CN112183847A - QRNN improved Stacking algorithm-based train running wind speed probability prediction method - Google Patents

QRNN improved Stacking algorithm-based train running wind speed probability prediction method Download PDF

Info

Publication number
CN112183847A
CN112183847A CN202011021235.0A CN202011021235A CN112183847A CN 112183847 A CN112183847 A CN 112183847A CN 202011021235 A CN202011021235 A CN 202011021235A CN 112183847 A CN112183847 A CN 112183847A
Authority
CN
China
Prior art keywords
test
wind speed
train
training set
time point
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202011021235.0A
Other languages
Chinese (zh)
Other versions
CN112183847B (en
Inventor
何耀耀
肖经凌
王云
张婉莹
朱建华
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hefei University of Technology
Original Assignee
Hefei University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hefei University of Technology filed Critical Hefei University of Technology
Priority to CN202011021235.0A priority Critical patent/CN112183847B/en
Publication of CN112183847A publication Critical patent/CN112183847A/en
Application granted granted Critical
Publication of CN112183847B publication Critical patent/CN112183847B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a QRNN improved Stacking algorithm-based train running wind speed probability prediction method, which comprises the following steps: 1: collecting wind speed related data of a wind speed observation station in a dangerous area along a railway, and dividing a training set and a test set; 2: selecting an autoregressive AR model, and respectively training the models by using a Support Vector Machine (SVM) and a Radial Basis Function (RBF) neural network as base learners; 3: forming a new training set by the predictions of the three models on the training set, and forming a new test set by the predictions on the test set; 4: the meta-learner selects a quantile regression neural network (QNN) method, obtains prediction results under different quantile points in a new data set, and combines kernel density estimation to obtain probabilistic prediction of wind speed. The invention can give full play to the learning performance of each learner and simultaneously obtain the probability density of the wind speed predicted value, thereby providing more reliable early warning information under the condition of fully considering the random time sequence and the high nonlinearity of the wind speed.

Description

QRNN improved Stacking algorithm-based train running wind speed probability prediction method
Technical Field
The invention belongs to the field of wind speed prediction, and particularly relates to a train running wind speed probability prediction method based on a QRNN improved Stacking algorithm.
Background
The safety running of the train can be seriously influenced by severe weather such as strong wind, and the early warning of the railway strong wind disaster accident is more severe due to the random time sequence and the high nonlinearity of the wind speed. The strong wind disaster can also induce the occurrence of more disasters such as collapse, debris flow and the like, and the wind speed prediction of some dangerous areas of the railway is of great importance in order to guarantee the life and property safety of people and reduce the negative social influence.
The existing wind speed prediction methods are roughly divided into statistical methods and machine learning methods, and the statistical methods mainly comprise autoregressive models, moving average models, autoregressive integral moving average models and the like; the machine learning method mainly comprises a support vector machine model, an artificial neural network model and the like. However, these single methods have obvious defects in terms of wind speed with strong randomness, cannot accurately capture wind speed characteristics in different environments, have large errors, and have no method superior enough to adapt to variable wind speeds, so that more and more integrated methods are built accordingly. Although the integration method improves the result accuracy, only the point prediction result of the wind speed is obtained, and the influence of various uncertain factors on the fluctuation of the wind speed is difficult to reflect. In the technical background of the integrated method, a special prediction method aiming at the wind speed probability is urgently needed to be provided, and hidden risks in a prediction result are reduced. Therefore, the practical problems that high accuracy of wind speed prediction is realized, prediction errors are reduced, and probability density prediction is provided for safe running of trains in a strong wind environment are needed to be solved urgently.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, and provides a train running wind speed probability prediction method based on a QNN improved Stacking algorithm, so that the learning performance of each learner can be fully exerted, and the probability density of a wind speed prediction value is obtained, so that more reliable early warning information is provided under the condition of fully considering the random time sequence and high nonlinearity of the wind speed.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention discloses a QNN improved Stacking algorithm-based train running wind speed probability prediction method which is characterized by comprising the following steps of:
step 1, collecting wind speed related data of a wind speed observation station in a dangerous area along a railway, converting all the data into dimensionless pure numerical values, and dividing a training set and a test set;
step 1.1, collecting historical wind speed and lambda influence factors of wind speed observation wind stations in dangerous areas along a railway and carrying out normalization processing to obtain a normalized data set;
step 1.2, delaying the historical wind speed and the lambda influence factors in the normalized data set by an m period to obtain m (lambda +1) explanation variables, and recording the m explanation variables as X ═ Xi}i＝m+1,m+2,…,nAnd one response variable is noted as Y ═ Yi}i＝m+1,m+2,…,nWherein x isiAn interpretation variable data set representing the ith time point and having: x is the number ofi＝{xi,j}j＝1,2,…,m(λ+1)，xi,jFor the j-th interpretation variable data at the i-th time point, yiThe data of the response variable of the ith time point are obtained, n is the total number of samples, and m is the lag order;
dividing a data set (X, Y) consisting of explanatory variables and response variables into a training set (X)train,Ytrain) And test set (X)test,Ytest) Wherein X istrainIs an explanatory variable in the training set, YtrainIs a response variable in the training set, XtestIs an explanatory variable in the test set, YtestIs a response variable in the test set;
step 2, training the training set (X)train,Ytrain) And testingCollection (X)test,Ytest) Respectively as the input of the autoregressive model, the support vector machine and the RBF neural network model, thereby correspondingly obtaining the autoregressive model in the training set (X)train,Ytrain) Is recorded as output of
Figure BDA0002700695620000021
Autoregressive model in test set (X)test,Ytest) Is recorded as output of
Figure BDA0002700695620000022
Support vector machine model in training set (X)train,Ytrain) Is recorded as output of
Figure BDA0002700695620000023
In the test set (X)test,Ytest) Is recorded as output of
Figure BDA0002700695620000024
And RBF neural network model in training set (X)train,Ytrain) Is recorded as output of
Figure BDA0002700695620000025
In the test set (X)test,Ytest) Is recorded as output of
Figure BDA0002700695620000026
Wherein the content of the first and second substances,
Figure BDA0002700695620000027
is an autoregressive model in the training set (X)train,Ytrain) The wind speed output at the upper i-th time point,
Figure BDA0002700695620000028
is an autoregressive model in the test set (X)test,Ytest) Wind speed output at the last ith moment point;
Figure BDA0002700695620000029
is that the support vector machine model is in the training set (X)train,Ytrain) The wind speed output at the upper i-th time point,
Figure BDA00027006956200000210
is that the support vector machine model is in the test set (X)test,Ytest) Wind speed output at the last ith moment point;
Figure BDA00027006956200000211
is a RBF neural network model in a training set (X)train,Ytrain) The wind speed output at the upper i-th time point,
Figure BDA00027006956200000212
is the RBF neural network model in the test set (X)test,Ytest) Wind speed output at the last ith moment point;
step 3, order
Figure BDA00027006956200000213
The new training set is denoted as (U)i,Ytrain) The new test set is represented as (Z)i,Ytest)；
Step 4, using a QRNN improved Stacking algorithm, obtaining prediction results under different quantiles in a new training set and a central test set, and obtaining probabilistic prediction of wind speed by combining nuclear density estimation so as to make early warning judgment;
step 4.1, utilize the new training set (U)i,Ytrain) Training the QRNN model to obtain a trained QRNN model; the new test set (Z)i,Ytest) Inputting the QRNN model after training, thereby obtaining a new test set (Z)i,Ytest) The result of the conditional quantile prediction of (1) is recorded as
Figure BDA0002700695620000034
Wherein the content of the first and second substances,
Figure BDA0002700695620000035
representing a response variable YtestIn interpreting variable ZiAt the r-th quantile ofrThe conditional quantile of (c);
step 4.2 order intermediate variables
Figure BDA0002700695620000036
And as the input of a KDE method for estimating the kernel density, a probability density function f (q) of any q point is obtained by using the formula (1):
Figure BDA0002700695620000031
in the formula (1), d is the bandwidth, E (·) is an Epanechnikov kernel function with the average value of 0 and the integral of 1;
4.3, discretizing the probability density function of each time point to obtain H predicted values, and performing reverse normalization on the H predicted values to finally obtain the wind speed predicted value of each time point, wherein the wind speed predicted value of the ith time point is represented as
Figure BDA0002700695620000032
And the corresponding probability is noted as fi,1,fi,2,…,fi,h,…,fi,H，fi,hAn h-th predicted value representing an i-th time point;
step 4.4, setting the train speed limit threshold value to T levels, and further setting T +1 threshold value set for making a great dealt}t＝1,2,…,T+1The judgment basis of the early warning of the running wind speed of the train is
Figure BDA0002700695620000033
tAndt+1a lower threshold and an upper threshold representing the t-th interval, respectively; calculating the sum of the probabilities of all the predicted values of the corresponding time points in the corresponding threshold interval according to all the predicted values of each time point, taking the sum as the probability of the occurrence of the graded deceleration or parking events of the corresponding time points, and recording the probability as pi (n:)t)。
Compared with the prior art, the invention has the beneficial effects that:
1. the method adopts a QRNN improved Stacking method, selects an autoregressive AR model, selects a Support Vector Machine (SVM) and a Radial Basis Function (RBF) neural network as base learners to respectively train the models, selects a Quantile Regression Neural Network (QRNN) method as a meta-learner, obtains prediction results under different quantile points, combines kernel density estimation to obtain probabilistic prediction of the wind speed, and overcomes the problems that a single method in the prior art has larger error and can only obtain point prediction, thereby leading early warning information to be more reliable.
2. The three base learners are respectively an autoregressive model, a support vector machine model and an RBF neural network model, the autoregressive model is a linear model, and the support vector machine and the RBF neural network are non-linear models. Due to the instability and randomness of the wind speed, the nonlinear model and the linear model have different advantages when applied to wind speed prediction, so that the linear model and the nonlinear model are combined, the complementation of model performance is realized, and the prediction precision is improved.
The Stacking generally adopts a linear method as an element learner, a nonlinear method QRNN is used as the element learner, the three base learners are stacked and learned, the Stacking algorithm is improved, and the learning performance of a linear model and a nonlinear model is fully exerted. The QNN method combines the advantages of a neural network and quantile regression, not only can reveal the integral condition distribution of response variables, but also can process complex nonlinear problems, and has strong superiority in prediction, thereby providing a more accurate wind speed prediction value.
4. The probability density prediction of the wind speed is obtained by the nuclear density method, so that more effective information can be provided for the wind speed prediction, and a new thought and method are developed for the wind speed prediction of safe train running under the strong wind condition.
Drawings
FIG. 1 is an overall flow diagram of the process of the present invention;
FIG. 2 is a schematic structural diagram of the Stacking system of the present invention.
Detailed Description
In this embodiment, a method for predicting the probability of the train running wind speed based on the QRNN improved Stacking algorithm is performed according to the following steps, as shown in fig. 1:
step 1, collecting wind speed related data of a wind speed observation station in a dangerous area along a railway, converting all the data into dimensionless pure numerical values, and dividing a training set and a test set;
step 1.1, collecting historical wind speed and lambda influence factors of wind speed observation wind stations in dangerous areas along a railway and carrying out normalization processing to obtain a normalized data set;
step 1.2, delaying the historical wind speed and the lambda influence factors in the normalized data set by an m period to obtain m (lambda +1) explanation variables, and recording the m explanation variables as X ═ Xi}i＝m+1,m+2,…,nAnd one response variable is noted as Y ═ Yi}i＝m+1,m+2,…,nWherein x isiAn interpretation variable data set representing the ith time point and having: x is the number ofi＝{xi,j}j＝1,2,…,m(λ+1)，xi,jFor the j-th interpretation variable data at the i-th time point, yiThe data of the response variable of the ith time point are obtained, n is the total number of samples, and m is the lag order;
dividing a data set (X, Y) consisting of explanatory variables and response variables into a training set (X)train,Ytrain) And test set (X)test,Ytest) Wherein X istrainIs an explanatory variable in the training set, YtrainIs a response variable in the training set, XtestIs an explanatory variable in the test set, YtestIs a response variable in the test set;
step 2, training set (X)train,Ytrain) And test set (X)test,Ytest) Respectively as the input of an autoregressive model, a support vector machine and an RBF neural network model, wherein the autoregressive model is obtained by the following formula (1):
yi＝α+φ1yi-1+φ2yi-2+…+φmyi-m+wi (1)
in the formula (1), α ═ μ (1-. phi.)1-…-φm) And α is a constant vector, where φ1,φ2,…,φmIs the autoregressive coefficient, phimThe expression is the autoregressive coefficient before the m-th independent variable, μ isSequence yi}i＝m+1,m+2,…,nThe mean value of (a); { wi}i＝m+1,m+2,…,nIs a white noise sequence, where wiRepresents the ith white noise;
the gaussian support vector machine model is obtained by equation (2):
Figure BDA0002700695620000051
in the formula (2), b is an offset term, αi，
Figure BDA0002700695620000052
Is the ith Lagrange multiplier and satisfies alphai≥0,
Figure BDA0002700695620000053
exp(-||xt,j-xi,j||2/2σ2) Is a Gaussian kernel expression where | is 2-norm, where two sample points x are representedt,j、xi,jThe inner product between, gamma is an adjustable kernel parameter, and sigma is a kernel width coefficient;
the RBF neural network model is obtained by equation (3):
Figure BDA0002700695620000054
wk,iis to initialize the weight from the hidden layer to the output layer, ckIs center, σ2Is the variance;
thereby obtaining a corresponding autoregressive model in the training set (X)train,Ytrain) Is recorded as output of
Figure BDA0002700695620000055
Autoregressive model in test set (X)test,Ytest) Is recorded as output of
Figure BDA0002700695620000056
Support vector machine model in training set (X)train,Ytrain) Is recorded as output of
Figure BDA0002700695620000057
In the test set (X)test,Ytest) Is recorded as output of
Figure BDA0002700695620000058
And RBF neural network model in training set (X)train,Ytrain) Is recorded as output of
Figure BDA0002700695620000059
In the test set (X)test,Ytest) Is recorded as output of
Figure BDA00027006956200000510
Wherein the content of the first and second substances,
Figure BDA00027006956200000511
is an autoregressive model in the training set (X)train,Ytrain) The wind speed output at the upper i-th time point,
Figure BDA00027006956200000512
is an autoregressive model in the test set (X)test,Ytest) Wind speed output at the last ith moment point;
Figure BDA00027006956200000513
is that the support vector machine model is in the training set (X)train,Ytrain) The wind speed output at the upper i-th time point,
Figure BDA00027006956200000514
is that the support vector machine model is in the test set (X)test,Ytest) Wind speed output at the last ith moment point;
Figure BDA00027006956200000517
is a RBF neural network model in a training set (X)train,Ytrain) The wind speed output at the upper i-th time point,
Figure BDA00027006956200000515
is the RBF neural network model in the test set (X)test,Ytest) Wind speed output at the last ith moment point;
step 3, order
Figure BDA00027006956200000516
The new training set is denoted as (U)i,Ytrain) The new test set is represented as (Z)i,Ytest)；
Step 4, a QRNN improved Stacking algorithm is used, and a Stacking structure schematic diagram in the invention is shown in FIG. 2; obtaining prediction results under different quantiles in a new training set and a new testing set, and obtaining probabilistic prediction of wind speed by combining nuclear density estimation so as to make early warning judgment;
step 4.1, using the formula (4) in the new training set (U)i,Ytrain) Training the QRNN model to obtain a trained QRNN model;
Figure BDA0002700695620000061
v is obtained according to a gradient optimization algorithmiAnd WiOf the optimum parameter, Vi＝{vi,k,s1, | K ═ 1,2, ·, K; s is a weight matrix connecting the input layer and the hidden layer at the ith time point, where v is the weight matrixi,k,sRepresenting the connection weight between the kth input layer node and the s-th hidden layer node of the ith time point. Wi＝{wi,sS1, 2, S represents the connection weight vector between the hidden layer and the output layer at the ith time point, wi,sThe connection weight of the s-th hidden layer node and the output layer node representing the ith time point. Tau isrDenotes the r-th quantile andre (0,1), R1, 2, R represents the number of quantile points, g1(. is an activation function of the hidden layer, g2(. is the activation function of the output layer;
new test set (Z)i,Ytest) After input trainingQRNN model, so as to obtain the new test set (Z)i,Ytest) The result of the conditional quantile prediction of (1) is recorded as
Figure BDA0002700695620000062
Wherein the content of the first and second substances,
Figure BDA0002700695620000063
representing a response variable YtestIn interpreting variable ZiAt the r-th quantile ofrThe conditional quantile of (c);
step 4.2 order intermediate variables
Figure BDA0002700695620000066
And as the input of the kernel density estimation KDE method, obtaining the probability density function f (q) of any q point by using the formula (5):
Figure BDA0002700695620000064
in the formula (1), d is the bandwidth, E (·) is an Epanechnikov kernel function with the average value of 0 and the integral of 1;
4.3, discretizing the probability density function of each time point to obtain H predicted values, and performing reverse normalization on the H predicted values to finally obtain the wind speed predicted value of each time point, wherein the wind speed predicted value of the ith time point is represented as
Figure BDA0002700695620000065
And the corresponding probability is noted as fi,1,fi,2,…,fi,h,…,fi,H，fi,hAn h-th predicted value representing an i-th time point;
step 4.4, the train speed limit threshold is set to T levels, and T +1 threshold sets for a great dealt}t＝1,2,…,T+1The judgment basis of the early warning of the running wind speed of the train is
Figure BDA0002700695620000071
tAndt+1a lower threshold and an upper threshold representing the t-th interval, respectively; calculating the sum of the probabilities of all the predicted values of the corresponding time points in the corresponding threshold interval according to all the predicted values of each time point, taking the sum as the probability of the occurrence of the graded deceleration or parking events of the corresponding time points, and recording the probability as pi (n:)t). And after calculating the probability of the occurrence of the graded deceleration or stopping event at the time point, reminding the train dispatching center to make corresponding preparations for deceleration and stopping, thereby providing guarantee for the safe running of the train.

Claims (1)

1. A QRNN improved Stacking algorithm-based train running wind speed probability prediction method is characterized by comprising the following steps:
step 1, collecting wind speed related data of a wind speed observation station in a dangerous area along a railway, converting all the data into dimensionless pure numerical values, and dividing a training set and a test set;
step 1.1, collecting historical wind speed and lambda influence factors of wind speed observation wind stations in dangerous areas along a railway and carrying out normalization processing to obtain a normalized data set;
step 1.2, delaying the historical wind speed and the lambda influence factors in the normalized data set by an m period to obtain m (lambda +1) explanation variables, and recording the m explanation variables as X ═ Xi}i＝m+1,m+2,…,nAnd one response variable is noted as Y ═ Yi}i＝m+1,m+2,…,nWherein x isiAn interpretation variable data set representing the ith time point and having: x is the number ofi＝{xi,j}j＝1,2,…,m(λ+1)，xi,jFor the j-th interpretation variable data at the i-th time point, yiThe data of the response variable of the ith time point are obtained, n is the total number of samples, and m is the lag order;
dividing a data set (X, Y) consisting of explanatory variables and response variables into a training set (X)train,Ytrain) And test set (X)test,Ytest) Wherein X istrainIs an explanatory variable in the training set, YtrainIs a response variable in the training set, XtestIs an explanatory variable in the test set, YtestIs test-focusedA response variable;
step 2, training the training set (X)train,Ytrain) And test set (X)test,Ytest) Respectively as the input of the autoregressive model, the support vector machine and the RBF neural network model, thereby correspondingly obtaining the autoregressive model in the training set (X)train,Ytrain) Is recorded as output of
Figure FDA0002700695610000011
Autoregressive model in test set (X)test,Ytest) Is recorded as output of
Figure FDA0002700695610000012
Support vector machine model in training set (X)train,Ytrain) Is recorded as output of
Figure FDA0002700695610000013
In the test set (X)test,Ytest) Is recorded as output of
Figure FDA0002700695610000014
And RBF neural network model in training set (X)train,Ytrain) Is recorded as output of
Figure FDA0002700695610000015
In the test set (X)test,Ytest) Is recorded as output of
Figure FDA0002700695610000016
Wherein the content of the first and second substances,
Figure FDA0002700695610000017
is an autoregressive model in the training set (X)train,Ytrain) The wind speed output at the upper i-th time point,
Figure FDA0002700695610000018
is an autoregressive model in the test set (X)test,Ytest) Wind speed output at the last ith moment point;
Figure FDA0002700695610000019
is that the support vector machine model is in the training set (X)train,Ytrain) The wind speed output at the upper i-th time point,
Figure FDA00027006956100000110
is that the support vector machine model is in the test set (X)test,Ytest) Wind speed output at the last ith moment point;
Figure FDA00027006956100000111
is a RBF neural network model in a training set (X)train,Ytrain) The wind speed output at the upper i-th time point,
Figure FDA00027006956100000112
is the RBF neural network model in the test set (X)test,Ytest) Wind speed output at the last ith moment point;
step 3, order
Figure FDA0002700695610000021
The new training set is denoted as (U)i,Ytrain) The new test set is represented as (Z)i,Ytest)；
Step 4, using a QRNN improved Stacking algorithm, obtaining prediction results under different quantiles in a new training set and a central test set, and obtaining probabilistic prediction of wind speed by combining nuclear density estimation so as to make early warning judgment;
step 4.1, utilize the new training set (U)i,Ytrain) Training the QRNN model to obtain a trained QRNN model; the new test set (Z)i,Ytest) Inputting the QRNN model after training, thereby obtaining a new test set (Z)i,Ytest) The result of the conditional quantile prediction of (1) is expressed as { QYtest(τr|Zi)}r＝1,2,…,R(ii) a Wherein Q isYtest(τr|Zi) Representing a response variable YtestIn interpreting variable ZiAt the r-th quantile ofrThe conditional quantile of (c);
step 4.2 order the intermediate variable Q (τ)r)＝QYtest(τr|Zi) R1, 2.. times, R, and used as an input to the kernel density estimation KDE method, a probability density function f (q) of an arbitrary q point is obtained using equation (1):
Figure FDA0002700695610000022
in the formula (1), d is the bandwidth, E (·) is an Epanechnikov kernel function with the average value of 0 and the integral of 1;
4.3, discretizing the probability density function of each time point to obtain H predicted values, and performing reverse normalization on the H predicted values to finally obtain the wind speed predicted value of each time point, wherein the wind speed predicted value of the ith time point is represented as
Figure FDA0002700695610000023
And the corresponding probability is noted as fi,1,fi,2,…,fi,h,…,fi,H，fi,hAn h-th predicted value representing an i-th time point;
step 4.4, setting the train speed limit threshold value to T levels, and further setting T +1 threshold value set for making a great dealt}t＝1,2,…,T+1The judgment basis of the early warning of the running wind speed of the train is
Figure FDA0002700695610000024
tAndt+1a lower threshold and an upper threshold representing the t-th interval, respectively; calculating the sum of the probabilities of all the predicted values of the corresponding time points in the corresponding threshold interval according to all the predicted values of each time point, taking the sum as the probability of the occurrence of the graded deceleration or parking events of the corresponding time points, and recording the probability as pi (n:)t)。
CN202011021235.0A 2020-09-25 2020-09-25 QRNN improved Stacking algorithm-based train running wind speed probability prediction method Active CN112183847B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011021235.0A CN112183847B (en) 2020-09-25 2020-09-25 QRNN improved Stacking algorithm-based train running wind speed probability prediction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011021235.0A CN112183847B (en) 2020-09-25 2020-09-25 QRNN improved Stacking algorithm-based train running wind speed probability prediction method

Publications (2)

Publication Number Publication Date
CN112183847A true CN112183847A (en) 2021-01-05
CN112183847B CN112183847B (en) 2022-09-20

Family

ID=73943404

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011021235.0A Active CN112183847B (en) 2020-09-25 2020-09-25 QRNN improved Stacking algorithm-based train running wind speed probability prediction method

Country Status (1)

Country Link
CN (1) CN112183847B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114420290A (en) * 2022-01-14 2022-04-29 中国科学院地理科学与资源研究所 Oncomelania snail density prediction method and system based on Relieff-SVM

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102063641A (en) * 2010-10-14 2011-05-18 北京大学 Method for forecasting wind speed of high speed railway line
KR101326678B1 (en) * 2012-10-10 2013-11-08 (주)동양화학 On-chip transformer balun and the mathod for manufaturing wire structure used the same
CN103400210A (en) * 2013-08-13 2013-11-20 广西电网公司电力科学研究院 Short-term wind-speed combination forecasting method
CN110066895A (en) * 2019-04-10 2019-07-30 东北大学 A kind of blast-melted quality section prediction technique based on Stacking
CN110685857A (en) * 2019-10-16 2020-01-14 湘潭大学 Mountain wind turbine generator behavior prediction model based on ensemble learning
US20200111174A1 (en) * 2018-10-04 2020-04-09 Yishen Wang Probabilistic Load Forecasting via Point Forecast Feature Integration
CN111612262A (en) * 2020-01-15 2020-09-01 长沙理工大学 Wind power probability prediction method based on quantile regression
CN111695666A (en) * 2020-05-26 2020-09-22 河海大学 Wind power ultra-short term conditional probability prediction method based on deep learning

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102063641A (en) * 2010-10-14 2011-05-18 北京大学 Method for forecasting wind speed of high speed railway line
KR101326678B1 (en) * 2012-10-10 2013-11-08 (주)동양화학 On-chip transformer balun and the mathod for manufaturing wire structure used the same
CN103400210A (en) * 2013-08-13 2013-11-20 广西电网公司电力科学研究院 Short-term wind-speed combination forecasting method
US20200111174A1 (en) * 2018-10-04 2020-04-09 Yishen Wang Probabilistic Load Forecasting via Point Forecast Feature Integration
CN110066895A (en) * 2019-04-10 2019-07-30 东北大学 A kind of blast-melted quality section prediction technique based on Stacking
CN110685857A (en) * 2019-10-16 2020-01-14 湘潭大学 Mountain wind turbine generator behavior prediction model based on ensemble learning
CN111612262A (en) * 2020-01-15 2020-09-01 长沙理工大学 Wind power probability prediction method based on quantile regression
CN111695666A (en) * 2020-05-26 2020-09-22 河海大学 Wind power ultra-short term conditional probability prediction method based on deep learning

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
MADASTHU SANTHOSH等: "Short-term wind speed forecasting approach using Ensemble Empirical Mode Decomposition and Deep Boltzmann Machine", 《SUSTAINABLE ENERGY, GRIDS AND NETWORKS》 *
李永刚等: "基于Stacking融合的短期风速预测组合模型", 《电网技术》 *
杨荣新等: "基于Stacking模型融合的光伏发电功率预测", 《计算机***应用》 *
胡梦月等: "基于改进AdaBoost.RT和KELM的风功率预测方法研究", 《电网技术》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114420290A (en) * 2022-01-14 2022-04-29 中国科学院地理科学与资源研究所 Oncomelania snail density prediction method and system based on Relieff-SVM
CN114420290B (en) * 2022-01-14 2022-11-04 中国科学院地理科学与资源研究所 Method and system for predicting oncomelania density based on Relieff-SVM

Also Published As

Publication number Publication date
CN112183847B (en) 2022-09-20

Similar Documents

Publication Publication Date Title
WO2021196367A1 (en) Performance-related ground motion risk analysis method employing three-layer dataset-based neural network
CN113191215B (en) Rolling bearing fault diagnosis method integrating attention mechanism and twin network structure
Zhang et al. Tropical cyclone intensity classification and estimation using infrared satellite images with deep learning
CN110059716B (en) Construction of CNN-LSTM-SVM network model and MOOC conjugate prediction method
CN106022954B (en) Multiple BP neural network load prediction method based on grey correlation degree
CN114509685A (en) Method and system for constructing thermal runaway prediction model of lithium ion battery
CN103714382B (en) A kind of multi-index comprehensive evaluation method for reliability of urban rail train security detection sensor network
Wu et al. A hybrid support vector regression approach for rainfall forecasting using particle swarm optimization and projection pursuit technology
CN104008164A (en) Generalized regression neural network based short-term diarrhea multi-step prediction method
CN111538076A (en) Earthquake magnitude rapid estimation method based on deep learning feature fusion
CN112465199B (en) Airspace situation assessment system
CN110210660B (en) Ultra-short-term wind speed prediction method
CN105574489A (en) Layered stack based violent group behavior detection method
CN112966853A (en) Urban road network short-term traffic flow prediction method based on space-time residual error mixed model
CN110533007A (en) A kind of vehicle-mounted strain of bridge influences the identification of line feature intelligent and extracting method
CN116822920B (en) Flow prediction method based on cyclic neural network
Wang et al. A remaining useful life prediction model based on hybrid long-short sequences for engines
CN112836937A (en) Flood disaster loss evaluation method based on entropy weight and BP neural network technology
CN112183847B (en) QRNN improved Stacking algorithm-based train running wind speed probability prediction method
CN112149884A (en) Academic early warning monitoring method for large-scale students
Xu et al. Global attention mechanism based deep learning for remaining useful life prediction of aero-engine
Wang et al. CNN-BiLSTM-attention model in forecasting wave height over South-East China Seas
CN112203311B (en) Network element abnormity diagnosis method, device, equipment and computer storage medium
CN116842358A (en) Soft measurement modeling method based on multi-scale convolution and self-adaptive feature fusion
CN115391523A (en) Wind power plant multi-source heterogeneous data processing method and device

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