CN110263447A - A kind of loading spectrum Extrapolation method based on shot and long term memory network - Google Patents
A kind of loading spectrum Extrapolation method based on shot and long term memory network Download PDFInfo
- Publication number
- CN110263447A CN110263447A CN201910551153.8A CN201910551153A CN110263447A CN 110263447 A CN110263447 A CN 110263447A CN 201910551153 A CN201910551153 A CN 201910551153A CN 110263447 A CN110263447 A CN 110263447A
- Authority
- CN
- China
- Prior art keywords
- loading spectrum
- layer
- shot
- long term
- term memory
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/15—Vehicle, aircraft or watercraft design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Aviation & Aerospace Engineering (AREA)
- Automation & Control Theory (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Machine Translation (AREA)
Abstract
The present invention relates to loading spectrum processing technology field, specially a kind of loading spectrum Extrapolation method based on shot and long term memory network, comprising the following steps: S100: load modal data is obtained;S200: load modal data is subjected to data processing, divides training set and test set;S300: it is based on LSTM algorithm, is trained using training set, obtains input sample data model;S400: extrapolation calculating is carried out to load modal data using test set by obtained input sample data model.A kind of loading spectrum Extrapolation method based on shot and long term memory network provided by the invention enables to extrapolation loading spectrum preferably to express the information of former loading spectrum, and can solve human factor in the prior art influences big, the incomplete problem of loading spectrum data representation.
Description
Technical field
The present invention relates to loading spectrum processing technology field, specially a kind of loading spectrum extrapolation based on shot and long term memory network
Method.
Background technique
In the development process of automobile or components, reliability is required to progress actual road test and tests.According to damage
The principle of equal effects, under known users use environment automobile load input, theoretically can by test site according to a certain percentage
Various enhanced road surfaces are mixed, the load input under user's operating condition out is reappeared.It, can be shorter by the enhanced road surface at test site
Reliability compliance test is completed in time, is achieved the purpose that reduce test period, is shortened the R&D cycle.For research and development cost and when
Between the considerations of, general user's use environment automobile load input, loading spectrum acquisition will not be carried out on the basis of target mileage, and
It is to be acquired by type according to user's road ratio distribution situation, under the premise of sample size is enough, carries out outside loading spectrum
It pushes away, realizes that the loading spectrum under target mileage obtains.
Current more typical Extrapolation method includes: parameter extrapolation, by mileage quantile extrapolation, peak value (Peak
Over Threshold, POT) extrapolation, rainflow matrix extrapolation [1-3] etc..Wherein the principle of parameter extrapolation is to obtain to carry
Lotus composes equal, amplitude dimensional probability distribution function, based on probability density function and extrapolation target mileage, will tire out accordingly
The product frequency is extrapolated.POT extrapolation thinks to obey peak Distribution more than the peak value of threshold value in loading spectrum time series, by right
More than the fitting of the probability density function of the peak value of threshold value, extrapolated based on probability density function to peak value.Outside rainflow matrix
Loading spectrum is first obtained rainflow matrix by rain-flow counting by pushing manipulation, and the threshold calculations of extrapolation is selected to pass through grade from rainflow matrix
Then density obtains the limit by accumulation rainflow matrix and carries out rainflow matrix estimation extrapolation.The above method can be according to loading spectrum
Feature or application purpose are selected, but often introduced when being fitted with distribution function or threshold value being arranged it is artificial because
Element, and simple distribution function cannot express highly complex load spectrum information, extrapolation loading spectrum can be generated compared with original load spectrum
It is lost compared with multi information, extrapolating results need to be then converted to input of the time-domain program spectrum as next step.
Summary of the invention
The invention is intended to provide a kind of loading spectrum Extrapolation method based on shot and long term memory network, extrapolation load is enabled to
Spectrum preferably expresses the information of former loading spectrum, and can solve human factor in the prior art influences big, loading spectrum data representation not
Complete problem.
In order to solve the above-mentioned technical problem, the application provides the following technical solutions:
A kind of loading spectrum Extrapolation method based on shot and long term memory network, comprising the following steps:
S100: load modal data is obtained;
S200: load modal data is subjected to data processing, divides training set and test set;
S300: it is based on LSTM algorithm, is trained using training set, obtains input sample data model;
S400: extrapolation calculating is carried out to load modal data using test set by obtained input sample data model.
In technical solution of the present invention, using LSTM shot and long term memory network algorithm model, using computer learning technology into
The building of row sample data model and load modal data carry out extrapolation calculating, do not need that parameter artificially is arranged, the load postponed outside
Spectrum can preferably retain the information of former loading spectrum on frequency domain figure and rain flow graph, can solve human factor shadow in the prior art
It rings big, load modal data and retains incomplete problem.
Further, the unit of the LSTM algorithm includes input layer, output layer, forgets layer and state update step.
The content transmitted by forgeing layer to a upper unit carries out the processing of selectivity, is updated by state update step
The memory state of upper unit transmission, is handled by data of the input layer to input, output layer according to the input of input layer,
The memory state that state update step updates generates the output of active cell.
Further, the input layer controls output, the relational expression of the input layer are as follows: i by sigmoid functiont=σ
(Wi·[ht-1, Xt]+bi), itFor the output of input layer, σ is sigmoid function, WiFor the weight of input layer, ht-1It is upper one
The output of LSTM unit is as a result, XtFor the input of current LSTM unit, biFor the biasing of input layer.
By sigmoid function control output in the range of [0,1], control new information be added into number.
Further, the state update step is controlled by tanh function and is exported, the relational expression of state update step are as follows: For the output of state update step, tanh is tanh function, WcFor state update step
Weight, bcFor the biasing of state update step.
In new memory state CtBefore generation, interim memory state can be first generatedIt is by upper unit ht-1Shadow
It rings, is controlled and exported by nonlinear activation function tanh.
Further, the forgetting layer is controlled by sigmoid function and is exported, and forgets the relational expression of layer are as follows: ft=σ (Wf·
[ht-1, Xt]+bf), ftFor the output for forgeing layer, WfFor the weight for forgeing layer, bfFor the biasing for forgeing layer.
It is similar with input layer, through the control output of sigmoid function in the range of [0,1], control the note of a upper unit
State is recalled to the influence degree of current memory unit.
Further, the relational expression of the output layer are as follows:
Wherein, htFor the output of current LSTM unit, Ct-1For the memory state of a upper LSTM unit, CtFor current LSTM
The memory state of unit, WoFor the weight of output layer, boFor the biasing of output layer.
Filter out redundancy present in unit, if having reached threshold value when output, just by the output of the valve with work as
The calculated result of front layer is multiplied, and using obtained result as next layer of input;If not up to threshold value, " forgetting " output
As a result.
Further, the S300 is specifically included:
S310: building sample data model structure;
S320: the data of the data exported after training and test set are compared, and calculate the cost function of two groups of data;
S330: being adjusted each layer weight and biasing according to the calculated result of S320, until cost function converge to.
Further, use Squared Error Loss as cost function in the S320.
Further, the data processing in the S200 includes one of normalization, standardization and regularization or a variety of.
Further, it is 100, time step 20ms that the LSTM algorithm, which defines neuronal quantity, and batch processing size is
60, Learning Step 0.0001.
Detailed description of the invention
Fig. 1 is the flow chart in a kind of loading spectrum Extrapolation method embodiment based on shot and long term memory network of the present invention;
Fig. 2 is the knot of a unit in a kind of loading spectrum Extrapolation method embodiment based on shot and long term memory network of the present invention
Structure schematic diagram;
Fig. 3 is the variation of the biasing in a kind of loading spectrum Extrapolation method embodiment based on shot and long term memory network of the present invention
Figure;
Fig. 4 is the variation of the weight in a kind of loading spectrum Extrapolation method embodiment based on shot and long term memory network of the present invention
Figure;
Fig. 5 is the original load of pebble path in a kind of loading spectrum Extrapolation method embodiment based on shot and long term memory network of the present invention
Lotus spectrum;
Fig. 6 is that Belgian road is original in a kind of loading spectrum Extrapolation method embodiment based on shot and long term memory network of the present invention
Loading spectrum;
Fig. 7 is that pebble path extrapolation carries in a kind of loading spectrum Extrapolation method embodiment based on shot and long term memory network of the present invention
Lotus spectrum;
Fig. 8 is Belgian road extrapolation in a kind of loading spectrum Extrapolation method embodiment based on shot and long term memory network of the present invention
Loading spectrum;
Fig. 9 is pebble path loading spectrum in a kind of loading spectrum Extrapolation method embodiment based on shot and long term memory network of the present invention
Extrapolation comparison diagram;
Figure 10 is Belgian road road in a kind of loading spectrum Extrapolation method embodiment based on shot and long term memory network of the present invention
Loading spectrum extrapolation comparison diagram;
Figure 11 is original load spectrum in a kind of loading spectrum Extrapolation method embodiment based on shot and long term memory network of the present invention
Rain flow graph;
Figure 12 is in a kind of loading spectrum Extrapolation method embodiment based on shot and long term memory network of the present invention outside LSTM method
Push away loading spectrum rain flow graph;
Figure 13 is in a kind of loading spectrum Extrapolation method embodiment based on shot and long term memory network of the present invention
Expanechekov kernel function extrapolation rain flow graph;
Figure 14 is Circle core letter in a kind of loading spectrum Extrapolation method embodiment based on shot and long term memory network of the present invention
Number extrapolation rain flow graph;
Figure 15 is mean value kernel function in a kind of loading spectrum Extrapolation method embodiment based on shot and long term memory network of the present invention
Extrapolation rain flow graph;
Figure 16 is amplitude kernel function in a kind of loading spectrum Extrapolation method embodiment based on shot and long term memory network of the present invention
Extrapolation rain flow graph.
Specific embodiment
It is further described below by specific embodiment:
Embodiment one
As shown in Figure 1, a kind of loading spectrum Extrapolation method based on shot and long term memory network of the present embodiment, comprising:
S100: load modal data is obtained;
S200: load modal data is subjected to data processing, divides training set and test set;Data processing in S200 includes
One of normalization, standardization and regularization are a variety of.Regularization is used in the present embodiment.
S300: it is based on LSTM algorithm, is trained using training set, obtains input sample data model;S300 is specifically wrapped
It includes:
S310: building sample data model structure;
S320: the data of the data exported after training and test set are compared, and calculate the cost function of two groups of data;
S330: being adjusted each layer weight and biasing according to the calculated result of S320, until cost function convergence.
Using Squared Error Loss as cost function in S320.
S400: extrapolation calculating is carried out to load modal data using test set by obtained input sample data model.
In the present embodiment, it is 100, time step 20ms that LSTM algorithm, which defines neural unit quantity, batch processing size
It is 60, Learning Step 0.0001.As shown in Fig. 2, each neural unit of LSTM algorithm includes input layer, output layer, forgetting
Layer and state update step.Input layer controls output by sigmoid function in the range of [0,1], to control new information
The number being added into.The relational expression of input layer are as follows: it=σ (Wi·[ht-1, Xt]+bi), itFor the output of input layer, σ is
Sigmoid function, WiFor the weight of input layer, ht-1For a upper LSTM unit output as a result, XtFor current LSTM unit
Input, biFor the biasing of input layer.
State update step is controlled by tanh function and is exported, the relational expression of state update step are as follows: For the output of state update step, tanh is tanh function, WcFor state update step
Weight, bcFor the biasing of state update step.
Forget layer and output is controlled by sigmoid function, forgets the relational expression of layer are as follows: ft=σ (Wf·[ht-1, Xt]+bf), ft
For the output for forgeing layer, wfFor the weight for forgeing layer, bfFor the biasing for forgeing layer.It is similar with input layer, forget layer and passes through
The control output of sigmoid function controls shadow of the memory state to current memory unit of a upper unit in the range of [0,1]
The degree of sound.
Output layer filters out redundancy present in unit, if having reached threshold value when output, just by the defeated of the valve
It is multiplied out with the calculated result of current layer, and using obtained result as next layer of input;If not up to threshold value, " lose
Forget " output result relational expression are as follows:
Wherein, htFor the output of current LSTM unit, Ct-1For the memory state of a upper LSTM unit, CtFor current LSTM
The memory state of unit, WoFor the weight of output layer, boFor the biasing of output layer.
Specifically, in the present embodiment, by taking the extrapolation of the loading spectrum of enhanced road surface as an example, at sample car respectively core wheel and damping
Acceleration transducer, spring and part connecting rod position are installed on device, foil gauge is installed, several in certain proving ground are main special
Sign is strengthened road and is acquired, and the loading spectrum of multi collect enhanced road surface is needed, and effective loading spectrum is finally taken to carry out rejecting surprise
The pretreatments such as dissimilarity, elimination trend term, filtering.Various test traffic informations are as shown in table 1 below:
Mainly strengthen road road conditions in 1 test site of table:
In the present embodiment, LSTM extrapolation mould is established using the artificial intelligence Open-Source Tools TensorFlow issued based on Google
The characteristics of type, TensorFlow is that calculating task is indicated using figure, and in figure a running node obtains 0 or multiple tensors
It executes calculating, generates 0 or multiple tensors, each tensor is a typed Multidimensional numerical.TensorFlow is in meeting
Figure is executed in the context of words, using tensor representation data, state is safeguarded by variable.
TensorFlow, pandas are imported by python environment first in the present embodiment, numpy module will utilize csv
The time domain modal data that format is saved uses data-numpy.mean by pandas.read_csv function read-in programme
(data))/numpy.std (data) function is standardized.It is by LSTM model encapsulation in TensorFlow
LSTMCell module, it is 100, time step 20ms that neuronal quantity will be defined in the present embodiment, and batch processing size is 60,
Learning Step is 0.0001, calls directly the module after initialization input and output interface, is instructed using the process of Fig. 1
Practice.In the present embodiment by NVIDIA provide CUDA as hardware support, training process ratio with about 20 times fastly of common CP U,
In entire machine-learning process, by tensorboard it can be seen that in entire learning process parameter variation, such as Fig. 3 and
Shown in Fig. 4, when study accuracy rate reaches 98%, the income for continuing study becomes smaller, can deconditioning.
It can be extrapolated using the hyper parameter model that training obtains, 10 times of extrapolating results of part road vertical acceleration are such as
Shown in Fig. 5 to Fig. 8.
LSTM method extrapolated data and initial data spectrum curve are compared, part road spectrum curve such as Fig. 9 and
Shown in Figure 10.
On spectrogram, extrapolation front and back data spectrum has good consistency in total form, utilizes Pearson correlation
Y-factor method Y tests to consistency.Pearson correlation coefficient ρX, YIt is the degree for measuring array X and Y linear correlation, coefficient
Value always between -1.0 to 1.0, the variable close to 0 is known as non-correlation, and being referred to as close to 1 or -1 has strong phase
Guan Xing.Spectrum curve before and after extrapolating, the results are shown in Table after the inspection of Pearson correlation coefficient method is shown, the results are shown in Table 2.Through
The spectrum curve of LSTM method extrapolation front and back data has strong correlation, illustrates that this method can have very the spectrum signature of initial data
High learning rate.
The correlation of the extrapolation of table 2 front and back data spectrum curve
Road surface | Pearson correlation coefficient |
Pebble path | 0.99 |
Belgian road | 0.95 |
Become the long wave paths of wave square | 0.97 |
Washboard road | 0.96 |
Resonance road | 0.97 |
Broken stone road | 0.99 |
In order to which the technical effect to the application is verified, the application is also compared with existing Extrapolation method, right
The characteristics of extrapolation of loading spectrum, several Extrapolation methods generallyd use, is as shown in table 3:
The comparison of several Extrapolation methods of table 3
Based on be respectively adopted in LSTM Extrapolation method and norm of nonparametric kernel density Extrapolation method expanechekov kernel function,
Circle kernel function, mean value kernel function, amplitude kernel function extrapolate to loading spectrum collected on pebble path, several method
Rain flow graph comparison, see Figure 11 to Figure 16.
The distribution that rain flow graph from Figure 11 to Figure 16 can be seen that norm of nonparametric kernel density extrapolation tends to " monokaryon " feature,
LSTM method extrapolation distribution have " multicore " feature, the distribution situation of the latter and to it is each to the reproduction effect become estranged a little it is more preferable,
LSTM method may be selected in Practical Project to extrapolate.
Embodiment two
The difference between this embodiment and the first embodiment lies in defining neuronal quantity is 200, time step in the present embodiment
A length of 30ms, batch processing size are 80, Learning Step 0.0001.
The above are merely the embodiment of the present invention, the field that invention case study on implementation without being limited thereto is related to is known in scheme
Specific structure and the common sense such as characteristic do not describe excessively herein, one skilled in the art know the applying date or preferential
All ordinary technical knowledges of technical field that the present invention belongs to before Quan can know the prior art all in the field, and
And there is the ability for applying routine experiment means before the date, what one skilled in the art can provide in the application
Under enlightenment, this programme is improved and implemented in conjunction with self-ability, and some typical known features or known method should not become
One skilled in the art implement the obstacle of the application.It should be pointed out that for those skilled in the art, not taking off
Under the premise of from structure of the invention, several modifications and improvements can also be made, these also should be considered as protection scope of the present invention,
These all will not influence the effect and patent practicability that the present invention is implemented.This application claims protection scope should be with its right
It is required that content subject to, the records such as specific embodiment in specification can be used for explaining the content of claim.
Claims (10)
1. a kind of loading spectrum Extrapolation method based on shot and long term memory network, comprising the following steps: S100: obtaining loading spectrum number
According to;S200: load modal data is subjected to data processing, divides training set and test set;S300: it is based on LSTM algorithm, uses instruction
Practice collection to be trained, obtains input sample data model;S400: test set pair is used by obtained input sample data model
Load modal data carries out extrapolation calculating.
2. a kind of loading spectrum Extrapolation method based on shot and long term memory network according to claim 1, it is characterised in that: institute
The unit for stating LSTM algorithm includes input layer, output layer, forgets layer and state update step.
3. a kind of loading spectrum Extrapolation method based on shot and long term memory network according to claim 2, it is characterised in that: institute
It states input layer and controls output, the relational expression of the input layer are as follows: i by sigmoid functiont=σ (Wi·[ht-1,Xt]+bi),
itFor the output of input layer, σ is sigmoid function, WiFor the weight of input layer, ht-1For the output knot of a upper LSTM unit
Fruit, XtFor the input of current LSTM unit, biFor the biasing of input layer.
4. a kind of loading spectrum Extrapolation method based on shot and long term memory network according to claim 3, it is characterised in that: institute
It states state update step and output, the relational expression of state update step is controlled by tanh function are as follows: For the output of state update step, tanh is tanh function, WcFor the weight of state update step, bcFor state update step
Biasing.
5. a kind of loading spectrum Extrapolation method based on shot and long term memory network according to claim 4, it is characterised in that: institute
It states and forgets layer by the control output of sigmoid function, forget the relational expression of layer are as follows: ft=σ (Wf·[ht-1,Xt]+bf), ftTo forget
The output of layer, WfFor the weight for forgeing layer, bfFor the biasing for forgeing layer.
6. a kind of loading spectrum Extrapolation method based on shot and long term memory network according to claim 5, it is characterised in that: institute
State the relational expression of output layer are as follows:
ot=σ (Wo·[ht-1,Xt]+bo);
ht=ot*tanh(Ct);
Wherein, htFor the output of current LSTM unit, Ct-1For the memory state of a upper LSTM unit, CtFor current LSTM unit
Memory state, WoFor the weight of output layer, boFor the biasing of output layer.
7. a kind of loading spectrum Extrapolation method based on shot and long term memory network according to claim 1, it is characterised in that: institute
S300 is stated to specifically include:
S310: building sample data model structure;
S320: the data of the data exported after training and test set are compared, and calculate the cost function of two groups of data;
S330: being adjusted each layer weight and biasing according to the calculated result of S320, until cost function converge to.
8. a kind of loading spectrum Extrapolation method based on shot and long term memory network according to claim 7, it is characterised in that: institute
It states in S320 using Squared Error Loss as cost function.
9. a kind of loading spectrum Extrapolation method based on shot and long term memory network according to claim 1, it is characterised in that: institute
Stating the data processing in S200 includes one of normalization, standardization and regularization or a variety of.
10. a kind of loading spectrum Extrapolation method based on shot and long term memory network according to claim 2, it is characterised in that:
It is 100, time step 20ms that the LSTM algorithm, which defines neuronal quantity, and batch processing size is 60, and Learning Step is
0.0001。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910551153.8A CN110263447A (en) | 2019-06-24 | 2019-06-24 | A kind of loading spectrum Extrapolation method based on shot and long term memory network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910551153.8A CN110263447A (en) | 2019-06-24 | 2019-06-24 | A kind of loading spectrum Extrapolation method based on shot and long term memory network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110263447A true CN110263447A (en) | 2019-09-20 |
Family
ID=67921109
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910551153.8A Pending CN110263447A (en) | 2019-06-24 | 2019-06-24 | A kind of loading spectrum Extrapolation method based on shot and long term memory network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110263447A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111222199A (en) * | 2019-11-13 | 2020-06-02 | 中国汽车工程研究院股份有限公司 | Key index selection and equivalent calculation method during association of user and test field |
CN112389436A (en) * | 2020-11-25 | 2021-02-23 | 中汽院智能网联科技有限公司 | Safety automatic driving track-changing planning method based on improved LSTM neural network |
CN116663434A (en) * | 2023-07-31 | 2023-08-29 | 江铃汽车股份有限公司 | Whole vehicle load decomposition method based on LSTM deep neural network |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106407649A (en) * | 2016-08-26 | 2017-02-15 | 中国矿业大学(北京) | Onset time automatic picking method of microseismic signal on the basis of time-recursive neural network |
CN106934184A (en) * | 2017-04-25 | 2017-07-07 | 吉林大学 | A kind of Digit Control Machine Tool load Extrapolation method based on the extension of time domain load |
CN109558971A (en) * | 2018-11-09 | 2019-04-02 | 河海大学 | Intelligent landslide monitoring device and method based on LSTM shot and long term memory network |
-
2019
- 2019-06-24 CN CN201910551153.8A patent/CN110263447A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106407649A (en) * | 2016-08-26 | 2017-02-15 | 中国矿业大学(北京) | Onset time automatic picking method of microseismic signal on the basis of time-recursive neural network |
CN106934184A (en) * | 2017-04-25 | 2017-07-07 | 吉林大学 | A kind of Digit Control Machine Tool load Extrapolation method based on the extension of time domain load |
CN109558971A (en) * | 2018-11-09 | 2019-04-02 | 河海大学 | Intelligent landslide monitoring device and method based on LSTM shot and long term memory network |
Non-Patent Citations (4)
Title |
---|
冯小平 等: "《通信对抗原理》", 31 August 2009, 西安电子科技大学出版社 * |
刘彦龙 等: "基于挡位的汽车传动系载荷谱提取与外推", 《重庆理工大学学报( 自然科学)》 * |
曾敬 等: "汽车试验场在场车辆总数趋势预测", 《汽车工程学报》 * |
王斌杰 等: "基于载荷谱提升转向架构疲劳可靠性研究", 《铁道学报》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111222199A (en) * | 2019-11-13 | 2020-06-02 | 中国汽车工程研究院股份有限公司 | Key index selection and equivalent calculation method during association of user and test field |
CN111222199B (en) * | 2019-11-13 | 2022-06-21 | 中国汽车工程研究院股份有限公司 | Key index selection and equivalent calculation method during association of user and test field |
CN112389436A (en) * | 2020-11-25 | 2021-02-23 | 中汽院智能网联科技有限公司 | Safety automatic driving track-changing planning method based on improved LSTM neural network |
CN116663434A (en) * | 2023-07-31 | 2023-08-29 | 江铃汽车股份有限公司 | Whole vehicle load decomposition method based on LSTM deep neural network |
CN116663434B (en) * | 2023-07-31 | 2023-12-05 | 江铃汽车股份有限公司 | Whole vehicle load decomposition method based on LSTM deep neural network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2022077587A1 (en) | Data prediction method and apparatus, and terminal device | |
CN110163261A (en) | Unbalanced data disaggregated model training method, device, equipment and storage medium | |
CN106022954B (en) | Multiple BP neural network load prediction method based on grey correlation degree | |
CN103853786B (en) | The optimization method and system of database parameter | |
CN110263447A (en) | A kind of loading spectrum Extrapolation method based on shot and long term memory network | |
CN111860982A (en) | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU | |
CN110321603A (en) | A kind of depth calculation model for Fault Diagnosis of Aircraft Engine Gas Path | |
CN109002917A (en) | Total output of grain multidimensional time-series prediction technique based on LSTM neural network | |
CN113743016B (en) | Engine residual life prediction method based on self-encoder and echo state network | |
CN105808689A (en) | Drainage system entity semantic similarity measurement method based on artificial neural network | |
CN113591954A (en) | Filling method of missing time sequence data in industrial system | |
CN114220458B (en) | Voice recognition method and device based on array hydrophone | |
CN103675914B (en) | Use existing ground type earthquake instant analysis system and the method thereof of neural network | |
CN112700326A (en) | Credit default prediction method for optimizing BP neural network based on Grey wolf algorithm | |
CN109101717A (en) | Solid propellant rocket Reliability Prediction Method based on reality with the study of fuzzy data depth integration | |
CN113687433A (en) | Bi-LSTM-based magnetotelluric signal denoising method and system | |
CN111898734A (en) | NMR (nuclear magnetic resonance) relaxation time inversion method based on MLP (Multi-layer linear programming) | |
CN112883522A (en) | Micro-grid dynamic equivalent modeling method based on GRU (generalized regression Unit) recurrent neural network | |
Hu et al. | pRNN: A recurrent neural network based approach for customer churn prediction in telecommunication sector | |
CN116703464A (en) | Electric automobile charging demand modeling method and device, electronic equipment and storage medium | |
CN109146055A (en) | Modified particle swarm optimization method based on orthogonalizing experiments and artificial neural network | |
CN116542701A (en) | Carbon price prediction method and system based on CNN-LSTM combination model | |
CN108647714A (en) | Acquisition methods, terminal device and the medium of negative label weight | |
Regazzoni et al. | A physics-informed multi-fidelity approach for the estimation of differential equations parameters in low-data or large-noise regimes | |
CN110533109A (en) | A kind of storage spraying production monitoring data and characteristic analysis method and its 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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190920 |