CN109063939B - Wind speed prediction method and system based on neighborhood gate short-term memory network - Google Patents
Wind speed prediction method and system based on neighborhood gate short-term memory network Download PDFInfo
- Publication number
- CN109063939B CN109063939B CN201811296424.1A CN201811296424A CN109063939B CN 109063939 B CN109063939 B CN 109063939B CN 201811296424 A CN201811296424 A CN 201811296424A CN 109063939 B CN109063939 B CN 109063939B
- Authority
- CN
- China
- Prior art keywords
- wind speed
- causal relationship
- neighborhood
- term memory
- tree
- 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.)
- Active
Links
Images
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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Environmental & Geological Engineering (AREA)
- Atmospheric Sciences (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Biodiversity & Conservation Biology (AREA)
- Ecology (AREA)
- Environmental Sciences (AREA)
- Molecular Biology (AREA)
- Development Economics (AREA)
- Computational Linguistics (AREA)
- Game Theory and Decision Science (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (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 belongs to the technical field of wind speed prediction, and discloses a wind speed prediction method and a system based on a neighborhood gate long-term and short-term memory network, wherein linear and nonlinear correlations among variables are explored by respectively adopting a Pearson correlation coefficient and a maximum information coefficient to screen a wind speed correlation factor; on the basis of correlation analysis, a Glan's cause and effect relationship test is utilized to explore the cause and effect relationship of the wind speed and the wind speed factor in the statistical sense; dividing the causal relationship structure into 5 types, and unifying all types of causal relationships into an equivalent tree causal relationship structure by a decomposition-virtual variable-pruning method; and aiming at the causal relationship structure of the equivalent tree, a long-term and short-term memory network model based on a neighborhood gate is provided to predict the wind speed. The forecasting method (NLSTM) accurately considers the causal relationship between the wind speed and the wind speed factor, effectively improves the forecasting precision of the wind speed, and plays a vital role in wind power application and power grid dispatching.
Description
Technical Field
The invention belongs to the technical field of wind speed prediction, and particularly relates to a wind speed prediction method and system based on a neighborhood gate length short-term memory network.
Background
Currently, the current state of the art commonly used in the industry is such that:
wind energy is a promising renewable clean energy source and has received widespread attention in recent years from all over the world. More and more wind power is connected to the power system, so that the power system becomes unreliable, which is caused by strong fluctuation and strong randomness of wind speed. Therefore, accurately predicting wind speed plays a crucial role in the utilization of wind energy and efficient scheduling of power systems. Wind speed is affected by many meteorological factors, including factors such as air pressure, temperature, humidity, etc. The wind speed prediction is difficult due to the complex relationship among the factors, and the accuracy of the wind speed prediction by the traditional machine learning method is limited.
Deep learning methods long short term memory networks (LSTM) have a high prediction accuracy when solving time series prediction problems like wind speed, but LSTM is often used as a black box model, which makes the model less interpretable. The correlation of the wind speed influence factors is analyzed through characteristic engineering, and the causal relationship between the wind speed influence factors is cleared, so that the wind speed prediction precision is improved, and the interpretability of the model is enhanced. Therefore, how to analyze the causal relationship between the wind speed and the related factors and accurately consider the causal relationship into the LSTM is a theoretical and practical engineering problem to be solved urgently so as to improve the wind speed prediction accuracy and enhance the interpretability of the model.
In the feature engineering, commonly used correlation analysis methods include a graph method, a correlation coefficient method, a covariance method, a maximum information coefficient method, and the like. Common causal relationship analysis methods include theoretical analysis, transmission entropy and glovey causal relationship test.
In the present invention, the pearson correlation coefficient is used to explore the linear correlation between the factors, and the maximum information coefficient is used to explore the non-linear correlation between the factors. The glange causal relationship test is used to explore causal relationships between factors.
In summary, the problems of the prior art are as follows:
the causal relationship structure types among the factors are complex and various, and few scholars classify the causal relationship structures at present, so that how to scientifically and completely classify all the causal relationship structure types is also a problem in the prior art.
How to conveniently and effectively use the causal relationship structure after classification by using the case which is not available for reference at present, so that unifying the causal relationship structure after classification into a universal causal relationship structure is also a problem faced by the prior art.
In the prior art, the wind speed is difficult to predict due to the complex relationship among factors, and the accuracy of predicting the wind speed by using the traditional machine learning method is limited.
In the prior art, because the LSTM only has one characteristic input interface, the LSTM can only input all factors without distinction and cannot accurately consider the causal relationship obtained through characteristic engineering, and the causal relationship structure of the wind speed factor cannot be accurately considered into the LSTM.
The difficulty and significance for solving the technical problems are as follows:
the difficulty with classifying causal structures is how the classification can encompass all causal structure types. The completeness of classification is therefore the basis for the subsequent techniques.
The difficulty in unifying the causal structures is how to find the common points among the classified causal structures to obtain a common structure. This generic structure may not only represent all types of causal structures but also needs to be an easily predictable structure. Therefore, the representativeness and operability of the unified causal relationship structure play a role in the invention.
After the technical problem is solved, the significance is brought as follows:
in order to enable the LSTM to have the capability of accurately considering the obtained wind speed causal relationship structure, the invention provides a long-short term memory Network (NLSTM) based on a neighborhood gate. NLSTM differs from LSTM in network structure, and correctly deducing forward and backward propagation formulas of NLSTM is a difficulty in realizing NLSTM. The accurate realization of NLSTM is the guarantee for improving the wind speed prediction precision.
Because the causal relationship structures of wind speeds in different areas may be different, it is also one of the difficulties to popularize the technology of the present invention. Therefore, the designed NLSTM can be correspondingly changed according to different causal relationship structures, and the popularization of the NLSTM is very favorable.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a wind speed prediction method and system based on a neighborhood gate length short-term memory network, which can accurately consider the causal relationship structure between wind speed and wind speed influence factors and can obtain a high-precision wind speed prediction result.
The invention is realized in such a way that a wind speed prediction method based on a neighborhood gate length short-term memory network comprises the following steps: the wind speed prediction method based on the neighborhood gate long and short term memory network screens wind speed related factors by respectively adopting Pearson related coefficients and linear and nonlinear correlations between maximum information coefficient analysis variables;
analyzing the causal relationship of the wind speed and the wind speed factor in the statistical significance by utilizing the Glan's causal relationship on the basis of the correlation analysis; dividing the causal relationship structure into 5 types, and unifying all types of causal relationships into an equivalent tree causal relationship structure by a decomposition-virtual variable-pruning method;
and predicting the wind speed of the causal relationship structure of the equivalent tree through a long-term and short-term memory network model based on the neighborhood gate.
The method specifically comprises the following steps:
(2) Linear and non-linear correlations between wind speed and possibly wind speed influencing factors are analyzed using Pearson correlation coefficient (MIC) and Maximum Information Coefficient (MIC), respectively, to obtain wind speed correlation factor [ x ]1,x2,,…,xn]The influence factor of the absolute value of the Pearson correlation coefficient with the wind speed or the maximum information coefficient of 0.5 or more can be used as the wind speed correlation factor.
(3) Method for detecting and exploring wind speed Y and wind speed related factor [ x ] by using Glanberg causal relationship1,x2,,…,xn]Causal relationship in a statistical sense.
(4) According to the shape of the causal relationship among the wind speed and the wind speed related factors, the causal relationship structure is divided into five structures including a central hub, a chain structure, a ring structure, a tree structure and a network structure. Through careful analysis, it can be found that the central hub structure and the chain structure are special cases of the tree structure in the horizontal extension and the vertical extension, respectively, and the ring structure can be decomposed into a series of chain structures, so that the first three kinds of causal relationship structures can be converted into the tree causal relationship structure.
(5) The network-like structure is a general form of a causal structure. Decomposing a network structure into a plurality of chain structures in the direction of an arrow of inverse causal relationship from wind speed, replacing and distinguishing factors in a plurality of decomposition lines by virtual variables (the variables have the same attribute with actual factors but different numbers and are virtualized for distinguishing the same factor in different decomposition lines), combining all the decomposition lines into a tree structure (the recombined tree is very huge), and pruning the tree structure according to the size of computing resources to obtain a final equivalent tree structure, so that all types of causal relationship structures can be converted into the equivalent tree causal relationship structure.
(6) Constructing a training set D consisting of a wind speed factor and a wind speed according to the causal relationship structure of the equivalent treeTa=[xTa,YTa]And test set D consisting of predictor onlyTe=[xTe]And carrying out normalization processing on the data.
(7) Constructing a long-short term memory Network (NLSTM) based on a neighborhood gate according to the causal relationship of an equivalent tree, and setting parameters of the NLSTM, including the number n of nodes of an input layeriNumber of hidden layer nodes nhNumber of nodes of output layer noFixed learning rate η, batch size T, number of training rounds Ep.
(8) Adam optimization algorithm combined with mini-batch mechanism is adopted in training set DTaNLSTM was trained.
(9) Test set DTeAnd inputting the wind speed prediction result into a trained NLSTM for prediction to obtain a wind speed prediction result y.
Further, in step (8), the step of information forward propagation of the t-th time period and the calculation formula are as follows:
a. each node completes the forward propagation of standard LSTM independently
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 propagates forward along the tree from the leaf node to the 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 f isit,iit,ait,Cit,oit,hit,Rit,NitAnd yitRespectively, a forgetting gate, an input gate, an information state, a cell state, an output gate, a hidden layer output, a central pivot, a neighborhood and a predicted value of the ith node in a time period t; n is1itAnd n2itAll are neighborhood gates of the ith node in the time period t; pijIs the jth child node of the ith node; tanh and σ are the tan h and sigmoid activation functions, respectively; the symbols sum represent matrix multiplication and multiplication between matrix elements, respectively; the remaining variables are all intermediate variables.
Further, in step (8), the step of t-th period error back propagation and the calculation formula are as follows:
a. defining the most common square error function as the target to be optimized
b. Calculating errors of output layers
c. Counter-propagating errors from root node to leaf node against tree direction
d. Using Adam optimization algorithm with [ w ]Lh,i,wLx,i,bL,i]And [ wy,by]To update [ wLh,wLx,bL]And [ wy,by](ii) a For generality, the weight is denoted by the symbol W, the gradient of the weight is denoted by W, and the general formula for Adam to update the weight is:
mti=β1·mti-1+(1-β1)·Wti(70)
vti=β2·vti-1+(1-β2)·(Wti)2(71)
wherein EtAs an error function, ytAnd Ytβ for predicted and observed values, respectively1,β2And Adam's parameters, default to 0.9,0.999 and 10, respectively-8(ii) a ti is the current update times of the weight W, and is distinguished from the time period t;
calculating a predicted value by forward propagation according to the formula, and then updating the weight by backward propagation, which is called primary updating; a total iteration of Ep rounds, each round of which will train set DTaAnd (5) taking batches with the size of T for training, and finishing updating once in each batch.
Another object of the present invention is to provide a computer program for implementing the wind speed prediction method based on the neighborhood gate length short term memory network.
The invention also aims to provide an information data processing terminal for realizing the wind speed prediction method based on the neighborhood gate length short-term memory network.
It is another object of the present invention to provide a computer-readable storage medium, comprising instructions which, when executed on a computer, cause the computer to perform the wind speed prediction method based on a neighborhood gate short term memory network.
The invention also aims to provide a wind speed prediction control system based on the neighborhood gate length short-term memory network, which realizes the wind speed prediction method based on the neighborhood gate length short-term memory network.
The invention also aims to provide the power equipment for utilizing the wind energy by predicting the wind speed, which is provided with the wind speed prediction control system based on the neighborhood gate length short-term memory network.
In summary, the advantages and positive effects of the invention are:
the invention provides a wind speed prediction method based on a neighborhood gate length short-term memory network, which analyzes the causal relationship between wind speed and wind speed factors through characteristic engineering and converts the causal relationship structure into an equivalent tree causal relationship structure by adopting a decomposition-virtual variable-pruning method. NLSTM significantly enhances model interpretability, and can accurately consider the equivalent tree causal structure, which the prior art cannot accurately consider.
The model NLSTM provided by the invention has good universality and easy popularization, and a corresponding network structure can be transformed according to different wind speed causal relationship structures.
The wind speed prediction result obtained by the model provided by the invention has high precision, and the wind speed prediction result can be analyzed from the comparison of the wind speed prediction indexes in the attached table 4 and can also be visually seen from the attached table 5.
Drawings
FIG. 1 is a flow chart of a method for predicting wind speed of a neighborhood gate length short term memory network according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of an Equivalence Tree causal relationship structure provided by an embodiment of the present invention;
fig. 3 is a structural diagram of an NLSTM network according to an embodiment of the present invention;
FIG. 4 is a diagram illustrating a causal relationship of wind speed in a Xinjiang Fuzi station case and an equivalent tree thereof according to an embodiment of the present invention;
FIG. 5 is a comparative graph of wind speed prediction results of the Xinjiang Fuchun station case provided by the embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The method firstly adopts a 'decomposition-virtual variable-pruning' method to convert all types of causal relationship structures into a uniform equivalent tree structure, and then establishes a long-short term memory Network (NLSTM) based on a neighborhood gate corresponding to the equivalent tree structure to predict the wind speed. The structure of NLSTM is the same as the equivalent tree structure, so that the causal relationship structure among factors can be accurately considered.
Fig. 1 is a general flowchart of a wind speed prediction method of a neighborhood gate length short-term memory network according to the present invention, which specifically includes the following steps:
(2) Linear and non-linear correlations between wind speed and possibly wind speed influencing factors are analyzed using Pearson correlation coefficient (MIC) and Maximum Information Coefficient (MIC), respectively, to determine the wind speedObtaining the wind speed related factor [ x1,x2,,…,xn]The influence factor of the absolute value of the Pearson correlation coefficient with the wind speed or the maximum information coefficient of 0.5 or more can be used as the wind speed correlation factor.
(3) Method for detecting and exploring wind speed Y and wind speed related factor [ x ] by using Glanberg causal relationship1,x2,,…,xn]Causal relationship in a statistical sense.
(4) According to the shape of the causal relationship among the wind speed and the wind speed related factors, the causal relationship structure is divided into five structures including a central hub, a chain structure, a ring structure, a tree structure and a network structure. Through careful analysis, it can be found that the central hub structure and the chain structure are special cases of the tree structure in the horizontal extension and the vertical extension, respectively, and the ring structure can be decomposed into a series of chain structures, so that the first three kinds of causal relationship structures can be converted into the tree causal relationship structure.
(5) The network-like structure is a general form of a causal structure. Decomposing a network structure into a plurality of chain structures in the direction of an arrow of inverse causal relationship from wind speed, replacing and distinguishing factors in a plurality of decomposition lines by virtual variables (the variables have the same attribute with actual factors but different numbers and are virtualized for distinguishing the same factor in different decomposition lines), combining all the decomposition lines into a tree structure (the recombined tree is very huge), and pruning the tree structure according to the size of computing resources to obtain a final equivalent tree structure, so that all types of causal relationship structures can be converted into the equivalent tree causal relationship structure.
(6) Constructing a training set D consisting of a wind speed factor and a wind speed according to the causal relationship structure of the equivalent treeTa=[xTa,YTa]And test set D consisting of predictor onlyTe=[xTe]And carrying out normalization processing on the data.
(7) Constructing a long-short term memory Network (NLSTM) based on a neighborhood gate according to the causal relationship of an equivalent tree, and setting parameters of the NLSTM, including the number n of nodes of an input layeriNumber of hidden layer nodes nhNumber of nodes of output layer noFixed learning rate η, batchSize T, number of training rounds Ep. The weights of the nodes are initialized according to parameters, 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) Adam optimization algorithm combined with mini-batch mechanism is adopted in training set DTaNLSTM was trained. The implementation of NLSTM involves forward propagation of information and backward propagation of errors.
The step and the calculation formula of the forward propagation of the information of the t-th time interval are as follows:
a. each node completes the forward propagation of standard LSTM independently
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 propagates forward along the tree from the leaf node to the 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 f isit,iit,ait,Cit,oit,hit,Rit,NitAnd yitRespectively, a forgetting gate, an input gate, an information state, a cell state, an output gate, a hidden layer output, a central pivot, a neighborhood and a predicted value of the ith node in a time period t; n is1itAnd n2itAll are neighborhood gates of the ith node in the time period t; pijIs the jth child node of the ith node; tan h and σ are tan h and sigma activation functions, respectively; the symbols sum represent matrix multiplication and multiplication between matrix elements, respectively; the remaining variables are all intermediate variables.
The step and the calculation formula of the error back propagation in the t period are as follows:
a. defining the most common square error function as the target to be optimized
b. Calculating errors of output layers
c. Counter-propagating errors from root node to leaf node against tree direction
d. Using Adam optimization algorithm with [ w ]Lh,i,wLx,i,bL,i]And [ wy,by]To update [ wLh,wLx,bL]And [ wy,by](ii) a For generality, the weight is denoted by the symbol W, the gradient of the weight is denoted by W, and the general formula for Adam to update the weight is:
mti=β1·mti-1+(1-β1)·Wti(107)
vti=β2·vti-1+(1-β2)·(Wti)2(108)
wherein EtAs an error function, ytAnd Ytβ for predicted and observed values, respectively1,β2And Adam's parameters, default to 0.9,0.999 and 10, respectively-8. ti is the current update times of the weight W, and is distinguished from the time period t.
The remaining variables are synonymous with the previously mentioned variables, and the previously non-mentioned variables are intermediate variables, and no specific meaning is required.
According to the formula, the predicted value is calculated by forward propagation, and then the updating weight is updated by backward propagation, which is called once updating. A total iteration of Ep rounds, each round of which will train set DTaAnd (5) taking batches with the size of T for training, and finishing updating once in each batch.
(9) Test set DTeAnd inputting the wind speed prediction result into a trained NLSTM for prediction to obtain a wind speed prediction result y.
FIG. 2 is a diagram illustrating an Equivalence Tree cause and effect relationship structure;
fig. 3 shows a structure diagram of the NLSTM network.
The use of the present invention is further described below in conjunction with specific experiments.
The method takes meteorological data of Xinjiang Fuziji sites as objects, and the data adopts the meteorological data of one month from 7 and 15 days in 2018 to 8 and 14 days in 2018. The data time step is 1 hour, 744 time periods are totally, 595 time periods before division are taken as a training set, and 149 time periods after division are taken as a test set. The meteorological data includes a total of 20 factors as shown in table 1. And selecting the values of the first two time periods of each factor as the characteristics of the current time period. Pearson's Correlation Coefficient (PCC) and Maximum Information Coefficient (MIC) among the 20 factors were calculated as shown in Table 2. Based on the correlation analysis, the respective factors were subjected to the glange causal relationship test, as shown in table 3. The causal relationships associated with the average wind speed (AWS,13) are plotted in fig. 4(a) and converted to two equivalent tree structures as shown in fig. 4(b) and (d) based on computational resource size.
To validate the predictive performance of NLSTM, the following four models were constructed to predict average wind speed and compared:
LSTM-1: the method adopts standard LSTM, and is characterized in that [5,6,7,8,9,11,15,13] is obtained without considering the cause and effect relationship;
② LSTM-2: the method adopts standard LSTM, and only takes the characteristics of [13 ];
③ NLSTM-1: the method adopts NLSTM, and adopts an equivalent tree causal relationship structure shown in FIG. 4 (b);
(iv) NLSTM-2: the method adopts NLSTM, and adopts an equivalent tree causal relationship structure shown in FIG. 4 (d);
to avoid randomness, 4 models were run 20 times each. Table 4 lists the evaluation indices for the four models to predict the average wind speed. The evaluation indexes adopt Root Mean Square Error (RMSE) and mean absolute error percentage (MAPE), and the smaller the two indexes are, the higher the prediction precision is. As can be seen from Table 4, the prediction accuracy of NLSTM-2 and NLSTM-1 is higher than that of both LSTM-1 and LSTM-2, which shows that the method NLSTM of the present invention is superior to standard LSTM. The higher prediction accuracy of NLSTM-2 compared with NLSTM-1 shows that the closer the equivalent tree causal relationship structure is to the real causal relationship structure, the higher the accuracy of the obtained wind speed prediction result is under the permission of computing resources. The difference in prediction accuracy of the 4 models can be seen more clearly in fig. 5.
TABLE 1 Meteorological factor description Table
TABLE 2 correlation analysis table for Xinjiang Fuyun station
TABLE 3 analysis table of causal relationship between Xinjiang Fuyuntang Glandujie
TABLE 4 comparison table of wind speed prediction indexes of Xinjiang Fuyun station
In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When used in whole or in part, can be implemented in a computer program product that includes one or more computer instructions. When loaded or executed on a computer, cause the flow or functions according to embodiments of the invention to occur, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, the computer instructions may be transmitted from one website site, computer, server, or data center to another website site, computer, server, or data center via wire (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL), or wireless (e.g., infrared, wireless, microwave, etc.)). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that includes one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk (SSD)), among others.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (5)
1. A wind speed prediction method based on a neighborhood gate long and short term memory network is characterized in that the wind speed prediction method based on the neighborhood gate long and short term memory network respectively adopts Pearson correlation coefficient and maximum information coefficient to analyze linear and nonlinear correlation among variables to screen wind speed correlation factors;
analyzing the causal relationship of the wind speed and the wind speed factor in the statistical significance by utilizing the Glan's causal relationship on the basis of the correlation analysis; classifying the causal relationship structures, and unifying all types of causal relationships into an equivalent tree causal relationship structure by a decomposition-virtual variable-pruning method;
predicting the wind speed of the causal relationship structure of the equivalent tree through a long-term and short-term memory network model based on a neighborhood gate;
the wind speed prediction method based on the neighborhood gate length short-term memory network specifically comprises the following steps:
(2) Respectively analyzing linear and nonlinear correlations between wind speed and possible wind speed influence factors by using the Pearson correlation coefficient MIC and the maximum information coefficient MIC to obtain a wind speed correlation factor [ x ]1,x2,,…,xn]The influence factors of which the absolute value of the Pearson correlation coefficient or the maximum information coefficient of the wind speed is more than 0.5 are taken as the wind speed correlation factors;
(3) analysis of wind speed Y and related factor of wind speed [ x ] by using Glanberg causal relationship1,x2,,…,xn]Causal relationships in a statistical sense;
(4) dividing the causal relationship structure into five structures including a central hub, a chain structure, a ring structure, a tree structure and a network structure according to the shape of the causal relationship between the wind speed and the wind speed related factors; the central hub structure and the chain structure are special cases of the horizontal expansion and the vertical expansion of the tree structure respectively, the ring structure is decomposed into a series of chain structures, and the central hub structure, the chain structure and the ring structure are all converted into tree causal structures;
(5) decomposing a network structure into a plurality of chain structures from the wind speed according to the inverse causal relationship, replacing and distinguishing factors existing in a plurality of decomposition lines by using virtual variables, combining all the decomposition lines into a tree structure, and pruning the tree structure according to the size of calculation resources to obtain a final equivalent tree structure, so that all types of causal relationship structures are converted into equivalent tree causal relationship structures;
(6) constructing a training set D consisting of a wind speed factor and a wind speed according to the causal relationship structure of the equivalent treeTa=[xTa,YTa]And test set D consisting of predictor onlyTe=[xTe]And carrying out normalization processing on the data;
(7) constructing a long-short term memory network NLSTM based on a neighborhood gate according to the causal relationship of an equivalent tree, and setting parameters of the NLSTM, including the number n of nodes of an input layeriNumber of hidden layer nodes nhNumber of nodes of output layer noFixed learning rate η, batch size T, number of training rounds Ep;
(8) adam optimization algorithm combined with mini-batch mechanism is adopted in training set DTaTraining NLSTM;
(9) test set DTeAnd inputting the wind speed prediction result into a trained NLSTM for prediction to obtain a wind speed prediction result y.
2. The wind speed prediction method based on the neighborhood gate length short-term memory network as claimed in claim 1, wherein in the step (8), the step of forward propagation of the information of the t-th time period and the calculation formula are as follows:
a. each node completes the forward propagation of standard LSTM independently
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 propagates forward along the tree from the leaf node to the root node
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 f isit,iit,ait,Cit,oit,hit,Rit,NitAnd yitRespectively, a forgetting gate, an input gate, an information state, a cell state, an output gate, a hidden layer output, a central pivot, a neighborhood and a predicted value of the ith node in a time period t; n is1itAnd n2itAll are neighborhood gates of the ith node in the time period t; pijIs the jth child node of the ith node; tanh and σ are the tan h and sigmoid activation functions, respectively; the symbols sum represent matrix multiplication and multiplication between matrix elements, respectively; the remaining variables are all intermediate variables.
3. The wind speed prediction method based on the neighborhood gate length short-term memory network as claimed in claim 1, wherein in the step (8), the step of error back propagation in the t-th period and the calculation formula are as follows:
a. defining the most common square error function as the target to be optimized
b. Calculating errors of output layers
c. Counter-propagating errors from root node to leaf node against tree direction
Wherein EtAs an error function, ytAnd YtRespectively, predicted values and observed values.
4. The wind speed prediction method based on the neighborhood gate length short term memory network as claimed in claim 3, wherein the step of updating the weight of the t-th time period comprises:
using Adam optimization algorithm with [ w ]Lh,i,wLx,i,bL,i]And [ wy,by]To update [ wLh,wLx,bL]And [ wy,by](ii) a For generality, the weight is denoted by the symbol W, the gradient of the weight is denoted by W, and the general formula for Adam to update the weight is:
mti=β1·mti-1+(1-β1)·Wti(33)
vti=β2·vti-1+(1-β2)·(Wti)2(34)
β1,β2and Adam's parameters, default to 0.9,0.999 and 10, respectively-8(ii) a ti is the current update times of the weight W, and is distinguished from the time period t;
calculating a predicted value by forward propagation according to the formula, and then updating the weight by backward propagation, which is called primary updating; a total iteration of Ep rounds, each round of which will train set DTaAnd (5) taking batches with the size of T for training, and finishing updating once in each batch.
5. A computer-readable storage medium comprising instructions that, when executed on a computer, cause the computer to perform the method for wind speed prediction based on a neighborhood gate short term memory network of any of claims 1-4.
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 CN109063939A (en) | 2018-12-21 |
CN109063939B true 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) |
Families Citing this family (9)
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 |
CN110598911B (en) * | 2019-08-21 | 2022-06-21 | 同济大学 | Wind speed prediction method for wind turbine of wind power plant |
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 |
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 |
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 |
CN112949201B (en) * | 2021-03-17 | 2023-03-21 | 华翔翔能科技股份有限公司 | Wind speed prediction method and device, electronic equipment and storage medium |
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 |
Citations (6)
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 |
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 |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107239859B (en) * | 2017-06-05 | 2018-05-08 | 国网山东省电力公司电力科学研究院 | Heating load forecasting method based on series connection shot and long term memory Recognition with Recurrent Neural Network |
CN108280551B (en) * | 2018-02-02 | 2022-07-26 | 华北电力大学 | Photovoltaic power generation power prediction method utilizing long-term and short-term memory network |
CN108711847B (en) * | 2018-05-07 | 2019-06-04 | 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 |
-
2018
- 2018-11-01 CN CN201811296424.1A patent/CN109063939B/en active Active
Patent Citations (6)
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 |
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 |
Also Published As
Publication number | Publication date |
---|---|
CN109063939A (en) | 2018-12-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109063939B (en) | Wind speed prediction method and system based on neighborhood gate short-term memory network | |
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 | |
US7512570B2 (en) | Artificial intelligence analyzer and generator | |
CN108733508B (en) | Method and system for controlling data backup | |
CN112488395A (en) | Power distribution network line loss prediction method and system | |
CN110348622B (en) | Temperature prediction method and system based on machine learning and electronic equipment | |
CN112187554B (en) | Operation and maintenance system fault positioning method and system based on Monte Carlo tree search | |
CN111738520A (en) | System load prediction method fusing isolated forest and long-short term memory network | |
CN103336791A (en) | Hadoop-based fast rough set attribute reduction method | |
CN112396231B (en) | Modeling method and device for space-time data, electronic equipment and readable medium | |
CN113988477A (en) | Photovoltaic power short-term prediction method and device based on machine learning and storage medium | |
Chen et al. | Combined probabilistic forecasting method for photovoltaic power using an improved Markov chain | |
CN111985719A (en) | Power load prediction method based on improved long-term and short-term memory network | |
CN112162860A (en) | CPU load trend prediction method based on IF-EMD-LSTM | |
Girimurugan et al. | Application of deep learning to the prediction of solar irradiance through missing data | |
CN116822803A (en) | Carbon emission data graph construction method, device and equipment based on intelligent algorithm | |
AU2021106200A4 (en) | Wind power probability prediction method based on quantile regression | |
Shao et al. | Short‐Term Load Forecasting Based on EEMD‐WOA‐LSTM Combination Model | |
CN117688362A (en) | Photovoltaic power interval prediction method and device based on multivariate data feature enhancement | |
CN116112379B (en) | Dynamic prediction method for directed link of multidimensional service sharing equipment of data center | |
CN113837475B (en) | Method, system, equipment and terminal for forecasting runoff probability of directed graph deep neural network | |
US20230206084A1 (en) | Method, device, and program product for managing knowledge graphs | |
CN112560213B (en) | System modeling method and system based on model system engineering and hyper-network theory | |
CN115310355A (en) | Multi-energy coupling-considered multi-load prediction method and system for comprehensive energy system |
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 |