CN109359733A - A kind of dynamical system operating status modeling method based on variation self-encoding encoder - Google Patents
A kind of dynamical system operating status modeling method based on variation self-encoding encoder Download PDFInfo
- Publication number
- CN109359733A CN109359733A CN201811222309.XA CN201811222309A CN109359733A CN 109359733 A CN109359733 A CN 109359733A CN 201811222309 A CN201811222309 A CN 201811222309A CN 109359733 A CN109359733 A CN 109359733A
- Authority
- CN
- China
- Prior art keywords
- encoding encoder
- node
- modeling method
- operating status
- variation self
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
- G06N3/065—Analogue means
-
- 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)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Neurology (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
Abstract
The dynamical system operating status modeling method based on variation self-encoding encoder that the invention proposes a kind of.The data of the sensor monitoring on various parts are established into connection using the form of figure based on node insertion, side insertion, coder module, decoder module and Gumbel constraint, the fitting function on side is arrived using encoder combination Gumbel constraint study.Then the system running state of last moment is combined to be inferred to the system running state of subsequent time by the fitting function.The difference of the state and time of day inferred using BP algorithm backpropagation predicts system mode to learn optimal expression function.
Description
Technical field
The present invention relates to internet area, specially a kind of dynamical system operating status modeling based on variation self-encoding encoder
Method.
Background technique
Dynamical system operating status modeling method can establish effective simulation, study between the complication system of multiagent
Complicated dynamic interaction relationship between main body and main body.Solve the effective EVOLUTION ANALYSIS for being directed to complication system state.For traffic
The system that system, crowd behaviour, physical system etc. can be abstracted as graph structure has explains effect well.On this basis,
The support of Evolution Environment can be provided for intensified learning, the support as the technology for having demand for system mode prediction.Most connect
Nearly technology of the invention has:
(1), it stacks self-encoding encoder technology: multiple self-encoding encoders being stacked up one by one to form stacking self-encoding encoder, by
Layer study abstract characteristics, receive input data, are then reconstructed by stacking self-encoding encoder to it, belong to typical nerve net
Network black box is suitable for arbitrary system.
(2), variation self-encoding encoder technology: it is converted to certain distribution observed in data solely using neural network model
Vertical normal distribution, reconvert are gone back.Decoder use can be individually taken out after the completion of VAE model training, it is defeated in a decoder
Enter data needed for normal distribution generates.
(3), depth confidence network technology: limited Boltzmann machine successively being stacked and forms depth confidence network, successively training,
Upper one layer of input is reconstructed using less neuron to learn the rule between implicit data, is come finally by overall situation fine tuning
The case where falling into local optimum is avoided, complication system rule learning is suitable for.
Wherein, it stacks self-encoding encoder and is suitable for any system, but its expression effect is not too in actual use
It is good.Be directed to particular problem, stack self-encoding encoder need specific aim optimization can practical application, however optimization is then needed to it
Structure etc. is modified.Variation self-encoding encoder is advanced optimized to self-encoding encoder, is carried out to its compressive state
Constraint and character separation, effect are better than self-encoding encoder.But its predictive ability is insufficient.It is directed to reconstruct data.Depth is set
The same predictive ability of communication network is insufficient, is directed to the learning rules from data, however its usage scenario be limited to for
In the reasonability judgement of data.
Summary of the invention
In order to solve shortcoming and defect in the prior art, the invention proposes a kind of dynamics based on variation self-encoding encoder
System running state modeling method.Modelling of Dynamic System is divided into two parts, a part of by dynamic relationship between multiagent
It practises, another part predicts system mode based on the relationship learnt.Finally using predict come system mode and
True system mode, which is compared, carries out weight adjustment, rule optimization to neural network.
The technical solution of the present invention is as follows:
A kind of dynamical system operating status modeling method based on variation self-encoding encoder.Based on node insertion, side insertion, compile
Code device module, decoder module, comprising the following steps:
Step (1) is embedded in nervous layer based on the insertion of figure neural network design node and side.
Step (2) designs modeling method proposed in this paper based on node insertion, side insertion and variation self-encoding encoder.
Step (3), using encoder to multidimensional time-series data modeling, learn between dimension (node) and dimension node
Relationship (side distribution function), i.e. relationship between sensor and sensor.
Step (4), side distribution function approximate fits Gumbel distribution characteristics.Addition Gumbel noise is distributed for side.
Step (5), the system mode based on the system mode at side distribution function and upper moment prediction subsequent time.
Step (6), multiple prediction steps are superimposed to form decoder, for specific requirements predicted time length come really
Surely step number is predicted.
Step (7), exclusive use decoder predict system in future state, by input initial system state come pre-
Survey the system state change situation of the following set time.
Beneficial effects of the present invention:
(1) by using figure neural network, it is (linear that modeling expression effectively is carried out to the relationship between different data dimension
Correlation or non-linear dependencies).
(2) study for side distribution function is realized using variation self-encoding encoder structure, controls the shape output and input
Formula, effectively from the pattern information excavated in data between different nodes.
It (3), can be under conditions of known initial state for dynamical system future state by the way that decoder is used alone
It is predicted, for the abstract of real physical system, intensified learning even depth learning model instruction can be effectively applied to
In white silk.
Detailed description of the invention
Fig. 1 is that the present invention is based on the system construction drawings of the dynamical system operating status modeling method of variation self-encoding encoder
Fig. 2 is to need to learn to the relational graph between sensor.
Fig. 3 is the basic unit of the building of figure neural network, and information is transmitted to side from node and is transmitted to node again.
Fig. 4 is decoder internal structural map, and the stacking of multiple Single-step Predictions forms multi-step prediction.
Fig. 5 is that the present invention is based on the overview flow charts of the dynamical system operating status modeling method of variation self-encoding encoder
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
As shown in Figure 1, the system knot of the dynamical system operating status modeling method of the invention based on variation self-encoding encoder
Structure includes three modules: encoder, Gumbel sampling, decoder.
Below with reference to Fig. 1, Fig. 2, Fig. 3 and Fig. 4, to the dynamical system operating status modeling method based on variation self-encoding encoder
Detailed process be described in detail:
Step (1) pre-processes input data, rejects the data dimension that entropy is lower than certain threshold value, then logarithm
According to being cleaned, missing values completion, abnormality value removing, the operation such as normalization improves the quality of data.
Step (2) converts a vector for each attribute of data by node insertion.
Step (3), by while insertion by between node and node while potential relation vector.
Step (4), using connection between layers neural network based by each node and to side and section
Point establishes connection, and the last layer is hidden layer edge-vector.
Step (5) constructs Gumbel distribution function toward addition Gumbel constraint on hidden layer edge-vector, and more accurately simulation is true
Real distribution.
Step (6) is connected based on the side learnt, uses the data at first moment of input data as initial shape
State predicts next system mode, generates prediction data.
Step (7), the error based on prediction data and truthful data are iterated optimization to model.
Dynamical system operating status modeling method based on variation self-encoding encoder of the invention, by using figure neural network
(linear dependence or non-linear dependencies) are modeled to the relationship between main body and main body.Use variation self-encoding encoder knot
Structure controls the form output and input, difference is effectively excavated from data to realize the study for side distribution function
Pattern information between node.The data distribution characteristics of real system are more in line with using Gumbel distribution.Pass through exclusive use
Decoder can predict dynamical system future state under conditions of known initial state, for real physics
System is abstracted, and can be effectively applied in the training of intensified learning even depth learning model.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Within mind and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (1)
1. a kind of dynamical system operating status modeling method based on variation self-encoding encoder.Mainly comprise the steps of:
Step (1) is embedded in nervous layer based on the insertion of figure neural network design node and side.
Step (2) designs modeling method proposed in this paper based on node insertion, side insertion and variation self-encoding encoder.
Step (3), the pass using encoder to multidimensional time-series data modeling, between study dimension (node) and dimension node
It is (side distribution function), i.e. relationship between sensor and sensor.
Step (4), side distribution function approximate fits Gumbel distribution characteristics.
Step (5), the system mode based on the system mode at side distribution function and upper moment prediction subsequent time.
Step (6), multiple prediction steps are superimposed to form decoder.
Step (7), exclusive use decoder predict system in future state.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811222309.XA CN109359733A (en) | 2018-10-19 | 2018-10-19 | A kind of dynamical system operating status modeling method based on variation self-encoding encoder |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811222309.XA CN109359733A (en) | 2018-10-19 | 2018-10-19 | A kind of dynamical system operating status modeling method based on variation self-encoding encoder |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109359733A true CN109359733A (en) | 2019-02-19 |
Family
ID=65345913
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811222309.XA Pending CN109359733A (en) | 2018-10-19 | 2018-10-19 | A kind of dynamical system operating status modeling method based on variation self-encoding encoder |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109359733A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110059894A (en) * | 2019-04-30 | 2019-07-26 | 无锡雪浪数制科技有限公司 | Equipment state assessment method, apparatus, system and storage medium |
| CN111950690A (en) * | 2019-05-15 | 2020-11-17 | 天津科技大学 | Efficient reinforcement learning strategy model with self-adaptive capacity |
-
2018
- 2018-10-19 CN CN201811222309.XA patent/CN109359733A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110059894A (en) * | 2019-04-30 | 2019-07-26 | 无锡雪浪数制科技有限公司 | Equipment state assessment method, apparatus, system and storage medium |
| CN110059894B (en) * | 2019-04-30 | 2020-06-26 | 无锡雪浪数制科技有限公司 | Equipment state evaluation method, device, system and storage medium |
| CN111950690A (en) * | 2019-05-15 | 2020-11-17 | 天津科技大学 | Efficient reinforcement learning strategy model with self-adaptive capacity |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103514366B (en) | Urban air quality concentration monitoring missing data recovering method | |
| CN103914735B (en) | A kind of fault recognition method and system based on Neural Network Self-learning | |
| CN109255505B (en) | Short-term load prediction method of multi-model fusion neural network | |
| CN104270281B (en) | A kind of Web Service QoS combination forecasting methods | |
| CN104951836A (en) | Posting predication system based on nerual network technique | |
| CN108053052B (en) | A kind of oil truck oil and gas leakage speed intelligent monitor system | |
| CN105825271A (en) | Satellite fault diagnosis and prediction method based on evidential reasoning (ER) | |
| CN114919559B (en) | System and method for predicting residual service life of brake system based on digital twin | |
| CN114299723A (en) | Traffic flow prediction method | |
| CN109767006A (en) | A kind of failure prediction system and method based on bayesian belief networks | |
| CN113609955A (en) | Three-phase inverter parameter identification method and system based on deep learning and digital twinning | |
| CN109359733A (en) | A kind of dynamical system operating status modeling method based on variation self-encoding encoder | |
| CN113671421A (en) | Transformer state evaluation and fault early warning method | |
| CN104951871A (en) | Intelligent evaluation method and system for power simulation training | |
| Laite et al. | Behavioral analysis of agent based service channel design using neural networks | |
| CN116822722A (en) | Water level prediction method, system, device, electronic equipment and medium | |
| Shi et al. | The integrated methodology of rough set theory and artificial neural-network for construction project cost prediction | |
| Carpenter | Production-optimization strategy using a hybrid genetic algorithm | |
| Mahdavi et al. | Prediction of oil production with: data mining, neuro-fuzzy and linear Regression | |
| KR20220009058A (en) | Learning method of artificial intelligence part for monitoring system of cable-supported bridge and the artificial intelligence part | |
| Zamani et al. | Seismic response prediction of open-and closed-loop structural control systems using a multi-state-dependent parameter estimation approach | |
| Thair et al. | Reservoir operation by artificial neural network model (Mosul dam-Iraq, as a case study) | |
| CN113965467B (en) | Power communication system reliability assessment method and system based on neural network | |
| Di et al. | Multi-scale internet traffic prediction using wavelet neural network combined model | |
| Bathla et al. | Evaluating product system behavior using soft computing in product structure modeling |
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 | ||
| CB03 | Change of inventor or designer information |
Inventor after: Zhang Yafei Inventor after: Zhang Weishan Inventor before: Zhang Weishan Inventor before: Zhang Yafei |
|
| CB03 | Change of inventor or designer information | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190219 |
|
| WD01 | Invention patent application deemed withdrawn after publication |