CN109615027A - A kind of line of high-speed railway wind speed feature extraction intelligent Forecasting - Google Patents
A kind of line of high-speed railway wind speed feature extraction intelligent Forecasting Download PDFInfo
- Publication number
- CN109615027A CN109615027A CN201811643482.7A CN201811643482A CN109615027A CN 109615027 A CN109615027 A CN 109615027A CN 201811643482 A CN201811643482 A CN 201811643482A CN 109615027 A CN109615027 A CN 109615027A
- Authority
- CN
- China
- Prior art keywords
- wind speed
- model
- forecasting
- value
- feature
- 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
- 238000000605 extraction Methods 0.000 title claims abstract description 10
- 239000013598 vector Substances 0.000 claims abstract description 103
- 238000012216 screening Methods 0.000 claims abstract description 56
- 238000000034 method Methods 0.000 claims abstract description 38
- 238000013480 data collection Methods 0.000 claims abstract description 20
- 238000013528 artificial neural network Methods 0.000 claims description 46
- 238000012549 training Methods 0.000 claims description 46
- 239000000203 mixture Substances 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 8
- 238000001914 filtration Methods 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 7
- 239000012141 concentrate Substances 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 238000003064 k means clustering Methods 0.000 claims description 5
- 241001269238 Data Species 0.000 claims description 3
- 235000009413 Ratibida columnifera Nutrition 0.000 claims description 3
- 241000510442 Ratibida peduncularis Species 0.000 claims description 3
- 230000003111 delayed effect Effects 0.000 claims description 3
- 238000003780 insertion Methods 0.000 claims description 3
- 230000037431 insertion Effects 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 230000001537 neural effect Effects 0.000 claims description 3
- 210000005036 nerve Anatomy 0.000 claims description 2
- 238000003825 pressing Methods 0.000 claims description 2
- 238000004458 analytical method Methods 0.000 abstract description 4
- 239000000284 extract Substances 0.000 abstract description 3
- 238000010219 correlation analysis Methods 0.000 abstract description 2
- 238000010801 machine learning Methods 0.000 description 4
- 238000003062 neural network model Methods 0.000 description 4
- 238000013277 forecasting method Methods 0.000 description 3
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical compound [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 229910052742 iron Inorganic materials 0.000 description 1
- 238000012067 mathematical method Methods 0.000 description 1
- 238000000053 physical method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012731 temporal analysis Methods 0.000 description 1
- 238000000700 time series analysis Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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"
-
- G06Q50/40—
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Economics (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Strategic Management (AREA)
- Evolutionary Biology (AREA)
- Human Resources & Organizations (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (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 invention discloses a kind of line of high-speed railway wind speed feature extraction intelligent Forecastings, this method carries out K-means cluster using wind speed different characteristic, and select optimum clustering number K, Wind speed model sample is divided into K class, then 100 forecasting wind speed feature pre-selection models are established respectively for K cluster classification, using mathematical analysis and correlation analysis screening forecasting wind speed feature pre-selection model and establish K forecasting wind speed characteristic model group and K forecasting wind speed normalizing model, ultimate analysis prediction wind velocity vector and wind speed time series data collection between correlation and restore true wind speed, this method has pre- altimetry precision, feature intelligent extracts, it is adaptable, the high feature of robustness, it is suitably applied the fields such as the line of high-speed railway safety under strong wind atmosphere.
Description
Technical field
The present invention relates to a kind of line of high-speed railway wind speed feature extraction intelligent Forecastings.
Background technique
In recent years, forecasting wind speed rapidly becomes one of domestic and international popular research field.Wind speed forecasting method is in strong wind atmosphere
Under line of high-speed railway secure context play the role of increasing and extensive use.Wind speed forecasting method is applied to big vane
Along Railway safety under border, the following wind speed of accurate prediction, which can provide more to give warning in advance for train, in real time handles the time, is formed
Safe driving scheme guarantees traffic safety.
Wind speed will receive the influence of the factors such as season, temperature, height above sea level, with very strong randomness and complicated non-linear
Feature.Wind speed forecasting method mainly has learning method, physical method and statistical method at present.Machine learning method obtains in recent years
Predicted that preferably common model has to fast development, therefore using wind speed of the machine learning method to complex nonlinear
Artificial neural network, Kalman filtering, time series analysis etc..
To solve the problems, such as that the non-stationary in forecasting wind speed is nonlinear, wind speed feature is effectively extracted, wind speed is carried out high-precision
The prediction of degree, urgent need provide a kind of line of high-speed railway wind speed feature extraction intelligent Forecasting, realize the high-precision to wind speed,
The prediction of strong adaptability, high robust.
Summary of the invention
The present invention provides a kind of line of high-speed railway wind speed feature extraction intelligent Forecastings, it is intended that passing through
Wind speed characteristics under clustering algorithm and machine learning algorithm analysis line of high-speed railway strong wind atmosphere, and the wind speed feature extracted, build
A variety of neural network models are found, line of high-speed railway wind speed is accurately and effectively predicted.
A kind of line of high-speed railway wind speed feature extraction intelligent Forecasting, comprising the following steps:
Step 1: original wind speed time series data collection is obtained by air measuring station；
Air measuring station is set in line of high-speed railway designated position, original wind speed time series data is obtained by air measuring station
Collection；
The original wind speed time series data collection is made of n original air speed datas altogether；When wherein Δ t is wind-speed sample
Between be spaced；N is sampling number, and n is at least more than 500；
Step 2: establishing Wind speed model sample A；
The Wind speed model sample A includes Wind speed model training sample AaWith Wind speed model Screening Samples Ab；
Establishing Wind speed model sample A, detailed process is as follows:
Step A1: Kalman filtering is carried out to original wind speed time series data collection, obtains filtered wind speed time sequence
Column data collection and Time sequence of noise data set；
Step A2: it concentrates the wind speed at 4 moment of arbitrary continuation as one group of wind velocity vector wind speed time series data, obtains
To the Wind speed model sample A for including n-3 group wind velocity vector；
The air speed value of synchronization is likely to occur in the different wind velocity vectors；
Step A3: pressing sample time order, and from Wind speed model sample A, the wind velocity vector for choosing preceding 3/4 forms wind speed mould
Type training sample Aa, remaining wind velocity vector composition Wind speed model Screening Samples Ab；
Step 3: K-means clustering method is used, each group of wind velocity vector in Wind speed model sample A is clustered,
Optimum clustering number K is chosen, K cluster centre time series is obtained；
Step 4: according to the Wind speed model sample A for clustering classification for K has been gathered, by Wind speed model training sample AaIt is divided into
The Wind speed model training subsample { A of K corresponding cluster classificationa,1,Aa,2,…,Aa,K, by Wind speed model Screening Samples AbIt is divided into K
A corresponding Wind speed model screens subsample { Ab,1,Ab,2,…,Ab,K}；
Step 5: for K Wind speed model training subsample, it is special to establish forecasting wind speed of the K kind based on Elman neural network
Sign pre-selection each 100, model；
The forecasting wind speed feature pre-selection model output value based on Elman neural network is named as the pre-selection of wind speed feature
Value；
Step 6: K kind is preselected model each 100 based on the forecasting wind speed feature of Elman neural network respectively by screening model
A screening is to each 10；
Step 7: 10 forecasting wind speed features pre-selection model group of every kind of cluster classification is obtained at wind speed predicted characteristics model group
To K forecasting wind speed characteristic model group；
The forecasting wind speed characteristic model group includes that 10 forecasting wind speed features preselect model, each forecasting wind speed character modules
Type group output data groups are named as wind speed feature vector at the vector that length is 10；
Step 8: being directed to K forecasting wind speed characteristic model group, establish the K forecasting wind speed normalizings based on wavelet neural network
Model；
The forecasting wind speed normalizing model output value is set as wind speed normalizing value；
Step 9: real-time prediction of wind speed；
Acquisition current time and recent air speed data, data sampling time interval should keep using with step 1 in real time
Original wind speed time series data collection wind-speed sample time interval Δ t it is consistent, carry out the detailed process of real-time prediction of wind speed
It is as follows:
Step B1: setting forecasting wind speed object time T is obtained and object time T immediate at least 200 by air measuring station
The original air speed value at a moment forms original wind speed time series, enters step B2；
Step B2: Kalman filtering is carried out to original wind speed time series, obtains filtered wind speed time series, is entered
Step B3；
Step B3: it is obtained in calculation of wind speed time series with the air speed value at object time T immediate 4 moment and step 3
The Euclidean distance of K cluster centre time series out, to select the forecasting wind speed of corresponding cluster classification special apart from minimum principle
Levy model group, using in wind speed time series with the air speed value at object time T immediate 3 moment be used as input data, input
Into forecasting wind speed characteristic model group, output obtains the wind speed feature vector that length is 10, enters step B4；
Step B4: using wind speed feature vector as input data, it is input to the forecasting wind speed normalizing mould of corresponding cluster classification
Type, output wind speed normalizing value, enters step B5；
Step B5: whether the moment where judgment step B4 output wind speed normalizing value is greater than object time T, if so, entering step
Otherwise rapid B7 enters step B6；
Step B6: the wind speed normalizing value that step B4 is exported is inserted into wind speed time series first place, updates wind speed time sequence
Column, return step B3；
Step B7: in calculation of wind speed time series with the air speed value at object time T immediate 100 moment composition to
Between the vector of the air speed value composition at 100 moment of wind speed time series data collection arbitrary continuation obtained in amount and step 2
Pearson correlation coefficient, select the highest wind speed time series data of Pearson correlation coefficient to concentrate one group of 100 moment
The vector of air speed value composition, when the filter noise that the Time sequence of noise data set in step 2 corresponds to the moment is attached to wind speed
Between in sequence and object time T immediate 100 moment, i.e. reduction prediction of wind speed authenticity completes the wind speed of object time
Prediction.
Further, it is described use K-means clustering method, to each group of wind velocity vector in Wind speed model sample A into
Row cluster, chooses optimum clustering number K, obtaining K cluster centre time series, detailed process is as follows:
Step 2.1: setting K-means cluster numbers K ', K '=1；
Step 2.2: using Euclidean distance as similarity measurements flow function, to each group of wind speed in Wind speed model sample A
Vector is clustered, and a cluster classification of K ' and a cluster centre time series of K ' are obtained；
The calculation formula of the Euclidean distance isWherein, X=
(x1,x2,x3,x4), Y=(y1,y2,y3,y4) it is respectively two different wind velocity vectors, d (X, Y) is wind velocity vector X, between Y
Euclidean distance；
Step 2.3: calculating the error sum of squares SSE (K ') of K-means cluster；
Wherein, K ' is cluster numbers, CiIt is ith cluster classification, p is CiOne group of wind velocity vector in class, miIt is CiClass is gathered
Class center；
Step 2.4: judging whether K ' value is equal to 20, if so, entering step 2.5, otherwise, enable K '=K '+1, then return
Step 2.2；
Step 2.5: each α (K ') as K '=2 to 19 and β (K ') value are calculated separately, subsequently into step 2.6；
α (K ')=arctan (SSE (K ')-SSE (K '+1)), α ∈ (0 °, 90 °), K ' ∈ [2,19]
β (K ')=arctan (SSE (K ')-SSE (K ' -1)), β ∈ (90 °, 180 °), K ' ∈ [2,19]；
Step 2.6: calculating separately each α (K ')+β (K ') value as K '=2 to 19, choose so that α (K ')+β (K ')
K ' value when maximum obtains K cluster centre time series as optimum clustering number K.
Further, described for K Wind speed model training subsample, establish wind speed of the K kind based on Elman neural network
Predicted characteristics preselect each 100 processes of model are as follows: successively will be before every group of wind velocity vector in K Wind speed model training subsample
The air speed value at 3 moment trains Elman neural network as output data as input data, the air speed value at the 4th moment,
The Elman neural network obtained for each Wind speed model training subsample training is randomly provided 100 kinds of different values respectively
Elman neural network parameter obtains K kind based on the forecasting wind speed feature of Elman neural network and preselects each 100, model, altogether
The 100*K forecasting wind speed features based on Elman neural network preselect model.
Further, the 100*K forecasting wind speed feature pre-selection models based on Elman neural network are equal in training
It is 3 that input layer number, which is arranged, and it is [5,10] that range, which is arranged, in hidden layer node number, and output layer node number is 1, accepts layer
Node number is 7, implies layer functions and uses Sigmoid function, the training of network uses BP algorithm, and maximum number of iterations is set as
800, training learning rate setting range is [0.001,0.5], threshold value 0.005.
Further, the screening model, the respectively pre- modeling of forecasting wind speed feature by K kind based on Elman neural network
Detailed process is as follows to each 10 for each 100 screenings of type:
Step 5.1: K Wind speed model successively being screened into subsample { Ab,1,Ab,2,…,Ab,KIn every group of wind velocity vector before
The air speed value at 3 moment is separately input to 100 forecasting wind speed features pre-selection model of corresponding cluster classification as input data
In；
Step 5.2: preliminary screening forecasting wind speed feature preselects model；
Successively calculate what the wind speed feature preset value that each forecasting wind speed feature pre-selection model exports chronologically formed
One-dimensional vector and Wind speed model Screening Samples AbThe mean square error of the one-dimensional vector of the air speed value composition at middle corresponding moment, is eliminated equal
The forecasting wind speed feature that square error is greater than mean square error mean value preselects model；
Step 5.3: establishing screening vector；
It is utilized respectively in K cluster centre time series and establishes K with the air speed value at immediate 3 moment at current time
The screening vector that length is 3；
Step 5.4: updating screening vector；
Respectively using the air speed data at 3 moment before K screening vector as input data, it is input to corresponding cluster classification
Forecasting wind speed feature preselects in model, and vector first position is screened in the wind speed feature preset value insertion at the 4th moment of output,
Remainder data sequence of positions is successively delayed, and repeats this step until K screening vector length reaches 200；
Step 5.5: screening forecasting wind speed feature again and preselect model；
Calculate separately the skin of the obtained screening vector of forecasting wind speed feature pre-selection model of every kind of cluster classification between any two
The inferior related coefficient of that, obtains K Pearson correlation coefficient matrix, calculates separately between each screening vector and other screening vectors
Pearson correlation coefficient accumulated value, it is pre- to be phased out the maximum corresponding forecasting wind speed feature of Pearson correlation coefficient accumulated value
Modeling type, the forecasting wind speed feature pre-selection model that classification is clustered until every kind are 10 remaining.
Further, described to be directed to K forecasting wind speed characteristic model group, establish the K wind speed based on wavelet neural network
Predict that detailed process is as follows for normalizing model:
Step 6.1: successively using the air speed value at 3 moment before every group of wind velocity vector in Wind speed model sample A as input
Data are input in the forecasting wind speed characteristic model group of corresponding cluster classification, and output obtains wind speed feature vector；
Step 6.2: the wind speed feature vector for successively obtaining step 6.1 is right in Wind speed model sample A as input data
The air speed value at the 4th moment answered is obtained as output data, training wavelet neural network for K forecasting wind speed character modules
The K forecasting wind speed normalizing models based on wavelet neural network of type group.
Further, the forecasting wind speed normalizing model based on wavelet neural network is respectively provided with input layer section in training
Point number is 10, and hidden layer Wavelet Element number is 20, and output layer node number is 1；Hidden layer wavelet neural member uses Mexican
Hat wavelet function, output node layer use Sigmoid function, and the maximum number of iterations in training process is set as 500, training
Learning rate is 0.04, threshold value 0.002.
Beneficial effect
The present invention provides a kind of line of high-speed railway wind speed feature extraction intelligent Forecasting, this method is directed to high-speed iron
Complicated, non-linear, non-stationary wind speed under curb line strong wind atmosphere, by clustering algorithm and machine learning algorithm, to wind speed number
According to clustering is carried out, Wind speed model sample is divided into K class, and to every kind of clustering cluster by the cluster numbers K for selecting Clustering Effect best
Wind speed model sample establish 100 forecasting wind speed features pre-selection models, it is good to distinguish screening effect by Mathematical Method
10 model groups build up K forecasting wind speed characteristic model group and extract wind speed feature, then establish K forecasting wind speed normalizing model,
The feature output valve of forecasting wind speed characteristic model group is subjected to normalizing and restores the authenticity of wind speed finally by correlation analysis；
This method is clustered air speed data using clustering algorithm, and individually constructs a variety of differences for every kind of clustering cluster
Neural network model efficiently extracts wind speed feature；It is directed to wind speed different characteristic type during forecasting wind speed and carries out cluster point
Analysis, establishes forecasting wind speed characteristic model group, finally exports to wind speed feature and carries out normalizing, it is possible to prevente effectively from the non-stationary of wind speed
Property variation there is pre- altimetry precision, feature intelligent to extract, spy adaptable, robustness is high to find out wind speed characteristic value
Point is suitably applied the fields such as the line of high-speed railway safety under strong wind atmosphere.
Detailed description of the invention
Fig. 1 is the flow diagram of the method for the invention.
Specific embodiment
Below in conjunction with attached drawing and example, the present invention is described further.
As shown in Figure 1, a kind of line of high-speed railway wind speed feature extraction intelligent Forecasting, comprising the following steps:
Step 1: original wind speed time series data collection is obtained by air measuring station；
Air measuring station is set in line of high-speed railway designated position, original wind speed time series data collection is obtained by air measuring station
W (t)=[w (t=Δ t), w (t=2* Δ t) ..., w (t=n* Δ t)]；
The original wind speed time series data collection is made of n original air speed datas altogether；When wherein Δ t is wind-speed sample
Between be spaced；N is sampling number, and n is at least more than 500；
Step 2: establishing Wind speed model sample A；
The Wind speed model sample A includes Wind speed model training sample AaWith Wind speed model Screening Samples Ab；
Establishing Wind speed model sample A, detailed process is as follows:
Step A1: Kalman filtering is carried out to original wind speed time series data collection, obtains filtered wind speed time sequence
Column data collection w ' (t)=[w ' (t=Δ t), w ' (t=2* Δ t) ..., w ' (t=n* Δ t)] and Time sequence of noise data set
No (t)=[no (t=Δ t), no (t=2* Δ t) ..., no (t=n* Δ t)]；
Step A2: it concentrates the wind speed at 4 moment of arbitrary continuation as one group of wind velocity vector wind speed time series data, obtains
To the Wind speed model sample A for including n-3 group wind velocity vector；
The air speed value of synchronization is likely to occur in the different wind velocity vectors, the Wind speed model sample A is indicated are as follows:
Wherein, every a line in Wind speed model sample A indicates one group of wind velocity vector, total n-3 group wind velocity vector；
Step A3: sample time order is pressed, from Wind speed model sample A, before selectionGroup wind velocity vector group becomes a common practice
Fast model training sample Aa, remaining wind velocity vector composition Wind speed model Screening Samples Ab；
Step 3: K-means clustering method is used, each group of wind velocity vector in Wind speed model sample A is clustered,
Optimum clustering number K is chosen, obtains K cluster centre time series, detailed process is as follows:
Step 2.1: setting K-means cluster numbers K ', K '=1；
Step 2.2: using Euclidean distance as similarity measurements flow function, to each group of wind speed in Wind speed model sample A
Vector is clustered, and a cluster classification of K ' and a cluster centre time series of K ' are obtained；
The calculation formula of the Euclidean distance isWherein, X=
(x1,x2,x3,x4), Y=(y1,y2,y3,y4) it is respectively two different wind velocity vectors, d (X, Y) is wind velocity vector X, between Y
Euclidean distance；
Step 2.3: calculating the error sum of squares SSE (K ') of K-means cluster；
Wherein, K ' is cluster numbers, CiIt is ith cluster classification, p is CiOne group of wind velocity vector in class, miIt is CiClass is gathered
Class center；
Step 2.4: judging whether K ' value is equal to 20, if so, entering step 2.5, otherwise, enable K '=K '+1, then return
Step 2.2；
Step 2.5: each α (K ') as K '=2 to 19 and β (K ') value are calculated separately, subsequently into step 2.6；
α (K ')=arctan (SSE (K ')-SSE (K '+1)), α ∈ (0 °, 90 °), K ' ∈ [2,19]
β (K ')=arctan (SSE (K ')-SSE (K ' -1)), β ∈ (90 °, 180 °), K ' ∈ [2,19]；
Step 2.6: calculating separately each α (K ')+β (K ') value as K '=2 to 19, choose so that α (K ')+β (K ')
K ' value when maximum obtains K cluster centre time series as optimum clustering number K；
Step 4: according to the Wind speed model sample A for clustering classification for K has been gathered, by Wind speed model training sample AaIt is divided into
The Wind speed model training subsample { A of K corresponding cluster classificationa,1,Aa,2,…,Aa,K, by Wind speed model Screening Samples AbIt is divided into K
A corresponding Wind speed model screens subsample { Ab,1,Ab,2,…,Ab,K}；
Step 5: for K Wind speed model training subsample, it is special to establish forecasting wind speed of the K kind based on Elman neural network
Sign pre-selection each 100, model；
The forecasting wind speed feature pre-selection model output value based on Elman neural network is named as the pre-selection of wind speed feature
Value；
It is described to train subsample for K Wind speed model, establish forecasting wind speed feature of the K kind based on Elman neural network
Preselect each 100 processes of model are as follows: successively by 3 moment before every group of wind velocity vector in K Wind speed model training subsample
Air speed value as input data, the air speed value at the 4th moment is as output data, training Elman neural network, for each
The Elman neural network that Wind speed model training subsample training obtains is randomly provided the Elman nerve of 100 kinds of different values respectively
Network parameter obtains K kind based on the forecasting wind speed feature of Elman neural network and preselects each 100, model, and total 100*K are based on
The forecasting wind speed feature of Elman neural network preselects model, indicates are as follows:
Wherein, M_Elman indicates that the 100*K forecasting wind speed features based on Elman neural network preselect model, every a line
Indicate 100 forecasting wind speed features pre-selection model of same cluster type, total K row；
The 100*K forecasting wind speed feature pre-selection models based on Elman neural network are respectively provided with input in training
Node layer number is 3, and it is [5,10] that range, which is arranged, in hidden layer node number, and output layer node number is 1, accepts node layer number
It is 7, implies layer functions and use Sigmoid function, the training of network uses BP algorithm, and maximum number of iterations is set as 800, training
It is [0.001,0.5], threshold value 0.005 that range, which is arranged, in learning rate；
Step 6: K kind is preselected model each 100 based on the forecasting wind speed feature of Elman neural network respectively by screening model
A screening is to each 10, and detailed process is as follows:
Step 5.1: K Wind speed model successively being screened into subsample { Ab,1,Ab,2,…,Ab,KIn every group of wind velocity vector before
The air speed value at 3 moment is separately input to 100 forecasting wind speed features pre-selection model of corresponding cluster classification as input data
In；
Step 5.2: preliminary screening forecasting wind speed feature preselects model；
Successively calculate what the wind speed feature preset value that each forecasting wind speed feature pre-selection model exports chronologically formed
One-dimensional vector and Wind speed model Screening Samples AbThe mean square error of the one-dimensional vector of the air speed value composition at middle corresponding moment, is eliminated equal
The forecasting wind speed feature that square error is greater than mean square error mean value preselects model；
Step 5.3: establishing screening vector；
It is utilized respectively in K cluster centre time series and establishes K with the air speed value at immediate 3 moment at current time
The screening vector that length is 3；
Step 5.4: updating screening vector；
Respectively using the air speed data at 3 moment before K screening vector as input data, it is input to corresponding cluster classification
Forecasting wind speed feature preselects in model, and vector first position is screened in the wind speed feature preset value insertion at the 4th moment of output,
Remainder data sequence of positions is successively delayed, and repeats this step until K screening vector length reaches 200；
Step 5.5: screening forecasting wind speed feature again and preselect model；
Calculate separately the skin of the obtained screening vector of forecasting wind speed feature pre-selection model of every kind of cluster classification between any two
The inferior related coefficient of that, obtains K Pearson correlation coefficient matrix P1,P2,…,PK, i-th of Pearson correlation coefficient matrix PiTable
It is shown as:
Wherein, PiThe screening vector that the forecasting wind speed feature pre-selection model of expression ith cluster classification obtains is between any two
Pearson correlation coefficient matrix,Indicate the 100th in the forecasting wind speed feature pre-selection model of ith cluster classification
The Pearson correlation coefficient between screening vector that model and the 2nd model obtain；
The accumulated value for calculating separately the Pearson correlation coefficient between each screening vector and other screening vectors, that is, distinguish
To K Pearson correlation coefficient matrix P1,P2,…,PKIn each row sum, for example, i-th of Pearson correlation coefficient
Matrix PiThe accumulated value of Pearson correlation coefficient of i-th of forecasting wind speed feature pre-selection model be
It is phased out the maximum corresponding forecasting wind speed feature pre-selection model of Pearson correlation coefficient accumulated value, is gathered until every kind
The forecasting wind speed feature pre-selection model of class classification is 10 remaining, indicates are as follows:
Wherein, M_Elman indicates the 10*K after the screening pre- modeling of forecasting wind speed feature based on Elman neural network
Type, every a line indicate 10 forecasting wind speed features pre-selection model of same cluster type, total K row；
Step 7: 10 forecasting wind speed features pre-selection model group of every kind of cluster classification is obtained at wind speed predicted characteristics model group
To K forecasting wind speed characteristic model group；
The forecasting wind speed characteristic model group includes that 10 forecasting wind speed features preselect model, each forecasting wind speed character modules
Type group output data groups are named as wind speed feature vector at the vector that length is 10, and i-th of forecasting wind speed characteristic model group indicates
For [m_elmani,1,m_elmani,1,…,m_elmani,10]；
Step 8: being directed to K forecasting wind speed characteristic model group, establish the K forecasting wind speed normalizings based on wavelet neural network
Model；
The forecasting wind speed normalizing model output value is set as wind speed normalizing value；
It is described to be directed to K forecasting wind speed characteristic model group, establish the K forecasting wind speed normalizing moulds based on wavelet neural network
Detailed process is as follows for type:
Step 6.1: successively using the air speed value at 3 moment before every group of wind velocity vector in Wind speed model sample A as input
Data are input in the forecasting wind speed characteristic model group of corresponding cluster classification, and output obtains wind speed feature vector；
Step 6.2: the wind speed feature vector for successively obtaining step 6.1 is right in Wind speed model sample A as input data
The air speed value at the 4th moment answered is obtained as output data, training wavelet neural network for K forecasting wind speed character modules
The K forecasting wind speed normalizing models based on wavelet neural network of type group；
The forecasting wind speed normalizing model based on wavelet neural network is respectively provided with input layer number in training
10, hidden layer Wavelet Element number is 20, and output layer node number is 1；Hidden layer wavelet neural member uses Mexican Hat small echo letter
Number, output node layer use Sigmoid function, and the maximum number of iterations in training process is set as 500, and training learning rate is
0.04, threshold value 0.002；
Step 9: real-time prediction of wind speed；
Acquisition current time and recent air speed data, data sampling time interval should keep using with step 1 in real time
Original wind speed time series data collection wind-speed sample time interval Δ t it is consistent, carry out the detailed process of real-time prediction of wind speed
It is as follows:
Step B1: setting forecasting wind speed object time T is obtained and object time T immediate at least 200 by air measuring station
The original air speed value at a moment forms original wind speed time series, enters step B2；
Step B2: Kalman filtering is carried out to original wind speed time series, obtains filtered wind speed time series, is entered
Step B3；
Step B3: it is obtained in calculation of wind speed time series with the air speed value at object time T immediate 4 moment and step 3
The Euclidean distance of K cluster centre time series out, to select the forecasting wind speed of corresponding cluster classification special apart from minimum principle
Levy model group, using in wind speed time series with the air speed value at object time T immediate 3 moment be used as input data, input
Into forecasting wind speed characteristic model group, output obtains the wind speed feature vector that length is 10, enters step B4；
Step B4: using wind speed feature vector as input data, it is input to the forecasting wind speed normalizing mould of corresponding cluster classification
Type, output wind speed normalizing value, enters step B5；
Step B5: whether the moment where judgment step B4 output wind speed normalizing value is greater than object time T, if so, entering step
Otherwise rapid B7 enters step B6；
Step B6: the wind speed normalizing value that step B4 is exported is inserted into wind speed time series first place, updates wind speed time sequence
Column, return step B3；
Step B7: in calculation of wind speed time series with the air speed value at object time T immediate 100 moment composition to
Wind speed time series data collection w ' (t) obtained in amount and step 2=[w ' (t=Δ t), w ' (t=2* Δ t) ..., w ' (t=
N* Δ t)] in arbitrary continuation 100 moment air speed value composition vector between Pearson correlation coefficient, select Pierre
The vector that the inferior highest wind speed time series data of related coefficient concentrates the air speed value at one group of 100 moment to form, will be in step 2
Time sequence of noise data set no (t)=[no (t=Δ t), no (t=2* Δ t) ..., no (t=n* Δ t)] in it is corresponding when
The filter noise at quarter was attached in wind speed time series and object time T immediate 100 moment, i.e. reduction prediction of wind speed is true
Reality completes the forecasting wind speed of object time.
The above is only the preferred embodiment of the technology of the present invention, it is noted that for the common skill of the art
For art personnel, without departing from the technical principles of the invention, several improvement and replacement can also be made, these improve and
Replacement also should be regarded as protection scope of the present invention.
Claims (7)
1. a kind of line of high-speed railway wind speed feature extraction intelligent Forecasting, which comprises the following steps:
Step 1: original wind speed time series data collection is obtained by air measuring station；
Air measuring station is set in line of high-speed railway designated position, original wind speed time series data collection is obtained by air measuring station；
The original wind speed time series data collection is made of n original air speed datas altogether；Wherein Δ t is between the wind-speed sample time
Every；N is sampling number, and n is at least more than 500；
Step 2: establishing Wind speed model sample A；
The Wind speed model sample A includes Wind speed model training sample AaWith Wind speed model Screening Samples Ab；
Establishing Wind speed model sample A, detailed process is as follows:
Step A1: Kalman filtering is carried out to original wind speed time series data collection, obtains filtered wind speed time series number
According to collection and Time sequence of noise data set；
Step A2: it concentrates the wind speed at 4 moment of arbitrary continuation as one group of wind velocity vector wind speed time series data, is wrapped
Wind speed model sample A containing n-3 group wind velocity vector；
Step A3: pressing sample time order, from Wind speed model sample A, chooses preceding 3/4 wind velocity vector composition Wind speed model instruction
Practice sample Aa, remaining wind velocity vector composition Wind speed model Screening Samples Ab；
Step 3: using K-means clustering method, each group of wind velocity vector in Wind speed model sample A is clustered, choose
Optimum clustering number K obtains K cluster centre time series；
Step 4: according to the Wind speed model sample A for clustering classification for K has been gathered, by Wind speed model training sample AaIt is a right to be divided into K
The Wind speed model training subsample { A of classification should be clusteredA, 1, AA, 2..., AA, K, by Wind speed model Screening Samples AbIt is a right to be divided into K
Wind speed model is answered to screen subsample { AB, 1, AB, 2..., AB, K}；
Step 5: for K Wind speed model training subsample, it is pre- to establish forecasting wind speed feature of the K kind based on Elman neural network
Each 100, modeling type；
The forecasting wind speed feature pre-selection model output value based on Elman neural network is named as wind speed feature preset value；
Step 6: screening model, the pre-selection of forecasting wind speed feature each 100, model by K kind based on Elman neural network sieve respectively
Choosing is to each 10；
Step 7: 10 forecasting wind speed features pre-selection model group of every kind of cluster classification obtains K at wind speed predicted characteristics model group
A forecasting wind speed characteristic model group；
The forecasting wind speed characteristic model group includes that 10 forecasting wind speed features preselect model, each forecasting wind speed characteristic model group
Output data groups are named as wind speed feature vector at the vector that length is 10；
Step 8: being directed to K forecasting wind speed characteristic model group, establish the K forecasting wind speed normalizing moulds based on wavelet neural network
Type；
The forecasting wind speed normalizing model output value is set as wind speed normalizing value；
Step 9: real-time prediction of wind speed；
Acquisition current time and recent air speed data, data sampling time interval should be kept and original used in step 1 in real time
The wind-speed sample time interval Δ t of beginning wind speed time series data collection is consistent, and carrying out real-time prediction of wind speed, detailed process is as follows:
Step B1: setting forecasting wind speed object time T, when obtaining at least 200 immediate with object time T by air measuring station
The original air speed value carved forms original wind speed time series, enters step B2；
Step B2: Kalman filtering is carried out to original wind speed time series, filtered wind speed time series is obtained, enters step
B3；
Step B3: the K obtained in calculation of wind speed time series with the air speed value at object time T immediate 4 moment and step 3
The Euclidean distance of a cluster centre time series, to select the forecasting wind speed characteristic model of corresponding cluster classification apart from minimum principle
Group is input to wind speed using the air speed value in wind speed time series with object time T immediate 3 moment as input data
In predicted characteristics model group, output obtains the wind speed feature vector that length is 10, enters step B4；
Step B4: using wind speed feature vector as input data, being input to the forecasting wind speed normalizing model of corresponding cluster classification, defeated
Outlet air is returned quickly a value, and B5 is entered step；
Step B5: whether the moment where judgment step B4 output wind speed normalizing value is greater than object time T, if so, B7 is entered step,
Otherwise, B6 is entered step；
Step B6: the wind speed normalizing value that step B4 is exported is inserted into wind speed time series first place, wind speed time series is updated, returns
Return step B3；
Step B7: in calculation of wind speed time series with the air speed value at object time T immediate 100 moment composition vector with
Skin between the vector of the air speed value composition at 100 moment of wind speed time series data collection arbitrary continuation obtained in step 2
The inferior related coefficient of that selects the highest wind speed time series data of Pearson correlation coefficient to concentrate the wind speed at one group of 100 moment
It is worth the vector of composition, the filter noise that the Time sequence of noise data set in step 2 corresponds to the moment is attached to wind speed time sequence
Column restore prediction of wind speed authenticity, complete the forecasting wind speed of object time with object time T immediate 100 moment.
2. the method according to claim 1, wherein described use K-means clustering method, to Wind speed model sample
Each group of wind velocity vector in this A is clustered, and is chosen optimum clustering number K, is obtained the specific mistake of K cluster centre time series
Journey is as follows:
Step 2.1: setting K-means cluster numbers K ', K '=1；
Step 2.2: using Euclidean distance as similarity measurements flow function, to each group of wind velocity vector in Wind speed model sample A
It is clustered, obtains a cluster classification of K ' and a cluster centre time series of K '；
The calculation formula of the Euclidean distance isWherein, X=(x1, x2,
x3, x4), Y=(y1, y2, y3, y4) be respectively two different wind velocity vectors, d (X, Y) is wind velocity vector X, between Y it is European away from
From；
Step 2.3: calculating the error sum of squares SSE (K ') of K-means cluster；
Wherein, K ' is cluster numbers, CiIt is ith cluster classification, p is CiOne group of wind velocity vector in class, miIt is CiIn the cluster of class
The heart；
Step 2.4: judging whether K ' value is equal to 20, if so, entering step 2.5, otherwise, enable K '=K '+1, then return step
2.2；
Step 2.5: each α (K ') as K '=2 to 19 and β (K ') value are calculated separately, subsequently into step 2.6；
α (K ')=arctan (SSE (K ')-SSE (K '+1)), α ∈ (0 °, 90 °), K ' ∈ [2,19]
β (K ')=arctan (SSE (K ')-SSE (K ' -1)), β ∈ (90 °, 180 °), K ' ∈ [2,19]；
Step 2.6: calculating separately each α (K ')+β (K ') value as K '=2 to 19, choose so that α (K ')+β (K ') is maximum
When K ' value as optimum clustering number K, obtain K cluster centre time series.
3. establishing K the method according to claim 1, wherein described train subsample for K Wind speed model
Forecasting wind speed feature of the kind based on Elman neural network preselects each 100 processes of model are as follows: successively instructs K Wind speed model
The air speed value at 3 moment is as input data before every group of wind velocity vector in white silk subsample, and the air speed value at the 4th moment is as defeated
Data out, training Elman neural network, the Elman neural network point obtained for each Wind speed model training subsample training
It is not randomly provided the Elman neural network parameter of 100 kinds of different values, obtains forecasting wind speed of the K kind based on Elman neural network
Feature preselects each 100, model, and the total 100*K forecasting wind speed features based on Elman neural network preselect model.
4. according to the method described in claim 3, it is characterized in that, the 100*K wind speed based on Elman neural network are pre-
Surveying feature pre-selection model and being respectively provided with input layer number in training is 3, and it is [5,10] that range, which is arranged, in hidden layer node number,
Output layer node number is 1, and accepting node layer number is 7, implies layer functions and uses Sigmoid function, the training of network uses
BP algorithm, maximum number of iterations are set as 800, and training learning rate setting range is [0.001,0.5], threshold value 0.005.
5. the method according to claim 1, wherein the screening model, is based on Elman nerve for K kind respectively
Detailed process is as follows to each 10 for each 100 screenings of forecasting wind speed feature pre-selection model of network:
Step 5.1: K Wind speed model successively being screened into subsample { AB, 1,AB, 2..., AB, KIn every group of wind velocity vector before 3 when
The air speed value at quarter is separately input in 100 forecasting wind speed features pre-selection model of corresponding cluster classification as input data；
Step 5.2: preliminary screening forecasting wind speed feature preselects model；
Successively calculate the pre-selection model output of each forecasting wind speed feature wind speed feature preset value chronologically form it is one-dimensional
Vector and Wind speed model Screening Samples AbThe mean square error of the one-dimensional vector of the air speed value composition at middle corresponding moment, eliminates mean square error
The forecasting wind speed feature that difference is greater than mean square error mean value preselects model；
Step 5.3: establishing screening vector；
It is utilized respectively the air speed value in K cluster centre time series with immediate 3 moment at current time and establishes K length
For 3 screening vector；
Step 5.4: updating screening vector；
Respectively using the air speed data at 3 moment before K screening vector as input data, it is input to the wind speed of corresponding cluster classification
Predicted characteristics preselect in model, and vector first position is screened in the wind speed feature preset value insertion at the 4th moment of output, remaining
Data position sequence is successively delayed, and repeats this step until K screening vector length reaches 200；
Step 5.5: screening forecasting wind speed feature again and preselect model；
Calculate separately the Pearson came of the obtained screening vector of forecasting wind speed feature pre-selection model of every kind of cluster classification between any two
Related coefficient obtains K Pearson correlation coefficient matrix, calculates separately the skin between each screening vector and other screening vectors
The accumulated value of your inferior related coefficient is phased out the maximum corresponding pre- modeling of forecasting wind speed feature of Pearson correlation coefficient accumulated value
Type, the forecasting wind speed feature pre-selection model that classification is clustered until every kind are 10 remaining.
6. establishing K the method according to claim 1, wherein described be directed to K forecasting wind speed characteristic model group
Detailed process is as follows for a forecasting wind speed normalizing model based on wavelet neural network:
Step 6.1: successively using the air speed value at 3 moment before every group of wind velocity vector in Wind speed model sample A as input data,
It is input in the forecasting wind speed characteristic model group of corresponding cluster classification, output obtains wind speed feature vector；
Step 6.2: the wind speed feature vector for successively obtaining step 6.1 is corresponding in Wind speed model sample A as input data
The air speed value at the 4th moment is obtained as output data, training wavelet neural network for K forecasting wind speed characteristic model group
K forecasting wind speed normalizing models based on wavelet neural network.
7. according to the method described in claim 6, it is characterized in that, the forecasting wind speed normalizing mould based on wavelet neural network
It is 10 that type, which is respectively provided with input layer number in training, and hidden layer Wavelet Element number is 20, and output layer node number is 1；It is hidden
Layer wavelet neural member uses Mexican Hat wavelet function, and output node layer uses Sigmoid function, in training process most
Big the number of iterations is set as 500, and training learning rate is 0.04, threshold value 0.002.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811643482.7A CN109615027B (en) | 2018-12-29 | 2018-12-29 | Intelligent prediction method for extracting wind speed characteristics along high-speed railway |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811643482.7A CN109615027B (en) | 2018-12-29 | 2018-12-29 | Intelligent prediction method for extracting wind speed characteristics along high-speed railway |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109615027A true CN109615027A (en) | 2019-04-12 |
| CN109615027B CN109615027B (en) | 2020-12-08 |
Family
ID=66016400
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811643482.7A Active CN109615027B (en) | 2018-12-29 | 2018-12-29 | Intelligent prediction method for extracting wind speed characteristics along high-speed railway |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109615027B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110650058A (en) * | 2019-10-08 | 2020-01-03 | 河南省云安大数据安全防护产业技术研究院有限公司 | Network traffic analysis method, device, storage medium and equipment |
| CN111291943A (en) * | 2020-03-09 | 2020-06-16 | 西南交通大学 | High-speed rail strong wind early warning method based on frequent sequence |
| CN112381420A (en) * | 2020-11-17 | 2021-02-19 | 平安普惠企业管理有限公司 | Risk early warning method, device and equipment based on desert wind speed and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070223841A1 (en) * | 2006-03-23 | 2007-09-27 | Weinzapfel Robert A | Methods of characterizing the probability of wind impact |
| CN102063641A (en) * | 2010-10-14 | 2011-05-18 | 北京大学 | Method for forecasting wind speed of high speed railway line |
| CN107895202A (en) * | 2017-10-24 | 2018-04-10 | 天津大学 | Short-term wind speed forecasting method based on various visual angles wind speed mode excavation |
| CN109063907A (en) * | 2018-07-27 | 2018-12-21 | 中南大学 | A kind of big step-ahead prediction method of line of high-speed railway extreme wind speed intelligence traversal |
-
2018
- 2018-12-29 CN CN201811643482.7A patent/CN109615027B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070223841A1 (en) * | 2006-03-23 | 2007-09-27 | Weinzapfel Robert A | Methods of characterizing the probability of wind impact |
| CN102063641A (en) * | 2010-10-14 | 2011-05-18 | 北京大学 | Method for forecasting wind speed of high speed railway line |
| CN107895202A (en) * | 2017-10-24 | 2018-04-10 | 天津大学 | Short-term wind speed forecasting method based on various visual angles wind speed mode excavation |
| CN109063907A (en) * | 2018-07-27 | 2018-12-21 | 中南大学 | A kind of big step-ahead prediction method of line of high-speed railway extreme wind speed intelligence traversal |
Non-Patent Citations (3)
| Title |
|---|
| HUI LIU 等: "Wind speed forecasting method based on deep learning strategy using empirical wavelet transform, long short term memory neural network and Elman neural network", 《ENERGY CONVERSION AND MANAGEMENT》 * |
| 刘辉 等: "基于小波分析法与滚动式时间序列法的风电场风速短期预测优化算法", 《中南大学学报(自然科学版)》 * |
| 潘迪夫 等: "基于时间序列分析和卡尔曼滤波算法的风电场风速预测优化模型", 《电网技术》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110650058A (en) * | 2019-10-08 | 2020-01-03 | 河南省云安大数据安全防护产业技术研究院有限公司 | Network traffic analysis method, device, storage medium and equipment |
| CN110650058B (en) * | 2019-10-08 | 2022-03-04 | 河南省云安大数据安全防护产业技术研究院有限公司 | Network traffic analysis method, device, storage medium and equipment |
| CN111291943A (en) * | 2020-03-09 | 2020-06-16 | 西南交通大学 | High-speed rail strong wind early warning method based on frequent sequence |
| CN111291943B (en) * | 2020-03-09 | 2022-03-08 | 西南交通大学 | High-speed rail strong wind early warning method based on frequent sequence |
| CN112381420A (en) * | 2020-11-17 | 2021-02-19 | 平安普惠企业管理有限公司 | Risk early warning method, device and equipment based on desert wind speed and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109615027B (en) | 2020-12-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105975931B (en) | A kind of convolutional neural networks face identification method based on multiple dimensioned pond | |
| CN106339775B (en) | The air heavily contaminated case method of discrimination clustered based on weather typing and meteorological element | |
| CN109615027A (en) | A kind of line of high-speed railway wind speed feature extraction intelligent Forecasting | |
| CN106446942A (en) | Crop disease identification method based on incremental learning | |
| CN110298266A (en) | Deep neural network object detection method based on multiple dimensioned receptive field Fusion Features | |
| CN107392232A (en) | A kind of flotation producing condition classification method and system | |
| CN107133960A (en) | Image crack dividing method based on depth convolutional neural networks | |
| CN105374033B (en) | SAR image segmentation method based on ridge ripple deconvolution network and sparse classification | |
| CN105913087A (en) | Object identification method based on optimal pooled convolutional neural network | |
| CN108921201B (en) | Dam defect identification and classification method based on feature combination with CNN | |
| CN110348376A (en) | A kind of pedestrian's real-time detection method neural network based | |
| CN107578423A (en) | The correlation filtering robust tracking method of multiple features hierarchical fusion | |
| CN105938564A (en) | Rice disease recognition method based on principal component analysis and neural network and rice disease recognition system thereof | |
| CN106446895A (en) | License plate recognition method based on deep convolutional neural network | |
| CN106650913A (en) | Deep convolution neural network-based traffic flow density estimation method | |
| CN108510521A (en) | A kind of dimension self-adaption method for tracking target of multiple features fusion | |
| CN107871136A (en) | The image-recognizing method of convolutional neural networks based on openness random pool | |
| CN107480649A (en) | A kind of fingerprint pore extracting method based on full convolutional neural networks | |
| CN107341463A (en) | A kind of face characteristic recognition methods of combination image quality analysis and metric learning | |
| CN101763502A (en) | High-efficiency method and system for sensitive image detection | |
| CN103593652B (en) | A kind of cucumber disease recognition methods processed based on cucumber leaves symptomatic picture | |
| CN109218134B (en) | Test case generation system based on neural style migration | |
| CN109635811A (en) | The image analysis method of spatial plant | |
| CN110417694A (en) | A kind of modulation mode of communication signal recognition methods | |
| CN106997475A (en) | A kind of insect image-recognizing method based on parallel-convolution neutral net |
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 |