CN108446802A - A kind of red tide prewarning method based on graph model structure - Google Patents
A kind of red tide prewarning method based on graph model structure Download PDFInfo
- Publication number
- CN108446802A CN108446802A CN201810243914.9A CN201810243914A CN108446802A CN 108446802 A CN108446802 A CN 108446802A CN 201810243914 A CN201810243914 A CN 201810243914A CN 108446802 A CN108446802 A CN 108446802A
- Authority
- CN
- China
- Prior art keywords
- red tide
- data
- sample
- stage
- distance
- 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.)
- Granted
Links
Classifications
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
- Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- General Engineering & Computer Science (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Development Economics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Game Theory and Decision Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to a kind of red tide prewarning methods based on graph model structure, the described method comprises the following steps：Step S1, data prediction；Step S11, weight calculation；Step S12, environmental factor dimensionality reduction；Step S2 executes DWFCM clustering algorithms；Step S21, cluster centre selection improve；Step S22, Euclidean distance weighting；Step S23, object function improve；Step S3 builds red tide diagram data model.Its advantage is shown：Red tide data can be classified by stage of development, and is stored in graph model by the stage；Red tide forecast personnel can quickly judge the stage residing for red tide, and targetedly propose control measure for the current generation, reduce economy and Ecological Loss；Facilitate that red tide forecast personnel are accurate, quick search is to data are needed, provides the support in quick and precisely comprehensive data for red tide forecast personnel, carry out comparison control using present data and historical data, predict the stage of development of red tide.
Description
Technical field
The present invention relates to red tide datagram model construction techniques fields, specifically, being a kind of to be built based on graph model
Red tide prewarning method.
Background technology
One of an important factor for red tide is influence marine environment, global Disaster And Prevention Measures of Red Tides takes place frequently, and endangers to ecology and economic system
Evil is very big.The marine site of China in 2003 finds red tide 119 times altogether, about 14550 square kilometres of cumulative area, caused by direct economy damage
Lose up to 42,810,000 yuan.In recent years, with the nearly development of coastal industrial or agricultural and the increase of population, industrial wastewater and urban life are dirty
Water be continually drained it is marine, cause seawater part nutritional profile increase, promote red tide plankton to increase significantly, lead to frequent occurrence, damage
It loses increasingly severe.
There is correlation technique to be studied in the prior art, has included mainly that red tide forecast method for early warning and graph model are built
Method.
1. red tide forecast method for early warning
Existing correlation technique can study red tide monitoring data at present, and following two categories can be divided by integrating：
Traditional red tide data classification method and machine learning classification method.Traditional sorting technique is divided into empirical method, statistic law sum number
Value method.Such methods model is simple, to a certain extent can be with forecasting and warning red tide, but accuracy and precision are not high enough, nothing
Method is applied to large-scale breakout of red tide.
In recent years, with the rise of artificial intelligence, carrying out research to red tide using the correlation technique of machine learning becomes master
Stream.
Wang Xing is strong et al., and the COSA algorithms that would be used for text cluster are combined with FCM Algorithms and introduce similar pass
System's pretreatment, is being improved, and is applied in red tide monitoring field, is achieved preferable accuracy rate and real-time
Zhang Chenghui et al. is directed to the deficiency of tradition FCM algorithms, it is proposed that a kind of improved FCM algorithms using sample and gather
Class center similarity relation determines influence coefficient of each sample to cluster centre so that cluster process is rapider, and cluster result is more
Add and stablizes accurately.
Nitin Muttil profits et al. model coastal red tide using genetic algorithm, achieve good effect.
Su Xinhong et al. establishes itself and temperature, precipitation, wind speed, air pressure and day using BP neural network artificial intelligence model
Divide according to geographical location with corresponding meteorological index according to the non-linear relation of 5 meteorological factors, and by these red tide case datas
Other input model is learnt, trained and is predicted, new approach is provided for the forecast of red tide.
Above method plays the forecasting and warning of red tide certain effect, but still in place of Shortcomings.First,
The generation of red tide is influenced by various environmental factors, is difficult to differentiate between each stage, has ambiguity.Secondly, red tide monitoring number
After classification, it is not stored effectively, forecast personnel use can not be supplied directly to.
2. graph model construction method
Nowadays diagram data model is widely used in various fields, such as community network, GIS-Geographic Information System, bioinformatics
Deng.
Ou Xiao equality people propose music data model GraMM and query language GraMQL based on figure, by musical database
It is modeled as individual big figure, also song can show as a vertex on figure to a first sound, and similarity between song turns over the relationship of singing
Etc. the side that can be shown as between music track vertex, the inquiry operation of music data is to search to meet given meaning on big figure
The subgraph of entry part.
Tang Dequan[Et al. propose based on the crime law study of graphical data mining algorithm and its application, utilize vertex representation
Criminal, node of graph information expression case is other, if crime personnel are in selection motivation, selection place, selecting object, crime means
As soon as having identical on equal main informations, then there is a line connection naturally, indicate that the two crime personnel connected there may be connection
System or clique, then excavate frequent subgraph, these Frequent tree minings are to finding out criminal activity rule and Safety actuality on the diagram
Development trend provides effective decision-making foundation.
Wu Ye et al. proposes a kind of multi-source geographical spatial data relational model MSGCM towards efficient retrieval, passes through extraction
Multi-source geographical spatial data spatial information, semantic description information, content description information and its incidence relation, construction feature element
Figure.And multi-source geospatial object is fused in uniform spaces based on association mode, by calculating the pass between different objects
Join intensity, build the correlation model of similar figure, provides based on keyword, is based on geographical location and object-based three kinds of issuers
Method.
Above graph model construction method only for respective FIELD Data, is not applied for all data.Red tide monitoring
Data have the characteristics that：First, red tide is interactional by various impact factors as a result, having between each impact factor
Complicated relationship is suitble to store it using graph model；Secondly, the impact factor of red tide is numerous, and data have higher-dimension；
Finally, it in order to give red tide forecast early warning personnel to provide quickly accurate aid decision, needs to deposit red tide monitoring data by the stage
It puts.These features, which result in storage red tide monitoring data, cannot directly apply mechanically other storage method, it is necessary to which development is a kind of for red
The graph model of damp monitoring data feature.
In order to reduce the influence that red tide is brought, the associated mechanisms of countries in the world all establish red tide in the neighbour marine site of oneself
Monitoring system carrys out forecasting and warning Disaster And Prevention Measures of Red Tides.Most monitoring system monitoring region is wide, and monitoring parameters are numerous, brings a large amount of
Red tide monitoring historical data.These historical datas can provide auxiliary and reference for the forecast of red tide forecast personnel.Therefore face
To these historical datas, how for it effective storage model is built, improve the inquiry accuracy and speed of red tide forecast personnel, precisely
Forecast the stage residing for red tide (initial period, developing stage, maintenance stage and extinction stage), and special for the biology in each stage
Sign takes corresponding control measure, is only the key for successfully studying red tide.
The method of currently used storage red tide monitoring data is deposited according to the difference in time, monitoring region or monitoring means
Storage is in relational database.This method table stable structure, can keep the consistency of data.But use relation data inventory
Storage red tide monitoring data have the following problems：First, a red tide be it is coefficient by a variety of impact factors as a result,
Relational database can only record data, can not establish between each impact factor, the relationship between each red tide stage of development；Its
Secondary, continuing to monitor for many years causes red tide monitoring data volume huge, and the plenty of time will be consumed when searching related data.
Chinese patent literature CN201010242351.5, the applying date 20100802, patent name is：A kind of easy red tide
Method for early warning includes the following steps：Following sea area illumination variation situation in a short time is first investigated, if its suitable or convenient sea area is red
Damp biological growth breeding, then monitor the cell quantity of red tide plankton in the presence of sea area；In these cases, if having one in sea area
Algal bloom biological cell quantity has reached its certain population advantage, then needs to judge that the suspicious red tide plankton gives birth to live water temperature again
Response situation is managed, the risk class that red tide occurs for sea area is then made；If being not present in sea area or thin without a kind of red tide plankton
Born of the same parents' quantity reaches its certain population advantage, there is no need to monitor water temperature again, can directly make the risk etc. that red tide occurs for sea area
Grade；If illumination variation situation is not suitable for red tide plankton growth in a short time in the following sea area, thin there is no need to monitor sea area red tide plankton
Born of the same parents' quantity and live water temperature, can directly make the risk class that red tide occurs for sea area.
The red tide prewarning method of above patent document, by the cell quantity of red tide plankton present in sea area, suspicious red
Damp biology judges sea area to three factors such as live water temperature physiological reaction and the possible situation of change of following sea area illumination in a short time
The risk class that red tide occurs in the recent period, to reach red tide prewarning purpose.The red tide prewarning method is suitable for each local sea area scene
Monitoring, the required factor judged of entire method is few and clear, and required data are obtained without need for large-scale instrument and equipment, and whole
A method does not need complicated operation and judgement yet.In terms of early warning effect, the red tide prewarning method sensitivity is higher, is judged
Early warning result reliability it is high, if can be promoted in each sea area and conscientiously applied to each She Hai economic entities, and producing
Corresponding red tide prewarning coping mechanism is established in journey, will greatly enhance its resisting risk ability and the market competitiveness.But it closes
Data can be classified in one kind, and sorted data are stored in graph model, to facilitate looking into for red tide data
It looks for, achievees the purpose that quick Exact Forecast, more natural the data with complicated incidence relation can be stored.But it closes
Red tide data can be classified in one kind by stage of development, and is stored in graph model by the stage；Red tide forecast personnel can
Quickly to judge the stage residing for red tide, and control measure is targetedly proposed for the current generation, reduces economic and ecology damage
It loses；Facilitate that red tide forecast personnel are accurate, quick search is to data are needed, is provided for red tide forecast personnel quick and precisely comprehensively several
According to upper support, comparison control is carried out using present data and historical data, predicts the technical solution of the stage of development of red tide
Then without corresponding open.
In conclusion needing one kind that red tide data can be classified by stage of development, and artwork is stored in by the stage
In type；Red tide forecast personnel can quickly judge the stage residing for red tide, and targetedly propose that prevention is arranged for the current generation
It applies, reduces economy and Ecological Loss；Facilitate that red tide forecast personnel are accurate, quick search is to data are needed, for red tide forecast personnel
Support in quick and precisely comprehensive data is provided, comparison control is carried out using present data and historical data, predicts red tide
Stage of development the red tide prewarning method based on graph model structure, and yet there are no report about this method for early warning.
Invention content
The purpose of the present invention is being directed to deficiency in the prior art, provide one kind can by red tide data by stage of development into
Row classification, and be stored in graph model by the stage；Red tide forecast personnel can quickly judge the stage residing for red tide, and for current
Stage targetedly proposes control measure, reduces economy and Ecological Loss；Facilitate that red tide forecast personnel are accurate, quick search arrives
Data are needed, the support in quick and precisely comprehensive data is provided for red tide forecast personnel, uses present data and history number
According to comparison control is carried out, the red tide prewarning method of the stage of development of red tide built based on graph model is predicted.
To achieve the above object, the technical solution adopted by the present invention is that：
A kind of red tide prewarning method based on graph model structure, the described method comprises the following steps：
Step S1, data prediction；
Step S11, weight calculation；
Step S12, environmental factor dimensionality reduction；
Step S2 executes DWFCM clustering algorithms；
Step S21, cluster centre selection improve；
Step S22, Euclidean distance weighting；
Step S23, object function improve；
Step S3 builds red tide diagram data model.
As a kind of perferred technical scheme, in step S1, sample data is normalized, data is made to be mapped to section
Between [0,1], normalized function is as follows：
Wherein, min (xi) and max (xi) be respectively sample data minimum value and maximum value.
As a kind of perferred technical scheme, the weight calculation in step S11 is specific as follows：
Assuming that input sample matrix X, shares m sample, n environmental factor, expression formula is：
Each sample is to the contribution degree of jth (1≤j≤n) a environmental factor：
Whole i sample datas are expressed as E to the contribution total amount of environmental factor jj：
Wherein, constantThis ensures that 0≤Ej≤ 1, i.e. EjIt is up to 1, when each sample number under some attribute
According to contribution degree reach unanimity when, EjTend to 1；When the sample data under the attribute is all equal, do not consider the attribute certainly
Effect in plan, the weight of the attribute is zero at this time.
As a kind of perferred technical scheme, occurred in a red tide by calculating each environmental factor in step S12
Feature power kind in journey, dimensionality reduction is carried out to the high-dimensional environment factor, and by ranking, the corresponding sample selection of environmental factor of the first two comes out,
Subsequent red tide monitoring data are carried out using the sample after dimensionality reduction to classify.
As a kind of perferred technical scheme, the cluster centre selection in step S12, which improves, is specially：
Step S121, if data set X={ xi, i=1 ..., n } and it is sample set, the minimum range threshold between each class is set
Value a；
Step S122, input：Data set X, output：C cluster centre；
Step S123 calculates the Euclidean distance between any two sample, generates Distance matrix D, will be apart from nearest two
A data sample is set to one kind, and takes the midpoint of two samples as first kind cluster centre
Step S123 sets minimum threshold of distance a, is searched and two sample distances in the first kind using Distance matrix D
It is all higher than the sample of a, and two nearest samples of selection Weighted distance are set to one kind in these samples, and takes two samples
Midpoint is as the second class cluster centre；
Step S124 is similarly looked in remaining sample and is both greater than the sample of a with the sample distance found, and select
It is a kind of to select the shortest two samples positioning of distance in these samples, and takes the terminal of two samples as its cluster centre；
Step S125, repeat step S124, until finding c cluster centre, if in sample be not present distance a with
Interior sample can suitably reduce α.
As a kind of perferred technical scheme, a weight, weight and Europe are assigned by each environmental factor in step S22
Formula distance combines, and obtains the weighted euclidean distance based on weight, and each sample is made to have more tendentious degree of membership, specific side
Method is as follows：
Step S221 defines related coefficient Tij, sample xiAnd xjRelated coefficient, i=1,2 ..., n, j=1,2 ..., m,
And Tij=Tft；
Step S222 defines weighted euclidean distance dij, d2 ij=| | (xi-xj)+Tij(yi-yj)||2
Step S223 defines weighted distance matrix D,It is that any two sample adds in X
Weigh the matrix of Euclidean distance, wherein dijIt is sample xiAnd xjBetween weighted euclidean distance.
As a kind of perferred technical scheme, the target letter of distance between sample data and cluster centre is considered in step S3
Number is：
Wherein,dijIt is viAnd vjBetween weighting it is European away from
From；
Object function need to meet claimed below：
μik∈ [0,1],
Using lagrange's method of multipliers to Jm(U, V, η) partial differential is sought in degree of membership and cluster centre and make etc.
Formula is equal to zero, and the more new formula for obtaining degree of membership and cluster centre is：
As a kind of perferred technical scheme, the algorithm steps after Further aim function are as follows：
Step S231, input：Data set X, cluster centre number c, end condition ε, Fuzzy Weighting Exponent m, minimum weight
Distance threshold a, iterations k=1；Output：C cluster
Step S232 calculates weighted distance matrix D, finds out c according to minimum weight distance threshold a and cluster centre number c
A cluster centre；
Step S233 resets iterations k=1, using the result of step 232 as initial cluster center vi, i=1,
2 ..., c；
Step S234, according to vi(k) formula is pressedCalculate degree of membership μij(k)；
Step S235, according to μij(k) formula is pressedContinue to change
For object function and cluster centre vi(k+1)；
Step S236, if ‖ vi(k+1)-viK ‖ >=ε, then return to step three, k=k+1；Otherwise, it end loop and obtains
Cluster result.
As a kind of perferred technical scheme, following steps are specifically included in step S4：
Red tide data are fallen into 5 types, are respectively by step S41：Red tide stage, initial period, developing stage, dimension do not occur
Hold stage and extinction stage；
Sorted red tide data are stored in graph model by step S42, with big figure and small figure structure point and side, on big figure
O'clock be made of a small figure, indicate that the red tide monitoring data set in a stage, session information store in the label, share 5
Stage；While indicating the correlation degree between each stage red tide monitoring data；Point on small figure indicates a red tide data, Bian You
The when and where structure incidence relation composition of red tide data；
Step S43 defines red tide data graph model, non-directed graph G=(VG, E, Mv), VG={ vi, i=1,2 ..., 5 } and it is a little
Set, red tide is shared | V | a stage, E={ eij, i, j=1,2 ..., | V |；i≠j；eij=ejiBe side set, Mv=
{scBe V label, indicate the stage of red tide, c=5；
Step S44 defines the degree of association between the red tide stage：Incidence relation between each stage is by formulaStructure, and be stored in the side E of big figure；
Step S44 defines each stage red tide datagram：VG=(v, e, mv), v={ v 'I, i=1,2 ..., n } it is current rank
One red tide data of section, e={ e 'ijIt is v 'tWith v 'jBetween side, mvThe label on v, indicate red tide data when
Between, place and attribute information；
Step S42 defines the degree of association between red tide data：By when and where information architecture, and it is stored in small figure
In the e of side.
The invention has the advantages that：
1, a kind of red tide prewarning method based on graph model structure of the invention, can by red tide data by stage of development into
Row classification, and be stored in graph model by the stage；Red tide forecast personnel can quickly judge the stage residing for red tide, and for current
Stage targetedly proposes control measure, reduces economy and Ecological Loss；Facilitate that red tide forecast personnel are accurate, quick search arrives
Data are needed, the support in quick and precisely comprehensive data is provided for red tide forecast personnel, uses present data and history number
According to comparison control is carried out, the stage of development of red tide is predicted.
2, assume known one group of red tide data, store it in the graph model that the present invention uses, can quickly judge to work as
Stage residing for preceding red tide, and take appropriate measures；Red tide forecast personnel want inquiry some day, a certain marine site it is all red
Damp historical data can go out accurately data with quick search.
3, the present invention is classified data using the DWFCM algorithms for red tide monitoring data, and will be sorted
Data are stored in graph model, to facilitate the lookup of red tide data, achieve the purpose that quick Exact Forecast.
4, the present invention first pre-processes red tide monitoring data by impact factor weight calculation and dimensionality reduction, then
By improving cluster centre selection, having obtained suitable red tide data classification to Euclidean distance weighted sum Further aim function
The red tide monitoring data classified finally are stored in graph model, are established on side between each phase data by DWFCM algorithms
Relationship.Experiment shows that sorting algorithm used in the present invention has higher operation when classifying to red tide monitoring data
The use of speed and accuracy, graph model improves the efficiency of inquiry, and providing efficient auxiliary to the forecasting and warning of red tide determines
Plan.
5, weighed kind by calculating the feature of each environmental factor in a red tide generating process, to the high-dimensional environment factor into
Row dimensionality reduction, by ranking, the corresponding sample selection of environmental factor of the first two comes out, and subsequent red tide is carried out using the sample after dimensionality reduction
Monitoring data are classified, and operation time is improved.
6, by selecting to improve to cluster centre so that the stabilization of object function iterations improves the operation speed of algorithm
Degree.
7, the present invention is that each environmental factor assigns a weight, and weight and Euclidean distance are combined, obtained based on weight
Weighted euclidean distance makes each sample have more tendentious degree of membership.
8, the improvement of the invention by object function, improved object function not only allow between sample and cluster centre
Distance, it is also contemplated that interrelated factor between each cluster centre calculates the minimum between sample and cluster centre
Maximum weighted Euclidean distance between weighted euclidean distance and each cluster centre so that inter- object distance is closer, and between class distance is got over
Far, cluster result will be more accurate.
9, diagram data model is a data model for indicating and storing information by vertex, side and its attribute, it
Flexibility more natural can store the data with complicated incidence relation, be provided to red tide forecast early warning personnel efficient
Aid decision, provide a kind of new memory module for red tide monitoring data.
10, the present invention is classified data according to stage of development by algorithm, then by sorted each phase data
As a vertex on big figure, the side between each phase data is built；Finally by each data structure on big figure vertex
It builds as a vertex on small figure, the similarity relation on small figure side is built with time and geographical location, to realize using artwork
Type stores red tide monitoring data.
Description of the drawings
Attached drawing 1 is a kind of overall framework schematic diagram of red tide prewarning method built based on graph model of the present invention.
Attached drawing 2 is red tide data graph model.
Specific implementation mode
It elaborates below in conjunction with the accompanying drawings to specific implementation mode provided by the invention.
Fig. 1 is please referred to, Fig. 1 is a kind of overall framework signal of red tide prewarning method built based on graph model of the present invention
Figure.Red tide data are pre-processed first, it is contemplated that the generation of red tide is formed by various environmental factors collective effect, if jointly
Bring calculating into, larger calculation amount will be led to, thus first calculate each environmental factor of red tide weight, and according to weight to environment because
Son carries out dimensionality reduction, selects shared maximum two environmental factors of weight and is subsequently calculated.Secondly, by the cluster to FCM
The heart, Euclidean distance and object function are improved, and obtain DWFCM algorithms, and red tide data are fallen into 5 types according to stage of development.Most
Sorted red tide data are stored in graph model afterwards, structure point and side.
A kind of red tide prewarning method based on graph model structure, the described method comprises the following steps：
Step S1, data prediction；
Step S11, weight calculation；
Step S12, environmental factor dimensionality reduction；
Step S2 executes DWFCM clustering algorithms；
Step S21, cluster centre selection improve；
Step S22, Euclidean distance weighting；
Step S23, object function improve；
Step S3 builds red tide diagram data model.
1. data prediction
First, it is contemplated that sample data may have different data target and unit, and sample data is normalized,
Data are made to be mapped between section [0,1].Normalized function is as follows：
Wherein, min (xi) and max (xi) be respectively sample data minimum value and maximum value.
1.1 weight calculation
Assuming that input sample matrix X, shares m sample, n environmental factor.
Each sample is to the contribution degree of jth (1≤j≤n) a environmental factor：
Whole i sample datas are expressed as E to the contribution total amount of environmental factor jj：
Wherein, constantThis ensures that 0≤Ej≤1, i.e. EjIt is up to 1.
It can be seen that from formula when the contribution degree of each sample data under some attribute reaches unanimity, FjTend to 1；Especially
It is when the sample data under the attribute is all equal, so that it may not consider effect of the attribute in decision, i.e. the category at this time
The weight of property is zero.
djIndicate the degree of consistency of each sample data contribution degree under j-th of environmental factor, dj=1-Ej。
Each attribute weight is：
1.2 environmental factor dimensionality reductions
Red tide is formed by a variety of environmental factor collective effects, and red tide of each environmental factor pair has
There is different influence degrees, classify if these environmental factors are all brought into, higher operation time will be caused.The present invention
Kind is weighed by calculating feature of each environmental factor in a red tide generating process, dimensionality reduction is carried out to the high-dimensional environment factor, it will
The ranking corresponding sample selection of environmental factor of the first two comes out, and subsequent red tide monitoring data point are carried out using the sample after dimensionality reduction
Class.
2.DWFCM
The selection of 2.1 cluster centres improves
FCM algorithms, which are exactly one, makes object function JmThe iterative solution process of minimum.In FCM algorithms, algorithm gathers
Class effect suffers from the influence of initial cluster center, and the random selection of initial cluster center results in object function iterations
It is unstable, and be easy so that the case where algorithmic statement is to local minimum, the calculating of a large amount of repeatedly Euclidean distances result in algorithm
The speed of service it is low.In view of the above problems, the present invention first improves the selection of initial cluster center.
If X={ xi, i=1 ..., n } and it is sample set, the minimum threshold of distance α between each class is set, is selected initial poly-
The algorithm steps at class center are as follows.
Algorithm 1:
Input：Data set X
Output：C cluster centre
Step 1：The Euclidean distance between any two sample is calculated, Distance matrix D is generated.By two of distance recently
Data sample is set to one kind, and takes the midpoint of two samples as first kind cluster centre；
Step 2：Minimum threshold of distance α is set, is searched with two samples distance in the first kind using Distance matrix D
Sample more than α, and two nearest samples of selection Weighted distance are set to one kind in these samples, and take in two samples
Point is used as the second class cluster centre；
Step 3：Similarly, it is looked in remaining sample and is both greater than the sample of α with the sample distance found, and selected
The shortest two samples positioning of distance is a kind of in these samples, and takes the terminal of two samples as its cluster centre；
Step 4：Step 3 is repeated, until finding c cluster centre.If there is no distances within α in sample
Sample can suitably reduce α.
2.2 Euclidean distances weight
In FCM algorithms, some sample xkWhich kind of is more likely to belong to, needs to be judged according to degree of membership size, and is subordinate to
It is in cluster process that the computational methods of category degree, which are by measuring the distance between sample and cluster, therefore according to range estimation ownership,
Important method.In red tide monitoring data, there are part edge samples, that is, are between two stages, and degree of membership distribution is equal
It is even, it can not directly judge which stage this sample belongs to by degree of membership.FCM algorithms based on traditional Euclidean distance, it is European away from
It is similarly acted on from assuming that each attribute plays during cluster.But in red tide during actually occurring, Mei Gehuan
There is different weighing factors, some environmental factors important work is played in cluster process for the generation of red tide of border factor pair
With, and the effect of some environmental factors is secondary or can be ignored.In view of the above problems, the present invention is each environmental factor
A weight is assigned, weight and Euclidean distance combine, and obtain the weighted euclidean distance based on weight, each sample is made to have more
Tendentious degree of membership.
It calculates first in a red tide generating process, the weight that each Environmental Factors red tide occurs, and by its band
Enter in Euclidean distance.
Define 1 correlation coefficient rij.Sample xiAnd xjRelated coefficient, i=1,2 ..., n, j=1,2 ..., m, and rij=
rij。
Define 2：Weighted euclidean distance dij。d2 ij=| | (xi-xj)+rij(yi-yj)||2
Define 3：Weighted distance matrix D.Be any two sample in X weighting it is European
The matrix of distance.Wherein dijIt is sample xiAnd xjBetween weighted euclidean distance.
3. object function improves
Original FCM algorithms only account for the distance between sample data and cluster centre, do not consider each cluster centre it
Between distance, inter- object distance is closer, and between class distance is remoter, and cluster result will be more accurate.
The object function for considering distance between sample data and cluster centre is：
Wherein,dijIt is viAnd vjBetween weighting it is European away from
From.Object function need to meet claimed below：
Using lagrange's method of multipliers to Jm(U, V, η) partial differential is sought in degree of membership and cluster centre and make etc.
Formula is equal to zero, and the more new formula for obtaining degree of membership and cluster centre is：
Improved object function not only allows for the distance between sample and cluster centre, it is also contemplated that each cluster centre
Between interrelated factor, calculate minimum weight Euclidean distance between sample and cluster centre and each cluster centre it
Between maximum weighted Euclidean distance so that each class inner distance is nearest, and distance is farthest between class, to accomplish more accurately to divide
Class.
Algorithm steps after Further aim function are as follows.
Algorithm 2:
Input：Data set X, cluster centre number c, end condition ε, Fuzzy Weighting Exponent m, minimum weight distance threshold α,
Iterations k=1.
Output：C cluster.
Step 1：According to minimum weight distance threshold α and cluster centre number c, weighted distance matrix D is calculated, finds out c
Cluster centre；
Step 2：Iterations k=1 is reset, using the result of step 1 as initial cluster center vi, i=1,2 ..., c；
Step 3：According to vi(k) it presses formula 10 and calculates degree of membership μij(k)；
Step 4：According to μij(k) it presses formula 9 and continues iterative target function and cluster centre vi(k+1)；
Step 5：If ‖ vi(k+1)-viK ‖ >=ε, then return to step three, k=k+1；Otherwise, end loop and gathered
Class result.
4. red tide datagram model construction
By the operation that 1.2.3 is saved, red tide data are classified according to the stage of generation, red tide data quilt in the present invention
It falls into 5 types, is respectively：Red tide stage, initial period, developing stage, maintenance stage do not occur and withers away the stage.
The red tide monitoring data with incidence relation are very suitable for storing it using graph model from each other.At this
In invention, o'clock being made of a small figure on big figure indicates that the red tide monitoring data set in a stage, session information are stored in mark
In label, 5 stages are shared in the present invention；While indicating the correlation degree between each stage red tide monitoring data.Point on small figure
A red tide data are indicated, while being made of the when and where structure incidence relation of red tide data.The graph model of structure such as Fig. 2 institutes
Show.
Define 4:Red tide data graph model：Non-directed graph G=(VG, E, Mv)。VG={ vi, i=1,2 ..., 5 } and it is the collection put
Close, red tide is shared | V | a stage, E={ eij, i, j=1,2 ..., | V |；i≠j；eij=ejiBe side set, Mv={ scBe
The label of v indicates the stage of red tide.In the present invention, c=5.
Define 5:The degree of association between the red tide stage：Incidence relation between each stage is built by formula 6, and is stored in
In the side E of big figure.
Define 6:Each stage red tide datagram：VG=(v, e, mv), v={ v 'i, i=1,2 ..., n } and it is the one of the current generation
Red tide data, e={ e 'ijIt is v 'iWith v 'jBetween side.mvIt is the label on v, the time of one red tide data of expression,
Point and attribute information.
Define 7:The degree of association between red tide data：By when and where information architecture, and it is stored in the side e of small figure.
The present invention is tested using the monitoring data at the Changjiang river port in May, -2014 in 2012, be respectively adopted FCM,
PCM and the improved DWFCM algorithms of the present invention classify to data, are then store in the graph model that the present invention is built.
1 FCM, PCM and DWFCM classification results of table compare
By table 1 it is found that the DWFCM methods that use of the present invention can obtain higher standard by less iterations
True rate carries out more accurate, quickly classification to red tide data.
It is just valuable that the data of storage are only applied to reality.Red tide data are stored in relational model and the present invention respectively
In the graph model used, and use identical inquiry, the result of comparison query.
(1) position keyword query：Location information includes longitude and latitude, in actual data acquisition, due to
Certain error and reality, physical location and site location may have deviation.In relational model, closed according to position
Key word is possible to can not find related data；In graph model, due to establishing the incidence relation between each data, Ke Yigen in advance
According to given position, the information of neighbouring position is inquired.
(2) time-critical word is inquired：Red tide data are a period of time sequences, by a determining time, in relational model
In can only inquire a data.In graph model, all data in correlation time can be inquired, red tide forecast people is facilitated
Member is completely forecast.
(3) environmental factor keyword query：In relational model, an environmental factor keyword can inquire it is all with
The matched data of this environmental factor, this will expend very more time.Red tide is formed by a variety of environmental factor collective effects
, a single environmental factor can not provide effective aid decision for red tide forecast personnel.It, can be with and in graph model
According to incidence relation, inquire the data of Related Environmental Factors and according to the classification of red tide stage of development and query rate it is higher.
A kind of red tide prewarning method based on graph model structure of the present invention can be carried out red tide data by stage of development
Classification, and be stored in graph model by the stage；Red tide forecast personnel can quickly judge the stage residing for red tide, and be directed to current rank
Section targetedly proposes control measure, reduces economy and Ecological Loss；Facilitate that red tide forecast personnel are accurate, quick search to need
Data are wanted, the support in quick and precisely comprehensive data is provided for red tide forecast personnel, uses present data and historical data
Comparison control is carried out, predicts the stage of development of red tide；Assuming that known one group of red tide data, store it in the figure that the present invention uses
In model, the stage residing for current red tide can be quickly judged, and take appropriate measures；Red tide forecast personnel want to inquire certain
One day, all red tide historical datas in a certain marine site, can go out accurately data with quick search；Diagram data model is one and passes through
Vertex, side and its attribute indicate and store the data model of information, its flexibility can be more natural to having complicated pass
The data of connection relationship are stored, and efficient aid decision is provided to red tide forecast early warning personnel, provide a kind of new be directed to
The memory module of red tide monitoring data；The present invention is classified data according to stage of development by algorithm, after then classifying
Each phase data as a vertex on big figure, build the side between each phase data；It finally will be on big figure vertex
Each data is configured to a vertex on small figure, and the similarity relation on small figure side is built with time and geographical location, to
It realizes and stores red tide monitoring data using graph model；The present invention use for red tide monitoring data DWFCM algorithms to data into
It has gone classification, and sorted data has been stored in graph model, to facilitate the lookup of red tide data, reached quick Exact Forecast
Purpose；The present invention first pre-processes red tide monitoring data by impact factor weight calculation and dimensionality reduction, then logical
It crosses and improves cluster centre selection, the DWFCM of suitable red tide data classification has been obtained to Euclidean distance weighted sum Further aim function
The red tide monitoring data classified finally are stored in graph model, the relationship between each phase data are established on side by algorithm.
Experiment show sorting algorithm used in the present invention when classifying to red tide monitoring data have the higher speed of service and
The use of accuracy, graph model improves the efficiency of inquiry, and efficient aid decision is provided to the forecasting and warning of red tide.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
Member, under the premise of not departing from the method for the present invention, can also make several improvement and supplement, these are improved and supplement also should be regarded as
Protection scope of the present invention.
Claims (9)
1. a kind of red tide prewarning method based on graph model structure, which is characterized in that the described method comprises the following steps：
Step S1, data prediction；
Step S11, weight calculation；
Step S12, environmental factor dimensionality reduction；
Step S2 executes DWFCM clustering algorithms；
Step S21, cluster centre selection improve；
Step S22, Euclidean distance weighting；
Step S23, object function improve；
Step S3 builds red tide diagram data model.
2. the red tide prewarning method according to claim 1 based on graph model structure, which is characterized in that right in step S1
Sample data is normalized, and so that data is mapped between section [0,1], normalized function is as follows：
Wherein, min (xi) and max (xi) be respectively sample data minimum value and maximum value.
3. the red tide prewarning method according to claim 1 based on graph model structure, which is characterized in that in step S11
Weight calculation is specific as follows：
Assuming that input sample matrix X, shares m sample, n environmental factor, expression formula is：
Each sample is to the contribution degree of jth (1≤j≤n) a environmental factor：
Whole i sample datas are expressed as E to the contribution total amount of environmental factor jj：
Wherein, constantThis ensures that 0≤Ej≤ 1, i.e. EjIt is up to 1, when the tribute of each sample data under some attribute
When degree of offering reaches unanimity, EjTend to 1；When the sample data under the attribute is all equal, do not consider the attribute in decision
Effect, the weight of the attribute is zero at this time.
4. the red tide prewarning method according to claim 1 based on graph model structure, which is characterized in that lead in step S12
The feature power kind for calculating each environmental factor in a red tide generating process is crossed, dimensionality reduction is carried out to the high-dimensional environment factor, will be arranged
The name corresponding sample selection of environmental factor of the first two comes out, and subsequent red tide monitoring data point are carried out using the sample after dimensionality reduction
Class.
5. the red tide prewarning method according to claim 1 based on graph model structure, which is characterized in that in step S12
Cluster centre selection, which improves, is specially：
Step S121, if data set X={ xi, i=1 ..., n } and it is sample set, the minimum threshold of distance α between each class is set；
Step S122, input：Data set X, output：C cluster centre；
Step S123 calculates the Euclidean distance between any two sample, generates Distance matrix D, by two nearest numbers of distance
It is set to one kind according to sample, and takes the midpoint of two samples as first kind cluster centre
Step S123 sets minimum threshold of distance α, is searched using Distance matrix D big with two samples distance in the first kind
In the sample of α, and two nearest samples of selection Weighted distance are set to one kind in these samples, and take the midpoint of two samples
As the second class cluster centre；
Step S124 is similarly looked in remaining sample and is both greater than the sample of α with the sample distance found, and select this
The shortest two samples positioning of distance is a kind of in a little samples, and takes the terminal of two samples as its cluster centre；
Step S125 repeats step S124, until finding c cluster centre, if there is no distances within α in sample
Sample can suitably reduce α.
6. the red tide prewarning method according to claim 1 based on graph model structure, which is characterized in that lead in step S22
It crosses each environmental factor and assigns a weight, weight and Euclidean distance combine, and obtain the weighted euclidean distance based on weight, make every
A sample has more tendentious degree of membership, and the specific method is as follows：
Step S221 defines correlation coefficient rij, sample xiAnd xjRelated coefficient, i=1,2 ..., n, j=1,2 ..., m, and rij
=rji；
Step S222 defines weighted euclidean distance dij, d2 ij=| | (xi-xj)+rij(yi-yj)||2
Step S223 defines weighted distance matrix D,It is the weighting Europe of any two sample in X
The matrix of formula distance, wherein dijIt is sample xiAnd xjBetween weighted euclidean distance.
7. the red tide prewarning method according to claim 1 based on graph model structure, which is characterized in that consider in step S3
The object function of distance is between sample data and cluster centre：
Wherein,0≤β≤1, a, b ∈ c, dijIt is viAnd vjBetween weighted euclidean distance；
Object function need to meet claimed below：
μik∈ [0,1],
Using lagrange's method of multipliers to JmThe equation etc. that (U, V, η) seeks partial differential and made in degree of membership and cluster centre
In zero, the more new formula for obtaining degree of membership and cluster centre is：
8. the red tide prewarning method according to claim 7 based on graph model structure, which is characterized in that Further aim function
Algorithm steps afterwards are as follows：
Step S231, input：Data set X, cluster centre number c, end condition ε, Fuzzy Weighting Exponent m, minimum weight distance
Threshold alpha, iterations k=1；Output：C cluster
Step S232 calculates weighted distance matrix D according to minimum weight distance threshold α and cluster centre number c, finds out c and gathers
Class center；
Step S233 resets iterations k=1, using the result of step 232 as initial cluster center vi, i=1,2 ..., c；
Step S234, according to vi(k) formula is pressedCalculate degree of membership μij(k)；
Step S235, according to μij(k) formula is pressedContinue iteration mesh
Scalar functions and cluster centre vi(k+1)；
Step S236, if | | vi(k+1)-viK | | >=ε, then return to step three, k=k+1；Otherwise, end loop and gathered
Class result.
9. the red tide prewarning method according to claim 7 based on graph model structure, which is characterized in that specific in step S4
Include the following steps：
Red tide data are fallen into 5 types, are respectively by step S41：The red tide stage does not occur, initial period, developing stage, maintains rank
Section and extinction stage；
Sorted red tide data are stored in graph model by step S42, and point and side, the point on big figure are built with big figure and small figure
It is made of a small figure, indicates that the red tide monitoring data set in a stage, session information store in the label, share 5 stages；
While indicating the correlation degree between each stage red tide monitoring data；Point on small figure indicates a red tide data, while by red tide
The when and where structure incidence relation composition of data；
Step S43 defines red tide data graph model, non-directed graph G=(VG, E, Mv), VG={ vi, i=1,2 ..., 5 } and it is the collection put
Close, red tide is shared | V | a stage, E={ eij, i, j=1,2 ..., | V |；i≠j；eij=ejiBe side set, Mv={ scBe
The label of V indicates the stage of red tide, c=5；
Step S44 defines the degree of association between the red tide stage：Incidence relation between each stage is by formulaStructure, and be stored in the side E of big figure；
Step S44 defines each stage red tide datagram：VG=(v, e, mv), v={ v 'i, i=1,2 ..., n } and it is the current generation
One red tide data, e={ e 'ijIt is v 'iWith v 'jBetween side, mvIt is the label on v, the time of one red tide data of expression,
Place and attribute information；
Step S42 defines the degree of association between red tide data：By when and where information architecture, and it is stored in the side e of small figure
In.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810243914.9A CN108446802B (en) | 2018-03-23 | 2018-03-23 | Red tide early warning method constructed based on graph model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810243914.9A CN108446802B (en) | 2018-03-23 | 2018-03-23 | Red tide early warning method constructed based on graph model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108446802A true CN108446802A (en) | 2018-08-24 |
| CN108446802B CN108446802B (en) | 2022-08-19 |
Family
ID=63196711
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810243914.9A Active CN108446802B (en) | 2018-03-23 | 2018-03-23 | Red tide early warning method constructed based on graph model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108446802B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112084716A (en) * | 2020-09-15 | 2020-12-15 | 河北省科学院地理科学研究所 | Red tide prediction and early warning method based on eutrophication comprehensive evaluation |
| CN113378682A (en) * | 2021-06-03 | 2021-09-10 | 山东省科学院自动化研究所 | Millimeter wave radar fall detection method and system based on improved clustering algorithm |
| CN114510875A (en) * | 2022-01-26 | 2022-05-17 | 国家海洋环境预报中心 | Circulation-current-situation red tide forecasting system and method based on multi-source meteorological data |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101901468A (en) * | 2010-08-02 | 2010-12-01 | 福州市海洋与渔业技术中心 | Simple red tide early warning method |
| CN102721655A (en) * | 2012-06-01 | 2012-10-10 | 天津科技大学 | Optimal distinguishing method for red tide water body and main algae thereof |
-
2018
- 2018-03-23 CN CN201810243914.9A patent/CN108446802B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101901468A (en) * | 2010-08-02 | 2010-12-01 | 福州市海洋与渔业技术中心 | Simple red tide early warning method |
| CN102721655A (en) * | 2012-06-01 | 2012-10-10 | 天津科技大学 | Optimal distinguishing method for red tide water body and main algae thereof |
Non-Patent Citations (7)
| Title |
|---|
| HU X, WANG D, QU H, ETC: "Prediction Research of Red Tide Based on Improved FCM", 《MATHEMATICAL PROBLEMS IN ENGINEERING》 * |
| 周忠海等: "基于应对赤潮灾害发生的预防预警措施的探讨", 《中国水运(下半月)》 * |
| 张承慧 等: "一种改进的FCM聚类算法及其在赤潮预测中的应用", 《山东大学学报(工学版)》 * |
| 李雪丁 等: "福建省赤潮预警系统研究与应用", 《海洋预报》 * |
| 欧晓平 等: "一个基于图的音乐数据模型与查询语言及其实现", 《计算机研究与发展》 * |
| 王兴强 等: "改进的CSFCM聚类算法及其在赤潮监测中的应用", 《计算机工程与应用》 * |
| 黄冬梅 等: "强关联海洋数据划分研究", 《计算机工程》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112084716A (en) * | 2020-09-15 | 2020-12-15 | 河北省科学院地理科学研究所 | Red tide prediction and early warning method based on eutrophication comprehensive evaluation |
| CN113378682A (en) * | 2021-06-03 | 2021-09-10 | 山东省科学院自动化研究所 | Millimeter wave radar fall detection method and system based on improved clustering algorithm |
| CN114510875A (en) * | 2022-01-26 | 2022-05-17 | 国家海洋环境预报中心 | Circulation-current-situation red tide forecasting system and method based on multi-source meteorological data |
| CN114510875B (en) * | 2022-01-26 | 2022-09-30 | 国家海洋环境预报中心 | Circulation-current-situation red tide forecasting system and method based on multi-source meteorological data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108446802B (en) | 2022-08-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109034448A (en) | Trajectory predictions method based on track of vehicle semantic analysis and deepness belief network | |
| CN103324954B (en) | Image classification method based on tree structure and system using same | |
| CN108446802A (en) | A kind of red tide prewarning method based on graph model structure | |
| Yan et al. | A stream prediction model based on attention-LSTM | |
| Khan et al. | Evaluating the performance of several data mining methods for predicting irrigation water requirement | |
| Gupta et al. | Clustering-Classification based prediction of stock market future prediction | |
| CN107748940A (en) | A kind of energy conservation potential Quantitative prediction methods | |
| CN115099450A (en) | Family carbon emission monitoring and accounting platform based on fusion model | |
| Li et al. | Daily streamflow forecasting based on flow pattern recognition | |
| Yan et al. | Climate-informed monthly runoff prediction model using machine learning and feature importance analysis | |
| Fattore et al. | Optimal scoring of partially ordered data, with an application to the ranking of smart cities | |
| Mao et al. | Naive Bayesian algorithm classification model with local attribute weighted based on KNN | |
| Wang et al. | Computer supported data-driven decisions for service personalization: a variable-scale clustering method | |
| Xu et al. | MM-UrbanFAC: Urban functional area classification model based on multimodal machine learning | |
| Aranay et al. | Deep active genetic learning-based assessment of lakes’ water quality using climate data | |
| Khan et al. | Irrigation water requirement prediction through various data mining techniques applied on a carefully pre-processed dataset | |
| Wang | Application of carbon emission prediction based on a combined neural algorithm in the control of coastal environmental pollution in China | |
| Yang | Intelligent informatization early warning analysis of agricultural economy based on support vector sequential regression model | |
| Zhu et al. | Priority power generation allocation method for reservoir power stations based on the priority index | |
| Jawla et al. | Crime Forecasting using Folium | |
| Segretier et al. | Evolutionary predictive modelling for flash floods | |
| Chen et al. | An approach of multi-element fusion method for harmful algal blooms prediction | |
| Chen et al. | Analysis of urban expansion and the driving forces in eastern coastal region of China | |
| Huang et al. | Graph model building for red tide data based on DWFCM clustering algorithm | |
| Geng et al. | Study on index model of tropical cyclone intensity change based on projection pursuit and evolution strategy |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |