Evaluation and prediction method for odor intensity and odor type of sewage plant

Abstract

The invention relates to a sewage plant odor intensity and odor type evaluation and prediction method, which comprises the following steps: sampling in a sewage plant, analyzing a gas sample, and performing olfactory discrimination analysis on the sample to obtain a GC-IMS spectrogram, odor intensity and odor type; processing the sample data, carrying out frame selection on peaks in the spectrogram, carrying out integration on the peak area of the selected signal peak, and exporting a csv file; the different csv files are arranged and combined into a dataset.csv file; performing data preprocessing on the file, and arranging data in the file obtained in the step 3 into an n * s matrix; selecting sample data as a training set and a test set, and establishing a random forest model; and applying the obtained data to a random forest model to obtain the odor intensity and the odor type. The method has the advantages of reasonable design and high reliability, and can more accurately predict the stink pollution characteristics of the specific point location and the surrounding environment of the sewage plant, thereby guiding the sewage plant to more effectively carry out stink pollution prevention and control.

Description

Technical Field

[0001] The invention relates to the technical field of sewage treatment plants, and in particular to a method for evaluating and predicting odor intensity and odor type in sewage treatment plants. Background Art

[0002] Odor pollution, as one of the seven major environmental hazards, seriously reduces environmental quality, brings discomfort to surrounding residents and even endangers human health, and causes harm to the ecological environment. Since odor (smell) pollution is different from traditional pollution, it has the characteristics of subjective perceptibility, complexity of perceptual variation, and easy diffusion of pollution, which makes it extremely difficult to accurately detect and scientifically evaluate. Sewage treatment plants are typical sources of odor (smell) pollution, with the characteristics of both point sources and non-point sources (unorganized emissions). The odor characteristics of different process units in the plant area are significantly different, and there are technical difficulties such as difficulty in accurately identifying key odor pollutants, difficulty in qualitatively evaluating odor characteristics, and difficulty in accurately tracing the source.

[0003] At present, the main methods used to detect odor (peculiar smell) pollution include instrumental analysis and sensory measurement. Instrumental analysis can qualitatively and quantitatively determine the types and chemical concentrations of odor (peculiar smell) compounds, but it cannot directly reflect the intensity level of odor (peculiar smell). Some pollutants can still produce strong odor stimulation at extremely low concentrations (below the detection limit of the instrument), and the odor stimulation after the superposition of multiple pollutants that do not exceed the threshold or have weak odors cannot be ignored. Therefore, it is still necessary to use sensory measurement methods to directly evaluate the degree of stimulation of odor pollutants on the human olfactory organs. Sensory measurement also has problems such as time-consuming, laborious and possible health risks to olfactory identification personnel. In addition, although the measurement methods are different, the sensory measurement of odor (peculiar smell) in various countries around the world basically uses two indicators: odor intensity and odor concentration. There is a lack of identification of odor types, which is not conducive to the accurate identification and prevention and control of odor pollution. Some studies have attempted to combine the two methods of instrumental analysis and sensory measurement. For example, the analysis method of gas chromatography-mass spectrometry (GC-MS-ODP) equipped with a sniffing cup is used, but it is time-consuming, labor-intensive and has high instrument costs. The electronic nose equipped with multiple odorant sensors can be used to semi-quantitatively determine odorous (smell) substances, and can be combined with sensory evaluation results, and algorithms such as partial least squares (PLS) can be used to correlate and analyze sensor response data with sensory evaluation results, thereby realizing the prediction of odor concentration. However, due to the limited sensitivity and resolution of electronic nose sensors, it is still difficult to accurately characterize, identify and predict odor characteristics (including odor intensity and odor type). Gas chromatography-ion mobility spectrometry (GC-IMS) is a technology used to separate and detect gas phase compounds. It combines the effective separation capability of gas chromatography (GC) and the rapid response capability of ion mobility spectrometry (IMS). The sensitivity can reach ppb level and can identify a variety of volatile and semi-volatile substances, especially oxygen-containing or sulfur-containing volatile organic compounds with low olfactory thresholds. GC-IMS is widely used in military applications, such as analyzing explosives, illegal drugs and chemical warfare agents, as well as in food science, natural products, clinical science, process chemistry and environmental protection. GC-IMS for odor (smell) pollution feature analysis is expected to solve the technical bottleneck problem of GC-MS, electronic nose and other instrumental analysis methods in the rapid identification of odor pollution features.

[0004] The correlation between instrumental analysis and sensory evaluation data can be calculated using multiple regression algorithms and machine learning regression algorithms such as support vector machine regression (SVR), random forest (RF), and multi-layer perceptron (MLP). Support vector machine regression minimizes the model error by finding a model that maximizes the margin. This type of model attempts to maintain smoothness to the maximum extent possible without exceeding a certain threshold. The random forest algorithm combines multiple decision trees. Each time, the data set is randomly selected with replacement, and some features are randomly selected as input. Therefore, this algorithm is called the random forest algorithm. The advantages of this algorithm are: (1) The training can be highly parallelized, which has advantages in large sample training speed in the big data era; (2) It can process high-dimensional data without feature selection; (3) Due to the use of random sampling, the variance of the trained model is small and the generalization ability is strong; (4) It is insensitive to the missing of some features. The multi-layer perceptron algorithm is a classic feedforward neural network, a type of artificial neural network (ANN), and a neural network composed of multiple neuron models. Taking the three-layer multi-layer perceptron as an example, it consists of an input layer, an intermediate layer, and an output layer. It is widely used in tasks such as classification, regression, and sequence prediction. MLP can approximate complex nonlinear relationships by introducing nonlinear activation functions (such as ReLU, Sigmoid, Tanh), so it has strong performance in dealing with nonlinear problems. However, in actual operation, compared with the random forest model, the prediction results of the MLP model are not accurate when the amount of data in the data set is not large enough, mainly because the model tends to ignore the correlation between attributes in the data set and is prone to overfitting. The present invention is proposed based on the above research background, and aims to provide a method for evaluating and predicting the odor intensity and odor type of a sewage treatment plant to meet actual design and use requirements, and to improve the odor prevention and control efficiency of sewage treatment plants. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the object of the present invention is to provide a method for evaluating and predicting odor intensity and odor type in a sewage treatment plant, which has the advantages of reasonable design and high reliability, and can more accurately predict the odor pollution characteristics of specific points in a sewage treatment plant and the surrounding environment, thereby guiding the sewage treatment plant to carry out odor pollution prevention and control more effectively.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A method for evaluating and predicting odor intensity and odor type of a sewage treatment plant, the method comprising the following steps:

[0008] Step 1: Sampling is performed at sampling points in the sewage treatment plant and the samples are taken as gas samples. The sampling points include typical gas pollution sources and key emission ignition plant boundary sensitive points; the gas samples are analyzed by gas chromatography-ion mobility spectrometry and the odor hierarchy analysis method is used to perform olfactory analysis on the samples simultaneously, and the GC-IMS spectrum and odor intensity and odor type results are obtained for subsequent steps 2-6 to construct a random forest model. The horizontal axis of the GC-IMS spectrum is the drift time, and the vertical axis is the gas phase retention time. The position and peak intensity of the signal peak in the spectrum indicate the type and concentration of the volatile substances contained in the sample. The odor hierarchy analysis method is used by the olfactory judge to give the overall odor intensity value and the main odor type of the sample. The odor type is determined by the classification in the municipal sewage odor wheel diagram;

[0009] Step 2: Use the data processing software provided by the GC-IMS analyzer to process the sample data. Use the built-in peak circle tool of the software to select the signal peak in the GC-IMS spectrum, and automatically integrate the peak area of ​​the selected signal peak to obtain the peak intensity information. Export the data containing the peak position and peak intensity information obtained by selecting all the signal peaks into a csv file;

[0010] Step 3: Repeat step 2 for the GC-IMS data obtained from all samples, and then organize and merge the different csv files into a dataset.csv file;

[0011] Step 4: In order to facilitate the use of data in the subsequent step 5, the pandas library of Python is used to preprocess the data of the dataset.csv file. The preprocessing content is as follows: the data in the file obtained in step 3 is organized into an n×s matrix, where the number of matrix rows n represents the number of samples, each row corresponds to a sample, and the number of matrix columns s = 3×m+2, where 3 represents three features, namely, peak center position, peak width, and peak intensity; 2 represents two labels, namely, odor intensity and odor type, and m represents the number of signal peaks detected in all samples. The first s-2 columns represent the peak center position, peak width, and peak intensity corresponding to the m peaks, and the last two columns represent the odor intensity and odor type, respectively;

[0012] Step 5: Randomly select 3 / 4 of the sample data as a training set, and select the remaining 1 / 4 of the sample data as a test set; establish a random forest model, and use the training set for supervised training, use mean square error loss and cross entropy loss as loss functions in the odor intensity and odor type prediction tasks to evaluate the accuracy of the model, then use the training set for supervised training, and the test set is used to evaluate the accuracy of the model. According to the GC-IMS spectrum obtained in the actual application, the number of signal peaks, peak intensity, peak position, and peak width data obtained by operating in step 2 are used to adjust the number of trees in the random forest model, the maximum depth of the tree, and the minimum number of samples required for the leaf node. The grid search is to traverse all possible parameter combinations in the defined parameter grid by exhaustive search, and train the model for each combination, evaluate the performance, and finally select the best parameter combination;

[0013] Step 6: Perform steps 1-4 on the sample to be tested, and then apply the obtained data to the random forest model to obtain the odor intensity and odor type of the sample, thus completing the evaluation and prediction of the odor intensity and odor type of the sewage treatment plant.

[0014] As a further optimization of the above scheme, in the step 1, the odor hierarchy analysis method is used by the olfactory identification personnel to evaluate the sensory characteristics of the on-site odor pollution, and give the odor intensity and odor type results, wherein the odor intensity is divided into 0, 2, 4, 6, 8, 10, and 12 levels, wherein level 0 indicates no intensity and no odor, level 2 indicates weak intensity and the odor can be detected by people with odor sensitivity, level 4 indicates weak intensity and the odor can be detected by people with general sense, level 6 indicates medium intensity level and the odor is obviously detected, level 8 indicates strong intensity level and irritation during smell measurement, level 10 indicates strong intensity and unbearable for a long time, and level 12 indicates very strong intensity and unbearable; the odor type is described using the odor characteristic terms given in the odor wheel diagram established for the sewage treatment plant, and the odor type results are assigned and scaled as the target variable data set for subsequent analysis, wherein the odor wheel diagram includes three concentric circles, the first circle corresponds to the 8 common odor types in municipal sewage plants, the second circle is the subdivision of each odor type in the above 8 odor types, and the third circle is the typical substances corresponding to the subdivision of each odor type.

[0015] As a further optimization of the above scheme, the eight common odor types of the municipal sewage treatment plant are feces, corruption / sulfur, sour, earthy, medicinal, solvent, grass, and ammonia, which are assigned values ​​of 1-8 in the model respectively.

[0016] As a further optimization of the above scheme, in step 1, the detection results of volatile substances are extracted from the GC-IMS spectrum data, including the peak center point position, peak width, and peak intensity. When the signal peak is circled, a unified standard should be used, and the same set of peak circle methods should be applied to all samples to extract peak intensity information, and then redundant information is eliminated for characterizing the samples.

[0017] As a further optimization of the above scheme, the mean square error loss is a measure of the square average of the difference between the model prediction value and the actual value. Its purpose is to optimize the prediction ability of the model by minimizing the mean square error. It is used for odor intensity prediction. Ideally, when the predicted value is very close to the true value, the MSE will be close to zero. Therefore, the smaller the MSE, the better the model prediction result. The cross entropy loss is a measure of the difference between the predicted probability distribution and the true distribution. Its purpose is to improve the classification model's ability to distinguish categories by minimizing the cross entropy Cross-Entropy Loss. It is used for odor type prediction. The calculation formulas of the above MSE and Cross-Entropy Loss are as follows:

[0018]

[0019] Where: n is the number of samples; yi' is the predicted value of the i-th sample; y i is the actual value of the i-th sample.

[0020]

[0021] Where: n is the number of samples; K is the number of categories; y i,k Indicates that sample i belongs to the true label of category k (value is 0 or 1); It represents the probability that sample i is predicted to be k.

[0022] The method for evaluating and predicting odor intensity and odor type of a sewage treatment plant according to the present invention has the following beneficial effects:

[0023] (1) By extracting features (such as peak position, width, intensity, etc.) from the GC-IMS spectrum and combining it with principal component analysis (PCA) for dimensionality reduction when necessary, and applying machine learning regression algorithms to predict the odor intensity and odor type of the sample, it can provide more scientific and reliable data basis and evaluation results for environmental odor pollution control.

[0024] (2) The odor intensity and odor type of the sample can be predicted more accurately based on the instrumental measurement results, thus avoiding the physical harm of the odorist caused by the traditional sensory measurement method and being safer.

[0025] (3) Compared with traditional regression analysis methods, machine learning methods are faster, and the larger the data set, the more reliable the model prediction results.

[0026] (4) Modeling and training are carried out based on the odor pollution characteristics of sewage treatment plants, and the model is more accurate and targeted. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Attached Figure 1 It is a schematic flow chart of the method for evaluating and predicting odor intensity and odor type of a sewage treatment plant according to the present invention.

[0028] Attached Figure 2 This is an example of the GC-IMS spectrum of the sample of the present invention.

[0029] Attached Figure 3 This is an example diagram of the sample characteristic value information, odor intensity and odor type of the present invention.

[0030] Attached Figure 4 This is the source code diagram of the model of the present invention.

[0031] Attached Figure 5 This is a diagram showing the results of running the model of the present invention.

[0032] Attached Figure 6 Odor wheel diagram created for the sewage treatment plant of the present invention. DETAILED DESCRIPTION

[0033] The following is combined with Figure 1-6 The present invention provides a method for evaluating and predicting odor intensity and odor type in a sewage treatment plant in detail.

[0034] A method for evaluating and predicting odor intensity and odor type of a sewage treatment plant, the method comprising the following steps:

[0035] Step 1: Sampling is performed at sampling points in the sewage treatment plant and the samples are taken as gas samples. The sampling points include typical gas pollution sources and key emission ignition plant boundary sensitive points; the gas samples are analyzed by gas chromatography-ion mobility spectrometry and the odor hierarchy analysis method is used to perform olfactory analysis on the samples simultaneously, and the GC-IMS spectrum and odor intensity and odor type results are obtained for subsequent steps 2-6 to construct a random forest model. The horizontal axis of the GC-IMS spectrum is the drift time, and the vertical axis is the gas phase retention time. The position and peak intensity of the signal peak in the spectrum indicate the type and concentration of the volatile substances contained in the sample. The odor hierarchy analysis method is used by the olfactory judge to give the overall odor intensity value and the main odor type of the sample. The odor type is determined by the classification in the municipal sewage odor wheel diagram;

[0036] Step 2: Use the data processing software provided by the GC-IMS analyzer to process the sample data. Use the built-in peak circle tool of the software to select the signal peak in the GC-IMS spectrum, and automatically integrate the peak area of ​​the selected signal peak to obtain the peak intensity information. Export the data containing the peak position and peak intensity information obtained by selecting all the signal peaks into a csv file;

[0037] Step 3: Repeat step 2 for the GC-IMS data obtained from all samples, and then organize and merge the different csv files into a dataset.csv file;

[0038] Step 4: In order to facilitate the use of data in the subsequent step 5, the pandas library of Python is used to preprocess the data of the dataset.csv file. The preprocessing content is as follows: the data in the file obtained in step 3 is organized into an n×s matrix, where the number of matrix rows n represents the number of samples, each row corresponds to a sample, and the number of matrix columns s = 3×m+2, where 3 represents three features, namely, peak center position, peak width, and peak intensity; 2 represents two labels, namely, odor intensity and odor type, and m represents the number of signal peaks detected in all samples. The first s-2 columns represent the peak center position, peak width, and peak intensity corresponding to the m peaks, and the last two columns represent the odor intensity and odor type, respectively;

[0039] Step 5: Randomly select 3 / 4 of the sample data as a training set, and select the remaining 1 / 4 of the sample data as a test set; establish a random forest model, and use the training set for supervised training, use mean square error loss and cross entropy loss as loss functions in the odor intensity and odor type prediction tasks to evaluate the accuracy of the model, then use the training set for supervised training, and the test set is used to evaluate the accuracy of the model. According to the GC-IMS spectrum obtained in the actual application, the number of signal peaks, peak intensity, peak position, and peak width data obtained by operating in step 2 are used to adjust the number of trees in the random forest model, the maximum depth of the tree, and the minimum number of samples required for the leaf node. The grid search is to traverse all possible parameter combinations in the defined parameter grid by exhaustive search, and train the model for each combination, evaluate the performance, and finally select the best parameter combination;

[0040] Step 6: Perform steps 1-4 on the sample to be tested, and then apply the obtained data to the random forest model to obtain the odor intensity and odor type of the sample, thus completing the evaluation and prediction of the odor intensity and odor type of the sewage treatment plant.

[0041] In the step 1, the odor level analysis method is used by the olfactory identification personnel to evaluate the sensory characteristics of the on-site odor pollution, and give the odor intensity and odor type results, wherein the odor intensity is divided into 0, 2, 4, 6, 8, 10, and 12 levels, wherein level 0 indicates that the intensity is none and there is no odor, level 2 indicates that the intensity is weak and the odor can be detected by people with odor sensitivity, level 4 indicates that the intensity is weak and the odor can be detected by people with general sense, level 6 indicates that the intensity level is medium and the odor is obviously detected, level 8 indicates that the intensity level is strong and there is irritation when smelling, level 10 indicates that the intensity is strong and it is unbearable for a long time, and level 12 indicates that the intensity is very strong and unbearable; the odor type is described using the odor characteristic terms given in the odor wheel diagram established for the sewage treatment plant, and the odor type results are assigned and scaled as the target variable data set for subsequent analysis, and the odor wheel diagram includes three concentric circles, the first circle corresponds to the 8 common odor types in municipal sewage treatment plants, the second circle is the subdivision of each odor type in the above 8 odor types, and the third circle is the typical substances corresponding to the subdivision of each odor type.

[0042] The eight common odor types of the municipal sewage treatment plant are feces, corruption / sulfur, sour, earthy, medicinal, solvent, grass, and ammonia, which are assigned values ​​of 1-8 in the model.

[0043] In step 1, the volatile substance detection results are extracted from the GC-IMS spectrum data, including the peak center point position, peak width, and peak intensity. When the signal peak is circled, a unified standard should be used, and the same set of peak circle methods should be applied to all samples to extract peak intensity information, and then redundant information is eliminated for characterizing the samples.

[0044] The mean square error loss is a measure of the square average of the difference between the model prediction value and the actual value. Its purpose is to optimize the prediction ability of the model by minimizing the mean square error. It is used for odor intensity prediction. Ideally, when the predicted value is very close to the true value, the MSE will be close to zero. Therefore, the smaller the MSE, the better the model prediction result. The cross entropy loss is a measure of the difference between the predicted probability distribution and the true distribution. Its purpose is to improve the classification model's ability to distinguish between categories by minimizing the cross entropy Cross-EntropyLoss. It is used for odor type prediction. The calculation formulas of the above MSE and Cross-Entropy Loss are as follows:

[0045]

[0046] Where: n is the number of samples; yi' is the predicted value of the i-th sample; y i is the actual value of the i-th sample.

[0047]

[0048] Where: n is the number of samples; K is the number of categories; y i,k Indicates that sample i belongs to the true label of category k (value is 0 or 1); It represents the probability that sample i is predicted to be k.

[0049] The following is a further description of the method for evaluating and predicting odor intensity and odor type in a sewage treatment plant of the present invention in conjunction with two embodiments.

[0050] Example 1

[0051] On-site sampling was carried out at a sewage treatment plant. A total of 7 sampling points were set up. Two parallel samples were collected at each sampling point and repeated sampling was carried out twice on the same day. A total of 3 samplings were carried out, with a total of 84 samples. The odor intensity and odor type of the samples were determined by odor hierarchy analysis method, and the samples were analyzed and determined by GC-IMS to obtain the material spectrum (see Appendix Figure 2 ), a total of 84 sets of IMS data and sensory analysis results. Follow the steps below to build a random forest model:

[0052] (1) Processing sample data using the software provided with the GC-IMS analyzer, adding the sample data to the project file manager, assigning colors for differentiation, and integrating the peaks of the samples;

[0053] (2) Export the peak intensity data obtained by GC-IMS spectrum analysis software as a csv file;

[0054] (3) After repeating steps 2-3 for the GC-IMS spectrum data obtained from different sample analyses, the different csv files were sorted and merged into a dataset.csv file.

[0055] (4) Import the pandas and Scikit-learn libraries into the Python software and perform data preprocessing on the dataset.csv file; randomly select 3 / 4 of the sample data as the training set and 1 / 4 of the sample data as the test set; establish a random forest model, use mean square error loss and cross entropy loss as loss functions in the odor intensity prediction and odor type identification tasks, respectively, use the training set for supervised training, and the test set is used to evaluate the accuracy of the model.

[0056] Example 2

[0057] Take 50 samples of sewage plant ambient air with unknown odor intensity and odor type and hand them over to the tester for testing. The test steps are as follows:

[0058] (1) Use a 5L gas bag to collect samples, first pass the gas into the GC-IMS analyzer for analysis, and then perform olfactory analysis.

[0059] (2) Process the data using the data processing software (LAV) provided with the GC-IMS analyzer, add the sample data to the project file manager, and integrate the sample peaks;

[0060] (3) Export the peak intensity data obtained by LAV software as a csv file;

[0061] (4) Organize the olfactory identification results and input them into the csv file exported by the corresponding samples;

[0062] (5) The sample data was applied to the random forest model generated in Example 1 to predict the odor intensity and odor type of the sample. The model prediction results (see Appendix Figure 5 ) It can be seen that the mean square error between the predicted value and the true value is about 0.504, and the prediction effect is good.

[0063] The above description of the embodiments is to facilitate the understanding and application of the present invention by those skilled in the art. It is obvious that those skilled in the art can easily make various modifications to these embodiments and apply the general principles described herein to other embodiments without creative work. Therefore, the present invention is not limited to the embodiments herein, and improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the present invention should be within the protection scope of the present invention.

Claims

1. A method for evaluating and predicting odor intensity and odor type in a sewage treatment plant, characterized in that: The method comprises the following steps: Step 1: Sampling is performed at sampling points in the sewage treatment plant and the samples are taken as gas samples. The sampling points include typical gas pollution sources and key emission ignition plant boundary sensitive points; the gas samples are analyzed by gas chromatography-ion mobility spectrometry and the odor hierarchy analysis method is used to perform olfactory analysis on the samples simultaneously, and the GC-IMS spectrum and odor intensity and odor type results are obtained for subsequent steps 2-6 to construct a random forest model. The horizontal axis of the GC-IMS spectrum is the drift time, and the vertical axis is the gas phase retention time. The position and peak intensity of the signal peak in the spectrum indicate the type and concentration of the volatile substances contained in the sample. The odor hierarchy analysis method is used by the olfactory judge to give the overall odor intensity value and the main odor type of the sample. The odor type is determined by the classification in the municipal sewage odor wheel diagram; Step 2: Use the data processing software provided by the GC-IMS analyzer to process the sample data. Use the built-in peak circle tool of the software to select the signal peak in the GC-IMS spectrum, and automatically integrate the peak area of ​​the selected signal peak to obtain the peak intensity information. Export the data containing the peak position and peak intensity information obtained by selecting all the signal peaks into a csv file; Step 3: Repeat step 2 for the GC-IMS data obtained from all samples, and then organize and merge the different csv files into a dataset.csv file; Step 4: In order to facilitate the use of data in the subsequent step 5, the pandas library of Python is used to preprocess the data of the dataset.csv file. The preprocessing content is as follows: the data in the file obtained in step 3 is organized into an n×s matrix, where the number of matrix rows n represents the number of samples, each row corresponds to a sample, and the number of matrix columns s = 3×m+2, where 3 represents three features, namely, peak center position, peak width, and peak intensity; 2 represents two labels, namely, odor intensity and odor type, and m represents the number of signal peaks detected in all samples. The first s-2 columns represent the peak center position, peak width, and peak intensity corresponding to the m peaks, and the last two columns represent the odor intensity and odor type, respectively; Step 5: Randomly select 3 / 4 of the sample data as a training set, and select the remaining 1 / 4 of the sample data as a test set; establish a random forest model, and use the training set for supervised training, use mean square error loss and cross entropy loss as loss functions in the odor intensity and odor type prediction tasks to evaluate the accuracy of the model, then use the training set for supervised training, and the test set is used to evaluate the accuracy of the model. According to the GC-IMS spectrum obtained in the actual application, the number of signal peaks, peak intensity, peak position, and peak width data obtained by operating in step 2 are used to adjust the number of trees in the random forest model, the maximum depth of the tree, and the minimum number of samples required for the leaf node. The grid search is to traverse all possible parameter combinations in the defined parameter grid by exhaustive search, and train the model for each combination, evaluate the performance, and finally select the best parameter combination; Step 6: Perform steps 1-4 on the sample to be tested, and then apply the obtained data to the random forest model to obtain the odor intensity and odor type of the sample, thus completing the evaluation and prediction of the odor intensity and odor type of the sewage treatment plant.

2. The method for evaluating and predicting odor intensity and odor type of a sewage treatment plant according to claim 1, characterized in that: In the step 1, the odor level analysis method is used by the olfactory identification personnel to evaluate the sensory characteristics of the on-site odor pollution, and give the odor intensity and odor type results, wherein the odor intensity is divided into 0, 2, 4, 6, 8, 10, and 12 levels, wherein level 0 indicates that the intensity is none and there is no odor, level 2 indicates that the intensity is weak and the odor can be detected by people with odor sensitivity, level 4 indicates that the intensity is weak and the odor can be detected by people with general sense, level 6 indicates that the intensity level is medium and the odor is obviously detected, level 8 indicates that the intensity level is strong and there is irritation when smelling, level 10 indicates that the intensity is strong and it is unbearable for a long time, and level 12 indicates that the intensity is very strong and unbearable; the odor type is described using the odor characteristic terms given in the odor wheel diagram established for the sewage treatment plant, and the odor type results are assigned and scaled as the target variable data set for subsequent analysis, and the odor wheel diagram includes three concentric circles, the first circle corresponds to the 8 common odor types in municipal sewage treatment plants, the second circle is the subdivision of each odor type in the above 8 odor types, and the third circle is the typical substances corresponding to the subdivision of each odor type.

3. The method for evaluating and predicting odor intensity and odor type of a sewage treatment plant according to claim 2, characterized in that: The eight common odor types of the municipal sewage treatment plant are feces, corruption / sulfur, sour, earthy, medicinal, solvent, grass, and ammonia, which are assigned values ​​of 1-8 in the model.

4. The method for evaluating and predicting odor intensity and odor type of a sewage treatment plant according to claim 1, characterized in that: In step 1, the volatile substance detection results are extracted from the GC-IMS spectrum data, including the peak center point position, peak width, and peak intensity. When the signal peak is circled, a unified standard should be used, and the same set of peak circle methods should be applied to all samples to extract peak intensity information, and then redundant information is eliminated for characterizing the samples.

5. The method for evaluating and predicting odor intensity and odor type of a sewage treatment plant according to claim 1, characterized in that: The mean square error loss is a measure of the square average of the difference between the model prediction value and the actual value. Its purpose is to optimize the prediction ability of the model by minimizing the mean square error. It is used for odor intensity prediction. Ideally, when the predicted value is very close to the true value, the MSE will be close to zero. Therefore, the smaller the MSE, the better the model prediction result. The cross entropy loss is a measure of the difference between the predicted probability distribution and the true distribution. Its purpose is to improve the classification model's ability to distinguish between categories by minimizing the cross entropy Cross-Entropy Loss. It is used for odor type prediction. The calculation formulas of the above MSE and Cross-Entropy Loss are as follows: Where: n is the number of samples; yi' is the predicted value of the i-th sample; y i is the actual value of the i-th sample. Where: n is the number of samples; K is the number of categories; y i,k Indicates that sample i belongs to the true label of category k (value is 0 or 1); It represents the probability that sample i is predicted to be k.

Images