CN108828961A - Active Power Filter-APF sliding-mode control based on metacognition fuzzy neural network - Google Patents
Active Power Filter-APF sliding-mode control based on metacognition fuzzy neural network Download PDFInfo
- Publication number
- CN108828961A CN108828961A CN201811086430.4A CN201811086430A CN108828961A CN 108828961 A CN108828961 A CN 108828961A CN 201811086430 A CN201811086430 A CN 201811086430A CN 108828961 A CN108828961 A CN 108828961A
- Authority
- CN
- China
- Prior art keywords
- neural network
- metacognition
- layer
- apf
- active power
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
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
- G05B13/042—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 in which a parameter or coefficient is automatically adjusted to optimise the performance
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses a kind of Active Power Filter-APF sliding-mode controls based on metacognition fuzzy neural network, including following procedure：S1, establishes Active Power Filter-APF kinetics equation, S2, and design control law is：WhereinFor the estimated value for approaching f (x) acquisition using metacognition fuzzy neural network.Present invention introduces metacognition methods to carry out on-line tuning to structure of fuzzy neural network, it is updated according to the increase of tracking error design rule, parameter and regular Pruning algorithm dynamic adjusts structure of fuzzy neural network, can be improved compensation current tracking performance and system robustness of the active power filter system there are Parameter Perturbation and external interference.
Description
Technical field
The present invention relates to active power filter control technique fields, and in particular to one kind is based on metacognition fuzznet
The Active Power Filter-APF sliding-mode control of network.
Background technique
With the large-scale popularization and application of modern power electronics technology, various power electronics devices are more and more, harmonic wave, nothing
Function, imbalance etc. produce very big influence to electric system, have seriously affected power supply quality, reduce generating equipment, electricity consumption
The working performance and service life of equipment, or even jeopardize the safety of electric system.The side of additional filter is mainly used at present
Formula is administered, and filter is divided into passive filter and two kinds of active filter.Since passive filter presence can only compensate spy
The defects of determining harmonic wave, so being concentrated mainly on active filter to the Controlling research of electric energy problem now.Active filter can be right
Harmonic wave that frequency and amplitude all change carries out tracing compensation, can not only compensate each harmonic, may also suppress flickering, compensating reactive power,
Filtering characteristic is not influenced by system impedance simultaneously, and therefore, it has become the extensive hot spots studied and pay close attention to.
Have at present and various advanced control methods are applied in the control of Active Power Filter-APF, typically has adaptive
Control and sliding-mode control.On the one hand these advanced methods compensate for modeling error, on the other hand realize to active electric power
The compensation current follow-up control of filter.But self adaptive control is very low to the robustness of external disturbance, and system is easily made to become unstable
It is fixed.
It can be seen that above-mentioned existing Active Power Filter-APF is in use, it is clear that there are still there is inconvenient and defect, and urgently
Wait be further improved.In order to solve the problems, such as that existing Active Power Filter-APF exists in use, relevant manufactures are there's no one who doesn't or isn't
Painstakingly seek solution, but has no that applicable design is developed completion always for a long time.
Summary of the invention
It is an object of the invention to overcome deficiency in the prior art, propose a kind of based on metacognition fuzzy neural network
Active Power Filter-APF sliding-mode control, can be improved active power filter system that there are Parameter Perturbation and the external worlds is dry
Compensation current tracking performance and system robustness in the case of disturbing.
In order to solve the above technical problems, the present invention provides a kind of, the active electric power based on metacognition fuzzy neural network is filtered
Wave device sliding-mode control, characterized in that including following procedure：
S1, establishing Active Power Filter-APF kinetics equation is：
Wherein, x i1、i2Or i3, f (x) isOr
B isOrhkUncertain for the lump of bounded, u indicates control law, v1,v2,v3For points of common connection
Locate voltage, i1,i2,i3Compensation electric current, v are exported for filterdcFor DC capacitor voltage, LcFor Inductor, RcFor exchange
Side equivalent resistance；Lc1、Rc1Respectively system parameters Lc、RcNominal value；
S2 establishes sliding mode controller, design control law, to control Active Power Filter-APF；
Sliding-mode surface is designed as：
If instruction current is yd, then error be
E=x-yd (15)
Wherein, CsIt is normal number；
Design control law is：
Wherein,For the estimated value of unknown portions f (x), K is normal number, and sgn (S) is sign function,
To approach the estimated value that f (x) is obtained using metacognition fuzzy neural network, It is real-time
Weight,For the output of rules layer, T indicates transposition.
Preferably, the process for establishing Active Power Filter-APF kinetics equation is：
Mathematical model of the Active Power Filter-APF under abc coordinate system：
Wherein：；C is the capacitance of DC bus capacitor device, and t is time, dnkSwitch state function, n=0,1,2 ...,
7, k=1,2,3；
Consider that the mathematical model of Active Power Filter-APF when unknown external interference and Parameter Perturbation is represented by：
Wherein：G=[g1 g2 g3 g4]TFor extraneous unknown disturbance vector, Lc1、Rc1And C1Respectively system parameter is nominal
Value (this value is known), Δ L, Δ R and Δ C are respectively the variable quantity of parameter；
Formula (10) it is rewritable at：
Wherein,For design current control system, examine
Preceding 3 equations of worry formula (11)：
Above formula derivation is obtained：
For the sake of simplicity, it is denoted as following form：
Wherein, x i1、i2Or i3, f (x) isOr
B isOrhkFor Or
hkIt is uncertain for the lump of bounded, that is, there is unknown constant H>0, so that | hk|≤H, u indicate control law.
Preferably, switch state function dnk, it is defined as：
N is switching mode in above formula, and k is the number of phases,
Switch function ck, indicate the working condition of IGBT switch in Active Power Filter-APF, be defined as：
Wherein, k=1,2,3.
Preferably, metacognition fuzzy neural network uses four-layer network network structure, and each layer is respectively：Input layer, degree of membership letter
Several layers, rules layer and output layer.
Preferably, in metacognition fuzzy neural network：
First layer：The input element of input layerFor the element in tracing deviation vector e, i=1 ..., n；
The second layer：Subordinating degree function layer using Gauss type function as membership function,With
It is the center vector and sound stage width of the membership function of i-th of input variable, j-th of fuzzy set respectively,Indicate degree of membership letter
Number, i.e.,
Convenient for calculating, using NpiIt indicates the independent number of subordinating degree function, and defines adaptive parameter vector b and c points
The set of Gaussian subordinating degree function all sound stage widths and center vector is not represented, then
I.e.
WhereinRepresent the total number of subordinating degree function；
Third layer：Rules layer uses Fuzzy inferential engine, and the output of each node is all input signals of the node
Product, i.e.,
In formula, φk(k=1 ..., l) indicates k-th of output of rules layer,It represents between blurring layer and rules layer
Connection weight matrix, be taken as unit vector herein, l is the total number of rules layer；
4th layer：The node on behalf output variable of output layer.Each node y of output layero(o=1 ..., No) output
For the sum of all input signals of the node；Indicate the connection weight matrix between rules layer and output layer, then
Further, define metacognition fuzzy neural network controller output be：
Y=[y1 y2 … yl]=WTΦ=W1φ1+W2φ2+...+Wlφl。
Particularly, metacognition fuzzy neural network further includes that data study and data delete two kinds of self-control strategies：
It is data learning strategy first, the data learning process of metacognition fuzzy neural network is related to closest to current defeated
The online evolution and parameter for entering the rule of data update.
Fuzzy rule is gradually determined according to the following conditions, i.e., | | ei||>EaAnd ψ<Es.Wherein ψ is spherical surface potential energy, table
The novelty for showing input data is provided by following formula：
Wherein, EsAnd EaBe novelty and addition threshold value,EaIt can be according to following formula
Self-control is carried out,
Ea=δ Ea+(1-δ)||ei|| (20)
Wherein | | ei| | indicate tracking error.δ indicates slope factor；
When needing to be added new fuzzy rule ((l+1) rule), parameter initialization is,
Wherein κ is the Overlapping parameters of previously given fuzzy rule；
When | | ei||>ElWhen, adjust parameter of regularity.Threshold value ElIt is also that self-control is carried out according to tracking error, by as follows
Formula provides,
El=δ El+(1-δ)||ei|| (22)
In learning process, a certain rule may reduce the contribution degree of output；It in this case, should be from rule
Inessential rule is deleted in library to calculate to avoid excessive；The contribution degree of q rule is provided by following formula：
βq=Φq|eiWq|, i=1 ..., n (23)
Wherein,N represents the dimension of input.
If q rule is lower than threshold value E to the contribution degree of inputpThen delete this rule；
Followed by data deletion policies, particularly, current tracking error and last neural network iterative process
When error is close, without updating network parameter, overlearning is avoided, computation burden is reduced.
Compared with prior art, the beneficial effects obtained by the present invention are as follows being：The present invention is to reduce algorithm complexity to introduce member
Cognitive approach carries out on-line tuning to structure of fuzzy neural network, is increased according to tracking error design rule, parameter updates and rule
Then Pruning algorithm dynamic adjustment structure of fuzzy neural network, avoids overlearning and improves approximation capability.Meanwhile current tracking misses
When difference and close last neural network iterative process error, without updating network parameter, computation burden is reduced, is being had for algorithm
The practice of active power filter provides technical support.The method of the present invention, which can be improved active power filter system, to be existed
Compensation current tracking performance and system robustness in the case of Parameter Perturbation and external interference.
Detailed description of the invention
Fig. 1 is the structure chart of existing Active Power Filter-APF；
Fig. 2 is metacognition structure of fuzzy neural network figure；
Fig. 3 is the schematic diagram of control system of the present invention；
Fig. 4 is the MATLAB simulation result that Active Power Filter-APF uses the method for the present invention to be controlled.
Specific embodiment
The invention will be further described below in conjunction with the accompanying drawings.Following embodiment is only used for clearly illustrating the present invention
Technical solution, and not intended to limit the protection scope of the present invention.
In power grid application, the application of three-phase alternating current occupies the majority, so the present invention is mainly studied for three-phase three-wire system
The case where system.If Fig. 1 is the basic circuit topology diagram of existing phase three-wire three shunt voltage type Active Power Filter-APF,
vs1,vs2,vs3It is network voltage, is1,is2,is3It is source current, iL1,iL2,iL3It is load current, v1,v2,v3It is commonly connected
Voltage at point, i1,i2,i3Compensation electric current is exported for filter, C is DC bus capacitor, vdcFor DC capacitor voltage, idcIt is straight
Flow lateral capacitance electric current, LcFor Inductor, RcTo exchange side equivalent resistance.
Its working principle is that detection load current and network voltage are input to instruction current computing unit first, pass through finger
It enables current calculation unit that instruction current is calculated and is input to current control system, this instruction current and harmonic wave size of current are equal
Contrary, current control system calculates corresponding control according to designed control strategy according to compensation electric current and instruction current
Power processed carries out PWM modulation to control force and generates pwm signal, and IGBT in Active Power Filter-APF is controlled using pwm signal and is switched
The on-off of (i.e. switch S1-S6) can disappear compensation electric current injection power grid to change the size of filter output compensation electric current
Except harmonic wave.
And control method of the invention is the control strategy studied in current control system.
Data study in present invention combination fuzzy neural network and metacognitive strategy designs one kind with data deleting mechanism
Metacognition structure of fuzzy neural network, wherein data deleting mechanism is that current tracking error and upper sampling period error are close
When, without updating network parameter, overlearning is avoided, computation burden is reduced, and is increased according to tracking error design rule, ginseng
Number updates, rule is deleted and adjusts structure of fuzzy neural network with data deletion algorithm dynamic, increases threshold value when tracking error is greater than
When, new fuzzy rule is added, responds rapidly to the complex working conditions such as load sudden change；When tracking error is greater than self-adjusting threshold value, only
Distance input variable sound stage width regular recently, center vector, weight are updated, online amount of calculation is saved, is based on Lyapunov
The stability of Theory of Stability design parameter adaptive law realization closed-loop system；When contribution degree of a certain rule to output is smaller
When, this inessential rule is removed from rule base, realizes the automatic adjusument of controller architecture and parameter, is had better
Approximation effect.
Active Power Filter-APF sliding-mode control based on metacognition fuzzy neural network of the invention, as shown in figure 3,
Include the following steps：
Step S1 establishes Active Power Filter-APF kinetics equation.
According to the available following formula of Circuit theory and Kirchhoff's second law：
Wherein, vmMFor M point to 1,2,3 points of voltage (1,2,3 point be respectively Active Power Filter-APF abc three-phase output
Point), m=1,2,3, vMNFor the voltage of M point to N point, it is assumed that v1+v2+v3=0, i1+i2+i3=0, it is available
Insertion switch function ck, indicate the working condition of IGBT, be defined as：
Wherein, k=1,2,3.
In turn, available vkM=ckvdc。
So the kinetics equation (1) of Active Power Filter-APF can be rewritten as
It is further introduced into switch state function dnk, it is defined as：
Above formula shows dnkValue depend on switching mode n and number of phases k, in other words, depend on Active Power Filter-APF
Switch function ck.This is also interactional embodiment between three-phase current.
Switching mode n refers to eight kinds of switching modes of 6 IGBT, and n=0,1,2 ..., 7, IGBT conducting are denoted as 1, disconnect
It is denoted as 0, then one shares 8 kinds of switching modes.
1 switching mode of table
Meanwhile eight kinds of switching modes n=0,1,2 ..., 7, available c based on formula (5) and IGBTkAnd dnkBetween
Transformational relation be：
On the other hand, available in DC side：
It utilizesFormula (7) can be rewritten into：
Utilize i1+i2+i3=0, it is available：
Therefore, mathematical model of the Active Power Filter-APF under abc coordinate system：
Wherein：v1、v2、v3It is the voltage of points of common connection, i1、i2、i3It is the compensation electric current of Active Power Filter-APF, C is
The capacitance of DC bus capacitor device, vdcIt is the voltage of capacitor C, LcIt is Inductor, RcIt is equivalent resistance, t is time, dnk
It is switch state function, n=0,1,2 ..., 7, k=1,2,3.
Active Power Filter-APF not only will receive the influence of extraneous various unknown disturbances in actual operation, and use
The injection system elements such as inductance and filter capacitor can gradually aging, i.e. parameter presence perturbation in the process.In order to improve system external
The robustness of boundary disturbance and Parameter Perturbation, it is necessary to these influences are considered in system model.
Therefore consider that the mathematical model of Active Power Filter-APF when unknown external interference and Parameter Perturbation is represented by：
Wherein：G=[g1 g2 g3 g4]TFor extraneous unknown disturbance vector, Lc1、Rc1And C1Respectively system parameter is nominal
Value (this value is known), Δ L, Δ R and Δ C are respectively the variable quantity of parameter.
For the ease of analysis, formula (10) it is rewritable at：
Wherein,
For design current control system, preceding 3 equations of formula (11) are considered：
Because the second order that back needs to use electric current is led, above formula derivation is obtained：
It can be seen that there is no phases between ' 1 ', ' 2 ', ' 3 ' three-phases although this is a multi-input multi-output system
Mutual coupling item, so this multivariable Control can be turned to three single argument controls in the design process of current control system,
And in the symmetrical situation of parameter, it more can simplify as a single argument control problem.
For the sake of simplicity, it is denoted as following form：
Wherein, x i1、i2Or i3, f (x) isOr
B isOrhkFor Or
hkIt is uncertain for the lump of bounded, that is, there is unknown constant H>0, so that | hk|≤H, u indicate control law.
Step S2 establishes the Active Power Filter-APF sliding mode controller based on metacognition fuzzy neural network, design control
Rule, and utilize the unknown portions of metacognition fuzzy neural network approximation system.
Sliding-mode surface is designed as：
If instruction current is yd, then error is e=x-yd (15)
Wherein, CsIt is normal number.
Sliding-mode surface S derivation is obtained：
It enablesObtain Equivalent control law：
It is defined as follows liapunov function：To its derivation and (17) are brought into and can be obtained
Known toTherefore, system is stable.
Although the controller designed in (17) can guarantee that system is stablized, will not under normal conditions to save cost
Voltage sensor is set and measures v1、v2、v3, thus f (x) be it is unknown, optimal controller cannot achieve.
In view of neural network approaches the ability of arbitrary function, can be used metacognition neural network come to unknown portions into
Row estimation, and controller design is carried out using the estimated value of f (x).
The structure of metacognition fuzzy neural network is as shown in Figure 2.Wherein, eiIt is the input of metacognition fuzzy neural network, Y
It is the output of metacognition fuzzy neural network, W=[W1,W2...Wl]TFor weight vectors,For metacognition fuzzy neural network
Real-time weight, it is online to constantly update；Φ=[φ1,φ2,...,φl]TIt is the output of rules layer, l is the sum of fuzzy rule
Mesh.
Using four-layer network network structure, each layer is respectively：
First layer：Input layer
Each node of input layer is directly connect with each component of input quantity, and input quantity is passed to next layer.In Fig. 2,
e1..ei..enThe input for representing metacognition fuzzy neural network has n input, and i-th of input is ei, ei(i=1 ..., n) be
Element in tracing deviation vector.
The second layer：Subordinating degree function layer
Using Gauss type function as membership function,WithIt is i-th of input respectively
The center vector and sound stage width of the membership function of j-th of fuzzy set of variable,Indicate subordinating degree function, i.e.,
Convenient for calculating, using NpiIt indicates the independent number of subordinating degree function, and defines adaptive parameter vector b and c points
The set of Gaussian subordinating degree function all sound stage widths and center vector is not represented, i.e.,
WhereinRepresent the total number of subordinating degree function.
Third layer：Rules layer
The layer uses Fuzzy inferential engine, and the output of each node is the product of all input signals of the node, i.e.,
In formula, φk(k=1 ..., l) indicates k-th of output of rules layer,It represents between blurring layer and rules layer
Connection weight matrix, be taken as unit vector herein, l is the total number of rules layer.
4th layer：Output layer.
The node on behalf output variable of output layer.The output of each node yo (o=1 ..., No) of output layer is the section
The sum of all input signals of point；Indicate the connection weight matrix between rules layer and output layer, then
Further, define metacognition fuzzy neural network controller output be：
Y=[y1 y2 … yl]=WTΦ=W1φ1+W2φ2+...+Wlφl。
Particularly, it is proposed that metacognition fuzzy neural network model consider data study and data delete two kinds self adjust
Section strategy, helps to be effectively carried out real-time control task.
It is data learning strategy first.The data learning process of metacognition fuzzy neural network is related to closest to current defeated
The online evolution and parameter for entering the rule of data update.
Fuzzy rule is gradually determined according to the following conditions, i.e., | | ei||>EaAnd ψ<Es.Wherein ψ is spherical surface potential energy, table
The novelty for showing input data is provided by following formula：
Wherein, EsAnd EaBe novelty and addition threshold value,EaIt can be according to following formula
Self-control is carried out,
Ea=δ Ea+(1-δ)||ei|| (20)
Wherein | | ei| | indicate tracking error.δ indicates slope factor.
When needing to be added new fuzzy rule ((l+1) rule), parameter initialization is,
Wherein κ is the Overlapping parameters of previously given fuzzy rule.
When | | ei||>ElWhen, adjust parameter of regularity.Threshold value ElIt is also that self-control is carried out according to tracking error, by as follows
Formula provides,
El=δ El+(1-δ)||ei|| (22)
In learning process, a certain rule may reduce the contribution degree of output.It in this case, should be from rule
Inessential rule is deleted in library to calculate to avoid excessive.The contribution degree of q rule is provided by following formula：
βq=Φq|eiWq|, i=1 ..., n (23)
Wherein,N represents the dimension of input.
If q rule is lower than threshold value E to the contribution degree of inputpThen delete this rule.
Followed by data deletion policies.Particularly, current tracking error and last neural network iterative process
When error is close, without updating network parameter, overlearning is avoided, computation burden is reduced.
Active Power Filter-APF System with Sliding Mode Controller structure chart such as Fig. 3 based on metacognition fuzzy neural network of the invention
It is shown.
F (x) is the unknown dynamic characteristic of system, and metacognition fuzzy neural network is used to approach unknown function f (x), unknown letter
The network reconfiguration error function that number f (x) can be parameterized as ideal metacognition fuzzy neural network an output and bounded：
Ω=W*TΦ*+ε0, wherein W*Indicate ideal network weight, Φ*Indicate the ideal output of rules layer, ε0For neural network reconstruct
Error, T represent transposition,.Under ideal network weight, neural network reconstructed error is minimum, and uniform bound, | ε0|≤ε0E, ε0E
For the positive number of very little.Therefore, design controller is：
Wherein,For the reality output of metacognition fuzzy neural network,For real-time weight,For rule
The output of layer,For rules layer output error, it is denoted asT indicates that transposition, H are the positive coefficient of definition.
Step S3, is based on lyapunov function theory, and design adaptive law verifies the stability of system；
It is defined as follows liapunov function：
Wherein, c is center vector, and b is sound stage width, η1, η2, η3It is normal number, indicates learning rate；
For the error of estimative weight vector,
Work as system convergence, W will stay in that a constant.Accordingly, there existSo
Note
Obviously, V2It is the scalar of positive definite, is substituted into its derivation and by control law (24)
Wherein,
It will using Taylor series expansionBe converted to following form：
Wherein,B* and c* is the optimal value of b and c,WithIt is the estimated value of b* and c*, Onv
It is high-order term,
It willTaylor expansion substitute into (27)：
It enables?
It enables?
It enables?
Adaptive law (29)~(31) are substituted into (28) to obtain：
Assuming that ε, OhO is respectively present upper bound εE,OE, i.e., | ε |≤εE, | Oho|≤OEAs long as therefore making：K≥H+εE+OE,
It can guarantee：
Negative semidefinite demonstrate the Active Power Filter-APF System with Sliding Mode Controller based on metacognition fuzzy neural network
Stability.
In order to verify the feasibility of above-mentioned theory, emulation experiment has been carried out at Matlab.Simulation results show set
Count the effect of controller.
Simulation parameter is chosen as follows：
2 simulation parameter of table
Each parameter of controller is chosen as follows：Cs=1300, K=1000, Ea=5, Es=0.02, El=0, Ep=0.2, η1=
1, η2=0.01, η3=0.01.
In a t=0.1s identical nonlinear load in parallel again, remove newly-increased load in t=0.2s, such as Fig. 4 institute
Show, i in figureLRepresent load current, isRepresent power network current, icrefRepresent instruction current, icCompensation electric current is represented, error is represented
Instruction current and the deviation for compensating electric current.As can be known from Fig. 4, load sudden change, using control system designed by the present invention, power grid
Electric current only needs half period that can reach stable state, it was confirmed that designed control method has good dynamic effect.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, without departing from the technical principles of the invention, several improvements and modifications, these improvements and modifications can also be made
Also it should be regarded as protection scope of the present invention.
Claims (7)
1. the Active Power Filter-APF sliding-mode control based on metacognition fuzzy neural network, characterized in that including following mistake
Journey：
S1, establishing Active Power Filter-APF kinetics equation is：
Wherein, x i1、i2Or i3, f (x) isOrB
ForOrhkUncertain for the lump of bounded, u indicates control law, v1,v2,v3At points of common connection
Voltage, i1,i2,i3Compensation electric current, v are exported for filterdcFor DC capacitor voltage, LcFor Inductor, RcTo exchange side
Equivalent resistance；Lc1、Rc1Respectively system parameters Lc、RcNominal value；
S2 establishes sliding mode controller, design control law, to control Active Power Filter-APF；
Sliding-mode surface is designed as：
If instruction current is yd, then error be
E=x-yd (15)
Wherein, CsIt is normal number；
Design control law is：
Wherein,For the estimated value of unknown portions f (x), K is normal number, and sgn (S) is sign function,
To approach the estimated value that f (x) is obtained using metacognition fuzzy neural network, For metacognition mould
The real-time weight of neural network is pasted,For the output of rules layer in metacognition fuzzy neural network, T indicates transposition.
2. the Active Power Filter-APF sliding-mode control according to claim 1 based on metacognition fuzzy neural network,
It is characterized in that the process for establishing Active Power Filter-APF kinetics equation is：
Mathematical model of the Active Power Filter-APF under abc coordinate system：
Wherein：C is the capacitance of DC bus capacitor device, and t is time, dnkIt is switch state function, n=0,1,2 ..., 7, k=
1,2,3；
Consider that the mathematical model of Active Power Filter-APF when unknown external interference and Parameter Perturbation is represented by：
Wherein：G=[g1 g2 g3 g4]TFor extraneous unknown disturbance vector, Lc1、Rc1And C1The respectively nominal value of system parameter, Δ
L, Δ R and Δ C is respectively the variable quantity of parameter；
Formula (10) it is rewritable at：
Wherein,
For design current control system, preceding 3 equations of formula (11) are considered：
Above formula derivation is obtained：
It is denoted as following form：
Wherein, x i1、i2Or i3, f (x) isOrB
ForOrhkFor Or
hkUncertain for the lump of bounded, u indicates control law.
3. the Active Power Filter-APF sliding-mode control according to claim 2 based on metacognition fuzzy neural network,
It is characterized in that switch state function dnk, it is defined as：
N is switching mode in above formula, and k is the number of phases,
Switch function ck, indicate the working condition of IGBT switch in Active Power Filter-APF, be defined as：
Wherein, k=1,2,3.
4. the Active Power Filter-APF sliding-mode control according to claim 1 based on metacognition fuzzy neural network,
It is characterized in that metacognition fuzzy neural network uses four-layer network network structure, each layer is respectively：Input layer, subordinating degree function layer, rule
Then layer and output layer.
5. the Active Power Filter-APF sliding-mode control according to claim 4 based on metacognition fuzzy neural network,
It is characterized in that in metacognition fuzzy neural network：
First layer：The input element of input layerFor the element in tracing deviation vector e, i=1 ..., n；
The second layer：Subordinating degree function layer using Gauss type function as membership function,WithIt is i-th of input variable jth respectively
The center vector and sound stage width of the membership function of a fuzzy set,Indicate subordinating degree function, i.e.,
Wherein, i=1 ..., n, j=1 ..., Npi, using NpiIt indicates the independent number of subordinating degree function, and defines adaptive
Parameter vector b and c respectively represent the set of Gaussian subordinating degree function all sound stage widths and center vector, then
WhereinRepresent the total number of subordinating degree function；
Third layer：Rules layer uses Fuzzy inferential engine, and the output of each node is multiplying for all input signals of the node
Product, i.e.,
In formula, φk(k=1 ..., l) indicates k-th of output of rules layer,Represent the company between blurring layer and rules layer
Weight matrix is connect, is taken as unit vector herein, l is the total number of rules layer；
4th layer：Each node y of output layero(o=1 ..., No) output be all input signals of the node sum；Table
Show the connection weight matrix between rules layer and output layer, then
Further, define metacognition fuzzy neural network controller output be：
Y=[y1 y2 … yl]=WTΦ=W1φ1+W2φ2+...+Wlφl。
6. the Active Power Filter-APF sliding-mode control according to claim 1 based on metacognition fuzzy neural network,
It is characterized in that metacognition fuzzy neural network further includes that data study and data delete two kinds of self-control strategies.
7. the Active Power Filter-APF sliding-mode control according to claim 6 based on metacognition fuzzy neural network,
It is characterized in that data deletion policies are specifically, current tracking error and last neural network iterative process error are close
When, without updating network parameter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811086430.4A CN108828961B (en) | 2018-09-18 | 2018-09-18 | Sliding mode control method of active power filter based on element cognition fuzzy neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811086430.4A CN108828961B (en) | 2018-09-18 | 2018-09-18 | Sliding mode control method of active power filter based on element cognition fuzzy neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108828961A true CN108828961A (en) | 2018-11-16 |
| CN108828961B CN108828961B (en) | 2021-07-30 |
Family
ID=64149353
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811086430.4A Active CN108828961B (en) | 2018-09-18 | 2018-09-18 | Sliding mode control method of active power filter based on element cognition fuzzy neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108828961B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109709802A (en) * | 2018-12-14 | 2019-05-03 | 江南大学 | The control method of active electronic ladder circuit based on iterative learning control |
| CN112271729A (en) * | 2020-10-15 | 2021-01-26 | 河海大学常州校区 | Long-time memory fuzzy neural network sliding mode control method of active power filter |
| CN112947066A (en) * | 2021-01-26 | 2021-06-11 | 苏州连恺自动化有限公司 | Manipulator improved finite time inversion control method |
| CN114336742A (en) * | 2021-12-22 | 2022-04-12 | 浙江大学 | Solar photovoltaic controller based on self-adaptive neural network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104730921A (en) * | 2015-01-13 | 2015-06-24 | 河海大学常州校区 | Method for controlling fuzzy neural network of active power filter based on terminal sliding mode |
| CN107453362A (en) * | 2017-09-19 | 2017-12-08 | 河海大学常州校区 | Active Power Filter-APF inverting global sliding mode fuzzy control based on neutral net |
| CN108334679A (en) * | 2018-01-22 | 2018-07-27 | 河海大学常州校区 | Active Power Filter-APF total-sliding-mode control method based on double feedback fuzzy neural networks |
-
2018
- 2018-09-18 CN CN201811086430.4A patent/CN108828961B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104730921A (en) * | 2015-01-13 | 2015-06-24 | 河海大学常州校区 | Method for controlling fuzzy neural network of active power filter based on terminal sliding mode |
| CN107453362A (en) * | 2017-09-19 | 2017-12-08 | 河海大学常州校区 | Active Power Filter-APF inverting global sliding mode fuzzy control based on neutral net |
| CN108334679A (en) * | 2018-01-22 | 2018-07-27 | 河海大学常州校区 | Active Power Filter-APF total-sliding-mode control method based on double feedback fuzzy neural networks |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109709802A (en) * | 2018-12-14 | 2019-05-03 | 江南大学 | The control method of active electronic ladder circuit based on iterative learning control |
| CN112271729A (en) * | 2020-10-15 | 2021-01-26 | 河海大学常州校区 | Long-time memory fuzzy neural network sliding mode control method of active power filter |
| CN112271729B (en) * | 2020-10-15 | 2022-08-19 | 河海大学常州校区 | Long-time memory fuzzy neural network sliding mode control method of active power filter |
| CN112947066A (en) * | 2021-01-26 | 2021-06-11 | 苏州连恺自动化有限公司 | Manipulator improved finite time inversion control method |
| CN114336742A (en) * | 2021-12-22 | 2022-04-12 | 浙江大学 | Solar photovoltaic controller based on self-adaptive neural network |
| CN114336742B (en) * | 2021-12-22 | 2023-10-27 | 浙江大学 | Solar photovoltaic controller based on self-adaptive neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108828961B (en) | 2021-07-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109103884A (en) | Active Power Filter-APF back stepping control method based on metacognition fuzzy neural network | |
| CN108828961A (en) | Active Power Filter-APF sliding-mode control based on metacognition fuzzy neural network | |
| Fei et al. | Novel neural network fractional-order sliding-mode control with application to active power filter | |
| CN108334679A (en) | Active Power Filter-APF total-sliding-mode control method based on double feedback fuzzy neural networks | |
| CN107453362A (en) | Active Power Filter-APF inverting global sliding mode fuzzy control based on neutral net | |
| CN106169754B (en) | Active Power Filter-APF neural network dynamic PID total-sliding-mode control methods | |
| CN104730921B (en) | Method for controlling fuzzy neural network of active power filter based on terminal sliding mode | |
| Citro et al. | Designing inverters’ current controllers with resonance frequencies cancellation | |
| Narayana et al. | A Comparative Analysis of PI Controller and Fuzzy Logic Controller for Hybrid Active Power Filter Using Dual Instantaneous Power Theory | |
| CN109100937A (en) | Active Power Filter-APF total-sliding-mode control method based on two hidden-layer recurrent neural networks | |
| Fei et al. | Recurrent neural network fractional-order sliding mode control of dynamic systems | |
| CN110137971B (en) | Voltage stability control method for three-phase alternating current power spring | |
| Rathika et al. | Fuzzy logic–based approach for adaptive hysteresis band and dc voltage control in shunt active filter | |
| CN103293963B (en) | Active Power Filter-APF adaptive fuzzy inverting tracking and controlling method | |
| CN110266044B (en) | Microgrid grid-connected control system and method based on energy storage converter | |
| Tan et al. | Droop controlled microgrid with DSTATCOM for reactive power compensation and power quality improvement | |
| CN110429835A (en) | A kind of RBFNN segmentation on-line optimization Passive Shape Control system and method based on LCL filtering | |
| CN108233757A (en) | Inverter control method based on NEW TYPE OF COMPOSITE control algolithm | |
| Pournazarian et al. | Enhancing microgrid small-signal stability and reactive power sharing using ANFIS-tuned virtual inductances | |
| CN107623341B (en) | VSC inverter station mathematical model for supplying power to passive network and internal model controller | |
| CN109103885A (en) | Active Power Filter-APF metacognition fuzzy Neural Network Control Method | |
| CN104467741B (en) | Active filter Intelligent current tracking and controlling method based on T S obscurity model buildings | |
| Gupta et al. | A fuzzy controlled shunt active power filter for reducing current harmonics and reactive power compensation | |
| CN110350546B (en) | Control method of single-phase active power filter | |
| Chau et al. | Online control method with time-delay compensation for hybrid active power filter with injection circuit |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |