CN109188916A - A kind of oxygenerator compress control method based on fuzzy neural network - Google Patents
A kind of oxygenerator compress control method based on fuzzy neural network Download PDFInfo
- Publication number
- CN109188916A CN109188916A CN201811348848.8A CN201811348848A CN109188916A CN 109188916 A CN109188916 A CN 109188916A CN 201811348848 A CN201811348848 A CN 201811348848A CN 109188916 A CN109188916 A CN 109188916A
- Authority
- CN
- China
- Prior art keywords
- layer
- input
- function
- network
- fuzzy
- 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
- G05B13/0265—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion
- G05B13/0285—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion using neural networks and fuzzy logic
-
- 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
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D16/00—Control of fluid pressure
- G05D16/20—Control of fluid pressure characterised by the use of electric means
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Automation & Control Theory (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Fluid Mechanics (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Feedback Control In General (AREA)
Abstract
The present invention relates to a kind of oxygenerator compress control method based on fuzzy neural network, comprising the following steps: the first step constructs fuzzy neural network controller, which includes former piece network and consequent network；Former piece network includes 4 layers, this 4 layers are as follows: former piece network input layer is blurred layer, and fuzzy rule matching layer normalizes layer；Consequent network includes 3 layers, this 3 layers are as follows: consequent network input layer, consequent computation layer, controller output layer.Oxygenerator compress control method of the present invention based on fuzzy neural network, combine the advantage of fuzzy control and ANN Control, for carrying out dynamic realtime control to oxygen making set system pressure, it is able to solve the deficiency for the oxygenerator that works in a manner of time control switching or fixation pressure limit value, conducive to steady and continuous prepares the oxygen of high concentration.
Description
Technical field
The invention belongs to medical instruments fields, specifically, relating generally to a kind of oxygenerator compress control method, especially relate to
And a kind of oxygenerator compress control method based on fuzzy neural network.
Background technique
Currently, the middle-size and small-size oxygenerator of rehabilitation class mostly uses greatly molecular sieve pressure swing adsorption to carry out oxygen processed.Oxygen generation system
Pressure is significant for high-purity and oxygen making set system influence on system operation steady in a long-term of oxygen concentration processed whether stabilization.Due to molecular sieve material
Expect the regional power grid of the primary dispersive difference of characteristic, the technological ability distribution probability of existing industrial technology level, oxygenerator application
The system pressure of the collective effect of many factors such as fluctuation and environmental change, oxygenerator operation has non-linear and time-varying spy
Property, that is, it is easy to cause oxygen processed unstable.
What Current Domestic and the international middle-size and small-size oxygenerator of rehabilitation class used mainly controls switching bilateral with the time
Adsorbing tower with molecular sieve operation.Since oxygen making set system pressure has the characteristics that above-mentioned non-linear and time-varying, time control can not
Guarantee oxygenerator long-continued high-performance operation, adsorbing tower with molecular sieve hypertonia easily occurs and cause to penetrate or bilateral molecule
Sieve adsorption tower load uneven (i.e. when wherein side molecular sieve excess load, being unable to absorption nitrogen, oxygen concentration is caused to reduce), low
Effect operation.Least a portion of pressure control switching bilateral adsorbing tower with molecular sieve runs oxygenerator type, and use is also fixed pressure
Power limit value setting, it is difficult to which performance and operating condition difference for system individual and the operation fluctuation in complicated applications environment are corresponding
Ground is adjusted, and also limits the high performance operation of oxygen making set system.
In consideration of it, how to find a kind of suitable oxygen making set system characteristic and adapt to the intelligent control side of complicated running environment
Formula proposes that a kind of oxygenerator compress control method based on fuzzy neural network is the project of the invention to be studied.
Summary of the invention
The present invention provides a kind of oxygenerator compress control method based on fuzzy neural network, existing its purpose is to solve
There is technology not can guarantee the long-continued high-performance operation of oxygenerator, adsorbing tower with molecular sieve hypertonia easily occur and cause to penetrate,
Or the problem of bilateral adsorbing tower with molecular sieve load is uneven, fallback.
In order to achieve the above objectives, the technical solution adopted by the present invention is that: a kind of oxygenerator pressure based on fuzzy neural network
Force control method, comprising the following steps:
The first step constructs fuzzy neural network controller, which includes former piece network and consequent
Network；
Definition system input variable is x1And x2: x1=er=pd- p,Wherein, pdFor pressure set points,
P is pressure measurements, and er is pressure error, and ec is pressure error change rate, i.e. derivative of the er to time t；
The former piece network includes 4 layers, this 4 layers are respectively as follows:
First layer is former piece network input layer, has 2 nodes, respectively by input variable x1And x2It is sent to blurring
Layer；
The second layer, to be blurred layer, by input variable x1And x2Respectively it is divided into miA fuzzy subset, the fuzzy subset's
Subordinating degree function is all made of Gaussian function, expression formula such as following formula one:
Wherein,It is input variable xiQi(qi=1,2 ..., mi) a fuzzy subset subordinating degree function, i=1,2
For the serial number of input variable, miFor input variable xiFuzzy partition number,WithRespectively subordinating degree functionCenter
Value and width, the node sum for being blurred layer are
Third layer is fuzzy rule matching layer, the relevance grade α of every rule is calculated using multiplication operationj, expression formula is as follows
Formula two:
Wherein, q1j∈{1,2,…,m1, q2j∈{1,2,…,m2, q1jIn subscript in j represent input variable x1's
Q1The subordinating degree function of a fuzzy subset corresponds to j-th strip rule, q2jIn subscript in j represent input variable x2Q2
The subordinating degree function of a fuzzy subset corresponds to j-th strip rule；
αjIn subscript in j represent relevance grade corresponding to j-th strip rule；
J=1,2 ..., m, regular sum are
4th layer, to normalize layer, calculate normalized regular relevance gradeCalculating formula such as following formula three:
Wherein, j=1,2 ..., m；αsFor the relevance grade of every rule, with αjRepresentative meaning is identical, the α in formula threes
For stating the relevance grade summation to m rule, s=1,2 ..., m, use and αjDifferent symbols is with the calculating side to formula three
Formula statement is clear；
The consequent network includes 3 layers, this 3 layers are as follows:
First layer is consequent network input layer, has 3 nodes, wherein the input value of the 0th node is x0=1, the 1st
A and the 2nd node distinguishes input variable x1And x2；
The second layer is consequent computation layer, calculates the consequent u of each rulej, expression formula such as following formula four:
Wherein, j=1,2 ..., m；x0=1；wjlJ-th strip rule and input value x are corresponded to for consequent networklBetween connection
Power；Herein, l=0,1,2, xlRespectively x0、x1And x2；
Third layer is controller output layer, calculates the output u of fuzzy neural network controller, expression formula such as following formula five:
Wherein, j=1,2 ..., m；For normalized regular relevance grade；ujFor the consequent of each rule；
Second step is iterated calculating using the BP algorithm of error back propagation, obtains the control that the above-mentioned first step is established
The learning algorithm of consequent connection weight, regular former piece subordinating degree function central value and width in device processed, online adjustment in real time
Each parameter of controller in the first step makes oxygen making set system pressure stablize optimal setting value of the tracking based on expertise；
Taking error cost function E is such as following formula six:
The consequent is connected to the network weight wjlLearning algorithm be such as following formula seven:
The central value of each node subordinating degree function of the former piece network second layerLearning algorithm be such as following formula eight:
The width of each node subordinating degree function of the former piece network second layerLearning algorithm be such as following formula nine:
Wherein, j=1,2 ..., m；L=0,1,2；η is learning rate；β is factor of momentum；(k) time step, (k+1) are indicated
Indicate future time step, (k-1) indicates previous time step；
wjl(k+1)、wjl(k) and wjlIt (k-1) is respectively (k+1) time step, (k) time step and (k-1) time step
Connection weight；
P (k) is the pressure measurements of (k) time step, and p (k-1) is the pressure measurements of (k-1) time step；
U (k) is output of the fuzzy neural network controller in (k) time step, and u (k-1) is Fuzzy Neural-network Control
Output of the device in (k-1) time step；
AndRespectively subordinating degree function is in (k+1) time step, (k)
The central value of spacer step and (k-1) time step；
AndRespectively subordinating degree function is in (k+1) time step, (k)
The width of spacer step and (k-1) time step.
Related content in above-mentioned technical proposal is explained as follows:
1, in above scheme, the input function of the former piece network input layer node iAnd output functionRespectively
Are as follows:
Wherein, serial number of the subscript i for former piece network input layer node, i=1,2；
In I indicate input (Input),In O indicate output (Output)；fi (a1)In f represent correspond to
Input function relationship,In g represent corresponding output function relationship；Subscript (a1) expression former piece be (antecedent's
Initial a) network first tier.
2, in above scheme, in the blurring layer with input variable xiQiThe corresponding node of a fuzzy subset
Input functionAnd output functionIt is respectively as follows:
Wherein, subscript iqiMiddle qiI before is input variable xiSerial number, i=1,2；Subscript iqiIn qiIt indicates to correspond to
Input variable xiQiA fuzzy subset, qi=1,2 ..., mi；
WithIt respectively represents and input variable xiQiThe input function of the corresponding node of a fuzzy subset
Relationship and output function relationship.
3, in above scheme, the input function of the fuzzy rule matching layer node jAnd output functionRespectively
Are as follows:
Wherein, j=1,2 ..., m；
fj (a3)AndIn subscript j respectively represent the node j of corresponding fuzzy rule matching layer；
Subscript 1q1jMiddle q1j1 before represents input variable x1Serial number, q1jIn q1It representsCorrespond to
Input variable x1Q1A fuzzy subset, q1jIn j indicateFor the input of j-th of node of fuzzy rule matching layer；Subscript 2q2jMiddle q2j2 before represent input variable x2Serial number, q2jIn q2It representsCorresponding to input variable
x2Q2A fuzzy subset, q2jIn j indicateFor the input of j-th of node of fuzzy rule matching layer.
4, in above scheme, the input function of the normalization layer node jAnd output functionIt is respectively as follows:
Wherein, j=1,2 ..., m.
fj (a4)AndIn subscript j respectively represent the node j of corresponding normalization layer；
For the output function of fuzzy rule matching layer node, withRepresentative meaning is identical, in normalization layer
In the input function expression formula of node jFor stating the output function summation to m node of fuzzy rule matching layer, s=
1,2 ..., m, using withDifferent symbols calculates statement clearly with the input function to normalization layer node j.
5, in above scheme, the input function of the consequent network input layer node iAnd output functionRespectively
Are as follows:
Wherein, serial number of the subscript i for consequent network input layer node, i=0,1,2；x0=1.
6, in above scheme, the input function of the consequent computation layer node jAnd output functionRespectively
Are as follows:
Wherein, j=1,2 ..., m；
fj (c2)AndIn subscript j respectively represent the node j of rule of correspondence consequent computation layer.
7, in above scheme, the input function I of the controller output layer node(c3)And output function O(c3)It is respectively as follows:
Wherein, j=1,2 ..., m.
8, in above scheme, about the explanation for describing mathematic sign and formula of the present invention:
Mathematic sign and formula used by the content of present invention, according to chinese national standard GB3102.11-93 " physics section
Learn and technology used in mathematic sign " regulation selected and applied.The mathematic sign selected in the present invention is carried out
Explanation attached be listed in such as the following table 1.
9, in above scheme, illustrated using mathematic sign:
To describe mathematic sign and formula used by the content of present invention, according to chinese national standard GB3102.11-93
The regulation of " mathematic sign used in physics and technology " is selected and is applied.
Mathematic sign and its institute in GB3102.11-93 employed in the be classified as present specification and claims of table 1
Defined item No. and meaning.Symbol arrangement is by item No. sequence in GB3102.11-93.
Table 1 describes mathematic sign used in the present invention
10, in above scheme, the control method that the present invention designs is intelligent control, the structure of fuzzy neural network being based on
For Takagi-Sugeno (TS) model.
The learning functionality and adaptive functions of intelligent control, Controlling model is uncertain, energy in the non-linear and system of time-varying
Ideal effect is obtained, can be used for solving problem encountered in the control of oxygen making set system pressure.
Fuzzy control in intelligent control based on fuzzy set theory, Fuzzy Linguistic Variable and fuzzy logic inference, from
Apish fuzzy reasoning and decision in behavior.It can use expertise and be formed by fuzzy rule, be to dynamic time-varying
System is controlled.But fuzzy control is not easy to the study and adjustment of control parameter.
Another important branch ANN Control of intelligent control imitates human brain physiological system, can learn and adapt to non-thread
The dynamic characteristic of property uncertain system.But the mapping ruler of artificial neural network is invisible, is bad to utilize Expert Rules.
The oxygenerator compress control method based on fuzzy neural network that the present invention designs, combines fuzzy control and nerve
The advantage of network-control, for oxygen making set system pressure carry out dynamic realtime control, be able to solve with the time control switching or
Fixation pressure limit value mode works the deficiency of oxygenerator, conducive to steady and continuous prepares the oxygen of high concentration.
Detailed description of the invention
Attached drawing 1 is the oxygenerator control pressurer system structure chart of the invention based on fuzzy neural network；
Attached drawing 2 is Takagi-Sugeno (TS) structure of fuzzy neural network schematic diagram that the present invention uses.
Specific embodiment
The invention will be further described with reference to the accompanying drawings and embodiments:
A kind of embodiment: oxygenerator compress control method based on fuzzy neural network
The following steps are included:
The first step constructs fuzzy neural network controller, which includes former piece network and consequent
Network.
Referring to shown in attached drawing 1, FNNC is fuzzy neural network controller, and u (k) indicates control amount, and K is fuzznet
The coefficient of network control amount.
Fuzzy controller is using dual input (two dimension) single export structure in the present embodiment.
The former piece network generates fuzzy rule consequent for matching fuzzy rule former piece, consequent network.Fuzznet
Network controller architecture schematic diagram is referring to attached drawing 2.
Former piece network has four layers.
First layer, former piece network input layer.Former piece network input layer node sum is Na1=2, respectively with input variable x1
And x2It is connected, by input value x1And x2It is sent to following blurring layers.
In the present embodiment, the input function of former piece network input layer node iAnd output functionIt is respectively as follows:
Wherein, serial number of the subscript i for former piece network input layer node, i=1,2.
Functional symbol used by above two formula and upper target meaning are as follows:
In I indicate input (Input),In O indicate output (Output)；fi (a1)In f represent correspond to
Input function relationship,In g represent corresponding output function relationship；Subscript (a1) expression former piece be (antecedent's
The 1st layer of initial a) network.
Affiliated net is indicated herein in regard to input I, output O, input function relationship f and output function relationship g and with subscript
The expression way of the meaning of network layers is equally applicable in the specification and claims of the present invention to state former piece network institute
There are layer and all layers of consequent network, the difference that each corresponding functional symbol of layer is specifically stated is only that different layers in subscript
Indicated with respectively corresponding Arabic numerals, and consequent network in subscript (such as (c1)) with c (consequent consequence
Initial) it indicates.
Input variable x1And x2Itself be not belonging to fuzzy neural network, but in system structure be located at the 1st layer of former piece network it
Before, so indicating that symbol is expressed as with the 0th layerIn subscript (0).
The second layer is blurred layer.By input variable x1And x2Respectively be divided into 7 fuzzy subsets NB, NM, NS, O, PS, PM,
PB}。
The subordinating degree function of the fuzzy subset is all made of Gaussian function such as following formula one:
Wherein,It is input variable xiQi(qi=1,2 ..., mi) a fuzzy subset subordinating degree function, i=1,2
For the serial number of input variable, miFor input variable xiFuzzy partition number, m in the present embodiment1=m2=7.WithRespectively
Subordinating degree functionCentral value and width.It is described blurring layer node sum be
In the present embodiment, it is blurred in layer and input variable xiQiThe input letter of the corresponding node of a fuzzy subset
NumberAnd output functionIt is respectively as follows:
Wherein, subscript iqiMiddle qiI before is input variable xiSerial number, i=1,2；Subscript iqiIn qiIt indicates to correspond to
Input variable xiQiA fuzzy subset, qi=1,2 ..., mi；WithIt respectively represents and input variable xiQi
The input function relationship and output function relationship of the corresponding node of a fuzzy subset.
Third layer, fuzzy rule matching layer.This layer of each node represents a fuzzy rule, is calculated using multiplication operation every
The relevance grade of rule.
J-th strip rule relevance grade αjExpression formula be such as following formula two:
Wherein, q1j∈{1,2,…,m1, q2j∈{1,2,…,m2, q1jIn subscript in j represent input variable x1's
Q1The subordinating degree function of a fuzzy subset corresponds to j-th strip rule, q2jIn subscript in j represent input variable x2Q2
The subordinating degree function of a fuzzy subset corresponds to j-th strip rule.
αjIn subscript in j represent relevance grade corresponding to j-th strip rule；
J=1,2 ..., m, regular sum are
This layer of node sum is Na3=49.
In the present embodiment, the input function of the fuzzy rule matching layer node jAnd output functionRespectively
Are as follows:
Wherein, j=1,2 ..., m；
fj (a3)AndIn subscript j respectively represent the node j of corresponding fuzzy rule matching layer；
Subscript 1q1jMiddle q1j1 before represents input variable x1Serial number, q1jIn q1It representsCorrespond to
Input variable x1Q1A fuzzy subset, q1jIn j indicateFor the input of j-th of node of fuzzy rule matching layer；Subscript 2q2jMiddle q2j2 before represent input variable x2Serial number, q2jIn q2It representsCorresponding to input variable
x2Q2A fuzzy subset, q2jIn j indicateFor the input of j-th of node of fuzzy rule matching layer.
4th layer, normalize layer.The normalization of implementation rule relevance grade calculates.
The normalization relevance grade of corresponding j-th strip ruleCalculating formula such as following formula three:
Wherein, j=1,2 ..., m.
This layer of node sum is Na4=49.
In the present embodiment, the input function of layer node j is normalizedAnd output functionIt is respectively as follows:
Wherein, j=1,2 ..., m；fj (a4)AndIn subscript j respectively represent corresponding normalization layer
Node j；
For the output function of fuzzy rule matching layer node, withRepresentative meaning is identical, in normalization layer
In the input function expression formula of node jFor stating the output function summation to m node of fuzzy rule matching layer, s=
1,2 ..., m, using withDifferent symbols calculates statement clearly with the input function to normalization layer node j.
Consequent network is formed by 3 layers, comprising:
First layer, consequent network input layer.Consequent network input layer number of network nodes is Nc1=3.Wherein the 0th node is defeated
Entering value is x0=1, act on the constant term being to provide in fuzzy rule consequent；1st and the 2nd node difference input variable x1
And x2。
In the present embodiment, the input function of consequent network input layer node iAnd output functionIt is respectively as follows:
Wherein, subscript i is the serial number of consequent network input layer node；I=0,1,2；x0=1.
Input variable x1And x2Itself be not belonging to fuzzy neural network, but in system structure be located at the 1st layer of consequent network it
Before, so indicating that symbol is expressed as with the 0th layerIn subscript (0).
The second layer, consequent computation layer calculate the consequent of each rule.Shared Nc2=m=49 node, Mei Gejie
Point represents a rule.
Consequent expression formula such as following formula four:
Wherein, x0=1；wjlJ-th strip rule and input value x are corresponded to for consequent networklBetween connection weight, j=1,2 ...,
m。
In the present embodiment, the input function of consequent computation layer node jAnd output functionIt is respectively as follows:
Wherein, j=1,2 ..., m；fj (c2)AndIn subscript j respectively represent rule of correspondence consequent meter
Calculate the node j of layer.
Third layer, controller output layer calculate the output u of fuzzy neural network controller.
Export the expression formula such as following formula five of u:
In formula, m=49.
The layer has Nc3=1 node.
In the present embodiment, the input function I of controller output layer node(c3)And output function O(c3)It is respectively as follows:
Wherein, j=1,2 ..., m；For normalized regular relevance grade；ujFor the consequent of each rule.
Second step, learning algorithm are iterated calculating using the BP algorithm of error back propagation, obtain the above-mentioned first step
Consequent connection weight w in middle established controllerjl。
In the present embodiment, the study of the fuzzy neural network controller uses on-line study, application error backpropagation
BP algorithm.
Because of input variable x1And x2Fuzzy partition number all predefined, so after needing the parameter that learns to be mainly
The connection weight w of part networkjlAnd the central value of each node subordinating degree function of the former piece network second layerAnd width
Take error cost function E such as following formula six:
According to chain type Rule for derivation, obtain
In formula,
Consequent is connected to the network weight wjlLearning algorithm such as following formula seven:
Wherein,
J=1,2 ..., m；L=0,1,2；
η is learning rate, η ∈ [0,1]；
β is factor of momentum, β ∈ [0,1]；
(k) time step is indicated；(k+1) future time step is indicated, (k-1) indicates previous time step.
Consequent network connection weight is computed obtain after, can be by each connection weight parameter wjlIt is fixed, with calculate it is described before
The central value of each node subordinating degree function of the part network second layerAnd width
The central value of each node subordinating degree function of the former piece network second layerLearning algorithm such as following formula eight:
The width of each node subordinating degree function of the former piece network second layerLearning algorithm such as following formula nine:
In formula eight and formula nine, i=1,2；qi=1,2 ..., mi；
η is learning rate, η ∈ [0,1]；
β is factor of momentum, β ∈ [0,1].
The central value of subordinating degree functionIn learning algorithm expression formula (formula eight)And the width of subordinating degree functionIn learning algorithm expression formula (formula nine)It can be using the BP algorithm of error back propagation by following iterative calculation (chain
Formula rule seeks partial derivative, is iterated calculating):
The input function of former piece network third layerFor multiplication operation, whenRepresent j-th strip rule
When one input of node,
The label for being indicated input variable in above formula using v is subordinate to for what is inputted with the node for representing j-th strip rule in formula
Degree function number i is distinguished, clear to state, v=1, and 2；
Otherwise,
To obtain:
Ten acquired results of formula are substituted into formula eight, the central value of subordinating degree function can be obtained through iterating to calculate
11 acquired results of formula are substituted into formula nine, the width of subordinating degree function can be obtained through iterating to calculate
Oxygenerator compress control method of the present invention based on fuzzy neural network, combines fuzzy control and nerve
The advantage of network-control, for oxygen making set system pressure carry out dynamic realtime control, be able to solve with the time control switching or
Fixation pressure limit value mode works the deficiency of oxygenerator, conducive to steady and continuous prepares the oxygen of high concentration.
The above embodiments merely illustrate the technical concept and features of the present invention, and its object is to allow person skilled in the art
Scholar cans understand the content of the present invention and implement it accordingly, and it is not intended to limit the scope of the present invention.It is all according to the present invention
Equivalent change or modification made by Spirit Essence, should be covered by the protection scope of the present invention.
Claims (8)
1. a kind of oxygenerator compress control method based on fuzzy neural network, which comprises the following steps:
The first step constructs fuzzy neural network controller, which includes former piece network and consequent network；
Definition system input variable is x1And x2, x1=er=pd- p,Wherein, pdFor pressure set points, p is
Pressure measurements, er are pressure error, and ec is pressure error change rate, i.e. derivative of the er to time t；
The former piece network includes 4 layers, this 4 layers are respectively as follows:
First layer is former piece network input layer, has 2 nodes, respectively by the input variable x1And x2It is sent to following fuzzy
Change layer；
The second layer, to be blurred layer, by input variable x1And x2Respectively it is divided into miA fuzzy subset, the fuzzy subset's is subordinate to
Degree function is all made of Gaussian function, expression formula such as following formula one:
Wherein,It is input variable xiQi(qi=1,2 ..., mi) a fuzzy subset subordinating degree function, i=1,2 be defeated
Enter the serial number of variable, miFor input variable xiFuzzy partition number,WithRespectively subordinating degree functionCentral value and
Width, the node sum for being blurred layer are
Third layer is fuzzy rule matching layer, the relevance grade α of every rule is calculated using multiplication operationj, expression formula such as following formula two:
Wherein, q1j∈{1,2,…,m1, q2j∈{1,2,…,m2, j=1,2 ..., m, regular sum isThe
Four layers, to normalize layer, calculate normalized regular relevance gradeCalculating formula such as following formula three:
Wherein, j=1,2 ..., m；αsFor the relevance grade of every rule, with αjRepresentative meaning is identical, α in the formula threesWith
It sums in relevance grade of the statement to m rule, s=1,2 ..., m；
The consequent network includes 3 layers, this 3 layers are respectively as follows:
First layer is consequent network input layer, has 3 nodes, wherein the input value of the 0th node is x0=1, the 1st and
2 nodes distinguish input variable x1And x2；
The second layer is consequent computation layer, calculates the consequent u of each rulej, expression formula such as following formula four:
Wherein, j=1,2 ..., m；x0=1；L=0,1,2；wjlJ-th strip rule and input value x are corresponded to for consequent networklBetween
Connection weight；
Third layer is controller output layer, calculates the output u of fuzzy neural network controller, expression formula such as following formula five:
Wherein, j=1,2 ..., m；For normalized regular relevance grade；ujFor the consequent of each rule；
Second step is iterated calculating using the BP algorithm of error back propagation, obtains the controller that the above-mentioned first step is established
Middle consequent connection weight, regular former piece subordinating degree function central value and width learning algorithm adjust described the in real time online
Each parameter of controller in one step makes oxygen making set system pressure stablize optimal setting value of the tracking based on expertise；
Taking error cost function E is such as following formula six:
The consequent is connected to the network weight wjlLearning algorithm be such as following formula seven:
The central value of each node subordinating degree function of the former piece network second layerLearning algorithm be such as following formula eight:
The width of each node subordinating degree function of the former piece network second layerLearning algorithm be such as following formula nine:
Wherein, j=1,2 ..., m；L=0,1,2；η is learning rate；β is factor of momentum；(k) time step is indicated；(k+1) it indicates
Future time step；(k-1) previous time step is indicated；
wjl(k+1)、wjl(k) and wjl(k-1) be respectively (k+1) time step, (k) time step and (k-1) time step company
Connect weight；
P (k) is the pressure measurements of (k) time step, and p (k-1) is the pressure measurements of (k-1) time step；
U (k) is output of the fuzzy neural network controller in (k) time step, and u (k-1) is that fuzzy neural network controller exists
The output of (k-1) time step；
AndRespectively subordinating degree function is in (k+1) time step, (k) time step
And the central value of (k-1) time step；
AndRespectively subordinating degree function is in (k+1) time step, (k) time step
And the width of (k-1) time step.
2. the oxygenerator compress control method according to claim 1 based on fuzzy neural network, which is characterized in that described
The input function of former piece network input layer node iAnd output functionIt is respectively as follows:
Wherein, i=1,2；In I indicate input (Input),In O indicate output (Output)；fi (a1)In f
Corresponding input function relationship is represented,In g represent corresponding output function relationship；Subscript (a1) indicates former piece
(initial a) the network first tier of antecedent.
3. the oxygenerator compress control method according to claim 1 based on fuzzy neural network, which is characterized in that described
It is blurred in layer and input variable xiQiThe input function of the corresponding node of a fuzzy subsetAnd output functionIt is respectively as follows:
Wherein, i=1,2；qi=1,2 ..., mi；WithIt respectively represents and input variable xiQiA fuzzy subset's phase
The input function relationship and output function relationship of corresponding node.
4. the oxygenerator compress control method according to claim 1 based on fuzzy neural network, which is characterized in that described
The input function of fuzzy rule matching layer node jAnd output functionIt is respectively as follows:
Wherein, j=1,2 ..., m.
5. the oxygenerator compress control method according to claim 1 based on fuzzy neural network, which is characterized in that described
Normalize the input function of layer node jAnd output functionIt is respectively as follows:
Wherein, j=1,2 ..., m.
6. the oxygenerator compress control method according to claim 1 based on fuzzy neural network, which is characterized in that described
The input function of consequent network input layer node iAnd output functionIt is respectively as follows:
Wherein, i=0,1,2；x0=1.
7. the oxygenerator compress control method according to claim 1 based on fuzzy neural network, which is characterized in that described
The input function of consequent computation layer node jAnd output functionIt is respectively as follows:
Wherein, j=1,2 ..., m.
8. the oxygenerator compress control method according to claim 1 based on fuzzy neural network, which is characterized in that described
The input function I of controller output layer node(c3)And output function O(c3)It is respectively as follows:
Wherein, j=1,2 ..., m.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811348848.8A CN109188916A (en) | 2018-11-13 | 2018-11-13 | A kind of oxygenerator compress control method based on fuzzy neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811348848.8A CN109188916A (en) | 2018-11-13 | 2018-11-13 | A kind of oxygenerator compress control method based on fuzzy neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109188916A true CN109188916A (en) | 2019-01-11 |
Family
ID=64939723
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811348848.8A Pending CN109188916A (en) | 2018-11-13 | 2018-11-13 | A kind of oxygenerator compress control method based on fuzzy neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109188916A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115798618A (en) * | 2022-10-28 | 2023-03-14 | 武汉城市职业学院 | Optimization method and system for efficiently purifying hydrogen through pressure swing adsorption |
| CN116236881A (en) * | 2023-05-09 | 2023-06-09 | 威海威高健康科技有限公司 | Molecular sieve oxygen generation method and system with program self-adaption function |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH06332501A (en) * | 1993-05-24 | 1994-12-02 | Ishikawajima Harima Heavy Ind Co Ltd | Feedback controller and incinerator using the feedback controller |
| US20020083736A1 (en) * | 2000-12-28 | 2002-07-04 | Kotliar Igor K. | Mobile firefighting systems with breathable hypoxic fire extinguishing compositions for human occupied environments |
| CN103984277A (en) * | 2014-05-26 | 2014-08-13 | 保定迈卓医疗器械有限公司 | Monitoring and controlling system of full-digital oxygenerator |
| CN105800560A (en) * | 2016-05-16 | 2016-07-27 | 镇江市高等专科学校 | Medical oxygen generator with stable performance |
| CN104133372B (en) * | 2014-07-09 | 2016-09-28 | 河海大学常州校区 | Room temperature control algolithm based on fuzzy neural network |
| CN107291009A (en) * | 2017-06-27 | 2017-10-24 | 合肥康居人智能科技有限公司 | A kind of Domestic oxygen erator intelligence control system and its control method |
| CN108249404A (en) * | 2018-04-17 | 2018-07-06 | 湖州灵感电子科技有限公司 | A kind of domestic type oxygenerator |
-
2018
- 2018-11-13 CN CN201811348848.8A patent/CN109188916A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH06332501A (en) * | 1993-05-24 | 1994-12-02 | Ishikawajima Harima Heavy Ind Co Ltd | Feedback controller and incinerator using the feedback controller |
| US20020083736A1 (en) * | 2000-12-28 | 2002-07-04 | Kotliar Igor K. | Mobile firefighting systems with breathable hypoxic fire extinguishing compositions for human occupied environments |
| CN103984277A (en) * | 2014-05-26 | 2014-08-13 | 保定迈卓医疗器械有限公司 | Monitoring and controlling system of full-digital oxygenerator |
| CN104133372B (en) * | 2014-07-09 | 2016-09-28 | 河海大学常州校区 | Room temperature control algolithm based on fuzzy neural network |
| CN105800560A (en) * | 2016-05-16 | 2016-07-27 | 镇江市高等专科学校 | Medical oxygen generator with stable performance |
| CN107291009A (en) * | 2017-06-27 | 2017-10-24 | 合肥康居人智能科技有限公司 | A kind of Domestic oxygen erator intelligence control system and its control method |
| CN108249404A (en) * | 2018-04-17 | 2018-07-06 | 湖州灵感电子科技有限公司 | A kind of domestic type oxygenerator |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115798618A (en) * | 2022-10-28 | 2023-03-14 | 武汉城市职业学院 | Optimization method and system for efficiently purifying hydrogen through pressure swing adsorption |
| CN116236881A (en) * | 2023-05-09 | 2023-06-09 | 威海威高健康科技有限公司 | Molecular sieve oxygen generation method and system with program self-adaption function |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | An adaptive fractional-order BP neural network based on extremal optimization for handwritten digits recognition | |
| Ata et al. | An adaptive neuro-fuzzy inference system approach for prediction of tip speed ratio in wind turbines | |
| He et al. | Conditions for the convergence of evolutionary algorithms | |
| Han et al. | Modified constrained learning algorithms incorporating additional functional constraints into neural networks | |
| Li et al. | Hyperchaos in Hopfield-type neural networks | |
| CN109188916A (en) | A kind of oxygenerator compress control method based on fuzzy neural network | |
| Liu et al. | Hybrid learning particle swarm optimizer with genetic disturbance | |
| Oh et al. | Self-organizing neural networks with fuzzy polynomial neurons | |
| Balakrishnan et al. | On-line emission and economic load dispatch using adaptive Hopfield neural network | |
| CN106959613A (en) | Dynamical linearization adaptive control laws algorithm of the SISO systems based on recent renewal information | |
| Soodi et al. | STATCOM estimation using back-propagation, PSO, shuffled frog leap algorithm, and genetic algorithm based neural networks | |
| Niu et al. | A novel quantum neural network based on multi-level activation function | |
| Luo et al. | Pruning method for dendritic neuron model based on dendrite layer significance constraints | |
| Ding et al. | College English online teaching model based on deep learning | |
| Sharma et al. | ANFIS based control design for AGC of a hydro-hydro power system with UPFC and hydrogen electrolyzer units | |
| Ning et al. | Using chaotic neural network to forecast stock index | |
| Shuangxi et al. | Research on PID control based on BP neural network and its application | |
| Singh et al. | Fast approach to artificial neural network training and its application to economic load dispatch | |
| Zhang et al. | Study of prediction model on grey relational BP neural network based on rough set | |
| Hanai et al. | Modeling of total evaluation process of Ginjo sake using a fuzzy neural network | |
| CN107392306A (en) | The training learning method and system of the neutral net of pattern specifications parameter inference pattern | |
| CN114202063A (en) | Fuzzy neural network greenhouse temperature prediction method based on genetic algorithm optimization | |
| CN107272418A (en) | A kind of bionic intelligence control method for the neutral net that is self-regulated based on many intensions | |
| Li et al. | An improved double hidden-layer variable length incremental extreme learning machine based on particle swarm optimization | |
| CN111798037A (en) | Data-driven optimal power flow calculation method based on stacked extreme learning machine frame |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190111 |
|
| RJ01 | Rejection of invention patent application after publication |