CN109597449A - A kind of ultrasonic wave separating apparatus temprature control method neural network based - Google Patents
A kind of ultrasonic wave separating apparatus temprature control method neural network based Download PDFInfo
- Publication number
- CN109597449A CN109597449A CN201910092679.4A CN201910092679A CN109597449A CN 109597449 A CN109597449 A CN 109597449A CN 201910092679 A CN201910092679 A CN 201910092679A CN 109597449 A CN109597449 A CN 109597449A
- Authority
- CN
- China
- Prior art keywords
- layer
- neural network
- output
- control
- weight
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D23/00—Control of temperature
- G05D23/19—Control of temperature characterised by the use of electric means
- G05D23/30—Automatic controllers with an auxiliary heating device affecting the sensing element, e.g. for anticipating change of temperature
- G05D23/32—Automatic controllers with an auxiliary heating device affecting the sensing element, e.g. for anticipating change of temperature with provision for adjustment of the effect of the auxiliary heating device, e.g. a function of time
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Automation & Control Theory (AREA)
- Feedback Control In General (AREA)
- Control Of Temperature (AREA)
Abstract
The invention discloses a kind of ultrasonic wave separating apparatus temprature control methods neural network based；The present invention first passes through sensor and carries out data acquisition to the indices of thermostatic control system, to BP neural network training and feedforward control, then the temperature sampling signal in reaction kettle is filtered with filtering algorithm, RBF neural is trained, finally pid parameter is calculated with Fuzzy RBF Neural Network, then pid parameter is transmitted in PID controller, output cooling water valve must control signal, to control the aperture of cooling water valve.Present invention control speed is fast, robustness is high, has overshoot.
Description
Technical field
The present invention relates to thermostatic control algorithm field, specifically a kind of temprature control method neural network based.
Background technique
In recent years, ultrasonic wave separating apparatus has in fields such as traditional Chinese medicine extraction, ore pulp leaching, liquid handling, dispersion, emulsifications
Be widely applied.It is improving product efficiency, shortens the reaction time, reduces system energy consumption etc. and suffers from significant effect
Fruit.But will appear many problems in actual application, wherein temperature control is exactly one of them.Because of ultrasonic wavelength-division
Dissipating instrument to extract the optimum temperature of Chinese medicine is the excessively high stability that will affect extractant of temperature and influence between 40 DEG C to 60 DEG C
Extraction efficiency, so temperature controlling extent is an important index for considering separating apparatus working efficiency.
Temperature control system is often the complication system with non-linear, hysteresis quality, time variation, therefore how to be utilized
It is that we will solve the problems, such as that suitable temperature control system, which carries out temperature control,.Our common thermostatic control system masters now
It to be controlled based on PID control method, the advantages that PID approach principle is simple, adaptable, stability is strong.But it is traditional
PID control method carry out non-linear, hysteresis quality, the control of the temperature of time variation is to be difficult to obtain expected control effect.And
Exotherm rate is fast when due to ultrasonic equipment in actual operation, has to temperature controlled rapidity, stability, robustness
Very high request, so only being insufficient for requiring with PID control.
Summary of the invention
In view of the deficiencies of the prior art, the present invention proposes a kind of ultrasonic wave separating apparatus temperature neural network based to control
Method.
Ultrasonic wave separating apparatus can release a large amount of heat when decomposition in a kettle to material, by reacting
The cooling water pipeline of the periphery of kettle, the temperature in reaction kettle is reduced with the cooling water in thermostat, due to the flow velocity of cooling water
When bigger, the speed of cooling also can be faster, so that we can control the flow velocity of cooling water to control the temperature of reaction kettle.
In feedforward system, feedforward controller is constituted using neural network, so that neural network plays a feedforward compensation
The effect of device.Wherein input is inlet amount for the measured value of interference, exports the control signal for cooling water valve.Because feedovering
In system, system realizes that the condition compensated entirely is to disturbance, and when disturbance q is not zero, the temperature y of output is zero.So when being
When there was only feedforward system in system, it is trained whether the temperature value of output is zero as the training objective of neural network, temperature
The variation of output quantity will guide the adjustment of network weight and output, and temperature is finally made to export the influence of not disturbed momentum change.
In feedback system, temperature value is measured by the temperature sensor in reaction kettle, temperature signal is first passed through into filtering
Algorithm is filtered, and the temperature signal handled and desired temperature signal are then transmitted to Fuzzy RBF Neural Network
In, PID controller is sent to by the pid parameter that Fuzzy RBF Neural Network exports.Cooling water is then controlled according to pid algorithm
The amount of feed of flow valve controls cooling water flow.And then control the temperature of reaction kettle.
The step of the method for the present invention includes:
Step 1 carries out data acquisition by indices of the sensor to thermostatic control system.
With the inlet amount of flow sensor acquisition reaction material, the environment temperature in reaction kettle is acquired with temperature sensor,
With the size of flow in flow sensor acquisition cooling water pipe.
Step 2, BP neural network training and feedforward control.
In feedforward system, feedforward controller is constituted using neural network, so that neural network plays a feedforward compensation
The effect of device.Wherein input is inlet amount for the measured value of interference, exports the control signal for cooling water valve.Because feedovering
In system, system realizes that the condition compensated entirely is to disturbance, and when disturbance q is not zero, the temperature Y of output is zero.So when being
When there was only feedforward system in system, it is trained whether the temperature value of output is zero as the training objective of neural network, temperature
The variation of output quantity will guide the adjustment of network weight and output, and temperature is finally made to export the influence of not disturbed momentum change.
If Y is that output is temperature signal, GPD(S) transmission function between interference volume and controlled variable temperature, GPC(S) become for manipulation
(cooling water flow) is measured to the transmission function between controlled variable temperature, GffIt (S) is feedforward compensation function.After there is disturbance, make
Input of the q as BP neural network must be disturbed, to export a feed-forward control signals to control the aperture of cooling water valve.
Because in the actual production process since the characteristic of controlled device is sufficiently complex, there are many other disturbing factors,
So the mathematical model in ideal disturbance channel and control channel, i.e. G can not be gotPD(S) and GPC(S), to be difficult to find outIt can be by training so that passing through nerve net under the action of disturbance by the way of neural network
The adjusting of network can make the y of output be zero, to reach the effect of feedforward compensation.The following are BP neural network training steps:
STEP2.1, first weight, learning rate, factor of momentum in network are initialized, chooses N number of disturbing signal q conduct
Repetitive exercise of the sample in BP neural network, if the input vector of k-th of sample is
The input and output of hidden layer of STEP2.2, BP neural network are respectively
In formulaFor the corresponding weight of hidden layer, u is the corresponding neuron number of hidden layer.
If the extensive function of BP neural network hidden layer neuron is
The input of output layer of STEP2.3, BP neural network is
Output are as follows:
In formulaFor output layer weight.
STEP2.4, by error functionThat is r (k) is temperature in feedforward system
Desired output Y=0, y (k) are real output value of the system under interference.Whether error in judgement reaches requirement, and instruction is then exited in arrival
Practice, otherwise carries out in next step.
STEP2.5, for input sample, the partial gradient of each layer of neuron of retrospectively calculate, by formula
G ' (x)=g (x) (1-g (x)) in formula, f ' (x)=(1-f2(x))/2, WithAmendment weight respectively between output layer and hidden layer,
The weight of output layer and hidden layer is inscribed when K+1 are as follows:
Revised weight is calculated, next sample value is put into, goes to STEP2.3.
Step 3 is filtered the temperature sampling signal in reaction kettle with filtering algorithm
According to the sample frequency F of signals, with formula Ld=K × Fs×LpCalculate the window width L of median filterd, Lp
For burr pulse width, K is integer type window coefficient,.Obtain window width LdAfterwards, by weight vector W=[W1, W2...,
W2d+1] processing is weighted to data in window.It presses again
L (A)=Med [W1 ※ s (A-d), Wd+1※ s (A) ..., W2d+1※s(A+d)]
It carries out taking median operation, output again is carried out to signal, the processing result conduct after obtaining Weighted median filtering
The input signal of neural network.Wherein Med indicates to take the median of all digits in window, and ※ indicates that duplication is
Step 4 is trained RBF neural.
STEP4.1, the center c for initializing membership function0With width b0And the initial weight of network.Selected learning coefficient
And inertia coeffeicent.
STEP4.2, sampling system given value and reality output calculate error e (k) and error rate ec (k).K is indicated
K-th of sampled point.
STEP4.3, in RBF neural, if input vector X=[x1, x2]T, wherein x1For actual output temperature and phase
Error e (k) between prestige value, x2For error rate ec (k).
It is possible thereby to determine that the number of nodes of the Fuzzy RBF Neural Network second layer is 7 layers, i.e. l=7, the number of nodes of third layer
For 49, i.e. n=49.
First layer is input layer, that is, f1(i)=xi, i=1,2.
The second layer is blurring layer, and fuzzy membership functions selects Gaussian function, and obtained input component is in different moulds
Paste the corresponding degree of membership of Linguistic Value are as follows:
Wherein cij、bijThe center vector and width of the membership function of i-th of input variable, j-th of fuzzy set when respectively.
Third layer is fuzzy reasoning layer, the relevance grade of each node are as follows:
WhereinNiIt is the fuzzy partition function of i-th of input.
4th layer is output layer, i.e. f4(1), f4(2), f4(3) be respectively PID three parameter Kp, Ki, Kd。
Wherein wsdFor export node layer and each node of third layer connection weight matrix, s=1,2,3.The control of PID regulator
System rule are as follows:
Wherein KpFor proportionality coefficient, KiFor integral coefficient, KdFor differential coefficient.
Because of controller Δ u (k)=Kpx(1)+Kix(2)+KdX (3), wherein
X (1)=e (k)-e (k-1)
X (2)=e (k)
X (3)=e (k) -2e (k-1)+e (k-2)
Using increment type PID algorithm
U (k)=u (k-1)+Δ u (k)
Network approximate error target function is quadratic form
Wherein r (k) is desired value namely our given value of temperature, and y (k) is actual output temperature value.Judge E
Whether it is less than threshold epsilon, less than training is then exited, otherwise carries out in next step.
STEP4.4, according to gradient descent method, node sound stage width, node center and weight adjusting can obtain
Wherein η1For learning rate, α1For factor of momentum, value is between 0 to 1.
STEP4.5, next sampled data is taken, returns to STEP4.2.
Step 5 calculates pid parameter with Fuzzy RBF Neural Network.
RBF neural and fuzzy PID algorithm are combined, if r (k) is the Setting signal of input, y (k) is output letter
Number, then it is e (k)=r (k)-y (k) based on the improved PID control error of RBF neural
By x (1), x (2), x (3) are respectively indicated each parameter in pid algorithm.
X (1)=e (k)-e (k-1)
X (2)=e (k)
X (3)=e (k) -2e (k-1)+e (k-2)
By x (1), x (2), x (3), which is updated in incremental timestamp algorithm, to be solved, and obtains control algolithm
U (k)=u (k-1)+Kpx(1)+Kix(2)+Kdx(3)
The training of neural network is designated as
E (k) is brought into the parameter K of incremental digital PID control devicep, Ki, KdExpression formula in
Wherein, learning efficiency is indicated by η, the sensitivity information of system byIt indicates, to obtainWherein wjFor the output weight of RBF fuzzy neural network third layer after training,
cjAnd bjFor the central value and width of each node of fuzzy neural network third layer after training.Then pid parameter is transmitted to PID
In controller, output cooling water valve must control signal, to control the aperture of cooling water valve.
The present invention has effect compared with the existing technology: control speed is fast, and robustness is high, has overshoot.
Detailed description of the invention
Fig. 1 is temperature control system；
Fig. 2 is BP neural network structure chart；
Fig. 3 is fuzzy RBF neural network structure；
Fig. 4 is learning feed-forward control system；
Fig. 5 is advanced PID control strategy.
Specific embodiment
Present invention employs the temperature of the method for feedforward plus feedback control reaction kettle.Structure is as shown in Figure 1.
A kind of ultrasonic wave separating apparatus temprature control method neural network based, this method specifically includes the following steps:
Step 1 carries out data acquisition by indices of the sensor to thermostatic control system.
Data acquisition is carried out to the parameters index in thermostatic control system by sensor, is acquired with flow sensor
The inlet amount of reaction material acquires cooling water pipe with flow sensor with the environment temperature in temperature sensor acquisition reaction kettle
The size of middle flow.
Step 2, BP neural network training and feedforward control.
Training process can be divided into two kinds, i.e. on-line training and off-line training.So-called on-line training is in each sampling
Quarter is all modified the weight of neural network using temperature value output as error, constantly updates network.And off-line training is then
After certain disturbance occurs, entire adjustment process is sampled, to allow temperature value output to go to zero as training objective to net
Network is trained and updates its output, repeats this process until network output reaches requirement.Herein mainly using off-line training
Mode.
As shown in figure 4, it is temperature signal that Y, which is output, it is temperature signal, G for outputPDIt (S) is interference volume and controlled variable
Transmission function between temperature, GPCIt (S) is manipulating variable (cooling water flow) to the transmission function between controlled variable temperature,
GffIt (S) is feedforward compensation function.After there is disturbance, so that input of the disturbance q as BP neural network, thus one feedforward of output
Signal is controlled to control the aperture of cooling water valve.
Because in the actual production process since the characteristic of controlled device is sufficiently complex, there are many other disturbing factors,
So the mathematical model in ideal disturbance channel and control channel, i.e. G can not be gotPD(S) and GPC(S), to be difficult to find outIt can be by training so that passing through nerve net under the action of disturbance by the way of neural network
The adjusting of network can make the y of output be zero, to reach the effect of feedforward compensation.The following are BP neural network training steps:
STFP2.1, in BP neural network, initial input layer number of choosing is 1, i.e., input be disturbance quantity (material
Inlet amount) it is denoted as x (1).So the node number of hidden layer is 3, output quantity y is feed-forward control signals (flow of cooling water),
The structure that BP neural network is arranged is 1-3-1.As shown in Figure 2.
First weight, learning rate, factor of momentum in network are initialized, choose N number of disturbing signal q as sample in BP
Repetitive exercise in neural network, if input vector is
The input and output of hidden layer of STFP2.2, BP neural network are respectively
In formulaFor the corresponding weight of hidden layer, u is the corresponding neuron number of hidden layer.
If the extensive function of BP neural network hidden layer neuron is
The input and output of output layer of STEP2.3, BP neural network are respectively
In formulaFor output layer weight.
STEP2.4, by error functionThat is r (k) is temperature in feedforward system
Desired output Y=0, y (k) are real output value of the system under interference.Whether error in judgement reaches requirement, and instruction is then exited in arrival
Practice, otherwise carries out in next step.
STEP2.5, for input sample, the partial gradient of each layer of neuron of retrospectively calculate, by formula
In formula
G ' (x)=g (x) (1-g (x)), f ' (x)=(1-f2(x))/2,
WithAmendment weight respectively between output layer and hidden layer,
The weight of output layer and hidden layer is inscribed when K+1 are as follows:
Revised weight is calculated, next sample value is put into, goes to STEP2.3.
Step 3 is filtered the temperature sampling signal in reaction kettle with filtering algorithm
It is more so as to cause the burr of temperature signal since the instantaneous relase power of ultrasonic equipment is big, thus firstly the need of
The temperature signal of input is filtered.
According to the sample frequency F of signals, calculate the window width L of median filterd, generally take 2~4 times of burr pulses
Window width of the width as median filter, it may be assumed that
Ld=K × Fs×Lp
Wherein LpFor burr pulse width, K is integer type window coefficient, value range [2,4].
Obtain window width LdAfterwards, by weight vector W=[W1, W2..., W2d+1] place is weighted to data in window
Reason.
By L (A)=Med [W1※ s (A-d), Wd+1※ s (A) ..., W2d+1※ s (A+d)] it carries out taking median operation, to letter
Number carry out output again, input signal of the processing result as neural network after obtaining Weighted median filtering.Wherein Med table
Show that the median for taking all digits in window, ※ indicate that duplication is
This method can allow the window width in median filter freely according to the big freedom in minor affairs of burr width
It adjusts without taking a fixed value as traditional median filtering.
Step 4 is trained RBF neural.
STEP4.1, the center c for initializing membership function0With width b0And the initial weight of network.Selected learning coefficient
And inertia coeffeicent.
STEP4.2, sampling system given value and reality output calculate error e (k) and error rate ec (k).
STEP4.3, in RBF neural, as shown in figure 3, setting input vector X=[x1, x2]T, wherein x1It is practical defeated
Error e (k) between temperature and desired value out, x2For error rate ec (k).K indicates k-th of sampled point.
It is possible thereby to determine that the number of nodes of the Fuzzy RBF Neural Network second layer is 7 layers, i.e. l=7, the number of nodes of third layer
For 49, i.e. n=49.
First layer is input layer, that is, f1(i)=xi, i=1,2.
The second layer is blurring layer, and fuzzy membership functions selects Gaussian function, and obtained input component is in different moulds
Paste the corresponding degree of membership of Linguistic Value are as follows:
Wherein cij、bijThe center vector and width of the membership function of i-th of input variable, j-th of fuzzy set when respectively.
Third layer is fuzzy reasoning layer, the relevance grade of each node are as follows:
WhereinNiIt is the fuzzy partition function of i-th of input.
4th layer is output layer, i.e. f4(1), f4(2), f4(3) be respectively PID three parameter Kp, Ki, Kd。
Wherein wsdFor export node layer and each node of third layer connection weight matrix, s=1,2,3.The control of PID regulator
System rule are as follows:
Wherein KpFor proportionality coefficient, KiFor integral coefficient, KdFor differential coefficient.Since RBF neural has very strong convergence
Property and have very strong complementary characteristic with fuzzy logic, thus we can based on RBF neural for Fuzzy Adaptive PID
Algorithm optimizes.
Because of controller Δ u (k)=Kpx(1)+Kix(2)+KdX (3), wherein
X (1)=e (k)-e (k-1)
X (2)=e (k)
X (3)=e (k) -2e (k-1)+e (k-2)
Using increment type PID algorithm
U (k)=u (k-1)+Δ u (k)
Network approximate error target function is quadratic form
Wherein r (k) is desired value namely our given value of temperature, and y (k) is actual output temperature value.Judge E
Whether it is less than threshold epsilon, less than training is then exited, otherwise carries out in next step.
STEP4.4, according to gradient descent method, node sound stage width, node center and weight adjusting can obtain
Wherein η1For learning rate, α1For factor of momentum, value is between 0 to 1.
STEP4.5, next sampled data is taken, returns to STEP4.2.
Step 5 calculates pid parameter with Fuzzy RBF Neural Network.
RBF neural and fuzzy PID algorithm are combined, as shown in figure 5, setting r (k) as the Setting signal of input, y
(k) it is output signal, then is e (k)=r (k)-y (k) based on the improved PID control error of RBF neural
By x (1), x (2), x (3) are respectively indicated each parameter in pid algorithm.
X (1)=e (k)-e (k-1)
X (2)=e (k)
X (3)=e (k) -2e (k-1)+e (k-2)
By x (1), x (2), x (3), which is updated in incremental timestamp algorithm, to be solved, and obtains control algolithm
U (k)=u (k-1)+Kpx(1)+Kix(2)+Kdx(3)
The training of neural network is designated as
E (k) is brought into the parameter K of incremental digital PID control devicep, Ki, KdExpression formula in
Wherein, learning efficiency is indicated by η, the sensitivity information of system byIt indicates, to obtainWherein wjFor the output weight of RBF fuzzy neural network third layer after training, cj
And bjFor the central value and width of each node of fuzzy neural network third layer after training.Then pid parameter is transmitted to PID control
In device processed, output cooling water valve must control signal, to control the aperture of cooling water valve.
Claims (1)
1. a kind of ultrasonic wave separating apparatus temprature control method neural network based, which is characterized in that this method specifically include with
Lower step:
Step 1 carries out data acquisition by indices of the sensor to thermostatic control system；
With the inlet amount of flow sensor acquisition reaction material, the environment temperature in reaction kettle is acquired with temperature sensor, with stream
Quantity sensor acquires the size of flow in cooling water pipe；
Step 2, BP neural network training and feedforward control；
The following are BP neural network training steps:
STEP2.1, first weight, learning rate, factor of momentum in network are initialized, chooses N number of disturbing signal q as sample
Repetitive exercise in BP neural network, if the input vector of k-th of sample is
The input and output of hidden layer of STEP2.2, BP neural network are respectively
In formulaFor the corresponding weight of hidden layer, u is the corresponding neuron number of hidden layer；
If the extensive function of BP neural network hidden layer neuron is
The input and output of output layer of STEP2.3, BP neural network are respectively
In formulaFor output layer weight；
STEP2.4, by error functionThat is r (k) is expectation of the temperature in feedforward system
Y=0 is exported, y (k) is real output value of the system under interference；Whether error in judgement reaches requirement, and training is then exited in arrival,
Otherwise it carries out in next step；
STEP2.5, for input sample, the partial gradient of each layer of neuron of retrospectively calculate,
By formula
In formula, g ' (x)=g (x) (1-g (x)), f ' (x)=(1-f2(x))/2,
WithAmendment weight respectively between output layer and hidden layer,
The weight of output layer and hidden layer is inscribed when K+1 are as follows:
Revised weight is calculated, next sample value is put into, goes to STEP2.3；
Step 3 is filtered the temperature sampling signal in reaction kettle with filtering algorithm
According to the sample frequency F of signals, with formula Ld=K × Fs×LpCalculate the window width L of median filterd, LpFor hair
Pulse width is pierced, K is integer type window coefficient,；Obtain window width LdAfterwards, by weight vector W=[W1,W2,…,W2d+1] right
Data are weighted processing in window；L (A)=Med [W is pressed again1※s(A-d),Wd+1※s(A),…,W2d+1※ s (A+d)] it carries out
Median operation is taken, output again is carried out to signal, the processing result after obtaining Weighted median filtering is as the defeated of neural network
Enter signal；Wherein Med indicates to take the median of all digits in window, and ※ indicates that duplication is
Step 4 is trained RBF neural；
STEP4.1, the center c for initializing membership function0With width b0And the initial weight of network；It selectes learning coefficient and is used to
Property coefficient；
STEP4.2, sampling system given value and reality output calculate error e (k) and error rate ec (k)；
STEP4.3, in RBF neural, if input vector X=[x1,x2]T, wherein x1For actual output temperature and desired value
Between error e (k), x2For error rate ec (k)；K also illustrates that k-th of sample；
It is possible thereby to determine that the number of nodes of the Fuzzy RBF Neural Network second layer is 7 layers, i.e. l=7, the number of nodes of third layer is 49
It is a, i.e. n=49；
First layer is input layer, that is, f1(i)=xi, i=1,2；
The second layer is blurring layer, and fuzzy membership functions selects Gaussian function, and obtained input component is in different Vague languages
The corresponding degree of membership of speech value are as follows:
Wherein cij、bijThe center vector and width of the membership function of i-th of input variable, j-th of fuzzy set when respectively；
Third layer is fuzzy reasoning layer, the relevance grade of each node are as follows:
WhereinNiIt is the fuzzy partition function of i-th of input；
4th layer is output layer, i.e. f4(1),f4(2),f4(3) be respectively PID three parameter Kp, Ki, Kd；
Wherein wsdFor export node layer and each node of third layer connection weight matrix, s=1,2,3；The control of PID regulator is advised
Rule are as follows:
Wherein KpFor proportionality coefficient, KiFor integral coefficient, KdFor differential coefficient；
Because of controller Δ u (k)=Kpx(1)+Kix(2)+KdX (3), wherein
X (1)=e (k)-e (k-1)
X (2)=e (k)
X (3)=e (k) -2e (k-1)+e (k-2)
Using increment type PID algorithm
U (k)=u (k-1)+Δ u (k)
Network approximate error target function is quadratic form
Wherein r (k) is desired value namely our given value of temperature, and y (k) is actual output temperature value；Whether judge E
Less than threshold epsilon, less than training is then exited, otherwise carry out in next step；STEP4.4, according to gradient descent method, node sound stage width, node
Center and weight adjusting can obtain
Wherein η1For learning rate, α1For factor of momentum, value is between 0 to 1；
STEP4.5, next sampled data is taken, returns to STEP4.2；
Step 5 calculates pid parameter with Fuzzy RBF Neural Network；
RBF neural and fuzzy PID algorithm are combined, if r (k) is the Setting signal of input, y (k) is output signal, then
It is e (k)=r (k)-y (k) based on the improved PID control error of RBF neural
By x (1), x (2), x (3), which is updated in incremental timestamp algorithm, to be solved, and obtains control algolithm
U (k)=u (k-1)+Kpx(1)+Kix(2)+Kdx(3)
The training of neural network is designated as
E (k) is brought into the parameter K of incremental digital PID control devicep, Ki, KdExpression formula in
Wherein, learning efficiency is indicated by η, the sensitivity information of system byIt indicates, to obtainWherein wjFor the output weight of RBF fuzzy neural network third layer after training, cj
And bjFor the central value and width of each node of fuzzy neural network third layer after training；Then pid parameter is transmitted to PID control
In device processed, output cooling water valve must control signal, to control the aperture of cooling water valve.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910092679.4A CN109597449A (en) | 2019-01-30 | 2019-01-30 | A kind of ultrasonic wave separating apparatus temprature control method neural network based |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910092679.4A CN109597449A (en) | 2019-01-30 | 2019-01-30 | A kind of ultrasonic wave separating apparatus temprature control method neural network based |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109597449A true CN109597449A (en) | 2019-04-09 |
Family
ID=65967079
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910092679.4A Withdrawn CN109597449A (en) | 2019-01-30 | 2019-01-30 | A kind of ultrasonic wave separating apparatus temprature control method neural network based |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109597449A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110262582A (en) * | 2019-07-30 | 2019-09-20 | 中原工学院 | A kind of barotor temprature control method based on improvement RBF neural |
| CN110376879A (en) * | 2019-08-16 | 2019-10-25 | 哈尔滨工业大学（深圳） | A kind of PID type iterative learning control method neural network based |
| CN110554715A (en) * | 2019-10-25 | 2019-12-10 | 攀钢集团攀枝花钢铁研究院有限公司 | RBF neural network-based PID control method for hydrolysis process temperature of titanyl sulfate plus seed crystal |
| CN110647186A (en) * | 2019-10-25 | 2020-01-03 | 北京和隆优化科技股份有限公司 | Chloroethylene rectification temperature control method based on fuzzy neural network |
| CN111812968A (en) * | 2020-06-24 | 2020-10-23 | 合肥工业大学 | Fuzzy neural network PID controller-based valve position cascade control method |
| CN111963471A (en) * | 2020-08-14 | 2020-11-20 | 苏州浪潮智能科技有限公司 | Fan rotating speed control method and device |
| CN112953271A (en) * | 2021-03-11 | 2021-06-11 | 上海空间电源研究所 | Space high-power rectification system with active disturbance rejection |
| CN113342069A (en) * | 2021-06-09 | 2021-09-03 | 南京工业大学 | VOCs waste gas flow control method of fuzzy neural network PID (proportion integration differentiation) with heat balance feedback |
| CN113460308A (en) * | 2021-07-30 | 2021-10-01 | 中国农业大学 | Unmanned aerial vehicle variable pesticide application control system and method |
| CN114020063A (en) * | 2021-10-31 | 2022-02-08 | 浙江工业大学 | Fuzzy neural network-based traditional Chinese medicine decoction piece moistening temperature prediction control method |
| CN114296489A (en) * | 2021-12-04 | 2022-04-08 | 北京工业大学 | RBF-PID (radial basis function-proportion integration differentiation) municipal solid waste incineration process hearth temperature control method based on event triggering |
| CN114326375A (en) * | 2021-12-22 | 2022-04-12 | 江苏精瓷智能传感技术研究院有限公司 | Nitrogen oxygen sensor pump unit self-adaptive control system |
| CN114562713A (en) * | 2022-01-17 | 2022-05-31 | 中冶华天南京工程技术有限公司 | Main steam temperature control method and system for power generation boiler |
| CN115090200A (en) * | 2022-05-27 | 2022-09-23 | 福建龙氟化工有限公司 | Automatic batching system for preparing electronic grade hydrofluoric acid and batching method thereof |
| CN115145337A (en) * | 2022-07-27 | 2022-10-04 | 沈阳农业大学 | Plant factory temperature and humidity control method and system |
| CN117316356A (en) * | 2023-10-24 | 2023-12-29 | 中国民航大学 | Feedforward compensation regulation and control method for composite material component autoclave molding process parameters |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104833154A (en) * | 2015-05-28 | 2015-08-12 | 河海大学常州校区 | Chilled water loop control method based on fuzzy PID and neural internal model |
| CN106168815A (en) * | 2016-07-27 | 2016-11-30 | 中冶南方工程技术有限公司 | A kind of acid liquor temperature control system based on Neural network PID and method |
| CN106682735A (en) * | 2017-01-06 | 2017-05-17 | 杭州创族科技有限公司 | BP neural network algorithm based on PID adjustment |
| CN107800129A (en) * | 2017-10-30 | 2018-03-13 | 柳州铁道职业技术学院 | A kind of single Classification of Power Quality Disturbances research method |
| CN108958020A (en) * | 2018-05-29 | 2018-12-07 | 南京理工大学 | A kind of temprature control method based on RBF neural dynamic prediction PID |
| CN109034390A (en) * | 2018-08-07 | 2018-12-18 | 河北工业大学 | Phase angular amplitude PID adaptive approach based on BP neural network Three-Dimensional Magnetic feature measurement |
-
2019
- 2019-01-30 CN CN201910092679.4A patent/CN109597449A/en not_active Withdrawn
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104833154A (en) * | 2015-05-28 | 2015-08-12 | 河海大学常州校区 | Chilled water loop control method based on fuzzy PID and neural internal model |
| CN106168815A (en) * | 2016-07-27 | 2016-11-30 | 中冶南方工程技术有限公司 | A kind of acid liquor temperature control system based on Neural network PID and method |
| CN106682735A (en) * | 2017-01-06 | 2017-05-17 | 杭州创族科技有限公司 | BP neural network algorithm based on PID adjustment |
| CN107800129A (en) * | 2017-10-30 | 2018-03-13 | 柳州铁道职业技术学院 | A kind of single Classification of Power Quality Disturbances research method |
| CN108958020A (en) * | 2018-05-29 | 2018-12-07 | 南京理工大学 | A kind of temprature control method based on RBF neural dynamic prediction PID |
| CN109034390A (en) * | 2018-08-07 | 2018-12-18 | 河北工业大学 | Phase angular amplitude PID adaptive approach based on BP neural network Three-Dimensional Magnetic feature measurement |
Non-Patent Citations (3)
| Title |
|---|
| 宋宝军: "基于中值滤波的冲激引信毛刺脉冲消除方法", 《探测与控制学报》 * |
| 李琳: ""基于RBF神经网络无刷直流电机控制系统的基础研究"", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 韦皓 等: ""基于模糊RBF神经网络的自适应PID励磁控制"", 《企业科技与发展》 * |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110262582A (en) * | 2019-07-30 | 2019-09-20 | 中原工学院 | A kind of barotor temprature control method based on improvement RBF neural |
| CN110376879A (en) * | 2019-08-16 | 2019-10-25 | 哈尔滨工业大学（深圳） | A kind of PID type iterative learning control method neural network based |
| CN110554715A (en) * | 2019-10-25 | 2019-12-10 | 攀钢集团攀枝花钢铁研究院有限公司 | RBF neural network-based PID control method for hydrolysis process temperature of titanyl sulfate plus seed crystal |
| CN110647186A (en) * | 2019-10-25 | 2020-01-03 | 北京和隆优化科技股份有限公司 | Chloroethylene rectification temperature control method based on fuzzy neural network |
| CN111812968B (en) * | 2020-06-24 | 2022-04-22 | 合肥工业大学 | Fuzzy neural network PID controller-based valve position cascade control method |
| CN111812968A (en) * | 2020-06-24 | 2020-10-23 | 合肥工业大学 | Fuzzy neural network PID controller-based valve position cascade control method |
| CN111963471A (en) * | 2020-08-14 | 2020-11-20 | 苏州浪潮智能科技有限公司 | Fan rotating speed control method and device |
| CN112953271A (en) * | 2021-03-11 | 2021-06-11 | 上海空间电源研究所 | Space high-power rectification system with active disturbance rejection |
| CN113342069A (en) * | 2021-06-09 | 2021-09-03 | 南京工业大学 | VOCs waste gas flow control method of fuzzy neural network PID (proportion integration differentiation) with heat balance feedback |
| CN113460308A (en) * | 2021-07-30 | 2021-10-01 | 中国农业大学 | Unmanned aerial vehicle variable pesticide application control system and method |
| CN114020063A (en) * | 2021-10-31 | 2022-02-08 | 浙江工业大学 | Fuzzy neural network-based traditional Chinese medicine decoction piece moistening temperature prediction control method |
| CN114296489A (en) * | 2021-12-04 | 2022-04-08 | 北京工业大学 | RBF-PID (radial basis function-proportion integration differentiation) municipal solid waste incineration process hearth temperature control method based on event triggering |
| CN114296489B (en) * | 2021-12-04 | 2022-09-20 | 北京工业大学 | RBF-PID (radial basis function-proportion integration differentiation) municipal solid waste incineration process hearth temperature control method based on event triggering |
| CN114326375A (en) * | 2021-12-22 | 2022-04-12 | 江苏精瓷智能传感技术研究院有限公司 | Nitrogen oxygen sensor pump unit self-adaptive control system |
| CN114562713A (en) * | 2022-01-17 | 2022-05-31 | 中冶华天南京工程技术有限公司 | Main steam temperature control method and system for power generation boiler |
| CN114562713B (en) * | 2022-01-17 | 2024-04-09 | 中冶华天南京工程技术有限公司 | Main steam temperature control method and system for power generation boiler |
| CN115090200A (en) * | 2022-05-27 | 2022-09-23 | 福建龙氟化工有限公司 | Automatic batching system for preparing electronic grade hydrofluoric acid and batching method thereof |
| CN115145337A (en) * | 2022-07-27 | 2022-10-04 | 沈阳农业大学 | Plant factory temperature and humidity control method and system |
| CN117316356A (en) * | 2023-10-24 | 2023-12-29 | 中国民航大学 | Feedforward compensation regulation and control method for composite material component autoclave molding process parameters |
| CN117316356B (en) * | 2023-10-24 | 2024-05-17 | 中国民航大学 | Feedforward compensation regulation and control method for composite material component autoclave molding process parameters |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109597449A (en) | A kind of ultrasonic wave separating apparatus temprature control method neural network based | |
| CN109670580A (en) | A kind of data recovery method based on time series | |
| CN105807632B (en) | A kind of furnace temp controller design method based on improvement dahlin algorithm | |
| CN105911862B (en) | A kind of temperature Control of Electric Heater method | |
| CN110687800B (en) | Data-driven self-adaptive anti-interference controller and estimation method thereof | |
| CN110347192B (en) | Glass furnace temperature intelligent prediction control method based on attention mechanism and self-encoder | |
| CN108536005A (en) | One kind being based on Fuzzy Neural PID ship course keeping control device and its control method | |
| CN107868979A (en) | A kind of silicon single crystal diameter control method based on permanent casting speed control structure | |
| CN106054596B (en) | It is a kind of that setting method is optimized based on the PID controller parameter for improving performance indicator | |
| CN111812968A (en) | Fuzzy neural network PID controller-based valve position cascade control method | |
| CN103557037A (en) | Method for controlling rotating speed of steam turbine on basis of self-adaptive inverse control | |
| CN105676637B (en) | Molecular weight output PDF control methods based on Predictive function control | |
| CN108719516A (en) | One kind being based on RBF neural tea machine processing control parameter intelligence setting method | |
| Kergus et al. | Frequency-domain data-driven control design in the Loewner framework | |
| CN106842948A (en) | The method for optimally controlling of the HDP molecular distillation systems based on BP networks | |
| Chtourou et al. | Control of a bioreactor using a neural network | |
| CN108549213A (en) | A kind of speed ratio control method based on RBF neural PID | |
| CN117455320B (en) | QGMVPN model-based multi-kettle polymerization process melt quality index online prediction method | |
| CN113759700A (en) | Fractional order PID self-adaptive adjustment method based on particle swarm and neural network | |
| CN103675010B (en) | The industrial melt index soft measurement instrument of support vector machine and method | |
| CN106371321A (en) | PID control method for fuzzy network optimization of coking-furnace hearth pressure system | |
| CN114200840B (en) | Traditional Chinese medicine pharmaceutical process operation optimization method based on distributed model predictive control | |
| CN106940526A (en) | It is a kind of to go the pseudo- carbon fiber coagulation bath technique controlled | |
| CN1047671C (en) | Predictive controlling method and system thereof for controlling object being controlled | |
| Tsoneva et al. | Augmented Lagrange decomposition method for optimal control calculation of batch fermentation processes |
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 | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190409 |
|
| WW01 | Invention patent application withdrawn after publication |