CN109063939A - A kind of wind speed forecasting method and system based on neighborhood door shot and long term memory network - Google Patents
A kind of wind speed forecasting method and system based on neighborhood door shot and long term memory network Download PDFInfo
- Publication number
- CN109063939A CN109063939A CN201811296424.1A CN201811296424A CN109063939A CN 109063939 A CN109063939 A CN 109063939A CN 201811296424 A CN201811296424 A CN 201811296424A CN 109063939 A CN109063939 A CN 109063939A
- Authority
- CN
- China
- Prior art keywords
- wind speed
- neighborhood
- causality
- term memory
- door
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01W—METEOROLOGY
- G01W1/00—Meteorology
- G01W1/10—Devices for predicting weather conditions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Environmental & Geological Engineering (AREA)
- Life Sciences & Earth Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Development Economics (AREA)
- Biophysics (AREA)
- Game Theory and Decision Science (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Atmospheric Sciences (AREA)
- Biodiversity & Conservation Biology (AREA)
- Ecology (AREA)
- Environmental Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to forecasting wind speed technical fields, a kind of wind speed forecasting method and system based on neighborhood door shot and long term memory network is disclosed, Pearson correlation coefficient and maximum information coefficient is respectively adopted come the linear and nonlinear correlation probed between variable to screen wind speed correlation factor；The causality of wind speed and the wind speed factor in statistical significance is probed into using Granger causality test on the basis of correlation analysis；Causal structure is fallen into 5 types, and all types of causalities are unified for by a kind of equivalent tree causality structure by the method for " decomposition-dummy variables-beta pruning "；For equivalent tree causality structure, propose the shot and long term memory network model based on neighborhood door come prediction of wind speed.Forecasting procedure (NLSTM) of the invention accurately considers the causality between wind speed and the wind speed factor, effectively increases the precision of prediction of wind speed, the scheduling of application and power grid to wind-powered electricity generation is of crucial importance.
Description
Technical field
The invention belongs to forecasting wind speed technical field more particularly to a kind of wind speed based on neighborhood door shot and long term memory network
Prediction technique and system.
Background technique
Currently, the prior art commonly used in the trade is such that
Wind energy is a kind of promising renewable and clean energy resource, in recent years by from extensive concern all over the world.More
Carry out more wind power integration electric system, so that electric system becomes unreliable, this is by the strong fluctuation of wind speed and strong random
Caused by property.Therefore, accurate prediction of wind speed is of crucial importance the efficient scheduling of utilization and the electric system of wind energy.
Wind speed is influenced by many meteorological factors, including air pressure, temperature, the factors such as humidity.Complicated relationship makes between the factor
Forecasting wind speed becomes difficult, and the attainable precision of conventional machines learning method prediction of wind speed institute is limited.
Deep learning method shot and long term memory network (LSTM) is when solving the problems, such as the time series forecasting as wind speed
Precision of prediction with higher, but LSTM is generally regarded as black-box model and comes using this makes the interpretation of model die down.
Analyze the correlativity of the air speed influence factor by Feature Engineering and clear the causality between them to improve wind speed it is pre-
The interpretation for surveying precision and enhancing model is all highly beneficial.Therefore, how to analyze between wind speed and its correlation factor because
Fruit relationship simultaneously accurately accounts for this causality into LSTM, to improve forecasting wind speed precision and enhance the interpretation of model
It is the theory and Practical Project problem of urgent need to resolve.
In Feature Engineering, common correlation analysis has diagram method, correlation coefficient process, covariance method, maximum letter
Cease Y-factor method Y etc..Common Causality Analysis Approach has theoretical analysis, transfer entropy method and Granger causality test method.
In the present invention, Pearson correlation coefficient is used to probe into the linear dependence between the factor, maximum information coefficient
It is used to probe into the non-linear dependencies between the factor.Granger causality test is used to probe into the pass of the cause and effect between the factor
System.
In conclusion problem of the existing technology is:
The complicated multiplicity of causality structure type between the factor, current few scholars divide causality structure
Class, thus how science and completely to all causality structure types carry out classification and prior art problems faced.
How causality structure after classification conveniently and effectively uses currently without the case that can refer to, therefore will
It is also prior art problems faced that sorted causality structure, which is unified for the general causality structure of one kind,.
In the prior art, complicated relationship becomes difficult forecasting wind speed between the factor, conventional machines study side
The attainable precision of method prediction of wind speed institute is limited.
In the prior art, due to only one feature input interface of LSTM, LSTM can only be by all factor indifferences
Ground inputs and can not accurately consider the causality obtained by Feature Engineering, cannot be by the causality structure essence of the wind speed factor
Really take into account LSTM.
Solve the difficulty and meaning of above-mentioned technical problem:
About the classification of causality structure, it can include all causality structure types that difficult point, which is how to classify,.
Therefore the integrality of classification is the basis of subsequent technology.
About the unification of causality structure, difficult point be how to find sorted all types of causality structures it
Between common ground to obtain a kind of general structure.This universal architecture can not only represent all types of causality knots
Structure also needs to be the structure for being easy to predict.Therefore the unified representativeness of causality structure and operability play in the present invention
The effect formed a connecting link.
After solving above-mentioned technical problem, bring meaning are as follows:
For the wind speed causality structure for making LSTM have the ability accurately to consider to obtain, the invention proposes be based on neighborhood door
Shot and long term memory network (NLSTM).NLSTM is from network structure and LSTM produces difference, the correct forward direction for deriving NLSTM
It is the difficult point for realizing NLSTM with backpropagation formula.It is correct to realize that NLSTM is the guarantee for improving forecasting wind speed precision.
Since the causality structure of different zones wind speed may be different, technology of the invention is promoted
It is also one of difficult point.Therefore the NLSTM designed can carry out corresponding variation to NLSTM according to different causality structures
Popularization it is highly beneficial.
Summary of the invention
In view of the problems of the existing technology, the present invention provides a kind of wind speed based on neighborhood door shot and long term memory network
Prediction technique and system can accurately consider the causality structure between wind speed and the air speed influence factor, and can be had
High-precision forecasting wind speed result.
The invention is realized in this way a kind of wind speed forecasting method based on neighborhood door shot and long term memory network, comprising: institute
It states the wind speed forecasting method based on neighborhood door shot and long term memory network and Pearson correlation coefficient and maximum information coefficient is respectively adopted
Linear and nonlinear correlation between situational variables screens wind speed correlation factor；
On the basis of correlation analysis using Granger causality analysis wind speed and the wind speed factor in statistical significance
Causality；Causal structure is fallen into 5 types, and by decomposition-dummy variables-beta pruning method by it is all types of because
Fruit relationship is unified for a kind of equivalent tree causality structure；
Parity price tree causality structure carries out prediction of wind speed by the shot and long term memory network model based on neighborhood door.
It specifically includes:
(1) collect wind speed Y and may be wind speed impact factor data
(2) Pearson correlation coefficient (MIC) and maximum information coefficient (MIC) analysis wind speed and may be wind speed are utilized respectively
Linear and nonlinear correlation between impact factor, to obtain wind speed correlation factor [x1,x2,,…,xn], the Pierre with wind speed
Inferior related coefficient absolute value or maximum information coefficient can be used as wind speed correlation factor in 0.5 or more impact factor.
(3) wind speed Y and wind speed correlation factor [x are probed into using Granger causality test1,x2,,…,xn] anticipate in statistics
Causality in justice.
(4) according to shape causal between wind speed and wind speed correlation factor, causality structure is divided into center pivot
Knob, chain type is cyclic annular, tree-shaped and network-like totally five kinds of structures.It carefully analyzes it can be found that central hub structure and chain structure point
It is not tree in extending transversely and Longitudinal Extension special case, and cyclic structure can be decomposed into a series of chain type knot
Structure, so first three causality structure can be converted to tree-shaped causality structure.
(5) network-like structure is the general type of causality structure.Inverse causality arrow direction will since wind speed
Network-like structure decomposites a plurality of chain structure, and (and the practical factor has the factor dummy variables for being present in a plurality of decomposition line
The same attribute still numbers different variables, appears in the same Graph One factor in different decomposition line for differentiation and fictionalize the change come
Amount) it replaces and distinguishes, each item decomposition line is combined into tree (tree reconfigured is often very huge), according to computing resource
Size carries out beta pruning to tree, obtains final tree construction of equal value, therefore all types of causality structures are ok
Be converted to equivalent tree causality structure.
(6) according to equivalent tree causality structure, the training set D being made of the wind speed factor and wind speed is constructedTa=[xTa,
YTa] and the test set D being made of predictorTe=[xTe], and data are normalized.
(7) the shot and long term memory network (NLSTM) based on neighborhood door is constructed according to equivalent tree causality, NLSTM is set
Parameter, including input layer number ni, hidden layer number of nodes nh, output layer number of nodes no, fixed learning rate η, crowd size T, instruction
Practice wheel number Ep.
(8) using the Adam optimization algorithm in conjunction with mini-batch mechanism in training set DTaUpper trained NLSTM.
(9) by test set DTeIt is input in trained NLSTM and is predicted, obtain forecasting wind speed result y.
Further, in step (8), the step of t-th of period information propagated forward and calculation formula are as follows:
A. the propagated forward of each node elder generation complete independently standard LSTM
fit=σ (netf,i,t)=σ (wfh,i·hi,t-1+wfx,i·xit+bf,i) (38)
iit=σ (neti,i,t)=σ (wih,i·hi,t-1+wix,i·xit+bi,i) (39)
ait=tanh (neta,i,t)=tanh (wah,i·hi,t-1+wax,i·xit+ba,i) (40)
Cit=fit*Ci,t-1+iit*ait (41)
oit=σ (neto,i,t)=σ (woh,i·hi,t-1+wox,i·xit+bo,i) (42)
hit=oit*tanh(Cit) (43)
B. each node is along tree from leaf node propagated forward to root node
n1it=σ (netn1,i,t)=σ (wn1h,i·hi,t-1+wn1x,i·xit+bn1,i) (46)
n2it=σ (netn2,i,t)=σ (wn2h,i·hi,t-1+wn2x,i·xit+bn2,i) (47)
Nit=n1it*Rit+n2it*hit (48)
yt=σ (zt)=σ (wy·Nmt+by) (49)
Wherein fit,iit,ait,Cit,oit,hit,Rit,NitAnd yitIt is forgetting door of i-th of node in period t respectively, input
Door, information state, cell state, out gate, hidden layer output, central hub, neighborhood and predicted value；n1itAnd n2itAll being is
Neighborhood door of the i node in period t；PijIt is j-th of child node of i-th of node；Tanh and σ is tanh and sigmoid respectively
Activation primitive；Symbol and * respectively represent multiplication between matrix multiplication and matrix element；Remaining variable is all intermediate variable.
Further, in step (8), the step of t-th of period error back propagation and calculation formula are as follows:
A. the target that the most common squared error function optimizes as needs is defined
B. the error of output layer is calculated
C. against tree direction from root node to leaf node reverse propagated error
D. Adam optimization algorithm is used, with [δ wLh,i,δwLx,i,δbL,i] and [δ wy,δby] update [wLh,wLx,bL] and
[wy,by]；For versatility, weight is indicated with symbol W, the gradient of weight is indicated with δ W, and Adam updates the general formula of weight
Are as follows:
mti=β1·mti-1+(1-β1)·δWti (70)
vti=β2·vti-1+(1-β2)·(δWti)2 (71)
Wherein EtFor error function, ytAnd YtRespectively predicted value and observation.β1,β2It is Adam parameter with ε, defaults respectively
Take 0.9,0.999 and 10-8；Ti is that the current update times of weight W and period t are distinguished；
Predicted value is calculated according to above formula elder generation propagated forward, then backpropagation updates weight, it is referred to as primary to update；One
Iteration Ep wheel altogether, it is every to take turns training set DTaCriticizing for T size is taken to be trained, every batch of completes primary update.
Another object of the present invention is to provide the wind speed based on neighborhood door shot and long term memory network described in a kind of realize
The computer program of prediction technique.
Another object of the present invention is to provide the wind speed based on neighborhood door shot and long term memory network described in a kind of realize is pre-
The information data processing terminal of survey method.
Another object of the present invention is to provide a kind of computer readable storage mediums, including instruction, when it is in computer
When upper operation, so that computer executes the wind speed forecasting method based on neighborhood door shot and long term memory network.
Another object of the present invention is to provide the wind speed based on neighborhood door shot and long term memory network described in a kind of realize is pre-
The forecasting wind speed control system based on neighborhood door shot and long term memory network of survey method.
Another object of the present invention is to provide the wind speed based on neighborhood door shot and long term memory network described in a kind of carrying is pre-
Survey power equipment of the prediction of wind speed to wind energy utilization of control system.
In conclusion advantages of the present invention and good effect are as follows:
The present invention provides a kind of wind speed forecasting methods based on neighborhood door shot and long term memory network, pass through Feature Engineering point
The causality between wind speed and the wind speed factor is precipitated, using the method for " decomposition-dummy variables-beta pruning " by this causality knot
Structure is converted to equivalent tree causality structure.NLSTM significantly enhances the interpretation of model, it can accurately consider this equivalence
Set causality structure and the prior art can not accurately consider causality structure.
The universality of model NLSTM proposed by the present invention is good, easy to spread, can be according to different wind speed causality knots
Structure converts out corresponding network structure.
The forecasting wind speed result precision that model proposed by the present invention obtains is high, this point forecasting wind speed can refer to from subordinate list 4
It can analyze and obtain in target comparison, can also intuitively find out from attached drawing 5.
Detailed description of the invention
Fig. 1 is the wind speed forecasting method flow chart of neighborhood door shot and long term memory network provided in an embodiment of the present invention；
Fig. 2 is equivalent tree causality structural schematic diagram provided in an embodiment of the present invention；
Fig. 3 is provided in an embodiment of the present invention for NLSTM network structure；
Fig. 4 is case wind speed causality structure chart in fuyun station provided in an embodiment of the present invention and its tree construction of equal value
Figure；
Fig. 5 case forecasting wind speed comparative result figure in fuyun station provided in an embodiment of the present invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to embodiments, to the present invention
It is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to
Limit the present invention.
All types of causality structures are converted to system by the method that the present invention first uses " decomposition-dummy variables-beta pruning "
Then one tree construction of equal value establishes the corresponding shot and long term memory network based on neighborhood door for tree construction of equal value
(NLSTM) carry out prediction of wind speed.The structure of NLSTM is as tree construction of equal value, therefore it can accurately consider the cause and effect between the factor
Relational structure.
Fig. 1 show the overall flow figure of the wind speed forecasting method of neighborhood door shot and long term memory network of the present invention, specific to wrap
Include following steps:
(1) collect wind speed Y and may be wind speed impact factor data
(2) Pearson correlation coefficient (MIC) and maximum information coefficient (MIC) analysis wind speed and may be wind speed are utilized respectively
Linear and nonlinear correlation between impact factor, to obtain wind speed correlation factor [x1,x2,,…,xn], the Pierre with wind speed
Inferior related coefficient absolute value or maximum information coefficient can be used as wind speed correlation factor in 0.5 or more impact factor.
(3) wind speed Y and wind speed correlation factor [x are probed into using Granger causality test1,x2,,…,xn] anticipate in statistics
Causality in justice.
(4) according to shape causal between wind speed and wind speed correlation factor, causality structure is divided into center pivot
Knob, chain type is cyclic annular, tree-shaped and network-like totally five kinds of structures.It carefully analyzes it can be found that central hub structure and chain structure point
It is not tree in extending transversely and Longitudinal Extension special case, and cyclic structure can be decomposed into a series of chain type knot
Structure, so first three causality structure can be converted to tree-shaped causality structure.
(5) network-like structure is the general type of causality structure.Inverse causality arrow direction will since wind speed
Network-like structure decomposites a plurality of chain structure, and (and the practical factor has the factor dummy variables for being present in a plurality of decomposition line
The same attribute still numbers different variables, appears in the same Graph One factor in different decomposition line for differentiation and fictionalize the change come
Amount) it replaces and distinguishes, each item decomposition line is combined into tree (tree reconfigured is often very huge), according to computing resource
Size carries out beta pruning to tree, obtains final tree construction of equal value, therefore all types of causality structures are ok
Be converted to equivalent tree causality structure.
(6) according to equivalent tree causality structure, the training set D being made of the wind speed factor and wind speed is constructedTa=[xTa,
YTa] and the test set D being made of predictorTe=[xTe], and data are normalized.
(7) the shot and long term memory network (NLSTM) based on neighborhood door is constructed according to equivalent tree causality, NLSTM is set
Parameter, including input layer number ni, hidden layer number of nodes nh, output layer number of nodes no, fixed learning rate η, crowd size T, instruction
Practice wheel number Ep.According to the weight of each node of parameter initialization, including [wfh,i,wfx,i,bf,i],[wih,i,wix,i,bi,i],
[wah,i,wax,i,ba,i],[woh,i,wox,i,bo,i],[wn1h,i,wn1x,i,bn1,i],[wn2h,i,wn2x,i,bn2,i],[wrh,i,j,
wrx,i,j,br,i] and [wy,by]。
(8) using the Adam optimization algorithm in conjunction with mini-batch mechanism in training set DTaUpper trained NLSTM.NLSTM's
It realizes and is related to the backpropagation of the propagated forward and error of information.
The step of t-th of period information propagated forward and calculation formula are as follows:
A. the propagated forward of each node elder generation complete independently standard LSTM
fit=σ (netf,i,t)=σ (wfh,i·hi,t-1+wfx,i·xit+bf,i) (75)
iit=σ (neti,i,t)=σ (wih,i·hi,t-1+wix,i·xit+bi,i) (76)
ait=tanh (neta,i,t)=tanh (wah,i·hi,t-1+wax,i·xit+ba,i) (77)
Cit=fit*Ci,t-1+iit*ait (78)
oit=σ (neto,i,t)=σ (woh,i·hi,t-1+wox,i·xit+bo,i) (79)
hit=oit*tanh(Cit) (80)
B. each node is along tree from leaf node propagated forward to root node
n1it=σ (netn1,i,t)=σ (wn1h,i·hi,t-1+wn1x,i·xit+bn1,i) (83)
n2it=σ (netn2,i,t)=σ (wn2h,i·hi,t-1+wn2x,i·xit+bn2,i) (84)
Nit=n1it*Rit+n2it*hit (85)
yt=σ (zt)=σ (wy·Nmt+by) (86)
Wherein fit,iit,ait,Cit,oit,hit,Rit,NitAnd yitIt is forgetting door of i-th of node in period t respectively, input
Door, information state, cell state, out gate, hidden layer output, central hub, neighborhood and predicted value；n1itAnd n2itAll being is
Neighborhood door of the i node in period t；PijIt is j-th of child node of i-th of node；Tanh and σ is that tanh and sigma swashs respectively
Function living；Symbol and * respectively represent multiplication between matrix multiplication and matrix element；Remaining variable is all intermediate variable.
The step of t-th of period error back propagation and calculation formula are as follows:
A. the target that the most common squared error function optimizes as needs is defined
B. the error of output layer is calculated
C. against tree direction from root node to leaf node reverse propagated error
D. Adam optimization algorithm is used, with [δ wLh,i,δwLx,i,δbL,i] and [δ wy,δby] update [wLh,wLx,bL] and
[wy,by]；For versatility, weight is indicated with symbol W, the gradient of weight is indicated with δ W, and Adam updates the general formula of weight
Are as follows:
mti=β1·mti-1+(1-β1)·δWti (107)
vti=β2·vti-1+(1-β2)·(δWti)2 (108)
Wherein EtFor error function, ytAnd YtRespectively predicted value and observation.β1,β2It is Adam parameter with ε, defaults respectively
Take 0.9,0.999 and 10-8.Ti is that the current update times of weight W and period t are distinguished.
Remaining variables are consistent with above-mentioned variable meaning, are intermediate variable, nothing there are also the variable that front is not mentioned
It need to know concrete meaning.
Predicted value is calculated according to above formula elder generation propagated forward, then backpropagation updates weight, it is referred to as primary to update.One
Iteration Ep wheel altogether, it is every to take turns training set DTaCriticizing for T size is taken to be trained, every batch of completes primary update.
(9) by test set DTeIt is input in trained NLSTM and is predicted, obtain forecasting wind speed result y.
Fig. 2 show equivalent tree causality structural schematic diagram；
Fig. 3 show NLSTM network structure.
Application of the invention is further described below with reference to specific experiment.
For the present invention using fuyun website meteorological data as object, data use 15 days to 2018 July in 2018 of August 14
Day one totally month meteorological data.Data time step-length is 1 hour, totally 744 periods, and dividing preceding 595 periods is training set,
149 periods are test set afterwards.Meteorological data includes totally 20 factors as shown in Table 1.Choose each factor the first two period
Feature of the value as present period.Calculate the Pearson correlation coefficient (PCC) and maximum information coefficient between this 20 factors
(MIC), as shown in table 2.On the basis of correlation analysis, Granger causality test is carried out to each factor, such as 3 institute of table
Show.Will be with mean wind speed (AWS, 13) in relevant causality drafting and Fig. 4 (a), and be converted to according to computing resource size
Fig. 4 (b) and two kinds shown in (d) tree constructions of equal value.
In order to verify the estimated performance of NLSTM, following four model is constructed to predict mean wind speed and compare:
1. LSTM-1: method use standard LSTM, feature take [5,6,7,8,9,11,15,13] but do not consider they because
Fruit relationship；
2. LSTM-2: method uses standard LSTM, and feature only takes [13]；
3. NLSTM-1: method uses NLSTM, using equivalent tree causality structure shown in Fig. 4 (b)；
4. NLSTM-2: method uses NLSTM, using equivalent tree causality structure shown in Fig. 4 (d)；
In order to avoid randomness, 4 models are run 20 times.Table 4 lists the evaluation of four model prediction mean wind speeds
Index.Evaluation index is all value using root-mean-square error (RMSE) and mean absolute error percentage (MAPE), the two indexs
Smaller, precision of prediction is higher.From table 4, it can be seen that the precision of prediction ratio LSTM-1 and LSTM-2 of NLSTM-2 and NLSTM-1 is
Want high, this illustrates that method NLSTM of the invention is more excellent than standard LSTM.The higher explanation of precision of prediction of NLSTM-2 ratio NLSTM-1
In the case where computing resource allows, the equivalent tree causality structure wind speed that it is obtained closer to true causality structure
Prediction result precision is higher.The difference of 4 model prediction accuracies can be more clearly found out in Fig. 5.
1 meteorological factor of table illustrates table
2 fuyun station correlation analysis table of table
。
3 fuyun station Granger causality analytical table of table
4 fuyun station forecasting wind speed index contrast table of table
In the above-described embodiments, can come wholly or partly by software, hardware, firmware or any combination thereof real
It is existing.When using entirely or partly realizing in the form of a computer program product, the computer program product include one or
Multiple computer instructions.When loading on computers or executing the computer program instructions, entirely or partly generate according to
Process described in the embodiment of the present invention or function.The computer can be general purpose computer, special purpose computer, computer network
Network or other programmable devices.The computer instruction may be stored in a computer readable storage medium, or from one
Computer readable storage medium is transmitted to another computer readable storage medium, for example, the computer instruction can be from one
A web-site, computer, server or data center pass through wired (such as coaxial cable, optical fiber, Digital Subscriber Line (DSL)
Or wireless (such as infrared, wireless, microwave etc.) mode is carried out to another web-site, computer, server or data center
Transmission).The computer-readable storage medium can be any usable medium or include one that computer can access
The data storage devices such as a or multiple usable mediums integrated server, data center.The usable medium can be magnetic Jie
Matter, (for example, floppy disk, hard disk, tape), optical medium (for example, DVD) or semiconductor medium (such as solid state hard disk Solid
State Disk (SSD)) etc..
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (10)
1. a kind of wind speed forecasting method based on neighborhood door shot and long term memory network, which is characterized in that described long based on neighborhood door
The line between Pearson correlation coefficient and maximum information coefficient analysis variable is respectively adopted in the wind speed forecasting method of short-term memory network
Property and non-linear dependencies screen wind speed correlation factor；
On the basis of correlation analysis using Granger causality analysis wind speed and the wind speed factor in statistical significance because
Fruit relationship；Causal structure is classified, and passes through decomposition-dummy variables-beta pruning method for all types of causes and effects
Relationship is unified for a kind of equivalent tree causality structure；
Parity price tree causality structure passes through the shot and long term memory network model prediction wind speed based on neighborhood door.
2. the wind speed forecasting method as described in claim 1 based on neighborhood door shot and long term memory network, which is characterized in that described
Wind speed forecasting method based on neighborhood door shot and long term memory network specifically includes:
(1) collect wind speed Y and may be wind speed impact factor data
(2) it is utilized respectively Pearson correlation coefficient MIC and maximum information coefficient MIC analysis wind speed and may be the air speed influence factor
Between linear and nonlinear correlation, obtain wind speed correlation factor [x1,x2,,…,xn], the Pearson correlation coefficient with wind speed
The impact factor of absolute value or maximum information coefficient greater than 0.5 is used as wind speed correlation factor；
(3) wind speed Y and wind speed correlation factor [x is analyzed using Granger causality1,x2,,…,xn] in statistical significance because
Fruit relationship；
(4) according to shape causal between wind speed and wind speed correlation factor, causality structure is divided into central hub, chain
Formula, cyclic annular, tree-shaped and network-like totally five kinds of structures；Central hub structure and chain structure are tree respectively extending transversely
With the special case of Longitudinal Extension, cyclic structure is decomposed into a series of chain structure, central hub structure, chain structure and cyclic annular knot
Three kinds of causality structures of structure are converted into tree-shaped causality structure；
(5) network-like structure is decomposited into a plurality of chain structure against causality since wind speed, be present in a plurality of decomposition line
The factor with dummy variables replace and distinguish, by each item decomposition line be combined into tree, according to computing resource size to tree-shaped knot
Structure carries out beta pruning, obtains final tree construction of equal value, therefore all types of causality structures are converted into equivalent tree cause and effect
Relational structure；
(6) according to equivalent tree causality structure, the training set D being made of the wind speed factor and wind speed is constructedTa=[xTa,YTa] and
The test set D being only made of predictorTe=[xTe], and data are normalized；
(7) the shot and long term memory network NLSTM based on neighborhood door is constructed according to equivalent tree causality, the parameter of NLSTM is set,
Including input layer number ni, hidden layer number of nodes nh, output layer number of nodes no, fixed learning rate η, crowd size T, exercise wheel number
Ep；
(8) using the Adam optimization algorithm in conjunction with mini-batch mechanism in training set DTaUpper trained NLSTM；
(9) by test set DTeIt is input in trained NLSTM and is predicted, obtain forecasting wind speed result y.
3. the wind speed forecasting method as claimed in claim 2 based on neighborhood door shot and long term memory network, which is characterized in that step
(8) in, the step of t-th of period information propagated forward and calculation formula are as follows:
A. the propagated forward of each node elder generation complete independently standard LSTM
fit=σ (netf,i,t)=σ (wfh,i·hi,t-1+wfx,i·xit+bf,i)(1)
iit=σ (neti,i,t)=σ (wih,i·hi,t-1+wix,i·xit+bi,i)(2)
ait=tanh (neta,i,t)=tanh (wah,i·hi,t-1+wax,i·xit+ba,i)(3)
Cit=fit*Ci,t-1+iit*ait(4)
oit=σ (neto,i,t)=σ (woh,i·hi,t-1+wox,i·xit+bo,i)(5)
hit=oit*tanh(Cit)(6)
B. each node is along tree from leaf node propagated forward to root node
rijt=σ (netr,i,j,t)=σ (wrh,i,j·hi,t-1+wrx,i,j·xi,t+br,i,j)(Pi≠)(7)
n1it=σ (netn1,i,t)=σ (wn1h,i·hi,t-1+wn1x,i·xit+bn1,i)(9)
n2it=σ (netn2,i,t)=σ (wn2h,i·hi,t-1+wn2x,i·xit+bn2,i)(10)
Nit=n1it*Rit+n2it*hit(11)
yt=σ (zt)=σ (wy·Nmt+by)(12)
Wherein fit,iit,ait,Cit,oit,hit,Rit,NitAnd yitIt is forgetting door of i-th of node in period t respectively, input gate,
Information state, cell state, out gate, hidden layer output, central hub, neighborhood and predicted value；n1itAnd n2itAll being is i-th
Neighborhood door of the node in period t；PijIt is j-th of child node of i-th of node；Tanh and σ is tanh and sigmoid activation respectively
Function；Symbol and * respectively represent multiplication between matrix multiplication and matrix element；Remaining variable is all intermediate variable.
4. the wind speed forecasting method as claimed in claim 2 based on neighborhood door shot and long term memory network, which is characterized in that step
(8) in, the step of t-th of period error back propagation and calculation formula are as follows:
A. the target that the most common squared error function optimizes as needs is defined
B. the error of output layer is calculated
C. against tree direction from root node to leaf node reverse propagated error
Wherein EtFor error function, ytAnd YtRespectively predicted value and observation.
5. the wind speed forecasting method as claimed in claim 4 based on neighborhood door shot and long term memory network, which is characterized in that t
A period weight update the step of include:
Using Adam optimization algorithm, with [δ wLh,i,δwLx,i,δbL,i] and [δ wy,δby] update [wLh,wLx,bL] and [wy,by]；
For versatility, weight is indicated with symbol W, the gradient of weight is indicated with δ W, and Adam updates the general formula of weight are as follows:
mti=β1·mti-1+(1-β1)·δWti(33)
vti=β2·vti-1+(1-β2)·(δWti)2(34)
β1,β2It is Adam parameter with ε, default takes 0.9,0.999 and 10 respectively-8；Ti is current update times and the period of weight W
T is distinguished；
Predicted value is calculated according to above formula elder generation propagated forward, then backpropagation updates weight, it is referred to as primary to update；It changes altogether
It is taken turns for Ep, it is every to take turns training set DTaCriticizing for T size is taken to be trained, every batch of completes primary update.
6. a kind of forecasting wind speed side realized based on neighborhood door shot and long term memory network described in Claims 1 to 5 any one
The computer program of method.
7. a kind of wind speed forecasting method realized based on neighborhood door shot and long term memory network described in Claims 1 to 5 any one
Information data processing terminal.
8. a kind of computer readable storage medium, including instruction, when run on a computer, so that computer is executed as weighed
Benefit requires the wind speed forecasting method based on neighborhood door shot and long term memory network described in 1-5 any one.
9. a kind of wind speed forecasting method realized based on neighborhood door shot and long term memory network described in claim 1 based on neighborhood door
The forecasting wind speed control system of shot and long term memory network.
10. a kind of prediction for carrying the forecasting wind speed control system based on neighborhood door shot and long term memory network described in claim 9
Power equipment of the wind speed to wind energy utilization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811296424.1A CN109063939B (en) | 2018-11-01 | 2018-11-01 | Wind speed prediction method and system based on neighborhood gate short-term memory network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811296424.1A CN109063939B (en) | 2018-11-01 | 2018-11-01 | Wind speed prediction method and system based on neighborhood gate short-term memory network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109063939A true CN109063939A (en) | 2018-12-21 |
| CN109063939B CN109063939B (en) | 2020-08-18 |
Family
ID=64788951
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811296424.1A Active CN109063939B (en) | 2018-11-01 | 2018-11-01 | Wind speed prediction method and system based on neighborhood gate short-term memory network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109063939B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110210660A (en) * | 2019-05-27 | 2019-09-06 | 河海大学 | A kind of ultra-short term wind speed forecasting method |
| CN110567534A (en) * | 2019-09-10 | 2019-12-13 | 广东工业大学 | Method for predicting flow of combustion air outlet in glass melting furnace and related device |
| CN110598911A (en) * | 2019-08-21 | 2019-12-20 | 同济大学 | Wind speed prediction method for wind turbine of wind power plant |
| CN110731787A (en) * | 2019-09-26 | 2020-01-31 | 首都师范大学 | fatigue state causal network method based on multi-source data information |
| CN111582551A (en) * | 2020-04-15 | 2020-08-25 | 中南大学 | Method and system for predicting short-term wind speed of wind power plant and electronic equipment |
| CN111624681A (en) * | 2020-05-26 | 2020-09-04 | 杨祺铭 | Hurricane intensity change prediction method based on data mining |
| CN112365045A (en) * | 2020-11-09 | 2021-02-12 | 上海明华电力科技有限公司 | Main steam temperature intelligent prediction method based on big data |
| CN112949201A (en) * | 2021-03-17 | 2021-06-11 | 华翔翔能科技股份有限公司 | Wind speed prediction method and device, electronic equipment and storage medium |
| CN113466634A (en) * | 2021-08-20 | 2021-10-01 | 青岛鼎信通讯股份有限公司 | Ground fault waveform identification method based on fault indicator |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103746370A (en) * | 2013-12-20 | 2014-04-23 | 河海大学 | Wind-power-plant reliability modeling method |
| CN104217260A (en) * | 2014-09-19 | 2014-12-17 | 南京信息工程大学 | Combined filling system for measured wind speed loss values of multiple neighboring wind motors in wind field |
| CN104573363A (en) * | 2015-01-05 | 2015-04-29 | 南方电网科学研究院有限责任公司 | Spatial valuing method of design air speed of overhead transmission line of coastal region |
| CN104657791A (en) * | 2015-02-28 | 2015-05-27 | 武汉大学 | Wind power plant group wind speed distribution prediction method based on correlation analysis |
| CN105654207A (en) * | 2016-01-07 | 2016-06-08 | 国网辽宁省电力有限公司锦州供电公司 | Wind power prediction method based on wind speed information and wind direction information |
| CN107239859A (en) * | 2017-06-05 | 2017-10-10 | 国网山东省电力公司电力科学研究院 | The heating load forecasting method of Recognition with Recurrent Neural Network is remembered based on series connection shot and long term |
| CN108197394A (en) * | 2018-01-05 | 2018-06-22 | 上海电气分布式能源科技有限公司 | A kind of wind speed curve emulation mode |
| CN108280551A (en) * | 2018-02-02 | 2018-07-13 | 华北电力大学 | A kind of photovoltaic power generation power prediction method using shot and long term memory network |
| CN108615097A (en) * | 2018-05-10 | 2018-10-02 | 广东工业大学 | A kind of wind speed forecasting method, system, equipment and computer readable storage medium |
| CN108711847A (en) * | 2018-05-07 | 2018-10-26 | 国网山东省电力公司电力科学研究院 | A kind of short-term wind power forecast method based on coding and decoding shot and long term memory network |
-
2018
- 2018-11-01 CN CN201811296424.1A patent/CN109063939B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103746370A (en) * | 2013-12-20 | 2014-04-23 | 河海大学 | Wind-power-plant reliability modeling method |
| CN104217260A (en) * | 2014-09-19 | 2014-12-17 | 南京信息工程大学 | Combined filling system for measured wind speed loss values of multiple neighboring wind motors in wind field |
| CN104573363A (en) * | 2015-01-05 | 2015-04-29 | 南方电网科学研究院有限责任公司 | Spatial valuing method of design air speed of overhead transmission line of coastal region |
| CN104657791A (en) * | 2015-02-28 | 2015-05-27 | 武汉大学 | Wind power plant group wind speed distribution prediction method based on correlation analysis |
| CN105654207A (en) * | 2016-01-07 | 2016-06-08 | 国网辽宁省电力有限公司锦州供电公司 | Wind power prediction method based on wind speed information and wind direction information |
| CN107239859A (en) * | 2017-06-05 | 2017-10-10 | 国网山东省电力公司电力科学研究院 | The heating load forecasting method of Recognition with Recurrent Neural Network is remembered based on series connection shot and long term |
| CN108197394A (en) * | 2018-01-05 | 2018-06-22 | 上海电气分布式能源科技有限公司 | A kind of wind speed curve emulation mode |
| CN108280551A (en) * | 2018-02-02 | 2018-07-13 | 华北电力大学 | A kind of photovoltaic power generation power prediction method using shot and long term memory network |
| CN108711847A (en) * | 2018-05-07 | 2018-10-26 | 国网山东省电力公司电力科学研究院 | A kind of short-term wind power forecast method based on coding and decoding shot and long term memory network |
| CN108615097A (en) * | 2018-05-10 | 2018-10-02 | 广东工业大学 | A kind of wind speed forecasting method, system, equipment and computer readable storage medium |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110210660B (en) * | 2019-05-27 | 2022-07-22 | 河海大学 | Ultra-short-term wind speed prediction method |
| CN110210660A (en) * | 2019-05-27 | 2019-09-06 | 河海大学 | A kind of ultra-short term wind speed forecasting method |
| CN110598911A (en) * | 2019-08-21 | 2019-12-20 | 同济大学 | Wind speed prediction method for wind turbine of wind power plant |
| CN110598911B (en) * | 2019-08-21 | 2022-06-21 | 同济大学 | Wind speed prediction method for wind turbine of wind power plant |
| CN110567534A (en) * | 2019-09-10 | 2019-12-13 | 广东工业大学 | Method for predicting flow of combustion air outlet in glass melting furnace and related device |
| CN110567534B (en) * | 2019-09-10 | 2021-08-13 | 广东工业大学 | Method for predicting flow of combustion air outlet in glass melting furnace and related device |
| CN110731787A (en) * | 2019-09-26 | 2020-01-31 | 首都师范大学 | fatigue state causal network method based on multi-source data information |
| CN110731787B (en) * | 2019-09-26 | 2022-07-22 | 首都师范大学 | Fatigue state causal network method based on multi-source data information |
| CN111582551A (en) * | 2020-04-15 | 2020-08-25 | 中南大学 | Method and system for predicting short-term wind speed of wind power plant and electronic equipment |
| CN111582551B (en) * | 2020-04-15 | 2023-12-08 | 中南大学 | Wind power plant short-term wind speed prediction method and system and electronic equipment |
| CN111624681A (en) * | 2020-05-26 | 2020-09-04 | 杨祺铭 | Hurricane intensity change prediction method based on data mining |
| CN112365045A (en) * | 2020-11-09 | 2021-02-12 | 上海明华电力科技有限公司 | Main steam temperature intelligent prediction method based on big data |
| CN112949201A (en) * | 2021-03-17 | 2021-06-11 | 华翔翔能科技股份有限公司 | Wind speed prediction method and device, electronic equipment and storage medium |
| CN113466634A (en) * | 2021-08-20 | 2021-10-01 | 青岛鼎信通讯股份有限公司 | Ground fault waveform identification method based on fault indicator |
| CN113466634B (en) * | 2021-08-20 | 2023-12-29 | 青岛鼎信通讯股份有限公司 | Ground fault waveform identification method based on fault indicator |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109063939B (en) | 2020-08-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109063939A (en) | A kind of wind speed forecasting method and system based on neighborhood door shot and long term memory network | |
| Li et al. | A hybrid model based on synchronous optimisation for multi-step short-term wind speed forecasting | |
| Tian et al. | Short-term wind speed prediction based on improved PSO algorithm optimized EM-ELM | |
| Zhao et al. | Towards traffic matrix prediction with LSTM recurrent neural networks | |
| Tencaliec et al. | Flexible semiparametric generalized Pareto modeling of the entire range of rainfall amount | |
| Nichiforov et al. | Deep learning techniques for load forecasting in large commercial buildings | |
| CN103197983B (en) | Service component reliability online time sequence predicting method based on probability graph model | |
| CN106649479B (en) | Transformer state association rule mining method based on probability graph | |
| CN110059867B (en) | Wind speed prediction method combining SWLSTM and GPR | |
| CN111612262A (en) | Wind power probability prediction method based on quantile regression | |
| CN103279672B (en) | Short-term wind speed forecasting method based on noise-model support-vector regression technique | |
| Niu et al. | Short‐term wind speed hybrid forecasting model based on bias correcting study and its application | |
| CN111985719A (en) | Power load prediction method based on improved long-term and short-term memory network | |
| Zhao et al. | Spatio‐temporal Markov chain model for very‐short‐term wind power forecasting | |
| Souhe et al. | A hybrid model for forecasting the consumption of electrical energy in a smart grid | |
| CN113837432A (en) | Power system frequency prediction method driven by physics-data combination | |
| CN110738363B (en) | Photovoltaic power generation power prediction method | |
| Girimurugan et al. | Application of deep learning to the prediction of solar irradiance through missing data | |
| AU2021106200A4 (en) | Wind power probability prediction method based on quantile regression | |
| Souhe et al. | Forecasting of electrical energy consumption of households in a smart grid | |
| Zhang et al. | A method based on Lorenz disturbance and variational mode decomposition for wind speed prediction | |
| Zhang et al. | Network security situation prediction model based on EMD and ELPSO optimized BiGRU neural network | |
| Zeng et al. | Decision tree classification model for popularity forecast of Chinese colleges | |
| CN116911459A (en) | Multi-input multi-output ultra-short-term power load prediction method suitable for virtual power plant | |
| Heng et al. | A hybrid forecasting model based on empirical mode decomposition and the cuckoo search algorithm: a case study for power load |
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 |