrum curve can be selected according to actual needs. For details, refer to the above illustrate.

According to the above spectral data linear fitting distance matrix D, the spectral data linear fitting distance corresponding to any coal training sample can be compared with the preset distance threshold, and the microwave S parameter spectral initial curve greater than the distance threshold can be eliminated , to obtain the required microwave S-parameter spectrum information of the coal sample. During specific implementation, the preset distance threshold is 1.5×dis_m, wherein dis_m is the median distance after sorting the spectral data linear fitting distances in the region where the current raw coal sample is located.

In specific implementation, when the raw coal samples correspond to multiple collection areas, the initial curves of microwave S-parameter spectra of all coal training samples in each area need to be pre-processed, so that after the pre-processing is completed, the coal samples in all areas are microwaved The S-parameter spectrum curve constitutes the coal sample microwave S-parameter spectrum information, wherein the coal sample microwave S-parameter spectrum information is specifically a plurality of coal sample microwave S-parameter spectrum curves.

Further, the training data set made is divided according to the collection area of the coal training samples to obtain multiple sets of training data subsets;

Input the training data subset into the DW-KNN model with k-nearest neighbors according to the leave-one-out cross-validation method, and calculate the cross-validation mean value of the evaluation index at different k-nearest neighbor numbers, and select the k-nearest neighbor when the cross-validation mean value of the evaluation index is optimal Number, and according to the k-nearest neighbor number when the mean value of the selected evaluation index is cross-validated, the coal moisture content detection model based on DW-KNN is constructed.

In the embodiment of the present invention, since the raw coal samples are collected in different areas, after the training data set is produced through the above steps, the training data set is divided according to the collection area of the coal training samples to obtain multiple sets of training data subsets. The leave-one-out cross-validation method is specifically: take out any group of training data subsets divided by the collection area of the coal training samples as the leave-one-out cross-validation test data set, and the remaining groups as the leave-one-out cross-validation training data set, and leave The specific implementation and process of a cross-validation method are consistent with the existing ones and are well known to those skilled in the art, and will not be repeated here.

The training data subset is input into the DW-KNN model when the number of k nearest neighbors is according to the leave-one-out cross-validation method commonly used in the art, and the cross-validation mean value of the evaluation index when different k nearest neighbors are calculated respectively, and the evaluation index includes the coefficient of determination R ² , mean absolute error (MAE: Mean Absolute Error) and root mean square error (Root Mean Squared Error, RMSE), specifically, the number of k-nearest neighbors starts from small and gradually increases, respectively calculated and recorded in different k-nearest neighbors When the cross-validation mean value of the evaluation index is counted; when the evaluation index no longer has a large change, the operation is stopped, and the k-nearest neighbor number when the cross-validation mean value of the evaluation index is optimal is selected as the best choice of the model, and the coal based on DW-KNN is obtained. Moisture content detection model. During specific implementation, when the average value of the evaluation index cross-validation is optimal, generally, the value of the determination coefficient ^R2 is the largest, and it can be determined according to the value situation of the determination coefficient ^R2 whether it is when the average value of the evaluation index cross-validation is optimal; of course, It is also possible to use technical means to determine the optimal cross-validation mean value of the evaluation index, and the specific method and process for determining the optimal cross-validation average value of the evaluation index index are well known to those skilled in the art.

In the embodiment of the present invention, the number of k-nearest neighbors starts from 3, and is incremented by 1 each time until the evaluation index no longer changes significantly, and the calculation is stopped. The specific process is consistent with the existing ones and is well known to those skilled in the art.

Step 2, provide the coal sample with the moisture content to be detected, obtain the microwave S parameter spectrum information of the coal sample consistent with the microwave S parameter spectrum information of the coal training sample, and use the above-mentioned coal moisture content detection model based on DW-KNN to The obtained coal sample microwave S-parameter spectrum information is processed to obtain and output the moisture content of the coal sample.

During specific implementation, the microwave S parameter spectrum information of the coal sample that is consistent with the microwave S parameter spectrum information of the coal training sample is obtained, specifically referring to the coal sample that provides the moisture content to be detected, it is necessary to use a sieve with a 13mm sieve hole to The coal sample is screened, and the coal sample microwave S parameter spectral information of the coal sample is consistent with the coal training sample microwave S parameter spectral information for the coal sample after screening; and the coal sample microwave S parameter spectral information is consistent with the coal training sample microwave The S-parameter spectrum information is consistent, which means that the microwave S-parameter spectrum information of the coal sample and the microwave S-parameter spectrum information of the coal training sample are all tested at the same frequency point to obtain the corresponding spectrum data. When the number n of corresponding frequency points on the curve is 133, the microwave S-parameter spectrum information of the coal sample and the microwave S-parameter spectrum information of the coal training sample both include spectrum data corresponding to 133 frequency points.

When the coal moisture content detection model based on DW-KNN processes the microwave S-parameter spectrum information of coal samples to obtain the moisture content of coal samples, the following steps are included:

Step 2.1. Calculate the Euclidean distance between the microwave S-parameter spectrum information of the coal sample and the qth coal training sample δ _q

为煤炭标本微波S参数波谱信息第p个频率点相应的波谱数据，δ_qp为第q个煤炭训练样本δ_q的煤炭训练样本微波S参数波谱信息第p个频率点相应的波谱数据；in,

is the spectral data corresponding to the pth frequency point of the coal sample microwave S-parameter spectral information, δ _qp is the corresponding spectral data of the p-th frequency point of the coal training sample microwave S-parameter spectral information of the qth coal training sample δ _q ;

Specifically, when the total number of coal sample microwave S-parameter spectrum curves contained in the coal sample microwave S-parameter spectrum information is Q, the value range of q is 1-Q.

按升序排序，确定与前k个欧式距离分别对应的煤炭训练样本；Step 2.2, all the calculated Euclidean distances

Sort in ascending order to determine the coal training samples corresponding to the first k Euclidean distances;

对所述欧式距离

按升序排序后，根据上述确定的k近邻数，则选取前k个欧式距离以及与前k个欧式距离相对应的煤炭训练样本，在确定前k个煤炭训练样本后，能确定每个煤炭训练样本的煤炭训练样本水分含量标签以及煤炭训练样本在所述煤炭训练样本水分含量标签下的煤炭训练样本微波S参数波谱信息。Specifically, through step 1, Q Euclidean distances can be determined

For the Euclidean distance

After sorting in ascending order, according to the number of k neighbors determined above, select the first k Euclidean distances and the coal training samples corresponding to the first k Euclidean distances. After determining the first k coal training samples, each coal training sample can be determined The coal training sample moisture content label of the sample and the coal training sample microwave S-parameter spectrum information of the coal training sample under the coal training sample moisture content label.

以及相应的煤炭训练样本，计算得到并输出所述煤炭标本的水分含量为：Step 2.3, the first k Euclidean distances determined according to the selection

And the corresponding coal training sample, calculate and output the moisture content of the described coal sample as:

为煤炭标本的水分含量，ψ(δ_j)为确定前k个煤炭训练样本中第j个煤炭训练样本δ_j的煤炭训练样本水分含量标签，σ_j为确定前k个煤炭训练样本中第j个煤炭训练样本δ_j的距离权重，

为确定的前k个欧式距离内相应的第j个欧氏距离。in,

is the moisture content of the coal sample, ψ(δ _j ) is the moisture content label of the jth coal training sample δ _j in the first k coal training samples, and σ _j is the jth coal training sample j in the first k coal training samples. The distance weight of coal training samples δ _j ,

is the corresponding j-th Euclidean distance within the determined first k Euclidean distances.

In the embodiment of the present invention, the moisture content of the coal sample can be obtained by performing the above processing on the microwave S-parameter spectrum information of the coal sample.

Claims (5)

1. A DW-KNN-based coal moisture content rapid nondestructive testing method is characterized by comprising the following steps:

step 1, constructing a DW-KNN-based coal moisture content detection model, wherein when the DW-KNN-based coal moisture content detection model is constructed, a manufactured training data set comprises a coal training sample moisture content label of a coal training sample and coal training sample microwave S parameter spectrum information of the coal training sample under the coal training sample moisture content label, and the manufactured training data set is used for training the DW-KNN model to construct and obtain the DW-KNN-based coal moisture content detection model;

step 2, providing a coal sample to be detected for moisture content, acquiring coal sample microwave S parameter spectrum information of which the coal sample is consistent with coal training sample microwave S parameter spectrum information, and processing the acquired coal sample microwave S parameter spectrum information by using the DW-KNN-based coal moisture content detection model to obtain and output the moisture content of the coal sample;

in step 1, when a training data set is manufactured, the method specifically comprises the following steps:

step 1.1, collecting raw coal samples in different areas, screening all the collected raw coal samples, uniformly and flatly paving the screened raw coal samples in corresponding carriers, and drying the raw coal samples in the carriers to obtain experimental coal samples in different areas;

step 1.2, carrying out sealing cooling, water spraying stirring and sealing standing treatment on the obtained experimental coal samples in different areas in sequence to obtain coal training samples under a preset coal training sample moisture content label;

step 1.3, performing microwave testing on the obtained coal training samples to obtain coal sample microwave S parameter spectrum information of the coal training samples in corresponding areas under the coal training sample moisture content label;

in the step 1.3, when the coal training samples are subjected to microwave testing, spectrum data of corresponding frequency points under the frequency of 8.05 GHz-12.01 GHz are acquired and obtained for any coal training sample to obtain coal sample microwave S parameter spectrum information of the coal training sample, wherein all coal sample microwave S parameter spectrum data of one coal training sample form a microwave S parameter spectrum initial curve;

preprocessing microwave S parameter spectrum initial curves of all coal training samples from the same area to obtain coal sample microwave S parameter spectrum information of the coal training samples under a coal training sample moisture content label after preprocessing, wherein the preprocessing comprises the following steps:

step 1.3.1, performing linear fitting on any microwave S parameter spectrum initial curve, calculating the corresponding spectrum data distance between the microwave S parameter spectrum initial curve and the linear fitting straight line of the microwave S parameter spectrum initial curve to obtain a spectrum data linear fitting distance matrix D,

wherein, A [ m-1] [ i ] is the spectral data of the ith frequency point of the m-1 th microwave S parameter spectrum initial curve, A [ m-1] [ i ] is the fitted spectral data of the ith frequency point of a linear fitted straight line corresponding to the m-1 th microwave S parameter spectrum initial curve, n is the number of the frequency points corresponding to each microwave S parameter spectrum initial curve obtained by testing, and m is the number of the microwave S parameter spectrum initial curves;

step 1.3.2, comparing the linear fitting distance of the spectrum data in the linear fitting distance matrix D of the spectrum data with a preset distance threshold, and eliminating a microwave S parameter spectrum initial curve which is greater than the distance threshold to obtain a coal sample microwave S parameter spectrum curve corresponding to the current region;

the microwave S parameter spectrum curves of the coal samples in all the areas form the microwave S parameter spectrum information of the coal training samples of the training data set;

dividing the manufactured training data set according to the acquisition region of the coal training sample to obtain a plurality of groups of training data subsets;

inputting a training data subset into a DW-KNN model when k neighbors exist according to a leave-one-out cross validation method, respectively calculating cross validation mean values of evaluation indexes when the k neighbors exist at different numbers, selecting the k neighbors when the cross validation mean value of the evaluation indexes is optimal, and constructing and obtaining a DW-KNN-based coal moisture content detection model according to the k neighbors when the selected evaluation index cross validation mean value is optimal;

when the DW-KNN-based coal moisture content detection model processes microwave S parameter spectrum information of a coal specimen to obtain the moisture content of the coal specimen, the DW-KNN-based coal moisture content detection method comprises the following steps:

step 2.1, calculating microwave S parameter spectrum information of the coal sample and a qth coal training sample delta _q Euclidean distance of

Wherein,

spectral data corresponding to the p-th frequency point of the microwave S parameter spectral information of the coal sample delta _qp For the qth coal training sample δ _q Spectrum data corresponding to the p-th frequency point of the microwave S parameter spectrum information of the coal training sample;

step 2.2, all the Euclidean distances obtained by calculation

Sequencing in an ascending order, and determining coal training samples respectively corresponding to the first k Euclidean distances;

step 2.3, according to the first k Euclidean distances determined by selection

And corresponding coal training samples, and calculating and outputting the moisture content of the coal specimen as follows:

wherein,

moisture content of coal specimen, psi (delta) _j ) To determine the jth coal training sample delta in the first k coal training samples _j Coal training sample moisture content tag, σ _j To determine the jth coal training sample delta in the first k coal training samples _j The weight of the distance of (a) is,

is the corresponding j-th Euclidean distance in the determined first k Euclidean distances.

2. The DW-KNN-based coal moisture content rapid nondestructive testing method as claimed in claim 1, wherein in step 1.2, during water spraying and stirring, the water spraying quality determined according to the moisture content label of the preset coal training sample for any experimental coal sample is as follows:

wherein M is _add Is the mass of the water spray, M _c&m Is the mass of the mixture of the experimental coal sample and water, MC _before Is the moisture content, MC, of the experimental coal sample _after Is a coal sample moisture content label of a coal training sample.

3. The DW-KNN-based coal moisture content rapid nondestructive testing method according to claim 1, wherein the preset distance threshold is 1.5 x dis _ m, wherein dis _ m is a distance median after linear fitting distance sorting of spectral data of a region where the current raw coal sample is located.

4. The DW-KNN-based coal moisture content rapid nondestructive testing method as claimed in claim 1, wherein in step 1.1, a 13mm sieve is used for sieving the raw coal sample, and for the coal sample providing the moisture content to be detected, a 13mm sieve is also used for sieving the coal sample, and for the sieved coal sample, the coal sample microwave S parameter spectrum information of the coal sample is obtained, wherein the coal sample is consistent with the coal training sample microwave S parameter spectrum information.

5. The DW-KNN-based coal moisture content rapid nondestructive testing method according to claim 1 or 3, characterized by further comprising a microwave testing system for microwave testing, wherein the microwave testing system comprises a vector network analyzer (2) and two horn antennas (3) matched with the vector network analyzer (2), a sample container (4) for containing a coal sample or a coal sample is arranged between the two horn antennas (3) which are arranged in a positive correspondence manner, and the vector network analyzer (2) is electrically connected with the testing main controller (1).

Images