CN109615027B - Intelligent prediction method for extracting wind speed characteristics along high-speed railway - Google Patents
Intelligent prediction method for extracting wind speed characteristics along high-speed railway Download PDFInfo
- Publication number
- CN109615027B CN109615027B CN201811643482.7A CN201811643482A CN109615027B CN 109615027 B CN109615027 B CN 109615027B CN 201811643482 A CN201811643482 A CN 201811643482A CN 109615027 B CN109615027 B CN 109615027B
- Authority
- CN
- China
- Prior art keywords
- wind speed
- model
- preselection
- characteristic
- clustering
- 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
- 238000000034 method Methods 0.000 title claims abstract description 50
- 239000013598 vector Substances 0.000 claims abstract description 105
- 238000010606 normalization Methods 0.000 claims abstract description 32
- 238000003064 k means clustering Methods 0.000 claims abstract description 9
- 238000012216 screening Methods 0.000 claims description 62
- 238000013528 artificial neural network Methods 0.000 claims description 50
- 238000012549 training Methods 0.000 claims description 47
- 230000006870 function Effects 0.000 claims description 16
- 238000005070 sampling Methods 0.000 claims description 15
- 238000004422 calculation algorithm Methods 0.000 claims description 8
- 238000001914 filtration Methods 0.000 claims description 7
- 235000009413 Ratibida columnifera Nutrition 0.000 claims description 3
- 241000510442 Ratibida peduncularis Species 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 238000010248 power generation Methods 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 abstract description 3
- 238000010219 correlation analysis Methods 0.000 abstract description 2
- 238000010801 machine learning Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000007621 cluster analysis Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000000053 physical method Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012731 temporal analysis Methods 0.000 description 1
- 238000000700 time series analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G06Q50/40—
Abstract
The invention discloses an intelligent prediction method for extracting wind speed characteristics along a high-speed railway, which utilizes different characteristics of wind speed to perform K-means clustering, selects an optimal clustering number K, divides a wind speed model sample into K classes, respectively establishes 100 wind speed prediction characteristic preselection models aiming at the K clustering classes, screens the wind speed prediction characteristic preselection models by utilizing a mathematical analysis and correlation analysis method, establishes K wind speed prediction characteristic model groups and K wind speed prediction normalization models, and finally analyzes the correlation between a predicted wind speed vector and a wind speed time sequence data set and restores the real wind speed.
Description
Technical Field
The invention relates to an intelligent prediction method for extracting wind speed characteristics along a high-speed railway.
Background
In recent years, wind speed prediction has rapidly become one of the popular research fields at home and abroad. The wind speed prediction method has greater and greater functions and wide application in the aspect of safety along the high-speed railway in the strong wind environment. The wind speed prediction method is applied to railway line safety in a strong wind environment, more early warning processing time can be provided for a train by accurately predicting the future wind speed in real time, a safe driving scheme is formed, and driving safety is guaranteed.
The wind speed is influenced by factors such as season, temperature, altitude and the like, and has the characteristics of strong randomness and complex nonlinearity. At present, the wind speed prediction method mainly comprises a learning method, a physical method and a statistical method. Machine learning methods have been developed rapidly in recent years, so it is more desirable to predict complex nonlinear wind speeds using machine learning methods, and common models include artificial neural networks, kalman filtering, time series analysis, and the like.
In order to solve the problem of non-stationary nonlinearity in wind speed prediction, effectively extract wind speed characteristics and predict wind speed with high precision, an intelligent prediction method for extracting wind speed characteristics along a high-speed railway is urgently needed, and prediction of high precision, strong adaptability and high robustness of the wind speed is realized.
Disclosure of Invention
The invention provides an intelligent prediction method for extracting wind speed characteristics along a high-speed railway, which aims to analyze the wind speed characteristics under the strong wind environment along the high-speed railway through a clustering algorithm and a machine learning algorithm, establish various neural network models and accurately and effectively predict the wind speed along the high-speed railway by extracting the wind speed characteristics.
An intelligent prediction method for extracting wind speed characteristics along a high-speed railway comprises the following steps:
step 1: acquiring an original wind speed time series data set through a wind measuring station;
setting a wind measuring station at a specified position along a high-speed railway, and acquiring an original wind speed time sequence data set through the wind measuring station;
the original wind speed time-series data set consists of n original wind speed data; wherein Δ t is the wind speed sampling time interval; n is the sampling times, and n is at least more than 500;
step 2: establishing a wind speed model sample A;
the wind speed model sample A comprises a wind speed model training sample AaWind speed model screening sample Ab；
The specific process of establishing the wind speed model sample A is as follows:
step A1: performing Kalman filtering on the original wind speed time series data set to obtain a filtered wind speed time series data set and a noise time series data set;
step A2: wind speeds at any continuous 4 moments in the wind speed time sequence data set are used as a group of wind speed vectors to obtain a wind speed model sample A containing n-3 groups of wind speed vectors;
wind speed values at the same moment may occur in different wind speed vectors;
step A3: according to the sampling time sequence, selecting a wind speed vector of front 3/4 from the wind speed model sample A to form a wind speed model training sample AaAnd the residual wind speed vectors form a wind speed model screening sampleThis Ab；
And step 3: clustering each group of wind speed vectors in the wind speed model sample A by using a K-means clustering method, and selecting an optimal clustering number K to obtain K clustering center time sequences;
and 4, step 4: according to the wind speed model sample A which is already gathered into K cluster categories, the wind speed model training sample A isaWind speed model training subsamples (A) divided into K corresponding cluster categoriesa,1,Aa,2,…,Aa,KAnd (4) screening a sample A of the wind speed modelbScreening subsamples { A) into K corresponding wind speed modelsb,1,Ab,2,…,Ab,K}；
And 5: aiming at K wind speed model training subsamples, 100 wind speed prediction characteristic preselection models based on an Elman neural network are established;
the output value of the wind speed prediction characteristic preselection model based on the Elman neural network is named as a wind speed characteristic preselection value;
step 6: screening models, namely screening 100 wind speed prediction characteristic preselection models of K types based on the Elman neural network to 10 wind speed prediction characteristic preselection models respectively;
and 7: 10 wind speed prediction characteristic pre-selection models of each clustering category form a wind speed prediction characteristic model group to obtain K wind speed prediction characteristic model groups;
the wind speed prediction characteristic model group comprises 10 wind speed prediction characteristic preselection models, and a vector with the length of 10 is formed by output data of each wind speed prediction characteristic model group and is named as a wind speed characteristic vector;
and 8: aiming at K wind speed prediction characteristic model groups, K wind speed prediction normalization models based on wavelet neural networks are established;
the output value of the wind speed prediction normalization model is set as a wind speed normalization value;
and step 9: predicting the wind speed in real time;
acquiring wind speed data at the current moment and in the near term in real time, wherein the data sampling time interval is consistent with the wind speed sampling time interval delta t of the original wind speed time sequence data set used in the step 1, and the specific process of predicting the wind speed in real time is as follows:
step B1: setting a wind speed prediction target time T, acquiring original wind speed values of at least 200 moments closest to the target time T through a wind measuring station to form an original wind speed time sequence, and entering step B2;
step B2: performing Kalman filtering on the original wind speed time sequence to obtain a filtered wind speed time sequence, and entering the step B3;
step B3: calculating Euclidean distances between wind speed values at 4 moments closest to the target time T in the wind speed time series and the K clustering center time series obtained in the step (3), selecting a wind speed prediction characteristic model group corresponding to a clustering category according to a minimum distance principle, inputting the wind speed values at 3 moments closest to the target time T in the wind speed time series into the wind speed prediction characteristic model group as input data, outputting to obtain a wind speed characteristic vector with the length of 10, and entering the step B4;
step B4: inputting the wind speed characteristic vector serving as input data into a wind speed prediction normalization model corresponding to the cluster category, outputting a wind speed normalization value, and entering the step B5;
step B5: judging whether the time of the wind speed normalization value output in the step B4 is greater than the target time T, if so, entering a step B7, otherwise, entering a step B6;
step B6: inserting the wind speed normalization value output in the step B4 into the first position of the wind speed time sequence, updating the wind speed time sequence, and returning to the step B3;
step B7: and (3) calculating a Pearson correlation coefficient between a vector consisting of wind speed values at 100 moments closest to the target time T in the wind speed time series and a vector consisting of wind speed values at 100 moments arbitrarily continuous in the wind speed time series data set obtained in the step (2), selecting a vector consisting of wind speed values at a group of 100 moments in the wind speed time series data set with the highest Pearson correlation coefficient, adding the filter noise at the moment corresponding to the noise time series data set in the step (2) to the 100 moments closest to the target time T in the wind speed time series, namely restoring the truth of the predicted wind speed, and finishing the wind speed prediction at the target moment.
Further, the specific process of clustering each group of wind speed vectors in the wind speed model sample A by using the K-means clustering method, selecting the optimal clustering number K and obtaining the K clustering center time sequences is as follows:
step 2.1: setting a K-means clustering number K ', K' ═ 1;
step 2.2: clustering each group of wind speed vectors in the wind speed model sample A by using Euclidean distance as a similarity measurement function to obtain K 'clustering categories and K' clustering center time sequences;
the calculation formula of the Euclidean distance is
Wherein X is (X)1,x2,x3,x4),Y＝(y1,y2,y3,y4) Respectively two different wind speed vectors, d (X, Y) is the Euclidean distance between the wind speed vectors X and Y;
step 2.3: calculating the error square sum SSE (K') of the K-means cluster;
wherein K' is the number of clusters, CiIs the ith cluster class, p is CiA set of wind velocity vectors, m, in a classiIs CiCluster centers of the classes;
step 2.4: judging whether the value of K 'is equal to 20, if so, entering a step 2.5, otherwise, enabling K' +1, and then returning to the step 2.2;
step 2.5: calculating each α (K ') and β (K ') value when K ' ═ 2 to 19, respectively, and then proceeding to step 2.6;
α(K′)＝arctan(SSE(K′)-SSE(K′+1)),α∈(0°,90°),K′∈[2,19]
β(K′)＝arctan(SSE(K′)-SSE(K′-1)),β∈(90°,180°),K′∈[2,19]；
step 2.6: each value of α (K ') + β (K') when K '═ 2 to 19 is calculated, respectively, and the value of K' at which α (K ') + β (K') is the maximum is selected as the optimal clustering number K, resulting in K clustering center time series.
Further, the process of training the subsamples aiming at the K wind speed models and establishing 100 wind speed prediction characteristic preselection models based on the Elman neural network comprises the following steps: and sequentially taking the wind speed value at the first 3 moments of each group of wind speed vectors in the K wind speed model training subsamples as input data, taking the wind speed value at the 4 th moment as output data, training an Elman neural network, and randomly setting 100 Elman neural network parameters with different values for the Elman neural network obtained by training each wind speed model training subsample to obtain 100 wind speed prediction characteristic preselection models based on the Elman neural network, wherein the total number of the 100 Elman neural network-based wind speed prediction characteristic preselection models is 100.
Further, the 100 x K wind speed prediction feature preselection models based on the Elman neural network set the number of input layer nodes to be 3, the number of hidden layer nodes to be in a set range of [5,10], the number of output layer nodes to be 1, the number of receiving layer nodes to be 7, the hidden layer functions to be Sigmoid functions, the network training to be performed by the BP algorithm, the maximum iteration number to be 800, the training learning rate to be in a set range of [0.001,0.5], and the threshold to be 0.005.
Further, the screening model respectively screens 100 wind speed prediction characteristic preselection models of the K types based on the Elman neural network to 10 specific processes as follows:
step 5.1: screening subsamples { A) of K wind speed models in sequenceb,1,Ab,2,…,Ab,KWind speed values at the first 3 moments of each group of wind speed vectors in the wind power generation system are used as input data and are respectively input into 100 wind speed prediction characteristic preselection models corresponding to the clustering categories;
step 5.2: preliminarily screening a wind speed prediction characteristic preselection model;
sequentially calculating a one-dimensional vector formed by the wind speed characteristic preselected values output by each wind speed prediction characteristic preselected model according to the time sequence and a wind speed model screening sample AbThe mean square error of a one-dimensional vector formed by the wind speed values at the corresponding moments is eliminated, and a wind speed prediction characteristic preselection model with the mean square error larger than the mean square error is eliminated;
step 5.3: establishing a screening vector;
respectively establishing K screening vectors with the length of 3 by using the wind speed values of 3 moments closest to the current moment in the K clustering center time sequences;
step 5.4: updating the screening vector;
respectively taking the wind speed data of 3 moments before the K screening vectors as input data, inputting the input data into a wind speed prediction characteristic preselection model corresponding to the clustering category, inserting the output wind speed characteristic preselection value of 4 th moment into a first position of the screening vectors, sequentially delaying the positions of the rest data, and repeating the steps until the lengths of the K screening vectors reach 200;
step 5.5: screening the wind speed prediction characteristic preselection model again;
respectively calculating the Pearson correlation coefficient between every two screening vectors obtained by the wind speed prediction feature preselection model of each clustering class to obtain K Pearson correlation coefficient matrixes, respectively calculating the accumulated value of the Pearson correlation coefficient between each screening vector and other screening vectors, and gradually eliminating the wind speed prediction feature preselection model corresponding to the maximum Pearson correlation coefficient accumulated value until 10 wind speed prediction feature preselection models of each clustering class are remained.
Further, the specific process of establishing K wind speed prediction normalization models based on the wavelet neural network for the K wind speed prediction feature model groups is as follows:
step 6.1: sequentially inputting wind speed values of the first 3 moments of each group of wind speed vectors in the wind speed model sample A as input data into the wind speed prediction characteristic model groups of the corresponding clustering categories, and outputting to obtain wind speed characteristic vectors;
step 6.2: and (3) sequentially taking the wind speed characteristic vector obtained in the step (6.1) as input data, taking a corresponding wind speed value at the 4 th moment in the wind speed model sample A as output data, training a wavelet neural network, and obtaining K wind speed prediction normalization models based on the wavelet neural network for K wind speed prediction characteristic model groups.
Furthermore, the wind speed prediction normalization model based on the wavelet neural network is set to have the number of input layer nodes of 10, the number of hidden layer wavelet elements of 20 and the number of output layer nodes of 1 during training; the hidden layer wavelet neuron adopts a Mexican Hat wavelet function, the output layer node adopts a Sigmoid function, the maximum iteration number in the training process is set to be 500, the training learning rate is 0.04, and the threshold value is 0.002.
Advantageous effects
The invention provides an intelligent prediction method for extracting wind speed characteristics along a high-speed railway, which is used for carrying out cluster analysis on wind speed data by a clustering algorithm and a machine learning algorithm according to complex, nonlinear and non-steady wind speeds under a strong wind environment along the high-speed railway, selecting a clustering number K with the best clustering effect, dividing wind speed model samples into K classes, establishing 100 wind speed prediction characteristic preselection models for the wind speed model samples of each clustering cluster, respectively screening 10 model groups with good effects by a mathematical analysis method to establish K wind speed prediction characteristic model groups to extract wind speed characteristics, then establishing K wind speed prediction normalization models, normalizing the characteristic output values of the wind speed prediction characteristic model groups, and finally restoring the authenticity of the wind speed by correlation analysis;
the method utilizes a clustering algorithm to cluster wind speed data, and a plurality of different neural network models are independently constructed for each cluster, so that wind speed characteristics are effectively extracted; in the wind speed prediction process, clustering analysis is carried out according to different characteristic types of wind speeds, a wind speed prediction characteristic model group is established, finally, wind speed characteristic output is normalized, non-stationary change of the wind speed can be effectively avoided, and therefore a wind speed characteristic value is found.
Drawings
FIG. 1 is a schematic flow diagram of the process of the present invention.
Detailed Description
The invention will be further described with reference to the accompanying drawings and examples.
As shown in FIG. 1, the intelligent prediction method for extracting the wind speed characteristics along the high-speed railway comprises the following steps:
step 1: acquiring an original wind speed time series data set through a wind measuring station;
arranging an anemometer station at a designated position along a high-speed railway, and acquiring an original wind speed time series data set w (t) ═ w (t ═ Δ t), w (t ═ 2 ═ Δ t), …, w (t ═ n ═ Δ t) through the anemometer station;
the original wind speed time-series data set consists of n original wind speed data; wherein Δ t is the wind speed sampling time interval; n is the sampling times, and n is at least more than 500;
step 2: establishing a wind speed model sample A;
the wind speed model sample A comprises a wind speed model training sample AaWind speed model screening sample Ab；
The specific process of establishing the wind speed model sample A is as follows:
step A1: performing kalman filtering on the original wind speed time-series data set to obtain a filtered wind speed time-series data set w ' (t) ═ Δ t, w ' (t ═ 2 × Δ t), …, w ' (t ═ n × Δ t) ] and a noise time-series data set no (t) ═ no (t ═ Δ t), no (t ═ 2 × Δ t), …, no (t ═ n × Δ t) ];
step A2: wind speeds at any continuous 4 moments in the wind speed time sequence data set are used as a group of wind speed vectors to obtain a wind speed model sample A containing n-3 groups of wind speed vectors;
the wind speed value at the same moment may occur in different wind speed vectors, and the wind speed model sample a is represented as:
each row in the wind speed model sample A represents a group of wind speed vectors, and n-3 groups of wind speed vectors are total;
step A3: according to the sampling time sequence, selecting from the wind speed model sample A before
Wind speed model formed by combining wind speed vectorsTraining sample AaAnd the residual wind speed vectors form a wind speed model screening sample Ab；
And step 3: clustering each group of wind speed vectors in the wind speed model sample A by using a K-means clustering method, selecting an optimal clustering number K, and obtaining K clustering center time sequences, wherein the specific process is as follows:
step 2.1: setting a K-means clustering number K ', K' ═ 1;
step 2.2: clustering each group of wind speed vectors in the wind speed model sample A by using Euclidean distance as a similarity measurement function to obtain K 'clustering categories and K' clustering center time sequences;
the calculation formula of the Euclidean distance is
Wherein X is (X)1,x2,x3,x4),Y＝(y1,y2,y3,y4) Respectively two different wind speed vectors, d (X, Y) is the Euclidean distance between the wind speed vectors X and Y;
step 2.3: calculating the error square sum SSE (K') of the K-means cluster;
wherein K' is the number of clusters, CiIs the ith cluster class, p is CiA set of wind velocity vectors, m, in a classiIs CiCluster centers of the classes;
step 2.4: judging whether the value of K 'is equal to 20, if so, entering a step 2.5, otherwise, enabling K' +1, and then returning to the step 2.2;
step 2.5: calculating each α (K ') and β (K ') value when K ' ═ 2 to 19, respectively, and then proceeding to step 2.6;
α(K′)＝arctan(SSE(K′)-SSE(K′+1)),α∈(0°,90°),K′∈[2,19]
β(K′)＝arctan(SSE(K′)-SSE(K′-1)),β∈(90°,180°),K′∈[2,19]；
step 2.6: respectively calculating each value of alpha (K ') + beta (K') when K 'is 2-19, and selecting the value of K' when the value of alpha (K ') + beta (K') is maximum as an optimal clustering number K to obtain K clustering center time sequences;
and 4, step 4: according to the wind speed model sample A which is already gathered into K cluster categories, the wind speed model training sample A isaWind speed model training subsamples (A) divided into K corresponding cluster categoriesa,1,Aa,2,…,Aa,KAnd (4) screening a sample A of the wind speed modelbScreening subsamples { A) into K corresponding wind speed modelsb,1,Ab,2,…,Ab,K}；
And 5: aiming at K wind speed model training subsamples, 100 wind speed prediction characteristic preselection models based on an Elman neural network are established;
the output value of the wind speed prediction characteristic preselection model based on the Elman neural network is named as a wind speed characteristic preselection value;
the process of training the subsamples aiming at the K wind speed models and establishing 100 wind speed prediction characteristic preselection models based on the Elman neural network comprises the following steps: sequentially taking the wind speed value of 3 moments before each group of wind speed vectors in K wind speed model training subsamples as input data, taking the wind speed value of 4 moments as output data, training an Elman neural network, respectively and randomly setting 100 Elman neural network parameters with different values for the Elman neural network obtained by training each wind speed model training subsample, obtaining 100 wind speed prediction characteristic preselection models based on the Elman neural network, and expressing 100 Elman neural network-based wind speed prediction characteristic preselection models in total as follows:
wherein M _ Elman represents 100 x K wind speed prediction characteristic pre-selection models based on the Elman neural network, and each row represents 100 wind speed prediction characteristic pre-selection models of the same cluster type, and the total number of the rows is K;
the 100 x K wind speed prediction characteristic preselection models based on the Elman neural network are all set with the number of input layer nodes being 3, the number of hidden layer nodes being set in a range of [5,10], the number of output layer nodes being 1, the number of carrying layer nodes being 7, the hidden layer functions adopting Sigmoid functions, the network training adopting BP algorithm, the maximum iteration number being 800, the training learning rate being set in a range of [0.001,0.5], and the threshold being 0.005;
step 6: screening the models, namely screening 100 wind speed prediction characteristic preselection models of K types based on the Elman neural network to 10 models respectively, wherein the specific process is as follows:
step 5.1: screening subsamples { A) of K wind speed models in sequenceb,1,Ab,2,…,Ab,KWind speed values at the first 3 moments of each group of wind speed vectors in the wind power generation system are used as input data and are respectively input into 100 wind speed prediction characteristic preselection models corresponding to the clustering categories;
step 5.2: preliminarily screening a wind speed prediction characteristic preselection model;
sequentially calculating a one-dimensional vector formed by the wind speed characteristic preselected values output by each wind speed prediction characteristic preselected model according to the time sequence and a wind speed model screening sample AbThe mean square error of a one-dimensional vector formed by the wind speed values at the corresponding moments is eliminated, and a wind speed prediction characteristic preselection model with the mean square error larger than the mean square error is eliminated;
step 5.3: establishing a screening vector;
respectively establishing K screening vectors with the length of 3 by using the wind speed values of 3 moments closest to the current moment in the K clustering center time sequences;
step 5.4: updating the screening vector;
respectively taking the wind speed data of 3 moments before the K screening vectors as input data, inputting the input data into a wind speed prediction characteristic preselection model corresponding to the clustering category, inserting the output wind speed characteristic preselection value of 4 th moment into a first position of the screening vectors, sequentially delaying the positions of the rest data, and repeating the steps until the lengths of the K screening vectors reach 200;
step 5.5: screening the wind speed prediction characteristic preselection model again;
calculating each cluster separatelyObtaining the Pearson correlation coefficient between every two screening vectors obtained by the wind speed prediction characteristic preselection model of the category to obtain K Pearson correlation coefficient matrixes P1,P2,…,PKI th Pearson correlation coefficient matrix PiExpressed as:
wherein, PiA Pearson correlation coefficient matrix between every two screening vectors obtained by the wind speed prediction characteristic preselection model representing the ith clustering class,
representing the Pearson correlation coefficient between the screening vectors obtained by the 100 th model and the 2 nd model in the wind speed prediction feature preselection model of the ith cluster category;
respectively calculating the accumulated value of the Pearson correlation coefficient between each screening vector and other screening vectors, namely respectively calculating K Pearson correlation coefficient matrixes P1,P2,…,PKIs summed, e.g., the ith Pearson correlation coefficient matrix PiThe accumulated value of the Pearson correlation coefficients of the ith wind speed prediction characteristic preselection model is
Gradually eliminating the wind speed prediction characteristic preselection model corresponding to the maximum accumulated value of the Pearson correlation coefficients until 10 wind speed prediction characteristic preselection models of each cluster category remain, and expressing as follows:
wherein, M _ Elman represents 10 screened wind speed prediction characteristic preselection models based on the Elman neural network, each line represents 10 wind speed prediction characteristic preselection models of the same cluster type, and the total number of the lines is K;
and 7: 10 wind speed prediction characteristic pre-selection models of each clustering category form a wind speed prediction characteristic model group to obtain K wind speed prediction characteristic model groups;
the wind speed prediction characteristic model group comprises 10 wind speed prediction characteristic preselection models, each wind speed prediction characteristic model group outputs data to form a vector with the length of 10, the vector is named as a wind speed characteristic vector, and the ith wind speed prediction characteristic model group is expressed as [ m _ elman ]i,1,m_elmani,1,…,m_elmani,10]；
And 8: aiming at K wind speed prediction characteristic model groups, K wind speed prediction normalization models based on wavelet neural networks are established;
the output value of the wind speed prediction normalization model is set as a wind speed normalization value;
the specific process of establishing K wind speed prediction normalization models based on the wavelet neural network aiming at the K wind speed prediction characteristic model groups is as follows:
step 6.1: sequentially inputting wind speed values of the first 3 moments of each group of wind speed vectors in the wind speed model sample A as input data into the wind speed prediction characteristic model groups of the corresponding clustering categories, and outputting to obtain wind speed characteristic vectors;
step 6.2: sequentially taking the wind speed characteristic vector obtained in the step 6.1 as input data, taking a corresponding wind speed value at the 4 th moment in the wind speed model sample A as output data, training a wavelet neural network, and obtaining K wind speed prediction normalization models based on the wavelet neural network for K wind speed prediction characteristic model groups;
the wind speed prediction normalization model based on the wavelet neural network is characterized in that the number of input layer nodes is 10, the number of hidden layer wavelet elements is 20, and the number of output layer nodes is 1 during training; the hidden layer wavelet neuron adopts a Mexican Hat wavelet function, the output layer node adopts a Sigmoid function, the maximum iteration number in the training process is set to be 500, the training learning rate is 0.04, and the threshold value is 0.002;
and step 9: predicting the wind speed in real time;
acquiring wind speed data at the current moment and in the near term in real time, wherein the data sampling time interval is consistent with the wind speed sampling time interval delta t of the original wind speed time sequence data set used in the step 1, and the specific process of predicting the wind speed in real time is as follows:
step B1: setting a wind speed prediction target time T, acquiring original wind speed values of at least 200 moments closest to the target time T through a wind measuring station to form an original wind speed time sequence, and entering step B2;
step B2: performing Kalman filtering on the original wind speed time sequence to obtain a filtered wind speed time sequence, and entering the step B3;
step B3: calculating Euclidean distances between wind speed values at 4 moments closest to the target time T in the wind speed time series and the K clustering center time series obtained in the step (3), selecting a wind speed prediction characteristic model group corresponding to a clustering category according to a minimum distance principle, inputting the wind speed values at 3 moments closest to the target time T in the wind speed time series into the wind speed prediction characteristic model group as input data, outputting to obtain a wind speed characteristic vector with the length of 10, and entering the step B4;
step B4: inputting the wind speed characteristic vector serving as input data into a wind speed prediction normalization model corresponding to the cluster category, outputting a wind speed normalization value, and entering the step B5;
step B5: judging whether the time of the wind speed normalization value output in the step B4 is greater than the target time T, if so, entering a step B7, otherwise, entering a step B6;
step B6: inserting the wind speed normalization value output in the step B4 into the first position of the wind speed time sequence, updating the wind speed time sequence, and returning to the step B3;
step B7: a vector consisting of wind speed values at 100 moments closest to the target time T in the wind speed time series and the wind speed time series data set w '(T) obtained in step 2 are calculated as [ w' (T is Δ T) ], and w '(T ═ 2 × Δ T), …, and w' (T ═ n × Δ T) ] for the pearson correlation coefficient between vectors composed of wind speed values at any 100 consecutive times, selecting a vector composed of wind speed values at a set of 100 times in the wind speed time-series data set with the highest pearson correlation coefficient, and adding the filtered noise at the corresponding time in the noise time-series data set no (T) (no (T ═ Δ T), no (T ═ 2 × Δ T), …, and no (T ═ n × Δ T) in step 2 to the 100 times at which the wind speed time-series is closest to the target time T, that is, restoring the predicted wind speed, thereby completing the prediction of the wind speed at the target time.
The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and substitutions can be made without departing from the technical principle of the present invention, and these modifications and substitutions should also be regarded as the protection scope of the present invention.
Claims (7)
1. An intelligent prediction method for extracting wind speed characteristics along a high-speed railway is characterized by comprising the following steps:
step 1: acquiring an original wind speed time series data set through a wind measuring station;
setting a wind measuring station at a specified position along a high-speed railway, and acquiring an original wind speed time sequence data set through the wind measuring station;
the original wind speed time-series data set consists of n original wind speed data; wherein Δ t is the wind speed sampling time interval; n is the sampling times, and n is at least more than 500;
step 2: establishing a wind speed model sample A;
the wind speed model sample A comprises a wind speed model training sample AaWind speed model screening sample Ab；
The specific process of establishing the wind speed model sample A is as follows:
step A1: performing Kalman filtering on the original wind speed time series data set to obtain a filtered wind speed time series data set and a noise time series data set;
step A2: wind speeds at any continuous 4 moments in the wind speed time sequence data set are used as a group of wind speed vectors to obtain a wind speed model sample A containing n-3 groups of wind speed vectors;
step A3: according to the sampling time sequence, selecting a wind speed vector of front 3/4 from the wind speed model sample A to form a wind speed model training sample AaAnd the residual wind speed vectors form a wind speed model screening sample Ab；
And step 3: clustering each group of wind speed vectors in the wind speed model sample A by using a K-means clustering method, and selecting an optimal clustering number K to obtain K clustering center time sequences;
and 4, step 4: according to the wind speed model sample A which is already gathered into K cluster categories, the wind speed model training sample A isaWind speed model training subsamples (A) divided into K corresponding cluster categoriesa，1，Aa，2，…，Aa，KAnd (4) screening a sample A of the wind speed modelbScreening subsamples { A) into K corresponding wind speed modelsb，1，Ab，2，…，Ab，K}；
And 5: aiming at K wind speed model training subsamples, 100 wind speed prediction characteristic preselection models based on an Elman neural network are established;
the output value of the wind speed prediction characteristic preselection model based on the Elman neural network is named as a wind speed characteristic preselection value;
step 6: screening models, namely screening 100 wind speed prediction characteristic preselection models of K types based on the Elman neural network to 10 wind speed prediction characteristic preselection models respectively;
and 7: 10 wind speed prediction characteristic pre-selection models of each clustering category form a wind speed prediction characteristic model group to obtain K wind speed prediction characteristic model groups;
the wind speed prediction characteristic model group comprises 10 wind speed prediction characteristic preselection models, and a vector with the length of 10 is formed by output data of each wind speed prediction characteristic model group and is named as a wind speed characteristic vector;
and 8: aiming at K wind speed prediction characteristic model groups, K wind speed prediction normalization models based on wavelet neural networks are established;
the output value of the wind speed prediction normalization model is set as a wind speed normalization value;
and step 9: predicting the wind speed in real time;
acquiring wind speed data at the current moment and in the near term in real time, wherein the data sampling time interval is consistent with the wind speed sampling time interval delta t of the original wind speed time sequence data set used in the step 1, and the specific process of predicting the wind speed in real time is as follows:
step B1: setting a wind speed prediction target time T, acquiring original wind speed values of at least 200 moments closest to the target time T through a wind measuring station to form an original wind speed time sequence, and entering step B2;
step B2: performing Kalman filtering on the original wind speed time sequence to obtain a filtered wind speed time sequence, and entering the step B3;
step B3: calculating Euclidean distances between wind speed values at 4 moments closest to the target time T in the wind speed time series and the K clustering center time series obtained in the step (3), selecting a wind speed prediction characteristic model group corresponding to a clustering category according to a minimum distance principle, inputting the wind speed values at 3 moments closest to the target time T in the wind speed time series into the wind speed prediction characteristic model group as input data, outputting to obtain a wind speed characteristic vector with the length of 10, and entering the step B4;
step B4: inputting the wind speed characteristic vector serving as input data into a wind speed prediction normalization model corresponding to the cluster category, outputting a wind speed normalization value, and entering the step B5;
step B5: judging whether the time of the wind speed normalization value output in the step B4 is greater than the target time T, if so, entering a step B7, otherwise, entering a step B6;
step B6: inserting the wind speed normalization value output in the step B4 into the first position of the wind speed time sequence, updating the wind speed time sequence, and returning to the step B3;
step B7: and (3) calculating a Pearson correlation coefficient between a vector consisting of wind speed values at 100 moments closest to the target time T in the wind speed time series and a vector consisting of wind speed values at 100 moments arbitrarily continuous in the wind speed time series data set obtained in the step (2), selecting a vector consisting of wind speed values at a group of 100 moments in the wind speed time series data set with the highest Pearson correlation coefficient, adding the filter noise at the moment corresponding to the noise time series data set in the step (2) to the 100 moments closest to the target time T in the wind speed time series, namely restoring the truth of the predicted wind speed, and finishing the wind speed prediction at the target moment.
2. The method according to claim 1, wherein the K-means clustering method is used for clustering each group of wind speed vectors in the wind speed model sample a, and an optimal clustering number K is selected, so that a specific process for obtaining K clustering center time sequences is as follows:
step 2.1: setting a K-means clustering number K ', K' ═ 1;
step 2.2: clustering each group of wind speed vectors in the wind speed model sample A by using Euclidean distance as a similarity measurement function to obtain K 'clustering categories and K' clustering center time sequences;
the calculation formula of the Euclidean distance is
Wherein X is (X)1，x2，x3，x4)，Y＝(y1，y2，y3，y4) Respectively two different wind speed vectors, d (X, Y) is the Euclidean distance between the wind speed vectors X and Y;
step 2.3: calculating the error square sum SSE (K') of the K-means cluster;
wherein K' is the number of clusters, CiIs the ith cluster class, p is CiA set of wind velocity vectors, m, in a classiIs CiCluster centers of the classes;
step 2.4: judging whether the value of K 'is equal to 20, if so, entering a step 2.5, otherwise, enabling K' +1, and then returning to the step 2.2;
step 2.5: calculating each α (K ') and β (K ') value when K ' ═ 2 to 19, respectively, and then proceeding to step 2.6;
α(K′)＝arctan(SSE(K′)-SSE(K′+1))，α∈(0°，90°)，K′∈[2，19]
β(K′)＝arctan(SSE(K′)-SSE(K′-1))，β∈(90°，180°)，K′∈[2，19]；
step 2.6: each value of α (K ') + β (K') when K '═ 2 to 19 is calculated, respectively, and the value of K' at which α (K ') + β (K') is the maximum is selected as the optimal clustering number K, resulting in K clustering center time series.
3. The method as claimed in claim 1, wherein the training of the subsamples for the K wind speed models, and the process of establishing 100 each of the K Elman neural network-based wind speed prediction feature preselection models is as follows: and sequentially taking the wind speed value at the first 3 moments of each group of wind speed vectors in the K wind speed model training subsamples as input data, taking the wind speed value at the 4 th moment as output data, training an Elman neural network, and randomly setting 100 Elman neural network parameters with different values for the Elman neural network obtained by training each wind speed model training subsample to obtain 100 wind speed prediction characteristic preselection models based on the Elman neural network, wherein the total number of the 100 Elman neural network-based wind speed prediction characteristic preselection models is 100.
4. The method according to claim 3, wherein the 100 x K wind speed prediction feature preselection models based on the Elman neural network are trained by setting the number of input layer nodes to be 3, the number of hidden layer nodes to be in a range of [5,10], the number of output layer nodes to be 1, the number of receiving layer nodes to be 7, the hidden layer functions to be Sigmoid functions, the network training to be performed by BP algorithm, the maximum iteration number to be 800, the training learning rate to be in a range of [0.001,0.5], and the threshold value to be 0.005.
5. The method as claimed in claim 1, wherein the specific process of screening the K kinds of wind speed prediction characteristic preselection models based on the Elman neural network from 100 to 10 respectively is as follows:
step 5.1: screening subsamples { A) of K wind speed models in sequenceb，1，Ab，2，…，Ab，KWind speed values at the first 3 moments of each group of wind speed vectors in the wind power generation system are used as input data and are respectively input into 100 wind speed prediction characteristic preselection models corresponding to the clustering categories;
step 5.2: preliminarily screening a wind speed prediction characteristic preselection model;
sequentially calculating a one-dimensional vector formed by the wind speed characteristic preselected values output by each wind speed prediction characteristic preselected model according to the time sequence and a wind speed model screening sample AbThe mean square error of a one-dimensional vector formed by the wind speed values at the corresponding moments is eliminated, and a wind speed prediction characteristic preselection model with the mean square error larger than the mean square error is eliminated;
step 5.3: establishing a screening vector;
respectively establishing K screening vectors with the length of 3 by using the wind speed values of 3 moments closest to the current moment in the K clustering center time sequences;
step 5.4: updating the screening vector;
respectively taking the wind speed data of 3 moments before the K screening vectors as input data, inputting the input data into a wind speed prediction characteristic preselection model corresponding to the clustering category, inserting the output wind speed characteristic preselection value of 4 th moment into a first position of the screening vectors, sequentially delaying the positions of the rest data, and repeating the steps until the lengths of the K screening vectors reach 200;
step 5.5: screening the wind speed prediction characteristic preselection model again;
respectively calculating the Pearson correlation coefficient between every two screening vectors obtained by the wind speed prediction feature preselection model of each clustering class to obtain K Pearson correlation coefficient matrixes, respectively calculating the accumulated value of the Pearson correlation coefficient between each screening vector and other screening vectors, and gradually eliminating the wind speed prediction feature preselection model corresponding to the maximum Pearson correlation coefficient accumulated value until 10 wind speed prediction feature preselection models of each clustering class are remained.
6. The method according to claim 1, wherein the specific process of establishing K wavelet neural network-based wind speed prediction normalization models for the K wind speed prediction feature model groups is as follows:
step 6.1: sequentially inputting wind speed values of the first 3 moments of each group of wind speed vectors in the wind speed model sample A as input data into the wind speed prediction characteristic model groups of the corresponding clustering categories, and outputting to obtain wind speed characteristic vectors;
step 6.2: and (3) sequentially taking the wind speed characteristic vector obtained in the step (6.1) as input data, taking a corresponding wind speed value at the 4 th moment in the wind speed model sample A as output data, training a wavelet neural network, and obtaining K wind speed prediction normalization models based on the wavelet neural network for K wind speed prediction characteristic model groups.
7. The method according to claim 6, wherein the wind speed prediction normalization model based on the wavelet neural network sets the number of input layer nodes to be 10, the number of hidden layer wavelet elements to be 20 and the number of output layer nodes to be 1 during training; the hidden layer wavelet neuron adopts a Mexican Hat wavelet function, the output layer node adopts a Sigmoid function, the maximum iteration number in the training process is set to be 500, the training learning rate is 0.04, and the threshold value is 0.002.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811643482.7A CN109615027B (en) | 2018-12-29 | 2018-12-29 | Intelligent prediction method for extracting wind speed characteristics along high-speed railway |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811643482.7A CN109615027B (en) | 2018-12-29 | 2018-12-29 | Intelligent prediction method for extracting wind speed characteristics along high-speed railway |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109615027A CN109615027A (en) | 2019-04-12 |
| CN109615027B true CN109615027B (en) | 2020-12-08 |
Family
ID=66016400
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811643482.7A Active CN109615027B (en) | 2018-12-29 | 2018-12-29 | Intelligent prediction method for extracting wind speed characteristics along high-speed railway |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109615027B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110650058B (en) * | 2019-10-08 | 2022-03-04 | 河南省云安大数据安全防护产业技术研究院有限公司 | Network traffic analysis method, device, storage medium and equipment |
| CN111291943B (en) * | 2020-03-09 | 2022-03-08 | 西南交通大学 | High-speed rail strong wind early warning method based on frequent sequence |
| CN112381420A (en) * | 2020-11-17 | 2021-02-19 | 平安普惠企业管理有限公司 | Risk early warning method, device and equipment based on desert wind speed and storage medium |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070223841A1 (en) * | 2006-03-23 | 2007-09-27 | Weinzapfel Robert A | Methods of characterizing the probability of wind impact |
| CN102063641A (en) * | 2010-10-14 | 2011-05-18 | 北京大学 | Method for forecasting wind speed of high speed railway line |
| CN107895202B (en) * | 2017-10-24 | 2022-03-04 | 天津大学 | Short-term wind speed prediction method based on multi-view wind speed mode mining |
| CN109063907B (en) * | 2018-07-27 | 2020-06-05 | 中南大学 | Intelligent traversal large-step-length prediction method for maximum wind speed along high-speed railway |
-
2018
- 2018-12-29 CN CN201811643482.7A patent/CN109615027B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN109615027A (en) | 2019-04-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110263846B (en) | Fault diagnosis method based on fault data deep mining and learning | |
| CN108388927B (en) | Small sample polarization SAR terrain classification method based on deep convolution twin network | |
| CN108197648B (en) | Hydroelectric generating set fault diagnosis method and system based on LSTM deep learning model | |
| CN110287983B (en) | Single-classifier anomaly detection method based on maximum correlation entropy deep neural network | |
| CN109615027B (en) | Intelligent prediction method for extracting wind speed characteristics along high-speed railway | |
| CN110232319B (en) | Ship behavior identification method based on deep learning | |
| CN108985252B (en) | Improved image classification method of pulse depth neural network | |
| CN109726802B (en) | Machine learning prediction method for wind speed in railway and wind farm environment | |
| CN109145516B (en) | Analog circuit fault identification method based on improved extreme learning machine | |
| CN111352977B (en) | Time sequence data monitoring method based on self-attention bidirectional long-short term memory network | |
| CN111292525A (en) | Traffic flow prediction method based on neural network | |
| CN107832789B (en) | Feature weighting K nearest neighbor fault diagnosis method based on average influence value data transformation | |
| CN111275165A (en) | Network intrusion detection method based on improved convolutional neural network | |
| CN111815806A (en) | Method for preprocessing flight parameter data based on wild value elimination and feature extraction | |
| CN112257741A (en) | Method for detecting generative anti-false picture based on complex neural network | |
| CN110110785B (en) | Express logistics process state detection and classification method | |
| CN113109782B (en) | Classification method directly applied to radar radiation source amplitude sequence | |
| CN112966595B (en) | Sensor network data abnormity judgment method based on graph neural network | |
| CN110633741B (en) | Time sequence classification method based on improved impulse neural network | |
| CN110020680B (en) | PMU data classification method based on random matrix theory and fuzzy C-means clustering algorithm | |
| CN112422546A (en) | Network anomaly detection method based on variable neighborhood algorithm and fuzzy clustering | |
| CN111797979A (en) | Vibration transmission system based on LSTM model | |
| CN115357862B (en) | Positioning method in long and narrow space | |
| Sobrevilla et al. | Daily weather forecast in Tiwi, Albay, Philippines using artificial neural network with missing values imputation | |
| CN108021873A (en) | A kind of EEG signals epilepsy sorting technique and system for clustering asymmetric mutual information |
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 |