CN113326975B - Ultrahigh prediction method for track irregularity based on random oscillation sequence gray model - Google Patents
Ultrahigh prediction method for track irregularity based on random oscillation sequence gray model Download PDFInfo
- Publication number
- CN113326975B CN113326975B CN202110493436.9A CN202110493436A CN113326975B CN 113326975 B CN113326975 B CN 113326975B CN 202110493436 A CN202110493436 A CN 202110493436A CN 113326975 B CN113326975 B CN 113326975B
- Authority
- CN
- China
- Prior art keywords
- height
- sequence
- prediction
- residual error
- neural network
- 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
- 230000010355 oscillation Effects 0.000 title claims abstract description 43
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000013528 artificial neural network Methods 0.000 claims abstract description 89
- 241001206881 Myrmeleon inconspicuus Species 0.000 claims abstract description 39
- 238000012937 correction Methods 0.000 claims abstract description 33
- 238000012545 processing Methods 0.000 claims abstract description 27
- 238000007781 pre-processing Methods 0.000 claims abstract description 13
- 238000010606 normalization Methods 0.000 claims abstract description 4
- 238000012549 training Methods 0.000 claims description 67
- 230000009467 reduction Effects 0.000 claims description 35
- 238000012360 testing method Methods 0.000 claims description 29
- 230000009466 transformation Effects 0.000 claims description 29
- 230000006870 function Effects 0.000 claims description 28
- 230000004913 activation Effects 0.000 claims description 15
- 238000001514 detection method Methods 0.000 claims description 12
- 238000013519 translation Methods 0.000 claims description 12
- 238000005457 optimization Methods 0.000 claims description 11
- 238000011161 development Methods 0.000 claims description 8
- 238000009825 accumulation Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 230000009471 action Effects 0.000 claims description 5
- 230000001133 acceleration Effects 0.000 claims description 4
- 241000257303 Hymenoptera Species 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 230000002087 whitening effect Effects 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 2
- 241000258923 Neuroptera Species 0.000 claims 1
- 230000003534 oscillatory effect Effects 0.000 claims 1
- 230000007547 defect Effects 0.000 abstract description 3
- 241000282320 Panthera leo Species 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000005299 abrasion Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
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
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06395—Quality analysis or management
-
- 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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/40—Business processes related to the transportation industry
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- Development Economics (AREA)
- General Business, Economics & Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Educational Administration (AREA)
- Game Theory and Decision Science (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a random oscillation sequence gray model-based track irregularity ultrahigh prediction method, which comprises the following steps: a data preprocessing step: carrying out mean value processing on the detected height deviations of the left rail surface and the right rail surface to obtain an equidistant average height deviation sequence; a preliminary prediction step: carrying out random oscillation sequence gray prediction based on a gray model to obtain a preliminary prediction height deviation; and a prediction correction step: correcting the preliminary predicted height residual error based on the height residual error average value to obtain a corrected height residual error, and performing normalization processing; and (3) optimizing an Elman neural network: optimizing the initial weight and the threshold of the Elman neural network through the ant lion algorithm to further obtain an optimized Elman neural network; ultra-high prediction step: and obtaining an orbit prediction correction height residual error based on the optimized Elman neural network. The method overcomes the defect of non-ideal prediction results of the random oscillation sequence by combining the random oscillation sequence gray model and the Elman neural network, so that the ultrahigh prediction results are more accurate.
Description
Technical Field
The invention relates to the field of track detection, in particular to a random oscillation sequence gray model-based track irregularity ultrahigh prediction method.
Background
The detection of the running safety condition of the urban rail transit is an important part for guaranteeing the running of the rail, the existing method can carry out dynamic accurate detection on the parameters of the rail, but how to analyze and predict the quality of the rail from the data detected on the rail is very important for the research on the rail detection.
Most of the current researches are used for predicting the comprehensive quality TQI of the track, and from the experimental result, the prediction of the comprehensive quality TQI is more accurate by a combination method of gray prediction and a neural network;
the super-high value is different from TQI which is a random oscillation sequence, but the traditional grey prediction is only suitable for data types with exponential growth, the prediction effect on the random oscillation sequence is poor, and the variation trend of the sequence cannot be fitted, so that the error of the prediction result on the random oscillation sequence is larger.
Disclosure of Invention
In order to overcome the defects and shortcomings of the prior art, the invention provides a random oscillation sequence gray model-based track irregularity ultrahigh prediction method, which realizes the initial prediction of an ultrahigh sequence and can accurately predict the variation trend of the ultrahigh sequence and the attached ultrahigh sequence.
The invention also provides a system for predicting the track irregularity based on the random oscillation sequence gray model.
In order to achieve the first object, the invention adopts the following technical scheme:
a random oscillation sequence gray model-based track irregularity ultrahigh prediction method comprises the following steps:
a data preprocessing step: carrying out mean value processing on the detected height deviation of the left rail surface and the right rail surface to obtain an equidistant average height deviation sequence;
a preliminary prediction step: performing random oscillation sequence gray prediction on the equidistant average height deviation sequence based on a gray model to obtain a preliminary prediction height deviation;
and a prediction correction step: obtaining a preliminary predicted height residual error based on the preliminary predicted height deviation and the original data, calculating a height residual error average value, correcting the preliminary predicted height residual error based on the height residual error average value to obtain a corrected height residual error, and performing normalization processing on the preliminary predicted height residual error and the corrected height residual error;
and (3) optimizing an Elman neural network: dividing a training set and a test set based on the preliminary prediction height residual error and the correction height residual error, building an Elman neural network, training based on the preliminary prediction height residual error and the correction height residual error of the training set, and optimizing an initial weight and a threshold of the Elman neural network through an ant lion algorithm to obtain an optimized Elman neural network;
ultra-high prediction step: and inputting the preliminary prediction height residual error based on the optimized Elman neural network to obtain the orbit prediction correction height residual error.
As a preferred technical solution, the data preprocessing step specifically includes: and acquiring the height deviation of the preset interval at equal detection time intervals, and carrying out average value processing on the height deviation to obtain the average height deviation representing the interval.
As a preferred technical solution, the preliminary prediction step specifically includes the following steps:
transformation of random oscillation sequence:
detecting and detecting the detection sections at the same time to obtain an equidistant average height deviation sequence, wherein the equidistant average height deviation sequence is set as follows: x(0)={X(0)(1),X(0)(2),…,X(0)(n)};
Translating the equidistant average height deviation sequence to a full value, and setting the following parameters:
Max=max{x(0)(k)|k,k∈{1,2,…,n}};
Min=min{x(0)(k)|k,k∈{1,2,…,n}};
max represents the maximum value in the original equidistant average height deviation sequence, Min represents the minimum value in the original equidistant average height deviation sequence, T represents the maximum oscillation ratio of the original equidistant average height deviation sequence, and n is the maximum value of the sequence;
the sequence obtained by performing accelerated exponential transformation on the equidistant average height deviation sequence is a monotonically increasing height deviation sequence, namely:
XD1={x(1)d1,x(2)d1,…,x(n)d1};
wherein x (k) d1K is 1,2,3 … n, which is the k-th monotonically increasing height deviation value after the acceleration exponential transformation;
to XD1And performing geometric mean generation transformation on the sequence to obtain a geometric mean processing height deviation sequence which keeps monotonous increase but the increase amplitude is reduced, namely an ideal height deviation sequence, namely:
XD2={x(1)d2,x(2)d2,…,x(n)d2};
x(k)d2the k-th geometric mean processing height deviation value is defined as k, wherein k is 1, …, n is a positive integer;
judging the grade ratio:
if the stage ratio k (i) is in the intervalIf so, the ideal height deviation sequence is directly used for a prediction model;
if the step ratio k (i) is not satisfiedSelecting a proper translation coefficient z to carry out integral translation on the sequence so as to meet the level ratio k (i) in the intervalThe method comprises the following steps:
whereinIndicating the ideal height deviation after the ith translation,representing the ideal height deviation before the ith translation;
and establishing a gray model to solve the predicted height deviation.
As a preferred technical scheme, the establishing of the gray model for solving the predicted height deviation specifically comprises the following steps:
to sequence XD2Performing primary accumulation to obtain:
X(1)={x(1)(1),x(1)(2),…,x(1)(n)};
establishing a GM (1,1) model, wherein a whitening formal differential equation is as follows:
the discretized difference equation is:
x(k)d2+az(1)(k)=u;
wherein a is a development coefficient, and u is a gray acting amount;
let z(1)(k) Represents the average of the kth value and the previous value in the ideal altitude deviation sequence:
and (3) solving a development coefficient a and a gray action amount u by using a least square method:
(a,u)T=(BTB)-1BTY.
wherein B represents an ideal height deviation average matrix, and Y represents an ideal height deviation original data matrix;
substituting the development coefficient a and the gray action quantity u into a differential equation to solve to obtain an unreduced first prediction height deviation:
when i < n, the differential processing sequence is processedSubtracting the sequences to obtain an ideal altitude deviation accumulation sequence, and sequentially performing decrement, geometric mean transformation and accelerated exponential transformation to restore to a second predicted altitude deviation:
the first reduction, the second reduction and the third reduction are respectively reduced by decrement, geometric mean transformation and accelerated exponential transformation;
the first reduction is specifically as follows:
the second reduction is specifically as follows:
the third reduction is specifically:
when i is larger than n, the first reduction and the second reduction are performed in sequence, if X is larger than n(0)(n+1)<X(0)(n), then performing a fourth reduction:
if X is(0)(n+1)>X(0)(n), then a fifth reduction is performed:
reducing to a third predicted altitude deviation, wherein the preliminary predicted altitude deviation comprises the first predicted altitude deviation, the second predicted altitude deviation and the third predicted altitude deviation, and the first predicted altitude deviation, the second predicted altitude deviation and the third predicted altitude deviation are obtainedIndicating the ith value in the sequence of differential processing.
As a preferred technical scheme, the construction of the Elman neural network specifically comprises the following steps:
establishing a first layer as an input layer, inputting the preprocessed data into a network, and determining an input numerical value and an input node number, wherein the input numerical value is a preliminary prediction height residual error;
establishing a second layer as a hidden layer, determining the number of nodes according to an empirical formula, and selecting a nonlinear function as a first activation function according to the range of input numerical values;
establishing a third layer as a receiving layer, receiving a feedback signal from the hidden layer, wherein each hidden layer node is connected with a corresponding node, and the hidden layer state at the previous moment and the network input at the current moment are taken as the input of the hidden layer through connection memory;
establishing a fourth layer as an output layer, selecting a linear transfer function as a second activation function, and outputting a neural network calculation result;
and learning the neural network by using a reverse error propagation algorithm, and continuously updating the weight value by using a gradient descent method.
As a preferred technical solution, the empirical formula specifically includes:wherein n isiTo input the number of nodes, noM is any integer value between 1 and 10 for the number of output nodes.
As a preferred technical solution, the first activation function selects a hyperbolic tangent sigmoid function, where the hyperbolic tangent sigmoid function is expressed as:
wherein x1E represents a natural constant, which is an input variable of the hidden layer;
the second activation function is:
f(x2)=x2;
wherein x is2Representing the input variables of the output layer.
As a preferred technical scheme, the optimization of the initial weight and the threshold of the Elman neural network by the ant lion algorithm specifically comprises the following steps:
a condition initialization step:
the initial weight and the threshold value of the neural network are coded by adopting a real number coding mode, and the coding formula is as follows:
S=R×S1+S1×S1+S1×S2+S1+S2,
wherein R is the number of input nodes, S1 is the number of hidden layer nodes, and S2 is the number of output nodes;
taking the weight and threshold coding number of the Elman neural network as a population, initially setting ant lion and ant as the same population and the same population size, setting the number of each individual in the population as a random number in a preset individual number interval, and setting the weight and threshold of the neural network corresponding to each individual in the population;
a first neural network training step:
training parameter initialization: taking the preliminary predicted height residual as training input, taking the corresponding corrected height residual as a training result, and evaluating the ant position based on the network training error;
screening the elite lion: according to the updating of ant positions, giving and evaluating values, and selecting the ant lion with the highest evaluating value as the elite ant lion;
judging the training degree: inputting a test height residual error, outputting to obtain a first test predicted height residual error, and adjusting training according to whether the first test predicted height residual error reaches a first preset residual error accuracy value or not;
when the first test predicted height residual does not reach a first preset residual precision value, ant lion optimization is carried out, and the first neural network training step is repeated;
when the first test prediction height residual reaches a first preset residual precision value, executing a parameter updating step;
updating parameters: reassigning the training parameters: assigning the optimal weight and the optimal threshold value after the ant lion algorithm is optimized to the weight parameter and the threshold value parameter corresponding to the neural network, namely replacing the initial weight and the threshold value;
ant lion optimization: after the elite ant lion captures the ants, the ant lion position is updated to be the ant position, the ant position updating is influenced by the elite ant lion, and finally the optimal weight and the threshold are found;
a second neural network training step:
carrying out Elman neural network training, and updating the weight parameter and the threshold parameter based on the network training error;
and judging a training ending condition, adjusting training according to whether the second test predicted height residual reaches a second preset residual precision value, ending the second neural network training step when the second test predicted height residual reaches the second preset residual precision value, and otherwise, repeating the second neural network training step.
In order to achieve the second object, the invention adopts the following technical scheme:
a random-oscillatory-sequence-grey-model-based ultra-high prediction system of rail irregularity, comprising: the system comprises a data preprocessing module, a preliminary prediction module, a prediction correction module, a neural network module and an ultrahigh prediction module;
the data preprocessing module is used for carrying out mean value processing on the detected height deviation of the left rail surface and the right rail surface to obtain an equidistant average height deviation sequence;
the preliminary prediction module is used for carrying out random oscillation sequence gray prediction on the equidistant average height deviation sequence based on a gray model to obtain preliminary prediction height deviation;
the prediction correction module is used for obtaining a preliminary prediction height residual error based on the preliminary prediction height deviation and original data, calculating a height residual error average value, correcting the preliminary prediction height residual error based on the height residual error average value to obtain a corrected height residual error, and normalizing the preliminary prediction height residual error and the corrected height residual error;
the neural network module is used for dividing a training set and a test set based on the preliminary prediction height residual error and the correction height residual error, building an Elman neural network, training based on the preliminary prediction height residual error and the correction height residual error of the training set, and optimizing an initial weight and a threshold of the Elman neural network through an ant lion algorithm so as to obtain an optimized Elman neural network;
and the ultrahigh prediction module is used for inputting a preliminary prediction height residual error based on the optimized Elman neural network to obtain a track prediction correction height residual error.
Compared with the prior art, the invention has the following advantages and beneficial effects:
(1) the method overcomes the defect that the prediction result of the traditional grey model on the random oscillation sequence is not ideal by combining the random oscillation sequence grey model and the Elman neural network, enriches the single index prediction method on the rail irregularity, solves the error problem of the prediction result of the random oscillation sequence grey model by adopting the Elman neural network optimized by the ant lion algorithm, enables the ultrahigh prediction result to be more accurate, predicts the real ultrahigh historical detection data on the method, and shows that the method has better prediction capability and higher prediction precision by the prediction result.
(2) According to the Elman neural network optimization method, the initial weight and the threshold of the Elman neural network are optimized through the ant lion algorithm, the problem that the Elman neural network is easy to fall into a local optimal solution is solved, and the optimized Elman neural network achieves the global optimal correction effect.
Drawings
FIG. 1 is a flowchart illustrating the steps of a random oscillation gray model-based superelevation prediction method in example 1 of the present invention;
fig. 2 is a diagram of a training procedure for optimizing an Elman neural network in the ultrahigh prediction method based on the random oscillation gray model in embodiment 1 of the present invention;
fig. 3 is a schematic structural diagram of an optimized Elman neural network in embodiment 1 of the present invention.
Detailed Description
In the description of the present disclosure, it should be noted that the terms "first", "second", and "third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance. Also, the use of the terms "a," "an," or "the" and similar referents do not denote a limitation of quantity, but rather denote the presence of at least one. The word "comprising" or "comprises", and the like, means that the element or item appearing before the word, includes the element or item listed after the word and its equivalent, but does not exclude other elements or items.
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the invention.
Examples
Example 1
As shown in fig. 1, the present embodiment provides a method for predicting superelevation of track irregularity based on a random oscillation sequence gray model, which includes the following steps:
a data preprocessing step: carrying out mean value processing on the detected height deviation of the left rail surface and the right rail surface to obtain an equidistant average height deviation sequence;
a preliminary prediction step: performing random oscillation sequence gray prediction on the equidistant average height deviation sequence based on a gray model to obtain a preliminary prediction height deviation;
and a prediction correction step: obtaining a preliminary prediction height residual error based on the preliminary prediction height deviation and original data, calculating a height residual error average value, correcting the preliminary prediction height residual error based on the height residual error average value to obtain a corrected height residual error, and normalizing the preliminary prediction height residual error and the corrected height residual error;
and (3) optimizing an Elman neural network: dividing a training set and a test set based on the preliminary prediction height residual error and the correction height residual error, building an Elman neural network, training based on the preliminary prediction height residual error and the correction height residual error of the training set, and optimizing an initial weight and a threshold of the Elman neural network through an ant lion algorithm to further obtain an optimized Elman neural network;
ultra-high prediction step: and inputting the preliminary prediction height residual error based on the optimized Elman neural network to obtain a track prediction correction height residual error, wherein the height deviation conforming to the natural change trend of track abrasion is obtained by adding the preliminary prediction height to the track prediction correction height residual error.
In this embodiment, the data preprocessing step specifically includes: and acquiring the height deviation of the preset interval at equal detection time intervals, carrying out average value processing on the height deviation to obtain the average height deviation representing the interval, wherein the preset interval is 200m, and in addition, the preset interval can be adjusted according to the actual condition.
In this embodiment, the preliminary prediction step specifically includes the following steps:
transformation of random oscillation sequence:
detecting and detecting the detection sections at the same time to obtain an equidistant average height deviation sequence, wherein the equidistant average height deviation sequence is as follows: x(0)={X(0)(1),X(0)(2),…,X(0)(n) }; the equidistant average height deviation sequence belongs to a super high sequence value, and because the super high sequence value is different in positive and negative, the original super high average value sequence is translated to a full value, and the following parameters are set:
Max=max{x(0)(k)|k,k∈{1,2,…,n}};
Min=min{x(0)(k)|k,k∈{1,2,…,n}};
wherein Max represents the maximum value in the original equidistant average height deviation sequence, Min represents the minimum value in the original equidistant average height deviation sequence, T represents the maximum oscillation ratio of the original equidistant average height deviation sequence, and n is the maximum value of the sequence;
the sequence obtained by performing accelerated exponential transformation on the equidistant average height deviation sequence is a monotonically increasing height deviation sequence, namely:
XD1={x(1)d1,x(2)d1,…,x(n)d1};
wherein x (k) d1Where k is 1,2,3 … n, which is the k-th monotonically increasing height deviation value after being subjected to the acceleration exponential transformation, in this embodiment, the acceleration exponential transformation specifically is:
x(k)d1=x(0)(k)Tk-1,k=1,2,3…n;
to XD1And performing geometric mean generation transformation on the sequence to obtain a geometric mean processing height deviation sequence which keeps monotonous increase but the increase amplitude is reduced, namely an ideal height deviation sequence, namely:
XD2={x(1)d2,x(2)d2,…,x(n)d2};
x(k)d2and k is the k-th geometric mean processing height deviation value, k is 1 and …, and n is a positive integer. In this embodiment, the geometric mean transformation specifically includes:
judging the grade ratio:
if the stage ratio k (i) is in the intervalThen the ideal height deviation sequence can be directly used for a prediction model;
if not, a proper translation coefficient z needs to be selected to perform integral translation on the sequence, and further the level ratio k (i) is satisfied in the intervalThe method comprises the following steps:
whereinIndicating the ideal height deviation after the ith translation,representing the ideal height deviation before the ith translation;
establishing a gray model:
for sequence XD2Performing primary accumulation to obtain:
X(1)={x(1)(1),x(1)(2),…,x(1)(n)};
establishing a GM (1,1) model, wherein a whitening formal differential equation is as follows:
the discretized difference equation is:
x(k)d2+az(1)(k)=u;
wherein a is a development coefficient, and u is a gray acting amount;
let z(1)(k) Represents the average of the kth value and the previous value in the ideal altitude deviation sequence:
and (3) solving a development coefficient a and a gray action amount u by using a least square method:
(a,u)T=(BTB)-1BTY.
wherein B represents an ideal height deviation average matrix, and Y represents an ideal height deviation original data matrix:
substituting the parameters a and u into a differential equation to solve to obtain an unreduced first prediction height deviation:
when i < n, the sequence of differential processing is appliedSubtracting the sequences to obtain an ideal altitude deviation accumulation sequence, and sequentially performing decrement, geometric mean transformation and accelerated exponential transformation to restore to a second predicted altitude deviation:
during actual application, sequentially executing first reduction, second reduction and third reduction, wherein the first reduction, the second reduction and the third reduction respectively correspond to decrement, geometric mean transformation and accelerated exponential transformation reduction;
the first reduction is specifically:
the second reduction is specifically:
the third reduction is specifically:
when i > n, performing the first reduction and the second reduction in sequence, if X(0)(n+1)<X(0)(n), then performing a fourth reduction:
if X(0)(n+1)>X(0)(n), then a fifth reduction is performed:
reducing the height deviation to a third predicted height deviation, wherein the preliminary predicted height deviation comprises a first predicted height deviation, a second predicted height deviation and a third predicted height deviation;
as shown in fig. 2 and 3, the Elman neural network is built, and the method specifically comprises the following steps:
establishing a first layer as an input layer, inputting the preprocessed data into a network, and determining an input numerical value and an input node number, wherein the input numerical value is a preliminary prediction height residual error;
establishing a second layer as a hidden layer, determining the number of nodes according to an empirical formula, and selecting a nonlinear function as a first activation function according to the range of input numerical values;
establishing a third layer as a receiving layer, receiving a feedback signal from the hidden layer, wherein each hidden layer node is connected with a corresponding node, and the hidden layer state at the previous moment and the network input at the current moment are used as the input of the hidden layer through connection memory;
establishing a fourth layer as an output layer, selecting a linear transfer function as a second activation function, and outputting a neural network calculation result;
and learning the neural network by using a reverse error propagation algorithm, and continuously updating the weight value by a gradient descent method.
In this embodiment, the empirical formula is specifically:wherein n isiFor the number of input nodes, noM is any integer value between 1 and 10 for the number of output nodes.
In this embodiment, the first activation function is selected from a hyperbolic tangent sigmoid function, where the hyperbolic tangent sigmoid function is expressed as:
wherein x1E represents a natural constant, which is an input variable of the hidden layer;
in this embodiment, the second activation function is the following function:
f(x2)=x2;
wherein x2Representing the input variables of the output layer.
In addition, the first activation function and the second activation function may adopt other activation functions, and are not limited herein.
With reference to fig. 2 and 3, the optimization of the initial weight and the threshold of the Elman neural network by the ant lion algorithm specifically includes the following steps:
a condition initialization step:
the initial weight and the threshold value of the neural network are coded by adopting a real number coding mode, and the coding formula is as follows:
S=R×S1+S1×S1+S1×S2+S1+S2,
wherein R is the number of input nodes, S1 is the number of hidden layer nodes, and S2 is the number of output nodes;
taking the weight and threshold coding number of the Elman neural network as a population, initially setting ant lion and ant as the same population and the same population size, setting the number of each individual of the population as a random number in a preset individual number interval, and setting the weight and threshold of the neural network corresponding to each individual in the population, wherein the preset individual number interval in the embodiment adopts [ -3,3 ];
a first neural network training step:
initializing training parameters: taking the preliminary prediction height residual error as training input, taking the corresponding correction height residual error as a training result, and evaluating the ant position based on the network training error;
screening the elite lion: according to the updating of ant positions, giving and evaluating values, and selecting the ant lion with the highest evaluating value as the elite ant lion;
judging the training degree: inputting a test height residual error, outputting to obtain a first test prediction height residual error, and adjusting training according to whether the first test prediction height residual error reaches a first preset residual error precision value, wherein the test height residual error is obtained by a test set;
when the first test predicted height residual does not reach a first preset residual precision value, ant lion optimization is carried out, and the first neural network training step is repeated;
when the first test prediction height residual reaches a first preset residual precision value, executing a parameter updating step;
updating parameters: reassigning the training parameters: assigning the optimal weight and the optimal threshold value after the ant lion algorithm is optimized to the weight parameter and the threshold value parameter corresponding to the neural network, namely replacing the initial weight and the threshold value;
ant lion optimization: after the elite ant lion captures the ants, the ant lion position is updated to be the ant position, the ant position updating is influenced by the elite ant lion, and finally the optimal weight and the threshold are found;
a second neural network training step:
carrying out Elman neural network training, and updating the weight parameter and the threshold parameter based on the network training error;
and judging a training ending condition, adjusting training according to whether the second test predicted height residual reaches a second preset residual precision value, ending the second neural network training step when the second test predicted height residual reaches the second preset residual precision value, and otherwise, repeating the second neural network training step.
Example 2
The embodiment provides a system for predicting the superelevation of the rail irregularity based on a random oscillation sequence gray model, which comprises: the system comprises a data preprocessing module, a preliminary prediction module, a prediction correction module, a neural network module and an ultrahigh prediction module;
the data preprocessing module is used for carrying out mean processing on the detected height deviation of the left rail surface and the right rail surface to obtain an equidistant average height deviation sequence;
the preliminary prediction module is used for carrying out random oscillation sequence gray prediction on the equidistant average height deviation sequence based on a gray model to obtain preliminary prediction height deviation;
the prediction correction module is used for obtaining a preliminary prediction height residual error based on the preliminary prediction height deviation and original data, calculating a height residual error average value, correcting the preliminary prediction height residual error based on the height residual error average value to obtain a corrected height residual error, and normalizing the preliminary prediction height residual error and the corrected height residual error;
the neural network module is used for dividing a training set and a test set based on the preliminary prediction height residual error and the correction height residual error, building an Elman neural network, training the preliminary prediction height residual error and the correction height residual error based on the training set, and optimizing an initial weight and a threshold of the Elman neural network through an ant lion algorithm to further obtain an optimized Elman neural network;
and the ultrahigh prediction module is used for inputting the preliminary prediction height residual error based on the optimized Elman neural network to obtain the track prediction correction height residual error.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.
Claims (9)
1. A track irregularity ultrahigh prediction method based on a random oscillation sequence gray model is characterized by comprising the following steps:
a data preprocessing step: carrying out mean value processing on the detected height deviations of the left rail surface and the right rail surface to obtain an equidistant average height deviation sequence;
a preliminary prediction step: performing random oscillation sequence gray prediction on the equidistant average height deviation sequence based on a gray model to obtain a preliminary prediction height deviation;
the preliminary prediction step specifically comprises the following steps:
transformation of random oscillation sequence:
detecting and detecting the detection sections at the same time to obtain an equidistant average height deviation sequence, wherein the equidistant average height deviation sequence is set as follows: x(0)={X(0)(1),X(0)(2),…,X(0)(n)};
Translating the equidistant average height deviation sequence to a full value, and setting the following parameters:
Max=max{x(0)(k)|k,k∈{1,2,…,n}};
Min=min{x(0)(k)|k,k∈{1,2,…,n}};
max represents the maximum value in the original equidistant average height deviation sequence, Min represents the minimum value in the original equidistant average height deviation sequence, T represents the maximum oscillation ratio of the original equidistant average height deviation sequence, and n is the maximum value of the sequence;
the sequence obtained by performing accelerated exponential transformation on the equidistant average height deviation sequence is a monotonically increasing height deviation sequence, namely:
XD1={x(1)d1,x(2)d1,…,x(n)d1};
wherein x (k) d1K is 1,2,3 … n, which is the k-th monotonically increasing height deviation value after the acceleration exponential transformation;
for XD1And performing geometric mean generation transformation on the sequence to obtain a geometric mean processing height deviation sequence which keeps monotonous increase but the increase amplitude is reduced, namely an ideal height deviation sequence, namely:
XD2={x(1)d2,x(2)d2,…,x(n)d2};
x(k)d2the k-th geometric mean processing height deviation value is defined as k, wherein k is 1, …, n is a positive integer;
and a prediction correction step: obtaining a preliminary prediction height residual error based on the preliminary prediction height deviation and original data, calculating a height residual error average value, correcting the preliminary prediction height residual error based on the height residual error average value to obtain a corrected height residual error, and normalizing the preliminary prediction height residual error and the corrected height residual error;
and (3) optimizing an Elman neural network: performing normalization processing based on the preliminary prediction height residual error and the correction height residual error, dividing a training set and a test set, building an Elman neural network, performing training based on the preliminary prediction height residual error and the correction height residual error of the training set, and optimizing an initial weight and a threshold of the Elman neural network through an ant lion algorithm to obtain an optimized Elman neural network;
ultra-high prediction step: and inputting the preliminary prediction height residual error based on the optimized Elman neural network to obtain the orbit prediction correction height residual error.
2. The method for ultrahigh prediction of rail irregularity based on random oscillation sequence gray model according to claim 1, wherein the data preprocessing step is specifically: and acquiring the height deviation of the preset interval at equal detection time intervals, and carrying out average value processing on the height deviation to obtain the average height deviation representing the interval.
3. The method for ultrahigh prediction of orbit irregularity based on stochastic oscillatory sequence gray model of claim 1, wherein the preliminary prediction step comprises the following steps:
judging the grade ratio:
if the step ratio k (i) is in the intervalIf so, the ideal height deviation sequence is directly used for a prediction model;
if the step ratio k (i) is not satisfiedSelecting a proper translation coefficient z to carry out integral translation on the sequence so as to meet the level ratio k (i) in the intervalThe method comprises the following steps:
whereinIndicating the ideal height deviation after the ith translation,representing the ideal height deviation before the ith translation;
and establishing a gray model to solve the predicted height deviation.
4. The ultrahigh prediction method of track irregularity based on the random oscillation sequence gray model as claimed in claim 3, wherein the establishing of the gray model to solve the predicted height deviation specifically comprises the following steps:
to sequence XD2Performing primary accumulation to obtain:
X(1)={x(1)(1),x(1)(2),…,x(1)(n)};
establishing a GM (1,1) model, wherein a whitening form differential equation is as follows:
the discretized difference equation is:
x(k)d2+az(1)(k)=u;
wherein a is a development coefficient, and u is a gray acting amount;
let z(1)(k) Represents the average of the kth value and the previous value in the ideal altitude deviation sequence:
and (3) solving a development coefficient a and a gray action amount u by using a least square method:
(a,u)T=(BTB)-1BTY.
wherein B represents an ideal height deviation average matrix, and Y represents an ideal height deviation original data matrix;
substituting the development coefficient a and the gray action quantity u into a differential equation to solve to obtain an unreduced first prediction height deviation:
when i < n, differentiateProcessing sequenceSubtracting the sequences to obtain an ideal altitude deviation accumulation sequence, and sequentially performing decrement, geometric mean transformation and accelerated exponential transformation to restore to a second predicted altitude deviation:
the first reduction, the second reduction and the third reduction are respectively corresponding to decrement, geometric mean transformation and accelerated exponential transformation reduction;
the first reduction is specifically:
the second reduction is specifically as follows:
the third reduction is specifically as follows:
when i is larger than n, the first reduction and the second reduction are performed in sequence, if X is larger than n(0)(n+1)<X(0)(n), then a fourth reduction is performed:
if X(0)(n+1)>X(0)(n), then a fifth reduction is performed:
reducing the height deviation to a third predicted height deviation, wherein the preliminary predicted height deviation comprises the first predicted height deviation, the second predicted height deviation and the third predicted height deviation, and the third predicted height deviationIndicating the ith value in the sequence of differential processing.
5. The method for predicting the track irregularity in the ultrahigh environment based on the random oscillation sequence gray model according to claim 1, wherein the construction of the Elman neural network specifically comprises the following steps:
establishing a first layer as an input layer, inputting the preprocessed data into a network, and determining an input numerical value and an input node number, wherein the input numerical value is a preliminary prediction height residual error;
establishing a second layer as a hidden layer, determining the number of nodes according to an empirical formula, and selecting a nonlinear function as a first activation function according to the range of input numerical values;
establishing a third layer as a receiving layer, receiving a feedback signal from the hidden layer, wherein each hidden layer node is connected with a corresponding node, and the hidden layer state at the previous moment and the network input at the current moment are taken as the input of the hidden layer through connection memory;
establishing a fourth layer as an output layer, selecting a linear transfer function as a second activation function, and outputting a neural network calculation result;
and learning the neural network by using a reverse error propagation algorithm, and continuously updating the weight value by a gradient descent method.
7. The stochastic oscillation sequence gray model-based ultrahigh prediction method of orbital irregularity according to claim 5, wherein the first activation function is selected from a hyperbolic tangent sigmoid function, wherein the hyperbolic tangent sigmoid function is represented as:
wherein x is1E represents a natural constant, which is an input variable of the hidden layer;
the second activation function is:
f(x2)=x2;
wherein x2Representing the input variables of the output layer.
8. The ultrahigh prediction method of orbit irregularity based on the random oscillation sequence gray model as claimed in claim 1, wherein the optimization of the initial weight and threshold of the Elman neural network by the ant lion algorithm specifically comprises the following steps:
a condition initialization step:
the initial weight and the threshold value of the neural network are coded by adopting a real number coding mode, and the coding formula is as follows:
S=R×S1+S1×S1+S1×S2+S1+S2,
wherein R is the number of input nodes, S1 is the number of hidden layer nodes, and S2 is the number of output nodes;
taking the weight and threshold coding number of the Elman neural network as a population, initially setting ant lion and ant as the same population and the same population size, setting the number of each individual in the population as a random number in a preset individual number interval, and setting the weight and threshold of the neural network corresponding to each individual in the population;
a first neural network training step:
training parameter initialization: taking the preliminary prediction height residual error as training input, taking the corresponding correction height residual error as a training result, and evaluating the ant position based on the network training error;
screening elite ant lions: according to the updating of ant positions, giving and evaluating values, and selecting the ant lion with the highest evaluating value as the elite ant lion;
judging the training degree: inputting a test height residual error, outputting to obtain a first test prediction height residual error, and adjusting training according to whether the first test prediction height residual error reaches a first preset residual error precision value;
when the first test predicted height residual does not reach a first preset residual precision value, ant lion optimization is carried out, and the first neural network training step is repeated;
when the first test prediction height residual reaches a first preset residual precision value, executing a parameter updating step;
and updating parameters: reassigning the training parameters: assigning the optimal weight and the optimal threshold value after the ant lion algorithm optimization to the weight parameter and the threshold value parameter corresponding to the neural network, namely replacing the initial weight and the threshold value;
ant lion optimization: after the elite ant lion captures the ants, the ant lion position is updated to be the ant position, the ant position updating is influenced by the elite ant lion, and finally the optimal weight and the threshold are found;
a second neural network training step:
carrying out Elman neural network training, and updating weight parameters and threshold parameters based on network training errors;
and judging a training ending condition, adjusting training according to whether the second test predicted height residual reaches a second preset residual precision value, ending the second neural network training step when the second test predicted height residual reaches the second preset residual precision value, and otherwise, repeating the second neural network training step.
9. An ultrahigh prediction system of rail irregularity based on a random oscillation sequence gray model, comprising: the device comprises a data preprocessing module, a preliminary prediction module, a prediction correction module, a neural network module and an ultrahigh prediction module;
the data preprocessing module is used for carrying out mean value processing on the detected height deviation of the left rail surface and the right rail surface to obtain an equidistant average height deviation sequence;
the preliminary prediction module is used for carrying out random oscillation sequence gray prediction on the equidistant average height deviation sequence based on a gray model to obtain preliminary prediction height deviation;
the grey prediction of the random oscillation sequence is carried out on the equidistant average height deviation sequence based on a grey model, and the method specifically comprises the following steps:
transformation of random oscillation sequence:
detecting and detecting the detection sections at the same time to obtain an equidistant average height deviation sequence, wherein the equidistant average height deviation sequence is as follows: x(0)={X(0)(1),X(0)(2),…,X(0)(n)};
Translating the equidistant average height deviation sequence to a full value, and setting the following parameters:
Max=max{x(0)(k)|k,k∈{1,2,…,n}};
Min=min{x(0)(k)|k,k∈{1,2,…,n}};
max represents the maximum value in the original equidistant average height deviation sequence, Min represents the minimum value in the original equidistant average height deviation sequence, T represents the maximum oscillation ratio of the original equidistant average height deviation sequence, and n is the maximum value of the sequence;
the sequence obtained by performing accelerated exponential transformation on the equidistant average height deviation sequence is a monotonically increasing height deviation sequence, namely:
XD1={x(1)d1,x(2)d1,…,x(n)d1};
wherein x (k) d1Where k is 1,2,3 … n, which is the kth accelerated exponential transformThe later monotonically increasing height deviation value;
for XD1And performing geometric mean generation transformation on the sequence to obtain a geometric mean processing height deviation sequence which keeps monotonous increase but the increase amplitude is reduced, namely an ideal height deviation sequence, namely:
XD2={x(1)d2,x(2)d2,…,x(n)d2};
x(k)d2the k is the k-th geometric mean processing height deviation value, k is 1, …, n is a positive integer;
the prediction correction module is used for obtaining a preliminary prediction height residual error based on the preliminary prediction height deviation and original data, calculating a height residual error average value, correcting the preliminary prediction height residual error based on the height residual error average value to obtain a corrected height residual error, and normalizing the preliminary prediction height residual error and the corrected height residual error;
the neural network module is used for dividing a training set and a testing set after normalization processing based on the preliminary prediction height residual error and the correction height residual error, building an Elman neural network, training based on the preliminary prediction height residual error and the correction height residual error of the training set, and optimizing an initial weight and a threshold of the Elman neural network through a ant lion algorithm so as to obtain an optimized Elman neural network;
and the ultrahigh prediction module is used for inputting the preliminary prediction height residual error based on the optimized Elman neural network to obtain an orbit prediction correction height residual error.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110493436.9A CN113326975B (en) | 2021-05-07 | 2021-05-07 | Ultrahigh prediction method for track irregularity based on random oscillation sequence gray model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110493436.9A CN113326975B (en) | 2021-05-07 | 2021-05-07 | Ultrahigh prediction method for track irregularity based on random oscillation sequence gray model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113326975A CN113326975A (en) | 2021-08-31 |
CN113326975B true CN113326975B (en) | 2022-07-22 |
Family
ID=77414291
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110493436.9A Active CN113326975B (en) | 2021-05-07 | 2021-05-07 | Ultrahigh prediction method for track irregularity based on random oscillation sequence gray model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113326975B (en) |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111310990A (en) * | 2020-02-07 | 2020-06-19 | 北京联合大学 | Improved gray combination model-based track quality prediction method and system |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170091615A1 (en) * | 2015-09-28 | 2017-03-30 | Siemens Aktiengesellschaft | System and method for predicting power plant operational parameters utilizing artificial neural network deep learning methodologies |
CN111581832B (en) * | 2020-05-12 | 2023-01-03 | 哈尔滨工程大学 | Improved gray Elman neural network hovercraft motion prediction method based on ARMA model correction |
-
2021
- 2021-05-07 CN CN202110493436.9A patent/CN113326975B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111310990A (en) * | 2020-02-07 | 2020-06-19 | 北京联合大学 | Improved gray combination model-based track quality prediction method and system |
Non-Patent Citations (1)
Title |
---|
基于分解集成策略和多目标优化算法的混合预测模型在电力负荷预测中的应用;吕勤 等;《河北电力技术》;20201231;正文第24-27、39页 * |
Also Published As
Publication number | Publication date |
---|---|
CN113326975A (en) | 2021-08-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110263866B (en) | Power consumer load interval prediction method based on deep learning | |
CN108900346B (en) | Wireless network flow prediction method based on LSTM network | |
CN111563706A (en) | Multivariable logistics freight volume prediction method based on LSTM network | |
CN107590565A (en) | A kind of method and device for building building energy consumption forecast model | |
CN111310990A (en) | Improved gray combination model-based track quality prediction method and system | |
Gao et al. | Mean-entropy-based shadowed sets: a novel three-way approximation of fuzzy sets | |
CN110163743A (en) | A kind of credit-graded approach based on hyperparameter optimization | |
CN117977576B (en) | Platform load prediction method based on multi-scale historical load data | |
CN115935834A (en) | History fitting method based on deep autoregressive network and continuous learning strategy | |
CN113255963A (en) | Road surface use performance prediction method based on road element splitting and deep learning model LSTM | |
CN115840240A (en) | Automatic driving positioning method and system based on LSTM deep reinforcement learning | |
CN117349797A (en) | Aircraft fault detection method and system based on artificial intelligence | |
CN112766537A (en) | Short-term electric load prediction method | |
CN110260914B (en) | Engineering safety monitoring system region division method based on measuring point space-time characteristics | |
CN118114118A (en) | CNDT-based fault diagnosis method for typical weapon equipment | |
CN113326975B (en) | Ultrahigh prediction method for track irregularity based on random oscillation sequence gray model | |
CN118017482A (en) | Flexible climbing capacity demand analysis method based on prediction error feature extraction | |
CN113159395A (en) | Deep learning-based sewage treatment plant water inflow prediction method and system | |
CN112858644A (en) | Concrete carbonization depth prediction method and prediction system | |
CN116628444A (en) | Water quality early warning method based on improved meta-learning | |
CN116542139A (en) | Method and device for predicting roughness of liquid jet polishing surface | |
CN105787507A (en) | LS-SVMs online learning method based on budget support vector set | |
CN116934603A (en) | Logging curve missing segment completion method and device, storage medium and electronic equipment | |
CN115982555A (en) | Sedimentation prediction method based on fuzzy information granulation and dynamic neural network | |
CN109376957A (en) | A kind of prediction technique of thermal power plant's load |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |