Electronic nose instrument operable in both scheduled and on-demand modes and method for online real-time detection and analysis of multi-component odors

Abstract

An electronic nose instrument operable in both scheduled and on-demand modes and a method for online real-time detection and analysis of multi-component odors. A hardware unit of the electronic nose instrument mainly comprises: a gas-sensitive sensor array module (I), a headspace sampling module (II), a pressurization cylinder (III), a computer control and analysis module (IV), a backup power supply (V), and a clean air cylinder (VI). A main housing integrates the first four components. Within a cycle time T=180-600 s, the pressurization cylinder (III) significantly increases a gas-sensitive response by means of short-term pressure multiplication. The gas-sensitive sensor array obtains a 50-dimensional sensing sample for single detection. A large odor dataset X comprises online detection data from the electronic nose instrument, and offline detection data from olfactometry and chromatography etc. The detection data is decomposed into multiple single-concentration sub-tasks. A machine learning cascade model is formed by multiple learning groups consisting of single neurons, and shallow and deep neural networks. The electronic nose instrument can flexibly achieve online real-time identification of odor pollutants and multi-component concentration estimation and prediction.

Description

A fixed and active two-convenient electronic nose instrument and a malodor multi-component online real-time detection and analysis method TECHNICAL FIELD

[0001] The present application is a fixed and active two-convenient electronic nose instrument and a malodor multi-component online real-time detection and analysis method, based on the GB14554 "Malodor Pollutant Emission Standard" (draft for comments) published by the Ministry of Ecology and Environment on December 3, 2018, aiming at the monitoring and analysis of malodor pollution status in chemical industry parks, landfill sites, sewage treatment plants, livestock and poultry farms, and adjacent residential areas, involving intelligent sensing, machine learning, computer, analytical chemistry, ecological environment and other scientific and technological fields, mainly solving the problems of unsatisfactory sensitivity and selectivity of existing electronic nose devices, poor online real-time capability, poor applicability, and information selection uncertainty, to flexibly realize the continuous online or on-site real-time detection and analysis of malodor pollutants. BACKGROUND

[0002] Green environmental protection industry is a strategic emerging industry identified in the "Twelfth Five-Year" and "Fourteenth Five-Year" National Economic and Social Development Plan. On December 27, 2023, the "Opinions of the Central Committee of the Communist Party of China and the State Council on Fully Promoting the Construction of Beautiful China" proposed to "establish a modern ecological environment monitoring system and focus on solving malodor, cooking fume and other pollution problems." On September 19, 2022, the "Fourteenth Five-Year Plan for Scientific and Technological Innovation in the Field of Ecological Environment" proposed by the Ministry of Science and Technology, the Ministry of Ecology and Environment and other five departments proposed to "research atmospheric malodor pollution online monitoring and impact assessment technology." On March 3, 2020, the "Guiding Opinions on Building a Modern Environmental Governance System" (hereinafter referred to as the "Guiding Opinions") proposed by the Central and State Offices proposed to implement the "double random, one public" supervision mode, strengthen monitoring capacity building, achieve full coverage of environmental quality, pollution sources and ecological conditions monitoring, promote the development of precise, rapid and portable monitoring equipment, ensure that monitoring data is "true, accurate and complete", and comprehensively improve the level of monitoring automation, standardization and informatization. These are important starting points for the present application.

[0003] Odor, also known as malodor, is a short form of odor pollution, which refers to offensive odor and generally refers to all gas substances that stimulate the olfactory organ and cause people to feel unpleasant and harm the living environment. Odor pollutants exist widely in biological chemical industry and pharmaceutical industry, garbage and sewage treatment, livestock breeding and other enterprises, and adjacent residential areas. One of the current situations of odor / malodor pollution in China is that there are many emission sources, the odor composition is complex, the national standard is lagging behind, residents complain frequently, and accidents occur occasionally. For many years, the proportion of environmental complaints has been more than 20%, becoming the second largest source of environmental complaints. The second situation is that the monitoring results meet the national standard, but the residents cannot tolerate it, and the complaints continue. Odor is a serious public nuisance and a serious threat to health. The current Environmental Protection Law of the People's Republic of China lists it as one of the seven major pollutants to be prevented and controlled, and odor pollution control has been included in the "Twelfth Five-Year" and "Fourteenth Five-Year" national environmental protection plans.

[0004] On December 3, 2018, the Ministry of Ecological Environment released the GB14554 Odor Pollutant Emission Standard (Draft for Comments), which stipulates that the odor pollutant emission control indicators consist of eight single component concentrations of hydrogen sulfide H2S, ammonia NH3, methyl mercaptan CH4S, dimethyl disulfide C2H6S2, methyl sulfide C2H6S, carbon disulfide CS2, trimethylamine C3H9N, and styrene C8H8, and one overall odor concentration, i.e. 8+1 quantitative indicators. GB14554-1993 has been implemented for 31 years and has been unable to adapt to the current odor pollution situation, but the new standard has not been issued for a long time, indicating that the problem of odor pollution qualitative and quantitative analysis is very complex. Many experts call for the revision and improvement of odor emission and monitoring standards to build a prevention and control standard system.

[0005] Both the current standard and the new standard (draft for comments) of GB14554 stipulate that the three-point comparison odor bag method is used to determine the odor concentration, the gas chromatography method is used to determine the concentrations of H2S, CH4S, C2H6S, C2H6S2, C3H9N and C8H8, and the spectrophotometry method is used to determine the concentrations of NH3 and CS2. According to the industry standard HJ1262-2022 "Environmental Air and Waste Gas-Odor Determination-Three-Point Comparison Odor Bag Method" implemented in 2023, "odor concentration" is the dilution factor when the odor sample collected on site is continuously diluted with odorless clean air to the olfactory threshold of the odorant in the olfactory chamber; the EU standard EN17325-2003 uses OU (odor unit) value to measure. HJ1262 standardizes the selection of odorants, sample collection, manual dilution, and manual olfaction.

[0006] The dilemma of the sensory method for determining the odor concentration is that many odor substances either have no olfactory threshold or have different values given by different countries, regions or organizations. The selection is entirely dependent on the subjective judgment of the panelists after smelling [6-7]. In 2015, the National Environmental Protection Odor Pollution Control Key Laboratory in Tianjin organized 30 panelists (13 men and 17 women) to determine the olfactory threshold of 40 odor substances [7]. The NH3 olfactory threshold in Tianjin is 5 times different from the value in Japan, the H2S value is nearly 3 times different, the trimethylamine value is 28.12 times different, and the n-pentanoic acid value is even 65.67 times different. These facts at least indicate two problems: (1) the sensory determination of the odor concentration is complex, time-consuming and labor-intensive; (2) the olfactory threshold is not consistent and objective. Human olfaction can sense more than 4,000 odor-causing components, and an odor often contains dozens or hundreds of them. The olfactory method is harmful to health, and the conventional instrumental method such as chromatography is complex in operation and harsh in testing conditions. The common defects of the two methods are high cost, low efficiency and poor real-time performance, which do not meet the requirements of the Guidance. Therefore, the machine olfaction method and instrument are particularly attractive [8-9].

[0007] Olfaction is a complex sense of a large number of olfactory cells in the nasal cavity of an organism, and is the most mysterious of the five senses. Odor is invisible and intangible, and is a mixture of dozens, hundreds or even thousands of components, which is a complex perception of the olfactory system of an organism. In 2004, Axel and Buck, winners of the Nobel Prize in Physiology or Medicine, elucidated the working principle of the biological olfactory system: about 1,000 receptor cells perceive complex odor molecules to form a pattern or sample of information, and the brain memorizes and recognizes thousands of odors according to the pattern or sample

[0010] . The biological olfactory system inspires and enlightens the machine olfactory system, i.e. electronic nose, and the working principle has been imitated [11-12]. The machine olfactory system uses an array of multiple gas sensors with partially overlapping sensing ranges to perceive and analyze simple and complex odors in multiple scales to determine the category of flavoring substances, authenticity, freshness, odor intensity, odor grade, etc. Before AD 2000, the machine olfaction research focused on the identification of simple gases, and today the application objects are mainly qualitative and quantitative analysis of complex odors, such as the determination of the quality of spices, tea, wine, milk and oil; the judgment of the maturity of fruits and the freshness of meat products; the monitoring of air and water; the diagnosis of diseases; the discovery of drugs and bacteria; etc. In summary, odor is one of the basic properties of a substance, and there is no absolutely odorless substance.

[0008] One of the trends of machine olfaction is to develop gas sensors with high sensitivity and high selectivity. It is encouraging that the gas sensing sensitivity of SnO2 metal oxide semiconductor (MOS) material has reached 10 -8The second trend is to develop new working principles and new structures of electronic nose instruments guided by typical applications. An array of multiple gas sensors with overlapping sensing ranges is used to optimize the design of the structure of the gas sensor array and other functional components. A small, integrated, and automated electronic nose instrument is developed, and its working parameters are optimized to significantly improve the overall selectivity and sensitivity of the instrument. The third trend is to establish a large dataset of complex odors, invent simple and efficient information enhancement techniques and machine learning methods, including gas sensor response drift compensation, to achieve online real-time identification of complex odors and prediction of multiple concentration indicators.

[0009] Machine olfaction methods have broad application prospects. The relevant search results are as follows: (1) Papers. There were only 60 papers before the 1990s, more than 500 papers in the 2000s, and nearly 13,000 papers in total, indicating that this research has been widely carried out since the 2000s. (2) Invention patents. More than 2,500 international patents and more than 700 domestic patents were mostly applied for or granted in the past 10 years, showing that machine olfaction intellectual property protection has been taken seriously. (3) Technical standards. In 2020, the IEEE Standards Association released technical standards related to artificial olfaction devices and systems. (4) Domestic related work is basically laboratory research using foreign electronic nose products [9]. The above results show that machine olfaction theory and application research needs to be further developed.

[0010] Online real-time monitoring, estimation, and prediction of 8+1 concentration indicators of malodor pollutants are basic requirements for the construction of smart cities, indicating the urgency of machine olfaction theory and technology for industrial applications. The harsh reality is that machine olfaction research, which began in the early 1960s and flourished in the 1980s, is still in the laboratory stage as a whole and is not yet mature, with almost no industrial applications

[0012] . For example, the FOX odor fingerprint analyzer from France entered the Chinese market in 2007, with a price of $150,000 or more than 1 million RMB, and was only used for laboratory offline detection, which cannot be used for online real-time detection of malodor sites. The parent company has stopped production and withdrawn from the Chinese market.

[0011] Machine olfaction systems were initially considered to be an instrument composed of an array of multiple gas sensors with overlapping performance and appropriate pattern recognition methods, which can identify simple and complex odors, and now more emphasis is placed on odor measurement and analysis capabilities

[0012] Unfortunately, the research of machine olfaction has been confined to the laboratory for more than 60 years. Typical gas sensitive materials include MOS, electro-chemical (EC), conducting polymer (CP), quartz microbalance (QMB), surface acoustic wave (SAW) and photo ionization detector (PID)

[0011] . Considering sensitivity, response speed, commercialization, life, size, simplicity and stability, the MOS gas sensor represented by SnO2 is the most widely used. The present application takes this as the basis for developing a two convenient odor electronic nose instrument.

[0012] The research of machine olfaction should be based on the revelation of gas sensitive perception mechanism, take the gas sensitive sensor array and machine learning method as the starting point, and be guided by industrial application. The three are interdependent and cannot be neglected, otherwise it will fall into the wrong area. For example, what is the micro perception mechanism of the MOS gas sensitive material represented by SnO2? From the perspective of material properties, there are grain boundary potential barrier changes, grain surface charge changes, etc. There is no consensus

[0011] . If we study the gas sensitive perception mechanism from the perspective of industrial application, we will find a new world.

[0013] Here are the meanings of several terms: on-site real-time - one-time or limited-time detection and analysis of odor or source material volatile gas on site, sample preparation and detection at the same time, detection and analysis cycle 5-10 min; continuous online - continuous multiple detection and analysis of odor gas on site, detection and analysis cycle still 5-10 min; identification / classification - determine the fragrance type, origin or true or false of the odor and its source material; estimate - quantify the current odor intensity level and main component concentration; predict - quantify the recent odor intensity level and component concentration. The present application takes continuous online detection, estimation and prediction of odor pollutants as the main application background, supplemented by on-site real-time identification and estimation, so as to organically combine the continuous online and on-site real-time two working modes of the electronic nose instrument.

[0014] According to the Web of Science core database, there are only about 200 papers on the application of machine olfaction methods to malodor detection analysis, less than 2% of the total number of papers in this field, and most of them are offline detection and data processing of indoor air, water, and soil samples[8-9]. The existing malodor monitoring system uses a "one-point-one-nose" camera, which is inevitably subjected to wind, sun, and rain for a long time, causing fatigue and "poisoning" of the gas-sensitive element, resulting in short service life, high cost, low sensitivity, and few concentration indicators, and a series of problems. The installation of the French RQ-BOX electronic nose at the Shanghai Laogang landfill and the installation of 102 German Olfosense electronic noses at the Tianjin Dagang Industrial Zone at a cost of 1483.5 million yuan have been repeatedly included in the top of environmental protection reports[5]. These primary applications do not consider the fundamental problems of poor selectivity of MOS elements, low sensitivity of EC elements compared to MOS by 1-2 orders of magnitude, and poor diversity of PID elements, and have limited effect

[0011] .

[0015] The working mode and performance level of existing malodor monitoring devices are far from the "random, accurate, portable, and fast" principles of the "Guidance". What are the key scientific problems and technical bottlenecks of electronic noses for malodor detection and analysis applications? First, the mechanism of gas-sensitive perception is not well understood, and there is a lack of research on the characteristics of complex odor composition, and the laboratory perception performance research is disconnected from the application. Second, there is a lack of research on big data and machine learning from the perspective of complex odor analysis needs, resulting in low online real-time analysis capability of olfactory simulation technology. Third, research has not been guided by industrial applications, resulting in the fact that olfactory simulation methods have not been able to leave the laboratory. In order to apply olfactory simulation methods and technology to online real-time detection and analysis of malodor pollutants, we must solve the following scientific and technical problems:

[0016] (A) Mechanism of gas-sensitive perception, concentration enhancement, and information selection problems guided by typical industrial applications

[0017] Theoretical and practical disconnection exists in the research on the mechanism of gas-sensitive perception. On the one hand, material researchers are working hard to find more sensitive gas-sensitive materials and develop more advanced manufacturing processes; on the other hand, there are countless gas compounds, and it is too arbitrary to conclude whether a material is sensitive or not based on laboratory tests of a limited number of compounds

[0011] . Therefore, it is particularly important to study the mechanism of gas-sensitive perception guided by applications.

[0018] The complex odor characteristics are that (i) there are many components and they are changeable. Except for a few inorganics such as H2S and NH3, most are organics, i.e. volatile organic compounds (VOC). (ii) Some compounds have very low odor threshold, but their contribution to odor intensity is very large; vice versa. (iii) Some compounds have very small contribution to odor intensity, but the gas sensing element is very sensitive; vice versa. Therefore, revealing the gas sensing perception mechanism must be from both the gas sensing element and the application object, and cannot be biased, especially when a material has good sensitivity performance. It is necessary to focus on the specific application object and application conditions.

[0019] One of the scientific problems to be solved is to establish odor big data and reveal the change rule of the micro-macro characteristics of the MOS gas sensing element and the typical compound molecular weight, structure, functional group, intensity, etc. (i) 8 single components and odor concentration specified in GB14554; (ii) 5 standard odor liquids and n-butanol concentration specified in HJ1262; (iii) ethanol concentration. Similar to the human body BMI index, we conceive that useful rules can be found by normalizing the molecular weight, structure, functional group, etc.

[0020] From the perspective of instrument design and industrial application, how to learn from existing gas pre-concentration technology and Langnuir adsorption and desorption principles is studied. By increasing the pressure for a short time, more trace odor molecules can be adsorbed on the surface of the sensitive film, so as to improve the sensitivity of gas sensing perception. Whether the steady-state response peak of a gas sensor to 1ppm gas A and 10ppm gas B will be completely equal? It is possible. This example tells us a truth: it is not feasible to select a single information from a single response curve, because it does not meet the triangle stability principle. Therefore, we should study how to select multiple information components from a single response curve, so as to meet the triangle stability principle and help improve selectivity and sensitivity.

[0021] (B) Simple and efficient, flexible and variable machine learning model and algorithm to solve many complex odor online real-time analysis problems

[0022] In terms of industrial application, the use of electronic nose instrument is limited if it is only used for odor identification. More importantly, it can perform online real-time estimation and prediction of odor intensity, quality level and concentration of multiple main components. The machine learning method in the olfactory simulation system has its own particularity compared with image, speech and other fields, that is, the machine learning model and algorithm not only need to complete the identification of odor and its source material, but especially need to complete the current state estimation and future state prediction of complex odor. It needs to flexibly complete multiple tasks such as identification, identification + estimation, estimation, estimation + prediction. This invention focuses on the latter two.

[0023] In view of the diversity of odor components and concentrations and the variability of atmospheric environment, it is difficult to establish a reliable analysis model based on small data sets. It must be pointed out that the relevant papers are basically in the laboratory using conventional single model and algorithm to identify small sample and few categories of odor offline [8-9, 11-12], which is far from the requirements and expectations of people for industrial applications of electronic nose instruments, and indirectly explains why it has not been out of the laboratory.

[0024] Big data has profoundly changed people's way of life and thinking. In order to flexibly perform identification, identification + estimation, estimation + prediction and other multi-type analysis tasks, we must propose simple and efficient machine learning models and algorithms, combine offline learning of odor big data with online learning of recent small data sets, and achieve the purpose of multi-parameter online real-time analysis and prediction. Therefore, the scientific problems to be solved include: ① How to divide the whole analysis problem of multiple complex odors into multiple relatively simple binary classification, single concentration and multiple single time series sub-data set analysis problems based on the "divide and conquer" principle and the "carry forward and start later" rule; ② How to use as many simple and effective machine learning sub-models and algorithms of different types as possible to learn the aforementioned multiple sub-data sets according to the Occam razor law - "simple and effective" law; ③ Discover simple and effective decision rules to form flexible machine learning models to complete multi-type odor analysis tasks.

[0025] (C) Machine olfactory system multi-working parameter combination optimization and online real-time industrial application problem

[0026] The present application project is oriented by the urgent need for instruments to replace human odor analysis, by improving the multi-parameter intelligent analysis capability of machine olfactory system, and by realizing the industrial application of complex odor multi-index online real-time analysis.

[0027] One of the scientific and technical problems to be solved here is: from the perspective of machine olfactory system design and industrial application, simulate human breathing, study how to optimize the combination of sample size, volatile gas capacity, sample flow rate, pressure, temperature, and time length, and achieve rapid cyclic adsorption and desorption of sensitive membrane to complex odor with large sample size and large sample flow rate, internal precision "invariable" to external "change", online real-time, fixed and mobile, indoor and outdoor, to significantly improve the gas sensitivity and repeatability.

[0028] Another scientific and technical problem to be solved is how to establish a large-scale multi-scale field gas sensing-based odor big data by means of a new gas sensing mechanism and a simple and efficient machine learning method, support the analysis results with detailed and reliable analysis results, and incorporate the machine olfactory method into the industry standard for online real-time analysis of malodor pollution state and various malodor pollutants, solve the virtualization problem of various malodor pollutant indicators in the existing national / industry standard of malodor field, and realize real industrial application.

[0029] The present application breaks through the industrial application technology bottleneck of olfactory simulation method and instrument according to the general requirement of "demand-driven, breakthrough bottleneck", detects high content and micro trace components, reveals the change rule of complex odor process of malodor, proposes a new model and algorithm of machine learning, realizes multi-scale online real-time detection and multi-state parameter intelligent analysis of malodor, and meets the urgent needs of environmental protection industry.

[0030] REFERENCES

[0031] [1] Wang Zhi, Xiao Haichuan, 7 deaths and 4 injuries in a gas poisoning accident in Shouguang, Shandong, a biotechnology company, Xinhua News Agency, www.news.cn / local / 20241013 / 3e97b718ff564a61b568b5b16b0690f3 / c.html, October 13, 2024.

[0032] [2] Xu Hui, Xu Hui, Chengdu Dayi residents have complained about the odor many times, response: emissions meet the standard still have odor, will solve in many ways, Pingo News, August 21, 2024, m.thepaper.cn / newsDetail_forward_28480792.

[0033] [3] Tang Yijun, a dairy farm in Luyang, Jinshan, the stench of the sky makes villagers unbearable, don't open the window all day, there is also a peculiar smell in the room, People's Network, August 21, 2023, sh.people.com.cn / n2 / 2023 / 0821 / c176737-40539775.html.

[0034] [4] Diao Fanchao, high volume of complaints about odor and peculiar smell, the Ministry of Environmental Protection will make relevant arrangements for "odor and peculiar smell control", Pingo News, July 14, 2023, www.thepaper.cn / newsDetail_forward_23841935.

[0035] [5] Ministry of Ecology and Environment, Air Environment Department, Letter on the issuance of the report on the analysis of national odor pollution complaints in 2018-2020, 2021.07.26, wzq1.mee.gov.cn / xxgk2018 / xxgk / sthjbsh / 202108 / t20210802_853623.html.

[0036] [6] Xue L, Odor pollution, how to achieve "national standard" and "people's standard" double standard? China Environment APP, 2023.07.20.

[0037] [7] Wang H, Zhai Z, Geng J, et al., Determination of olfactory threshold values of 40 typical odorants, Journal of Safety and Environment, 2015, 15(6):348-351.

[0038] [8] Ma HL, Wang T, Li BL, et al., A low-cost and efficient electronic nose system for quantification of multiple indoor air contaminants utilizing HC and PLSR, Sensors & Actuators: B. Chemical, 2022, 350:130768.

[0039] [9] He M, Xiong LJ, Han SY, et al., Rapid detection of mixed odor intensity level in composting plants based on E-nose, IEEE Sensors Journal, 2023, 23(22):27795-27803.

[0040]

[0010] Buck L, Axel R, A novel multigene family may encode odorant receptors: a molecular basis for odor recognition, Cell, 1991, 65(1):175-187.

[0041]

[0011] Boeker P, On 'Electronic Nose' methodology, Sensors & Actuators B-Chemical, 2014, 204:2-17.

[0042]

[0012] Covington JA, Marco S, Persaud KC, et al., Artificial olfaction in the 21 st century, IEEE Sensors Journal, 2021, 21(11): 12969-12990.

[0043] SUMMARY

[0044] The present application is based on the existing invention patents "A multi-point centralized online monitoring and analysis system and method for malodorous gas" (see patent application number: 2018104716131) and "Big data driven multi-point centralized electronic nose instrument online analysis method for malodorous gas" (see patent application number: 2018104717083), and a fixed and active two-convenient electronic nose instrument and a malodorous multi-component online real-time detection and analysis method are invented to solve the problem of malodorous pollution site or long-term continuous online monitoring or short-term real-time detection, to realize the online real-time quantification estimation and prediction of odor concentration and the concentration of multiple malodorous compounds.

[0045] In order to achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0046] The hardware of the fixed and active two-convenient electronic nose instrument mainly includes a gas sensitive sensor array module I, a gas automatic headspace sampling module II, a supercharged cylinder III, a computer control and analysis module IV, a backup power supply V, and a clean air bottle VI. Among them, the gas sensitive sensor array module I, the gas automatic headspace sampling module II, the supercharged cylinder III, and the computer control and analysis module IV are the four components of the main machine box. The software mainly includes a computer operation platform, an upper and lower computer communication, and a machine learning model and algorithm; the electronic nose instrument flexibly realizes fixed or mobile detection of malodorous pollution site, malodorous type identification, and quantitative estimation and prediction of the concentration of multiple malodorous compounds.

[0047] The gas sensitive sensor array module I includes a gas sensitive sensor array I-1, a gas sensitive sensor array annular working cavity I-2, a heat insulation layer I-3, a partition plate I-4, a fan I-5, a resistance heating wire I-6, and a temperature sensor I-7, which are located in the upper right part of the electronic nose instrument main machine box. The environmental temperature and humidity sensor I-8 is located at the upper right corner of the inner wall of the front of the main machine box;

[0048] The gas automatic headspace sampling gas path module II includes: a micro vacuum pump II-1, a flow meter II-2, a throttle valve II-3, a first two-position two-way electromagnetic valve II-4, a second two-position two-way electromagnetic valve II-5, a third two-position two-way electromagnetic valve II-6, a fourth two-position two-way electromagnetic valve II-7, a fifth two-position two-way electromagnetic valve II-8, and a purifier II-9. The pressurized cylinder III includes: a piston III-1, a piston rod III-2, a cylinder body III-3, a compression spring III-4, an O-shaped sealing ring III-5, an inner snap spring III-6, a front cover III-7, and a bump pad III-8. The gas automatic headspace sampling gas path module II and the pressurized cylinder III are located at the lower right of the main machine box.

[0049] The main component units of the computer control and analysis module IV include: an A / D data acquisition card IV-1, a driving and control circuit board IV-2, a computer mainboard IV-3, a 4-way precision DC voltage stabilizer IV-4, a WIFI board card IV-5, and a touch display screen IV-6, which are located at the left side of the main machine box.

[0050] The clean air bottle VI includes: a 40L clean air large bottle VI-1 and 2-6L clean air small bottles VI-2. The 40L clean air large bottle VI-1 is mainly used for long-term continuous online detection of malodorous sites for more than 2 hours. The 2-6L clean air small bottles VI-2 are used for limited-time short-term online real-time detection of malodorous sites for less than 2 hours.

[0051] The gas sensitive sensor array I-1 is composed of 16 gas sensitive elements with partially overlapping sensing ranges. The gas sampling period of the dual-purpose electronic nose instrument for a malodorous observation point is T = 180-600s, and the default value is T = 300s. To meet the principle of triangular stability, in the kth period corresponding to the time kT, the computer control and analysis module IV selects the steady-state voltage peak value v i,0 (k), the corresponding peak time t i,0 (k), and the area A i,0 (k) under the curve from the response curves of the gas sensitive sensor array I-1 with a time length of 60s, i = 1, 2, …, 16. Thus, the computer control and analysis module IV extracts 3*16 = 48 response components from the 16 response curves of the gas sensitive sensor array I-1 in the period kT, and adds the local environmental temperature value T em,0 (k) and the environmental humidity value H um,0 (k) at that time, so that the electronic nose instrument obtains an initial sensing vector x0(k) = (v 1,0 (k), …, v 16,0 (k), t 1,0 (k), …, t 16,0 (k), A 1,0 (k), …, A16,0 (k),T em,0 (k),H um,0 (k)) T ∈R 50 , called initial sample, is stored in the data file of the computer mainboard IV-3 hard disk, and the data x0(k) is sent to the designated fixed / mobile terminal and cloud through the WIFI board card IV-5.

[0052] The dual-purpose electronic nose instrument learns the corresponding relationship between the gas sensor array response sample and the pollution type and the 8+1 concentration control index of the specified hydrogen sulfide H2S, ammonia NH3, methyl mercaptan CH4S, dimethyl disulfide C2H6S2, methyl sulfide C2H6S, carbon disulfide CS2, trimethylamine C3H9N, styrene C8H8 and one overall type odor gas concentration in GB14554-1993 "Odor Pollutant Emission Standard" through multiple odor pollution sites or long-term continuous online or short-term multiple detection, forming an odor big data set X.

[0053] In the learning stage, the machine learning cascade model software of the computer control and analysis module IV learns the odor big data set X offline to determine its structure and parameters, and master the corresponding relationship between the gas sensor array response sample and the pollution type and the 8+1 concentration control index. In the decision-making stage, according to the current perception vector x0(k) of the gas sensor array I-1, the machine learning cascade model online real-time quantitatively estimates the 8+1 odor pollutant concentration control index values specified by GB14554, and determines the odor pollution type as needed. Further, the machine learning cascade model predicts the future 1h perception response of the gas sensor array I-1 through online real-time learning of the long time sequence perception vector generated by the gas sensor array I-1 in the last 72 cycles about 6h, and accordingly online real-time predicts the 8+1 odor pollutant concentration control index values in the future 1h.

[0054] The gas sensor array I-1 and its annular working cavity I-2 are located in a constant temperature room at 55±0.1℃. Within the default value T=300s of the gas sampling period, the gas sensor array I-1 and its annular working cavity I-2 undergo the following five stages in turn according to the different gas species and flow rates: ① Precise calibration: under the purging action of 1,000ml / min clean air, the gas sensor array I-1 gradually and precisely recovers to the standard state, lasting for 60s. ② Equilibrium: there is no gas flow in the annular working cavity I-2 of the gas sensor array, lasting for 10s. ③ Headspace sampling: under the suction action of the micro vacuum pump II-1, the gas sensor array I-1 produces a sensitive response to 1,000ml / min of the measured malodorous gas, lasting for 30s. ④ Pressure multiplication: under the thrust action of 0.2Mpa clean air, 78ml of malodorous gas in the pressure cylinder III is compressed back to the annular working cavity I-2 of the gas sensor array, and the pressure in the cavity increases by about 1 time in a short time, so the sensitive response of the gas sensor array I-1 is further increased, lasting for 30s. ⑤ Waste gas discharge: under the suction action of the micro vacuum pump II-1, the waste gas in the annular working cavity I-2 of the gas sensor array, the pressure cylinder III and the related gas path pipeline is discharged to the atmosphere, lasting for 170s. As the malodorous waste gas is discharged, the gas sensor array I-1 gradually and approximately recovers to the initial state. If the gas sampling period T>300s, the waste gas discharge time is correspondingly extended, and the flow rates and time lengths of the other four stages remain unchanged; for example, if T=480s, the waste gas discharge time length is 170+(480-300)=350s; if T<300s, the waste gas discharge time length is 170+(T-300)<170s.

[0055] According to the ideal gas state equation p1V1=p2V2 under constant temperature conditions and the classical Langmuir isothermal low-pressure adsorption equation q=γpq m / (1+γp), where p1, p2 and p are pressure, V1 and V2 are volume, q is adsorption capacity, and q mFor saturated adsorption amount, γ is equilibrium constant. The present application realizes the increase of micro trace component adsorption amount with the increase of pressure by means of the pressurized cylinder III. The specific measures are as follows: designing a small volume and small section gas sensitive sensor array ring working cavity I-2 with a middle diameter of φ140mm and two kinds of section sizes of 14x10mm and 16x20mm and a net volume of 78ml; designing a one-way introduction ultra-thin piston type pressurized cylinder III, the piston III-1 diameter, i.e. cylinder diameter φ50mm, the piston rod III-2 diameter φ18mm, the theoretical maximum stroke 40mm, and the theoretical net volume of the cylinder 78ml. The working mode of the pressurized cylinder III is that the piston III-1 and the piston rod III-2 are retracted under the driving of 0.2Mpa clean air and are reset by means of the compression spring III-4 on the piston rod III-2. In the ⑤ waste gas discharge stage of the gas sampling period kT and the ① accurate calibration, ② equilibrium and ③ headspace sampling stages of the next period (k+1)T, the piston rod III-2 is extended under the tension of the compression spring III-4, the piston III-1 returns to the initial position, the maximum net volume of the cylinder reaches 78ml, and the 78ml of the measured malodorous gas is stored. In the ④ pressure doubling stage, the piston III-1 and the piston rod III-2 move the maximum stroke of 40mm under the driving of 0.2Mpa clean air, the compression spring III-4 is compressed, the static volume of the cylinder gradually changes from 78ml to 0, and the 78ml of the measured malodorous gas stored in the cylinder body III-3 is compressed back to the gas sensitive sensor array ring working cavity I-2. The electronic nose instrument realizes the doubling of the sensitive film adsorption amount of the micro trace component, the equivalent increase of the concentration of the micro trace component of the malodorous gas and the doubling of the sensitive sensing sensitivity by means of the pressurized cylinder III to make the pressure in the gas sensitive sensor array ring working cavity I-2 short-term doubled.

[0056] The pressurized cylinder III is placed at the rear end of the gas outlet of the gas sensitive sensor array ring working cavity I-2, all the gas flowing through the gas sensitive sensor array ring working cavity I-2 flows through the pressurized cylinder III, and finally is discharged to the atmosphere as waste gas. The pressurized cylinder III is provided with two gas inlets and one gas outlet: the gas inlet 1 communicates with the gas outlet of the gas sensitive sensor array ring working cavity I-2, the gas inlet 2 communicates with the gas outlet of the clean air bottle VI by means of the two-position two-way electromagnetic valve II-8, and the gas outlet of the pressurized cylinder III communicates with the flow control and waste gas discharge gas path by means of the two-position two-way electromagnetic valve II-5. The micro vacuum pump II-1 sucks the malodorous gas molecules remaining in the inner wall of the gas sensitive sensor array ring working cavity I-2, the pressurized cylinder III and the related gas path pipeline for a long time at a flow rate of 6,000ml / min, and the gas sensitive sensor array I-1 is thus preliminarily restored to the reference state. Under the continuous blowing of 1,000ml / min clean air for 60s, the gas sensitive sensor array I-1 is accurately restored to the reference state.

[0057] The electronic nose instrument is flexibly used in (i) field without alternating current power supply and (ii) field with alternating current power supply, and flexibly performs (A) short-term limited-time real-time detection and analysis task and (B) long-term continuous online detection and analysis task for several hours, 1 day, several days or even several months. Working scenario (i) + task (A): the instrument uses a backup power supply - 25.2V lithium battery pack V with a nominal capacity of 10Ah, and the battery pack weighs 2.5kg and can work continuously for more than 3h and can be charged; at the same time, the instrument also uses a 2-6L aluminum alloy seamless clean air small bottle VI-2 with a maximum working pressure of 15MPa, a size of φ111*359mm-φ140*629mm, and a self-weight of 2.8-7.2kg (valve 0.8Kg not included), and the maximum continuous working time is more than 2.5h calculated by the clean air accurate calibration time of 60s. Working scenario (ii) + task (A): the instrument uses alternating current power supply and only uses 2-6L aluminum alloy seamless clean air small bottle VI-2 to carry out short-term detection and analysis of malodor pollution for about 2h. Working scenario (ii) + task (B): the instrument uses alternating current power supply, and at the same time uses a 40L clean air large bottle VI-1 with a maximum working pressure of 15MPa, a size of φ219*1450mm, and a self-weight of 45kg, and still calculates the maximum continuous working time by the clean air accurate calibration time of 60s, and 1 large bottle of clean air can work continuously for about 20 days, and timely replacement of the used clean air large bottle VI-1 can continuously online detect and analyze for a longer time.

[0058] Based on the existing manufacturing process level and existing products of gas sensitive elements, the measures to significantly improve the gas sensitive sensing sensitivity and selectivity from the aspects of instrument structure design and industrial application include: (i), 14*10mm, 16*20mm two kinds of small cross-section size and medium diameter φ140mm and 78ml small volume gas sensitive sensor array ring-shaped working cavity I-2 structure design, 1,000ml / min headspace sampling large flow and 78ml net volume of booster cylinder III make the pressure in the gas sensitive sensor array ring-shaped working cavity I-2 multiply in a short time, and the optimization combination of the three makes the gas sensitive sensitivity multiply.(ii), select the three sensing components of steady-state peak value, peak time and curve area from the single gas sensitive sensor voltage response curve to significantly improve the gas sensitive selectivity.

[0059] From the beginning of the 6th second of the equilibrium stage ② of the kth gas sampling period, including the whole headspace sampling and pressure multiplication stage ③, to the end of the 5th second of the waste gas discharge stage ⑤, the computer control and analysis module IV records and stores the sensing information of the gas sensitive sensor array I-1 for 70s. From the beginning of the 6th second to the end of the 35th second of the waste gas discharge stage ⑤ of this period, the electronic nose instrument enters the information processing and analysis stage for 30s, and the computer control and analysis module IV selects the steady-state peak value v i,0(k), peak time t i,0 (k), area A under curve i,0 (k) These 3 initial perception information components, 3*16=48 perception information components selected from 16 gas sensitive sensor voltage response curves, plus on-site environment temperature and humidity T em,0 (k) and H um,0 (k) components, constitute initial sample x0(k)∈R 50 The modular machine learning cascade model determines the malodor pollution type, quantitatively estimates and predicts 8+1 concentration control index values accordingly.

[0060] In the ③ headspace sampling stage, the first two-position two-way electromagnetic valve II-4 and the second two-position two-way electromagnetic valve II-5 are turned on, the third two-position two-way electromagnetic valve II-6, the fourth two-position two-way electromagnetic valve II-7 and the fifth two-position two-way electromagnetic valve II-8 are turned off, the piston rod III-2 of the booster cylinder III extends under the elastic force of the compression spring III-4, and the cylinder body III-3 thus generates a storage space of 78 ml. Under the suction action of the micro vacuum pump II-1, the malodor gas flows through the first two-position two-way electromagnetic valve II-4, the gas sensitive sensor array annular working cavity I-2 and the gas sensitive sensor array I-1, the cylinder body III-3, the second two-position two-way electromagnetic valve II-5, the throttle valve II-3 and the flow meter II-2 in turn at a flow rate of 1,000 ml / min, and is finally discharged into the atmosphere, lasting for 30 s. In the process of malodor gas flow, the gas sensitive sensor array I-1 generates a sensitive response.

[0061] In the ④ pressure multiplication stage, the fifth two-position two-way electromagnetic valve II-8 is turned on, and the first two-position two-way electromagnetic valve II-4, the second two-position two-way electromagnetic valve II-5, the third two-position two-way electromagnetic valve II-6 and the fourth two-position two-way electromagnetic valve II-7 are turned off. Under the thrust of 0.2 MPa clean air, the piston rod III-2 of the booster cylinder III retracts, and the 78 ml of malodor gas stored in the cylinder body III-3 is compressed back into the gas sensitive sensor array annular working cavity I-2, lasting for 30 s; according to the ideal gas state equation p1V1=p2V2 under constant temperature conditions, the pressure of the gas sensitive sensor array annular working cavity I-2 gradually increases by 1 times, and according to the Langmuir isothermal low-pressure adsorption equation, the response of the gas sensitive sensor array I-1 increases by about 1 times on the basis of the ③ headspace sampling stage.

[0062] In the ⑤ exhaust gas discharge stage, the second two-position two-way electromagnetic valve II-5 and the fourth two-position two-way electromagnetic valve II-7 are turned on, the first two-position two-way electromagnetic valve II-4, the third two-position two-way electromagnetic valve II-6 and the fifth two-position two-way electromagnetic valve II-8 are turned off, the piston rod III-2 is extended again due to the elastic force of the compression spring III-4, and the cylinder body III-3 generates 78ml space again. Under the suction action of the micro vacuum pump II-1, the residual exhaust gas in the gas sensor array annular working cavity I-2 and the cylinder body III-3 is extracted at a theoretical flow rate of 6,000ml / min, and the gas sensor array I-1 gradually recovers to the initial reference state; then, in the next cycle of ① accurate calibration stage, the second two-position two-way electromagnetic valve II-5 and the third two-position two-way electromagnetic valve II-6 are turned on, the first two-position two-way electromagnetic valve II-4, the fourth two-position two-way electromagnetic valve II-7 and the fifth two-position two-way electromagnetic valve II-8 are turned off, the gas sensor array I-1 is accurately restored to the reference state under the action of 1,000ml / min clean air blowing for 60s, and the residual malodorous molecules in the cylinder body III-3 and the inner wall of the related pipeline are completely removed.

[0063] The present application takes 8 kinds of malodorous pollutants specified in GB14554, hydrogen sulfide H2S, ammonia NH3, methyl mercaptan CH4S, dimethyl disulfide C2H6S2, dimethyl sulfide C2H6S, carbon disulfide CS2, trimethylamine C3H9N, styrene C8H8, 5 kinds of standard malodorous liquid specified in HJ1262-2022 for olfactory qualification examination, methyl cyclopentanone C6H 10 O, β-phenylethanol C8H 10 O, γ-undecanolactone C 11 H 20 O2, β-methyl indole C9H9N, isovaleric acid C5H 10 O2 and n-butanol C4H 10 O14 kinds of typical malodorous compounds are analyzed, 0.1-100ppm single or mixed samples are prepared in turn, the volatile gas of the sample is detected by an electronic nose instrument, and the relationship between the multi-dimensional perception information of the gas sensor array I-1 and the molecular weight, structure, functional group micro characteristics and odor concentration of the 14 kinds of malodorous compounds is established.

[0064] The malodor big data set X includes: (i), the off-line detection data of the electronic nose instrument on the above-mentioned 14 typical malodor compound samples; (ii), a large amount of on-line real-time detection data of the electronic nose instrument at a plurality of malodor pollution sites; (iii), the odor concentration data determined by a group of olfactory discriminators in the laboratory on a large number of malodor samples collected from the site; (iv), the laboratory component analysis data of a large number of malodor samples collected from the site by color, mass spectrometry and other conventional instruments; (v), the atmospheric temperature / humidity data at the malodor pollution detection site; (vi), the adjacent resident complaint data; (vii), the correspondence data of the multi-dimensional sensing sample of the gas sensitive sensor array I-1 with the malodor type and the 8+1 concentration control indicators.

[0065] In the kth gas sampling period, the computer control and analysis module IV simultaneously selects the steady-state peak value v i,0 (k), the peak time t i,0 (k), the total area A i,0 (k) under the curve from the voltage response curve of the gas sensitive sensor i.

[0066] (a), the steady-state peak value v i (k):

[0067] In formula (1), v max = 10.0 V is the working loop voltage value of the gas sensitive sensor.

[0068] (b), the peak time t i (k):

[0069] In formula (2), t max = 30+30=60s is the sum of the time lengths of the ③ headspace sampling and the ④ pressure multiplication two stages after the computer control and analysis module IV issues the “headspace sampling” instruction.

[0070] (c), the total area A i (k) under the curve:

[0071] In formula (3), it is assumed that the maximum area is t max * v max = 10*60=600.

[0072] In addition, it is assumed that the highest ambient temperature is 45℃, the normalized ambient temperature is T em (k)=T em,0 (k) / 45; the normalized ambient humidity is directly the relative humidity value, that is, H um (k)=H um,0 (k) / 100.

[0073] The normalized sample is x(k) = (v1(k),..., v 16 (k), t1(k),..., t 16 (k), A1(k),..., A 16 (k), T em (k), H um (k)) T = (x1(k),..., x 16 (k), x 17 (k),..., x 32 (k), x 33 (k),..., x 48 (k), x 49 (k), x 50 (k)) T ∈ R 50 , and each input component after normalization is dimensionless.

[0074] The present application flexibly uses the "divide and conquer" strategy to transform the pollution type recognition and 8+1 concentration control index estimation and prediction problems implied in the large odor data set X into multiple single-target recognition and multiple single-concentration index estimation and prediction problems, that is, the n odor pollution type recognition problems are decomposed into C n 2 = n(n-1) / 2 binary classification sub-problems, and the n type r = 8+1 odor pollution concentration index estimation and prediction problems are decomposed into n x r single-index one-to-one estimation and prediction problems. According to the Occam "simple and effective" law, n(n-1) / 2+nr "three-person" teams are formed by single neurons with priority 1, single-output single-hidden-layer neural networks with priority 2, and single-output deep neural networks with priority 3, corresponding to n(n-1) / 2 binary classification sub-set numbers + nr single-concentration index sub-set numbers; the team members learn the n(n-1) / 2+nr subsets formed by the "divide and conquer" decomposition of the large odor data set X in order of priority.

[0075] The basic parameters of the neural network learning process are set as follows: based on the conventional Sigmoid, ReLU, and Softmax activation functions, a new type of Sigmoid activation function f(σ) = 3(1+exp(-σ / 3)) -1 is discovered, σ is the total input of a certain hidden node or output node; a dynamic learning mechanism and algorithm that allows the most necessary samples to participate is established; the input components are proportionally enlarged to [0, 6], the target output components for classification are enlarged from {0, 1} to {0, 3}, the target output components for concentration estimation and prediction are enlarged from (0, 1) to (0, 3), the learning step factor η = 0.05 / N, where N is the number of training subset samples, the learning termination condition is root mean square error δ ≤ 0.075, and the maximum number of learning steps is ζmax = 10,000. Single-output single-hidden-layer neural network hidden nodes s = 8, deep neural network uses a peer-based single-hidden-layer network layer-by-layer learning algorithm.

[0076] According to the Occam's Razor law, in any "three-person" group, once the priority is 1, the single neuron learning accuracy - root mean square error δ meets the requirements, it automatically becomes the winner, and the remaining 2 members do not need to learn, and so on; thus, each group selects the winner with simple structure and effective problem solving to participate in decision-making; n(n-1) / 2 classification winners form n voting decision groups, each consisting of n-1 winners, and each classification winner participates and must participate in 2 voting decision groups; nr concentration estimation and prediction winners form n quantitative prediction groups, each consisting of r winners.

[0077] The active two-way electronic nose instrument detects real-time or continuous online detection of malodor pollution sites, and the computer control and analysis module IV performs the online real-time analysis task of the concentration index value of 8+1 malodor pollutants specified in GB14554-1993, which is divided into the following 3 cases according to the size of the continuous detection period k: (a) k = 1: corresponding to sample x(1), the first level of the machine learning cascade model - the winning group Ω j of the n voting decision groups j Corresponding to the quantitative prediction group Λ j 8+1 winners of the 8+1 concentration control index values.

[0078] (b) 1 < k ≤ 72: corresponding to samples x(2) ~ x(72), the first level of the cascade model no longer votes, and the malodor pollution type defaults to Ω j , the second level - the 8+1 winners of the quantitative prediction group Λ j continue to estimate the 8+1 concentration control index values one by one.

[0079] (c) k > 72: this is equivalent to the continuous detection and sample acquisition time length kT > 6h when T = 300s, the first level of the cascade model still does not vote, and the malodor pollution type defaults to Ω j ; follow the "before and after" rule, use the long time series response of the gas sensitive sensor array I-1 in the last 72 periods (about 6h) to predict the 8+1 concentration index values of the malodor pollutants in the next 1h, focusing on predicting the 8+1 concentration control index values of the 2nd(k = 2, 10min), 6th(k = 6, 30min) and

[0080] 12th(k = 12, 60min) period.

[0081] When the number of consecutive sampling cycles (k>72) of a gas sample from a malodorous pollution point using a dual-use electronic nose instrument is constant or continuous, the gas sensor array generates a long-term normalized data matrix {X(k-71),…,X(k)} with a response length of Δk=72 in the most recent 72 cycles. Each component of this matrix is ​​automatically generated from a portion of its own time series, forming three target outputs with a duration of Δk=72. The short-time series training subarray. The i-th long-time series component {x} with Δk = 72. i (k-71),…,x i (k-τ),…,x i (k)}∈R 72 For example, as shown in Tables 1a-c, X is generated. i (k+2), X i (k+6), X i (k+12) these 3 The short time series convolution training subsets, all with a dimension of 12 and sample numbers of 59, 55 and 49 respectively, are used for online learning by three single-output shallow neural networks with a 12-5-1 structure, so as to predict the value of the i-th perceptual component at future time (k+2)T, (k+6)T and (k+12)T starting from kT.

[0082] To accelerate the learning speed, we assume and normalize the peak time t i The time series of the gas sensor sensing component corresponding to (k) remains unchanged at times (k+2)T, (k+6)T, and (k+12)T, and no time series convolution training subset is generated; therefore, there are 2*16*3+2*3=102 short time series convolution training subset matrices, which are learned online by 102 single-output shallow neural networks with a 12-5-1 structure within 30s, and based on this, the normalized time series response {x(k+2),x(k+6),x(k+12)} of the gas sensor array that will occur in the next k+2 period (10min), k+6 period (30min), and k+12 period (60min) are predicted online; and Ω j The corresponding quantitative prediction group Λ j Based on this, the values ​​of 8+1 concentration control indicators will be predicted sequentially at the future times (k+2)T, (k+6)T, and (k+12)T.

[0083] The electronic nose instrument for both fixed and live odor detection and analysis performs online real-time detection and analysis of multiple components of malodorous substances, including the following steps:

[0084] (1) Preparation: (i), on-site real-time working scene with AC power supply, carrying the main box and 2-6L clean air small bottle VI-2 these two components; (ii), AC power supply and continuous online working scene, fixed installation main box and 40L clean air large bottle VI-1 these two components; (iii), field working scene without AC power supply, carrying the main box, backup power supply-25.2V lithium battery V and 2-6L clean air bottle VI-2 these three components.

[0085] (2) Start: the instrument is preheated for 30 minutes; set "gas sampling period T", the default value T=5min.

[0086] The second two-position two-way solenoid valve II-5 and the fourth two-position two-way solenoid valve II-7 are turned on, the first two-position two-way solenoid valve II-4, the third two-position two-way solenoid valve II-6 and the fifth two-position two-way solenoid valve II-8 are turned off, the piston rod III-2 is extended due to the elastic force of the compression spring III-4, and the cylinder body III-3 generates a space of 78ml. Under the suction of the micro vacuum pump II-1, the residual waste gas in the gas sensor array annular working cavity I-2 and the cylinder body III-3 is extracted at a flow rate of 6,000ml / min, and the gas sensor array I-1 gradually recovers to the reference state.

[0087] (3) Start of the online real-time detection and analysis period of malodorous gas:

[0088] (3.1) Malodorous gas detection:

[0089] In a gas sampling period T, depending on the different types and flow rates of gases, the gas sensor array module I goes through the following five stages in turn: ① accurate calibration, ② balance, ③ headspace sampling, ④ pressure multiplication, ⑤ waste gas discharge; Table 2 shows the gas type, sampling time length, starting time, flow rate, and on / off status of the five two-position two-way solenoid valves at T=300s.

[0090] From the beginning of the 6th second to the end of the 35th second in the T-⑤ waste gas discharge stage, the electronic nose instrument enters the information processing and analysis stage for 30s; from the beginning of the 6th second in the ② balance stage, including the entire ③ headspace sampling and ④ pressure multiplication stages, to the end of the 5th second in the ⑤ waste gas discharge stage, the computer control and analysis module IV records and stores the sensing information of the gas sensor array I-1, lasting for 70s,

[0091] Table 2, working parameters of the gas sensor array module I and related on / off status of the electromagnetic valves in the gas sampling period T=300s (default)

[0092] Then select the steady-state peak value vi,0 (k), the peak time t i,0 (k), the area under the curve A i,0 (k) These 3 perception information components, 16*3=48 information components of the gas sensor array I-1 plus the ambient temperature T at that time and place em,0 (k) and the ambient humidity H um,0 (k) form the initial sample x0(k)∈R 50 .

[0093] (3.2) Malodorous gas analysis:

[0094] (a) The k=1 gas sampling period: the electronic nose instrument obtains the initial sample x0(1) by detecting a certain malodorous pollution point for the first time, normalizes it to x(1), and the machine learning cascade model of the computer control and analysis module IV simultaneously performs malodorous pollution type identification and 8+1 typical malodorous compound concentration index quantification estimation and prediction.

[0095] (b) The k=2-71 gas continuous sampling period: the electronic nose instrument performs k<72 period continuous online real-time detection on a certain malodorous pollution point, obtains the normalized finite time sequence sample set {x(1), x(2), …, x(k<72)} including x(1), and the malodorous pollution type is still k=1. The first level of the machine learning cascade model—the winning group represented by the voting identification, and the second level of the cascade model—only needs to perform 8+1 typical compound concentration index quantification estimation on the quantification prediction group corresponding to the winning group.

[0096] (c) k≥72 gas continuous sampling period: the electronic nose instrument performs k≥72 period continuous online real-time detection on a certain malodorous pollution point, obtains the normalized time sequence response sample set {x(1), …, x(72), x(73), …, x(k)}; the latest 72 continuous samples are intercepted to form a moving long time sequence sample set X(k-τ)={x(k-τ)}∈R 72×50 , τ=0,1,…,71. For example, a recent Δk=72 moving long time sequence is {x(5), …, x(72), …, x(k=76)}; next, for predicting the response values of the future k+2, k+6, k+12 periods, 102 supervised short time sequence convolution training subsets {X i (k+2)∈R 59×12 ,X i (k+6)∈R 55×12 ,X i (k+12)∈R 49×12}, i = 1,..., 16, 33,..., 50; third step, 102 12-5-1 single-output neural networks learn online in real time from the 102 short time series convolution training subset. As soon as the learning is completed, the 102 12-5-1 single-output neural networks immediately predict the future k+2, k+6, k+12 period response component values of the gas sensor array and the temperature and humidity values according to the following formula short time series convolution training subset. As soon as the learning is completed, the 102 12-5-1 single-output neural networks immediately predict the future k+2, k+6, k+12 period response component values of the gas sensor array and the temperature and humidity values according to the following formula i (k-11), x i (k-10),..., x i (k-1), x i (k)}-predict the future k+2, k+6, k+12 period response component values of the gas sensor array and the temperature and humidity values, and combine them with the peak time component that did not participate in the prediction, to generate sample prediction values for the future k+2, k+6, k+12 period Finally, the second level of the machine learning cascade model, the quantitative prediction group corresponding to the winning group, predicts the 8+1 concentration control index values at the corresponding time according to the three sample prediction values.

[0097] (4) Once the gas sampling period kT at the same malodor pollution point ends, or enters the next period (k+1)T, repeat process (3), or move to the next malodor pollution point to start new detection and analysis.

[0098] It is particularly pointed out here that an important capability of the electronic nose instrument used in the process of online real-time detection and analysis of malodor pollution is that once a perception vector x(k) is obtained through real-time detection on site, the instrument can estimate the 8+1 concentration control index values of the malodor pollutants on site in real time within 1 min, and it is relatively secondary to know what type of malodor pollution it is. Here, this can be achieved by the second level of the machine learning cascade model, the quantitative prediction group. BRIEF DESCRIPTION OF DRAWINGS

[0099] Figure 1 is a schematic diagram of the technical route of the development of a fixed and mobile two-purpose electronic nose instrument and the method for online real-time detection and analysis of malodor multi-components according to the present application.

[0100] Figure 2 is a schematic diagram of the working principle of a multifunctional malodor electronic nose instrument according to the present application: the position of malodor gas headspace sampling, continuous online working state with alternating current power supply.

[0101] Figure 3 is a schematic diagram of the working principle of a multifunctional malodor electronic nose instrument according to the present application: the position of pressure multiplication, field real-time working state without alternating current power supply.

[0102] Figure 4 is a schematic diagram of the gas sensor array arrangement and the gas sensor array annular working chamber structure of the present application, a fixed-living two-convenient electronic nose instrument and malodor multi-component online real-time detection and analysis method.

[0103] Figure 5 is a schematic diagram of the pressure multiplication device, i.e. the pressure boosting cylinder structure of the present application, a fixed-living two-convenient electronic nose instrument and malodor multi-component online real-time detection and analysis method; (a), the pressure boosting cylinder piston rod retracted position, the gas sensor annular working chamber is in the pressure boosting state; (b), the pressure boosting cylinder piston rod extended position, the gas sensor annular working chamber is in the preliminary recovery, accurate calibration and headspace sampling state.

[0104] Figure 6 is a schematic diagram of the fixed-living two-convenient malodor electronic nose instrument of the present application, a fixed-living two-convenient electronic nose instrument and malodor multi-component online real-time detection and analysis method; (a), the front view of the fixed-living two-convenient malodor electronic nose instrument; (b), the back view of the fixed-living two-convenient malodor electronic nose instrument.

[0105] Figure 7 is a schematic diagram of the working scene of the fixed-living two-convenient malodor electronic nose instrument of the present application, a fixed-living two-convenient electronic nose instrument and malodor multi-component online real-time detection and analysis method; (a), the long-term online detection scene with alternating current; (b), the short-term real-time on-site detection scene in the field without alternating current.

[0106] Figure 8 is a schematic diagram of the gas sensor array working chamber I-2 gas sampling parameters and gas sensing response changes within a period of T = 360 s of the present application, a fixed-living two-convenient electronic nose instrument and malodor multi-component online real-time detection and analysis method.

[0107] Figure 9 is a schematic diagram of the basic component units of the machine learning cascade model of the present application, a fixed-living two-convenient electronic nose instrument and malodor multi-component online real-time detection and analysis method; they are: single neuron, single output single hidden layer neural network and single output deep neural network.

[0108] Figure 10 is a schematic diagram of the new Sigmoid activation function combined with moderate amplification of input and output components to significantly speed up the learning speed of the neural network of the present application, a fixed-living two-convenient electronic nose instrument and malodor multi-component online real-time detection and analysis method.

[0109] Figure 11 is a schematic diagram of the first level of the machine learning cascade model for malodor type identification of the present application, a fixed-living two-convenient electronic nose instrument and malodor multi-component online real-time detection and analysis method; the voting identification group component unit schematic diagram.

[0110] Figure 12 is a schematic diagram of the machine learning cascade model for both malodor type identification and 8+1 malodor pollutant concentration control index online real-time quantitative estimation of the present application, a fixed-living two-convenient electronic nose instrument and malodor multi-component online real-time detection and analysis method.

[0111] Figure 13 is a schematic diagram of the machine learning cascade model of the application, an active and portable electronic nose instrument and a method for online real-time detection and analysis of malodor multi-component, for online real-time quantification and prediction of 8+1 malodor pollutant concentration control indicators. DETAILED DESCRIPTION

[0112] The application will be further described in detail below with reference to the accompanying drawings.

[0113] Figure 1 is a schematic diagram of the technical route for developing the active and portable electronic nose instrument of the application and the technical route for online real-time detection and analysis of malodor multi-component.

[0114] The specific invention includes: (1) a new method of revealing gas sensing mechanism and multi-information selection fusion. The law between gas sensing and odor micro-macro characteristics is revealed, a new method of micro-trace component concentration enhancement and gas sensing multi-information selection fusion is proposed, and high selectivity and high sensitivity sensing of complex odor is realized from the perspectives of instrument design and application objects. (2) A new model and algorithm of flexible machine learning. A large malodor data set is established, the "divide and conquer" strategy, Occam's simple and effective law, and the law of carrying forward are comprehensively used, offline and online learning are combined, and the near future and the current perceived information are combined. A flexible machine learning cascade model and algorithm of single neuron, shallow neural network, and deep neural network fusion is proposed, and malodor multi-type recognition and malodor pollutant multi-concentration indicator estimation and prediction are flexibly realized. (3) Multi-parameter optimization combination and industrial application of active and portable electronic nose instrument. The application is characterized by large sample flow, small working cavity volume, and fast adsorption and desorption. The multi-working parameter optimization combination method of the instrument can significantly improve the sensitivity, selectivity, and stability of the electronic nose instrument, and realize online real-time detection, recognition, and multi-concentration indicator estimation and prediction of malodor.

[0115] Figure 2 is a schematic diagram of the working principle of the multifunctional active and portable malodor electronic nose instrument of the application, which is in the position of headspace sampling of malodor gas and has a continuous online working state with alternating current. The main hardware components of the electronic nose instrument include: gas sensitive sensor array module I, gas automatic headspace sampling module II, booster cylinder III, computer control and analysis module IV, backup power supply - 25.2V lithium battery group V, 40L clean air large bottle VI-1, 2-6L clean air small bottle VI-2. Among them, the gas sensitive sensor array module I, the gas automatic headspace sampling module II, the booster cylinder III, and the computer control and analysis module IV form the main machine box.

[0116] Figure 3 is a schematic diagram of the working principle of the multifunctional static and dynamic foul odor electronic nose instrument of the present application, which is in the foul odor gas pressure multiplication position and in the field real-time working state without alternating current power supply. The main components of the gas sensor array module I include: gas sensor array I-1, gas sensor array annular working cavity I-2, heat insulation layer I-3, partition I-4, fan I-5, resistance heating wire I-6, and temperature sensor I-7, which are located in the upper right part of the main machine box.

[0117] The automatic headspace sampling gas path module II includes: micro vacuum pump II-1, flow meter II-2, throttle valve II-3, first two-position two-way electromagnetic valve II-4, second two-position two-way electromagnetic valve II-5, third two-position two-way electromagnetic valve II-6, fourth two-position two-way electromagnetic valve II-7, fifth two-position two-way electromagnetic valve II-8, and purifier II-9. The booster cylinder III includes: piston III-1, piston rod III-2, cylinder body III-3, compression spring III-4, O-shaped sealing ring III-5, inner snap spring III-6, front cover III-7, and anti-collision pad III-8. The automatic headspace sampling gas path module II and the booster cylinder III are located in the lower right part of the main machine box.

[0118] The main components of the computer control and analysis module IV include: A / D data acquisition card IV-1, drive and control circuit board IV-2, computer mainboard IV-3, 4-way precision DC voltage regulator IV-4, WIFI board card IV-5, and touch display screen IV-6, which are located on the left side of the main machine box.

[0119] The clean air bottle VI includes: 40L clean air large bottle VI-1 and 2-6L clean air small bottle VI-2. The 40L clean air large bottle VI-1 is mainly used for foul odor long-term continuous online detection for more than 2h, and the 2-6L clean air small bottle VI-2 is used for foul odor short-term limited-time online real-time detection for less than 2h.

[0120] Figure 4(a) shows that the 16 gas sensors of the gas sensor array I-1 are arranged in a ring shape. Figures 4(b)-(e) show that the diameter of the gas sensor array annular working cavity I-2 is φ140mm, the two kinds of ring sections are 16*20mm and 14*10mm, and the net volume is about V0=78ml. The main components include working cavity bottom plate I-2-1, working cavity cover I-2-2, sensor socket I-2-3, internal hexagonal screw I-2-4, O-shaped sealing ring I-2-5, and food-grade odorless sealing glue I-2-6.

[0121] Figure 5 is a structure diagram of the gas pressure multiplication device - the pressurized cylinder III of the gas sensor array ring-shaped working cavity I-2. The main components of the pressurized cylinder III are: the piston rod III-1, the piston III-2, the cylinder body III-3, the compression spring III-4, the O-ring III-5, the inner snap spring III-6, the front cover III-7, and the anti-collision pad III-8. The pressurized cylinder III is a one-way introduction piston type ultra-thin cylinder. The diameter of the piston III-1 is φ50mm, the diameter of the piston rod III-2 is φ18mm, the maximum stroke is 40mm, and the theoretical net volume of the cylinder is 78ml. The piston III-1 and the piston rod III-2 are extended under the pushing action of 0.2Mpa clean air, and are reset with the help of the compression spring III-4 on the piston rod III-2. During the pressure multiplication stage, the piston III-1 and the piston rod III-2 move the maximum stroke of 40mm under the pushing action of 0.2Mpa clean air, and the compression spring III-4 is compressed. The static volume of the cylinder gradually changes from 78ml to 0, that is, the 78ml of the measured malodorous gas is temporarily compressed back into the gas sensor array ring-shaped working cavity I-2, so that the pressure in the ring-shaped working cavity I-2 is temporarily multiplied, so as to increase the adsorption amount of the gas sensitive membrane to the trace amount of malodorous components, which is equivalent to increasing the concentration of the trace amount of components, and realizing the multiplication of the gas sensitive sensing sensitivity. In the remaining four stages of a gas sampling period T, the piston III-1 returns to the initial position with the help of the compression spring III-4, and the maximum net volume of the cylinder returns to 78ml. In particular, the pressurized cylinder III is provided with an air inlet 1 to allow the measured malodorous gas to enter the cylinder body III-3, so as to increase the pressure inside the gas sensor array ring-shaped working cavity I-2; and an air inlet 2 is provided to push the piston III-1 to move.

[0122] Figure 6(a) is a three-dimensional appearance of the main box of the malodorous electronic nose instrument in the data analysis state. The main box has an outer shape size of 420 (width) x 380 (height) x 230mm (depth). The gas sensor array module I is located at the upper right part, the gas automatic headspace sampling module II and the pressurized cylinder III are located at the lower right part, and the computer control and analysis module IV is located at the left side. A 14" wide screen touch display screen (size: width w x height h x thickness t = 312 x 205 x 5mm) is arranged on the front side, which is used for online real-time display of the sensing curve of the gas sensor array I-1 to the malodorous pollutants and the analysis results of the machine learning method. A flow meter is arranged to display the flow change online in real time. Here, the local environmental temperature and humidity at the time are also measured and recorded. Figure 6(b) is a back view of the instrument. A mouse, a keyboard, a USB, a display, an Internet interface socket, two air inlets for clean air and malodorous gas, a waste gas outlet, an alternating current power supply socket and a direct current power supply socket are arranged on the back of the instrument.

[0123] Fig. 7(a) shows the long-term on-line detection scenario of the fixed-active two-convenient malodor electronic nose instrument with alternating current. The instrument main box plus a 40L clean air bottle VI-1 can continuously detect and analyze malodor pollutants for several days or even tens of days. The role of the operator is to replace the almost empty clean air bottle VI-1 with a new one in time. Fig. 7(b) shows the short-term on-site real-time detection scenario of the fixed-active two-convenient malodor electronic nose instrument in the field without alternating current. The instrument main box is equipped with 2-6L clean air bottles VI-2 and a standby DC power supply—25.2V lithium battery pack V, which can detect malodor pollutants in the field for a limited number of times, with a continuous working time of up to 2h. If necessary, a mobile AC power supply can be used instead of the standby DC power supply, and 3-6L clean air bottles can be carried. It should be pointed out that even in the scenario with alternating current, if the detection time is short, the fixed-active two-convenient malodor electronic nose instrument of the present application can use the instrument main box plus small 2-6L clean air bottles VI-2 to carry out short-term malodor detection and analysis.

[0124] Fig. 8 is a schematic diagram of the gas sampling type parameters of the working cavity I-2 of the fixed-active two-convenient electronic nose instrument of the present application and the sensing response change of one of the gas sensitive sensors in one gas sampling period T=360s. According to Fig. 8(a), one gas sampling period T=300s includes five stages in turn: ① precise calibration: under the action of 1,000ml / min clean air blowing for 60s, the gas sensitive sensor array I-1 is precisely restored to the standard state; ② equilibrium: the gas sensitive sensor array ring-shaped working cavity I-2 has no gas flow for 10s; ③ headspace sampling: under the action of the micro vacuum pump II-1 for 30s, the gas sensitive sensor array I-1 generates a sensitive response to 1,000ml / min of the measured malodor gas; ④ pressure multiplication: under the action of 0.2Mpa clean air thrust for 30s, the pressure cylinder III increases the pressure of the 78ml malodor gas accumulated in itself by 1 times, so that the response of the gas sensitive sensor array I-1 is further increased; ⑤ waste gas discharge: under the action of the micro vacuum pump II-1 at a flow rate of 6,000ml / min for 170s, the waste gas of the gas sensitive sensor array ring-shaped working cavity I-2, the pressure cylinder III and the related pipeline is pumped out, and the gas sensitive sensor array I-1 gradually returns to the initial state.

[0125] According to Fig. 8(b), within a gas sampling cycle T, only 70s of the gas sensor response information is recorded by the computer control and analysis module IV, including the last 5s of the equilibrium stage, the sum of 60s of the complete headspace sampling stage and the pressure multiplication stage, and the first 5s of the waste gas discharge stage. In the 30s information selection and analysis interval, the computer control and analysis module IV selects the steady-state voltage peak value v i,0 (k), the peak time t i,0 (k), the curve area A i,0 (k) These three original perception components, i = 1, 2, …, 16. Thus, the computer control and analysis module IV extracts 3*16 = 48 response components from the 16 response curves of the gas sensor array I-1 in a cycle T, obtaining a 48-dimensional initial perception vector or sample x0(k) = (…, v i,0 (k), …, t i,0 (k), …, A i,0 (k), …) T ∈R 48 , which is stored in the data file of the computer mainboard IV-3 hard disk, and the sample data x(k) is sent to the designated fixed / mobile terminal and cloud through the WIFI board card IV-5, as the main basis for the machine learning cascade model to identify the type of malodor pollution and quantitatively estimate and predict 8+1 concentration control indicators. The computer control and analysis module IV gives the results of malodor type identification and 8+1 concentration control indicator quantitative estimation and prediction in the 30s information selection and analysis interval.

[0126] The present application flexibly completes the four major tasks of online real-time analysis of malodor pollution with a flexible machine learning cascade model: ① type identification; ② type identification and 8+1 concentration control indicator estimation; ③ 8+1 concentration control indicator estimation; ④ 8+1 concentration control indicator estimation and prediction. The present application creatively uses a flexible machine learning cascade model to flexibly complete the above four major malodor analysis tasks, embodying "flexibility", by comprehensively using the "divide and conquer" strategy, Occam's simple and effective law, and the rule of carrying forward and starting, combining offline and online learning, and combining current and recent perception information. In the specific implementation process, the present application decomposes the n malodor pollution type identification problem into n(n-1) / 2 binary classification sub-problems, decomposes the n type r(=8+1) malodor pollutant concentration indicator estimation and prediction problem into n×r single indicator estimation and prediction sub-problems, and uses single neuron, single output single hidden layer neural network and single output deep neural network as the basic members of the machine learning cascade model to solve the n(n-1) / 2+n×r sub-problems one by one according to priority.

[0127] Figure 9 shows the structure of the members of the machine learning cascade model - a single neuron, a single-output shallow neural network and a single-output deep neural network. The single shallow neural network is single-output single-hidden-layer, and the single deep neural network is single-output 3-hidden-layer. The activation functions of all hidden and output layers of the neuron, the shallow neural network and the deep neural network are f(σ) = 3(1 + exp(-σ / 3)) -1 . Figure 10 is a schematic diagram of the combination of the new sigmoid activation function and the moderate amplification of the input and output component values. The new sigmoid activation function f(σ) = 3(1 + exp(-σ / 3)) -1 replaces the original f0(σ) = (1 + exp(-σ)) -1 . After that, the significant change range of the first-order partial derivative is greatly widened, and the input component is increased from [0, 1] to [0, 6], and the output component is increased from {0, 1} to {0, 3}, which makes the learning speed of the neural network increase by more than one order of magnitude, and the generalization ability is greatly improved!

[0128] Based on the Occam's simple and effective law, the difficulty of the task is different, and the complexity of the neural network is also different. The present application learns a subtask in the order of priority of (i) neuron, (ii) single-output shallow neural network, (iii) single-output deep neural network, and uses the dynamic learning and error backpropagation algorithm with the most necessary samples, and the learning accuracy is uniformly set to root mean square error p = 0.05, and the maximum learning factor η max = 5 / N, N is the number of samples of a certain subset. Once a member with priority reaches the set accuracy, the learning process of the group is terminated, and the winner is generated, which again reflects flexibility. If a single neuron can complete a subtask, a more complex single-output shallow neural network will not be used, and so on. Therefore, a flexible machine learning cascade model is composed of winners from multiple groups, which include neurons, single-output shallow neural networks and single-output deep neural networks.

[0129] Figure 11 is a schematic diagram of the composition unit of the first level - the voting recognition group of the machine learning cascade model for malodor type recognition of the present application. Assuming there are n types of malodor, n voting recognition groups are needed, each group having n-1 members. In the learning phase, n(n-1) / 2 single-output neural network models ω jk (j, k = 1, 2, …, n and j ≠ k) learn n(n-1) / 2 normalized binary classification subsets in order of priority, and the winners participate in the subsequent voting decision. In the decision-making phase, a winner needs and only needs the votes of two recognition groups, for example, the winner ω jk participates in the voting recognition groups Ω j and Ω kThe voting process; according to the majority voting rules, among the n voting groups, the group with the most votes determines the type of sample x.

[0130] Figure 12 shows the machine learning cascade model of the present invention, which is used for both odor type identification and online real-time quantitative estimation of 8+1 odor pollutant concentration control indicators—the winning group Ω. j and the corresponding quantitative prediction group Λ j Schematic diagram. Quantitative prediction group Λ j The system consists of winners from a simple training subset of single concentrations that can be normalized through learning and achieve a set learning accuracy. It estimates the concentrations of 8+1 pollutants for each type of odor, j.

[0131] Figure 13 is a schematic diagram of the machine learning cascade model of the present invention for online real-time quantitative prediction of 8+1 odor pollutant concentration control indicators. This model is still cascaded in form, but it no longer performs type recognition; the type is assumed to be known, and it only performs continuous online quantitative prediction of the 8+1 odor pollutant concentration control indicators. The core of this invention is that the first level of the machine learning cascade model—the convolutional neural network layer—learns online in real-time the long-term sequence response data matrix X(k-τ)∈R generated by the gas-sensitive sensor array over the past 6 hours (Δk=72). 72×50 Based on this, the short-time series response of the gas-sensitive sensor array is predicted within the next 1 hour. On this basis, the second stage of the cascade model—the quantitative prediction group with known odor type—predicts the concentration control index values ​​of 8+1 odor pollutants within the next 1 hour based on prior knowledge obtained from offline learning of the odor big data set X.

[0132] The specific steps to achieve the above-mentioned invention objectives are as follows:

[0133] (1) Supervised generation of a short-time-series convolutional training subset. This is achieved by generating a long-time-series response data matrix X(k-τ) = {x(k-τ)} ∈ R from a gas-sensitive sensor array and local temperature and humidity sensors. 72×50 τ = 0, 1, ..., 71 (adjustable), automatically generates several outputs with target values. Short-time-series convolutional training subset. To shorten online real-time learning time, only 8+1 concentration control index values ​​are selectively predicted at future times (k+2)T, (k+6)T, and (k+12)T. Correspondingly, for each long-time series response component x... i (k-τ)∈R 72 Automatically generate 3 files containing the target output. Short time series convolution training subset X i (k+2), X i (k+6) and X i (k+12). The perceived component value x for predicting the future (k+2)T periods.i (k+2) is generated by x i (k-τ) The first training sample pair is {x i (k-13), x i (k-12), …, x i (k-3), x i (k-2)}-->x i (k) The second training sample pair is {x i (k-14), x i (k-13), …, x i (k-4), x i (k-3)}-->x i (k-1), until the 59th training sample pair {x i (k-71), x i (k-70), …, x i (k-61), x i (k-60)}-->x i (k-58). Therefore, the short time series training subset X i (k+2)∈R 59×12 There are 59 samples, and the dimension is 12. Similarly, X i (k+6)∈R 55×12 , X i (k+12)∈R 49×12 The specific data subset generated is shown in the “SUMMARY”-Table 1.

[0134] To speed up the learning speed, we only learn and predict the perception response time series of the normalized 16 steady-state voltage peak values v i (k), 16 area under the curve A i (k), the ambient temperature value T em (k) and the ambient humidity value H um (k), and the value corresponding to the peak time t i (k) remains unchanged, because the latter changes little in the next 1h. Therefore, the short time series convolution training subset is automatically generated, a total of 2*16*3+2*3=102.

[0135] (2) Convolutional neural network online real-time learning and prediction of gas-sensitive perception response and temperature and humidity values in the next 12 cycles 1h

[0136] The present application uses 102 single-output neural networks with a structure of 12-5-1 to form the first level of the machine learning cascade model shown in FIG. 13, a convolutional neural network layer, to learn the above-mentioned 102 short time series convolution training subsets one by one in real time. The convolutional neural network layer here is different from the single-output single-hidden-layer neural networks of the voting identification group and the quantitative prediction group of FIGS. 11 and 12 in terms of function, input and hidden node number, training subset, winner selection, etc., but the same in terms of single hidden layer, new Sigmoid activation function, learning accuracy and step factor.

[0137] After learning is completed, the 102 12-5-1 single-output neural networks predict 102 gas sensor array responses and temperature and humidity values according to the short time series {x i (k-11), x i (k-10), …, x i (k-1), x i (k)} one by one, i.e., {x i (k+2), x i (k+6), x i (k+12)}, i = 1, …, 16, 33, …, 50. These predicted sensory component values are the main body, plus the recent future peak time component values equal to their current time values, i.e., {x i (k+2) = x i (k), x i (k+6) = x i (k), x i (k+12) = x i (k)}, i = 17, …, 32, the predicted values of the sensory vectors of the future periods (k+2)T, (k+6)T, (k+12)T are generated.

[0138] (3) Prediction of 8+1 concentration control index values of malodorous pollutants in the future 12 periods within 1h

[0139] The second level of the machine learning cascade network shown in FIG. 13, the quantitative prediction group Λ j , predicts the 8+1 concentration control index values of malodorous pollutants according to the predicted .

[0140] It is particularly pointed out that an important capability of the electronic nose instrument for industrial application in the process of online real-time detection and analysis of malodor pollution is that once a perception vector x(k) is obtained through on-site real-time detection, the instrument can estimate the 8+1 concentration control index values of the malodor pollutants on-site in real time within 1 min, and it is relatively secondary to know what kind of malodor pollution it is. Here, this can be achieved by the second level-quantitative prediction group of the machine learning cascade model shown in FIG. 13.

Claims

The application discloses a fixed and active two-purpose electronic nose instrument and a method for detecting and analyzing multiple components of odor in real time. The electronic nose instrument hardware comprises a gas sensitive sensor array module I, a gas automatic headspace sampling module II, a pressurized cylinder III, a computer control and analysis module IV, a backup power supply V, and a clean air bottle VI; wherein the gas sensitive sensor array module I, the gas automatic headspace sampling module II, the pressurized cylinder III, and the computer control and analysis module IV are four components of a main machine box; the software mainly comprises a computer operation platform, upper and lower computer communication, and a machine learning model and algorithm; the electronic nose instrument flexibly realizes fixed or mobile detection of malodorous pollution sites, malodorous type identification, and quantitative estimation and prediction of concentrations of various malodorous compounds; The gas sensitive sensor array module I comprises a gas sensitive sensor array I-1, a gas sensitive sensor array annular working cavity I-2, a heat insulation layer I-3, a partition plate I-4, a fan I-5, a resistance heating wire I-6, and a temperature sensor I-7, which are located at the upper right part in the main machine box; an environment temperature and humidity sensor I-8 is located at the upper right corner of the inner wall of the front of the main machine box; The gas automatic headspace sampling gas path module II comprises a miniature vacuum pump II-1, a flowmeter II-2, a throttle valve II-3, a first two-position two-way electromagnetic valve II-4, a second two-position two-way electromagnetic valve II-5, a third two-position two-way electromagnetic valve II-6, a fourth two-position two-way electromagnetic valve II-7, a fifth two-position two-way electromagnetic valve II-8, and a purifier II-9; the pressurized cylinder III comprises a piston III-1, a piston rod III-2, a cylinder body III-3, a compression spring III-4, an O-shaped sealing ring III-5, an inner snap spring III-6, a front cover III-7, and a bump stop pad III-8; the gas automatic headspace sampling gas path module II and the pressurized cylinder III are located at the lower right part in the main machine box; The main component of the computer control and analysis module IV mainly comprises an A / D data acquisition card IV-1, a driving and control circuit board IV-2, a computer mainboard IV-3, a 4-way precision DC stabilized power supply IV-4, a WIFI board card IV-5, and a touch display screen IV-6, which are located at the left side in the main machine box; The clean air bottle VI comprises a 40L clean air large bottle VI-1 and 2-6L clean air small bottles VI-2; the 40L clean air large bottle VI-1 is mainly used for long-term continuous online detection of malodorous sites for more than 2 hours; and the 2-6L clean air small bottles VI-2 are used for limited-time short-term online real-time detection of malodorous sites for less than 2 hours; The gas sensor array I-1 is composed of 16 gas sensitive elements with partially overlapped sensing ranges; the gas sampling period of the dual active electronic nose instrument for one malodor observation point is T = 180-600s, and the default value is T = 300s; in order to meet the triangular stability principle, in the kth period corresponding to the time kT, the computer control and analysis module IV selects the steady-state voltage peak value v i,0 (k) corresponding to the peak time t i,0 (k) corresponding to the area A i,0 (k) of the response curve from the gas sensor array I-1 with a time length of 60s in the kth period; thus, the computer control and analysis module IV extracts 3*16 = 48 response components from the 16 response curves of the gas sensor array I-1 in the period kT, and adds the ambient temperature value T em,0 (k) and the ambient humidity value H um,0 (k) at that time and that place, so that the electronic nose instrument obtains an initial sensing vector x0(k) = (v 1,0 (k), …, v 16,0 (k), t 1,0 (k), …, t 16,0 (k), A 1,0 (k), …, A 16,0 (k), T em,0 (k), H um,0 (k)) T ∈R 50 , which is called an initial sample and is stored in the data file of the hard disk of the computer mainboard IV-3, and the data x0(k) is sent to the designated fixed / mobile terminal and cloud through the WIFI board card IV-5; The fixed and mobile electronic nose instrument realizes long-term continuous online or short-term multiple detection of various malodorous pollution sites, forms a malodorous big data set X, and establishes a corresponding relationship between a gas sensitive sensor array response sample and a malodorous type and 8+1 malodorous pollutant concentration control indexes, i.e., 8 single components of hydrogen sulfide H2S, ammonia NH3, methyl mercaptan CH4S, dimethyl disulfide C2H6S2, dimethyl sulfide C2H6S, carbon disulfide CS2, trimethylamine C3H9N, and styrene C8H8, and 1 overall type odor concentration specified in GB14554-1993 “Malodorous Pollutant Discharge Standard”. In the learning stage, the computer control and analysis module IV learns the large data set X of malodor off-line to determine its own structure and parameters, and to master the corresponding relationship between the response of the gas sensor array to the sample and the pollution type and the 8+1 concentration control indicators; in the decision-making stage, according to the current perception vector x0(k) of the gas sensor array I-1, the machine learning cascade model quantitatively estimates the 8+1 concentration control indicators of the malodor pollutants specified in GB14554 in real time, and determines the malodor pollution type as needed; further, the machine learning cascade model predicts the perception response of the gas sensor array I-1 in the next 1h in real time by learning the long-time sequence perception vector generated by the gas sensor array I-1 in the last 72 cycles of about 6h, and accordingly predicts the 8+1 concentration control indicators of the malodor pollutants in the next 1h in real time. The fixed active two-way electronic nose instrument and the malodor multi-component online real-time detection and analysis method according to claim 1 are characterized in that, The gas sensor array I-1 and its annular working cavity I-2 are located in a constant-temperature room at 55±0.1℃; in the gas sampling period default value T=300s, the gas sensor array I-1 and its annular working cavity I-2 successively experience the following 5 stages according to the difference in gas type and flow rate: ① accurate calibration: under the action of 1,000ml / min clean air blowing, the gas sensor array I-1 gradually and accurately recovers to the standard state for 60s; ② equilibrium: the gas sensor array annular working cavity I-2 has no gas flow for 10s; ③ headspace sampling: under the action of the micro vacuum pump II-1, the gas sensor array I-1 generates a sensitive response to 1,000ml / min of the measured malodor gas for 30s; ④ pressure multiplication: under the action of 0.2Mpa clean air thrust, 78ml of malodor gas in the pressure cylinder III is pressed back to the gas sensor array annular working cavity I-2, the pressure in the cavity increases by about 1 time for a short time, and the sensitive response of the gas sensor array I-1 is further increased for 30s; ⑤ waste gas discharge: under the action of the micro vacuum pump II-1, the waste gas in the gas sensor array annular working cavity I-2, the pressure cylinder III and the related gas path pipeline is discharged to the atmosphere for 170s; as the malodor waste gas is discharged, the gas sensor array I-1 gradually and approximately recovers to the initial state; if the gas sampling period T>300s, the waste gas discharge time is correspondingly extended, and the flow rates and time lengths of the other 4 stages remain unchanged; for example, if T=480s, the waste gas discharge time length is 170+(480-300)=350s; if T<300s, the waste gas discharge time length is 170+(T-300)<170s. ​ The fixed active two-way electronic nose instrument and the malodor multi-component online real-time detection and analysis method according to claim 1 are characterized in that, According to the ideal gas state equation p1V1 = p2V2 under constant temperature condition and the classical Langmuir isothermal low-pressure adsorption equation q = γpq / (1 + γp), wherein p1, p2 and p are pressure, V1 and V2 are volume, q is adsorption amount, q m is saturated adsorption amount, and γ is equilibrium constant m The present application increases the adsorption amount of micro-amount components with the increase of pressure by means of the pressurized cylinder III. The specific measures are as follows: designing a small-volume and small-section gas-sensitive sensor array annular working cavity I-2 with a medium diameter of φ140 mm and two section sizes of 14x10 mm and 16x20 mm and a net volume of 78 ml; designing a one-way introduction ultra-thin piston type pressurized cylinder III, wherein the diameter of the piston III-1, i.e. the cylinder diameter, is φ50 mm, the diameter of the piston rod III-2 is φ18 mm, the theoretical maximum stroke is 40 mm, and the theoretical net volume of the cylinder is 78 ml; the working mode of the pressurized cylinder III is that the piston III-1 and the piston rod III-2 are retracted under the driving of 0.2 Mpa clean air and are reset by means of the compression spring III-4 on the piston rod III-2; during the ⑤ exhaust emission stage of the gas sampling period kT and the ① accurate calibration, ② equilibrium and ③ headspace sampling stages of the next period (k+1)T, the piston rod III-2 is extended under the tension of the compression spring III-4, the piston III-1 returns to the initial position, and the maximum net volume of the cylinder reaches 78 ml, which is used for storing 78 ml of the measured malodorous gas; during the ④ pressure doubling stage, the piston III-1 and the piston rod III-2 move the maximum stroke of 40 mm under the driving force of 0.2 Mpa clean air, the compression spring III-4 is compressed, the static volume of the cylinder gradually changes from 78 ml to 0, and the 78 ml of the measured malodorous gas stored in the cylinder body III-3 is compressed back to the gas-sensitive sensor array annular working cavity I-2; the electronic nose instrument doubles the pressure in the gas-sensitive sensor array annular working cavity I-2 for a short period of time by means of the pressurized cylinder III, so as to increase the adsorption amount of the sensitive membrane to the micro-amount components, which is equivalent to increasing the concentration of the micro-amount components of the malodorous gas and doubling the sensing sensitivity of the gas-sensitive sensor array. The fixed-active two-way electronic nose instrument and the method for detecting and analyzing multiple components of malodors in real time on line according to claim 1 are characterized in that, The booster cylinder III is arranged at the rear end of the gas outlet of the gas sensor array annular working cavity I-2, all the gas flowing through the gas sensor array annular working cavity I-2 flows through the booster cylinder III, and finally is discharged into the atmosphere as waste gas; the booster cylinder III is provided with two gas inlets and one gas outlet: the gas inlet 1 is communicated with the gas outlet of the gas sensor array annular working cavity I-2, the gas inlet 2 is communicated with the gas outlet of the clean air bottle VI by means of the two-position two-way electromagnetic valve II-8, and the gas outlet of the booster cylinder III is communicated with the flow control and waste gas discharge gas path by means of the two-position two-way electromagnetic valve II-5; the micro vacuum pump II-1 sucks the odor gas molecules remaining in the gas sensor array annular working cavity I-2, the booster cylinder III and the related gas path pipeline for a long time at a flow rate of 6,000 ml / min, and the gas sensor array I-1 is thus preliminarily restored to the reference state; under the continuous purging action of 1,000 ml / min clean air for 60 s, the gas sensor array I-1 is accurately restored to the reference state. The fixed active two-way electronic nose instrument and the malodor multi-component online real-time detection and analysis method according to claim 1 are characterized in that, The electronic nose instrument is flexibly used in (i) a field scene without alternating current and (ii) a field scene with alternating current, and flexibly performs (A) a short-term limited-time real-time detection and analysis task and (B) a long-term continuous online detection and analysis task lasting for several hours, 1 day, several days or even several months; the working scene (i) + task (A): the instrument uses a backup power supply - 25.2V lithium battery pack V, the nominal capacity is 10Ah, the battery pack weighs 2.5kg, and the continuous working time is >3h, which can be charged; at the same time, the instrument also uses a 2-6L aluminum alloy seamless clean air bottle VI-2, the maximum working pressure is 15MPa, the size is φ111*359mm-φ140*629mm, the air bottle weighs 2.8kg-7.2kg (valve 0.8Kg not included), and the maximum continuous working time is >2.5h calculated by the clean air accurate calibration time of 60s; the working scene (ii) + task (A): the instrument uses an alternating current power supply and only uses a 2-6L aluminum alloy seamless clean air bottle VI-2 to carry out short-term detection and analysis of odor pollution for about 2h; the working scene (ii) + task (B): the instrument uses an alternating current power supply, and at the same time uses a 40L clean air bottle VI-1, the maximum working pressure is 15MPa, the size is φ219*1450mm, the air bottle weighs 45kg, still calculated by the clean air accurate calibration time of 60s, 1 large bottle of clean air can work for about 20 days, and timely replacement of the used clean air bottle VI-1 can continuously detect and analyze for a longer time. The fixed-active two-way electronic nose instrument and the method for detecting and analyzing multiple components of malodors in real time on line according to claim 1 are characterized in that, Based on the existing manufacturing process level and existing products of gas sensitive element, the measures to significantly improve the sensitivity and selectivity of gas sensitive perception from the aspects of instrument structure design and industrial application include: (i), the I-2 structure design of the 14*10mm and 16*20mm small cross-section size and medium diameter φ140mm and 78ml small volume gas sensitive sensor array annular working cavity, the 1,000ml / min headspace sampling large flow and the 78ml net volume booster cylinder III make the pressure in the gas sensitive sensor array annular working cavity I-2 increase by several times in a short time, and the optimized combination of the three makes the gas sensitive sensitivity increase by several times; (ii), the three perception components of steady-state peak value, peak time and curve area are selected from the single gas sensitive sensor voltage response curve to significantly improve the gas sensitive selectivity. The fixed-active two-way electronic nose instrument and the method for detecting and analyzing multiple components of malodors in real time on line according to claim 1 are characterized in that, From the beginning of the 6th second of the ② equilibrium phase of the kth gas sampling cycle, through the ③ headspace sampling and ④ pressure multiplication phases, and until the end of the 5th second of the ⑤ waste gas discharge phase, the computer control and analysis module IV records and stores the sensing information of the gas sensor array I-1, for a duration of 70 seconds; from the beginning of the 6th second of the ⑤ waste gas discharge phase of the cycle until the end of the 35th second, the electronic nose instrument enters an information processing and analysis phase lasting 30 seconds, and the computer control and analysis module IV selects the steady-state peak value v i,0 (k) from each of the 70-second voltage response curves of the gas sensors i,0 (k) from each of the 70-second voltage response curves of the gas sensors i,0 (k) from each of the 70-second voltage response curves of the gas sensors, plus the on-site environmental temperature and humidity T em,0 (k) from each of the 70-second voltage response curves of the gas sensors, plus the on-site environmental temperature and humidity T um,0 (k) from each of the 70-second voltage response curves of the gas sensors, plus the on-site environmental temperature and humidity T 50 ; The modular machine learning cascade model determines the odor pollution type, quantitatively estimates and predicts 8+1 concentration control index values. The fixed active two-way electronic nose instrument and the method for on-line real-time detection and analysis of multi-component malodor according to claim 1, characterized in that, in ③In the headspace sampling stage, the first and second two-way solenoid valve II-4 and the second two-way solenoid valve II-5 are turned on, the third two-way solenoid valve II-6, the fourth two-way solenoid valve II-7 and the fifth two-way solenoid valve II-8 are turned off, the piston rod III-2 of the booster cylinder III extends under the elastic force of the compression spring III-4, and the cylinder body III-3 thus generates a 78ml storage space; under the suction of the micro vacuum pump II-1, the odor gas flows through the first and second two-way solenoid valve II-4, the gas sensitive sensor array annular working cavity I-2 and the gas sensitive sensor array I-1, the cylinder body III-3, the second two-way solenoid valve II-5, the throttle valve II-3 and the flowmeter II-2 in turn at a flow rate of 1,000ml / min, and finally is discharged into the atmosphere, lasting for 30s; during the flow of the odor gas, the gas sensitive sensor array I-1 generates a sensitive response. The fixed active two-way electronic nose instrument and the method for on-line real-time detection and analysis of multi-component malodor according to claim 1, characterized in that, in ④In the pressure multiplication stage, the fifth two-way solenoid valve II-8 is turned on, and the first two-way solenoid valve II-4, the second two-way solenoid valve II-5, the third two-way solenoid valve II-6 and the fourth two-way solenoid valve II-7 are turned off, under the thrust of 0.2MPa clean air, the piston rod III-2 of the booster cylinder III retracts, and the 78ml odor gas stored in the cylinder body III-3 is pressed back to the gas sensitive sensor array annular working cavity I-2, lasting for 30s; according to the ideal gas state equation p1V1=p2V2 under constant temperature conditions, the pressure in the gas sensitive sensor array annular working cavity I-2 gradually increases by 1 times, and according to the Langmuir isothermal low pressure adsorption equation, the response of the gas sensitive sensor array I-1 increases by about 1 times based on the response in the ③ headspace sampling stage. The fixed active two-way electronic nose instrument and the malodor multi-component online real-time detection and analysis method according to claim 1 are characterized in that, In the ⑤ exhaust stage, the second two-position two-way solenoid valve II-5 and the fourth two-position two-way solenoid valve II-7 are turned on, the first two-position two-way solenoid valve II-4, the third two-position two-way solenoid valve II-6 and the fifth two-position two-way solenoid valve II-8 are turned off, the piston rod III-2 is extended again due to the elastic force of the compression spring III-4, and the cylinder body III-3 generates a space of 78ml again; under the suction of the miniature vacuum pump II-1, the residual exhaust gas in the gas sensor array annular working cavity I-2 and the cylinder body III-3 is extracted at a theoretical flow rate of 6,000ml / min, and the gas sensor array I-1 gradually recovers to the initial baseline state; then, in the ① accurate calibration stage of the next cycle, the second two-position two-way solenoid valve II-5 and the third two-position two-way solenoid valve II-6 are turned on, the first two-position two-way solenoid valve II-4, the fourth two-position two-way solenoid valve II-7 and the fifth two-position two-way solenoid valve II-8 are turned off, and the gas sensor array I-1 is accurately recovered to the baseline state under the blowing action of 1,000ml / min clean air for 60s, and the residual malodorous molecules in the cylinder body III-3 and the inner wall of the related pipeline are completely removed. The fixed-active two-way electronic nose instrument and the method for detecting and analyzing multiple components of malodors in real time on line according to claim 1 are characterized in that, The application takes 8 kinds of foul odor pollutants specified in GB14554, hydrogen sulfide H2S, ammonia NH3, methyl mercaptan CH4S, dimethyl disulfide C2H6S2, dimethyl sulfide C2H6S, carbon disulfide CS2, trimethylamine C3H9N, styrene C8H8, 5 kinds of standard smelly liquid specified in HJ1262-2022 for olfactory qualification examination, methyl cyclopentanone C6H 10 O, beta-phenylethanol C8H 10 O, gamma-undecanolactone C 11 H 20 O2, beta-methyl indole C9H9N, isovaleric acid C5H 10 O2 and n-butanol C4H 10 O, a total of 14 typical foul odor compounds are analysis objects, and 0.1-100 ppm The single or mixed sample is detected by the electronic nose instrument, and the relationship between the multi-dimensional perception information of the gas sensor array I-1 and the molecular weight, structure, functional group micro-characteristics and odor concentration macro-characteristics of the 14 kinds of malodorous compounds is established. The malodor big data set X includes: (i), the off-line detection data of the electronic nose instrument on the above-mentioned 14 kinds of typical malodorous compound samples; (ii), a large amount of on-line real-time detection data of the electronic nose instrument in multiple malodor pollution sites; (iii), the odor concentration data determined by the smell discrimination of the smell discrimination team in the laboratory on a large number of malodorous samples collected in the field; (iv), the laboratory component analysis data of a large number of malodorous samples collected in the field by color, mass spectrometry and other conventional instruments; (v), the atmospheric temperature / humidity data in the malodor pollution detection site; (vi), the adjacent resident complaint data; (vii), the corresponding relationship data between the multi-dimensional perception sample of the gas sensor array I-1 and the malodor type and 8+1 concentration control indicators. The fixed active two-way electronic nose instrument and the method for on-line real-time detection and analysis of multi-component malodor according to claim 1, characterized in that, in In the kth gas sampling period, the computer control and analysis module IV selects the steady-state peak value v i,0 (k), the time t of peak appearance i,0 (k), the total area A under the curve i,0 (k) The three initial perception information components, the normalization value calculation formula is respectively: (a), steady state peak v i (k): In formula (1), v max = 10.0 V is the working loop voltage value of the gas sensor (b), peak time t i (k): In formula (2), t max = 30 + 30 = 60 s is the sum of the time lengths of the two stages of headspace sampling and pressure multiplication after the computer control and analysis module IV issues a "headspace sampling" instruction. (c), total area under the curve A i (k): In formula (3), the maximum area is assumed to be t max *v max = 10 * 60 = 600; Further, assuming an environmental maximum temperature of 45 °C, the normalized environmental temperature is T em (k) = T em,0 (k) / 45; the normalized environmental humidity is directly the relative humidity value, i.e. H um (k) = H um,0 (k) / 100; The normalized samples are x(k) = (vi(k),..., v 16 (k), ti(k),..., ti 16 (k), Ai(k),..., Ai 16 (k), Ti em (k), Hi um (k)) T = (xi(k),..., xi 16 (k), xi 17 (k),..., xi 32 (k), xi 33 (k),..., xi 48 (k), xi 49 (k), xi 50 (k)) T ∈ R 50 , each input component is dimensionless after normalization. The fixed active two-way electronic nose instrument and the malodor multi-component online real-time detection and analysis method according to claim 1 are characterized in that, The present application flexibly uses the "divide and conquer" strategy to transform the pollution type recognition and 8+1 pollution concentration control index estimation and prediction problems implied by the large odor data set X into multiple single-target recognition and multiple single-concentration index estimation and prediction problems, that is, the n odor pollution type recognition problems are divided into C n 2 =n(n-1) / 2 binary classification sub-problems, and the n type r=8+1 odor pollution concentration index estimation and prediction problems are divided into n x r single-index one-to-one estimation and prediction problems; according to the Occam "simple and effective" law, a group of "three people" consisting of a single neuron with priority 1, a single-output single-hidden-layer neural network with priority 2, and a single-output deep neural network with priority 3, corresponding to n(n-1) / 2 binary classification subset numbers + nr single-concentration index subset numbers; the members of the group learn the n(n-1) / 2+nr subsets formed by the "divide and conquer" decomposition of the large odor data set X in order of priority. The basic parameter settings of the neural network learning process are as follows: based on the conventional Sigmoid, ReLU, Softmax activation function, a new type of Sigmoid activation function f(σ) = 3(1 + exp(-σ / 3)) is found -1 , σ is the total input of a certain hidden node or output node; a dynamic learning mechanism and algorithm that allows the most necessary samples to participate is established; the input components are proportionally enlarged to [0, 6], the target output components for classification are enlarged from {0, 1} to {0, 3}, the target output components for concentration estimation and prediction are enlarged from (0, 1) to (0, 3), the learning step factor η = 0.05 / N, where N is the number of training subset samples, the learning termination condition is root mean square error δ ≤ 0.075, the maximum number of learning steps is ζ max = 10,000; the single-output single-hidden-layer neural network has 8 hidden nodes s; the deep neural network adopts a layer-by-layer learning algorithm based on peer single-hidden-layer networks; According to the Occam's simple and effective law, in any "three-person" group, once the single neuron learning accuracy-root mean square error δ of the priority 1 meets the requirements, it becomes the winner automatically, and the remaining 2 members do not need to learn any more, and so on; thus, one winner with simple structure and effective problem solving is generated in each group to participate in decision-making; n(n-1) / 2 classification winners form n voting decision groups, each consisting of n-1 winners, and each classification winner participates in and must participate in 2 voting decision groups; nr concentration estimation and prediction winners form n quantitative prediction groups, each consisting of r winners. ​ The fixed-active two-way electronic nose instrument and the method for detecting and analyzing multiple components of malodors in real time on line according to claim 1 are characterized in that, The fixed and living two-convenient electronic nose instrument detects the real-time or continuous online of the malodorous pollution site, and the computer control and analysis module IV performs the online real-time analysis task of the concentration index value of 8+1 kinds of malodorous pollutants specified in GB14554-1993, which is divided into the following three cases according to the size of the continuous detection cycle number k: (a) k = 1: corresponding to sample x(1), machine learning cascade model first stage - winning group Ω of n voting decision groups j determine the malodor pollution type, cascade model second stage - corresponding quantified prediction group Λ j of 8+1 winners - estimate 8+1 concentration control index values j ​ (b) 1 < k < 72: corresponding to samples x(2) ~ x(72), the first stage of the cascade model no longer votes, and the malodor pollution type defaults to Ω j , the second stage - the 8+1 winners of the quantization prediction group Λ j continue to estimate the 8+1 concentration control indicator values one by one; (c) k > 72: This corresponds to kT > 6h when the sample is continuously detected and taken at T = 300s, the first stage of the cascade model still does not vote, and the default odor pollution type is Ω j ; Follow the "before and after" rule, use the long time series response generated by the gas sensitive sensor array I-1 in the last 72 cycles (about 6h) to predict the future 1h of the multi-concentration index value of the odor pollution, focusing on predicting the 8+1 concentration control index values of the future k = 2 (10min), k = 6 (30min) and k = 12 cycles (60min). The fixed-active two-way electronic nose instrument and the method for detecting and analyzing multiple components of malodors in real time on line according to claim 1 are characterized in that, When the number of consecutive sampling cycles (k>72) of the electronic nose instrument for a gas sample from an odorous pollution point is constant, the gas-sensitive sensor array generates a long-term normalized data matrix {X(k-71),…,X(k)} with a response length of Δk=72 in the most recent 72 cycles. Each component is automatically generated from a portion of its own time series to form three short-term training subarrays with a duration of ▽k=12. The i-th long-term component {x} with a duration of Δk=72... i (k-71),…,x i (k-τ),…,x i (k)}∈R 72 For example, as shown in Tables 1a-c, X is generated. i (k+2), X i (k+6), X i These three training subsets of short time series convolutions with a dimension of 12 (k+12) and sample sizes of 59, 55 and 49 respectively are used for online learning by three single-output shallow neural networks with a 12-5-1 structure, so as to predict the value of the i-th perceptual component at future times (k+2)T, (k+6)T and (k+12)T starting from kT. To speed up the learning, assume the normalized time to peak t i (k) The corresponding gas sensitive sensor sensing component time series remain unchanged at time (k+2)T, (k+6)T, (k+12)T, no time series convolution training subset is generated; therefore, there are 2*16*3+2*3=102 short time series convolution training subset arrays, which are learned online by 102 single-output shallow neural networks with a structure of 12-5-1 one by one within 30s, and the normalized time series response {x(k+2), x(k+6), x(k+12)} of the gas sensitive sensor array about to occur in the future k+2 period, i.e. the 10th minute, the k+6 period, i.e. the 30th minute, and the k+12 period, i.e. the 60th minute, is predicted online accordingly; and Ω j The corresponding quantized prediction group Λ j Accordingly, the 8+1 concentration control index values at the future (k+2)T, (k+6)T, (k+12)T time points are predicted in turn. The fixed active two-way electronic nose instrument and the malodor multi-component online real-time detection and analysis method according to claims 1-15 are characterized in that, The fixed and living two-convenient electronic nose instrument detects the real-time or continuous online of the malodorous pollution site, and the computer control and analysis module IV performs the online real-time analysis task of the concentration index value of 8+1 kinds of malodorous pollutants specified in GB14554-1993, which is divided into the following three cases according to the size of the continuous detection cycle number k: (1) Preparation: (i), there is an alternating current power supply for real-time work scene, carry the main box and 2-6L clean air bottle VI-2; (ii), there is an alternating current power supply and continuous online work scene, fixed installation main box and 40L clean air bottle VI-1; (iii), there is no alternating current power supply for outdoor work scene, carry the main box, spare power supply-25.2V lithium battery V and 2-6L clean air bottle VI-2; (2) Start: instrument preheating 30 minutes; set "gas sampling period T", default value T=5min; The second two-position two-way electromagnetic valve II-5 and the fourth two-position two-way electromagnetic valve II-7 are turned on, the first two-position two-way electromagnetic valve II-4, the third two-position two-way electromagnetic valve II-6 and the fifth two-position two-way electromagnetic valve II-8 are turned off, the piston rod III-2 is stretched out due to the elastic force of the compression spring III-4, and the cylinder body III-3 generates 78ml space; under the suction of the micro vacuum pump II-1, the residual waste gas in the gas sensitive sensor array ring working cavity I-2 and the cylinder body III-3 is extracted at a flow rate of 6,000ml / min, and the gas sensitive sensor array I-1 gradually recovers to the baseline state; (3) Start of the malodorous gas online real-time detection and analysis cycle: (3.1) Malodorous gas detection: In a gas sampling period T, according to the different types and flow rates of gases, the gas sensitive sensor array module I goes through the following five stages in turn: ① accurate calibration, ② balance, ③ headspace sampling, ④ pressure multiplication, ⑤ waste gas discharge; Table 2 shows the gas type, sampling time length, starting time, flow rate, and on / off status of the five two-position two-way electromagnetic valves at T=300s; Table 2, Gas sampling period T = 300 s (default), working parameters of gas sensor array module I and related electromagnetic valve on / off status From the beginning of the 6th second of the waste gas discharging phase T-⑤ until the end of the 35th second, the electronic nose instrument enters the information processing and analysis phase lasting 30 seconds; from the beginning of the 6th second of the balance phase ②, including the entire headspace sampling and pressure multiplication phase ③ and ④, until the end of the 5th second of the waste gas discharging phase ⑤, the computer control and analysis module IV records and stores the sensing information of the gas sensitive sensor array I-1, lasting 70 seconds, and then selects the steady-state peak value v i,0 (k) from each gas sensitive sensor voltage response curve in the time period of 70 seconds i,0 (k), the area under the curve A i,0 (k) These three sensing information components, 16*3=48 information components of the gas sensitive sensor array I-1 plus the ambient temperature T em,0 (k) and the ambient humidity H um,0 (k) together form the initial sample x0(k)∈R 50 ; (3.2) Malodorous gas analysis: (a) The first k=1 gas sampling period: the electronic nose instrument detects a certain malodorous pollution point for the first time to obtain the initial sample x0(1), which is normalized as x(1), and the computer control and analysis module IV machine learning cascade model simultaneously performs malodorous pollution type identification and 8+1 typical malodorous compound concentration index quantification estimation and prediction; (b) The second k=2-71 gas continuous sampling period: the electronic nose instrument detects a certain malodorous pollution point for k<72 cycles of continuous online real-time detection to obtain the normalized finite time sequence samples {x(1), x(2), …, x(k<72)} including x(1), and the malodorous pollution type is still represented by the winning group of the first level of the machine learning cascade model, and the second level of the cascade model only needs to perform 8+1 typical compound concentration index quantification estimation. (c) k≥72 gas continuous injection cycles: the electronic nose instrument detects a certain malodorous pollution point for k≥72 cycles continuously online in real time, and obtains a normalized time sequence response sample set {x(l),...,x(72),x(73),...,x(k)}; the latest 72 continuous samples are intercepted to form a moving long time sequence sample set X(k-t)={x(k-t)}e R with Ak=72 72×50 , t=0,1,...,71; for example, a recent Ak=72 moving long time sequence is {x(5),...,x(72),...,x(k=76)}; next, for predicting the response values of the k+2, k+6, k+12 cycles in the future, 102 supervised Ak=12 short time sequence convolution training subsets {X i (k+2)e R 59×12 ,X i (k+6)e R 55×12 ,X i (k+12)e R 49×12} are generated from X(k-t), i=1,...,16,33,...,50; thirdly, 102 12-5-1 single output neural networks learn the 102 Ak=12 short time sequence convolution training subsets online in real time; after the learning is completed, the 102 12-5-1 single output neural networks immediately predict the response component values and temperature and humidity values of the gas sensitive sensor array in the k+2, k+6, k+12 cycles in the future according to the Ak=12 short time sequences {x i (k-11),x i (k-10),...,x i (k-1),x i (k)} and compare them with the actual values of the k+2, k+6, k+12 cycles in the future, respectively, to obtain the prediction accuracy of the 102 12-5-1 single output neural networks The sample prediction values of the future k+2, k+6, k+12 periods are generated in combination with the peak time component not participating in the prediction Finally, the second stage of the machine learning cascade model, the quantitative prediction group corresponding to the winning group, predicts the 8+1 concentration control index values at the corresponding time according to the three sample prediction values; (4) Once the gas sampling period kT of the same malodorous pollution point ends, or enters the next period (k+1) T, the process (3) is repeated, or it is transferred to the next malodorous pollution point to start new detection and analysis.

Images