CN110414713A - A kind of runoff real-time predicting method based on synchronous data flow compression - Google Patents
A kind of runoff real-time predicting method based on synchronous data flow compression Download PDFInfo
- Publication number
- CN110414713A CN110414713A CN201910564887.XA CN201910564887A CN110414713A CN 110414713 A CN110414713 A CN 110414713A CN 201910564887 A CN201910564887 A CN 201910564887A CN 110414713 A CN110414713 A CN 110414713A
- Authority
- CN
- China
- Prior art keywords
- runoff
- data
- representative sample
- prediction
- effect factors
- 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 37
- 230000006835 compression Effects 0.000 title claims abstract description 27
- 238000007906 compression Methods 0.000 title claims abstract description 27
- 230000001360 synchronised effect Effects 0.000 title claims abstract description 27
- 230000000694 effects Effects 0.000 claims abstract description 65
- 238000012417 linear regression Methods 0.000 claims abstract description 21
- 238000013480 data collection Methods 0.000 claims abstract description 16
- 230000008569 process Effects 0.000 claims abstract description 10
- 238000001514 detection method Methods 0.000 claims abstract description 8
- 238000004364 calculation method Methods 0.000 claims abstract description 3
- 238000009825 accumulation Methods 0.000 claims description 12
- 230000003993 interaction Effects 0.000 claims description 6
- 238000012423 maintenance Methods 0.000 claims description 6
- 230000001550 time effect Effects 0.000 claims description 4
- 238000011160 research Methods 0.000 claims description 3
- 230000004069 differentiation Effects 0.000 claims description 2
- 239000011159 matrix material Substances 0.000 claims description 2
- 239000000203 mixture Substances 0.000 claims description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 abstract description 16
- 230000008859 change Effects 0.000 description 8
- 238000011161 development Methods 0.000 description 4
- 238000007726 management method Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 238000012271 agricultural production Methods 0.000 description 1
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000003651 drinking water Substances 0.000 description 1
- 235000020188 drinking water Nutrition 0.000 description 1
- 230000008020 evaporation Effects 0.000 description 1
- 238000001704 evaporation Methods 0.000 description 1
- 239000013505 freshwater Substances 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 239000008235 industrial water Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000001556 precipitation Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000010792 warming Methods 0.000 description 1
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"
-
- 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
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The invention discloses a kind of runoff real-time predicting methods based on synchronous data flow compression, first by collecting impact factor feature relevant to runoff in basin, then it constructs representative sample data collection and represents whole data, in this way, pass through the K item record building linear regression model (LRM) nearest with sample to be predicted, and treat forecast sample and carry out Runoff Forecast, improve the accuracy of Runoff Forecast.Simultaneously, in view of the influence of weather and mankind's activity is constantly changing, so that the relation schema of effect factors and runoff is constantly changing, the present invention detects effect factors data using the concept drift detection method based on Statisti-cal control process, see whether it is developed, in case of developing, then a data set is reinitialized, which further increases the accuracys of prediction.In addition, the present invention is compressed using synchronous data flow, the calculation amount of comparison procedure is reduced, to realize the real-time Accurate Prediction to runoff under changing environment, is utilized for water resources management and provides technical support.
Description
Technical field
The invention belongs to hydrographic water resource application fields, more specifically, are related to a kind of based on synchronous data flow compression
Runoff real-time predicting method.
Background technique
Water resource is most basic, most important natural resources required for human social development, be human social economy and
Required source and forever motive force in Sustainable Development of Ecological Environment.Wherein the freshwater resources such as runoff are the important of water resource
It constitutes, is the main source of the daily required drinking water of the mankind, domestic water and industrial water etc..Meanwhile runoff is basin water
The important link of circulation, the Evolution of runoff be basin flood control take precautions against drought, basin water resources Sustainable Exploitation, utilization, planning with
The important evidence of management.With the rapid development of social economy, a large amount of hydraulic engineering, traffic engineering have been built in many basins
Deng, while the significant changes that the continuous promotion of urbanization rate, watershed system occur, directly affect the production confluence rule in basin
Rule and water storage, with water and water consumption condition.Under Background of Global Warming, each ground temperature also has different degrees of raising, different
The tendency variation that the precipitation in area also shows or increases or subtract.In addition, the Extreme Weather Events frequently occurred in recent years are such as
Arid, storm flood etc. produce extreme influence to the normal water circulation in each basin.Drought forccast, flood warning, flood peak
The tasks such as volume forecasting and industrial and agricultural production and people's life are closely bound up.Carrying out accurate Runoff Forecast has these tasks
Significance and value, and realize that water resources management, water resource optimal allocation and water resource sustainability are opened under changing environment
The important science support of hair utilization, guarantee social economy's fast and stable development etc..
Traditional Runoff Simulation prediction technique is mostly carried out with the processing mode of static data: firstly, going through according to certain
The parameter of the given hydrological model of history data calibration;Secondly, being carried out by using runoff of the calibrating patterns to future period corresponding
Prediction.The hypotheses of this mode are that the relationship between impact factor and runoff is metastable.However, due to weather
The influence of variation and mankind's activity, the relationship of weather, mankind's activity and runoff often constantly change at any time.How to change
Diameter stream carries out the hot issue that accurately simulation and forecast is current hydrographic water resource field concern under environment.
Summary of the invention
It is an object of the invention to overcome the deficiencies of the prior art and provide a kind of runoffs based on synchronous data flow compression
Real-time predicting method accurately predicts the runoff under changing environment with realizing.
For achieving the above object, the present invention is based on the runoff real-time predicting method of synchronous data flow compression, features
It is, comprising the following steps:
(1), each meteorological site and the closely related feature of runoff in collection research basin
(2), data prediction counts the data that each website is collected into, and carries out completion to missing values
(3), it initializes one and records the representative sample formed by certain item number history effect factors and run-off
Data set.
(4), for the feature of the current effect factors got, found in representative sample set away from
From K nearest neighbours, building Lasso linear regression model (LRM) is recorded using this K item, and using the model to current runoff
It is predicted.After getting true diameter flow valuve, which is inserted into representative sample set.
(5), dynamic data set is safeguarded: when having new runoff to be predicted to be predicted and get the true runoff at the moment
After value, the sample weights in representative sample data set are updated based on prediction deviation.When Relative Error is small
When certain threshold value, it is believed that correct to the prediction of the point, the weight for participating in K nearest-neighbors data of prediction increases;Conversely,
The weight for participating in K nearest-neighbors data of prediction reduces.After having updated weight every time, representative sample set is worked as
The middle lesser point of weight removes.
If representative sample data collection scale exceeds specified size, threshold size can be according to operating system here
Hardware capabilities determine, are compressed using based on synchrodata flow compression method: by every impact factor record be considered as feature to
Quantity space a little or an object, using synchronization principles, interaction relationship between simulated object finally makes similar
Object is gathered in an accumulation point, replaces all similar effect factors to record using the accumulation point, i.e., the accumulation point is
The average runoff of one effect factors record, all similar effect factors records is the true diameter of the accumulation point
Flow valuve deletes all similar impact factor records, to update representative sample data collection, to achieve the purpose that compression.
Due to climate change and the continuous variation of mankind's activity, the relation schema of effect factors and runoff also can be with
Dynamic change, cause current runoff model that may be very different with history runoff model.Therefore, under these conditions, make
It cannot need to carry out concept drift detection based on historical data with the prediction that new effect factors record carries out runoff,
To safeguard the dynamic data set for representing current effect factors and runoff relationship.
Meanwhile using the concept drift detection method based on Statisti-cal control process, detect what representative sample data was concentrated
Whether the relation schema of effect factors and runoff is developed;Specifically, being recorded to effect factors, dynamic
Safeguard and record the prediction error in a nearest period in consecutive data block, the size of data block according to concrete application and
It is fixed;To the prediction error Loss of the every data of the data blockiIt is counted, calculates error mean μ and standard deviation sigma.If worked as
The preceding moment is more than defined threshold value to the prediction error of runoff, then it is assumed that the relation schema of effect factors and runoff occurs
It develops, then empties representative sample data collection, then initialized according to step (3), carry out diameter according still further to step (4)
Stream prediction.
The object of the present invention is achieved like this.
The present invention is based on the runoff real-time detection methods of synchronous data flow compression, related to runoff in basin by collecting
Impact factor feature, then construct representative sample data collection and represents entirety data, in this way, by with it is to be predicted
The nearest K item record building linear regression model (LRM) of sample, and treat forecast sample and carry out Runoff Forecast, improve Runoff Forecast
Accuracy.Simultaneously, it is contemplated that the influence of weather and mankind's activity is constantly changing, so that effect factors and runoff
Relation schema constantly changing, the present invention using based on Statisti-cal control process concept drift detection method detect diameter
Impact factor data are flowed, see whether it is developed, in case of developing, then reinitialize a data set, in this way
Further improve the accuracy of prediction.In addition, the present invention is compressed using synchronous data flow, the calculating of comparison procedure is reduced
Amount, predicts runoff to realize in real time.
Detailed description of the invention
Fig. 1 is that the present invention is based on a kind of specific embodiment processes of runoff real-time predicting method of synchronous data flow compression
Figure;
Fig. 2 is the schematic diagram of synchronous compression in the present invention, wherein yiIndicate state recording, PiIndicate compressed state note
Record;
Fig. 3 is that the present invention is based on a kind of realities of specific embodiment of runoff real-time predicting method of synchronous data flow compression
When predict run-off system framework figure.
Specific embodiment
A specific embodiment of the invention is described with reference to the accompanying drawing, so that those skilled in the art is more preferable
Ground understands the present invention.Requiring particular attention is that in the following description, when the detailed description of known function and design
When perhaps can desalinate main contents of the invention, these descriptions will be ignored herein.
Fig. 1 is that the present invention is based on a kind of specific embodiment processes of runoff real-time predicting method of synchronous data flow compression
Figure.
In the present embodiment, as shown in Figure 1, the present invention is based on the runoff real-time predicting method packets of synchronous data flow compression
Include following steps:
S1: in collection research basin with the closely related feature of runoff
Each meteorological site is collected and the closely related effect factors feature of runoff out of basin.
Closely related feature includes: long sequence Daily rainfall amount, temperature, water surface evaporation, wind speed, air with runoff
Humidity, intensity of solar radiation and basin Outlet Section test day by day stream runoff etc..
S2: data prediction
In the specific implementation process, there may be multiple meteorological sites in basin, thus the data being collected into there may be
More parts, equalization processing can be carried out to the data for the node at the same time that multiple meteorological sites are collected into, to be represented
The effect factors feature in entire basin.
In addition, the initial data being collected into may have the problems such as shortage of data, some interpolation methods can be used,
For example linear interpolation etc. carries out completion to missing data.
S3: initialization representative sample data collection
For entire basin, one observational characteristic comprising certain item number of initialization and corresponding run-off record.
S4: runoff is predicted in real time
Once get the observation of the basin a certain moment each effect factors, then it can be each according to the moment
The observation of a effect factors is found with the moment observation in representative sample data collection apart from nearest K item
Record records training Lasso linear regression model (LRM) using this K item, and using the Lasso linear regression model (LRM) to the moment
Runoff is predicted.After getting true diameter flow valuve, which is inserted into representative sample set.
What runoff was predicted in real time method particularly includes:
4.1) the observation x of the current time effect factors of acquisition, is found respectively apart from representative sample data collection
Nearest K data point, that is, history effect factors record x in conjunction1、x2、…、xkAnd corresponding runoff records y1、
y2、…、yk.Enable Y=(y1,y2,…,yk)T, X=(x1,x2,…,xk)T.As follows, it calculates
Wherein,For Lasso linear regression model (LRM) coefficient, α is penalty factor, and d is the feature dimensions of effect factors
Degree.X and Y is respectively the K item record that distance moment effect factors observational characteristic to be predicted is nearest in representative sample set
Impact factor observation and runoff measured value composition matrix;
4.2), Runoff Forecast
After obtaining Lasso linear regression model (LRM) coefficient, for the observation x of current time effect factors, benefit
The predicted value y of current time runoff is calculated with formula the following:
S5: dynamic data set maintenance
5.1), representative sample data set updates
After thering is new runoff to be predicted to be predicted and getting the true diameter flow valuve at the moment, it is based on prediction deviation
Sample weights in representative sample data set are updated.When Relative Error is less than certain threshold value, it is believed that right
The prediction of the point is correct, and the weight for participating in K nearest-neighbors data of prediction increases;Conversely, participating in K arest neighbors of prediction
The weight for occupying data reduces.After having updated weight every time, weight in representative sample set is less than lesser point and is moved
It removes.
In the present embodiment, the specific update mode of representative sample data set is as follows:
5.1.1), y is enabledrealIt is predicted the true diameter flow valuve at moment, enables ypredFor the prediction diameter flow valuve for being predicted the moment, meter
Calculate the Relative Error for being predicted moment runoff:
After getting the prediction deviation l for being predicted runoff, the update of representative sample weight is carried out using following rule,
Method is as follows:
If first, prediction error l is less than or equal to 0.1, it is believed that prediction is accurate, and K of prediction are participated in representative sample
Nearest neighbor weight w1、w2、…、wkRespectively plus one;
If second, prediction error l is greater than 0.1, it is believed that prediction error, participate in representative sample K of prediction it is nearest
Neighbor weight w1、w2、…、wkSubtract one respectively;
5.1.2), wherein weight is less than -3 representative sample by the representative sample set after updating for weight
Data are removed from representative sample.
5.2) representative sample set Real Time Compression
If representative sample data collection scale exceeds specified size (determining according to the hardware capabilities of operating system), adopt
Compressed with based on synchrodata flow compression method: by every impact factor record be considered as characteristic vector space a little or
One object, using synchronization principles, interaction relationship between simulated object finally makes similar object be gathered in one
Accumulation point replaces all similar effect factors to record using the accumulation point, i.e., the accumulation point be a runoff influence because
The average runoff of subrecord, all similar effect factors records is the true diameter flow valuve of the accumulation point, deletes all phases
As impact factor record, to update representative sample data collection, to achieve the purpose that compression.
In the present embodiment, specific as follows based on synchrodata flow compression method:
5.2.1), each effect factors in data set are recorded, indicates x with feature vector herei, it is considered as spy
Levy vector space a little or an object;
5.2.2), the neighbor objects Nb that each object and distance are εε(xi) interact, interaction models are as follows:
WhereinIndicate that the i-th effect factors record xiThe value at (t+1) moment in jth dimension; Nbε
(xi) indicate to record x with effect factorsiCentered on, the range of Euclidean distance ε removes xiOuter set of data points, | Nbε(x)|
Indicate Nbε(xi) include effect factors record item number;
5.2.3), by repeatedly interacting, similar positive effect factors record be will accumulate in together, have phase
Same value.As shown in Fig. 2, effect factors record x in Fig. 2 (a)1And x3It is recorded in together in the effect factors of surrounding
Under step effect, the direction being directed toward to arrow is mobile, repeatedly after effect, effect factors record in each ash chromosphere by
Assemble together, respectively accumulation point P1、P2.And effect factors record x2And x4Surrounding is remembered without effect factors
Record, then be always maintained at constant, is directly expressed as accumulation point P3、P4, final all records are up to a stable state, as schemed
Shown in 2 (b).
5.2.4), finally for interaction after data set, represented using accumulation point or synchronous point initial data i.e. as
One effect factors record, so that complete paired data stream is effectively compressed.Meanwhile in compressed data set
It is each, store its feature vector xcAnd the runoff mean value y of its corresponding original data recordc.Calculation formula is as follows:
Wherein xcFor the corresponding feature vector of c-th of synchronous point, CiFor xcThe collection of the included reset condition record of synchronous point
It closes, NcFor CiIn state recording number, yiFor xiCorresponding true diameter flow valuve.Finally by synchronous compression, we are available
4 data points, the form of expression as shown in Fig. 2 (b) are as follows:
D={ (xc,yc) | c=1,2,3,4 }
It mode obtained based on synchronous compression in view of this can use new effect factors record and re-compressed, institute
Can infinitely be compressed in principle.Therefore, by synchronous compression, potential unlimited and real-time effect factors can be remembered
Record is managed, real-time servicing representative sample data collection.
5.3), the differentiation of relationship detects between effect factors and runoff
Due to climate change and the continuous variation of mankind's activity, the relation schema of effect factors and runoff also can be with
Dynamic change, cause current runoff model that may be very different with history runoff model.Therefore, under these conditions, make
It cannot need to carry out concept drift detection based on historical data with the prediction that new effect factors record carries out runoff,
To safeguard the dynamic data set for representing current effect factors and runoff relationship.
In the present invention, in entire dynamic data maintenance process, detected using the concept drift based on Statisti-cal control process
Whether method, the relation schema that detection representative sample data is concentrated are mutated.Specifically, in the whole process, dimension
The consecutive data block of a fixed size window is protected, which is arranged according to concrete scene.Count number in the data block
According to prediction error Loss mean value and standard deviation, if a certain moment prediction error be more than defined threshold value, then it is assumed that runoff
Relationship between impact factor and runoff is mutated.If mutating, representative sample data collection is emptied, and utilize
Effect factors record in the data block is reinitialized.
In the present embodiment, concept drift detection method is specific as follows:
5.3.1), effect factors being recorded, Dynamic Maintenance simultaneously records the prediction error in a consecutive data block,
The size of data block is depending on concrete application;
5.3.2), to the prediction error Loss of the every data of data blockiIt is counted, calculates error mean μ and standard
Poor σ.Wherein
Lossi=ypred-yreal
In the present embodiment, the loss Loss at current time is calculatedkIf the error amount and number of the moment Runoff Forecast
It is greater than three times mean square deviation according to the difference of the error mean μ in block, it may be assumed that | Lossk- μ | > 3 σ, then it is assumed that the moment runoff influence because
Relationship between son and runoff is integrally mutated, so historical data concentration has not been suitable for current Runoff Forecast.
Therefore by clear history data set, initialization is then re-started.
Fig. 3 is that the present invention is based on a kind of diameters of specific embodiment of runoff real-time predicting method of synchronous data flow compression
Flow the system framework figure of prediction.
As shown in figure 3, including the following steps that (1) works as in representative sample set when the system that runoff is predicted in real time is run
In find and recorded with data to be predicted apart from nearest K item;(2) using the K item record building Lasso linear regression mould selected
Type simultaneously treats prediction data progress Runoff Forecast;(3) current representative sample data collection data volume is excessive, synchronizes pressure
Contracting;(4) current data set has large change, then reinitializes data set.
Although the illustrative specific embodiment of the present invention is described above, in order to the skill of the art
Art personnel understand the present invention, it should be apparent that the present invention is not limited to the ranges of specific embodiment, to the general of the art
For logical technical staff, if various change in the spirit and scope of the present invention that the attached claims limit and determine,
These variations are it will be apparent that all utilize the innovation and creation of present inventive concept in the column of protection.
Claims (2)
1. a kind of runoff real-time predicting method based on synchronous data flow compression, which comprises the following steps:
(1), each meteorological site and the closely related feature of runoff in collection research basin;
(2), data prediction counts the data that each website is collected into, and carries out completion to missing values;
(3), it initializes one and records the representative sample data formed by certain item number history effect factors and run-off
Collection;
(4), for the feature of the current effect factors got, it is nearest that distance is found in representative sample set
K neighbours based on K item record building Lasso linear regression model (LRM), and predict current runoff using the model, In
After getting true diameter flow valuve, which is inserted into representative sample set;
(5), dynamic data set safeguard: when there is new runoff to be predicted to be predicted and get the moment true diameter flow valuve it
Afterwards, the sample weights in representative sample data set are updated based on prediction deviation;When Relative Error is less than one
When determining threshold value, it is believed that prediction is correct, and the weight for participating in K nearest-neighbors data of prediction increases;Conversely, participating in K of prediction
The weight of nearest-neighbors data reduces;After having updated weight every time, by the lesser point of weight in representative sample set
It removes;
If representative sample data collection scale exceeds specified size, specific threshold value can be true according to the hardware capabilities of operating system
It is fixed, it is compressed using based on synchrodata flow compression method;
Meanwhile using the concept drift detection method based on Statisti-cal control process, the runoff that representative sample data is concentrated is detected
Whether the relation schema of impact factor and runoff is developed, and in case of developing, then empties representative sample data collection,
Then it is initialized according to step (3), carries out Runoff Forecast according still further to step (4).
2. runoff real-time predicting method according to claim 1, which is characterized in that in step (4), the selection is represented
Property sample set carry out runoff real-time prediction and step (5) in, it is described to representative sample set carry out dynamic data set
Maintenance:
2.1) the real-time prediction that representative sample set carries out runoff, is chosen;
It is nearest in representative sample data set that the observation x of the current time effect factors of acquisition is found first
K data point, that is, history effect factors record x1、x2、…、xkAnd corresponding runoff records y1、y2、…、yk, enable Y=
(y1,y2,…,yk)T, X=(x1,x2,…,xk)T, then calculate as follows
Wherein,For Lasso linear regression model (LRM) coefficient, α is penalty factor, and d is the characteristic dimension of effect factors, X and Y
The influence of distance moment effect factors observational characteristic to be predicted is nearest respectively in representative sample set K item record because
The matrix of sub- observation and runoff measured value composition;
After obtaining Lasso linear regression model (LRM) coefficient, for the observation x of current time effect factors, under utilization
The predicted value y of face formula calculating current time runoff:
2.2) dynamic data set maintenance, is carried out to representative sample set;
Step 1: being updated to representative sample set;
Firstly, enabling yrealFor the true diameter flow valuve for being predicted the moment, y is enabledpredFor the prediction diameter flow valuve for being predicted the moment, quilt is calculated
The Relative Error of prediction time runoff:
After getting the prediction deviation l for being predicted runoff, the update of representative sample weight is carried out using following rule, method is such as
Under:
If first, prediction error l is less than or equal to 0.1, it is believed that prediction is accurate, and the K predicted is participated in representative sample recently
Neighbor weight w1、w2、…、wkRespectively plus one;
If second, prediction error l is greater than 0.1, it is believed that prediction error participates in K nearest neighbours of prediction in representative sample
Weight w1、w2、…、wkSubtract one respectively;
Then the representative sample set after updating for weight, the representative sample data by wherein weight less than -3 is from generation
It is removed in table sample;
Step 2: representative sample set Real Time Compression;
Each effect factors record in data set (is indicated into x with feature vector firsti) it is considered as the one of characteristic vector space
Point or an object;
The neighbor objects Nb for being then ε by each object and distanceε(xi) interact, interaction models are as follows:
WhereinIndicate that the i-th effect factors record xiThe value at (t+1) moment in jth dimension;Nbε(xi) indicate
X is recorded with effect factorsiCentered on (a bit of characteristic vector space), the range of Euclidean distance ε removes xiOuter data point set
It closes, | Nbε(x) | indicate Nbε(xi) include effect factors record item number;
By repeatedly interacting, similar effect factors record be will accumulate in together, and value having the same is finally directed to
Data set after interaction is represented initial data using accumulation point, synchronous point and recorded as an effect factors, thus
Complete paired data stream is effectively compressed, and for each point in compressed data set, stores its feature vector xcAnd it
The runoff mean value y of corresponding original data recordc, calculation formula is as follows:
Wherein xcFor the corresponding feature vector of c-th of synchronous point, CiFor xcThe set of the included reset condition record of synchronous point, NcFor
CiIn state recording number, yiFor xiCorresponding true diameter flow valuve;
It mode obtained based on synchronous compression in view of this can use new effect factors record and re-compressed, therefore, led to
Synchronous compression is crossed, potential unlimited and real-time effect factors can be recorded and be managed, real-time servicing representative sample
Data set;
Step 3: the differentiation of relationship detects between effect factors and runoff;
Effect factors are recorded first, Dynamic Maintenance simultaneously records the prediction error in a consecutive data block, data block
Size is depending on concrete application;
Then to the prediction error Loss of the every data of data blockiIt is counted, calculates error mean μ and standard deviation sigma, wherein
Lossi=ypred-yreal
Calculate the loss Loss at current timekIf error mean μ's in the error amount and data block of the moment Runoff Forecast
Difference is greater than three times meansquaredeviationσ, it may be assumed that | Lossk- μ | > 3 σ, then it is assumed that the relationship between the moment effect factors and runoff
Entirety is mutated, so historical data concentration has not been suitable for current Runoff Forecast, therefore by clear history data
Collection, then re-starts initialization.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910564887.XA CN110414713A (en) | 2019-06-27 | 2019-06-27 | A kind of runoff real-time predicting method based on synchronous data flow compression |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910564887.XA CN110414713A (en) | 2019-06-27 | 2019-06-27 | A kind of runoff real-time predicting method based on synchronous data flow compression |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110414713A true CN110414713A (en) | 2019-11-05 |
Family
ID=68358396
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910564887.XA Pending CN110414713A (en) | 2019-06-27 | 2019-06-27 | A kind of runoff real-time predicting method based on synchronous data flow compression |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110414713A (en) |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP3029149A1 (en) * | 2014-12-03 | 2016-06-08 | Fundacio Centre de Regulacio Genomica | Method of the identification of gene regulation targets |
US20170103148A1 (en) * | 2014-03-27 | 2017-04-13 | Nec Corporation | System-analyzing device, analysis-model generation method, system analysis method, and system-analyzing program |
CN106709588A (en) * | 2015-11-13 | 2017-05-24 | 日本电气株式会社 | Prediction model construction method and equipment and real-time prediction method and equipment |
JP2017130116A (en) * | 2016-01-21 | 2017-07-27 | 日本電信電話株式会社 | Regression analysis device, regression analysis method, and regression analysis program |
CN107292098A (en) * | 2017-06-15 | 2017-10-24 | 河海大学 | Medium-and Long-Term Runoff Forecasting method based on early stage meteorological factor and data mining technology |
US20170364933A1 (en) * | 2014-12-09 | 2017-12-21 | Beijing Didi Infinity Technology And Development Co., Ltd. | User maintenance system and method |
US20180373993A1 (en) * | 2017-06-23 | 2018-12-27 | University Of Alaska Fairbanks | Method Of Predicting Streamflow Data |
CN109272146A (en) * | 2018-08-23 | 2019-01-25 | 河海大学 | A kind of Forecasting Flood method corrected based on deep learning model and BP neural network |
CN109388847A (en) * | 2018-08-24 | 2019-02-26 | 河海大学 | A kind of streamflow change attribution technological synthesis selection method |
-
2019
- 2019-06-27 CN CN201910564887.XA patent/CN110414713A/en active Pending
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170103148A1 (en) * | 2014-03-27 | 2017-04-13 | Nec Corporation | System-analyzing device, analysis-model generation method, system analysis method, and system-analyzing program |
EP3029149A1 (en) * | 2014-12-03 | 2016-06-08 | Fundacio Centre de Regulacio Genomica | Method of the identification of gene regulation targets |
US20170364933A1 (en) * | 2014-12-09 | 2017-12-21 | Beijing Didi Infinity Technology And Development Co., Ltd. | User maintenance system and method |
CN106709588A (en) * | 2015-11-13 | 2017-05-24 | 日本电气株式会社 | Prediction model construction method and equipment and real-time prediction method and equipment |
JP2017130116A (en) * | 2016-01-21 | 2017-07-27 | 日本電信電話株式会社 | Regression analysis device, regression analysis method, and regression analysis program |
CN107292098A (en) * | 2017-06-15 | 2017-10-24 | 河海大学 | Medium-and Long-Term Runoff Forecasting method based on early stage meteorological factor and data mining technology |
US20180373993A1 (en) * | 2017-06-23 | 2018-12-27 | University Of Alaska Fairbanks | Method Of Predicting Streamflow Data |
CN109272146A (en) * | 2018-08-23 | 2019-01-25 | 河海大学 | A kind of Forecasting Flood method corrected based on deep learning model and BP neural network |
CN109388847A (en) * | 2018-08-24 | 2019-02-26 | 河海大学 | A kind of streamflow change attribution technological synthesis selection method |
Non-Patent Citations (1)
Title |
---|
刘志辉: "一个适于新疆河流日径流预测的实时预报模型", 《新疆大学学报(自然科学版)》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Li et al. | Effects of urbanization on direct runoff characteristics in urban functional zones | |
Apreda et al. | A climate vulnerability and impact assessment model for complex urban systems | |
Chen | Prediction algorithm of PM2. 5 mass concentration based on adaptive BP neural network | |
Zhang et al. | Using Markov chains to analyze changes in wetland trends in arid Yinchuan Plain, China | |
Guo et al. | A multi-variable grey model with a self-memory component and its application on engineering prediction | |
CN106685427B (en) | A kind of sparse signal reconfiguring method based on consistency on messaging | |
Kalyankar et al. | Data mining technique to analyse the metrological data | |
CN102830446B (en) | Intelligent meteorological station system capable of forecasting meteorological data | |
Das et al. | FORWARD: a model for forecasting reservoir water dynamics using spatial Bayesian network (SpaBN) | |
Jie et al. | Selecting proper method for groundwater interpolation based on spatial correlation | |
Zhou et al. | Evaluating spatiotemporal variation of groundwater depth/level in Beijing plain, a groundwater-fed area from 2001 to 2010 | |
Peng et al. | Research on the prediction of the water demand of construction engineering based on the BP neural network | |
Jeihouni et al. | Response of the Shabestar Plain aquifer to climate-change scenarios through statistical and hybrid soft computing techniques | |
Muluye | Comparison of statistical methods for downscaling daily precipitation | |
Udo et al. | Climate change impacts on runoff regimes at a river basin scale in Central Vietnam | |
Xiaoxun et al. | Research on wind speed behavior prediction method based on multi-feature and multi-scale integrated learning | |
Hou et al. | Spatial simulation and prediction of air temperature based on cnn-lstm | |
Bashar et al. | Investigation of factors affecting rural drinking water consumption using intelligent hybrid models | |
Zhiming et al. | Risk evaluation of urban rainwater system waterlogging based on neural network and dynamic hydraulic model | |
CN202735528U (en) | Intelligent meteorological station system capable of predicting meteorological data | |
CN105808867B (en) | A method of the building height optimization design based on simulated annealing | |
CN110414713A (en) | A kind of runoff real-time predicting method based on synchronous data flow compression | |
Liu et al. | Effect of Land Use and Drainage System Changes on Urban Flood Spatial Distribution in Handan City: A Case Study | |
Shirisha et al. | Real-time flow forecasting in a watershed using rainfall forecasting model and updating model | |
CN114707753B (en) | Regional LSTM flood forecasting method |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20191105 |
|
WD01 | Invention patent application deemed withdrawn after publication |