CN108896706B - The foul gas multiple spot centralization electronic nose instrument on-line analysis of big data driving - Google Patents
The foul gas multiple spot centralization electronic nose instrument on-line analysis of big data driving Download PDFInfo
- Publication number
- CN108896706B CN108896706B CN201810471708.3A CN201810471708A CN108896706B CN 108896706 B CN108896706 B CN 108896706B CN 201810471708 A CN201810471708 A CN 201810471708A CN 108896706 B CN108896706 B CN 108896706B
- Authority
- CN
- China
- Prior art keywords
- gas
- gas sensor
- sensor array
- response
- solenoid valve
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/0004—Gaseous mixtures, e.g. polluted air
- G01N33/0009—General constructional details of gas analysers, e.g. portable test equipment
- G01N33/0027—General constructional details of gas analysers, e.g. portable test equipment concerning the detector
- G01N33/0031—General constructional details of gas analysers, e.g. portable test equipment concerning the detector comprising two or more sensors, e.g. a sensor array
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/0004—Gaseous mixtures, e.g. polluted air
- G01N33/0009—General constructional details of gas analysers, e.g. portable test equipment
- G01N33/0027—General constructional details of gas analysers, e.g. portable test equipment concerning the detector
- G01N33/0031—General constructional details of gas analysers, e.g. portable test equipment concerning the detector comprising two or more sensors, e.g. a sensor array
- G01N33/0034—General constructional details of gas analysers, e.g. portable test equipment concerning the detector comprising two or more sensors, e.g. a sensor array comprising neural networks or related mathematical techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G08—SIGNALLING
- G08C—TRANSMISSION SYSTEMS FOR MEASURED VALUES, CONTROL OR SIMILAR SIGNALS
- G08C17/00—Arrangements for transmitting signals characterised by the use of a wireless electrical link
- G08C17/02—Arrangements for transmitting signals characterised by the use of a wireless electrical link using a radio link
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/30—Services specially adapted for particular environments, situations or purposes
- H04W4/38—Services specially adapted for particular environments, situations or purposes for collecting sensor information
Landscapes
- Engineering & Computer Science (AREA)
- Chemical & Material Sciences (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Biochemistry (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Food Science & Technology (AREA)
- Medicinal Chemistry (AREA)
- Analytical Chemistry (AREA)
- Artificial Intelligence (AREA)
- Computer Networks & Wireless Communication (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Combustion & Propulsion (AREA)
- Signal Processing (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Investigating Or Analyzing Materials By The Use Of Electric Means (AREA)
- Investigating Or Analyzing Materials By The Use Of Fluid Adsorption Or Reactions (AREA)
Abstract
The present invention relates to the foul gas multiple spot centralization electronic nose instrument on-line analysis of big data driving, stench big data includes: that the gas sensor array on-line checking data of stench standard sample and pollution scene, laboratory smell and distinguish that the conventional instruments offline inspection data such as data, color/mass spectrum and resident complain data;The present invention first regards a variety of odorant pollutant concentration sealings and forecasting problem as multiple gas sensor responses forecasting problem one by one, then regards a variety of concentration values forecasting problem one by one as;Machine learning model is made of modularization convolutional neural networks layer and modularization deep neural network level connection;The response of convolutional neural networks layer on-line study gas sensor array recent times sequence, and imminent response is predicted accordingly;Deep neural network layer off-line learning stench big data is responsible for a variety of odorant pollutant concentration of prediction.Analysis method of the invention can realize the circulation On-line Estimation and prediction of a variety of odorant pollutant concentration Con trolling index values in multiple monitoring points.
Description
Technical field
The present invention-big data driving foul gas multiple spot centralization electronic nose instrument on-line analysis, Environment Oriented
The market surpervision demand of protection and administrative department, it is raw towards industrial park, rubbish and sewage diposal area, farm, neighbouring resident
The on-line monitoring and analysis demand in the odor pollutions regions such as area living, be related to environmental protection, analytical chemistry, computer, artificial intelligence,
The technical fields such as big data mainly solve stench electronic nose instrument to the on-line monitoring of a variety of odorant pollutants in odor pollution region
With the On-line Estimation and forecasting problem of a variety of concentration Con trolling index values.
Background technique
" stench " refers in particular to unpleasant stink, is the olfactory organoleptic of all stimulation people, and damage human habitat makes us being difficult to
It endures or the common name of offending smell, sometimes referred to as " peculiar smell ".Odorant pollutant refers in particular to all foul gas substances, refers to all
Distribute the substance of foul odour.Odorant pollutant is widely present in petrochemical industry, rubbish and sewage treatment, pharmacy, cultivation etc. one
It cuts with the enterprise of exhaust gas discharge and adjacent to residential block, distribution is very wide, and coverage is very big.China's odor pollution status is to discharge
Source is numerous, and stink odor is complicated, and national standard lag, resident complains and takes place frequently.
Odor pollution evaluation object is foul gas, and evaluation method, which is divided into, smells the method for distinguishing and instrumental method.GB14554-93
" emission standard for odor pollutants " regulation, odorant pollutant emission control index include a kind of qualitative dimensionless odor concentration and 8
The quantitative single component concentration of kind, i.e. trimethylamine (C3H9N), styrene (C8H8), hydrogen sulfide (H2S), methyl mercaptan (CH4S), first sulphur
Ether (C2H6S), methyl disulfide (C2H6S2), ammonia (NH3), carbon disulfide (CS2).In addition, GB/T18883-2002 " Interior Space makings
Amount standard " special recommendation sulfur dioxide (SO2) and total volatile organic compounds (Total volatile organic
Compound, TVOC) the quantitative Con trolling index of this 2 kinds of concentration.At this stage, odor pollution assessment indicator system is mainly by above-mentioned a kind
Qualitative index and 10 kinds of quantitative targets are constituted.GB14554 regulation measures odor concentration triangle odor bag method, measurement
C3H9N、C8H8、H2S、CH4S、C2H6S、C2H6S2Concentration gas chromatography measures NH3And CS2Concentration uses spectrophotometry;
GB/T18883 regulation, measurement TVOC concentration use gas chromatography;GB/T15262-94 regulation, measures SO2Using spectrophotometric
Method.
" odor concentration " refers toCollection in worksiteFoul smell sample existLaboratoryExtremely with odorless clean air serial dilutionIt smells and distinguishes MemberThe extension rate of odor threshold, EU criteria EN17325-2003 OU (odor unit) value metric.Currently, odor concentration
Standard discrimination method mainly by the nose for smelling the person of distinguishing!The countries and regions such as China, America and Europe, Japan and Korea S are such.Implement 25
The national standard GB/T14675-93 " surrounding air-stench measurement-triangle odor bag method " in year, which has been standardized, to be smelt the person's of distinguishing choice, dislikes
The acquisition of odour sample and Sample intraocular, which dilute and smell, distinguishes three links such as measurement.The country such as America and Europe, Australia, New Zealand uses
Dynamic olfactometer dilutes foul smell sample.
GB/T14675 and HJ905 regulation, foul gas sample are first used sampling bottle or odorless airbag at the scene by staff
(such as 10L) acquisition, then transports back to smell and distinguishes room, then aspirated by a certain percentage with syringe and move to odorless airbag (such as 3L) simultaneously
It is diluted with odorless clean air, is finally distinguished that group member smells by smelling and distinguished.Triangle odor bag method core first is that: foul smell sample
After dilution is primary, smelling the person of distinguishing for one needs 3 3L airbags of smelling, wherein 1 is to have foul smell bag after diluting, another 2 are no foul smell
Bag, and can therefrom identify foul smell bag.
" selecting the full subjective judgement after smelling the person's of distinguishing smelling wrong to choosing ".Although GB/T14675 has implemented 25 years, status
It is that many odorants or the olfact difference provided without olfact or country variant or tissue are very big.2015,
Tianjin Environmental Science Institute national environmental protection odor pollution control key lab is from the expectation for having more statistical significance, tissue 30
Name smells the person of distinguishing (male 13 people, 17 people of female) and has carried out odor threshold measurement to 40 kinds of odorants.The result shows that NH3Odor threshold
5 times are differed with Japan, H2Nearly 3 times of S difference, trimethylamine differs 28.12 times, and positive valeric acid differs 65.67 times, etc..The above results
At least illustrate two problems: (1) determine odor concentration to smell the process of distinguishing very complicated, smell and comment a cost very big;(2) each list in various countries
The odorant odor threshold itself that position provides is not objective, does not have repeatability.
For triangle odor bag method as defined in GB/T14675 although ordinary people's impression can be embodied, operability is very poor, does
It once smells and distinguishes that test needs largely to sample and smell the personnel of distinguishing, cost is very high, is particularly unsuited for low concentration and smelling for noxious material is distinguished.
Smelling for triangle odor bag method comments result quality to be selected by 1. spot sampling point;2. sampling apparatus;3. laboratory condition;4. smelling
The person's of distinguishing ability and state;5. odor concentration and Initial dilution multiple;Distinguish that the factors such as time and fatigue influence 6. smelling, it is therein
It artificial sample, manual dilution and manually smells the method for distinguishing there are many limitations.
Due to smelling the method for distinguishing and conventional instrument analytic approach poor in timeliness, cost is high;It is harmful to the human body due also to smelling the method for distinguishing, smells and distinguish
As a result not objective, olfactory analog-Electronic Nose Technology is therefore especially noticeable with instrument.
Electronic Nose Technology has a extensive future, development trend first is that, development is highly sensitive, highly selective sensor for gas
Part, to realize the qualitative and quantitative detection and analysis of smell.It is encouraging that SnO2Semiconductor gas sensors device sensitivity is up to 10-9V/V (ppb) order of magnitude directly generates the response of V step voltage to smell, is not required to secondary amplification, this online prison to odorant pollutant
Survey is very attractive.Electronic Nose Technology development trend second is that, to have multiple and different type air-sensitives of necessary sensitivity
Element forms array, improves the selectivity to test object using data analysing method emphatically, realizes identification, the intensity of smell
Estimation and key component prediction.
Electronic nose theory and application research coordinate indexing result is as follows: (1) document.Only more than 60 before nineteen ninety, 2000
Add up more than 500 before year, add up to 6, more than 000 piece now, illustrates electronic nose research expansion extensively in recent years.(2) patent.
500 remainder world patents of invention and 100 remainder country patents of invention are open and authorization in nearly 5 years mostly, show olfactory analog
Intellectual property protection has been taken seriously.(3) technical standard.International standard database HIS there is no product related with olfactory analog
Technical standard.(4) it applies.Domestic most work carry out laboratory research with foreign countries' commercialization electronic nose.The above results are said
Bright, olfactory analog-electronic nose theory and application research urgently gos deep into.
ISI database query result shows that electronic nose method is applied to the text that environmental malodors gas process is detected with analyzed
It offers seldom, only more than 130 piece, less than the 2% of electronic nose document sum, and is mostly the offline inspection of Interior Space gas and water, rustic taste
With laboratory data processing;Not yet discovery odorant pollutant scene electronic nose on-line monitoring report, there is no mature stench electronics
Nose instrument commodity.
In the case where environmental protection administrative department, the Chinese government is leading, domestic some chemical industrial parks, refuse landfill, sewage treatment plant etc.
Discharge of pollutant sources unit uses the commercialization electronics of Airsense company, Germany and alpha MOS company, France by bid
Nose.This two product is by 4 metal-oxide semiconductor (MOS)s (Metal Oxide Semiconductor, MOS), 4 electrochemistry
(Electrochemical, EC), 1 photoion (Photoionization Detector, PID) gas sensor are array, are
Specifically for Chinese market exploitation, that there are monitoring standards in actual application is inconsistent, analysis model is not applicable, stability
With a series of problems, such as consistency is poor, equipment and operating cost are high.The stench monitoring system of domestic topology Zhi Xin company is with 1
PID and 8 EC gas sensor forms array, and focal point is placed on offset minimum binary (partial least squares, PLS)
Algorithm and the transmission of data cloud, attempt to make a decision compared with standard sample according to sample, do not account for foul gas
Complicated component and environment variability.
In order to which Electronic Nose Technology and instrument are used for foul gas on-line monitoring and analysis, we must solve following ask
Topic:
1, odor concentration and its key component concentration prediction problem based on big data and artificial intelligence
Human society is in big data and artificial intelligence epoch, healthy big data, financial big data, traffic big data, quotient
Sparetime university's data, gene big data etc. profoundly change people's lives and working method.In China, environment big data has been mentioned
Upper agenda, environmental protection administrative department, government energetically push in.
Due to foul odour complexity and environment variability, small data and conventional method of analysis are not enough to effectively establish estimation
With the mathematical model of prediction foul gas Multiple components.There is no stench electronic nose instrument to generate a large amount of odor pollution on-the-spot test
Gas sensor array response data, do not smell the personnel of distinguishing the laboratory of a large amount of stench samples smelt and distinguish data, without chromaticness
The conventional instruments such as spectrum attempt to lean on gas sensor array and simple mathematics merely to the offline inspection data of a large amount of stench samples
Model estimates that odor concentration with multiple pollutant ingredient is impossible.Moral, method electronic nose are exactly to do so, and are thus generated
Monitoring data effect it is extremely limited, it might even be possible to say it is incredible.
We should with gas sensor array response data, smell and distinguish data, chromaticness spectrum and the conventional instruments point such as spectrophotometric
Based on analysing data, foul gas big data is established, artificial intelligence theory and algorithm is furtherd investigate, is excavated from stench big data
The useful informations such as key component concentration out, to realize that electronic nose instrument refers to the control of above-mentioned 10+1 kind main odorant pollutant concentration
Target is predicted in real time.
2, stench electronic nose automation equipment and intelligent Problems
Odor polluting source is numerous, and foul gas constituent is numerous, environmental change multiterminal, and odorant pollutant form of export is many
It is more.We should abandon the distributing monitoring mode of " one point one nose ", the optimization of research gas sensor array with merge and multiple spot collection
Chinese style precision automatic sample handling system, invention and small, light-weight, the easy to operate novel stench electronic nose instrument of development size.Reason
Thinking situation is, a stench electronic nose instrument is able to achieve specific region (for example, area 4km2Within) multiple observation points while
On-line monitoring, can fixed point monitoring, also move point monitoring, certainly daily 24 hours as unit of the moon or even year are continuous
Monitoring;It is proposed that simple and effective machine learning model and algorithm realize 24 hours companies to aforementioned 10+1 kind odorant pollutant concentration
Monitoring data and analysis result are transferred to monitoring center and various in real time by continuous estimation and prediction, and utilize wireless WIFI technology
Terminal realizes that the odor pollution based on Internet remotely controls.
Summary of the invention
The present invention is in existing patent of invention " a kind of machine olfaction device and its olfactory analog test method " (referring to patent
Application number: 02111046.8), " a kind of machine olfaction odor distinguishing method based on modular combination neural network " (referring to special
Sharp application number: 03141537.7), " a kind of olfactory analog instrument and a variety of smell qualitative and quantitative analysis methods " (referring to patent Shen
Please number: 201010115026.2) and " a kind of multichannel integrates olfactory analog instrument and biological fermentation process on-line analysis "
On the basis of (referring to application number: 201310405315.X), a kind of foul gas multiple spot centralization electricity of big data driving is invented
Sub- nose instrument on-line analysis, to solve the long-term on-line monitoring and a variety of foul gas of the multiple monitoring points in odor pollution region
The on-line prediction problem of concentration Con trolling index.
To achieve the goals above, the foul gas multiple spot centralization electronic nose instrument of a kind of big data of the invention driving
On-line analysis, including stench electronic nose instrument, gas sampling probe II, external vacuum pump III, surrounding air purification device
IV, pure air V, gas pipeline, electronics Hygrothermograph VI, central control room VII and multiple fixation/mobile terminals are realized
The long-term on-line monitoring of 10, odor pollution region monitoring point and the On-line Estimation of a variety of odorant pollutant concentration Con trolling index values
With prediction.
Stench electronic nose instrument includes gas sensor array constant temperature operating room, multiple spot centralization foul gas automatic sampling
System, computer control and the big component part of data analysis system three.Concrete composition unit includes: (a) gas sensor array
Constant temperature operating room: gas sensor array I-1, thermal insulation layer I-2, resistance heating wire I-3, fan I-4 are located at stench electronic nose instrument
Device upper right side;(b) multiple spot centralization foul gas automatic sample handling system: the bi-bit bi-pass electromagnetism of control cleaning ambient air on-off
Valve I-5 controls 10 two-position two-way solenoid valve I-6-1~I-6-10 of 10 monitoring point foul gas on-off, shows external true
The pressure vacuum gauge I-7 of sky pump III working condition, control foul gas flow through the bi-bit bi-pass electricity of gas sensor array I-1
It is flow cycled to control annular working chamber internal gas locating for gas sensor array I-1 by magnet valve I-8, gas buffer room I-9
Two-position two-way solenoid valve I-10, throttle valve I-11, flowmeter I-12 control the two-position two-way solenoid valve I- of pure air on-off
13, built-in miniature vacuum pump I-14 are located at stench electronic nose instrument lower right;(c) computer control and data analysis system: meter
Calculate mainboard I-15, data collecting card, that is, A/D plate I-16, display I-17, driving and controlling circuits module I -18, multi-channel DC
Power supply I-19 is located on the left of stench electronic nose instrument.
Stench electronic nose instrument is T to the single monitoring point foul gas sampling period0=180-300s, implied value T0=
240s.Gas sensor array I-1 generates one 16 dimension response vector to the monitoring point one-shot measurement.Computer control and data
Analysis system refers to the foul smell olfactory concentration of the monitoring point, GB14554 according to this response vector, with machine learning cascade model
Fixed ammonia, hydrogen sulfide, carbon disulfide, trimethylamine, methyl mercaptan, methyl sulfide, dimethyl disulfide, styrene totally 8 kinds of compounds, GB/
The T18883 specified total 10+1 odorant pollutant concentration Con trolling index value of sulfur dioxide and total volatile organic compounds carries out
Analysis in real time and prediction, and by monitoring data and prediction result by wireless Internet teletransmission to central control room with
Specified fixation/mobile terminal.
Stench electronic nose instrument predicts future t+1, t+2 and t+3 moment foul smell olfactory concentration with machine learning cascade model
With 10 kinds of odorant pollutant concentration Con trolling index values.The machine learning cascade model first order-convolutional neural networks
(Convolutional neural network, CNN) layer is responsible for predicting t+1, t+2 and t+3 moment gas sensor array I-1
Response to a monitoring point foul gas is based on the current time t's and gas sensor array I-1 occurred in the recent period
Response time sequence.The machine learning cascade model second level-deep neural network (Deep neural network, DNN) layer
It further predicts t+1, t+2 and t+3 moment foul smell olfactory concentration and 10 kinds of odorant pollutant concentration Con trolling index values, is based on
The foul gas big data and the machine learning cascade model first order-convolutional neural networks layer predicted value of long-term accumulation.
Stench electronic nose instrument is to the long-term on-line monitoring of 10, odor pollution region monitoring point and a variety of odorant pollutants
The on-line prediction of concentration Con trolling index value, comprising the following steps:
(1) be switched on: instrument preheats 30min;Click " air purifier is opened " option of on-screen menu, surrounding air purification dress
It sets IV to start to purify room air locating for stench electronic nose instrument, be continued working for a long time until operator clicks that " air is net
Until the option of change device pass ".
Under the swabbing action of built-in minipump I-14, cleaning ambient air with the flow of 6,500mL/min successively
Two-position two-way solenoid valve I-5, gas sensor array I-1, two-position two-way solenoid valve I-10 are flowed through, outdoor is then discharged to;
The intracavitary temperature of annular working locating for gas sensor array I-1 reaches constant 55 ± 0.1 DEG C from room temperature.
Click " external vacuum pump is opened " option of on-screen menu;External vacuum pump III is with the extraction flow of 250-280L/min
Linear distance is reached 2.5km in 1min by internal diameter φ 10mm stainless steel pipes by the final vacuum of amount and 100-120mbar
Some monitoring point foul gas be drawn into stench electronic nose instrument, flow successively through corresponding two-position two-way solenoid valve, vacuum
Pressure gauge I-7 and gas surge chamber I-8, is then vented directly to outdoor;External vacuum pump III persistently aspirates foul gas, directly
Until operator clicks " external vacuum pump pass " option of on-screen menu.
Modify foul gas " single sampling period T of on-screen menu0" setting, implied value T0=240s.10 monitoring point stenches
The gas circulating sampling period is T=10T0。
(2) the foul gas circulating sampling period starts: clicking " starting to detect " button of on-screen menu, stench electronic nose instrument
Device successively carries out circulatory monitorings to 10 monitoring points, and computer control and data analysis system automatically generate 10 in specified folder
A data file, to store gas sensor array I-1 respectively to the response data of 10 monitoring point foul gas.
(3) the monitoring point k foul gas list sampling period, k=1,2 ..., 10.Here T is taken0=240s:
(3.1) gas sensor array tentatively restores: monocycle T00-155s, in the pumping of built-in minipump I-14
Under suction effect, cleaning ambient air is with the flow of 6,500mL/min followed by two-position two-way solenoid valve I-5, gas sensor
Annular working chamber locating for array I-1, two-position two-way solenoid valve I-10, are then discharged to outdoor.It is purified in 6,500mL/min
Under the action of surrounding air, the heat gathered in annular working chamber locating for gas sensor array I-1 is pulled away, and is adhered to gas
The foul gas molecule of dependent sensor sensitivity film surface and inner wall of the pipe is tentatively washed away, and gas sensor array I-1 is tentatively extensive
Normal condition is arrived again, lasts 155s.External vacuum pump III is persistently aspirated;10 two-position two-way solenoid valve I-6-1~I-6-10 are only
There is I-6-k conducting, remaining 9 disconnections.External vacuum pump III is persistently aspirated.
(3.2) pure air Accurate Calibration: in monocycle T0156-185s, two-position two-way solenoid valve I-13 conducting, two
Two-way electromagnetic valve I-5, I-8 and I-10 are disconnected for position, and two-position two-way solenoid valve I-6-1~I-6-10 keeps the state of step (3.1).
Under the swabbing action of built-in minipump I-14, pure air is with the flow of 1,000ml/min followed by bi-bit bi-pass
Electromagnetic valve I -13, gas pipeline, gas sensor array I-1, throttle valve I-11, flowmeter I-12, minipump I-14, so
After be discharged to outdoor.Pure air makes gas sensor array I-1 Exact recovery to normal condition;Last 30s;External vacuum
Pump III is persistently aspirated;
(3.3) it balances: in monocycle T0186-190s, two-position two-way solenoid valve I-5, I-8, I-10, I-13 are disconnected, and two
Position two-way electromagnetic valve I-6-1~I-6-10 keeps the state of step (3.1).Annular working locating for gas sensor array I-1
Intracavitary no gas flowing;From monocycle T0From the quarter that the 186s, that is, equilibrium state starts, computer control and data analysis system
The real-time response data of start recording gas sensor array I-1, and it is inner to be stored in specified temporary file " temp.txt ";It goes through
When 5s.External vacuum pump III is persistently aspirated.
(3.4) monitoring point k foul gas Head-space sampling: in monocycle T0190-220s, two-position two-way solenoid valve I-8 are led
Logical, 3 two-position two-way solenoid valves I-5, I-13 and I-10 are disconnected, and two-position two-way solenoid valve I-6-1~I-6-10 keeps step
(3.1) state.Under built-in minipump I-14 swabbing action, foul gas in the I-8 of gas buffer room with flow 1,
000ml/min flows successively through annular working chamber locating for gas sensor array I-1, throttle valve I-11, flowmeter I-12, built-in
Minipump I-14, is finally discharged to outdoor.The sensitive response that gas sensor array I-1 is generated continues to be recorded in interim
File " temp.txt " is inner, lasts 30s.External vacuum pump III is persistently aspirated.
(3.5) gas sensor array rinses: in monocycle T0221-240s, two-position two-way solenoid valve I-5 and I-10
Conducting, two-position two-way solenoid valve I-8 and I-13 are disconnected, under built-in minipump I-14 swabbing action, flow 6,500ml/
The cleaning ambient air of min is followed by two-position two-way solenoid valve I-5, gas sensor array I-1, two-position two-way solenoid valve
Then I-10 is discharged to outdoor.At the same time, if k < 10, two-position two-way solenoid valve I-6-k+1 conducting, 10 two two
Remaining 9 disconnections of three-way electromagnetic valve I-6-1~I-6-10, external vacuum pump transfer to aspirate the foul gas of monitoring point k+1;If k
=10, then k+1=1 is enabled, next foul gas circulating sampling period is transferred to, external vacuum pump transfers to aspirate monitoring point k=1's
Foul gas.Due to the effect of cleaning ambient air, the interior heat gathered of annular working chamber locating for gas sensor array I-1
It is pulled away, the foul gas molecule for being adhered to gas sensor sensitivity film surface and inner wall of the pipe is tentatively washed away, gas sensing
Device array I-1 is gradually restored to normal condition, lasts 20s.Wherein:
(a) in monocycle T0221-230s, gas sensor array response data continue to be recorded in temporary file
" temp.txt " is inner, lasts 10s.To the end 230s, computer control stops recording gas sensor battle array with data analysis system
Column response data.
(b) in monocycle T0231-240s, computer control successively carry out following three operations with data analysis system:
(b1) feature extraction: from the quarter of 231s, each gas is extracted from the temporary file " temp.txt " of duration 45s is inner
The maximum steady state response of dependent sensor and minimum steady-state response value, with the difference of maximum steady state response and minimum steady-state response value
As each gas sensor current time t to the response characteristic component x of monitoring point k foul gasi(t), i=1,2 ..., 16,
And it is recorded in corresponding data file.
(b2) gas sensor array response prediction: the machine learning cascade model convolutional neural networks of the first order -16*3
The gas sensor battle array occurred in the period according to [t-18, t], [t-19, t-1] and [t-20, t-2] before current time t
Column time series response vector realizes automatic measure on line, and predicts future T, 2T and 3T moment gas sensor array I-1 accordingly
Response.
(b3) foul gas concentration Con trolling index value is predicted: the machine learning cascade model second level -10+1 depth nerve
The response for the gas sensor array I-1 that network is predicted according to 16*3 convolutional neural networks of the cascade model first order, into
The 10+1 item odorant pollutant concentration Con trolling index value of one-step prediction monitoring point k, is shown by display, and will monitoring and
Prediction result is transmitted to central control room VII and multiple fixation/mobile terminals by Internet network.
(3.6) the monitoring point k foul gas list sampling period terminates: returning to step (3.1), monitoring point k+1 foul gas list
Sampling period starts;If k+1 > 10, the monitoring point k=1 for being transferred to next foul gas circulating sampling period starts.
(4) step (3.1)~(3.6) are repeated, stench electronic nose instrument realizes the circulation to 10 monitoring point foul gas
It monitors on-line, the prediction of identification and 10+1 odorant pollutant Con trolling index values.
Foul gas large data sets include: (1) gas sensor array I-1 to refuse landfill, sewage treatment plant including
A large amount of odorant pollutant scenes on-line checking of chemical industrial park, pharmaceutical factory, farm, neighbouring residential block including flavors and fragrances factory
Data.(2) gas sensor array I-1 is to the laboratory offline inspection data of a large amount of stench standard sample head space Volatile Gas,
In include GB/T14675 specify bata-phenethyl alcohol, isovaleric acid, methyl-cyclopentanone, peach aldehyde, Beta-methyl indoles this 5
The kind smelly liquid of standard;GB14554 specified ammonia, hydrogen sulfide, carbon disulfide, trimethylamine, methyl mercaptan, methyl sulfide, dimethyl disulfide, benzene
The standard stench of the sulfur dioxide that ethylene and GB/T18883 the are specified various concentration that totally 9 kinds of single component odorant pollutants are prepared
Sample further includes the blending constituent standard stench sample that a variety of single compounds of various concentration are prepared.(3) GB/T14675 and HJ
Dewar bottle as defined in 905 and foul smell bag in a large amount of odorant pollutant spot samplings, and transport back immediately smell it is immeasurable obtained from distinguishing room
Guiding principle odor concentration is smelt offline distinguishes data.(4) Tenax GC/TA adsorption tube odorant pollutant spot sampling as defined in GB/T18883,
The total volatile organic compounds data and spectrophotometer laboratory that gas chromatograph laboratory offline inspection obtains are examined offline
The sulfur dioxide data measured.(5) odorant pollutant spot sampling as defined in GB/T14676-14680,8 kinds of odor pollutants
Gas chromatograph, mass spectrograph and spectrophotometer laboratory offline inspection data.(6) odor polluting source adjacent domain resident complains
Data.
According to " dividing and rule " strategy, the machine learning cascade model first order singly exports single hidden layer convolutional Neural with 16*3 group
Network predicts the response of t+1, t+2 and t+3 moment each gas sensor one by one.To T0For=240s, it is equivalent to from current
Moment, t was counted, and predicted the response at following T, 2T and 3T moment.
With monocycle T0=240s, 3 single list hidden layer convolutional neural networks that export predict t+1, t+2 and t+3 moment respectively
For the response of gas sensor i:
(a) single to export single hidden layer convolutional neural networks CNNi1Predict the response of t+1 moment gas sensor i:
If convolutional neural networks CNNi118 time of day response times that study gas sensor i has occurred before t moment
Sequence, time delay length Δ t=9, then input number of nodes mi=9, take Hidden nodes hi=5, output node number ni=1.Convolutional Neural
Network C NNi1The pretreated gas sensor i response time sequence data collection X of on-line studyi1Are as follows:
Target output are as follows:
di1=(xi(t) xi(t-1) xi(t-2) xi(t-3) xi(t-4) xi(t-5) xi(t-6) xi(t-7) xi(t-
8) xi(t-9))T∈R10,
This mode is equivalent to convolutional neural networks CNNi1Learn nearest 1 occurred for 12 hours 18 gas sensor i
Response time sequence is tieed up, 10 9 dimension response time sequences are generated, is i.e. sample number is Ni1=10.Convolutional neural networks CNNi1It is hidden
Layer and output layer activation functions are Sigmoid correction functionLearnt using error back propagation algorithm, study because
Son is ηi=5/Ni1=0.2.Data set Xi1D is exported with targeti1It is proportional to transform to range [0,3].Convolutional neural networks
CNNi1In 10s after on-line study, one 9 dimension response time sequence according to the nearest period:
xi1=(xi(t-8) xi(t-7) xi(t-6) xi(t-5) xi(t-4) xi(t-3) xi(t-2) xi(t-1) xi
(t))T∈R9
Predict the response x of t+1 moment gas sensor ii(t+1);Work as T0When=240s, it is equivalent to prediction future 40min
The response of gas sensor i.
(b) single to export single hidden layer convolutional neural networks CNNi2With CNNi3Predict the sound of t+2 and t+3 moment gas sensor i
It answers:
Convolutional neural networks CNNi2And CNNi3Structure is still are as follows: mi=9, hi=5, ni=1;The pretreated number of on-line study
According to collection Xi2And Xi3It is respectively as follows:
With
That is Xi2And Xi3Equally there are 10 9 dimension response time sequences, sample number is Ni1=10;Convolutional neural networks CNNi2
With CNNi3The time series and CNN of foundation in the target output and prediction for learning the stagei1It is identical.Work as T0When=240s, quite
In the 12 hour responses that have occurred of the study gas sensor i before 40min and 80min, t+2 and t+3 moment air-sensitive is predicted
The response x of sensor ii(t+2) and xi(t+3), it is respectively equivalent to the sound of prediction gas sensor i future 80min and 120min
It answers.
According to " dividing and rule " strategy, ammonia, hydrogen sulfide, carbon disulfide, trimethylamine, methyl mercaptan, methyl sulfide, methyl disulfide
Ether, styrene, sulfur dioxide, total volatile organic compounds and the total 10+1 odorant pollutant concentration control of foul smell olfactory concentration
Index value entirety forecasting problem is broken down into 11 single concentration values forecasting problem one by one, and the machine learning cascade model second level is with 10
+ 1 three hidden layer deep neural network module of single output predicts this 10+1 odorant pollutant Con trolling index value respectively.Single output is deep
Degree neural metwork training integrate for the gas sensor array (I-1) of stench electronic nose instrument to the smelly liquid/gas sample of standard and largely
The foul gas big data that live on-line checking obtains is polluted, target output is smelt for foul smell distinguishes that value and chromaticness spectrum are normal with spectrophotometric
It advises instrument off-line measurement value and resident complains data.
Three hidden layer deep neural network DNN of single single outputjUsing layer-by-layer off-line learning mode from bottom to top;First He
Using single hidden layer equity neural network structure, i.e., hidden layer-output layer power of single hidden layer equity neural network when second hidden layer learns
Value is directly equal to its input layer-hidden layer weight, and target output is directly equal to its input, inputs component and output component according to special
Component size is proportional transforms to range [0,3] for sign.The hidden layer activation functions of single hidden layer equity neural network are Sigmoid amendment
FunctionLearnt using error back propagation algorithm, Studying factors ηj=1/Nj, abandoned after study hidden
Layer-output layer.
Assuming that t+1 moment concentration value yj(t+1) it is predicted, j-th of single output deep neural network DNNjIt is based on
Predicated response { x of 16 convolutional neural networks to t+1 moment gas sensor array (I-1)1(t+1),x2(t+1),…,x16
(t+1) } y, is predictedj(t+2) and yj(t+3) 16 convolutional neural networks are based on respectively to ring the prediction at t+2 and t+3 moment
Answer (x1(t+2),x2(t+2),…,x16(t+2))TWith (x1(t+3),x2(t+3),…,x16(t+3))T。
If actually entering is the current response vector (x of gas sensor array1(t),x2(t),…,x16(t))T, when necessary may be used
Along with t moment temperature and humidity value, then deep neural network DNNjReality output be to foul gas ingredient j existing concentration value yj
(t) estimation.
Detailed description of the invention
Fig. 1 is the present invention-big data driving foul gas multiple spot centralization electronic nose instrument on-line analysis-evil
Pass between smelly electronic nose instrument development, machine learning cascade model and algorithm and odorant pollutant on-line checking and prediction three
It is block diagram.
Fig. 2 is the present invention-big data driving foul gas multiple spot centralization electronic nose instrument on-line analysis-evil
Smelly electronic nose instrument and the monitoring of odor pollution region multiple spot centralization and analysis system operation principle schematic diagram (Head-space sampling shape
State).
Fig. 3 is the present invention-big data driving foul gas multiple spot centralization electronic nose instrument on-line analysis-volume
Product neural network CNNi1Predict that t+1 moment (such as future 40min) gas sensor i responds xi(t+1) schematic diagram.
Fig. 4 is that the present invention-big data driving foul gas multiple spot centralization electronic nose instrument on-line analysis-is true
Determine deep neural network DNNjThe reciprocity neural network structure and learning process schematic diagram of kth layer.
Fig. 5 is the present invention-big data driving foul gas multiple spot centralization electronic nose instrument on-line analysis-machine
Device learns cascade model and predicts t+1 moment (such as future 40min) a variety of odorant pollutant concentration Con trolling index value schematic diagrames.
Specific embodiment
The present invention is described in further detail below in conjunction with the accompanying drawings.
Fig. 1 is a kind of present invention-big data driving foul gas multiple spot centralization electronic nose instrument on-line analysis side
Between method-stench electronic nose instrument development, machine learning model and algorithm and odorant pollutant on-line checking and prediction three
Relationship block diagram.
The present invention is first chemically, physical angle analyzes the characteristics of odorant pollutant and gas sensor.Stench
Gas composition ingredient is numerous and complicated, often containing it is tens of so that it is hundreds of cause smelly ingredient, existing organic principle also have it is inorganic at
Point;Some odor pollutants are big to odor concentration contribution but actual concentration may be very low, gas sensor response therefore very little;Some
Odor pollutant contributes very little to odor concentration but actual concentration may be very high, and gas sensor is therefore very big;Vice versa.It is comprehensive
Consider sensitivity, selectivity, response speed, stability, commercialization, miniaturization, service life, the factors such as cost, present invention selection by
MOS type, EC type and PID type gas sensor form small-sized gas sensitive sensor array module.To avoid the wind outside monitoring region room
Expose to the sun and rain, the present invention proposes that critical component is located at indoor foul gas multiple spot centralization monitoring mode and develops stench accordingly
Electronic nose instrument.In view of the factors such as odorant pollutant complicated component and monitoring field environmental change multiterminal, therefore the present invention mentions
Establish stench big data out, and propose new machine learning cascade model come realize to the on-line monitorings of a variety of odorant pollutants with
Prediction.
According to Fig. 1, foul gas big data includes: the gas sensor array I-1 of (1) stench electronic nose instrument to a large amount of
The laboratory offline inspection data of stench standard sample head space Volatile Gas, including bata-phenethyl alcohol, isovaleric acid, methyl ring penta
The smelly liquid of 5 kinds of standards such as ketone, peach aldehyde, Beta-methyl indoles and C3H9N、C8H8、H2S、CH4S、C2H6S、C2H6S2、NH3、
CS2、SO2It further include a variety of by various concentration Deng the various concentration single component standard stench sample that 9 kinds of malodorous compounds are prepared
The blending constituent standard stench sample that single compound is prepared;(2) gas sensor array I-1 is to a large amount of odorant pollutant scenes
On-line checking data;(3) the odor concentration laboratory of a large amount of odorant pollutants is smelt offline distinguishes data;(4) a large amount of odor pollutions
The TVOC and the inspection of above-mentioned 9 kinds of odor pollutants that gas chromatograph, mass spectrograph and the spectrophotometer laboratory offline inspection of object obtain
Measured data;(5) odor polluting source adjacent domain resident complains data.
Fig. 2 is that stench electronic nose instrument and the monitoring of odor pollution region multiple spot centralization are illustrated with analysis system working principle
Figure.The monitoring of odor pollution region multiple spot centralization and analysis system include stench electronic nose instrument, 10 outdoor-monitoring point II-1
~II-1, external vacuum pump III, surrounding air purification device IV, pure air V, electronics Hygrothermograph VI, central control room
VII and its multiple fixation/mobile terminals realize that the long-term on-line monitoring of 10, odor pollution region monitoring point and foul gas are more
The on-line prediction of kind concentration Con trolling index value.The position of gas circuit and solenoid valve at this time is first monitoring point II-1 foul gas
It is sucked into stench electronic nose instrument, therefore gas sensor array I-1 generates the working condition of sensitive response.
The component units of stench electronic nose instrument include:
(a) gas sensor array constant temperature operating room: gas sensor array I-1 and its annular working chamber, thermal insulation layer I-
2, resistance heating wire I-3, fan I-4 are located at stench electronic nose instrument upper right side.
(b) multiple spot centralization foul gas automatic sample handling system: the bi-bit bi-pass electromagnetism of control cleaning ambient air on-off
Valve I-5 controls 10 two-position two-way solenoid valve I-6-1~I-6-10 of 10 monitoring point foul gas on-off, shows external true
The pressure vacuum gauge I-7 of sky pump III working condition, control foul gas flow into gas sensor array I-1 and its annular working
Intracavitary two-position two-way solenoid valve I-8, gas buffer room I-9 control the intracavitary stench of annular working of gas sensor array I-1
Gas and pure air 6,500ml/min and 1,000ml/min flow cycled two-position two-way solenoid valve I-10, throttle valve I-
11, flowmeter I-12 control two-position two-way solenoid valve I-13, the built-in miniature vacuum pump I-14 of pure air on-off, are located at and dislike
Smelly electronic nose instrument lower right.
(c) computer control and data analysis system: computer motherboard I-15, data collecting card, that is, A/D plate I-16, display
Device I-17, driving and controlling circuits module I -18, multi-channel dc power supply I-19 are located on the left of stench electronic nose instrument.
In single sampling period T0In the response data of duration 45s, the stable state maximum value of single gas sensor i response curve
Uimax(t) and minimum value Uimin(t) difference is extracted as characteristic component xi(t)=Uimax(t)-Uimin(t), gas sensor battle array
Therefore column generate response vector x (t)=(x of one 16 dimension1(t),…,xi(t),…,x16(t))T∈R16.In data record knot
In 10s after beam, i.e. 10s after surrounding air rinse stage, the machine learning grade gang mould of computer control and data analysis system
Type predicts 10+1 odorant pollutant concentration Con trolling index values according to response vector x (t).
According to " dividing and rule " strategy, the machine learning cascade model first order-convolutional neural networks (Convolutional
Neural network, CNN) layer single export single hidden layer convolutional neural networks and predicts each gas sensor one by one using multiple
Response.Fig. 3 is (such as future at prediction t+1 moment
Table 1 (a), convolutional neural networks CNNi1Time series training set Xi1
Table 1 (b), convolutional neural networks CNNi1Predict that t+1 moment gas sensor array responds xi(t+1) time series
Response sample xi(t)
40min) gas sensor i response xi(t+1) convolutional neural networks CNNi1Structural schematic diagram.Table 1 (a) is
CNNi1Time series training set Xi1∈R10×9, share 10 samples, dimension 9.Training set Xi1Time series span is [t-
18, t-1], in T0=240s and T=10T0In the case where, it is equivalent to CNNi1Learn gas sensor i before 12 hours currently to
The response that current time has occurred.According to table 1 (a), CNNi1A learning sample be equivalent to gas sensor i length Δ t=9
A time response series.Table 2 (b) provides CNNi1The time series response sample x used when prediction1=(xi(t-8),…,
xi(t))T∈R9。
Convolutional neural networks CNNi1The response time at 18 moment that study gas sensor i has occurred before t moment
Sequence, time delay length Δ t=9, then input number of nodes mi=9, take Hidden nodes hi=5, output node number ni=1;Convolutional Neural
Network C NNi1The pretreated response time sequence data collection X of on-line study gas sensor ii1, as shown in table 1 (a).CNNi1
Hidden node and output node activation functions are sigmoid correction functionIt is carried out using error back propagation algorithm
Study, Studying factors ηi1=5/Ni1=0.5, maximum number of iterations 10,000.Table 2 (a) and 2 (b) input and output component it is equal
It is proportional to transform to range [0,3].
Convolutional neural networks CNNi1Online learn is completed in the 10s time after gas sensor array surrounding air rinse stage
It practises, and the time series response sample x provided according to table 2 (b)i(t)=(xi(t-8),…,xi(t))TTo predict t+1 moment gas
The response x of dependent sensor ii(t+1)。
Present invention convolutional neural networks CNNi2And CNNi3Prediction gas sensor i will be in (such as future at t+2 moment respectively
80min) and the t+3 moment (such as future 120min) response xi(t+2) and xi(t+3)。CNNi2And CNNi3Structure and
Practise parameter and NNi1It is identical.Table 2 and table 3 give the time series training set X of this 2 convolutional neural networksi2∈R10×9And Xi3
∈R10×9.This 2 convolutional neural networks still use such as 1 (b)
Table 2, convolutional neural networks CNNi2Time series training set Xi2
Table 3, convolutional neural networks CNNi3Time series training set Xi3
Shown in CNNi1Identical time series response sample xi(t) t+2 and t+3 moment gas sensor i is predicted
Response xi(t+2)、xi(t+3).With Xi1Time span be [t-18, t-1] compare, Xi2And Xi3Time span be respectively
[t-19, t-2] and [t-20, t-3], it is remote away from t, so, CNNi2And CNNi3Predicted value confidence level compared with CNNi1It is low.
CNNi1、CNNi2And CNNi3Online learn is completed in the rear 10s of gas sensor array surrounding air rinse stage
It practises and predicts.Therefore, when carrying out t+1, t+2 and t+3 time of day response one by one to all 16 response curves of gas sensor array
When prediction, present invention employs 3*16 convolutional neural networks;If only predicting t+1 time of day response, 16 single output convolution are only needed
Neural network.
The multiple concentration value entirety forecasting problems of foul gas are decomposed into multiple lists according to " dividing and rule " strategy by the present invention
One concentration value forecasting problem one by one, with the multiple single output deep neural network (Deep in the machine learning cascade model second level-
Neural network, DNN) predict multiple single concentration values one by one, answering for machine learning model and algorithm is effectively reduced
Miscellaneous degree.Single output DNN number is equal to the foul gas concentration Con trolling index number to be predicted, corresponds.For example, to predict nothing
Dimension odor concentration OU value, NH3、H2S、CS2、C3H9N、CH4S、C2H6S、C2H6S2、C8H8、SO2Deng 9 kinds of odorant pollutant concentration and
TVOC concentration then needs 10+1 single output DNNs.What one single output DNN learnt is foul gas big data, and input value is
Gas sensor array detection data and stench electronic nose instrument scene data of the Temperature and Humidity module, target output distinguish value, chromaticness spectrum to smell
Equal conventional instruments off-line measurement value and resident complain data.Some samples only have gas sensor array in foul gas big data
It responds and distinguishes that the off-line measurements values such as value, chromaticness spectrum and resident complain data without smelling, do not participate in study.
One single output DNNjThere are 3 hidden layers, hidden layer and output layer to correct activation functions using SigmoidFirst and second hidden layers are characterized transformation (coding) layer, from bottom to top layer-by-layer off-line learning mode, knot
Structure and weighting parameter are determined with single hidden layer equity neural network.Fig. 4 is to determine DNNj+ 1 hidden layer weight of kth layer-kth and threshold
The reciprocity neural network learning process schematic of value.Fig. 4 (a) shows the output node number and input of a reciprocity neural network
Number of nodes is equal, is linear activation functions, and hidden layer-output layer weight and threshold value are directly equal to its input layer-hidden layer,
Target output is directly equal to it and actually enters.It is single to export DNN after Fig. 4 (b) shows the equity neural network learningj?
K+1 layers of Hidden nodes are equal to the Hidden nodes of the equity neural network, and+1 hidden layer weight of kth layer-kth is right equal to this with threshold value
Equal neural network input layers-hidden layer.If DNNjSample number be Nj, then reciprocity neural network learning factor η=2/Nj, maximum
Iterative steps τmax=10 000.Characteristic component and target output proportional transform to range [0,3.0].Single output DNNj's
Third hidden layer is Nonlinear Mapping layer, and j-th of concentration Con trolling index value of foul gas is fitted together with single output unit j.
Fig. 5 is that machine learning cascade model predicts that t+1 moment (such as future 40min) a variety of odorant pollutant concentration are shown
It is intended to.According to Fig. 5, the machine learning cascade model first order singly exports single hidden layer convolutional neural networks with 16*3 group and predicts t+ one by one
1, the response of t+2 and t+3 moment each gas sensor.10+1 three hidden layers of single output in the machine learning cascade model second level
Deep neural network predicts this 10+1 odorant pollutant Con trolling index value respectively.
Assuming that DNNjFoul gas concentration value y a kind of to the t+1 momentj(t+1) it is predicted, is then based on 16 CNNi1
(i=1,2 ..., 16) to gas sensor array t+1 moment predicated response (x1(t+1),x2(t+1),…,x16(t+1))TWith work as
Preceding moment temperature and humidity value;DNNjPredict yj(t+2) 16 CNN are based oni2To t+2 moment predicated response (x1(t+2),x2(t+
2),…,x16(t+2)) T and current time temperature and humidity value, etc..
If actually entering is the current response vector (x of gas sensor array1(t),x2(t),…,x16(t))T, when necessary may be used
Along with t moment temperature and humidity value, then deep neural network DNNjReality output be to foul gas ingredient j existing concentration value yj
(t) estimation.
Claims (4)
1. a kind of foul gas multiple spot centralization electronic nose instrument on-line analysis of big data driving, characterized in that including
Stench electronic nose instrument, gas sampling probe (II), external vacuum pump (III), surrounding air purification device (IV), pure air
(V), gas pipeline, electronics Hygrothermograph (VI), central control room (VII) and multiple fixation/mobile terminals realize that stench is dirty
Contaminate the long-term on-line monitoring of 10, region monitoring point and the On-line Estimation and prediction of a variety of odorant pollutant concentration Con trolling index values;
The stench electronic nose instrument include gas sensor array constant temperature operating room, multiple spot centralization foul gas automatically into
Sample system, computer control and the big component part of data analysis system three;Concrete composition unit includes: (a) gas sensor battle array
Column constant temperature operating room: gas sensor array (I-1), thermal insulation layer (I-2), resistance heating wire (I-3), fan (I-4) are located at and dislike
Smelly electronic nose instrument upper right side;(b) multiple spot centralization foul gas automatic sample handling system: the of control cleaning ambient air on-off
11 two-position two-way solenoid valves (I-5) control the first~the tenth of 10 monitoring point foul gas on-off totally 10 bi-bit bi-pass
Solenoid valve (I-6-1)~(I-6-10) shows the pressure vacuum gauge (I-7) of external vacuum pump (III) working condition, controls stench
Gas flows through the 12nd two-position two-way solenoid valve (I-8) of gas sensor array (I-1), and gas buffer room (I-9) controls gas
The flow cycled 13rd two-position two-way solenoid valve (I- of annular working chamber internal gas locating for dependent sensor array (I-1)
10), throttle valve (I-11), flowmeter (I-12) control the 14th two-position two-way solenoid valve (I-13) of pure air on-off, interior
Minipump (I-14) is set, stench electronic nose instrument lower right is located at;(c) it computer control and data analysis system: calculates
Mainboard (I-15), data collecting card, that is, A/D plate (I-16), display (I-17), driving and controlling circuits module (I-18) are more
Road DC power supply (I-19) is located on the left of stench electronic nose instrument;
The stench electronic nose instrument is T to the single monitoring point foul gas sampling period0=180-300s, implied value T0=
240s, gas sensor array (I-1) generate one 16 dimension response vector to the monitoring point one-shot measurement;Computer control and number
According to analysis system according to this response vector, with machine learning cascade model to the foul smell olfactory concentration of the monitoring point, GB14554
Specified ammonia, hydrogen sulfide, carbon disulfide, trimethylamine, methyl mercaptan, methyl sulfide, dimethyl disulfide, styrene totally 8 kinds of compounds,
GB/T18883 specified sulfur dioxide and the total 10+1 odorant pollutant concentration Con trolling index value of total volatile organic compounds into
Row analysis in real time and prediction, and monitoring data and prediction result are passed through into wireless Internet teletransmission to central control room
With specified fixation/mobile terminal;
Stench electronic nose instrument predicts future t+1, t+2 and t+3 moment foul smell olfactory concentration and 10 with machine learning cascade model
Kind odorant pollutant concentration Con trolling index value;The machine learning cascade model first order-convolutional neural networks (Convolutional
Neural network, CNN) layer be responsible for predict t+1, t+2 and t+3 moment gas sensor array (I-1) to a monitoring point
The response of foul gas is based on the response time sequence of current time t and the gas sensor array (I-1) occurred in the recent period
Column;The machine learning cascade model second level-deep neural network (Deep neural network, DNN) layer further predicts t
+ 1, t+2 and t+3 moment foul smell olfactory concentration and 10 kinds of odorant pollutant concentration Con trolling index values, are based on long-term accumulation
Foul gas big data and the machine learning cascade model first order-convolutional neural networks layer predicted value;
Stench electronic nose instrument is to the long-term on-line monitoring of 10, odor pollution region monitoring point and a variety of odorant pollutant concentration
The on-line prediction of Con trolling index value, comprising the following steps:
(1) be switched on: instrument preheats 30min;Click " surrounding air purification device is opened " option of on-screen menu, surrounding air purification
Device (IV) starts to purify room air locating for stench electronic nose instrument, is continued working for a long time until operator clicks " ring
Until the option of border air cleaning unit pass ";
Under the swabbing action of built-in minipump (I-14), cleaning ambient air is successively flowed with the flow of 6,500mL/min
Through the 11st two-position two-way solenoid valve (I-5), gas sensor array (I-1), the 13rd two-position two-way solenoid valve (I-10), so
After be discharged to outdoor;The intracavitary temperature of annular working locating for gas sensor array (I-1) reaches constant 55 from room temperature ±
0.1℃;
Click " external vacuum pump is opened " option of on-screen menu;External vacuum pump (III) is with the extraction flow of 250-280L/min
With the final vacuum of 100-120mbar, linear distance is reached 2.5km's in 1min by internal diameter φ 10mm stainless steel pipes
Some monitoring point foul gas is drawn into stench electronic nose instrument, flows successively through corresponding two-position two-way solenoid valve, vacuum pressure
Power table (I-7) and gas surge chamber (I-9), are then vented directly to outdoor;External vacuum pump (III) persistently aspirates effluvium
Body, until operator clicks " external vacuum pump pass " option of on-screen menu;
Modify foul gas " single sampling period T of on-screen menu0" setting, implied value T0=240s;10 monitoring point foul gas
The circulating sampling period is T=10T0;
(2) the foul gas circulating sampling period starts: click " starting to detect " button of on-screen menu, stench electronic nose instrument according to
Secondary to carry out circulatory monitorings to 10 monitoring points, computer control and data analysis system automatically generate 10 numbers in specified folder
According to file, to store gas sensor array (I-1) respectively to the response data of 10 monitoring point foul gas;
(3) the monitoring point k foul gas list sampling period, k=1,2 ..., 10;Here T is taken0=240s:
(3.1) gas sensor array tentatively restores: monocycle T00-155s, in the suction of built-in minipump (I-14)
Under effect, cleaning ambient air is with the flow of 6,500mL/min followed by the 11st two-position two-way solenoid valve (I-5), air-sensitive
Annular working chamber locating for sensor array (I-1), the 13rd two-position two-way solenoid valve (I-10), are then discharged to outdoor;
Under the action of 6,500mL/min cleaning ambient air, gathered in annular working chamber locating for gas sensor array (I-1)
Heat is pulled away, and the foul gas molecule for being adhered to gas sensor sensitivity film surface and inner wall of the pipe is tentatively washed away, air-sensitive
Sensor array (I-1) is tentatively restored to normal condition, lasts 155s;External vacuum pump (III) is persistently aspirated;
Only (I-6-k) conducting in first~the tenth this 10 two-position two-way solenoid valve (I-6-1)~(I-6-10), remaining 9
It disconnects;External vacuum pump (III) is persistently aspirated;
(3.2) pure air Accurate Calibration: in monocycle T0156-185s, the conducting of the 14th two-position two-way solenoid valve (I-13),
11st two-position two-way solenoid valve (I-5), the 12nd two-position two-way solenoid valve (I-8) and the 13rd two-position two-way solenoid valve (I-
10) it disconnects, the first~the tenth two-position two-way solenoid valve (I-6-1)~(I-6-10) keeps the state of step (3.1);Built-in micro-
Under the swabbing action of type vacuum pump (I-14), pure air is with the flow of 1,000ml/min followed by the 14th bi-bit bi-pass
Solenoid valve (I-13), gas pipeline, gas sensor array (I-1), throttle valve (I-11), flowmeter (I-12), micro vacuum
It pumps (I-14), is then discharged to outdoor;Pure air makes gas sensor array (I-1) Exact recovery to normal condition;It goes through
When 30s;External vacuum pump (III) is persistently aspirated;
(3.3) it balances: in monocycle T0186-190s, the 11st two-position two-way solenoid valve (I-5), the 12nd bi-bit bi-pass electricity
Magnet valve (I-8), the 13rd two-position two-way solenoid valve (I-10), the 14th two-position two-way solenoid valve (I-13) disconnect, and the first~the
Ten two-position two-way solenoid valves (I-6-1)~(I-6-10) keeps the state of step (3.1);Gas sensor array (I-1) is locating
The intracavitary no gas flowing of annular working;From monocycle T0From the quarter that the 186s, that is, equilibrium state starts, computer control and number
According to the real-time response data of analysis system start recording gas sensor array (I-1), and it is stored in specified temporary file
" temp.txt " is inner;Last 5s;External vacuum pump (III) is persistently aspirated;
(3.4) monitoring point k foul gas Head-space sampling: in monocycle T0190-220s, the 12nd two-position two-way solenoid valve (I-
8) it is connected, the 11st two-position two-way solenoid valve (I-5), the 14th two-position two-way solenoid valve (I-13) and the 13rd bi-bit bi-pass electricity
This 3 disconnections of magnet valve (I-10), the first~the tenth two-position two-way solenoid valve (I-6-1)~(I-6-10) keep step (3.1)
State;Under built-in minipump (I-14) swabbing action, foul gas in gas buffer room (I-9) with flow 1,
000ml/min flows successively through annular working chamber locating for gas sensor array (I-1), throttle valve (I-11), flowmeter (I-
12), built-in miniature vacuum pump (I-14), is finally discharged to outdoor;The sensitive response that gas sensor array (I-1) generates continues
It is inner to be recorded in temporary file " temp.txt ", lasts 30s;External vacuum pump (III) is persistently aspirated;
(3.5) gas sensor array rinses: in monocycle T0221-240s, the 11st two-position two-way solenoid valve (I-5) and
13 two-position two-way solenoid valves (I-10) conducting, the 12nd two-position two-way solenoid valve (I-8) and the 14th two-position two-way solenoid valve
(I-13) disconnect, under built-in minipump (I-14) swabbing action, flow 6, the cleaning ambient air of 500ml/min with according to
It is secondary to flow through the 11st two-position two-way solenoid valve (I-5), gas sensor array (I-1), the 13rd two-position two-way solenoid valve (I-
10), then it is discharged to outdoor;At the same time, if k < 10 ,+1 two-position two-way solenoid valve of kth (I-6-k+1) conducting, the
One~the tenth this 10 two-position two-way solenoid valve (I-6-1)~(I-6-10) remaining 9 are disconnected, and external vacuum pump transfers to aspirate
The foul gas of monitoring point k+1;If k=10 enables k+1=1, it is transferred to next foul gas circulating sampling period, it is external true
The foul gas of sky pump then suction monitoring point k=1;Due to the effect of cleaning ambient air, gas sensor array (I-1) institute
The heat gathered in the annular working chamber at place is pulled away, and is adhered to the effluvium of gas sensor sensitivity film surface and inner wall of the pipe
Body molecule is tentatively washed away, and gas sensor array (I-1) is gradually restored to normal condition;Last 20s;Wherein:
(a) in monocycle T0221-230s, gas sensor array response data continue to be recorded in temporary file " temp.txt "
In, last 10s;To the end 230s, computer control stops recording gas sensor array response data with data analysis system;
(b) in monocycle T0231-240s, computer control successively carry out following three operations with data analysis system:
(b1) feature extraction: from the quarter of 231s, each air-sensitive biography is extracted from the temporary file " temp.txt " of duration 45s is inner
The maximum steady state response of sensor and minimum steady-state response value, using the difference of maximum steady state response and minimum steady-state response value as
Response characteristic component x of each gas sensor current time t to monitoring point k foul gasi(t), i=1,2 ..., 16, and remember
Record is in corresponding data file;
(b2) gas sensor array response prediction: the machine learning cascade model convolutional neural networks foundation of the first order -16*3
When the gas sensor array that [t-18, t], [t-19, t-1] and [t-20, t-2] has occurred in the period before current time t
Between sequence response vector, realize automatic measure on line, and predict future T, 2T and 3T moment gas sensor array (I-1) accordingly
Response;
(b3) foul gas concentration Con trolling index value is predicted: the machine learning cascade model second level -10+1 deep neural network
The response for the gas sensor array (I-1) that the 16*3 convolutional neural networks according to the cascade model first order are predicted, into one
The 10+1 item odorant pollutant concentration Con trolling index value for walking prediction monitoring point k, is shown by display, and by monitoring and in advance
It surveys result and central control room (VII) and multiple fixation/mobile terminals is transmitted to by Internet network;
(3.6) the monitoring point k foul gas list sampling period terminates: returning to step (3.1), the sampling of monitoring point k+1 foul gas list
Period starts;If k+1 > 10, the monitoring point k=1 for being transferred to next foul gas circulating sampling period starts;
(4) step (3.1)~(3.6) are repeated, stench electronic nose instrument is realized online to the circulation of 10 monitoring point foul gas
It monitors, the prediction of identification and 10+1 odorant pollutant Con trolling index values.
2. the foul gas multiple spot centralization electronic nose instrument on-line analysis side of big data driving according to claim 1
Method, characterized in that foul gas large data sets include: (1) gas sensor array (I-1) to refuse landfill, sewage treatment
Factory, the chemical industrial park including flavors and fragrances factory, pharmaceutical factory, farm, a large amount of odorant pollutants of neighbouring residential block are live
On-line checking data;(2) gas sensor array (I-1) examines the laboratory of a large amount of stench standard sample head space Volatile Gas offline
Measured data, the bata-phenethyl alcohol specified including GB/T14675, isovaleric acid, methyl-cyclopentanone, peach aldehyde, β-first
This 5 kinds of smelly liquid of standard of base indoles;GB14554 specified ammonia, hydrogen sulfide, carbon disulfide, trimethylamine, methyl mercaptan, methyl sulfide, diformazan
Disulfide, the styrene and GB/T18883 specified sulfur dioxide various concentration that totally 9 kinds of single component odorant pollutants are prepared
Standard stench sample further includes the blending constituent standard stench sample that a variety of single compounds of various concentration are prepared;(3)GB/
Dewar bottle as defined in T14675 and HJ 905 and foul smell bag in a large amount of odorant pollutant spot samplings, and transport back immediately smell distinguish room and
Obtained dimensionless odor concentration is smelt offline distinguishes data;(4) Tenax GC/TA adsorption tube odor pollution as defined in GB/T18883
Object spot sampling, the total volatile organic compounds data and spectrophotometer that gas chromatograph laboratory offline inspection obtains are real
Test the sulfur dioxide data that room offline inspection obtains;(5) odorant pollutant spot sampling as defined in GB/T14676-14680,8 kinds
Gas chromatograph, mass spectrograph and the spectrophotometer laboratory offline inspection data of odor pollutant;(6) odor polluting source proximity
Domain resident complains data.
3. the foul gas multiple spot centralization electronic nose instrument on-line analysis side of big data driving according to claim 1
Method, characterized in that according to " dividing and rule " strategy, the machine learning cascade model first order singly exports single hidden layer convolution with 16*3 group
Neural network predicts the response of t+1, t+2 and t+3 moment each gas sensor one by one;To T0For=240s, be equivalent to from
Current time, t was counted, and predicted the response at following T, 2T and 3T moment;
With monocycle T0=240s, 3 single list hidden layer convolutional neural networks that export predict that t+1, t+2 and t+3 moment air-sensitive pass respectively
For the response of sensor i:
A) single to export single hidden layer convolutional neural networks CNNi1Predict the response of t+1 moment gas sensor i:
If convolutional neural networks CNNi118 time of day response time serieses that study gas sensor i has occurred before t moment,
Time delay length Δ t=9, then input number of nodes mi=9, take Hidden nodes hi=5, output node number ni=1;Convolutional neural networks
CNNi1The pretreated gas sensor i response time sequence data collection X of on-line studyi1Are as follows:
Target output are as follows:
di1=(xi(t) xi(t-1) xi(t-2) xi(t-3) xi(t-4) xi(t-5) xi(t-6) xi(t-7) xi(t-8)
xi(t-9))T∈R10, this mode is equivalent to convolutional neural networks CNNi1What study had occurred for gas sensor i nearest 12 hours
1 18 dimension response time sequence generates 10 9 dimension response time sequences, i.e. sample number is Ni1=10;Convolutional neural networks
CNNi1Hidden layer and output layer activation functions be Sigmoid correction functionUsing error back propagation algorithm
It practises, Studying factors ηi=5/Ni1=0.2;Data set Xi1D is exported with targeti1It is proportional to transform to range [0,3];Convolution
Neural network CNNi1In 10s after on-line study, one 9 dimension response time sequence according to the nearest period:
xi1=(xi(t-8) xi(t-7) xi(t-6) xi(t-5) xi(t-4) xi(t-3) xi(t-2) xi(t-1) xi(t))T
∈R9
Predict the response x of t+1 moment gas sensor ii(t+1);Work as T0When=240s, it is equivalent to prediction future 40min air-sensitive
The response of sensor i;
B) single to export single hidden layer convolutional neural networks CNNi2With CNNi3Predict the response of t+2 and t+3 moment gas sensor i:
Convolutional neural networks CNNi2And CNNi3Structure is still are as follows: mi=9, hi=5, ni=1;The pretreated data set of on-line study
Xi2And Xi3It is respectively as follows:
With
That is Xi2And Xi3Equally there are 10 9 dimension response time sequences, sample number is Ni1=10;Convolutional neural networks CNNi2With
CNNi3The time series and CNN of foundation in the target output and prediction for learning the stagei1It is identical;Work as T0When=240s, it is equivalent to
Learn the 12 hour responses that have occurred of the gas sensor i before 40min and 80min, prediction t+2 and t+3 moment air-sensitive passes
The response x of sensor ii(t+2) and xi(t+3), it is respectively equivalent to the sound of prediction gas sensor i future 80min and 120min
It answers.
4. the foul gas multiple spot centralization electronic nose instrument on-line analysis side of big data driving according to claim 1
Method, characterized in that according to " dividing and rule " strategy, ammonia, hydrogen sulfide, carbon disulfide, trimethylamine, methyl mercaptan, methyl sulfide, diformazan two
Thioether, styrene, sulfur dioxide, total volatile organic compounds and the total 10+1 odorant pollutant concentration control of foul smell olfactory concentration
Index value entirety forecasting problem processed is broken down into 11 single concentration values, and forecasting problem, the machine learning cascade model second level are used one by one
10+1 three hidden layer deep neural network modules of single output predict this 10+1 odorant pollutant Con trolling index value respectively;Single output
Deep neural network training set is the gas sensor array (I-1) of stench electronic nose instrument to the smelly liquid/gas sample of standard and big
Amount pollutes the foul gas big data that live on-line checking obtains, and target output is smelt for foul smell distinguishes value and chromaticness spectrum and spectrophotometric
Conventional instrument off-line measurement value and resident complain data;
Three hidden layer deep neural network DNN of single single outputjUsing layer-by-layer off-line learning mode from bottom to top;First and second
Using single hidden layer equity neural network structure when hidden layer learns, i.e., hidden layer-output layer weight of single hidden layer equity neural network is straight
It connects and is equal to its input layer-hidden layer weight, target output is directly equal to its input, inputs component and output component according to feature point
Size is proportional transforms to range [0,3] for amount;The hidden layer activation functions of single hidden layer equity neural network are Sigmoid correction functionLearnt using error back propagation algorithm, Studying factors ηj=1/Nj, hidden layer-is abandoned after study
Output layer;
Assuming that t+1 moment concentration value yj(t+1) it is predicted, j-th of single output deep neural network DNNjIt is based on 16
Predicated response { x of the convolutional neural networks to t+1 moment gas sensor array (I-1)1(t+1),x2(t+1),…,x16(t+
1) y }, is predictedj(t+2) and yj(t+3) it is based on 16 convolutional neural networks respectively to the predicated response at t+2 and t+3 moment
(x1(t+2),x2(t+2),…,x16(t+2))TWith (x1(t+3),x2(t+3),…,x16(t+3))T;
If actually entering is the current response vector (x of gas sensor array1(t),x2(t),…,x16(t))T, can add again when necessary
Upper t moment temperature and humidity value, then deep neural network DNNjReality output be to foul gas ingredient j existing concentration value yj(t)
Estimation.
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810471708.3A CN108896706B (en) | 2018-05-17 | 2018-05-17 | The foul gas multiple spot centralization electronic nose instrument on-line analysis of big data driving |
US16/642,531 US10948467B2 (en) | 2018-05-17 | 2018-05-30 | Online centralized monitoring and analysis method for multi-point malodorous gases using electronic nose instrument |
PCT/CN2018/088913 WO2019218395A1 (en) | 2018-05-17 | 2018-05-30 | Online monitoring and analysis method for malodorous gas multi-point centralized electronic nose instrument |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810471708.3A CN108896706B (en) | 2018-05-17 | 2018-05-17 | The foul gas multiple spot centralization electronic nose instrument on-line analysis of big data driving |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108896706A CN108896706A (en) | 2018-11-27 |
CN108896706B true CN108896706B (en) | 2019-04-16 |
Family
ID=64342882
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810471708.3A Active CN108896706B (en) | 2018-05-17 | 2018-05-17 | The foul gas multiple spot centralization electronic nose instrument on-line analysis of big data driving |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108896706B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111443160A (en) * | 2020-01-23 | 2020-07-24 | 华东理工大学 | Gas-sensitive-gas chromatography information fusion and electronic nose instrument on-line analysis method |
Families Citing this family (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111282006A (en) * | 2018-12-07 | 2020-06-16 | 华东师范大学 | Plant deodorant and method for deodorizing by matching with unmanned aerial vehicle |
WO2021147274A1 (en) * | 2020-01-23 | 2021-07-29 | 华东理工大学 | Gas-sensing-gas-phase electronic nose instrument and fermentation-malodor multi-state parameter online analysis method |
CN111443159B (en) * | 2020-01-23 | 2021-02-12 | 湖州老恒和酿造有限公司 | Gas-sensitive-gas chromatography multi-source sensing and electronic nose instrument on-line detection method |
CN111426801B (en) * | 2020-05-09 | 2022-08-02 | 上海宁和环境科技发展有限公司 | Electronic nose learning and domesticating method and equipment thereof |
CN111624314A (en) * | 2020-06-30 | 2020-09-04 | 浙江三青环保科技有限公司 | Malodorous gas detection method based on principal component analysis and pattern recognition |
CN112733876A (en) * | 2020-10-28 | 2021-04-30 | 北京工业大学 | Soft measurement method for nitrogen oxides in urban solid waste incineration process based on modular neural network |
CN112433028B (en) * | 2020-11-09 | 2021-08-17 | 西南大学 | Electronic nose gas classification method based on memristor cell neural network |
CN113341066B (en) * | 2021-05-24 | 2022-04-08 | 西南石油大学 | Multi-sensor fusion technology-based on-line detection method and system for tetrahydrothiophene concentration |
CN113358702A (en) * | 2021-06-08 | 2021-09-07 | 无锡时和安全设备有限公司 | Pollution source monitoring system based on sensor array modularization |
CN113495123A (en) * | 2021-07-05 | 2021-10-12 | 北京雪域飞虹环保科技有限公司 | Mercaptan gas pattern recognition method based on multi-sensor array |
CN114544868B (en) * | 2022-01-20 | 2024-03-26 | 上海工程技术大学 | Gas detection method and system for eliminating influence of interference gas |
CN116087427A (en) * | 2023-02-08 | 2023-05-09 | 吉林大学 | Vehicle-mounted electronic nose drunk driving detection device |
CN117035562B (en) * | 2023-10-10 | 2024-01-30 | 云境商务智能研究院南京有限公司 | Environment-friendly intelligent monitoring method based on electric power big data and data analysis equipment |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1482453A (en) * | 2003-07-11 | 2004-03-17 | 华东理工大学 | Machine olfaction odor distinguishing method based on modularized composite neural net |
CN101000357A (en) * | 2007-01-08 | 2007-07-18 | 华东理工大学 | Olfactory instrument and olfactory analysing method of small automatic machine |
CN101101299A (en) * | 2007-06-25 | 2008-01-09 | 华东理工大学 | Parallel-tandem mode identification method and its uses in machine scent |
CN102353798A (en) * | 2011-07-01 | 2012-02-15 | 华东理工大学 | Biological fermentation process-oriented olfactory analog instrument and online analysis method |
CN103472094A (en) * | 2013-07-24 | 2013-12-25 | 华东理工大学 | Olfactory analog instrument and on-site analysis method for odor grade of specific substance |
CN103808904A (en) * | 2014-03-07 | 2014-05-21 | 北京拓扑智鑫科技有限公司 | Method for globally monitoring odor pollution gas in real time |
CN103824054A (en) * | 2014-02-17 | 2014-05-28 | 北京旷视科技有限公司 | Cascaded depth neural network-based face attribute recognition method |
CN105259318A (en) * | 2015-11-27 | 2016-01-20 | 北京拓扑智鑫环境科技股份有限公司 | Foul smell OU value prediction method and system based on meteorological parameters |
CN105486812A (en) * | 2015-12-28 | 2016-04-13 | 周俊杰 | Electronic nose stink grade assignment method in continuous environment air quality monitoring process and application |
CN107843695A (en) * | 2017-10-31 | 2018-03-27 | 华东理工大学 | Tobacco and the electronic nose instrument evaluation method of tobacco product aesthetic quality |
-
2018
- 2018-05-17 CN CN201810471708.3A patent/CN108896706B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1482453A (en) * | 2003-07-11 | 2004-03-17 | 华东理工大学 | Machine olfaction odor distinguishing method based on modularized composite neural net |
CN101000357A (en) * | 2007-01-08 | 2007-07-18 | 华东理工大学 | Olfactory instrument and olfactory analysing method of small automatic machine |
CN101101299A (en) * | 2007-06-25 | 2008-01-09 | 华东理工大学 | Parallel-tandem mode identification method and its uses in machine scent |
CN102353798A (en) * | 2011-07-01 | 2012-02-15 | 华东理工大学 | Biological fermentation process-oriented olfactory analog instrument and online analysis method |
CN103472094A (en) * | 2013-07-24 | 2013-12-25 | 华东理工大学 | Olfactory analog instrument and on-site analysis method for odor grade of specific substance |
CN103824054A (en) * | 2014-02-17 | 2014-05-28 | 北京旷视科技有限公司 | Cascaded depth neural network-based face attribute recognition method |
CN103808904A (en) * | 2014-03-07 | 2014-05-21 | 北京拓扑智鑫科技有限公司 | Method for globally monitoring odor pollution gas in real time |
CN105259318A (en) * | 2015-11-27 | 2016-01-20 | 北京拓扑智鑫环境科技股份有限公司 | Foul smell OU value prediction method and system based on meteorological parameters |
CN105486812A (en) * | 2015-12-28 | 2016-04-13 | 周俊杰 | Electronic nose stink grade assignment method in continuous environment air quality monitoring process and application |
CN107843695A (en) * | 2017-10-31 | 2018-03-27 | 华东理工大学 | Tobacco and the electronic nose instrument evaluation method of tobacco product aesthetic quality |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111443160A (en) * | 2020-01-23 | 2020-07-24 | 华东理工大学 | Gas-sensitive-gas chromatography information fusion and electronic nose instrument on-line analysis method |
CN111443160B (en) * | 2020-01-23 | 2021-02-12 | 湖州老恒和酿造有限公司 | Gas-sensitive-gas chromatography information fusion and electronic nose instrument on-line analysis method |
Also Published As
Publication number | Publication date |
---|---|
CN108896706A (en) | 2018-11-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108896706B (en) | The foul gas multiple spot centralization electronic nose instrument on-line analysis of big data driving | |
WO2019218395A1 (en) | Online monitoring and analysis method for malodorous gas multi-point centralized electronic nose instrument | |
CN108709955B (en) | A kind of foul gas multiple spot centralization on-line monitoring and analysis system and method | |
Gebicki | Application of electrochemical sensors and sensor matrixes for measurement of odorous chemical compounds | |
CN107677709B (en) | Method for testing semiconductor gas sensitive element with gas sensitive material preparation step | |
CN111443160B (en) | Gas-sensitive-gas chromatography information fusion and electronic nose instrument on-line analysis method | |
WO2021147274A1 (en) | Gas-sensing-gas-phase electronic nose instrument and fermentation-malodor multi-state parameter online analysis method | |
CN108760913A (en) | Atmospheric odor pollution traceability system and application | |
CN116340723B (en) | Rural water pollution quick tracing method and system based on big data | |
CN112304831B (en) | Mass conservation-based student behavior vs. PM in classroom2.5Concentration calculating method | |
CN110187055A (en) | A kind of factory's circle type monitor between different malodor source producers | |
Nicolas et al. | The electronic nose as a warning device of the odour emergence in a compost hall | |
CN114578011A (en) | Water quality monitoring method based on multi-sensor multi-source data fusion | |
CN209841839U (en) | Factory boundary type monitor used among different odor source manufacturers | |
Saini et al. | Indoor air quality monitoring with IoT: predicting PM10 for enhanced decision support | |
CN109030737A (en) | A kind of sewage foul gas detection system and sewage foul gas detection method | |
CN110057410B (en) | Method and device for measuring and calculating pollutant production of daily domestic sewage of per capita | |
Zarra et al. | Smart instrumental Odour Monitoring Station for the efficient odour emission management and control in wastewater treatment plants | |
CN113340943B (en) | Method for analyzing odor type and odor intensity in water body based on fingerprint | |
Lorwongtragool et al. | Portable artificial nose system for assessing air quality in swine buildings | |
Gongora et al. | Urban monitoring of unpleasant odors with a handheld electronic nose | |
CN107764793A (en) | Detection method of the electronic nose to aspergillus oryzae fermentation situation in bean paste yeast making process | |
CN111443159B (en) | Gas-sensitive-gas chromatography multi-source sensing and electronic nose instrument on-line detection method | |
CN112707508B (en) | BP-ANN-based agricultural sewage facility COD treatment effect prediction method, device and platform | |
JP2007218704A (en) | Smell discrimination method, smell measuring method, smell discrimination program, smell discrimination device and smell measuring instrument |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |