A method for predicting moisture content of solid waste based on infrared spectroscopy and machine learning

Through Fourier transform attenuated total reflection infrared spectroscopy technology and machine learning algorithm, the accuracy and efficiency of moisture content detection of multi-source solid waste is solved, and high-precision and fast non-destructive detection is achieved, suitable for multi-source in complex environments Solid waste.

CN115062265BActive Publication Date: 2024-11-22TONGJI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210552755.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-05-19
Publication Date
2024-11-22
Estimated Expiration
2042-05-19

AI Technical Summary

Technical Problem

The prior art is difficult to quickly and accurately predict the moisture content of multi-source solid waste, especially in complex environments. Traditional spectroscopy technology lacks efficient and accurate spectral data preprocessing and feature spectrum extraction, resulting in low model accuracy and inability to be applicable to Multiple sources of solid waste.

Method used

Fourier transform attenuated total reflection infrared spectroscopy technology combined with machine learning, through spectral data preprocessing and feature spectrum extraction, a multi-source solid waste moisture content regression prediction model is established, and a random forest regression, support vector regression and other algorithms are used to predict.

Benefits of technology

It realizes high-precision and rapid non-destructive testing of multi-source solid waste, improves the level of inspection automation, simplifies the operation process, reduces the detection time, and is suitable for online inspection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115062265B_ABST
    Figure CN115062265B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for predicting the moisture content of solid waste based on infrared spectroscopy and machine learning, comprising the following steps: collecting typical multi-source solid waste, using Fourier transform attenuated total reflection infrared spectroscopy technology to obtain solid waste spectral data; establishing a multi-source solid waste moisture content prediction model through spectral data preprocessing and characteristic spectrum extraction combined with machine learning algorithms; using the constructed model and the infrared spectrum of the solid waste sample to be tested to predict the moisture content of the solid waste. Compared with the traditional destructive measurement method of weighing before and after drying, the present invention does not require sample crushing, drying and weighing operations, can non-destructively predict the moisture content of multi-source solid waste, and has a faster detection speed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a rapid prediction technology for the moisture content of solid waste, and in particular to a method for predicting the moisture content of solid waste based on infrared spectroscopy and machine learning. Background Art

[0002] Moisture content is one of the important indicators of solid waste treatment and disposal, which determines the generation of leachate during the storage, transportation, biological treatment and landfill disposal of solid waste. Leachate is organic wastewater containing complex components such as high concentration of organic matter, heavy metal salts, suspended solids, etc. Once leaked, it will not only pollute the soil and surface water sources, but also pollute groundwater. In addition, the moisture content of solid waste also has a significant impact on the biological treatment and thermal treatment processes. Water, as a carrier, plays a key role in the metabolism of microorganisms. It can not only dissolve organic matter to facilitate microbial metabolism, but also be used to adjust the temperature of the reactor to achieve the best disposal conditions. During incineration treatment, due to the complex composition of solid waste and the large differences in physical and chemical properties, the calorific value of the garbage entering the furnace will fluctuate greatly. If the moisture content of the garbage entering the furnace is too high, it will not only affect the stable operation of the incinerator, but also increase the emission of pollutants, such as causing incomplete incineration, increasing the CO concentration in the flue gas, and promoting the generation of residual carbon, thereby creating conditions for the production of dioxins and polluting the atmospheric environment. At the same time, the high moisture content and low calorific value of solid waste will prolong the drying time during incineration, increase energy consumption and reduce efficiency. Therefore, moisture content is an important measurement indicator that helps to evaluate the treatability of solid waste and predict possible changes during treatment and disposal.

[0003] At present, the methods for determining moisture content can be mainly divided into direct method and indirect method. The direct measurement method mainly uses oven or vacuum drying to obtain the moisture content by mass difference subtraction. This method is time-consuming and will change the properties of volatile and heat-sensitive compounds in the sample. In order to solve the problems brought by the direct measurement method, a variety of indirect measurement methods have been developed, such as X-ray scattering, microwave, new sensors, infrared, near-infrared, laser-induced breakdown spectroscopy, visible light and other technologies. Compared with the direct measurement method, the indirect measurement method has the advantages of fast response speed and saving manpower and material resources. However, X-ray scattering, microwave and new sensor technology are not suitable for processing complex environmental samples due to their high requirements for the measurement environment, and their application in the field of solid waste has not yet been confirmed. Infrared, near-infrared, laser-induced breakdown spectroscopy, visible light and near-infrared methods are all spectral technologies. Due to their strong classification ability, spectral technology has been widely used in fine classification and identification work. Infrared spectra have fingerprint characteristics. The different absorption degrees of infrared light by samples determine the distribution position and intensity of functional groups, that is, the position and peak area of ​​the characteristic peaks corresponding to the functional groups, and the type and content of functional groups determine the properties of the samples to a certain extent. At the same time, since water molecule functional groups appear in any water-containing substance, it can be assumed that the water content of the sample is related to the absorption peak of water molecules, so that the water content of different samples can be predicted based on the differences in these absorption peaks.

[0004] After obtaining infrared spectrum data, it is necessary to establish a mathematical model between infrared spectrum and moisture content to obtain the final result. At present, the commonly used methods for establishing mathematical models are principal component regression (PCR) and partial least squares regression (PLS), such as:

[0005] (1) "A method for detecting moisture content of dry powder fire extinguishing agent based on near-infrared spectroscopy analysis" (CN103175806B) uses the partial least squares method to establish a calibration model for the functional relationship between the near-infrared spectrum of the calibration set and the true value of the moisture content of the corresponding dry powder fire extinguishing agent sample;

[0006] (2) "Method for Establishing an Infrared Spectrum Measurement Model for the Moisture Content of Strip Surface Coating" (CN103837491B) uses regression calculation on multiple measurement values ​​to obtain one or more regression linear curves of the infrared spectrum measurement of the moisture content of the strip surface coating, and verifies the infrared spectrum measurement accuracy of the curve to obtain an infrared spectrum measurement model that can reflect the moisture content of the strip surface coating;

[0007] (3) "Wood moisture content detection method based on infrared difference spectrum technology" (CN104390932B) uses the least squares linear regression method to construct a model for detecting wood moisture content using infrared difference spectrum technology by using the moisture content W and area AW-0 of each sample.

[0008] There are also those that use empirical formulas as mathematical models, such as: "Device and method for measuring soil nitrogen and moisture content" (CN104990900B), "A circuit, device and method for rapid nondestructive detection of rice leaf moisture content based on near-infrared spectroscopy" (CN108872141B), and "A method for exploring the optimal soil moisture content for detecting soil nitrogen using near-infrared spectroscopy" (CN108982407A).

[0009] However, these methods are not ideal in terms of applicable occasions and nonlinear fitting accuracy. Therefore, methods using deep learning to establish mathematical models have been developed on this basis, such as:

[0010] (1) "A method and instrument for measuring water content in oil products based on deep learning" (CN108982405B) obtains an infrared spectrum characteristic curve through an infrared spectrum acquisition module, and analyzes the infrared spectrum characteristic curve through an artificial intelligence analysis algorithm based on deep learning to obtain the water content of the oil product to be tested;

[0011] (2) “A nondestructive detection method for moisture content of fresh cocoons based on spectral imaging and deep learning technology” (CN110567889A) uses visible / near-infrared spectral data of cocoon layer moisture content and moisture content information to train the Xception-1D network model and then use the model to predict the moisture content of fresh cocoons.

[0012] However, the above methods are aimed at objects with relatively single components and cannot be directly applied to multi-source solid waste. In addition, the above two methods lack efficient and accurate spectral data preprocessing and feature spectrum extraction process. The collected spectral data are directly used to train the model after simple signal processing or smoothing based on roughness penalty. The effects of environmental noise, baseline offset, and light scattering on the quality of spectral data are ignored, and the redundant and collinear information in the spectral data is not processed, which will not only increase the amount of model data processing, but also reduce the accuracy of the model.

[0013] Multi-source solid wastes are of various types and complex compositions. Although spectroscopy can obtain a large amount of spectral information of water-containing samples, the results include not only the information after the superposition of various groups and substances, but also noise and various external interference information. In addition, the establishment of a moisture content prediction model requires the rapid acquisition of features related to the functional groups of water molecules from the spectral information. Faced with a large number of samples, it is unrealistic to rely solely on manual selection or simple data analysis methods. Machine learning algorithms have better performance than humans in detecting small changes in spectral data. Analysis combined with machine learning algorithms can realize the visualization of spectral data, highlight its identifiable features, and provide conditions for identification and classification tasks. However, due to the complexity of multi-source solid wastes and spectral data, there are currently few studies on quantitative prediction of solid waste properties. Therefore, non-destructive and rapid prediction of the moisture content of multi-source solid wastes has become a key demand technology. Summary of the invention

[0014] In response to the above-mentioned problems existing in the prior art, the present application provides a method for predicting the moisture content of solid waste based on infrared spectroscopy and machine learning, which can realize non-destructive and rapid moisture content detection of multi-source solid waste with a wide variety and complex composition.

[0015] The technical solution of the present invention is as follows:

[0016] 1. A method for predicting moisture content of solid waste based on infrared spectroscopy and machine learning, comprising the following steps:

[0017] S1. Collect typical multi-source solid waste samples and directly use Fourier transform attenuated total reflection infrared spectroscopy technology to obtain sample infrared spectrum data; classified by source, the typical solid waste samples include but are not limited to: domestic waste, construction waste, general industrial solid waste and agricultural solid waste; classified by type, the typical multi-source solid waste samples include but are not limited to: kitchen waste, paper, fabric, leather, plastic, rubber, wood and bamboo, metal, glass, brick, tile and ceramic, and ash and soil;

[0018] S2. Use an oven to dry the sample and then test its moisture content; the oven temperature range is 100-110°C;

[0019] S3, using spectral data preprocessing and feature spectrum extraction, and dividing the processed infrared spectrum data into a training set and a test set;

[0020] S4. Establishing a multi-source solid waste moisture content regression prediction model according to the moisture content of solid waste in the training set and its spectral information;

[0021] S5. Use the test set to test the model accuracy and further optimize the multi-source solid waste moisture content prediction model;

[0022] S6. Collect the solid waste samples to be tested, directly use Fourier transform attenuated total reflection infrared spectroscopy technology to obtain the infrared spectrum data of the samples, and use the constructed multi-source solid waste moisture content prediction model to predict the moisture content of the samples.

[0023] Furthermore, when collecting various types of multi-source solid waste components in step S1, the number of representative samples of each component is not less than 10.

[0024] Furthermore, the moisture content of each type of solid waste sample collected in step S1 ranges from 0.50% to 90.00%.

[0025] Furthermore, the collection range of the Fourier transform attenuated total reflection infrared spectrum in step 1 is 400 to 4000 cm -1 , the resolution is 1~8cm -1 ; The number of scans for a single sample was 8 to 16 times, and the average spectrum was taken to construct the machine learning model.

[0026] Furthermore, the spectral data preprocessing method in step S3 includes but is not limited to: ATR correction, first-order derivative, Savitzky–Golay filtering, and first-order derivative combined with Savitzky–Golay filtering; the spectral data preprocessing method is used to sharpen the spectral data information.

[0027] Furthermore, the characteristic spectrum selection in step S3 is to select two infrared spectrum bands related to water molecules as characteristic bands, namely 1500-1800cm -1 and 3000~3600cm -1 .

[0028] Furthermore, the training set and test set division method described in step S3 is based on a Random function to divide the data of each sample into a training set and a test set, wherein the number of training sets accounts for 60% to 80% of all sample data, and the number of test sets accounts for 20% to 40% of all sample data.

[0029] Furthermore, the machine learning model for predicting the moisture content of solid waste in step S4 is obtained by one of the following algorithms: random forest regression, support vector regression, partial least squares regression, and continuous projection algorithm combined with support vector regression.

[0030] Furthermore, when the test set is used to verify the accuracy of the model in step S5, iterative calculations are performed based on the algorithm development logic to calculate the root mean square error RMSE and the goodness of fit R of the model. 2 , that is, select the root mean square error value with a relatively small value and the goodness of fit value as the optimal model parameters; the model accuracy rate is not less than 0.90 based on the optimal model parameters; the root mean square error RMSE and the goodness of fit R 2The calculation formula is as follows:

[0031]

[0032]

[0033] In the formula, y i The moisture content value predicted by the model; t i is the actual moisture content value measured in the experiment; r i It is the average value of m true moisture contents measured experimentally.

[0034] Furthermore, in steps S3 to S6, the spectral data preprocessing, feature spectrum extraction, and the establishment, prediction, and evaluation of the regression machine learning algorithm use the scikit-learn library of Python software, and the auxiliary calculations use the numpy and pandas libraries.

[0035] The beneficial effects of the present invention are:

[0036] (1) For multi-source solid wastes with a wide variety and complex composition, a method combining Fourier transform attenuated total reflection infrared spectroscopy technology and machine learning technology is used for the first time to predict their moisture content. This method can provide a highly accurate moisture content prediction value, improve the automation level of moisture content detection of multi-source solid wastes, and has high practical value.

[0037] (2) The operation is simple, and the samples do not need to be pre-processed such as crushing, drying, and weighing;

[0038] (3) Non-destructive testing, does not require any chemical agents and does not damage the sample;

[0039] (4) Fast response speed. The analysis time of optical properties of solid waste only takes less than 1 minute, which improves the timeliness of the detection process compared with traditional testing methods.

[0040] (5) It can be used for online detection and is expected to make progress in the real-time analysis of the moisture content of multi-source solid waste, assist in evaluating the treatability of solid waste and predict changes in indicators during the treatment and disposal process. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart of solid waste moisture content prediction based on infrared spectroscopy and machine learning involved in the present invention;

[0042] Figure 2 The original infrared spectrum of the representative solid waste sample involved in Example 1 of the present invention;

[0043] Figure 3 The spectrum of the representative solid waste sample involved in Example 1 of the present invention is processed by ATR correction and SPA feature extraction results;

[0044] Figure 4 The predicted value and the actual value of the moisture content of the independently collected samples involved in Example 1 of the present invention;

[0045] Figure 5 The predicted value and the actual value of the moisture content of the independently collected samples involved in Example 2 of the present invention;

[0046] Figure 6 The predicted value and the actual value of the moisture content of the independently collected samples involved in Example 3 of the present invention;

[0047] Figure 7 The predicted value and the actual value of the moisture content of the independently collected samples involved in Example 4 of the present invention. DETAILED DESCRIPTION

[0048] The present invention is described in detail below in conjunction with the accompanying drawings and embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Embodiment 1:

[0050] The specific implementation process is as follows Figure 1 As shown, the following steps are included:

[0051] Step 1, collect 1272 typical representative samples of domestic waste, construction waste and general industrial solid waste, with a moisture content ranging from 0.50% to 90.00%. Specifically as follows: paper (216 samples, moisture content ranging from 1.70% to 6.43%), fabric (597 samples, moisture content ranging from 4.09% to 76.44%), leather (432 samples, moisture content ranging from 0.50% to 15.57%), wood and bamboo (27 samples, moisture content ranging from 7.09% to 11.21%). The collected representative samples do not need to be pretreated, and the Fourier transform attenuated total reflection infrared spectroscopy technology is used directly to obtain the sample infrared spectrum data. The results are as follows: Figure 2 The acquisition range of Fourier transform attenuated total reflection infrared spectrum is 400~4000cm -1 , the resolution is 4cm -1 , a single sample was scanned 16 times, and the average spectrum was taken to construct the machine learning model.

[0052] Step 2: Dry the sample in an oven at 105°C ± 5°C and then test its moisture content.

[0053] Step 3: Use the ATR correction algorithm to preprocess the sample spectrum data collected to sharpen the spectrum data information so that the sample spectrum and the spectrum library match better. The results are as follows: Figure 3 shown.

[0054] ATR correction preprocessing is to correct the attenuated total reflectance (ATR) spectrum to reduce the change in penetration depth caused by the absorption band shift. In order to better match the sample spectrum with the spectrum library, it is necessary to perform corrections based on the wavelength of the incident light, the refractive index of the crystal, the refractive index of the sample, and the incident angle α of the light beam at the crystal interface.

[0055] The two infrared spectrum bands related to water molecules in the preprocessed infrared spectrum data are extracted as characteristic band data, namely 1500~1800cm -1 and 3000~3600cm -1 The infrared spectral data (n=1272) after spectral data preprocessing and characteristic spectrum extraction and its actual moisture content measurement information are divided into data sets. Based on the Random function, the data of each sample is divided into a training set and a test set, where the number of training sets accounts for 70% of the total sample data, and the number of test sets accounts for 30% of the total sample data.

[0056] Step 4: According to the moisture content and spectral information of solid waste in the training set, a successive projection algorithm combined with SVR (SPA-SVR) model of moisture content is established. There are 27 feature vectors extracted by the SPA algorithm; the SVR parameter settings are C = 1469.39, gamma = 62.24.

[0057] Step 5: Use the test set to verify the model accuracy and further optimize the multi-source solid waste moisture content prediction model. The test set accuracy of the successive projections algorithm combined with SVR (SPA-SVR) model reached more than 0.96, and the root mean square error value was 3.80.

[0058] Step 6, 43 papers, 120 fabrics, 86 leathers and 6 woods were collected independently, with moisture contents ranging from 0.50 to 90.00%, totaling 255 solid waste samples to be tested. The infrared spectrum data of the samples were directly obtained by using Fourier transform attenuated total reflection infrared spectroscopy technology. After spectral data preprocessing and feature extraction, the data were directly input into the moisture content prediction model constructed above to output the moisture content of the unknown samples. The comparison results of the predicted value of the moisture content and the true value are shown in Figure 6. Figure 4As shown. Among them, the moisture content regression R 2 It can reach 0.96, and the root mean square error RMSE is 3.80.

[0059] In steps 1 to 6, the scikit-learn library of Python software was used for spectral data preprocessing, feature extraction, and the establishment of regression machine learning algorithms, and the numpy and pandas libraries were used for auxiliary calculations.

[0060] Example 2

[0061] This embodiment is basically the same as Embodiment 1, except that:

[0062] (1) Use the first-order derivative to preprocess the spectral data. The first-order derivative preprocessing is to find the first-order derivative of the spectrum, which is conducive to analyzing the shoulder peak in the original spectrum and can give the change rate of the entire spectrum. In the first-order derivative, the shoulder peak becomes narrower, so the spectrum peak is easier to identify, which can improve the accuracy of the regression prediction model;

[0063] (2) A random forest model (RF) was established based on the infrared spectral data after spectral data preprocessing and feature spectrum extraction and the actual measurement information of moisture content; the number of RF model classifiers was set to 22.

[0064] After preprocessing and feature extraction, the spectral data of the unknown solid waste sample to be tested is directly input into the moisture content prediction model constructed above to output the moisture content of the unknown sample. The comparison results of the predicted value and the true value of the moisture content are shown in the figure. Figure 5 As shown. Among them, the moisture content regression R 2 It can reach 0.95, and the root mean square error RMSE is 4.85.

[0065] Example 3

[0066] This embodiment is basically the same as Embodiment 1, except that:

[0067] (1) 87 unknown leather samples were taken as test objects, with a moisture content ranging from 0.50% to 15.57%;

[0068] (2) Preprocess the spectral data using Savitsky-Golay filtering. Savitsky-Golay filtering preprocessing uses a given number of data points (any odd value in the range of 3 to 51) and the degree of the polynomial (any value in the range of 1 to 6) to determine the smoothness of the spectrum to be converted, which is helpful in improving the peak shape masked by noise.

[0069] Finally, we get the moisture content regression R 2 The predicted value of water content can reach 0.98, and the root mean square error value RMSE is 0.72. The comparison results between the predicted value and the true value of water content are as follows: Figure 6 shown.

[0070] Example 4

[0071] This embodiment is basically the same as Embodiment 1, except that:

[0072] (1) 111 unknown fabric samples were selected as test objects, with a moisture content ranging from 4.09% to 76.44%;

[0073] (2) Preprocess the spectral data using the first-order derivative combined with Savitzky–Golay filtering. The first-order derivative combined with Savitzky–Golay filtering preprocessing combines the advantages of the two preprocessing methods, and the spectral data after the first-order derivative processing is further preprocessed by Savitzky–Golay filtering.

[0074] Finally, we get the moisture content regression R 2 The predicted value of water content can reach 0.90, and the root mean square error value RMSE is 5.92. The comparison results between the predicted value and the true value of water content are as follows: Figure 7 shown.

[0075] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and implementation modes, and can be fully applicable to various fields suitable for the present invention. For those familiar with the art, for those of ordinary skill in the art, various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to specific details.

Claims

1. A method for predicting moisture content of solid waste based on infrared spectroscopy and machine learning, characterized in that: The following steps are involved: S1. Collect typical multi-source solid waste samples, and directly use Fourier transform attenuated total reflection infrared spectroscopy technology to obtain sample infrared spectrum data; classified by source, the typical multi-source solid waste samples include but are not limited to: domestic waste, construction waste, general industrial solid waste and agricultural solid waste; classified by type, the typical multi-source solid waste samples include but are not limited to: kitchen waste, paper, fabric, leather, plastic, rubber, wood and bamboo, metal, glass, brick, tile and ceramic, and ash and soil; S2. Use an oven to dry the sample and then test its moisture content; the oven temperature range is 100-110°C; S3, using spectral data preprocessing and feature spectrum extraction, and dividing the processed infrared spectrum data into a training set and a test set; S4. Establishing a multi-source solid waste moisture content regression prediction model according to the moisture content of solid waste in the training set and its spectral information; S5. Use the test set to test the model accuracy and further optimize the multi-source solid waste moisture content prediction model; S6. Collect the solid waste samples to be tested, directly use Fourier transform attenuated total reflection infrared spectroscopy technology to obtain the infrared spectrum data of the samples, and use the constructed multi-source solid waste moisture content prediction model to predict the moisture content of the samples.

2. A method for predicting moisture content of solid waste based on infrared spectroscopy and machine learning according to claim 1, characterized in that: When collecting various types of multi-source solid waste components in step S1, the number of representative samples of each component is not less than 10.

3. The method for predicting moisture content of solid waste based on infrared spectroscopy and machine learning according to claim 1, characterized in that: The moisture content of each type of solid waste sample collected in step S1 ranges from 0.50% to 90.00%.

4. The method for predicting moisture content of solid waste based on infrared spectroscopy and machine learning according to claim 1, characterized in that: The acquisition range of the Fourier transform attenuated total reflection infrared spectrum in step 1 is 400-4000cm -1 , the resolution is 1~8cm -1 ; The number of scans for a single sample was 8 to 16 times, and the average spectrum was taken to construct the machine learning model.

5. The method for predicting moisture content of solid waste based on infrared spectroscopy and machine learning according to claim 1, characterized in that: The spectral data preprocessing method in step S3 includes but is not limited to: ATR correction, first-order derivative, Savitzky–Golay filtering, and first-order derivative combined with Savitzky–Golay filtering; the spectral data preprocessing method is used to sharpen the spectral data information.

6. The method for predicting moisture content of solid waste based on infrared spectroscopy and machine learning according to claim 1, characterized in that: The characteristic spectrum selection in step S3 is to select two infrared spectrum bands related to water molecules as characteristic bands, namely 1500-1800cm -1 and 3000~3600cm -1 .

7. The method for predicting moisture content of solid waste based on infrared spectroscopy and machine learning according to claim 1, characterized in that: The training set and test set division method described in step S3 is to divide the data of each sample into a training set and a test set based on a Random function, wherein the number of training sets accounts for 60% to 80% of the total sample data, and the number of test sets accounts for 20% to 40% of the total sample data.

8. The method for predicting moisture content of solid waste based on infrared spectroscopy and machine learning according to claim 1, characterized in that: The machine learning model for predicting the moisture content of solid waste described in step S4 is obtained by one of the following algorithms: random forest regression, support vector regression, partial least squares regression, and continuous projection algorithm combined with support vector regression.

9. The method for predicting moisture content of solid waste based on infrared spectroscopy and machine learning according to claim 1, characterized in that: When the test set is used to verify the accuracy of the model in step S5, iterative calculations are performed based on the algorithm development logic to calculate the root mean square error RMSE and goodness of fit R of the model. 2 , that is, select the root mean square error value with a relatively small value and the goodness of fit value as the optimal model parameters; the model accuracy rate is not less than 0.90 based on the optimal model parameters; the root mean square error RMSE and the goodness of fit R 2 The calculation formula is as follows: In the formula, y i The moisture content value predicted by the model; t i is the actual moisture content value measured in the experiment; r i It is the average value of m true moisture contents measured experimentally.

10. The method for predicting moisture content of solid waste based on infrared spectroscopy and machine learning according to claim 1, characterized in that: In steps S3 to S6, the scikit-learn library of Python software is used for spectral data preprocessing, feature spectrum extraction, and the establishment, prediction, and evaluation of regression machine learning algorithms, and the numpy and pandas libraries are used for auxiliary calculations.