CN103914622B - A kind of chemical leakage fast prediction alarm emergency response decision-making method - Google Patents
A kind of chemical leakage fast prediction alarm emergency response decision-making method Download PDFInfo
- Publication number
- CN103914622B CN103914622B CN201410136961.5A CN201410136961A CN103914622B CN 103914622 B CN103914622 B CN 103914622B CN 201410136961 A CN201410136961 A CN 201410136961A CN 103914622 B CN103914622 B CN 103914622B
- Authority
- CN
- China
- Prior art keywords
- leakage
- wind direction
- sensor
- scene
- data
- 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
Abstract
A kind of chemical leakage fast prediction alarm emergency response decision-making method of the present invention, diffusion model simulation is combined with neutral net and gas sensor system, is applied to quick early warning and the aid decision of industrial park poisonous gas leakage, including：Garden risk factors collection, for recognizing all kinds of leakage accidents that may occur；Numerical simulation, the accident to being likely to occur is simulated, and obtains the coverage of pernicious gas；Data screening, according to live part in actual sensor placement extraction and restructuring numerical simulation result；Neural metwork training, is trained using the data after screening to specific neural network model, to obtain the model parameter for particular industry garden and ambient conditions, and uses redundant data Verification；Sensing system is integrated with neural network model, and model is combined with sensor DCS.
Description
Technical field
The invention belongs to the safe early warning technical field in industrial production, more particularly to a kind of chemical leakage fast prediction
Alarm emergency response decision-making method.
Background technology
Many process industries can be used or produce some harmful poisonous, pernicious gases in process of production
（Such as chlorine, phosgene etc.）, once there is toxic and harmful leakage accident, the toxic and harmful for leaking out in these industrial parks
Serious harm may be caused to a range of mankind in periphery.When toxic and harmful leakage accident occurs, leakage
Material and substantially leak position can with comparalive ease determine that but the leakage rate or leak rate of pernicious gas are then difficult
Scene obtains.It is accident emergency in the diffusion tendency and coverage of the limited information prediction toxic gas of limited time utilization
The important step of response process.At present, traditional numerical d ispersion forecast model, including Gaussian plume model, Fluid Mechanics Computation
Model, and some more ripe diffusion simulations softwares are required for user to provide detailed source of leaks leakage rate, and
Needing the regular hour to calculate can just obtain result.Therefore, the leakage rate or leakage rate of source of leaks and calculating more long
These traditional gas diffusion models of time restriction accident emergency respond and aid decision during applications, they are more
Ground is used for the investigation and analysis after accident occurs.Due to historical reasons, China quite a few poisonous and harmful gas may occur
Revealed when there is severe toxic pernicious gas with the presence of residential block in the range of the 3~5km of industrial park periphery of body leakage accident
During accident, these toxic and harmfuls are likely to diffuse to outside the battery limit (BL) of industrial park, and threat is produced to residential block.Without fast
When speed effectively predicts the method for toxic and harmful range of scatter, policymaker often considers pernicious gas with the worst situation
Coverage, and predicting the outcome for worst case often means that government department needs to evacuate tens of thousands of or even ten tens of thousands of people, this is
It is very unpractical.Therefore a kind of method of fast and effeciently prediction harmful gases diffusion scope is responded and auxiliary for accident emergency
Help decision-making significant.
Process toxic and harmful factory all can targetedly near toxic and harmful storage tank set one or
Multiple harmful gas sensors, for having monitored whether that gas is leaked.These gas sensors are connected with control centre, there is provided
Toxic gas leakage alarm information.But many industrial parks are without the effect for playing gas sensor completely, accident hair at present
When raw, emergency response personnel can only obtain the species of gas leakage by sensor, instantaneous concentration, and can not be by these information
The distribution situation of gas leakage is rapidly predicted, and then takes effective control measure, formulate rational evacuation plan.
At present, the accident consequence analysis method for commonly using in the world mainly includes using different types of numerical model
（Gaussian plume model, luid mechanics computation model（CFD）And business prototype PHAST, FLACS etc. are to the accident that has occurred and that
Reappeared, be illustrated in the coverage of gas leakage under the conditions of accident leakage at that time（Including dead area, severely injured area and influence
Area）, and study influence of the Atmospheric Diffusion condition to gas leakage range of scatter and concentration distribution.But use Numerical Model Analysis
Shortcoming is the source strength that must be known by source of leaks（Leakage rate）And leakage form（Blast leakage/aperture leakage etc.）, with reference to
Meteorologic parameter and mass transfer diffusion equation are simulated, and model is complicated, calculate that time-consuming, it is impossible to for real-time under accident condition or
Person's fast prediction.
The content of the invention
In order to overcome the shortcoming of above-mentioned prior art, it is an object of the invention to provide a kind of chemical leakage fast prediction
Alarm emergency response decision-making method, risk analysis is carried out to industrial park, reveals Scene Simulation, is supplemented and is optimized garden in right amount and show
Some toxic and harmful sensing systems, the toxic gas alarm system of garden is combined with gas diffusion forecast analysis,
For accident emergency Response Decision provides technical support.
To achieve these goals, the technical solution adopted by the present invention is：
A kind of chemical leakage fast prediction alarm emergency response decision-making method, including risk factors collection, scene numerical value
Simulation, analog result screening, neural metwork training and training result and sensing system integrated totally five stages, wherein：
Risk factors collection includes：The risk factors of chemical industrial park are recognized, quantifies each risk elements and according to identification
Risk elements and its span, reasonable value simultaneously combine the various leakage scenes that may occur；
The scene numerical simulation stage includes：The leakage being likely to occur that will be constituted in garden risk factors collection step
Scene be simulated, with obtain it is different leakage scenes under reveal gas coverage；
Analog result screening includes：
A kind of key parameter including in 4 placement schemes of gas sensor is optimized, is obtained using simulation
Leakage Gas concentration distribution carries out simple optimizing, obtains most suitable sensor placement；
According to the sensor placement scheme for determining, the virtual detection number in the range of the measurable wind direction of sensor placement is extracted in
According to this and matching environment sensitive spot gas leakage diffusion data, training data is prepared according to the training logic of neutral net
And verification data；
Neural metwork training step includes：
The feedforward neural network for Function Fitting is set up, using ready training data as the defeated of neural metwork training
Enter output, calculating network parameter；
Inspection data importation is input into neutral net, as a result part and neural network prediction Comparative result, assessment is pre-
Survey precision；
Neutral net and sensor and garden control system integration step：
By the sunykatuib analysis of each risk source and training result and risk source and the geographical location information of environment sensitive spot
Combined in the form of database and caller, and the data-interface with sensing system is provided, realize from the accident-biography
The workflow of sensor alarm-model fast prediction-aid decision.
The garden risk identification part comprises the steps of：
Step one, recognized using risk assessment risks and assumptions and accident scenarios factor and judge may generation all kinds of leakages
The indispensable element of scene, including risk elements and accident scenarios key element.These key elements include being not limited to specified risk material species（Have
Malicious/inflammable gas or volatile liquid）, reserves/consumption, storage location（Space coordinates）, storage form（Pressure, temperature, if
It is standby）, wind speed, humidity, temperature, atmospheric stability, surrounding enviroment sensitive spot position（Space coordinates）Deng.
Step 2, the span for determining each risks and assumptions and accident situational factor respectively（For continuous variable）,
Sequence of values is divided with a fixed step size respectively in the span of each factor（Step-length must not exceed the factor span
10%）, finally by each valued combinations of different risks and assumptions/accident scenarios factor into the substantial amounts of leakage feelings that may occur
Scape.N kind risk factors are such as had, each risk factors has NiPlant value（Ni>=1, integer）, then havePlant leakage
Scene.
The numerical model that the scene numerical simulation stage uses includes：Gauss and class Gauss diffusion model (Gaussian
Plume Model), Fluid Mechanics Computation（CFD）Model and PHAST, FLACS, SLAB, ALOHA, HGSYSTEM etc. integrate mould
Type.Finally obtain J kinds leakage Scene Simulation result.
The analog result screening step needs to be laid out one group of 4 sensor optimization, and 4 sensors are using just
Polygonized structure, serial number is 1~4, and the spacing of adjacent numbering sensor is equal, and Optimization Steps are as follows：
Step one, the position according to source of leaks and garden layout, the test limit of sensor, it is determined that basic layout parameter
L1, the span of d and α.Wherein L1It is No. 1 distance of sensor distance source of leaks, d is sensor spacing, and α is 1-2 and 1-4
The 1/2. of sensor angle
Step 2, cardinal wind is determined according to wind frequency information, sensor is arranged symmetrically in cardinal wind leeward upwards, actual
Wind direction and cardinal wind angle are β.And take a higher value β0It is interval as optimization, by its it is discrete be actual wind direction sequence (β1,
β2,…,βM), sequence length is M.（Typically take β0=50 °, by interval [- 50 °, 50 °] the discrete wind direction sequence to be spaced 1 °）
Step 3, to layout parameter L1, d, α is separated into argument sequence in span, and length is respectively P1,P2,P3,
CompositionPlant placement scheme.
Step 4, to kth kind placement scheme (k=1,2 ..., K) use data screening algorithm, for jth kind Scene Simulation
As a result（j=1,2,…,J）, it is β to filter out the placement scheme in wind directionmEfficient layout number when (m=1,2 ..., M)
Step 5, different wind direction β are calculated to kth kind placement scheme (k=1,2 ..., K)mScene when (m=1,2 ..., M)
Relevance factorRecord meetsM=1,2, ..., the maximum wind direction scope W of Mk, take satisfaction { max (Wk),k=1,
2 ..., K layout for optimize be laid out.
The standard that efficient layout is judged in analog result screening step is：For any sensor layout, to specific
Every kind of leakage scene under wind direction, it is called in the wind direction to have 3 or more than 3 to cover the then layout by plume in 4 sensors
With the efficient layout under the conditions of leakage scene.
In affiliated analog result screening step, scene relevance factor is(m=1,2,…,Mk=1,2,…,
K)。
The analog result data screening part includes following filtering algorithm：
Step one, check sensor placement parameter L1, d, α, actual wind direction β, lower wind direction sensitivity source position (L2, Y2) and
Whether leakage scene sequence number J etc. meets the requirements
Step 2, the gas diffusion distance for revealing j-th scene (j=1,2 ..., J) analog result, concentration distribution expand
The time of dissipating and plume width enter row interpolation, it is determined that screening starting point
Step 3, calculating parameter γ, θ2,θ4,W1, and W2, wherein γ is 2, No. 4 sensors with source of leaks as summit and master
Air guiding angle, θ2,θ4For 2, No. 4 sensors with source of leaks as summit and actual wind direction angle, W1It is environment sensitive spot distance
The vertical range of plume center line, W2It is the plume half-breadth at environment sensitive spot
Whether step 4, inspection sensor placement are efficient layout under the leakage scene and actual wind direction, if so, after
It is continuous；If it is not, skipping and recording this leakage scene, continue step 2
Step 5, the pollution sources concentration of 3 effective sensor positions of extraction and diffusion time.
Whether step 6, inspection leakage gas can diffuse to lower wind direction environment sensitive spot, i.e. computing environment sensitive spot exists
Location parameter Q in j-th leakage scenej.If it can, continue step 7, if it could not, recording this sequence number and continuing step
Two.
The concentration of pernicious gas and diffusion time at step 7, the lower wind direction environment sensitive spot of extraction.
Step 8, by all sensing station concentration extracted and time value and lower wind direction sensitive spot concentration and time
Value is arranged and preserved.
Whether the described method that judges gas leakage and can diffuse to lower wind direction environment sensitive spot is：Computing environment sensitive spot
Location parameter Q in gas leakage plumej=W1/W2If, Qj≤ 1, then environment sensitive spot extended influence by gas leakage, instead
It is then unaffected.
Neural metwork training part implementation process is as follows：
Step one, determine network inputs, using live available parameter as input, including pressure store (p), wind speed (v),
Wind direction (β), temperature (T), humidity (h), atmospheric stability (S), the valid density measurement of 3 gas sensors of wind direction under risk source
Value (c1,c2,c3) and time of fire alarming (t1,t2,t3), the position (L of lower wind direction sensitive spot2,Y2)
Step 2, determine network export, at following wind direction environment sensitive spot pernicious gas reach concentration (c) and the time
T () is used as output.
Step 3, network internal structure determination, use feedforward neural network（BP networks）, comprising a non-linear hidden layer
（Sigmoid transmission functions）With a linear convergent rate layer（Linear transmission functions）, network is not comprising backfeed loop.
Step 4, according to input and output requirement, be ready for and output data matrix and superfluous using data screening algorithm
Remaining checking data.（Matrix column is each key element of input and output, the inputoutput data of behavior difference scene）
Step 5, using MATLAB Neural Network Toolbox, according to single unify training method neutral net is used it is defeated
Enter output data training, obtain network parameter.
Step 6, the difference exported as neutral net input, comparing cell output and redundancy using redundant input data,
Assessment prediction precision.
The neutral net comprises the following steps with sensing system integration section：
Step one, the area map for setting up the environment sensitive spot of wind direction under single risk source and cardinal wind.
Step 2, risk analysis is carried out to certain single risk source, recognize that possible leakage scene is simultaneously carried out to leakage scene
Simulation.
Step 3, the sensor placement optimization for performing data screening step, obtain the sensor placement under cardinal wind
The maximum of wind direction is applicable angle.
Step 4, under the risk source cardinal wind wind direction maximum carried out using all environment sensitive spots in angular range
Data screening, obtains one group of input and output matrix of neural metwork training.
Step 5, the neutral net input and output matrix obtained using data screening are trained to be assessed with precision of prediction.
Step 6, the neural network parameter that obtains will be trained to be fabricated to the application program that can be called automatically, make sensor
DCS data can be gathered by data-interface and be calculated using application program.
When step 7, true accident occur, the prefabricated application program of sensor alarm-startup-prediction leakage diffusion
The influence of environmentally sensitive point（Time for including whether to influence environment sensitive spot or gas leakage to reach environment sensitive spot and dense
Degree）- decision-maker judges whether to need to evacuate.
Compared with prior art, the present invention proposes a kind of the simple of sensor placement for processing toxic and harmful factory
The method for quick predicting of optimization method, and introducing and employment early stage risk identification, numerical simulation and neural metwork training, so as to
Accident can timely and effectively predict the range of scatter of toxic and harmful when occurring, be garden surrounding resident area evacuating personnel and
Protection provides decision support.On the basis of the existing toxic and harmful sensor in garden, for garden production technology and week
Surrounding environment, completes, to garden production process risk identification, to reveal Scene Simulation, improves sensor placement and carries out neutral net
Training, overcomes toxic gas leakage accident and the deficiency that alarm information is not enough to support accident emergency response occurs, it is ensured that
The applicability and accuracy of method.
Brief description of the drawings
Fig. 1 is the flow chart of one embodiment of the invention method.
Fig. 2 is one embodiment of the invention method application scenario schematic diagram.
Fig. 3 is the sensor placement Optimal Curve that one embodiment of the invention method is obtained.
Fig. 4 is the rudimentary algorithm that one embodiment of the invention method is used for data screening
Fig. 5 is the neural network structure figure of one embodiment of the invention method application.
Fig. 6 is verified for the neutral net redundant data of one embodiment of the invention method application.
Whether Fig. 7 is the judgement impacted to sensitizing range of the neutral net of one embodiment of the invention method application.
Specific embodiment
Describe embodiments of the present invention in detail with reference to the accompanying drawings and examples.
Fig. 1 is the flow chart of method according to an embodiment of the invention, side according to an embodiment of the invention
Method includes：
- risk factors collection step
Risks and assumptions and accident scenarios factor are recognized using risk assessment and judge all kinds of leakage scenes of possible generation
Indispensable element, including risk elements and accident scenarios key element.These key elements include being not limited to specified risk material species（Poisonous/easily
The gas or volatile liquid of combustion）, reserves/consumption, storage location（Space coordinates）, storage form（Pressure, temperature, equipment）、
Wind speed, humidity, temperature, atmospheric stability, surrounding enviroment sensitive spot position（Space coordinates）Deng.Then each risk is determined respectively
The span of the factor and accident situational factor（For continuous variable）, respectively with certain in the span of each factor
Step-length divides sequence of values（Step-length must not exceed the 10% of the factor span）, finally by different risks and assumptions/accident feelings
Leakage scene of each valued combinations of scape factor into substantial amounts of possible generation.Such as have n kind risk factors, each risk factors
There is NiPlant value（Ni>=1, integer）, then havePlant leakage scene.
- leakage Scene Simulation step
The step carries out Numerical-Mode using traditional gas diffusion model, all kinds of leakage scenes to being identified in previous step
Intend, obtain the toxic and harmful diffusion data and distribution situation under different situations.Available model includes Gauss and class Gauss
Diffusion model (Gaussian Plume Model), Fluid Mechanics Computation（CFD）Model and PHAST, FLACS, SLAB,
ALOHA, HGSYSTEM etc..
- valid data screen step
Close the position that Fig. 2 show the risk source-plume-sensor-lower wind direction sensitive spot of valid data screening step
System's figure.Analog result screening step needs to be laid out optimization to one group of 4 sensor, and 4 sensors use regular polygon
Structure, serial number is 1~4, and the spacing of adjacent numbering sensor is equal.Optimization Steps are as follows：
Step one, the position according to source of leaks and garden layout, the test limit of sensor, it is determined that basic layout parameter
L1, the span of d and α.Wherein L1It is No. 1 distance of sensor distance source of leaks, d is sensor spacing, and α is 1-2 and 1-4
The 1/2. of sensor angle
Step 2, cardinal wind is determined according to wind frequency information, sensor is arranged symmetrically in cardinal wind leeward upwards, actual
Wind direction and cardinal wind angle are β.And take a higher value β0It is interval as optimization, by its it is discrete be actual wind direction sequence (β1,
β2,…,βM), sequence length is M.（Typically take β0=50 °, by interval [- 50 °, 50 °] the discrete wind direction sequence to be spaced 1 °）
Step 3, to layout parameter L1, d, α is separated into argument sequence in span, and length is respectively P1,P2,P3,
CompositionPlant placement scheme.
Step 4, to kth kind placement scheme (k=1,2 ..., K) use data screening algorithm, for jth kind Scene Simulation
As a result（j=1,2,…,J）, it is β to filter out the placement scheme in wind directionmEfficient layout number when (m=1,2 ..., M)
Wherein efficient layout is defined as：For any sensor layout, to specific wind direction under every kind of leakage scene, in 4 sensors
There are 3 or more than 3 to be covered by plume（Plume concentration can be detected）Placement scheme.
Step 5, different wind direction β are calculated to kth kind placement scheme (k=1,2 ..., K)mScene when (m=1,2 ..., M)
Relevance factor(m=1,2 ..., M k=1,2 ..., K), record meetsM=1,2, ..., M's
Maximum wind direction scope Wk, take satisfaction { max (Wk), k=1,2 ..., K layout for optimize be laid out.
Fig. 3 is by L1In the 27 of the value composition of=[40,60,80] m, d=[20,30,40] m and α=[30 °, 45 °, 60 °]
The layout optimization result of placement scheme（Fig. 3 d）.Under conditions of being 95% in scene utilization rate, sensor placement scheme L1=40m, d
=20m and α=60 ° are laid out to optimize, and the maximum wind direction scope of application of the layout is 20 °~20 ° of ﹣.Determine that sensor is optimized
After layout, it is simulated result using the filtering algorithm shown in Fig. 4 and screens：
Step one, check sensor placement parameter L1, d, α, actual wind direction β, lower wind direction sensitivity source position (L2, Y2) and
Whether leakage scene sequence number j etc. meets the requirements
Step 2, the gas diffusion distance for revealing j-th scene (j=1,2 ..., J) analog result, concentration distribution expand
The time of dissipating and plume width enter row interpolation, it is determined that screening starting point
Step 3, calculating parameter γ, θ2,θ4,W1, and W2, wherein γ is 2, No. 4 sensors with source of leaks as summit and master
Air guiding angle, θ2,θ4For 2, No. 4 sensors with source of leaks as summit and actual wind direction angle, W1It is environment sensitive spot distance
The vertical range of plume center line, W2It is the plume half-breadth at environment sensitive spot
Whether step 4, inspection sensor placement are efficient layout under the leakage scene and actual wind direction（Effectively sense
Device number is more than 3）, if so, continuing；If it is not, skipping and recording this leakage scene, continue step 2
Step 5, the pollution sources concentration of 3 effective sensor positions of extraction and diffusion time.
Whether step 6, inspection leakage gas can diffuse to lower wind direction environment sensitive spot, i.e. computing environment sensitive spot exists
Location parameter Q in j-th leakage scenej=W1/W2.If it can, i.e. Qj≤ 1, continue step 7, if it could not, i.e. Qj>1, note
Record this sequence number and continue step 2.
The concentration of pernicious gas and diffusion time at step 7, the lower wind direction environment sensitive spot of extraction.
Step 8, by all sensing station concentration extracted and time value and lower wind direction sensitive spot concentration and time
Value is arranged and preserved.
- neural metwork training step
Neural metwork training part implementation process is as follows：
Step one, determine network inputs, using live available parameter as input, including pressure store (p), wind speed (v),
Wind direction (β), temperature (T), humidity (h), atmospheric stability (S), the valid density measurement of 3 gas sensors of wind direction under risk source
Value (c1,c2,c3) and time of fire alarming (t1,t2,t3), the position (L of lower wind direction sensitive spot2,Y2)
Step 2, determine network export, at following wind direction environment sensitive spot pernicious gas reach concentration (c) and the time
T () is used as output.
Step 3, network internal structure determination, as shown in figure 5, using feedforward neural network（BP networks）, it is non-comprising one
Linear hidden layer（Sigmoid transmission functions）With a linear convergent rate layer（Linear transmission functions）, network is not comprising feeding back to
Road.
Step 4, according to input and output requirement, be ready for and output data matrix and superfluous using data screening algorithm
Remaining checking data.（Matrix column is each key element of input and output, the inputoutput data of behavior difference scene）
Step 5, using MATLAB Neural Network Toolbox, according to single unify training method neutral net is used it is defeated
Enter output data training, obtain network parameter.
Step 6, the difference exported as neutral net input, comparing cell output and redundancy using redundant input data,
Assessment prediction precision.As Fig. 6 shows a neural network prediction situation for chlorine gas leakage embodiment.All subgraphs is vertical in figure
Coordinate is under specific leakage situation, according to the position of 3 sensor alarm information, weather condition and lower wind direction sensitizing range
Put, using neutral net to lower wind direction sensitizing range
The concentration of (a) chlorine（ppm）
B () chlorine reaches the time of sensitive position（s）
The half-breadth W2 of (c) sensitive position plume（m）
The vertical range W1 of (d) sensitive position actual wind direction of distance（m）
And the abscissa of Fig. 6 a~6d is then the result of Phast simulations, in theory, the fitting result of scatterplot should be straight in figure
Line y=x, actual result shows that the only neural network prediction result of chlorine gas concentration is less than normal compared to Phast analog results（Slope is about
It is 0.93）, predicting the outcome for chlorine arrival time, W1 and W2 be all very accurate.
The W1 and the value of W2 obtained using Fig. 6 c and 6d can predict whether lower wind direction sensitizing range receives the prestige of chlorine leakage
The side of body, as shown in fig. 7, abscissa is the quantity of the leakage scene used：
The relative error of (a) neutral net concentration prediction value and the Phast analogues value
B W1/W2 ratios that () calculates according to the W1 and W2 of neural network prediction in Fig. 6
C W1/W2 ratios that () obtains according to reconnaissance in Phast simulations and plume
D () will predict that W1/W2 and theory W1/W2 values subtract each other the logical consequence for obtaining after rounding
According to (a) it can be found that worst error of the neural network prediction value compared with theoretical value is no more than 30%, Fig. 7 b and 7c
Closely, its difference shows in 7d：2780 reveal scene in, have 2 leakage scenes under wind direction sensitive spot in theory
Threatened by chlorine gas leakage, but the result of neural network prediction shows that it is not on the hazard, and referred to as " fails to report ", and rate of failing to report is
0.072%；On the other hand, thanked in only 14 leakage scenes and share sensitive spot and be not on the hazard in theory, but neutral net is pre-
It receives threat to survey display, referred to as " reports by mistake ", and rate of false alarm is 0.504%.
Step 7, the neural network parameter that obtains will be trained to be fabricated to the application program that can be called automatically, make sensor
DCS data can be gathered by data-interface and be calculated using application program.
When step 8, true accident occur, the prefabricated application program of sensor alarm-startup-prediction leakage diffusion
The influence of environmentally sensitive point（Time for including whether to influence environment sensitive spot or gas leakage to reach environment sensitive spot and dense
Degree）- decision-maker judges whether to need to evacuate.
Claims (9)
1. a kind of chemical leakage fast prediction alarm emergency response decision-making method, including risk factors collection, scene Numerical-Mode
Plan, analog result screening, neural metwork training and training result and sensing system integrated totally five stages：
Risk factors collection includes：The risk factors of chemical industrial park are recognized, quantifies each risk elements and according to the risk of identification
Key element and its span, reasonable value simultaneously combine the various leakage scenes that may occur；
The scene numerical simulation stage includes：The leakage scene being likely to occur that will be constituted in garden risk factors collection step
Be simulated, with obtain it is different leakage scenes under reveal gas coverage；
Analog result screening includes：
A kind of key parameter including in 4 placement schemes of gas sensor is optimized, the leakage obtained using simulation
Gas concentration distribution carries out simple optimizing, obtains most suitable sensor placement；
According to determine sensor placement scheme, be extracted in the virtual detection data in the range of the measurable wind direction of sensor placement with
And matching environment sensitive spot gas leakage diffusion data, prepare training data and school according to the training logic of neutral net
Test data；
Neural metwork training step includes：
The feedforward neural network for Function Fitting is set up, the input of ready training data as neural metwork training is defeated
Go out, calculating network parameter；
Inspection data importation is input into neutral net, as a result part and neural network prediction Comparative result, assessment prediction essence
Degree；
Neutral net and sensor and garden control system integration step：
By the geographical location information of the sunykatuib analysis of each risk source and training result and risk source and environment sensitive spot with number
Combined according to the form of storehouse and caller, and the data-interface with sensing system is provided, realize from the accident-sensor
The workflow of alarm-model fast prediction-aid decision；
Characterized in that, the neutral net comprises the following steps with sensing system integration section：
Step one, the area map for setting up the environment sensitive spot of wind direction under single risk source and cardinal wind；
Step 2, risk analysis is carried out to certain single risk source, recognize that possible leakage scene is simultaneously simulated to leakage scene；
Step 3, the sensor placement optimization for performing data screening step, obtain sensor placement wind direction under cardinal wind
Maximum be applicable angle；
Step 4, under the risk source cardinal wind wind direction maximum carry out data using all environment sensitive spots in angular range
Screening, obtains one group of input and output matrix of neural metwork training；
Step 5, the neutral net input and output matrix obtained using data screening are trained to be assessed with precision of prediction；
Step 6, the neural network parameter that obtains will be trained to be fabricated to the application program that can be called automatically, make the DCS of sensor
Data can be gathered by data-interface and calculated using application program；
When step 7, true accident occur, the prefabricated application program of sensor alarm-startup-prediction leakage diffusion couple ring
Influence-the decision-maker of border sensitive spot judges whether to need to evacuate.
2. chemical leakage fast prediction alarm emergency response decision-making method according to claim 1, it is characterised in that institute
Garden risk factors collection part is stated to comprise the steps of：
Step one, recognized using risk assessment risks and assumptions and accident scenarios factor and judge may generation all kinds of leakage scenes
Indispensable element, including risk elements and accident scenarios key element；
Step 2, the span for determining each risks and assumptions and accident situational factor respectively, in the value model of each factor
Enclose interior respectively with fixed step size division sequence of values, wherein step-length must not exceed the 10% of the factor span, finally will not
The leakage scene that each valued combinations of risks and assumptions/accident scenarios factor together occur into substantial amounts of possibility, when total n kind wind
During dangerous factor, each risk factors has NiPlant value, wherein Ni>=1, and NiIt is integer, then hasPlant leakage feelings
Scape.
3. chemical leakage fast prediction alarm emergency response decision-making method according to claim 2, it is characterised in that institute
Stating the numerical model that the scene numerical simulation stage uses includes：Gauss and class Gauss diffusion model (Gaussian Plume
Model), Fluid Mechanics Computation (CFD) model and PHAST, FLACS, SLAB, ALOHA and HGSYSTEM Integrated Models, most
After obtain J kinds leakage Scene Simulation result.
4. chemical leakage fast prediction alarm emergency response decision-making method according to claim 2, it is characterised in that institute
State analog result screening step and one group of 4 sensor are laid out with optimization, and 4 sensors use regular polygon structure, it is suitable
Sequence numbering is 1~4, and the spacing of adjacent numbering sensor is equal, and Optimization Steps are as follows：
Step one, the position according to source of leaks and garden layout, the test limit of sensor, it is determined that basic layout parameter L1, d and α
Span, wherein L1It is No. 1 distance of sensor distance source of leaks, d is sensor spacing, and α is 1-2 and 1-4 sensors
The 1/2 of angle；
Step 2, cardinal wind is determined according to wind frequency information, sensor is arranged symmetrically in the leeward of cardinal wind upwards, actual wind
It is β to cardinal wind angle, and takes a higher value β0It is interval as optimization, by its it is discrete be actual wind direction sequence (β1,
β2,…,βM), sequence length is M；
Step 3, to layout parameter L1, d, α is separated into argument sequence in span, and length is respectively P1,P2,P3, compositionPlant placement scheme；
Step 4, to kth kind placement scheme use data screening algorithm, for jth kind Scene Simulation result, filter out the layout
Scheme is β in wind directionmWhen efficient layout numberWherein k=1,2 ..., K；J=1,2 ..., J；M=1,2 ..., M；
Step 5, wind direction βs different to the calculating of kth kind placement schememWhen scene relevance factorRecord meetsMaximum wind direction scope Wk, take satisfaction { max (Wk), k=1,2 ..., K layout be
Optimize layout.
5. chemical leakage fast prediction alarm emergency response decision-making method according to claim 4, it is characterised in that institute
State and judge that the standard of efficient layout is in analog result screening step：For any sensor layout, to specific wind direction under it is every
Leakage scene is planted, it is called under the conditions of the wind direction and leakage scene there are more than 3 to cover the then layout by plume in 4 sensors
Efficient layout.
6. the chemical leakage fast prediction alarm emergency response decision-making method according to claim 4 or 5, its feature exists
In in affiliated analog result screening step, scene relevance factor isWherein m=1,2 ..., M；K=1,
2,…,K。
7. chemical leakage fast prediction alarm emergency response decision-making method according to claim 6, it is characterised in that institute
State analog result data screening part and include following filtering algorithm：
Step one, inspection sensor placement parameter L1, d, α, actual wind direction β, lower wind direction sensitivity source position (L2,Y2) and leakage
Whether scene sequence number J meets the requirements；
Step 2, the gas diffusion distance for revealing j-th Scene Simulation result, concentration distribution, diffusion time and plume width
Enter row interpolation, it is determined that screening starting point, wherein j=1,2 ..., J；
Step 3, calculating parameter γ, θ2, θ4,W1, and W2, wherein γ be No. 2 or No. 4 sensors with source of leaks be summit and dominate
The angle of wind direction, θ2It is No. 2 sensors with the angle that source of leaks is summit and actual wind direction, θ4For No. 4 sensors are with source of leaks
Summit and the angle of actual wind direction, W1It is environment sensitive spot apart from the vertical range of plume center line, W2At environment sensitive spot
Plume half-breadth；
Whether step 4, inspection sensor placement are efficient layout under the leakage scene and actual wind direction, if so, continuing；If
It is not to skip and record this leakage scene, continues step 2；
Step 5, the pollution sources concentration of 3 effective sensor positions of extraction and diffusion time；
Whether step 6, inspection leakage gas can diffuse to lower wind direction environment sensitive spot, i.e. computing environment sensitive spot is at j-th
Location parameter Q in leakage scenej, if it can, continue step 7, if it could not, recording this sequence number and continuing step 2；
The concentration of pernicious gas and diffusion time at step 7, the lower wind direction environment sensitive spot of extraction；
It is step 8, all sensing station concentration extracted and time value and lower wind direction sensitive spot concentration and time value is whole
Manage and preserve.
8. chemical leakage fast prediction alarm emergency response decision-making method according to claim 7, it is characterised in that institute
Whether the method for judging gas leakage and diffusing to lower wind direction environment sensitive spot stated is：Computing environment sensitive spot is in gas leakage
Location parameter Q in plumej=W1/W2If, Qj≤ 1, then environment sensitive spot is extended influence by gas leakage, otherwise does not receive then
Influence.
9. chemical leakage fast prediction alarm emergency response decision-making method according to claim 1, it is characterised in that institute
State neural metwork training part implementation process as follows：
Step one, network inputs are determined, using live available parameter as input, including pressure store (p), wind speed (v), wind direction
(β), temperature (T), humidity (h), atmospheric stability (S), the valid density measured value of 3 gas sensors of wind direction under risk source
(c1,c2,c3) and time of fire alarming (t1,t2,t3), the position (L of lower wind direction sensitive spot2,Y2)；
Step 2, determine network export, at following wind direction environment sensitive spot pernicious gas reach concentration (c) and the time (t) make
It is output；
Step 3, network internal structure determination, are BP networks using feedforward neural network, comprising a non-linear hidden layer and
Individual linear convergent rate layer, non-linear hidden layer takes sigmoid transmission functions, and linear convergent rate layer takes linear transmission functions, and network is not
Comprising backfeed loop；
Step 4, according to input and output requirement, be ready for and output data matrix and redundancy using data screening algorithm
Checking data；
Step 5, using MATLAB Neural Network Toolbox, unify training method according to single defeated using input to neutral net
Go out data training, obtain network parameter；
Step 6, the difference exported as neutral net input, comparing cell output and redundancy using redundant input data, assessment
Precision of prediction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410136961.5A CN103914622B (en) | 2014-04-04 | 2014-04-04 | A kind of chemical leakage fast prediction alarm emergency response decision-making method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410136961.5A CN103914622B (en) | 2014-04-04 | 2014-04-04 | A kind of chemical leakage fast prediction alarm emergency response decision-making method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103914622A CN103914622A (en) | 2014-07-09 |
| CN103914622B true CN103914622B (en) | 2017-07-07 |
Family
ID=51040298
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410136961.5A Active CN103914622B (en) | 2014-04-04 | 2014-04-04 | A kind of chemical leakage fast prediction alarm emergency response decision-making method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103914622B (en) |
Families Citing this family (47)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104376361A (en) * | 2014-10-15 | 2015-02-25 | 南京航空航天大学 | Nuclear accident source item inversion method based on BP neural network algorithm |
| CN104346657A (en) * | 2014-11-10 | 2015-02-11 | 张江华 | Optimal evacuation method and system for dangerous chemical leakage |
| CN104680025B (en) * | 2015-03-12 | 2017-08-25 | 重庆科技学院 | Oil pumper parameter optimization method based on genetic algorithm extreme learning machine |
| CN104680023B (en) * | 2015-03-12 | 2017-08-29 | 重庆科技学院 | Oil pumper parameter optimization method based on multiobjective decision-making |
| CN104792935B (en) * | 2015-03-20 | 2017-03-15 | 北京华泰诺安探测技术有限公司 | One seed nucleus biochemical threat cloud analysis system |
| CN104792468B (en) * | 2015-04-07 | 2016-04-06 | 中国石油大学(华东) | A kind of oil refining apparatus gas detecting and alarming instrument optimization placement method |
| CN104950708B (en) * | 2015-06-02 | 2018-04-13 | 大连市环境监测中心 | A kind of air environment monitoring terminal and Environmental emergency monitoring deployment system |
| CN104880494A (en) * | 2015-06-02 | 2015-09-02 | 大连市环境监测中心 | Environmental monitoring terminal and quick deployment system of environmental emergency monitoring |
| CN106267620B (en) * | 2015-06-10 | 2019-06-28 | 张国基 | Chemical leakage handles vehicle |
| US20190066479A1 (en) * | 2015-11-19 | 2019-02-28 | Halliburton Energy Services, Inc. | Method and system for monitoring and predicting gas leak |
| CN106770930B (en) * | 2016-11-24 | 2019-04-16 | 北京蓝色星语科技有限公司 | A kind of method and system carrying out hazardous material detection using unmanned plane |
| CN106644284B (en) * | 2016-12-16 | 2019-07-12 | 北京伟瑞迪科技有限公司 | Automatic leakage monitoring and repairing system for chemical plant device |
| US10663367B2 (en) | 2017-07-28 | 2020-05-26 | International Business Machines Corporation | Adaptive sensing for a gas leak detection |
| CN107607671A (en) * | 2017-08-24 | 2018-01-19 | 慧感(上海)物联网科技有限公司 | A kind of gas leakage source localization method |
| CN108091111B (en) * | 2017-11-20 | 2018-11-13 | 淮阴工学院 | A kind of oil truck oil and gas leakage intelligent early-warning system |
| CN108596364B (en) * | 2018-03-29 | 2021-09-21 | 杭州电子科技大学 | Dynamic early warning method for major hazard source in chemical industry park |
| CN108733881B (en) * | 2018-04-13 | 2021-10-12 | 同济大学 | Treatment method for dealing with sudden indoor air pollution of public building |
| CN108733893A (en) * | 2018-04-25 | 2018-11-02 | 同济大学 | The public building burst pollution of coupling depth learning method is traced to the source |
| CN108932379A (en) * | 2018-06-26 | 2018-12-04 | 山东睿益环境技术有限公司 | A kind of evacuating personnel emergence treating method and its application |
| CN110874807A (en) * | 2018-09-03 | 2020-03-10 | 中国石油化工股份有限公司 | Inhalation poison leakage emergency response disposal card and use method and application thereof |
| CN109324155B (en) * | 2018-10-31 | 2022-03-01 | 中国石油天然气股份有限公司 | Refining device tail gas pollutant on-line monitoring data early warning method and device |
| CN109737309B (en) * | 2019-01-09 | 2021-05-04 | 慧感(上海)物联网科技有限公司 | Risk source and leakage tracing method and system based on risk identification |
| CN110046834B (en) * | 2019-05-13 | 2022-06-24 | 苏州大学 | Workplace harmful gas health risk quantitative evaluation system and evaluation method thereof |
| CN110531030B (en) * | 2019-08-16 | 2021-11-12 | 北京慧辰资道资讯股份有限公司 | Method and device for analyzing atmospheric pollutants based on Internet of things big data |
| CN110472359B (en) * | 2019-08-21 | 2023-06-23 | 东北大学 | High-temperature molten aluminum liquid and water contact explosion prevention and early warning method based on numerical simulation technology |
| CN110674983A (en) * | 2019-09-05 | 2020-01-10 | 辽宁工程技术大学 | Working face gas early warning method based on copula function tail correlation analysis |
| EP4031894B1 (en) * | 2019-09-18 | 2023-06-07 | Bayer Aktiengesellschaft | Predicting mri images by prediction model trained by supervised learning |
| CN111105086A (en) * | 2019-12-16 | 2020-05-05 | 茂盟（上海）工程技术股份有限公司 | Visual layout system and method for cooling, heating and power load prediction |
| CN111144055B (en) * | 2019-12-27 | 2024-02-02 | 苏州大学 | Method, device and medium for determining concentration distribution of toxic heavy gas leakage in urban environment |
| CN111157680B (en) * | 2019-12-31 | 2021-10-26 | 北京辰安科技股份有限公司 | Indoor volatile substance leakage tracing method and device |
| CN111127279B (en) * | 2019-12-31 | 2023-07-21 | 中国科学院合肥物质科学研究院 | Nuclear emergency decision system, method and storage medium based on monitoring experiment database |
| CN111583067A (en) * | 2020-03-31 | 2020-08-25 | 西南交通大学 | Urban underground large space construction safety early warning and emergency decision-making method and system |
| CN112330082B (en) * | 2020-08-13 | 2023-08-15 | 青岛科技大学 | Intelligent quantitative risk assessment method based on dynamic mechanism model |
| EP4195091A4 (en) * | 2020-09-11 | 2024-04-03 | Siemens Ltd China | Target available model-based environment prediction method and apparatus, program, and electronic device |
| CN112270502B (en) * | 2020-11-17 | 2021-06-01 | 北京三维天地科技股份有限公司 | Environment emergency task cooperative disposal platform based on artificial intelligence technology |
| WO2022126544A1 (en) * | 2020-12-17 | 2022-06-23 | 西门子（中国）有限公司 | Method and devices for determining pollution source, and computer-readable storage medium |
| CN113095552B (en) * | 2021-03-29 | 2023-12-26 | 中国海洋石油集团有限公司 | Ocean platform leakage combustible gas cloud cluster volume prediction method and system |
| CN113468815A (en) * | 2021-07-12 | 2021-10-01 | 南京航空航天大学 | Liquid hazardous chemical substance volatilization concentration estimation method based on Gaussian diffusion model and deep neural network |
| CN113624921B (en) * | 2021-08-04 | 2024-03-29 | 杭州谱育科技发展有限公司 | Multi-mode comprehensive pollution tracing method |
| CN114023043A (en) * | 2021-10-13 | 2022-02-08 | 厦门理工学院 | Safety early warning method for leaked gas |
| CN114001867A (en) * | 2021-10-19 | 2022-02-01 | 北京伟瑞迪科技有限公司 | Monitoring and coping method for gas leakage of park |
| CN114241723A (en) * | 2021-12-30 | 2022-03-25 | 武汉市三藏科技有限责任公司 | Environmental monitoring alarm device and method based on Gaussian atmospheric diffusion |
| CN115019482B (en) * | 2022-08-04 | 2022-11-11 | 武汉鼎业环保工程技术有限公司 | Dangerous gas itinerant early warning method and device based on big data |
| CN115600643B (en) * | 2022-10-17 | 2023-06-09 | 中国科学技术大学 | Method and system for rapidly predicting toxic gas |
| CN116403165B (en) * | 2023-06-05 | 2023-08-04 | 交通运输部水运科学研究所 | Dangerous chemical leakage emergency treatment method, dangerous chemical leakage emergency treatment device and readable storage medium |
| CN117290639B (en) * | 2023-11-24 | 2024-02-27 | 深圳大学 | Method and system for positioning and detecting material leakage in phase-change energy storage tank |
| CN117789422A (en) * | 2024-02-26 | 2024-03-29 | 江西依爱弘泰消防安全技术有限公司 | Combustible gas alarm control system and method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2317311A2 (en) * | 2009-09-11 | 2011-05-04 | SELEX Galileo Ltd | Sensing network and method |
| CN102313793A (en) * | 2011-09-07 | 2012-01-11 | 中国环境科学研究院 | Monitoring system of multi-component gas pollutants in air of chemical industrial park |
-
2014
- 2014-04-04 CN CN201410136961.5A patent/CN103914622B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2317311A2 (en) * | 2009-09-11 | 2011-05-04 | SELEX Galileo Ltd | Sensing network and method |
| CN102313793A (en) * | 2011-09-07 | 2012-01-11 | 中国环境科学研究院 | Monitoring system of multi-component gas pollutants in air of chemical industrial park |
Non-Patent Citations (2)
| Title |
|---|
| 工业园区事故风险评价研究;孙爱军;《http://wenku.baidu.com/view/e6e7bcf1551810a6f4248600.html?from=search》;20130724;第36-53页 * |
| 翟卓韬等.基于MapX的泄露事故救援决策系统.《计算机与应用化学》.2007,第24卷(第4期),第498-502页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103914622A (en) | 2014-07-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103914622B (en) | A kind of chemical leakage fast prediction alarm emergency response decision-making method | |
| Moen et al. | Comparison of k-ε models in gaseous release and dispersion simulations using the CFD code FLACS | |
| CN102314748B (en) | Poisonous gas leakage source-locating method based on wireless sensor network | |
| CN103196038B (en) | Gas ductwork source of leaks real-time positioning analyzes method and system | |
| KR101885350B1 (en) | System for integrated managment of air pollution | |
| CN102567808B (en) | Method for forecasting and warning accident consequence of major hazard installation by combining with real-time meteorological information | |
| CN102967689B (en) | A kind of Pollution Source Monitoring points distributing method | |
| KR101956188B1 (en) | Method for providing of analyzing information about hazardous chemicals dispersion | |
| CN104750949B (en) | A kind of poisoning individual risk's qualitative assessment of Leakage Gas and characterizing method | |
| Seo et al. | Deep neural network-based optimization framework for safety evacuation route during toxic gas leak incidents | |
| RU2680652C2 (en) | Method of integrated environmental monitoring | |
| Mousavi et al. | A sensitivity analysis of parameters affecting the hydrogen release and dispersion using ANOVA method | |
| CN112633553B (en) | Gas pipeline-hazardous chemical enterprise coupling hidden danger identification and risk assessment method and system | |
| Jeon et al. | Development of surrogate model using CFD and deep neural networks to optimize gas detector layout | |
| KR20220058673A (en) | System for producing 3-dimensional diffusion prediction map based on cyber physical for monitoring air pollution gas diffusion of environmental facilities, and method for the same | |
| Efthimiou et al. | CFD-RANS prediction of the dispersion of a hazardous airborne material released during a real accident in an industrial environment | |
| Rad et al. | A risk-based methodology for optimum placement of flammable gas detectors within open process plants | |
| Kopka et al. | Framework for stochastic identification of atmospheric contamination source in an urban area | |
| CN105510535B (en) | Chemical industrial park gas sensor Sector Optimization dispositions method based on field experiment | |
| Azari et al. | Extracting spatial patterns of urban gas pipeline risk considering social and structural parameters of urban blocks | |
| Sengupta et al. | Challenges for introducing risk assessment into land use planning decisions in an Indian context | |
| KR101919993B1 (en) | The preventing method of the spread of risk in the industrial complexes | |
| KR20170040908A (en) | Probabilistic gas explosion scenario calculation system and probabilistic gas explosion scenario calculation method using the same | |
| Гаврись et al. | Software-based method of determining the necessary population evacuation zone in case of a chemical accident | |
| CN115906706A (en) | High-precision risk analysis method for domino comprehensive risk of hazardous chemical tank area |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |