CN105955032A - Inverter control method for optimization of extreme learning machine on the basis of bat algorithm - Google Patents
Inverter control method for optimization of extreme learning machine on the basis of bat algorithm Download PDFInfo
- Publication number
- CN105955032A CN105955032A CN201610463606.8A CN201610463606A CN105955032A CN 105955032 A CN105955032 A CN 105955032A CN 201610463606 A CN201610463606 A CN 201610463606A CN 105955032 A CN105955032 A CN 105955032A
- Authority
- CN
- China
- Prior art keywords
- learning machine
- extreme learning
- vespertilio
- algorithm
- control method
- 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
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
Abstract
The invention provides an inverter control method for optimization of an extreme learning machine on the basis of a bat algorithm. The method comprises the following steps: S1, calculating an individual optimal value and a global optimal value by use of the bat algorithm; S2, according to the individual optimal value and the global optimal value, calculating an output weight by use of the extreme learning machine; S3, according to the output weight, controlling a control object transfer function G(S) of an inverter; and S4, outputting a result. According to the invention, online setting and optimization are carried out on PI parameters by use of the extreme learning machine, and correlation parameters of the extreme learning machine are optimized by use of the bat algorithm. The control strategy brought forward by the invention has the advantages of obtaining quite good inverter output waveforms, reducing the adjustment time of a system and improving dynamic response performance of the system.
Description
Technical field
The invention belongs to field of photovoltaic power generation, particularly relate to photovoltaic inverter control system.
Background technology
The control loop of domestic and international more than 95% still uses PID structure.Develop by leaps and bounds in control theory and technology
Today, PID controller is still widely used, and to be primarily due to its control structure simple, and stability is good, can
High by property, it is easy to accomplish etc. advantage, and much senior control is all based on PID control.And PID
Control effect and depend entirely on adjusting and optimizing of pid parameter, therefore, adjusting and optimized algorithm of pid parameter
It is particularly important.In order to realize optimum PID control, pid parameter optimized algorithm has become control both at home and abroad
One focus of theoretical research, owing to simplex method scheduling algorithm operand is big, and is easily absorbed in local optimum,
It is thus desirable to look for a kind of simple and efficient pid parameter optimized algorithm.In recent years, sending out along with computer technology
Exhibition, some new intelligent algorithms have obtained developing rapidly and have extensively applied, and particularly simulated evolutionary algorithm is resonable
Opinion research and applied research aspect are the most active.
At present, the research to pid parameter optimized algorithm is still continuing.PID is joined by useful basic genetic algorithm
Number is optimized, but has very important shortcoming when optimizing some challenges, and basic genetic is calculated
Method convergence rate is slow, easily precocious.In view of ant group algorithm strong robustness, be suitable to parallel processing, realize and operate
Simple advantage, it is proposed that ant group algorithm is applied to pid parameter optimization, but along with optimizing space dimensionality and ginseng
With the increase of search Formica fusca group, the search efficiency of this algorithm is substantially reduced, thus in hyperspace optimization problem
It is necessary to inquire into out a kind of more effectively ant group algorithm model, to improve search efficiency；Population (Particle
Swarm Optimization, PSO) algorithm be by Kennedy and Eberhart doctor in nineteen ninety-five by bird
The inspiration of types of populations behavioral study result, and a kind of based on swarm intelligence the evolutionary computation technique proposed.For
Overall situation version PSO algorithm is easily trapped into this shortcoming of Local Extremum, it is proposed that the local version of real coding
PSO algorithm, the solution using this algorithm search gained is more excellent than overall version algorithm, but speed is slower.
But in actual applications, there are some defects in neutral net self, as parameter be difficult to determine, over-fitting
Deng, have impact on the effect that pid parameter optimizes.Compared with traditional neural network, extreme learning machine (Extreme
Learning Machine, ELM) the input parameter of network need not be adjusted, there is good generalization ability, for
Pid parameter optimization provides a kind of new approach.But in application process, find that extreme learning machine algorithm there is also
Slowly and weights are sensitive, there is the deficiencies such as local minimum to initially connecting convergence rate.Its structure such as Fig. 1 institute
Show.
Summary of the invention
In order to control inverter more accurately, the present invention proposes a kind of based on the adaptive bat of frequency
The inverter control method of bat algorithm optimization extreme learning machine.Use extreme learning machine that PI parameter is carried out online
Adjust and optimize, and using Vespertilio method to optimize the relevant parameter of extreme learning machine.Make control strategy can obtain relatively
Good inverter output waveforms, reduces the adjustment time of system, improves the dynamic response performance of system.
More specifically, the invention provides a kind of inverter control based on Vespertilio algorithm optimization extreme learning machine
Method, comprises the following steps:
S1, Vespertilio algorithm is utilized to calculate individual optimal value and global optimum；
S2, according to individual optimal value and global optimum, limit of utilization learning machine calculates network weight；
S3, according to described network weight, control control object transmission function G (S) of inverter；
S4, output result.
In this application, by bat width algorithm is combined with extreme learning machine algorithm, it is achieved that generalization ability with
Balance between capability of fitting, overcome extreme learning machine algorithm is easily absorbed in local extremum, generalization ability of network performance
Difference and the shortcoming such as overfitting, the algorithm of proposition has and accelerates the response speed to unknown data, receives faster
Hold back speed and the advantage of more high-class prediction accuracy.
As preferably, step S1 includes:
S1.1, all Vespertilio individualities are initialized；
S1.2, calculate optimum input weight and biasing.
In this application, if do not initialized all of Vespertilio individuality, its initial value may result in
There is certain error or mistake occur in the result of output eventually.In order to reduce this error and avoid mistake, at meter
First all Vespertilio individualities should be initialized when calculating optimum input weight and biasing.
Wherein, described Vespertilio individuality refers to be determined the general model of PI parameter by engineering PI computing formula
Enclose.As preferably, the Vespertilio individuality in step S1.1 includes input weight that extreme learning machine calculates and partially
Put.
In this application, the input weight calculated by extreme learning machine and biasing is needed to carry out as Vespertilio individuality
Calculate so that when calculating, extreme learning machine and Vespertilio individual collections have common factor, improve accurately
Property.
As preferably, the Vespertilio algorithm in step S1 is frequency self adaptation Vespertilio algorithm.
There is the deficiency such as discretization, slow, the Premature Convergence of late convergence in standard bat width algorithm.Therefore, at this
In application, use frequency adaptive Vespertilio algorithm.Frequency adaptive Vespertilio algorithm (FSABA algorithm) exists
Search speed, convergence precision, robustness are good, should be readily appreciated that and be easily programmed, and have to break away from and be absorbed in local
The ability of optimal solution, efficiently solves that standard bat width algorithm discretization, late convergence be slow, Premature Convergence
Deng not enough.
As preferably, between step 2 and step 3, also include step SA；
Step SA is: calculate output weight according to optimum input weight and biasing.
In this application, by calculating output weight so that what output result obtained is an optimal value, enters
And control inverter and export with this optimal value.
As preferably, while performing step SA, repeated several times step S1 is to step S3.
In this application, can compare with set-point (i.e. input value) by measuring output valve (i.e. value of feedback),
Having error when, carry out PI regulation, make system be in steady statue.
As preferably, step S2 includes:
S2.1, extreme learning machine train several training samples；
S2.2, calculate the output weight of extreme learning machine.
As preferably, output weight is respectively Proportional coefficient K p and integral coefficient Ki.
In this application, extreme learning machine train several training samples and optimum output weight be respectively than
Example COEFFICIENT K p and integral coefficient Ki, belong to technological means conventional in the art.
Accompanying drawing explanation
Fig. 1 is the structured flowchart that extreme learning machine optimizes PI control parameter.
Fig. 2 is the structured flowchart of a kind of embodiment in the present invention.
Fig. 3 is the control method FB(flow block) of a kind of embodiment in the present invention.
Fig. 4 is the grid-connected waveform of the photovoltaic DC-to-AC converter three-phase current of PI controller.
Fig. 5 be in the present invention a kind of embodiment adjust PI parameter inverter output voltage waveform.
Fig. 6 be in the present invention a kind of embodiment adjust PI parameter inverter output current waveform.
1. Ui: input；2. Ue: the error inputting and exporting；3. Uo: output；4. Kp, Ki:PI parameter；
5. the control object transmission function of G (S): inverter
Detailed description of the invention
Embodiment one
The first embodiment of the present invention provides a kind of inversion based on Vespertilio algorithm optimization extreme learning machine
Device control method, as it is shown on figure 3, comprise the following steps:
S1, Vespertilio algorithm is utilized to calculate individual optimal value and global optimum；
S2, according to individual optimal value and global optimum, limit of utilization learning machine calculates network weight；
S3, according to described network weight, control control object transmission function G (S) of inverter；
S4, output result.
Concrete, step S1 includes:
S1.1, all Vespertilio individualities are initialized；
S1.2, calculate optimum input weight and biasing.
Step S2 includes:
S2.1, extreme learning machine train several training samples；
S2.2, calculate the output weight of extreme learning machine.
Output weight is respectively Proportional coefficient K p and integral coefficient Ki.
In the present embodiment, as in figure 2 it is shown, by bat width algorithm is combined with extreme learning machine algorithm,
Achieving the balance between generalization ability and capability of fitting, overcome extreme learning machine algorithm is easily absorbed in local
The shortcomings such as extreme value, generalization ability of network poor performance and overfitting, the algorithm of proposition has and accelerates unknown data
Response speed, faster convergence rate and the advantage of more high-class prediction accuracy.
Wherein, the Vespertilio individuality in step S1.1 includes input weight and the biasing that extreme learning machine calculates.
Because during Practical Calculation, without the input weight that extreme learning machine is calculated with biasing as bat
Bat individuality calculates so that when calculating, and does not have common factor in extreme learning machine and Vespertilio individual collections,
It is likely to result in final calculation result incorrect.
Concrete,
Assume N number of training sample set (xi,yi), i=1,2 ... N so extreme learning machine can be described as
In formula, ajInput weights for hidden layer node Yu input neuron；βjFor hidden layer node and output god
Output weights through unit；oiFor output valve；bjDeviation for jth hidden layer.
If the ELM that excitation function is g (x) can approach N number of training sample (x with zero-deviationi,yi), then have
H β=T
In formula
Its corresponding step is:
1, by the way of random assignment, get input weights ajWith hidden layer Node Offsetting bj；
2, formula (3) is utilized to solve the output matrix H of hidden layer；
3, show that its network exports weight according to (2) equations.
Wherein β represents the output matrix of network, and T represents the desired output of sample.
Randomly choose input weight in view of extreme learning machine and biasing can make the hidden layer node of extreme learning machine lose
Effect, basic ideas based on Vespertilio Algorithm for Training extreme learning machine are: before extreme learning machine starts training,
Have that optimizing ability is strong in view of Vespertilio algorithm, be difficult to be absorbed in the features such as local optimum, by excellent for Vespertilio group's algorithm
Input weight and the biasing of changing network replace randomly generating in original extreme learning machine, optimization are obtained
Result as extreme learning machine input weight and biasing train neutral net.
Outer voltage Kp in engineering experience, Ki
The computing formula of PI parameter is given above, and it is a kind of approximate solution under the conditions of ignoring a lot, therefore
PI parameter equation is not the most accurately to solve, then in this derivation PI engineering calculation formulas, primarily to
One big probable value of estimation PI parameter.
First pass through above engineering PI computing formula and determine the probable ranges of PI parameter, the parameter input of PI
Weights replace, and network output weight represents that control parameter controls effect to system.
In the present embodiment, preferably a kind of scheme is that the Vespertilio algorithm in step S1 is frequency self adaptation
Vespertilio algorithm.Because there is the deficiency such as discretization, slow, the Premature Convergence of late convergence in standard bat width algorithm.
So in this application, frequency adaptive Vespertilio algorithm is used.Frequency adaptive Vespertilio algorithm (FSABA
Algorithm) in search speed, convergence precision, robustness is good, should be readily appreciated that and be easily programmed, and have and break away from
Be absorbed in the ability of locally optimal solution, efficiently solve standard bat width algorithm discretization, late convergence slow,
The deficiencies such as Premature Convergence.
In the present embodiment, step SA can also be included between step 2 and step 3；
Step SA is: calculate output weight according to optimum input weight and biasing.
In the present embodiment, while performing step SA, repeated several times step S1 is to step S3.And
And output valve (i.e. value of feedback) will be measured accordingly compare with set-point (i.e. input value), there iing error
Time carry out PI regulation, make system be in steady statue.
Fig. 4 is the grid-connected waveform of photovoltaic DC-to-AC converter three-phase current, and as can be seen from the figure current waveform is the most just
String, but a little burr has been can be seen that from the end of current waveform, unsmooth, illustrate in electric current containing one
Quantitative harmonic wave, is unsatisfactory for grid-connected current harmonic standard requirement.Reason is during PI choice of parameters
The determination of system PI parameter is to carry out emulating by continuously attempting to the method for algebraically, because the number of times attempted has
Limit, therefore cannot determine whether to find the PI parameter of optimum, so remaining a need for more preferable algorithm to determine optimum PI
Parameter.
And utilize the voltage of inverter output that the method that the present invention provides controls, current waveform respectively such as Fig. 5 and
Shown in Fig. 6.From analogous diagram it can be seen that the output of inverter only when load changing output voltage fluctuate relatively
Greatly, i.e. being tended towards stability by output waveform after certain regulating time, dynamic responding speed is slow.
The respective embodiments described above are to realize the specific embodiment of the present invention, and those of ordinary skill in the art is permissible
Understand, and in actual applications, can to it, various changes can be made in the form and details, without departing from this
Bright spirit and scope.
Claims (8)
1. an inverter control method based on Vespertilio algorithm optimization extreme learning machine, it is characterised in that comprise the following steps:
S1, Vespertilio algorithm is utilized to calculate individual optimal value and global optimum；
S2, according to individual optimal value and global optimum, limit of utilization learning machine calculates output weight；
S3, according to described output weight, control control object transmission function G (S) of inverter；
S4, output result.
The inverter control method of Vespertilio algorithm optimization extreme learning machine the most according to claim 1, it is characterised in that described step S1 includes:
S1.1, all Vespertilio individualities are initialized；
S1.2, calculate optimum input weight and biasing.
The inverter control method of Vespertilio algorithm optimization extreme learning machine the most according to claim 2, it is characterised in that the Vespertilio individuality in described step S1.1 includes input weight and the biasing that described extreme learning machine calculates.
The inverter control method of Vespertilio algorithm optimization extreme learning machine the most according to claim 1, it is characterised in that the Vespertilio algorithm in described step S1 is frequency self adaptation Vespertilio algorithm.
The inverter control method of Vespertilio algorithm optimization extreme learning machine the most according to claim 1, it is characterised in that also include step SA between described step 2 and step 3；
Described step SA is: calculate output weight according to described optimum input weight and biasing.
The inverter control method of Vespertilio algorithm optimization extreme learning machine the most according to claim 5, it is characterised in that while performing described step SA, described in repeated several times, step S1 is to step S3.
The inverter control method of Vespertilio algorithm optimization extreme learning machine the most according to claim 1, it is characterised in that described step S2 includes:
S2.1, described extreme learning machine train optimum input weight and the biasing of several training samples, the i.e. output of Vespertilio algorithm；
S2.2, calculate the output weight of extreme learning machine.
The inverter control method of Vespertilio algorithm optimization extreme learning machine the most according to claim 1, it is characterised in that the output weight of described extreme learning machine is Proportional coefficient K p and integral coefficient Ki.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610463606.8A CN105955032A (en) | 2016-06-23 | 2016-06-23 | Inverter control method for optimization of extreme learning machine on the basis of bat algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610463606.8A CN105955032A (en) | 2016-06-23 | 2016-06-23 | Inverter control method for optimization of extreme learning machine on the basis of bat algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105955032A true CN105955032A (en) | 2016-09-21 |
Family
ID=56903862
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610463606.8A Pending CN105955032A (en) | 2016-06-23 | 2016-06-23 | Inverter control method for optimization of extreme learning machine on the basis of bat algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105955032A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106644977A (en) * | 2016-10-13 | 2017-05-10 | 天津工业大学 | Spectral variable selection method based on bat algorithm |
| CN108830418A (en) * | 2018-06-14 | 2018-11-16 | 东北电力大学 | A kind of Short-Term Load Forecasting Method |
| CN110868414A (en) * | 2019-11-14 | 2020-03-06 | 北京理工大学 | Industrial control network intrusion detection method and system based on multi-voting technology |
| CN111082660A (en) * | 2020-01-09 | 2020-04-28 | 湖南科技大学 | Output voltage control method of Buck converter based on ELM-PID |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104317195A (en) * | 2014-09-28 | 2015-01-28 | 重庆邮电大学 | Improved extreme learning machine-based nonlinear inverse model control method |
| CN104330968A (en) * | 2014-11-18 | 2015-02-04 | 重庆邮电大学 | Inverse model PID compound control method based on improved support vector regression |
| CN105243428A (en) * | 2015-09-07 | 2016-01-13 | 天津市市政工程设计研究院 | Bus arrival time prediction method through optimizing support vector machine based on bat algorithm |
-
2016
- 2016-06-23 CN CN201610463606.8A patent/CN105955032A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104317195A (en) * | 2014-09-28 | 2015-01-28 | 重庆邮电大学 | Improved extreme learning machine-based nonlinear inverse model control method |
| CN104330968A (en) * | 2014-11-18 | 2015-02-04 | 重庆邮电大学 | Inverse model PID compound control method based on improved support vector regression |
| CN105243428A (en) * | 2015-09-07 | 2016-01-13 | 天津市市政工程设计研究院 | Bus arrival time prediction method through optimizing support vector machine based on bat algorithm |
Non-Patent Citations (2)
| Title |
|---|
| 朱承: "杂草算法优化极限学习机的电力逆变器优化控制", 《电气应用》 * |
| 陈绍炜: "基于蝙蝠算法优化ELM的模拟电路故障诊断研究", 《基于蝙蝠算法优化ELM的模拟电路故障诊断研究》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106644977A (en) * | 2016-10-13 | 2017-05-10 | 天津工业大学 | Spectral variable selection method based on bat algorithm |
| CN108830418A (en) * | 2018-06-14 | 2018-11-16 | 东北电力大学 | A kind of Short-Term Load Forecasting Method |
| CN110868414A (en) * | 2019-11-14 | 2020-03-06 | 北京理工大学 | Industrial control network intrusion detection method and system based on multi-voting technology |
| CN111082660A (en) * | 2020-01-09 | 2020-04-28 | 湖南科技大学 | Output voltage control method of Buck converter based on ELM-PID |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103591637B (en) | A kind of central heating secondary network runing adjustment method | |
| CN105955032A (en) | Inverter control method for optimization of extreme learning machine on the basis of bat algorithm | |
| Niu et al. | Model turbine heat rate by fast learning network with tuning based on ameliorated krill herd algorithm | |
| CN101709869A (en) | Hybrid control method for superheat steam temperature system of coal-fired boiler | |
| CN105870942A (en) | Primary frequency regulation additional learning control method based on approximate dynamic programming algorithm | |
| CN110501909A (en) | The Fuzzy Predictive Control method of enhancing robust property based on disturbance observer | |
| CN101709867A (en) | Hybrid control method for drum water level system of coal-fired boiler | |
| CN101709863A (en) | Hybrid control method for furnace pressure system of coal-fired boiler | |
| CN104167968B (en) | A kind of Vector Control System of Induction Motor method | |
| CN104616072A (en) | Method for improving concentration of glutamic acid fermented product based on interval optimization | |
| Tian et al. | Application of improved whale optimization algorithm in parameter identification of hydraulic turbine at no-load | |
| CN115986839A (en) | Intelligent scheduling method and system for wind-water-fire comprehensive energy system | |
| Yin et al. | Expandable deep width learning for voltage control of three-state energy model based smart grids containing flexible energy sources | |
| Xu et al. | Design of type-2 fuzzy fractional-order proportional-integral-derivative controller and multi-objective parameter optimization under load reduction condition of the pumped storage unit | |
| CN113991760A (en) | Direct-current locking multi-resource emergency control optimization method for alternating-current and direct-current series-parallel receiving-end power grid | |
| CN107402515A (en) | A kind of method of optimized synchronization generator excited system pid parameter | |
| Patel et al. | Optimized cascade fractional‐order 3DOF‐controller for frequency regulation of a hybrid power system using marine predators algorithm | |
| CN105207220B (en) | A kind of tapping voltage regulation and control method based on progressive learning | |
| CN116300422A (en) | Hydropower unit control optimization method and terminal | |
| Wang et al. | An improved particle swarm optimization algorithm based on fuzzy PID control | |
| CN105141208A (en) | Conversion method for proportion-integration-differentiation (PID) correction link in generator excitation system model | |
| CN109244510A (en) | A kind of solid oxide fuel cell control method based on Unmarried pregnancy compensation | |
| CN115528750A (en) | Data model hybrid drive unit combination method for power grid safety and stability | |
| TW202009628A (en) | Maximum power tracking method for solar cell and system thereof suitable for real-time online environment | |
| CN105353614A (en) | Control method for speed-variable frequency-constant double-feed bidirectional tidal power generator system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160921 |
|
| RJ01 | Rejection of invention patent application after publication |