CN111209999A - Contact network performance degradation prediction method based on recurrent neural network - Google Patents
Contact network performance degradation prediction method based on recurrent neural network Download PDFInfo
- Publication number
- CN111209999A CN111209999A CN201811394868.9A CN201811394868A CN111209999A CN 111209999 A CN111209999 A CN 111209999A CN 201811394868 A CN201811394868 A CN 201811394868A CN 111209999 A CN111209999 A CN 111209999A
- Authority
- CN
- China
- Prior art keywords
- data
- neural network
- performance degradation
- network
- bow net
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- G06N3/04—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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a cyclic neural network-based catenary performance degradation prediction method, which comprises the following steps of: 1) inputting bow net detection data; 2) preprocessing bow net detection data; 3) extracting data characteristics by using a Hilbert-Huang transform analysis method; 4) training and prediction of the recurrent neural network: and training by adopting a long-term and short-term memory network according to neuron input data, finding out the rule of bow net detection data, and predicting the performance degradation trend of the overhead line system according to the rule. The method utilizes the long-term and short-term memory artificial neural network to predict the performance degradation condition of the contact network, and has high prediction output recognition rate and low network operation time.
Description
Technical Field
The invention relates to a method for predicting the performance degradation of a contact network, in particular to a method for predicting the performance degradation of the contact network based on a recurrent neural network.
Background
The overhead contact system is a main framework of the railway electrification engineering and is a special power transmission line which is erected along a railway line and supplies power to an electric locomotive. Is responsible for the important task of directly delivering the electric energy obtained from the traction power transformation to the electric locomotive for use. Therefore, the performance and the working state of the contact net directly influence the transportation capacity of the electrified railway.
At present, no good means for predicting the degradation of the contact performance exists in the market, and the existing method has the following defects: (1) the prediction accuracy is not high enough; (2) the prediction model has more parameters.
Disclosure of Invention
In order to solve the above problems, the present invention provides a method for predicting the degradation of the performance of a catenary based on a recurrent neural network, which specifically comprises the following steps:
s1, inputting bow net detection data: bow net test data include, but are not limited to, lead height, pull out value, pressure, hard spot, and contact line wear in contact geometry parameters of the catenary;
s2, data preprocessing: preprocessing bow net detection data, wherein preprocessing methods include but are not limited to filtering and data normalization;
s3, data feature extraction: the extraction method adopts a Hilbert-Huang transform analysis method, which comprises an empirical mode decomposition method and Hilbert spectrum analysis; specifically, an empirical mode decomposition method is adopted to decompose the bow net detection data, then the correlation of intrinsic mode function component data is calculated to obtain more definite characteristic information, all the characteristics are normalized and then are connected in series to form N-dimensional vector neuron input of a cyclic neural network;
s4, training and predicting a recurrent neural network: and training by adopting a long-term and short-term memory network according to neuron input data, finding out the rule of bow net detection data, and predicting the performance degradation trend of the contact net according to the rule.
Further, step S3 includes the following sub-steps:
s31, for input original signal X [ t ]]Firstly, finding out all extreme points, then using cubic spline function to interpolate all the extreme points and fitting them to obtain X [ t ]]Upper envelope of (2)Similarly, fit X [ t ]]Lower envelope of (2)
S32, calculating a mean line according to the upper envelope line and the lower envelope line:
s33, calculating the deviation of the original signal:
H[t]=X[t]-M[t];
s34. treating H [ t ]]As the original signal X [ t ]]Repeating the steps S31, S32 and S33, calculating K times if H [ t [ ]]Satisfies the formula:SDthe value is 0.2 to 0.3,
taking Ht as the eigenmode function component;
s35, the first 5 eigenmode function components are taken as a part of data characteristics, and are combined with the mean value and the variance of original data to form a combined characteristic F ═ IMF1,IMF2,IMF3,IMF4,IMF5,mean,var]。
Further, in step S4, the training of the long-short term memory network employs a time-based backward propagation algorithm.
Further, in step S4, the long-term and short-term memory network includes an input layer, a middle layer and an output layer, where the input layer is 3 × 7, the middle layer is 120, and the output layer is 2.
The invention has the beneficial effects that:
(1) the prediction output recognition rate is high, and the recognition accuracy is more than 93%;
(2) the network running time is low, and the time of one network running is less than 10 milliseconds.
Drawings
FIG. 1 is a schematic diagram of the present invention;
FIG. 2 is a diagram of the LSTM network architecture of the present invention;
fig. 3 is a network architecture diagram of the present invention.
Detailed Description
In order to more clearly understand the technical features, objects, and effects of the present invention, embodiments of the present invention will now be described with reference to the accompanying drawings.
The invention provides a cyclic neural network-based catenary performance degradation prediction method, which comprises the steps of inputting pantograph-catenary detection data, preprocessing data, extracting data characteristics and training and predicting a cyclic neural network as shown in figure 1, and specifically comprises the following steps:
(1) inputting bow net detection data: and inputting the height, the pull-out value, the pressure, the hard point and the contact line abrasion data in the contact geometric parameters of the contact network.
(2) Data preprocessing: and preprocessing the bow net detection data by adopting a filtering and data normalization method.
(3) Data feature extraction: considering that the geometric parameter data of the contact network are basically one-dimensional signals, the method adopts a Hilbert-Huang transform analysis method (HHT method), which is a novel method for processing non-stationary signals. The method consists of an empirical mode decomposition method (EMD method) and Hilbert spectrum analysis, and the core of the method is the empirical mode decomposition method. The HHT method is a time-frequency localization analysis method which has better adaptability than Fourier transform, wavelet transform and the like, obtains instantaneous frequency components with practical physical significance in signals, further realizes time-frequency analysis with high resolution, and has the characteristic of good adaptability. The invention adopts EMD method to decompose the detection data, and then calculates the correlation of IMF (Intrinsic Mode Function) component data, so as to obtain more definite characteristic information, and all the characteristics are normalized and concatenated together to form the neuron input of an N-dimensional vector of the recurrent neural network.
Empirical Mode Decomposition (EMD) principle:
for the input original signal X [ t ]]Firstly, finding out all extreme points, then using cubic spline function to interpolate all the extreme points and fitting them to obtain X [ t ]]Upper envelope of (2)Similarly, fit X [ t ]]Lower envelope line of
Calculating a mean line according to the upper envelope line and the lower envelope line:
calculate the deviation of the original signal:
H[t]=X[t]-M[t];
taking Ht as the original signal X t, repeating the above steps, calculating K times, if Ht satisfies the formula:
taking Ht as the eigenmode function component;
the first 5 eigenmode function components are taken as a part of data characteristics and are concatenated into a combined characteristic F ═ IMF by combining the mean value and the variance of original data1,IMF2,IMF3,IMF4,IMF5,mean,var]。
(4) Training and prediction of the recurrent neural network: the Recurrent Neural Network (RNN) is used for solving the relation between the input and output feedback before and after the network, and is mainly applied to time sequence data analysis. According to the method, an LSTM (Long short-term memory network) is adopted, the rule of historical data is found according to historical data of the pantograph catenary, and the performance degradation trend of the catenary is predicted.
1) LSTM network
LSTM is a special cyclic neural network, which can learn long-term dependency, and is different from the internal operation of the neuron in the traditional neural network, and the LSTM adopts a neuron internal structure as shown in FIG. 2, wherein X is input of the neuron, X [ t-1] is input at the previous moment, X [ t ] is input at the current moment, X [ t +1] is input at the next moment, σ is sigmod function, tanh () is hyperbolic tangent, and h [ t ] is output at the current moment of the neuron:
h [ t ], (h [ t-1], (W x (x [ t-1], h [ t-1 ]))) + sigmoid (W x (x [ t-1], h [ t-1 ]))) tanh +. A plurality of neurons A are strung together to form a network layer.
2) Training of LSTM
The training algorithm of LSMT adopts a Back Propagation t tau (Back Propagation t time) based on time, which is basically a BP algorithm developed according to a time dimension, Forward Propagation (Forward Propagation) is sequentially calculated once according to a time sequence, and Back Propagation (Back Propagation) is to transmit an accumulated residual error Back from the last time.
3) Network architecture of the present invention
The invention adopts 3 layers of LSTM network, wherein the input layer is 3X 7, the middle layer is 120, the output is 2, the network structure is shown in figure 3, wherein x [ t ] is the geometrical parameter characteristic of the current month, x [ t-1] is the geometrical parameter characteristic of the previous month, x [ t-2] is the geometrical parameter characteristic of the previous month, the output y [ t ] is the performance degradation grade of the current month, y [ t +1] is the performance degradation grade of the next month, the grade is divided into 5, and the performance degradation is more serious when the value is larger.
The foregoing is illustrative of the preferred embodiments of the present invention, and it is to be understood that the invention is not limited to the precise forms disclosed herein, and that various other combinations, modifications, and environments may be resorted to, falling within the scope of the invention as hereinafter claimed, and that changes may be made by those skilled in the art or by those who review this disclosure. And that modifications and variations may be effected by those skilled in the art without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (4)
1. A cyclic neural network-based catenary performance degradation prediction method is characterized by comprising the following steps:
s1, inputting bow net detection data: bow net test data include, but are not limited to, lead height, pull out value, pressure, hard spot, and contact line wear in contact geometry parameters of the catenary;
s2, data preprocessing: preprocessing bow net detection data, wherein preprocessing methods include but are not limited to filtering and data normalization;
s3, data feature extraction: the extraction method adopts a Hilbert-Huang transform analysis method, which comprises an empirical mode decomposition method and Hilbert spectrum analysis; specifically, an empirical mode decomposition method is adopted to decompose the bow net detection data, then the correlation of intrinsic mode function component data is calculated to obtain more definite characteristic information, all the characteristics are normalized and then are connected in series to form the neuron input of an N-dimensional vector of the recurrent neural network;
s4, training and predicting a recurrent neural network: and training by adopting a long-term and short-term memory network according to neuron input data, finding out the rule of bow net detection data, and predicting the performance degradation trend of the overhead line system according to the rule.
2. The cyclic neural network-based catenary performance degradation prediction method of claim 1, wherein the step S3 comprises the following substeps:
s31, for input original signal X [ t ]]Firstly, finding out all extreme points, then using cubic spline function to interpolate all the extreme points and fitting them to obtain X [ t ]]Upper envelope of (2)Similarly, fit X [ t ]]Lower envelope of (2)
S32, calculating a mean line according to the upper envelope line and the lower envelope line:
s33, calculating the deviation of the original signal:
H[t]=X[t]-M[t];
s34. treating H [ t ]]As the original signal X [ t ]]Repeating the steps S31, S32 and S33, calculating K times if H [ t [ ]]Satisfies the formula:SDtaking the value of 0.2-0.3, then H [ t ] is added]As an intrinsic mode function component;
s35, the first 5 eigenmode function components are taken as a part of data characteristics, and are combined with the mean value and the variance of original data to form a combined characteristic F ═ IMF1,IMF2,IMF3,IMF4,IMF5,mean,var]。
3. The method for predicting the performance degradation of the catenary based on the recurrent neural network as claimed in claim 1, wherein in step S4, the training of the long-short term memory network employs a time-based back propagation algorithm.
4. The method of claim 1, wherein in step S4, the long-term and short-term memory network comprises an input layer, an intermediate layer and an output layer, wherein the input layer is 3 × 7, the intermediate layer is 120, and the output layer is 2.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811394868.9A CN111209999B (en) | 2018-11-21 | 2018-11-21 | Contact network performance degradation prediction method based on recurrent neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811394868.9A CN111209999B (en) | 2018-11-21 | 2018-11-21 | Contact network performance degradation prediction method based on recurrent neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111209999A true CN111209999A (en) | 2020-05-29 |
CN111209999B CN111209999B (en) | 2023-04-07 |
Family
ID=70789397
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811394868.9A Active CN111209999B (en) | 2018-11-21 | 2018-11-21 | Contact network performance degradation prediction method based on recurrent neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111209999B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111652395A (en) * | 2020-06-12 | 2020-09-11 | 成都国铁电气设备有限公司 | Health assessment method for high-speed railway contact network equipment |
CN112214852A (en) * | 2020-10-09 | 2021-01-12 | 电子科技大学 | Degradation rate-considered turbine mechanical performance degradation prediction method |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060146432A1 (en) * | 2004-12-30 | 2006-07-06 | Andrei Khurshudov | Methods for detecting contact between a read-write head and the accessed disk surface in a hard disk drive |
CN102521651A (en) * | 2011-12-22 | 2012-06-27 | 北京交通大学 | Bow net contact force prediction method based on NARX neural networks |
CN203039801U (en) * | 2012-11-30 | 2013-07-03 | 成都唐源电气有限责任公司 | Digital video monitoring apparatus for dynamic operation of railway pantograph and catenary |
JP2014132419A (en) * | 2013-01-07 | 2014-07-17 | Nec Corp | Performance prediction device of virtualization system, performance prediction method and computer program |
-
2018
- 2018-11-21 CN CN201811394868.9A patent/CN111209999B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060146432A1 (en) * | 2004-12-30 | 2006-07-06 | Andrei Khurshudov | Methods for detecting contact between a read-write head and the accessed disk surface in a hard disk drive |
CN102521651A (en) * | 2011-12-22 | 2012-06-27 | 北京交通大学 | Bow net contact force prediction method based on NARX neural networks |
CN203039801U (en) * | 2012-11-30 | 2013-07-03 | 成都唐源电气有限责任公司 | Digital video monitoring apparatus for dynamic operation of railway pantograph and catenary |
JP2014132419A (en) * | 2013-01-07 | 2014-07-17 | Nec Corp | Performance prediction device of virtualization system, performance prediction method and computer program |
Non-Patent Citations (5)
Title |
---|
J. A. S. B. JAYASINGHE: "A new approach to condition monitoring of high voltage overhead lines using neural networks", 《1999 ELEVENTH INTERNATIONAL SYMPOSIUM ON HIGH VOLTAGE ENGINEERING》 * |
何栋: "高速铁路接触网状态退化评估", 《中国优秀硕士论文辑》 * |
宋冬利: "多状态多模式受电弓机械系统混合可靠性模型研究", 《中国优秀硕士论文辑》 * |
张爱静: "基于鼻羁压力的奶牛反刍识别方法研究", 《中国优秀硕士论文辑》 * |
施斌: "高速铁路接触网静动态性能退化特征分析", 《中国优秀硕士论文辑》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111652395A (en) * | 2020-06-12 | 2020-09-11 | 成都国铁电气设备有限公司 | Health assessment method for high-speed railway contact network equipment |
CN112214852A (en) * | 2020-10-09 | 2021-01-12 | 电子科技大学 | Degradation rate-considered turbine mechanical performance degradation prediction method |
Also Published As
Publication number | Publication date |
---|---|
CN111209999B (en) | 2023-04-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110728360B (en) | Micro-energy device energy identification method based on BP neural network | |
CN109815223B (en) | Completion method and completion device for industrial monitoring data loss | |
CN109884419B (en) | Smart power grid power quality online fault diagnosis method | |
CN110070229A (en) | The short term prediction method of home electrical load | |
Khodayar et al. | Robust deep neural network for wind speed prediction | |
CN111209999B (en) | Contact network performance degradation prediction method based on recurrent neural network | |
Ramesh Babu et al. | Improving forecast accuracy of wind speed using wavelet transform and neural networks | |
CN110826803A (en) | Electricity price prediction method and device for electric power spot market | |
CN111508526A (en) | Method and device for detecting audio beat information and storage medium | |
CN113705396B (en) | Motor fault diagnosis method, system and equipment | |
CN110726898A (en) | Power distribution network fault type identification method | |
CN110880010A (en) | Visual SLAM closed loop detection algorithm based on convolutional neural network | |
CN112257847A (en) | Method for predicting geomagnetic Kp index based on CNN and LSTM | |
CN112305379A (en) | Mode identification method and system for GIS insulation defect | |
CN112418175A (en) | Rolling bearing fault diagnosis method and system based on domain migration and storage medium | |
CN115758208A (en) | Traction converter fault diagnosis method and device, computer equipment and storage medium | |
CN114049305A (en) | Distribution line pin defect detection method based on improved ALI and fast-RCNN | |
CN110879927A (en) | Sea clutter amplitude statistical distribution field modeling method for sea target detection | |
CN110766060A (en) | Time series similarity calculation method, system and medium based on deep learning | |
CN115827335A (en) | Time sequence data missing interpolation system and method based on modal crossing method | |
Fu et al. | Research on gas outburst prediction model based on multiple strategy fusion improved snake optimization algorithm with temporal convolutional network | |
CN114626487A (en) | Line-variable relation checking method based on random forest classification algorithm | |
CN103279581B (en) | A kind of method utilizing compact video subject description to carry out video frequency searching | |
CN107862375A (en) | A kind of two stage equipment fault diagnosis method | |
CN114266201A (en) | Self-attention elevator trapping prediction method based on deep learning |
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 |