CN106548232A - The new network structure and parameter identification method of RBF AR models - Google Patents
The new network structure and parameter identification method of RBF AR models Download PDFInfo
- Publication number
- CN106548232A CN106548232A CN201610992237.1A CN201610992237A CN106548232A CN 106548232 A CN106548232 A CN 106548232A CN 201610992237 A CN201610992237 A CN 201610992237A CN 106548232 A CN106548232 A CN 106548232A
- Authority
- CN
- China
- Prior art keywords
- rbf
- models
- parameter
- parameter identification
- model
- 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
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 the new network structure and parameter identification method of RBF AR models.The method converts thereof into the novel broad sense RBF neural containing two hidden layers according to the design feature of RBF AR (radial basis function network style coefficients AutoRegressive) model itself.For, in the case of low signal-to-noise ratio, SNPOM (structured nonlinear parameter optimization) the method defect low to RBF AR Model Distinguish precision, the present invention establish RBF AR models（Generalized RBF）State of self-organization spatial model, adopt a kind of adaptive particle filter algorithm that core is estimated as with parameter optimization initial value and driving parameter noise statisticses, realize the parameter identification to RBF AR models, modeling accuracy and precision of prediction of the RBF AR models to noisy data can be effectively improved, On-line Estimation and real-time control to RBF AR models are realized, the parameter identification for RBF AR models provides new method.
Description
Technical field
The present invention relates to the new network structure and parameter identification method of a kind of RBF-AR models.
Background technology
Recently, researcher has found neutral net and some simple models couplings that high performance combination die is just obtained
Type.Typically there is the RBF-AR of the propositions such as Peng（X）(RBF network-style coefficients
Autoregressive model with exogenous variable) model, the model for nonlinear system modeling with
And control is there is provided effective solution.The advantage that the existing RBF network functions of RBF-AR models are approached also has State-Dependent certainly
Regression model describes non-thread sexual clorminance.Result of study shows that RBF-AR models are better than some other model on precision of prediction
Or method；In the case where similar precision of prediction is reached, the Center Number that RBF-AR models need is than being used alone RBF networks
Much less.
However, being exactly the estimation of its model parameter using a difficult point of RBF-AR models.In this regard, Peng etc. is proposed
Famous structuring nonlinear parameter optimization (structured nonlinear parameter optimization,
SNPOM) method, the linear ginseng that the method is come in Optimized model with least square method (least square method, LSM)
Number, optimizes nonlinear parameter with LMM (levenberg-marquardt method) methods, and the method can be greatly enhanced receipts
Hold back speed.Later, Gan etc. was easy to be absorbed in locally optimal solution for SNPOM methods, it is proposed that the overall situation-office of RBF-AR (X) model
Portion's method for mixing and optimizing, as SNPOM, it is initial to search for the optimum of SNPOM except for the difference that to combine evolution algorithm for its essence
Value.Although the method can improve parameter identification precision, computation complexity is greatly increased, and is taken huge.Meanwhile, above-mentioned two side
Method to observe data noise statisticses directly cannot recognize, and data by noise jamming it is serious when, identification effect is poor.
Therefore, the method for parameter estimation based on SNPOM has seriously constrained application of the RBF-AR models in actual industrial.
The content of the invention
To overcome the shortcomings of prior art, solve existing RBF-AR model parameter identification methods in the case of noise jamming its
The technical problem of accuracy and poor reliability, it is proposed that the new network structure of RBF-AR models and parameter identification method.This
It is bright according to RBF-AR model own structural characteristics, be reconstructed into the novel broad sense RBF neural containing two hidden layers, and set up
Corresponding state of self-organization spatial model, is adopted one kind and is estimated with parameter optimization initial value and driving parameter noise statisticses
For the adaptive particle filter algorithm of core, the accurate on-line identification to RBF-AR model parameters is realized, RBF-AR is effectively improved
Modeling accuracy and precision of prediction of the model to noisy data, realize the On-line Estimation and real-time control to RBF-AR models, are
The parameter identification of RBF-AR models provides new method.
The present invention solves the technical scheme of above-mentioned technical problem and comprises the following steps：
1) noisy data are modeled using RBF-AR models；
2) according to the RBF-AR models design feature of itself, a kind of novel broad sense RBF nets for containing two hidden layers are converted thereof into
Network；
3) RBF-AR models are set up（Broad sense RBF network）State of self-organization spatial model；
4) impact for primary condition and man made noise's statistical property to parameter identification precision, proposes one kind with parameter most
Optimized initial value and driving parameter noise statisticses are estimated as the adaptive particle filter algorithm of core；
5) above-mentioned adaptive particle filter algorithm is used in the parameter identification of RBF-AR models, and the identification with SNPOM methods
As a result contrasted.
Above-mentioned steps 1) in, the RBF-AR models of noisy data are represented by：
Wherein,To export,It is state vector, is made up of output variable,It is the order of model,
For the center of RBF networks,For scale factor,It is linear power
Weight,Represent vectorial 2 normal form.
Above-mentioned steps 2) in, RBF-AR modelsIndividual RBF networks have an identical center, and each RBF network
Center Number is all, therefore, can be byIndividual RBF networks are converted into a kind of containing the new of two hidden layers and an output node
Broad sense RBF network（See Fig. 1）.Two hidden layers of novel broad sense RBF networks have respectivelyIndividual hidden node（Center）With
Individual hidden node.
Above-mentioned steps 3) in, in order to be recognized to RBF-AR models using adaptive particle filter algorithm, Selection Model
In unknown parameter center and act temporarily as state variable, i.e.,(,;,), model exports the observation as wave filter.Set up following from group
Knit state-space model：
Wherein, original state, stochastic errorWithFor separate Gaussian Profile, i.e.,,,。
Above-mentioned steps 4) in, as parameter identification precision is severely limited by parameter original stateAnd parameter perturbation itemThe impact of variance, exploitation is a kind of to be estimated as the adaptive of core with parameter optimization initial value and parameter perturbation noise statisticses
Answer particle filter algorithm.
Above-mentioned steps 5) in, by step 4) in adaptive particle filter algorithm be applied to the parameter identification of RBF-AR models
In, and contrasted at the aspect such as estimated accuracy and time with SNPOM method for parameter estimation.
Compared with prior art, the present invention it is had the advantage that for：The present invention is according to RBF-AR model self structures
Feature, is reconstructed its network structure, establishes corresponding state of self-organization spatial model, adopts one kind initial with parameter optimization
Value and driving parameter noise statisticses are estimated as the adaptive particle filter algorithm of core, realize to RBF-AR model parameters
Accurate on-line identification.The inventive method solves existing RBF-AR model parameter identification methods（Such as SNPOM methods）In low letter
Make an uproar the technical problem such as its accuracy and poor reliability than in the case of, effectively increase modeling essence of the RBF-AR models to noisy data
Degree and precision of prediction, realize On-line Estimation and real-time control to RBF-AR models, are that the parameter identification of RBF-AR models is carried
New method is supplied, has been that RBF-AR models should provide real-time and validity in actual industrial, with larger economic valency
Value.
Below in conjunction with the accompanying drawings the present invention is made further instructions.
Description of the drawings
Fig. 1 is the new network structure chart of RBF-AR models of the present invention.
Specific embodiment
Fig. 1 is the new network structure chart of RBF-AR models of the present invention, and the network structure substantially belongs to a kind of broad sense
RBF networks, comprising two-layer hidden layer and a linear convergent rate weighting layer, wherein, two hidden layers have respectivelyIndividual hidden node（In
The heart）WithIndividual hidden node.Based on this new RBF network structures, choose the center of network and act temporarily as the state for system
Variable, and corresponding state of self-organization spatial model is set up, one kind is adopted with parameter optimization initial value and driving parameter noise
Statistical property is estimated as the adaptive particle filter algorithm of core, realizes the accurate on-line identification to RBF-AR model parameters.
By the non-linear and non-stationary time series of noise jamming, can be described as with RBF-AR models：
Wherein,To export,It is state vector, is made up of output variable,It is the order of model,
For the center of RBF networks,For scale factor,It is linear power
Weight,Represent vectorial 2 normal form.
In RBF-AR modelsIndividual RBF networks all have identical center, and the Center Number of each network is, because
This, can be by RBF-AR model conversions into a kind of novel broad sense RBF networks for containing two hidden layers and an output node（See Fig. 1）.For profit
RBF-AR models are recognized with adaptive particle filter algorithm, choose state variable
(,;,), model is exported as wave filter
Observation, set up following state of self-organization spatial model：
Wherein, original state, stochastic errorWithFor separate Gaussian Profile, i.e.,,,。
Based on above-mentioned state-space model, exploitation one kind is with parameter optimization initial value and parameter perturbation noise statisticses
Be estimated as the adaptive particle filter algorithm of core, estimate the parameter in RBF-AR models, and with SNPOM method for parameter estimation
Contrasted at the aspect such as estimated accuracy and estimation time.Hereby it is achieved that carrying out real-time online to the parameter of RBF-AR models
Adjustment and On-line Estimation, further realize carrying out real-time control and prediction to RBF-AR models, are RBF-AR models in actual work
Application in industry provides real-time and validity.
Claims (6)
- The new network structure of 1.RBF-AR models and parameter identification method, comprise the steps：1) noisy data are modeled using RBF-AR models；2) according to the RBF-AR models design feature of itself, a kind of novel broad sense RBF nets for containing two hidden layers are converted thereof into Network；3) RBF-AR models are set up（Broad sense RBF network）State of self-organization spatial model；4) impact for primary condition and man made noise's statistical property to parameter identification precision, proposes one kind with parameter most Optimized initial value and driving parameter noise statisticses are estimated as the adaptive particle filter algorithm of core；5) above-mentioned adaptive particle filter algorithm is used in the parameter identification of RBF-AR models, and the identification with SNPOM methods As a result contrasted.
- 2. the new network structure and parameter identification method of the RBF-AR models according to claim l, it is characterised in that：Institute State step 1) in, the RBF-AR models of noisy data are represented by：Wherein,To export,It is state vector, is made up of output variable,It is the order of model, For the center of RBF networks,For scale factor,It is linear power Weight,Represent vectorial 2 normal form；Generally, in above-mentioned RBF-AR models forIndividual different regression variable coefficient, the center of RBF networks may It is different；But in the application of some reality, in order to mitigate calculating intensity, all of RBF networks have allowed identical center, due to The difference of linear weight in RBF networks, the autoregressive coefficient in above-mentioned RBF-AR models are also just different.
- 3. the new network configuration and parameter identification method of the RBF-AR models according to claim l, it is characterised in that：Institute State step 2) in, RBF-AR modelsIndividual RBF networks all have identical center, and the Center Number of each RBF network For；Therefore, can be by RBF-AR model conversions into a kind of novel broad sense RBF networks for containing two hidden layers and an output node（See figure 1）；Two hidden layers of novel broad sense RBF networks have respectivelyIndividual hidden node（Center）WithIndividual hidden node.
- 4. the new network structure and parameter identification method of the RBF-AR models according to claim l, it is characterised in that：Institute State step 3) in, in order to recognize to RBF-AR models using adaptive particle filter algorithm, choose RBF-AR unknown-models Center and act temporarily as state variable, i.e.,(,;,), model exports the observation as wave filter, sets up following state of self-organization space Model：Wherein, original state, stochastic errorWithFor separate Gaussian Profile, i.e.,,,。
- 5. the new network structure and parameter identification method of the RBF-AR models according to claim l, it is characterised in that：Institute State step 4) in, as parameter identification precision is severely limited by parameter original stateAnd parameter perturbation itemThe shadow of variance Ring, develop a kind of adaptive particle filter that core is estimated as with parameter optimization initial value and parameter perturbation noise statisticses Algorithm.
- 6. the new network structure and parameter identification method of the RBF-AR models according to claim l, it is characterised in that：Institute State step 5) in, adaptive particle filter algorithm is applied in the parameter identification of RBF-AR models, and with SNPOM parameter Estimations Method is contrasted at the aspect such as estimated accuracy and time.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610992237.1A CN106548232A (en) | 2016-11-11 | 2016-11-11 | The new network structure and parameter identification method of RBF AR models |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610992237.1A CN106548232A (en) | 2016-11-11 | 2016-11-11 | The new network structure and parameter identification method of RBF AR models |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106548232A true CN106548232A (en) | 2017-03-29 |
Family
ID=58395738
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610992237.1A Pending CN106548232A (en) | 2016-11-11 | 2016-11-11 | The new network structure and parameter identification method of RBF AR models |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106548232A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107608712A (en) * | 2017-10-27 | 2018-01-19 | 北京小米移动软件有限公司 | Method for edition management and device |
| CN108964969A (en) * | 2018-05-07 | 2018-12-07 | 中国铁路总公司 | The high-speed railway signal system method for predicting of hybrid neural networks and AR model |
| CN109059911A (en) * | 2018-07-31 | 2018-12-21 | 太原理工大学 | A kind of GNSS, INS and barometrical data fusion method |
-
2016
- 2016-11-11 CN CN201610992237.1A patent/CN106548232A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107608712A (en) * | 2017-10-27 | 2018-01-19 | 北京小米移动软件有限公司 | Method for edition management and device |
| CN108964969A (en) * | 2018-05-07 | 2018-12-07 | 中国铁路总公司 | The high-speed railway signal system method for predicting of hybrid neural networks and AR model |
| CN108964969B (en) * | 2018-05-07 | 2021-12-07 | 中国铁路总公司 | High-speed railway signal system flow prediction method based on hybrid neural network and AR model |
| CN109059911A (en) * | 2018-07-31 | 2018-12-21 | 太原理工大学 | A kind of GNSS, INS and barometrical data fusion method |
| CN109059911B (en) * | 2018-07-31 | 2021-08-13 | 太原理工大学 | Data fusion method of GNSS, INS and barometer |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhou et al. | Short-term wind power prediction optimized by multi-objective dragonfly algorithm based on variational mode decomposition | |
| CN104239964B (en) | Ultra-short term wind speed forecasting method based on spectral clustering and genetic optimization ExtremeLearningMachine | |
| CN111461463B (en) | Short-term load prediction method, system and equipment based on TCN-BP | |
| CN110309603A (en) | A kind of short-term wind speed forecasting method and system based on wind speed characteristics | |
| Sargolzaei et al. | Neuro–fuzzy modeling tools for estimation of torque in Savonius rotor wind turbine | |
| CN106548232A (en) | The new network structure and parameter identification method of RBF AR models | |
| CN114363195B (en) | Network flow prediction and early warning method for time and frequency spectrum residual convolution network | |
| CN107622305A (en) | Processor and processing method for neutral net | |
| CN104504442A (en) | Neural network optimization method | |
| CN108229750A (en) | A kind of stock yield Forecasting Methodology | |
| CN101853480A (en) | Foreign exchange transaction method based on neural network prediction models | |
| CN106296434A (en) | A kind of Grain Crop Yield Prediction method based on PSO LSSVM algorithm | |
| Chen et al. | [Retracted] Application of Improved LSTM Algorithm in Macroeconomic Forecasting | |
| CN115034422A (en) | Wind power short-term power prediction method and system based on fluctuation identification and error correction | |
| Chen et al. | MULTICRITERION DECISION MAKING FOR FLOOD CONTROL OPERATIONS: THEORY AND APPLICATIONS 1 | |
| CN101900991A (en) | Composite PID (Proportion Integration Differentiation) neural network control method based on nonlinear dynamic factor | |
| CN110569966B (en) | Data processing method and device and electronic equipment | |
| CN117117850A (en) | Short-term electricity load prediction method and system | |
| Lavaei et al. | Dynamic analysis of structures using neural networks | |
| Lin et al. | An RBF‐based model with an information processor for forecasting hourly reservoir inflow during typhoons | |
| CN104570759B (en) | The quick Binomial Trees of control system midpoint orientation problem | |
| Li et al. | Smoothed deep neural networks for marine sensor data prediction | |
| Hettiarachchi et al. | Time series regression and artificial neural network approaches for forecasting unit price of tea at Colombo auction | |
| CN116579479B (en) | Wind farm power ultra-short-term prediction method, system, computer and storage medium | |
| CN113656919B (en) | Asymmetric rotor displacement field reconstruction method based on deep convolutional neural network |
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 | ||
| CB03 | Change of inventor or designer information |
Inventor after: Xi Yanhui Inventor after: Zhao Ting Inventor after: Zhang Xiaodong Inventor after: Peng Hui Inventor after: Xiao Hui Inventor after: Li Zewen Inventor before: Xi Yanhui Inventor before: Zhao Ting Inventor before: Zhang Xiaodong Inventor before: Peng Hui Inventor before: Xiao Hui Inventor before: Li Zewen |
|
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170329 |