Method for monitoring odor smell at sea based on coupling of pattern recognition and concentration simulation

Abstract

The application provides a malodorous smell walk monitoring method based on pattern recognition and concentration simulation coupling, characterized in that the malodorous smell walk monitoring is carried out in a pattern recognition and concentration simulation coupling mode, the source of the malodorous smell in the environment is preliminarily screened through pattern recognition, the corresponding odor concentration simulation quantitative model is automatically selected according to the pattern recognition result, and when the component of the malodorous gas in the environment changes, the corresponding model is reselected through pattern recognition judgment. The application solves the problem of low accuracy of the instrument in monitoring the odor concentration during the malodorous smell walk monitoring, improves the actual effect of the malodorous smell walk monitoring, improves the accuracy of the industry judged by the pattern recognition method, and meets the requirement of accurately judging the industry during the malodorous smell walk monitoring.

Description

TECHNICAL FIELD

[0001] The application provides a foul odor smell cruise monitoring method based on pattern recognition and concentration simulation coupling, and belongs to the technical field of environmental monitoring. BACKGROUND

[0002] Foul odor refers to all odor gases that stimulate the olfactory organ and cause people to feel unpleasant and damage the living environment. Foul odor / odor pollution is a typical nuisance pollution, which not only affects the quality of life and environmental comfort of residents, but also has direct harm to physiological health. In recent years, residents have complained very strongly, and it has become a serious social livelihood problem.

[0003] Foul odor / odor has a wide source, complex composition, and low environmental concentration. Its emission and diffusion have the characteristics of paroxysmal and instantaneous. Currently, foul odor / odor monitoring usually adopts manual on-site inspection and manual sampling offline testing, which not only has high time and labor costs, but also is difficult to obtain foul odor / odor information in time and comprehensively, and cannot quickly and accurately identify the pollution source and diagnose the cause, which is difficult to meet the needs of existing environmental management. With the development of VOCs cruise monitoring technology, cruise monitoring has become an important means of regional atmospheric pollution monitoring, investigation, tracing, and evaluation.

[0004] The odor concentration determined by sensory analysis is the most important index for foul odor monitoring, law enforcement, and evaluation at home and abroad. This index can comprehensively reflect the sensory effect of foul odor gas on the human body. Since it is a comprehensive sensory index, the standard method is to determine it by manual olfactory identification (HJ1262-2022, EN13725:2003, ASTM E544-10, etc.). However, this method cannot realize on-site, continuous, and online monitoring, and cannot be applied to foul odor / odor cruise monitoring work.

[0005] With the development of sensor technology, foul odor electronic nose (also known as foul odor online monitoring instrument) can simulate human olfactory sensory through sensor array to simulate odor concentration. However, due to the complexity of foul odor gas components, the sensor response of different foul odor gas electronic noses has great differences. Therefore, current domestic and foreign electronic nose research mainly focuses on simulating odor concentration of foul odor gas in specific industries such as garbage and sewage, forming an odor concentration simulation model for industry emission characteristics. Therefore, when applied to regional foul odor / odor cruise monitoring, especially in areas with many foul odor sources such as industrial parks and industrial clusters, the components of foul odor gas in the environment will change with time, place, and meteorological conditions. Generally, foul odor electronic nose can only simulate odor concentration by a model based on specific industry emission characteristics and gas samples when cruising in a complex environment. The monitoring data is quite different from the actual environmental odor concentration.

[0006] The general malodorous electronic nose mainly uses the malodorous gas of specific industries such as garbage and sewage to establish a malodorous concentration simulation quantitative model through fitting or training, forms a malodorous concentration simulation model for the emission characteristics of the industry, is built into the malodorous electronic nose, simulates the malodorous concentration based on the model established based on the emission characteristics of the specific industry and the gas sample, and converts the response signal into a malodorous concentration monitoring value. Therefore, when the malodorous electronic nose is on the move, the model closest to the malodorous odor characteristics in the area is usually selected through experience to simulate the malodorous concentration, and the simulation result is used as the monitoring data. The technology has the following technical defects:

[0007] (1) The general malodorous electronic nose simulates the malodorous concentration based on the model established based on the emission characteristics of the specific industry and the gas sample, and in the area malodorous odor on-the-go monitoring, especially in the industrial park, industrial cluster and other areas with many sources of malodorous odor, since the components of the malodorous gas in the environment will change with time, place and weather conditions, therefore, when the general malodorous electronic nose is on the move, the built-in malodorous concentration simulation quantitative model is often not matched with the components of the malodorous gas, resulting in a large difference between the monitoring data and the actual malodorous concentration in the environment.

[0008] (2) Pattern recognition extracts different sensor response signal characteristic values, uses machine learning or deep learning methods to distinguish or classify different types of samples or industries, but since the sensor response signal is related to not only the components of the sample but also the concentration of the sample, therefore, when the concentration changes, the response signal characteristics of different sensors will also change, even for the same industry, the response signal of the sensor to the malodorous gas of different concentration ranges also has a large difference. When the pattern recognition is trained, the different types or industry samples are in a fixed concentration range, therefore, the classification model established by the pattern recognition has a high success rate in this fixed concentration range, but it is difficult to guarantee the discrimination or classification effect for samples outside this fixed concentration range. Malodorous odor on-the-go monitoring is in the atmospheric environment, and the concentration range of the malodorous gas is large, and the malodorous gas often exceeds the fixed concentration range. Expanding the concentration range will make the pattern recognition difficulty increase exponentially and the discrimination success rate decrease, and it is difficult to meet the requirement of accurately discriminating the industry in the malodorous odor on-the-go monitoring process. SUMMARY

[0009] In view of the above technical problems, the present application provides a malodorous odor on-the-go monitoring method based on the coupling of pattern recognition and concentration simulation, which achieves the following purposes:

[0010] (1) The application solves the problem of low accuracy of instrument monitoring of odor concentration during odor walk monitoring, and improves the actual effect of odor walk monitoring. Through the coupling of pattern recognition and concentration simulation, the source of odor in the environment is preliminarily screened through pattern recognition, and the corresponding odor concentration simulation quantitative model is automatically selected according to the odor industry of pattern recognition, when the composition of the odor gas in the environment changes, the corresponding model is selected through pattern recognition, forming a semi-adaptive, qualitative and quantitative coupling odor walk monitoring method, thereby improving the accuracy of instrument monitoring of odor concentration during odor walk monitoring.

[0011] (2) The application establishes a method of pattern recognition training according to the characteristics of the response relationship between odor gas and sensor, improves the accuracy of using pattern recognition method to distinguish industries, and meets the requirements of accurately distinguishing industries during odor walk monitoring.

[0012] The specific technical solution is:

[0013] The odor walk monitoring method based on pattern recognition and concentration simulation coupling uses pattern recognition and concentration simulation coupling to carry out odor walk monitoring, preliminarily screens the source of odor in the environment through pattern recognition, and automatically selects the corresponding odor concentration simulation quantitative model according to the odor industry of pattern recognition, when the composition of the odor gas in the environment changes, the corresponding model is selected through pattern recognition.

[0014] Specifically, the following steps are included:

[0015] S1: Collecting concentration gradient odor samples by industry, testing sample odor concentration and odor intensity.

[0016] Collect high-concentration odor samples discharged by typical process links of each industry, including organized and unorganized links, and establish a concentration gradient sample library covering odor intensity 0-5 by diluting the samples, wherein each industry (TC in the application) has at least 6 concentration gradient samples corresponding to each odor intensity, two-thirds of which are used for training of qualitative and quantitative models, and one-third of which are used for verification of qualitative and quantitative models.

[0017] The odor concentration and odor intensity of the sample are measured by an olfactory tester;

[0018] S2: Test the odor electronic nose with industry concentration gradient odor samples, and obtain the relationship between the response signal of each sensor and the sample odor concentration and odor intensity through feature extraction.

[0019] The industry-specific concentration gradient odor samples are sequentially introduced into the malodor electronic nose for testing from low concentration to high concentration. According to the response speed of the malodor electronic nose, the response signals of each sensor are extracted as characteristic values when the response signals of the malodor electronic nose enter the stable stage wherein n is the sensor number, i is the corresponding industry, and j is the corresponding odor intensity, and recorded in the sensor response signal and odor concentration, intensity relationship table, wherein two-thirds of the concentration gradient sample library test data are used as the training database, and one-third of the concentration gradient sample library test data are used as the verification database.

[0020] S3: Establishing an industry classification qualitative model using pattern recognition method and placing it into the malodor electronic nose.

[0021] The support vector machine algorithm SVM is used as the pattern recognition algorithm, the data in the training database are used to train the industry classification qualitative model, and the kernel function is selected as the Gaussian kernel function, i.e.

[0022]

[0023] wherein are the feature vectors of the i1th and i2th industries under odor intensity j1 and j2, i1 and i2 can be the same industry, j1 and j2 must be different odor intensities, and σ is the Gaussian kernel function parameter for controlling the distance after mapping.

[0024] The feature vector is wherein is the industry i under odor intensity j, is the sensor response signal characteristic value of the n th sensor under odor intensity j of the i th industry sample obtained in step S2.

[0025] The qualitative model is verified, the data of the verification database are substituted into the qualitative model, and whether the industry discrimination result of the qualitative model is consistent with the actual industry is compared, and the discrimination success rate is verified by the following formula. If the discrimination success rate is ≥ 95%, it means that the qualitative model can distinguish most industries, and the next step is continued. If the discrimination success rate is < 95%, adjust the Gaussian kernel function parameter σ or collect more samples for training until the discrimination success rate is ≥ 95%.

[0026]

[0027] wherein RS is the industry discrimination success rate, N S is the number of successful discrimination, and N T is the total amount of verification data.

[0028] S4: Select the most suitable algorithm from the simulation quantitative algorithm library to establish the odor concentration simulation quantitative model under different odor intensity of each industry, and establish the quantitative model library corresponding to the industry, and put it into the electronic nose.

[0029] The partial least squares regression algorithm (PLSR), support vector regression algorithm (SVR), and artificial neural network algorithm (ANN) are used as the odor concentration simulation quantitative algorithm library. The training database is used to train the algorithm according to the data of different industry odor intensity levels 1-5, and the validation database is used to evaluate the effect. According to the evaluation results, the optimal algorithm with the smallest root mean square error (RMSE) is selected as the odor concentration simulation quantitative algorithm for the industry under the odor intensity.

[0030] The root mean square error (RMSE) calculation formula is as follows:

[0031]

[0032] Where RMSE i，j is the root mean square error of the model for the i-th industry and the j-th odor intensity, N is the number of validation samples, X k is the simulated odor concentration of the k-th sample, Y k is the measured odor concentration of the k-th sample.

[0033] The odor intensity of 0 level of malodorous gas is low, and the difference between different industries is small, so the quantitative model of odor intensity 0 level will be selected together with all the data of odor intensity 0 level in the training database, and all the data of odor intensity 0 level in the validation database will be verified together. The optimal algorithm with the smallest root mean square error (RMSE) is selected as the odor concentration simulation quantitative algorithm for the odor intensity 0 level.

[0034] Through training and verification, the optimal odor concentration simulation quantitative model for all industries under different odor intensities is obtained, and the quantitative model library corresponding to the industry is established.

[0035] S5: Use the electronic nose to walk, when a high value appears, combine with the odor intensity data of artificial field olfaction, use the qualitative model to automatically identify the industry and select the quantitative model, and determine the odor concentration in real time.

[0036] S5 includes the following steps:

[0037] (1) Through data collection and complaint analysis, the number and distribution of regional enterprises are mastered, the walking monitoring range, walking time period, and walking route are determined, and the odor walk monitoring is carried out under the same wind direction in the relatively concentrated time of odor complaint.

[0038] (2) Through data research, understand the industries of enterprises in the monitoring area, screen the industries related to malodor in the area, and select the above industries in the malodor electronic nose or sailing monitoring system. If the sailing monitoring system is selected, the system sends the selected industries to the malodor electronic nose.

[0039] (3) According to the planned route, synchronous field olfaction and continuous monitoring are carried out. Field olfaction uses olfactory personnel to determine the odor intensity of the atmospheric environment in real time, and continuous monitoring uses the malodor electronic nose to determine the odor concentration in real time.

[0040] (4) When high odor value appears, usually when the odor intensity is greater than or equal to level 2, the olfactory personnel input the odor intensity into the malodor electronic nose or sailing monitoring platform. If the sailing monitoring system is input, the system sends real-time odor intensity data to the malodor electronic nose, and the malodor electronic nose identifies the industry according to the odor intensity data using the qualitative model, and selects the quantitative model corresponding to the industry and the odor intensity to determine the odor concentration.

[0041] (5) If the environmental odor intensity does not change, the quantitative model is used to automatically and continuously monitor the odor concentration of the atmospheric environment, which is displayed on the malodor electronic nose and uploaded to the sailing monitoring system, realizing automatic and continuous monitoring of the odor concentration.

[0042] S6: When the environmental odor intensity changes, the industry is identified again using the qualitative model and the quantitative model is selected to monitor the odor concentration in real time.

[0043] When the environmental odor intensity changes, the olfactory personnel input the new odor intensity into the malodor electronic nose or sailing monitoring platform, repeat step (4) of S4 to identify the industry using the qualitative model, if the industry changes, select the quantitative model again, and continue step (5) of S4 to automatically and continuously monitor the odor concentration of the atmospheric environment.

[0044] S7: Draw a malodor map, analyze and arrange the sailing monitoring data, and form a malodor sailing monitoring report.

[0045] When the sailing monitoring is completed, the report preparation work is carried out, and the steps are as follows:

[0046] (1) Draw a malodor map that can reflect the sailing route, real-time monitoring results of odor concentration, position and other information. The odor map needs to visually display the odor concentration of different sailing points on the map. The visual display methods include color differentiation according to odor concentration, height differentiation by column chart, and color and column chart height differentiation.

[0047] (2) Analyze the underway monitoring data, combine the foul odor smell map and the on-site smell identification results of the smell identification personnel, and compile a foul odor smell event list, including the foul gas high value number (which needs to be marked on the foul odor smell map), the nearby enterprise name, the foul gas concentration, the foul gas intensity, the smell characteristics and the occurrence frequency (including frequent, intermittent and occasional), the foul gas intensity, the smell characteristics and the occurrence frequency are judged and recorded by the smell identification personnel on site, and the foul gas concentration is the data of the foul odor electronic nose monitoring.

[0048] (3) Analyze and describe the overall situation and characteristics of the foul odor smell underway monitoring area.

[0049] (4) Integrate the above work contents to form a foul odor smell underway monitoring report.

[0050] The technical effect brought by the technical scheme of the present application is:

[0051] (1) The present application forms a semi-adaptive, qualitative and quantitative coupling foul odor smell underway monitoring method through pattern recognition and concentration simulation coupling technology, solves the problem of low accuracy of monitoring foul gas concentration using instruments due to changes in environmental foul gas components during foul odor smell underway monitoring, and improves the actual effect of foul odor smell underway monitoring.

[0052] (2) According to the characteristics of the response relationship between foul odor gas and sensors, the present application establishes a method of pattern recognition training according to foul gas intensity, improves the accuracy of using pattern recognition method to identify industries, and meets the requirement of accurately identifying industries during foul odor smell underway monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 is a flowchart of the present application.

[0054] Figure 2 is an example of a foul odor smell map in the S7 step of the present application in the specific implementation mode. DETAILED DESCRIPTION

[0055] The present application uses pattern recognition and concentration simulation coupling, preliminarily screens the source of foul odor smell in the environment through pattern recognition, automatically selects the corresponding foul gas concentration simulation quantitative model according to the pattern recognition result, judges and reselects the corresponding model when the components of foul odor gas in the environment change, thereby improving the accuracy of foul gas concentration during foul odor electronic nose underway monitoring. The underway monitoring technical route is as shown in Figure 1 . The specific steps are:

[0056] S1: Collect concentration gradient foul gas samples by industry, test the foul gas concentration and the foul gas intensity of the samples.

[0057] According to the Notification on the Issue of <Analysis of National Odor Complaints in 2018-2020> issued by the Ministry of Ecology and Environment, combined with the typical odor source types in daily life and work, through the collection of garbage, sewage, livestock and poultry, rubber, plastic, chemical (which can be further subdivided into daily chemicals, pesticides, coatings, etc. according to the characteristics of the industry), pharmaceutical manufacturing and other industries that are prone to cause odor pollution, high-concentration odor samples of typical process links in each industry are collected, including organized and unorganized links. A concentration gradient sample library covering odor intensity 0-5 is established by diluting the samples, wherein each industry (TC in the present invention) has at least 6 samples at each concentration gradient corresponding to the odor intensity, two-thirds of which are used for training the qualitative model and the quantitative model, and one-third of which are used for verifying the qualitative model and the quantitative model.

[0058] The odor concentration and odor intensity of the sample are measured by an olfactant, the odor concentration is measured according to <Determination of Odor in Ambient Air and Waste Gas - Three-point Comparison Odor Bag Method> (HJ1262-2022), and the odor intensity is directly measured by an olfactant according to Table 1 standard for odor intensity classification.

[0059] Table 1 Odor Intensity Classification Table

[0060] Rank Description 0 No odor 1 Slight odor 2 Faint odor, but can determine what it is 3 Obvious odor 4 Strong odor 5 Very strong, almost unbearable odor

[0061] S2: Test the odor electronic nose using the concentration gradient odor samples of each industry, and obtain the relationship between the response signals of each sensor and the odor concentration and odor intensity of the sample through feature extraction.

[0062] The concentration gradient odor samples of each industry are sequentially introduced into the odor electronic nose for testing, and according to the response speed of the odor electronic nose (the response speed of different odor electronic noses is different, and the present invention is applicable to various types of odor electronic noses), when the response signal of the odor electronic nose enters the stable stage, the response signal of each sensor is extracted as a feature value Wherein n is the sensor number, i is the corresponding industry, j is the corresponding odor intensity, and the relationship between the sensor response signal and the odor concentration and intensity is recorded in the table, wherein two-thirds of the concentration gradient sample library is used for testing the data as a training database, and one-third of the concentration gradient sample library is used for testing the data as a verification database.

[0063] S3: Establish a qualitative model for industry classification using a pattern recognition method and insert it into the odor electronic nose.

[0064] The present invention uses a support vector machine algorithm (SVM) as a pattern recognition algorithm, uses the data in the training database to train the qualitative model for industry classification, and selects a Gaussian kernel function as the kernel function, i.e.

[0065]

[0066] wherein are the eigenvectors of the ith1 and ith2 industries under odor intensity j1 and j2, ith1 and ith2 can be the same industry, j1 and j2 must be different odor intensities, and σ is the Gaussian kernel function parameter for controlling the distance after mapping.

[0067] The eigenvectors are wherein is the odor intensity of the industry i under odor intensity j, is the sensor response signal eigenvalue of the sample of the industry i under odor intensity j obtained by the n sensor in the S2 step.

[0068] Further, the qualitative model is verified, the data of the verification database is substituted into the qualitative model, whether the industry discrimination result identified by the qualitative model corresponds to the actual industry is compared, the discrimination success rate is verified by using the following formula, if the discrimination success rate ≥ 95% indicates that the qualitative model can distinguish most industries, the next step is continued, if the discrimination success rate < 95% the Gaussian kernel function parameter σ is adjusted or more samples are collected for training, until the discrimination success rate ≥ 95% is reached.

[0069]

[0070] wherein RS is the industry discrimination success rate, N S is the number of successful discrimination, N T is the total amount of verification data.

[0071] S4: Under different odor intensities of each industry, the most suitable algorithm is selected from the simulation quantitative algorithm library to establish the odor concentration simulation quantitative model, and the quantitative model library corresponding to the industry is established and put into the electronic nose.

[0072] Since the emission characteristics of different industries of malodorous odor pollutants are different, the applicable odor concentration simulation quantitative model is also different, therefore, the partial least squares regression algorithm (PLSR), the support vector regression algorithm (SVR) and the artificial neural network algorithm (ANN) are used as the odor concentration simulation quantitative algorithm library, the training database is used to train the algorithm according to the data of different industries under odor intensity of 1-5 levels, the verification database is used to evaluate the effect, and the optimal algorithm with the smallest root mean square error (RMSE) is selected as the odor concentration simulation quantitative algorithm of the industry under the odor intensity according to the evaluation result.

[0073] The root mean square error (RMSE) calculation formula is as follows:

[0074]

[0075] where RMSE is the root mean square error of the model for the ith industry and jth odor intensity, N is the number of validation samples, X i，j k is the simulated odor concentration of the kth sample, Y k is the measured odor concentration of the kth sample.

[0076] Odor intensity of 0 level is low concentration, and there is little difference between different industries. Therefore, the quantitative model of odor intensity 0 level will be trained together with all data of odor intensity 0 level in the training database, and all data of odor intensity 0 level in the validation database will be validated together. The optimal algorithm with the smallest root mean square error (RMSE) is selected as the simulated quantitative algorithm of odor concentration of odor intensity 0 level.

[0077] Through training and validation, the optimal odor concentration simulation quantitative model of all industries at different odor intensities is obtained, and the quantitative model library corresponding to the industry is established, as shown in Table 2, which is built into the malodor electronic nose.

[0078] Table 2 Odor concentration simulation quantitative model library

[0079]

[0080] S5: Use the malodor electronic nose to walk, when a high value appears, combine with the odor intensity data of artificial field olfaction, use the qualitative model to automatically identify the industry and select the quantitative model, and real-time determine the odor concentration.

[0081] S5 includes the following steps:

[0082] (1) Through data collection and complaint analysis, the number and distribution of regional enterprises are mastered, the walking monitoring range, walking time period and walking route are determined, and the odor walk monitoring is carried out under the same wind direction in the relatively concentrated time of malodor odor complaints and wind speed 8 m / s. s The following, no precipitation weather.

[0083] (2) Through data research, the industries of enterprises in the walking monitoring area are understood, the industries related to malodor odor in the area are screened, and the above industries are selected in the malodor electronic nose or walking monitoring system. If selected in the walking monitoring system, the system sends the selected industry to the malodor electronic nose.

[0084] (3) According to the planned route, field olfaction and continuous monitoring are carried out simultaneously. Field olfaction uses olfaction staff to determine the odor intensity of the atmospheric environment in real time, and continuous monitoring uses the malodor electronic nose to determine the odor concentration in real time.

[0085] ​(4) When the high odor value appears, usually when the odor intensity is greater than or equal to level 2, the odor intensity is input into the electronic nose or the monitoring platform, and if the monitoring system is used, the system sends real-time odor intensity data to the electronic nose, and the electronic nose identifies the industry according to the odor intensity data using the qualitative model, and selects the quantitative model corresponding to the industry and the odor intensity to determine the odor concentration.

[0086] (5) If the environmental odor intensity does not change, the quantitative model is used to automatically and continuously monitor the atmospheric environmental odor concentration, which is displayed on the electronic nose and uploaded to the monitoring system, realizing automatic and continuous monitoring of the odor concentration.

[0087] S6: When the environmental odor intensity changes, the qualitative model is used to identify the industry and select the quantitative model, and the odor concentration is monitored in real time.

[0088] When the environmental odor intensity changes, the new odor intensity is input into the electronic nose or the monitoring platform, and the step (4) of S4 is repeated to identify the industry using the qualitative model, if the industry changes, the quantitative model is selected again, and the step (5) of S4 is continued to automatically and continuously monitor the atmospheric environmental odor concentration.

[0089] S7: Draw the odor map, analyze and arrange the data of the monitoring, and form the odor monitoring report.

[0090] When the monitoring is completed, the report is prepared, and the steps are as follows:

[0091] (1) Draw the odor map reflecting the monitoring route, real-time monitoring results of the odor concentration, and location information, and the odor map needs to visually display the odor concentration of different monitoring points on the map, and the visual display methods include color differentiation according to the odor concentration, height differentiation by column chart, and color and column chart height differentiation together, as shown in FIG. 1. Figure 2

[0092] (2) Analyze the monitoring data, combine the odor map and the on-site odor identification results of the odor identification personnel, and compile the odor event list, as shown in Table 3, including the odor high value number (which needs to be marked on the odor map), the nearby enterprise name, the odor concentration, the odor intensity, the odor characteristics, and the occurrence frequency (including frequent, intermittent, and occasional), the odor intensity, the odor characteristics, and the occurrence frequency are judged and recorded by the odor identification personnel on site, and the odor concentration is the monitoring data of the electronic nose.

[0093] Table 3 Odor Event List (Example)

[0094]

[0095] ​

[0096] (3) Analyze and describe the overall situation and characteristics of the mobile monitoring area for malodorous odors.

[0097] (4) Integrate the above work content to form a mobile monitoring report on malodorous substances.

[0098] The above embodiments are merely preferred embodiments of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for monitoring malodorous odor at sea based on coupling of pattern recognition and concentration simulation, characterized in that, The method comprises the following steps: S1: Collecting concentration gradient odor samples from different industries, testing the odor concentration and odor intensity of the samples; S2: Testing the odor samples on the electronic nose, and obtaining the relationship between the sensor response signal and the odor concentration and odor intensity of the samples through feature extraction; S3: Establishing an industry classification qualitative model using pattern recognition method and placing it in the electronic nose; S4: Selecting the most suitable algorithm from the quantitative algorithm library to establish an odor concentration simulation quantitative model under different odor intensities of each industry, and establishing a quantitative model library corresponding to the industry and placing it in the electronic nose; S5: Using the electronic nose for field monitoring, when a high value appears, combining the odor intensity data obtained by artificial field olfaction, using the qualitative model to automatically identify the industry and select the quantitative model, and real-time determining the odor concentration; The industry-specific concentration gradient odor samples are sequentially introduced into the malodor electronic nose for testing. According to the response speed of the malodor electronic nose, when the response signal of the malodor electronic nose enters the stable stage, the response signal of each sensor is extracted as the characteristic value wherein n is the sensor number, i is the corresponding industry, and j is the corresponding odor intensity, and recorded in the sensor response signal and odor concentration, intensity relationship table, wherein two-thirds of the concentration gradient sample library is used as the training database, and one-third of the concentration gradient sample library is used as the verification database; Specifically, the method comprises the following steps: (1) Through data collection and complaint analysis, the number and distribution of enterprises in the region are mastered, the monitoring range, monitoring time and monitoring route are determined, and odor field monitoring is carried out under the condition of relatively concentrated odor complaint time and same wind direction; (2) Through data research, the industries of the enterprises in the monitoring area are understood, the industries related to odor are screened, and the industries are selected in the electronic nose or the monitoring system, and if the industries are selected in the monitoring system, the system sends the selected industries to the electronic nose; (3) Field olfaction and continuous monitoring are carried out simultaneously according to the planned route, the odor intensity of the atmospheric environment is determined by the olfaction personnel in real time, and the odor concentration is determined by the electronic nose in real time; (4) When a high value of odor appears, usually when the odor intensity is greater than or equal to 2, the olfaction personnel input the odor intensity into the electronic nose or the monitoring platform, if the industries are selected in the monitoring system, the system sends the real-time odor intensity data to the electronic nose, the electronic nose identifies the industry according to the odor intensity data using the qualitative model, and selects the quantitative model corresponding to the industry to determine the odor concentration; (5) If the odor intensity of the environment does not change, the quantitative model is used to automatically and continuously monitor the odor concentration of the atmospheric environment, the odor concentration is displayed on the electronic nose and uploaded to the monitoring system, and the automatic and continuous monitoring of the odor concentration is realized. ​ ​ ​ S6: When the environmental odor intensity changes, re-use the qualitative model to identify the industry and select the quantitative model, and monitor the odor concentration in real time; When the environmental odor intensity changes, the panelist inputs the new odor intensity into the electronic nose or the underway monitoring platform, repeats step (4) of S5 to identify the industry using the qualitative model, re-selects the quantitative model if the industry changes, and continues step (5) of S5 to automatically and continuously monitor the atmospheric environmental odor concentration; S7: Draw the odor and smell map, analyze and arrange the underway monitoring data, and form the odor and smell underway monitoring report.

2. The method according to claim 1, wherein the method is characterized by, S3 specifically includes the following processes: Use the support vector machine algorithm SVM as the pattern recognition algorithm, train the industry classification qualitative model using the data in the training database, and select the Gaussian kernel function as the kernel function, that is: ， wherein , are the eigenvectors of the ith1and ith2industry at the odor intensity level of j1and j2, respectively, ith1and ith2may be the same industry, and j1and j2must be different odor intensities, is the parameter of the Gaussian kernel function used to control the distance after mapping; The characteristic vector is wherein is the odor intensity j of the industry i, is the sensor response signal characteristic value of the sample of odor intensity j of the industry i obtained by the n sensor in the S2 step; The qualitative model is verified, data of the verification database is substituted into the qualitative model, whether the industry discrimination result identified by the qualitative model corresponds to the actual industry is compared, and the following formula is used to verify the discrimination success rate. If the discrimination success rate is greater than or equal to 95%, it is indicated that the qualitative model can distinguish most of the industries, and the next step is continued. If the discrimination success rate is less than 95%, the Gaussian kernel function parameters are adjusted or more samples are collected for training until the discrimination success rate is greater than or equal to 95%. or more samples are collected for training until the discrimination success rate is greater than or equal to 95%. ， wherein RS is the industry discrimination success rate, N S is the number of successful discriminations, N T is the total amount of validation data.

3. The method according to claim 1, wherein the method is characterized by, S4 specifically includes the following processes: Use the partial least squares regression algorithm PLSR, the support vector regression algorithm SVR, and the artificial neural network algorithm ANN as the odor concentration simulation quantitative algorithm library, train the algorithms using the data in the training database according to different industry odor intensities of 1-5 levels, evaluate the effect using the verification database, and select the optimal algorithm with the smallest root mean square error RMSE as the odor concentration simulation quantitative algorithm for the industry under the odor intensity according to the evaluation result; The root mean square error RMSE calculation formula is as follows: ， wherein is the root mean square error of the model for the ith industry jth odor intensity, N is the number of validation samples, is the simulated odor concentration of the kth sample, is the measured odor concentration of the kth sample; The quantitative model for odor intensity 0 will train all data with odor intensity 0 together in the training database, and validate all data with odor intensity 0 together in the verification database, and select the optimal algorithm with the smallest root mean square error RMSE as the odor concentration simulation quantitative algorithm for odor intensity 0. Through training and verification, the optimal odor concentration simulation quantitative model for all industries under different odor intensities is obtained, and the quantitative model library corresponding to the industry is established.

4. The method according to claim 1, wherein the method is characterized by, In S7, when the underway monitoring is completed, the report preparation work is carried out, and the steps are as follows: (1) Draw the odor and smell map that can reflect the underway route, odor concentration real-time monitoring result, and position information. The odor and smell map needs to visually display the odor concentrations of different underway points on the map. The visual display methods include color differentiation according to odor concentration, height differentiation by column chart, and color and column chart height differentiation together; (2) Analyze the underway monitoring data, combine the odor and smell map and the panelist's on-site olfactory results, compile the odor and smell event list, including odor high value number, nearby enterprise name, odor concentration, odor intensity, odor characteristics, and occurrence frequency. The odor intensity, odor characteristics, and occurrence frequency are judged and recorded by the panelist on site, and the odor concentration is the data monitored by the electronic nose; (3) Analyze and describe the overall situation and characteristics of the odor and smell underway monitoring area; (4) Integrate the above work contents to form the odor and smell underway monitoring report.

Images