CN110083918A - A kind of vehicle-mounted part is short to face early warning dissemination method - Google Patents
A kind of vehicle-mounted part is short to face early warning dissemination method Download PDFInfo
- Publication number
- CN110083918A CN110083918A CN201910323863.5A CN201910323863A CN110083918A CN 110083918 A CN110083918 A CN 110083918A CN 201910323863 A CN201910323863 A CN 201910323863A CN 110083918 A CN110083918 A CN 110083918A
- Authority
- CN
- China
- Prior art keywords
- data
- gate layer
- early warning
- prediction
- vehicle
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000012549 training Methods 0.000 claims abstract description 9
- 238000007689 inspection Methods 0.000 claims description 8
- 238000004458 analytical method Methods 0.000 claims description 6
- 238000004891 communication Methods 0.000 claims description 5
- 230000008447 perception Effects 0.000 claims description 5
- 238000001556 precipitation Methods 0.000 claims description 5
- 230000001174 ascending effect Effects 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000003908 quality control method Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 2
- 238000012544 monitoring process Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 abstract description 5
- 238000010586 diagram Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 230000003993 interaction Effects 0.000 description 3
- 238000001514 detection method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000000717 retained effect Effects 0.000 description 2
- 238000012876 topography Methods 0.000 description 2
- 208000036119 Frailty Diseases 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 206010003549 asthenia Diseases 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008094 contradictory effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000004393 prognosis Methods 0.000 description 1
- 230000003134 recirculating effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01W—METEOROLOGY
- G01W1/00—Meteorology
- G01W1/10—Devices for predicting weather conditions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Tourism & Hospitality (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Life Sciences & Earth Sciences (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Environmental & Geological Engineering (AREA)
- Biomedical Technology (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Development Economics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Marketing (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- General Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Quality & Reliability (AREA)
- Ecology (AREA)
- Biodiversity & Conservation Biology (AREA)
- Atmospheric Sciences (AREA)
- Educational Administration (AREA)
- Primary Health Care (AREA)
- Environmental Sciences (AREA)
- Operations Research (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Face early warning dissemination method the invention discloses a kind of vehicle-mounted part is short, region to be predicted is averagely divided into several zonules by this method, each zonule center is arranged one equipped with the small workstation Mobile Meteorological Station based on LSTM network, site observation date is obtained by Mobile Meteorological Station, small workstation pre-processes conception of history measured data and site observation date, according to pretreated conception of history measured data training prediction model, pretreated site observation date is analyzed using prediction model, obtains the prediction data of some areas future time instance.Multi-zone supervision and control may be implemented in the present invention, intuitive easy-to-use；Solving the problems, such as data acquisition, there are error and data itself bulk redundancies, realize assessment fining, improve the accuracy of early warning.
Description
Technical field
The present invention relates to a kind of Predictive meteorological methods more particularly to a kind of vehicle-mounted part is short faces early warning dissemination method.
Background technique
Chinese population is numerous, climate variability, with a varied topography, ecological environment frailty, is natural calamity most serious in the world
One of country, multiple natural calamity repeatedly test the public safety system and social security ability of entire society.Meteorological calamity
The publication of evil warning information is prevented and reduced natural disasters the important component and element task of work as government, has been caused widely
Pay attention to.Currently, as an effective meteorological disaster warning information dissemination method disaster information can be issued in time, and be
It is most important that disaster management decision provides accurate assessment result.
Meteorological disaster has the characteristics that type is more, range is wide, sudden strong, brings to the early warning of meteorological disaster huge
Problem and challenge.The time that meteorological disaster occurs is not fixed, if it is pre- to rely solely on Long-range Numerical Weather Prediction model progress weather
It surveys, can not predict the weather condition of area within a short period of time accurately and in real time, timeliness is difficult to ensure；Meanwhile it is special
Geographical location and topography and geomorphology make diastrous weather sea land difference obvious, climate change but also the distribution of diastrous weather not
Uniform characteristics are more obvious, if issuing warning information with the forecast result of whole region, it will cause the empty reports in certain small areas
Or fail to report, accuracy is difficult to ensure.
Summary of the invention
Goal of the invention: the object of the present invention is to provide the sides that one kind can predict some areas weather prognosis accurately and in real time
Method provides strong support for live emergency processing and commanding in the rear scheduling.
Technical solution: to achieve the purpose of the present invention, a kind of vehicle-mounted part of the present invention is short to face early warning dissemination method,
The following steps are included:
(1) region to be predicted being averagely divided into several zonules, a Mobile Meteorological Station is arranged in each zonule center,
The Mobile Meteorological Station is equipped with the small workstation based on LSTM network.
Further, the step (1) specifically includes:
(11) it selectes grid system type: region to be predicted is averagely divided into multiple zonules:
Further, the generally selected regular polygon of the grid system type, is broadly divided into following a few classes: equilateral triangle,
Regular quadrangle, regular hexagon.
Preferably, the grid system type is chosen to be regular hexagon, i.e. grid unit shape is regular hexagon, therefore to
Survey region, which will be averaged, is divided into several regular hexagon zonules.
(12) it establishes and expands grid network coordinate system:
(121) using first movement meteorological observatory present position as origin, the horizontal direction of map is X-axis, and vertical direction is Y-axis
The rectangular coordinate system of auxiliary is established, first movement meteorological observatory present position is used to mark the center in initial cell domain,
Distribution is adjoined in other zonules；
(122) in rectangular coordinate system, note grid unit is positive the side n shape, and the circumradius for defining grid unit is R,
In n+1 node of any cell of grid coordinate (i, j), central node is identified with (i, j), and n boundary node is from upper right
Angle successively identifies counterclockwise.
(13) effective control boundary condition of grid is determined:
Further, the boundary condition are as follows:
Wherein, R is zonule grid circumradius, Rs, RcRespectively the perception radius of Mobile Meteorological Station sensing node and
Communication radius.
(14) different grid unit node coordinates are calculated；
(15) area information after dividing is shown on map, including zone number, disaster, disaster of prediction etc.
Grade.
(2) Mobile Meteorological Station obtains site observation date, by small workstation to conception of history measured data and field observation
Data are pre-processed.
Further, the data include to area locating information, meteorologic parameter, environment distributed intelligence；Wherein, the ring
Border distributed intelligence includes river in the region, mountainous region information.
Further, the pretreatment for using consistency analysis method to conception of history measured data and site observation date into
The control of row quality, including internal consistency monitoring, time consistency inspection, duration inspection and Horizontal consistency calculate.
Further, the Horizontal consistency is calculated as, and to same element, calculates it a certain to pre- according to following formula
Whether the standard observation value for surveying element in region is normal with the observation for judging the element:
Tt=(Xt-Qt,2/4)/(Qt,3/4-Qt,1/4)
Wherein, TtFor the standard observation value of element, XtFor the observation of a certain element of t moment, Qt,1/4、Qt,2/4And Qt,3/4For
Neighbouring Mobile Meteorological Station t moment is to element observation by the quartile value after ascending converse sequencing；Work as TtMore than statistics threshold
When value, then the X is removedtData.
(3) small workstation is according to pretreated conception of history measured data training prediction model, using prediction model to pre-
Treated, and site observation date is analyzed, and obtains the prediction data of some areas future time instance.
Further, the trained prediction model is building LSTM recirculating network framework, and the network architecture includes one
For carrying out pretreated input layer, several processors for stacking connection and a resultant layer for output to input data；
Each processor includes input gate, forgets door, out gate and tanh layers.
Step (3) specifically includes the following steps:
(31) in the input layer of LSTM model, data are pre-processed according to the following formula, it is excellent to obtain model prediction
Bad judgment basis:
xin=(fis,fid,fit)
yout=(fos,fod,fot)
ye=(fes,fed,fet)
Wherein, xin、yout、yeRespectively conception of history measured data, LSTM model output data, expected data, every group of data
Vector element is corresponding in turn to the precipitation, humidity and temperature value of synchronization from left to right.
Further, the conception of history measured data is the data at the history observation moment inputted when training, is relative to the phase
Hope what data were discussed.Subsequent time is confirmed behind given same time interval, the observation data of the subsequent time are immediate
Hope data.
Further, the LSTM model output data is the data for the subsequent time predicted according to conception of history measured data.
(32) in the forgetting gate layer of the alternation of bed of LSTM model, data are screened.It can be according to LSTM network
The characteristics of information is transmitted between combination, filtering and the small information of prediction result relevance, retain and are associated with big, more important information.
It by network training, is optimized using weight parameter, keeps prediction result more reasonable.
Output information predicts weight gt, shown in following formula:
gt=σ [wf(ht-1, xt)+bf],
Wherein, gtValue be 0~1 between, 0 is expressed as giving up the information completely, 1 indicate the information is fully retained,
In, t is current time, ht-1The output of gate layer, x are exported for last momenttThe input of gate layer, w are forgotten for current timefAnd bfFor
Forget gate layer respective weights and biasing；
(33) in the input gate layer of the alternation of bed of LSTM model, data reservation and update are carried out according to the following formula:
it=σ [wi(ht-1, xt)+bi]
Wherein, t is current time, ht-1The output of gate layer, x are exported for last momenttIt is defeated that gate layer is forgotten for current time
Enter, wiAnd biFor input gate layer respective weights and biasing；
(34) in the tanh layer of the alternation of bed of LSTM model, processing can be added to by creating one according to the following formula
New candidate value in device state:
Wherein, the processor is three layers of alternation of bed, and t is current time, ht-1The output of gate layer is exported for last moment,
xtGate layer input, w are forgotten for current timecAnd bcFor respective weights and biasing, function tanh standardizes numerical value to -1 to 1 area
Between；
(35) according to the following formula, the information prediction weight g exported according to step (32)tAnd the i in step (33)t, will take
Old processor state C with recall infot-1With the candidate value comprising new informationIn conjunction with more new processor:
(36) in the output gate layer of LSTM model, LSTM is generated based on updated processor state according to the following formula
Output valve:
ht=σ [Wo(ht-1,xt)+bo]tanhCt
Wherein, ht-1The output of gate layer, x are exported for last momenttGate layer input, w are forgotten for current timeoAnd boFor output
Gate layer respective weights and biasing；
(37) the output result of LSTM model is subjected to number according to time series it was predicted that and comparing expected data, and benefit
It realizes that network parameter updates with BPTT algorithm, obtains prediction model；Wherein, the parameter includes weight, biasing；
(38) by site observation date input prediction model, the grade of prediction data is exported, and corresponding small on map
Disaster loss grade is shown in region.
The utility model has the advantages that the invention has the following advantages that
1, space refines, and region division is carried out to weather warning some areas, using vehicle-mounted small workstation to collection
A variety of data for arriving carry out short forecasting, compensate for the low disadvantage of backstage long-term forecasting precision, improve accuracy of the forecast and
It is effective；
2, high speed is analyzed, is guarantee with quality control and consistency analysis, is become using the time and space in meteorological element
The rule connected each other between law and each element is that clue is analyzed, and includes satellite data, mobile meteorology to what is got
Conception of history measured data including Taiwan investment material or Beidou meteorology individual soldier etc. is pre-processed with site observation date, and it is superfluous to reduce data
It is remaining, the quality of data is ensured to greatly improve the speed of subsequent data analysis；
3, the time refines, and is predicted using shot and long term memory network (LSTM) the grade of local disaster, from sequence
The characteristic of information is extracted in data, while remaining the information with long-range dependence in step farther out, is solved nerve and is followed
The problems such as gradient disperse or gradient that loop network is generated because Memory windows are too long in practical applications are exploded, improves prediction
Temporal resolution and precision.
Detailed description of the invention
Fig. 1 is flow chart of the present invention；
Fig. 2 is the execution flow diagram of region division of the present invention；
Fig. 3 is the structural schematic diagram of LSTM model of the present invention；
Fig. 4 is three schematic diagram of the embodiment of the present invention.
Specific embodiment
Further description of the technical solution of the present invention with reference to the accompanying drawings and examples.
Embodiment one
Referring to Figure 1, it illustrates a kind of short method flow diagram for facing early warning dissemination method in vehicle-mounted part, this method includes
Following steps:
(1) region to be predicted being averagely divided into several zonules, a Mobile Meteorological Station is arranged in each zonule center,
The Mobile Meteorological Station is equipped with the small workstation based on LSTM network.
Fig. 2 is referred to, it illustrates the execution flow diagrams of region division of the present invention.It is excessive for early warning range, in advance
Inaccurate problem is reported, Mobile Meteorological Station is set in each street and onboard carries small workstation, big region is put down
It is divided into multiple zonules, comprising the following steps:
(11) selecting grid system type is equilateral triangle；
(12) using first movement meteorological observatory present position as origin, the horizontal direction of map is X-axis, and vertical direction is Y-axis
The rectangular coordinate system of auxiliary is established, first movement meteorological observatory present position is used to mark the center in initial cell domain；
(13) in rectangular coordinate system, the circumradius for defining grid unit is R, in any of grid coordinate (i, j)
In n+1 node of unit, (i, j) is center node,
A (i, R+j),Respectively 3 boundary points of equilateral triangle；
(14) effective control boundary condition of grid is determined according to the following formula:
Wherein, R is zonule grid circumradius, Rs, RcRespectively the perception radius of Mobile Meteorological Station sensing node and
Communication radius；
(15) different grid unit node coordinates are calculated；
(16) after determining first equilateral triangle zonule, " simple unit duplication " principle is utilized, is successively replicated identical
Equilateral triangle, so that it is adjoined distribution；
(17) area information after dividing is shown on map, including zone number, disaster, disaster of prediction etc.
Grade.
(2) Mobile Meteorological Station obtains site observation date, by small workstation to conception of history measured data and field observation
Data are pre-processed, and the data include area locating information to be predicted, meteorologic parameter, environment distributed intelligence；Wherein, described
Environment distributed intelligence includes river in the region, mountainous region information.
It include satellite data to what is got, the field observation number including Mobile Meteorological Station data or Beidou meteorology individual soldier etc.
Accordingly and historical data carries out quality control, judges whether meteorological data is reasonable according to consistency analysis method.Foundation meteorology,
Synoptic meteorology, climatology principle, in meteorological element time and space changing rule and each element between connect each other rule for line
Suo Jinhang analysis.The consistency analysis method includes inside detection consistency, time consistency, duration consistency and water
The inspection methods such as flat consistency, specific effect are as follows:
1) internal consistency monitors.According to meteorology principle, to physics spies certain in observational data to associated element
Between whether meet the inspection of certain rule.Rudimentary algorithm is the meteorology correlation based on two observation elements, checks whether and deposits
In contradictory phenomena.Such as wind direction and wind velocity consistency check, temperature and moisture consistency check etc..
2) time consistency inspection.It is corresponding for judging observational data in time, it is therefore an objective to detect observational data
Time rate of change identifies undesirable time rate of change.Such as provide that atmospheric pressure value difference is no more than in 6 hours in observation area
18hpa。
3) duration inspection.Within a certain period of time, many meteorological elements are observed at any time, the wave that geographical diversity occurs
It is dynamic, observation instrument is corresponded to if certain meteorological element is there is no variation or transmission device breaks down.For example, wind direction waves direction
Always it fixes, then can determine whether that machine breaks down.
4) Horizontal consistency calculates.Have the characteristics that continuity and uniformity using meteorological element NATURAL DISTRIBUTION, it will be a certain
Detected observational data with around it other wash adjacent to observation station data Jin row Bi compare Alto, to judge the element whether just
Often, specific formula for calculation is as follows:
Tt=(Xt-Qt,2/4)/(Qt,3/4-Qt,1/4)
Wherein, TtFor the standard observation value of element, XtFor the observation of a certain element of t moment, Qt,1/4、Qt,2/4And Qt,3/4For
Neighbouring Mobile Meteorological Station t moment is to element observation by the quartile value after ascending converse sequencing；Work as TtMore than statistics threshold
When value, then the X is removedtData.
It is different for rainfall hourly, rainfall is divided into 12 grades in advance, rainfall is 0~20 per hour
It is level-one when milliliter, is second level at 20~40 milliliters, 40~60 milliliters of whens are three-level, and 60~80 milliliters of whens are level Four, 80~100
It is Pyatyi when milliliter, 100~120 milliliters of whens are eight grades when being seven grades, 140~180 milliliters when being six grades, 120~140 milliliters,
It is ten grades when being nine grades, 200~220 milliliters at 180~200 milliliters, 220~240 milliliters of Shi Weishi level-ones, 240~260 milliliters
Shi Weishi second level.And according to four warning lines of rainfall grade classification, including cordon bleu line (1-3 grades), yellow early warning line (3-
5 grades), orange warning line (5-10 grades) and red early warning line (10-12 grades).
(3) small workstation is according to pretreated conception of history measured data training prediction model, using prediction model to pre-
Treated, and site observation date is analyzed, and obtains the prediction data at certain following moment of some areas.
Fig. 3 is referred to, it illustrates LSTM model structure schematic diagrames.A is a processor, including input gate in figure
Layer forgets gate layer, exports gate layer and tanh layers, and many processor combination stackeds constitute LSTM network.LSTM network struction
Journey can be divided mainly into three parts: input layer data prediction, and interaction layer data calculates, the output of output layer result.
(31) input layer data prediction:
By taking heavy rain as an example, input data selected by the present invention includes precipitation, humidity and temperature.Firstly, to precipitation and
Humidity data is sampled, sampling interval 5s, and number of samples is 200, and is stated in vector form, temperature then root
Depending on concrete condition, fixing tentatively is 25 degree.Data of the present invention can be divided into acquisition data, and (history acquires data and collection in worksite number
According to), expected data and LSTM model output data, wherein history acquisition data and expected data realize to the training of network and
Optimization, collection in worksite data realize the application that network is completed to training, to predict rain-level.LSTM model output data is then
For experimental result as algorithm superiority and inferiority judgment basis, all data is as follows:
xin=(fis,fid,fit)
yout=(fos,fod,fot)
ye=(fes,fed,fet)
Wherein, xin, yout, yeIt respectively corresponds input data and acquires data, LSTM model output data, expected data,
Vector is respectively precipitation, the numerical value of humidity and temperature corresponding to different moments in bracket.
(32) interaction layer data calculates:
Interaction layer data calculating is the parameter training process to 3 gate layer and one tanh layers, three-level LSTM neural network forecast
The nuclear structure of model.
1) forget gate layer, i.e. first alternation of bed, it determines which information current procedures should give up.This layer it is defeated
G outtBe a value be 0~1 between number, it determines which needs processor state has in previous step treatment process
It continues to transmit, formula is as follows:
gt=σ [wf(ht-1, xt)+bf]
Wherein, ht-1The output of gate layer, x are exported for last momenttGate layer input, w are forgotten for current timefAnd bfFor correspondence
Weight and biasing.
2) gate layer, i.e. second alternation of bed are inputted, which determines which information should be added into processor state.This
One layer of output determines which information will be retained and update in tanh layers of third alternation of bed, and formula is as follows:
it=σ [wi(ht-1, xt)+bi]
3) tanh layers, i.e. third alternation of bed, effect be creation one can be added to it is new in processor state
Candidate value, formula are as follows:
After calculating by 3 layers, the old processor state C of recall info is carriedt-1With the candidate value comprising new informationKnot
It closes, specific formula is as follows:
4) gate layer is exported, for generating the output valve of LSTM, formula institute specific as follows based on updated processor state
Show:
ht=σ [Wo(ht-1,xt)+bo]tanhCt
(33) by the output result of LSTM model according to time series, number is carried out it was predicted that and comparing expected data, and benefit
It realizes that network parameter updates with BPTT algorithm, obtains prediction model；Wherein, the parameter includes weight, biasing；
(34) by site observation date input prediction model, the grade of prediction data is exported, and corresponding small on map
Disaster loss grade is shown in region.
Embodiment two:
A kind of vehicle-mounted part provided in this embodiment is short to face early warning dissemination method, and this method removes the concrete operations of step (1)
Different outer, other steps are identical as the method and step provided in embodiment one, therefore only different piece is described.
(11) selecting grid system type is regular quadrangle；
(12) using first movement meteorological observatory present position as origin, the horizontal direction of map is X-axis, and vertical direction is Y-axis
The rectangular coordinate system of auxiliary is established, first movement meteorological observatory present position is used to mark the center in initial cell domain；
(13) in rectangular coordinate system, the circumradius for defining grid unit is R, in any of grid coordinate (i, j)
In n+1 node of unit, (i, j) is center node, and 4 boundary points of regular quadrangle are respectively as follows:
(14) effective control boundary condition of grid is determined according to the following formula:
Wherein, R is zonule grid circumradius, Rs, RcRespectively the perception radius of Mobile Meteorological Station sensing node and
Communication radius；
(15) different grid unit node coordinates are calculated；
(16) after determining first regular quadrangle zonule, " simple unit duplication " principle is utilized, is successively replicated identical
Regular quadrangle, so that it is adjoined distribution；
(17) area information after dividing is shown on map, including zone number, disaster, disaster of prediction etc.
Grade.
Embodiment three:
A kind of vehicle-mounted part provided in this embodiment is short to face early warning dissemination method, and this method removes the concrete operations of step (1)
Different outer, other steps are identical as the method and step provided in embodiment one, therefore only different piece is described.Such as figure
Shown in 4, the region division process that is provided using embodiment three, it can be achieved that detection zone to repeat area coverage minimum, therefore conduct
The preferred steps of present invention execution process.
(11) selecting grid system type is regular hexagon；
(12) using first movement meteorological observatory present position as origin, the horizontal direction of map is X-axis, and vertical direction is Y-axis
The rectangular coordinate system of auxiliary is established, first movement meteorological observatory present position is used to mark the center in initial cell domain,
Distribution is adjoined in other zonules；
(13) in rectangular coordinate system, the circumradius for defining grid unit is R, in any of grid coordinate (i, j)
In n+1 node of unit, (i, j) is center node, and 6 boundary points of regular hexagon are respectively as follows:
(14) effective control boundary condition of grid is determined according to the following formula:
Wherein, R is zonule grid circumradius, Rs, RcRespectively the perception radius of Mobile Meteorological Station sensing node and
Communication radius；
(15) different grid unit node coordinates are calculated；
(16) after determining first regular hexagon zonule, " simple unit duplication " principle is utilized, is successively replicated identical
Regular hexagon, so that it is adjoined distribution；
(17) area information after dividing is shown on map, including zone number, disaster, disaster of prediction etc.
Grade.
Claims (9)
1. a kind of vehicle-mounted part is short to face early warning dissemination method, which comprises the following steps:
(1) region to be predicted is averagely divided into several zonules, a Mobile Meteorological Station is arranged in each zonule center, described
Mobile Meteorological Station is equipped with the small workstation based on LSTM network；
(2) Mobile Meteorological Station obtains site observation date, by small workstation to conception of history measured data and site observation date
It is pre-processed；
(3) small workstation is according to pretreated conception of history measured data training prediction model, using prediction model to pretreatment
Site observation date afterwards is analyzed, and the prediction data of some areas future time instance is obtained.
2. vehicle-mounted part according to claim 1 is short to face early warning dissemination method, it is characterised in that: the data include wanting
Element: area locating information, meteorologic parameter, environment distributed intelligence；Wherein, the environment distributed intelligence include river in region,
Mountainous region information.
3. vehicle-mounted part according to claim 1 is short to face early warning dissemination method, which is characterized in that the step (1) is specific
Include:
(11) grid system type is selected, region to be predicted is averagely divided into multiple zonules；
(12) it establishes and expands grid coordinate system；
(13) effective control boundary condition of grid is determined；
(14) different grid unit node coordinates are calculated；
(15) shown on map divide after area information, including zone number, disaster, prediction disaster loss grade.
4. vehicle-mounted part according to claim 1 or 3 is short to face early warning dissemination method, it is characterised in that: the zonule
Shape is regular hexagon.
5. vehicle-mounted part according to claim 3 is short to face early warning dissemination method, which is characterized in that the step (12) is specific
Include:
(121) using grid coordinate system origin as first movement meteorological observatory present position, the horizontal direction of map is X-axis, Vertical Square
To the rectangular coordinate system for establishing auxiliary for Y-axis, first movement meteorological observatory present position is for marking in the domain of initial cell
Distribution is adjoined in heart position, other zonules；
(122) in rectangular coordinate system, note grid unit is positive the side n shape, and the circumradius for defining grid unit is R, in lattice
In n+1 node of any cell of net coordinate (i, j), central node is identified with (i, j), and n boundary node is pressed from the upper right corner
Counterclockwise successively identify.
6. vehicle-mounted part according to claim 3 is short to face early warning dissemination method, which is characterized in that described in step (13)
Boundary condition are as follows:
Wherein, R is zonule grid circumradius, Rs、RcThe respectively the perception radius and communication of Mobile Meteorological Station sensing node
Radius.
7. vehicle-mounted part according to claim 1 is short to face early warning dissemination method, which is characterized in that described pre- in step (2)
Processing are as follows: quality control, including inside one are carried out to conception of history measured data and site observation date using consistency analysis method
The monitoring of cause property, time consistency inspection, duration inspection and Horizontal consistency calculate.
8. vehicle-mounted part according to claim 7 is short to face early warning dissemination method, which is characterized in that the Horizontal consistency meter
It calculates are as follows: to same element, calculate the standard observation value of its element in a certain region to be predicted according to following formula to judge this
Whether the observation of element is normal:
Tt=(Xt-Qt,2/4)/(Qt,3/4-Qt,1/4)
Wherein, TtFor the standard observation value of element, XtFor the observation of a certain element of t moment, Qt,1/4、Qt,2/4And Qt,3/4It is neighbouring
Mobile Meteorological Station t moment is to element observation by the quartile value after ascending converse sequencing；Work as TtWhen more than statistical threshold,
Then remove the XtData.
9. vehicle-mounted part according to claim 1 is short to face early warning dissemination method, which is characterized in that step (3) specifically includes
Following steps:
(31) in the input layer of LSTM model, data are pre-processed according to the following formula, obtain model prediction superiority and inferiority
Judgment basis:
xin=(fis,fid,fit)
yout=(fos,fod,fot)
ye=(fes,fed,fet)
Wherein, xin、yout、yeRespectively conception of history measured data, LSTM model output data, expected data, the vector of every group of data
Element is corresponding in turn to the precipitation, humidity and temperature value of synchronization from left to right；According to the LSTM model output data
The data of the subsequent time of conception of history measured data prediction；The expected data is the observation data of subsequent time；
(32) in the forgetting gate layer of the alternation of bed of LSTM model, data are screened according to the following formula, output information is pre-
Survey weight gt, gt=σ [wf(ht-1, xt)+bf], value is between 0~1, and 0 is expressed as giving up the information completely, and 1 indicates complete
Retain the information, wherein t is current time, ht-1The output of gate layer, x are exported for last momenttGate layer is forgotten for current time
Input, wfAnd bfTo forget gate layer respective weights and biasing；
(33) in the input gate layer of the alternation of bed of LSTM model, data reservation and update are carried out according to the following formula:
it=σ [wi(ht-1, xt)+bi]
Wherein, t is current time, ht-1The output of gate layer, x are exported for last momenttGate layer input, w are forgotten for current timeiWith
biFor input gate layer respective weights and biasing；
(34) in the tanh layer of the alternation of bed of LSTM model, processor shape can be added to by creating one according to the following formula
New candidate value in state:
Wherein, the processor includes input gate layer, forgets gate layer, exports gate layer and tanh layers, and t is current time, ht-1It is upper
The output of one moment out gate layer, xtGate layer input, w are forgotten for current timecAnd bcFor respective weights and biasing, function tanh
Numerical value is standardized to -1 to 1 section；
(35) according to the following formula, the information prediction weight g exported according to step (32)tAnd the i in step (33)t, remember carrying
Recall the old processor state C of informationt-1With the candidate value comprising new informationIn conjunction with more new processor:
(36) in the output gate layer of LSTM model, the defeated of LSTM is generated based on updated processor state according to the following formula
It is worth out:
ht=σ [Wo(ht-1,xt)+bo]tanhCt
Wherein, ht-1The output of gate layer, x are exported for last momenttGate layer input, w are forgotten for current timeoAnd boTo export gate layer
Respective weights and biasing；
(37) by the output result of LSTM model according to time series carry out number it was predicted that and compare expected data, utilize BPTT
Algorithm realizes that network parameter updates, and obtains prediction model；Wherein, the parameter includes weight, biasing；
(38) by site observation date input prediction model, the grade of prediction data is exported, and the corresponding zonule on map
Interior display disaster loss grade.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910323863.5A CN110083918A (en) | 2019-04-22 | 2019-04-22 | A kind of vehicle-mounted part is short to face early warning dissemination method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910323863.5A CN110083918A (en) | 2019-04-22 | 2019-04-22 | A kind of vehicle-mounted part is short to face early warning dissemination method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110083918A true CN110083918A (en) | 2019-08-02 |
Family
ID=67416099
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910323863.5A Pending CN110083918A (en) | 2019-04-22 | 2019-04-22 | A kind of vehicle-mounted part is short to face early warning dissemination method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110083918A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111489525A (en) * | 2020-03-30 | 2020-08-04 | 南京信息工程大学 | Multi-data fusion meteorological prediction early warning method |
| CN113486481A (en) * | 2021-06-25 | 2021-10-08 | 同济大学 | Optimized setting method of environment-aware network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105388538A (en) * | 2015-11-23 | 2016-03-09 | 四川长虹电器股份有限公司 | Miniature weather prediction device and weather prediction method |
| CN106371153A (en) * | 2016-08-16 | 2017-02-01 | 合肥东上多媒体科技有限公司 | Unattended small automatic weather station |
| CN109240264A (en) * | 2018-09-14 | 2019-01-18 | 浙江大学 | A kind of pulping process energy consumption prediction technique based on LSTM neural network |
| CN109660290A (en) * | 2018-11-23 | 2019-04-19 | 南京信息工程大学 | A kind of multi rack structure emergency security system based on more satellite multibands |
-
2019
- 2019-04-22 CN CN201910323863.5A patent/CN110083918A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105388538A (en) * | 2015-11-23 | 2016-03-09 | 四川长虹电器股份有限公司 | Miniature weather prediction device and weather prediction method |
| CN106371153A (en) * | 2016-08-16 | 2017-02-01 | 合肥东上多媒体科技有限公司 | Unattended small automatic weather station |
| CN109240264A (en) * | 2018-09-14 | 2019-01-18 | 浙江大学 | A kind of pulping process energy consumption prediction technique based on LSTM neural network |
| CN109660290A (en) * | 2018-11-23 | 2019-04-19 | 南京信息工程大学 | A kind of multi rack structure emergency security system based on more satellite multibands |
Non-Patent Citations (1)
| Title |
|---|
| 贺佳佳等: "一种多时间尺度SVM局部短时临近降雨预测方法", 《气象》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111489525A (en) * | 2020-03-30 | 2020-08-04 | 南京信息工程大学 | Multi-data fusion meteorological prediction early warning method |
| CN113486481A (en) * | 2021-06-25 | 2021-10-08 | 同济大学 | Optimized setting method of environment-aware network |
| CN113486481B (en) * | 2021-06-25 | 2022-08-09 | 同济大学 | Optimized setting method of environment-aware network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11295214B2 (en) | Analysis system and hydrology management for basin rivers | |
| CN111047099B (en) | Regional torrential flood risk prediction method and system | |
| US11335179B1 (en) | Water environment risk prediction and early warning method | |
| CN105243435B (en) | A kind of soil moisture content prediction technique based on deep learning cellular Automation Model | |
| Du et al. | Optimal sensor placement and measurement of wind for water quality studies in urban reservoirs | |
| CN104899437A (en) | Early-warning method for heavy-rainfall type landslide hazard | |
| CN111428942B (en) | Line icing thickness prediction method for extracting micro-terrain factors based on variable grid technology | |
| CN113327022B (en) | Lightning protection safety risk management system and method | |
| Huang et al. | A new air quality forecasting model using data mining and artificial neural network | |
| CN107480781A (en) | The nuclear accident Source Term Inversion method of neutral net adaptive Kalman filter | |
| CN113554345B (en) | Earthquake landslide disaster area and disaster chain risk assessment method | |
| CN110083918A (en) | A kind of vehicle-mounted part is short to face early warning dissemination method | |
| CN104200082A (en) | Typhoon landing prediction method | |
| CN111144637A (en) | Regional power grid geological disaster forecasting model construction method based on machine learning | |
| CN113011631A (en) | Earthquake-measuring emergency mobile station site selection method | |
| CN116976227B (en) | Storm water increasing forecasting method and system based on LSTM machine learning | |
| Li | Evaluation method of ecological vulnerability of scenic spots based on entropy weight TOPSIS model | |
| CN112233381B (en) | Debris flow early warning method and system based on mechanism and machine learning coupling | |
| CN116771425A (en) | Detection method and system for spatial distribution of concentration of gas in coal mine tunnel | |
| Zhang et al. | Spatial patterns and controlling factors of the evolution process of karst depressions in Guizhou province, China | |
| CN107231668A (en) | Transmission mechanism model under forest fire based on gray scale association analysis | |
| Yang et al. | Effect of observation network design on meteorological forecasts of Asian dust events | |
| CN113505948B (en) | Future ecological security pattern prediction and optimization method based on Bayesian network | |
| Burlando et al. | Wind short-term forecast in port areas | |
| Hadipour et al. | Genetic programming for downscaling extreme rainfall events |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190802 |
|
| RJ01 | Rejection of invention patent application after publication |