CN105446132A - Sewage treatment prediction control method based on neural network - Google Patents
Sewage treatment prediction control method based on neural network Download PDFInfo
- Publication number
- CN105446132A CN105446132A CN201210011168.3A CN201210011168A CN105446132A CN 105446132 A CN105446132 A CN 105446132A CN 201210011168 A CN201210011168 A CN 201210011168A CN 105446132 A CN105446132 A CN 105446132A
- Authority
- CN
- China
- Prior art keywords
- value
- moment
- formula
- neural network
- input
- 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
Landscapes
- Feedback Control In General (AREA)
Abstract
The invention discloses a sewage treatment prediction control method based on a neural network, and belongs to the technical field of sewage treatment. The method is characterized in that the method solves a problem of the control of the dissolved oxygen concentration of a fifth partitioned region and the nitrate nitrogen concentration of a second partitioned region on the basis of the global layout, which is the same as the layout of a reference simulation model (Benchmark Simulation Model No.1, BSM1) developed by International Water Association (IWA), of an active sludge and waste water treatment device through employing the technology of neural network prediction control. The invention proposes a multi-variable prediction control system based on the neural network, and the system mainly comprises two parts: a neural network recognizer which is used for extracting output data of an object; and a neural network controller which is used for outputting a control variable. The method improves the adaptability and robustness of the system.
Description
Technical field
The present invention utilizes network response surface technology, at the Benchmark Simulation Model (BenchmarkSimulationModelNo.1 assisting (IWA) to develop with international water, BSM1), on the active sludge sewage-treatment plant total arrangement basis that layout is the same, the control problem of the nitrate of the 5th subregion dissolved oxygen concentration and the second subregion is solved.Sewage disposal process controls the important step as wastewater treatment, is the important branch in advanced manufacturing technology field, both belongs to water treatment field, belong to control field again.
Background technology
Along with the growth of national economy and the enhancing of Public environmental attitude, wastewater treatment automatic technology has welcome unprecedented opportunity to develop.Propose to study in country's medium & long term sci-tech development program and promote wastewater treatment and control new technology.Therefore, achievement in research of the present invention has broad application prospects.
Sewage disposal process is a complicated biochemical reaction process, is attended by conversion and the transmittance process of physical-chemical reaction, biochemical reaction, phase transition process and material and energy, and process is complicated, and modeling difficulty, causes its process control more difficult.A lot of scholar, on the basis of BSM1, has carried out large quantifier elimination to the process control of wastewater treatment.Find that fuzzy control lacks self-learning capability based on research, adaptability is poor, and control accuracy is not high; Model Predictive Control is the PREDICTIVE CONTROL based on mathematical model, and model accuracy is not high, and practical application is more difficult.And the neural network be made up of neuron has powerful non-linear mapping capability and learning functionality, PREDICTIVE CONTROL can be carried out to nonlinear system well.And the research that most research is only carried out for dissolved oxygen DO.But in A/O technique, DO (dissolved oxygen DO) concentration of aeration zone and the S of oxygen-starved area
nO(nitrate nitrogen) concentration is the important parameter affecting nitration denitrification process.DO excessive concentration in aeration zone, can cause the DO entering oxygen-starved area to increase, cannot ensure the anaerobic environment needed for denitrification, increase the consumption of oxygen-starved area organic carbon capable of being fast degraded, thus affect treatment effect.Equally, suitable oxygen-starved area S is maintained
nOconcentration, can the denitrification of efficiency utilization oxygen-starved area, avoids too high inner circulating reflux amount simultaneously, improves denitrogenation clearance and also reduces power consumption.The control of nitrate and the control of dissolved oxygen concentration are the important parameters improving sewage disposal system treatment effect.
Summary of the invention
The present invention obtains a kind of based on neural network sewage disposal process control method, is the Multivariable Predictive Control System based on neural network, mainly comprises two parts in system: the output data of neural network identifier-extraction object; Nerve network controller-output control variable.
It is characterized in that, comprise the following steps:
1, process controller;
Active sludge sewage-treatment plant total arrangement comprises biochemical reaction tank and second pond.Biochemical reaction tank part comprises 5 subregions altogether, and the first two subregion is oxygen-starved area, and rear three subregions are aeration zones.In each subregion, all with Q
krepresent flow, Z
krepresent the concentration of each component, Z=(S
i, S
s, X
i, X
s, X
b, H, X
p, S
nO, S
nH, S
nD, X
nD, S
aLK), S
irepresent the not biodegradable organic concentration of dissolubility, unit gCOD.m
-3; S
srepresent dissolubility biodegradable organic concentration, unit gCOD.m
-3; X
irepresent not biodegradable organic concentration, unit gCOD.m
-3; X
srepresent biodegradable organic concentration, unit gCOD.m
-3; X
b, Hrepresent active heterotroph biosolids concentration, unit gCOD.m
-3; X
prepresent the inert concentration that biosolids decay produces, unit gCOD.m
-3; S
nOrepresent the nitrate nitrogen concentration in water outlet, unit gN.m
-3; S
nHrepresent the NH in water outlet
4-N and NH
3the total concentration of-N, unit gN.m
-3; S
nDrepresent the biodegradable organic nitrogen concentration of dissolubility, unit gN.m
-3; X
nDrepresent the biodegradable organic nitrogen concentration of graininess, unit gN.m
-3; S
aLKrepresent basicity, unit mol.m
-3.The volume V of two unit in oxygen-starved area
1=V
2, the volume V of three unit in aeration zone
3=V
4=V
5.At the external reflux amount Q being sewage, returning from second pond of the first subregion input
rwith capacity of returns Q in the 5th subregion
asummation Q
1; At the input Q of the second subregion
2the amount Q flowed into by the first subregion
1; The input Q of the 3rd subregion
3the amount Q flowed into by the second subregion
2; The input Q of the 4th subregion
4the amount Q flowed into by the 3rd subregion
3; The input Q of the 5th subregion
5the amount Q flowed into by the 4th subregion
4; The input Q of second pond
fthe amount Q flowed into by the 5th subregion
5with interior capacity of returns Q
apoor Q
5-Q
a; Water outlet Q is divided into after sedimentation in secondary sedimentation tank
e, mud discharging Q
wwith external reflux amount Q
r.Controller 1 controls the dissolved oxygen concentration of the 5th subregion, and controller 2 controls the concentration of the nitrate nitrogen of the second subregion.
2, the k moment, the control of dissolved oxygen concentration and nitrate;
2.1 and 2.2 carry out simultaneously below, and 2.6 and 2.7 carry out simultaneously.
The control of 2.1 dissolved oxygen concentrations
Input is the value y of the dissolved oxygen concentration that k-1 moment actual sewage processing procedure exports
1and the oxygen conversion coefficient K of k-1 moment the 5th subregion (k-1)
lathe value u of 5
1(k-1); Output is the oxygen conversion coefficient K of k moment the 5th subregion
lathe value u of 5
1(k).Three layers of BP network are utilized to set up network response surface device NPC
1, the input of neural network is:
z
1(k)=[u
1(k-1),y
1(k-1)]
T(1)
In formula, u
1(k-1) the oxygen conversion coefficient K of the 5th subregion in k-1 moment is referred to
lathe value of 5; y
1(k-1) value after the dissolved oxygen concentration that actual sewage processing procedure exports is referred to.
Being expressed as follows of neural network:
In formula, i
1refer to input layer,
it is input layer i-th
1individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
1refer to neural network output layer neuron,
the transition function of hidden layer,
the weights of hidden layer to output layer.
Output is: u
1(k), u
1k () refers to the oxygen conversion coefficient K of the 5th subregion in k moment
lathe value of 5
The control of 2.2 nitrates
Input is the value y of the nitrate that k-1 moment actual sewage processing procedure exports
2(k-1) capacity of returns Q and in the k-1 moment
avalue u
2(k-1); Output is capacity of returns Q in the k moment
avalue u
2(k).
Adopt and set up network response surface device with three layers of BP network, the input of neural network is:
z
2(k)=[u
2(k-1),y
2(k-1)]
T(4)
In formula, u
2(k-1) capacity of returns Q in the k-1 moment is referred to
avalue; y
2(k-1) be actual sewage processing procedure export nitrate after value.
Being expressed as follows of neural network:
In formula, i
2refer to neural network input layer neuron,
it is input layer i-th
2individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
2refer to neural network output layer neuron,
the transition function of hidden layer,
the weights of hidden layer to output layer.
Output is: u
2(k), u
2k () refers to capacity of returns Q in the k moment
avalue.
2.3 neural network prediction dissolved oxygen concentration and nitrates
Input is the oxygen conversion coefficient K of k moment the 5th subregion
lathe value u of 5
1(k) and interior capacity of returns Q
avalue u
2the value y of k dissolved oxygen concentration that () and k-1 moment actual sewage processing procedure export
1and the value y of nitrate (k-1)
2(k-1).
Neural network adopts three layers of BP neural network, and it comprises input layer (i
3layer), hidden layer (h
3layer) and output layer (j
3layer), it is input as:
x(k)=[u
1(k),u
2(k),y
1(k-1),y
2(k-1)]
T(7)
Being expressed as follows of neural network:
In formula, i
3refer to input layer,
represent input layer i-th
3individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
3refer to neural network output layer neuron,
the transition function of hidden layer,
the weights of hidden layer to output layer.
Output is: the value of the dissolved oxygen concentration of etching system during prediction k
with the value of nitrate
obtain output valve matrix y^ (k).
2.4 by the oxygen conversion coefficient K of the 5th subregion in k moment
lathe value u of 5
1(k) and interior capacity of returns Q
avalue u
2k (), as the input of sewage disposal process, obtains matrix y (k) at the dissolved oxygen concentration in k moment and the value of nitrate.
The initial weight of 2.5 correction neural network identifier NNI, the performance index function of definition following formula:
In formula: e (k) is
with the error of y (k),
be the matrix of the output valve of etching system during prediction k, y (k) is the matrix of actual sewage processing procedure in the output valve in k moment.
Weights are upgraded by following formula
with
In formula, η is a positive learning rate, η ∈ (0,1].
2.6 couples of NPC
1the initial value of weights carry out on-line tuning, the performance index function of definition following formula:
In formula, 2 are output valves of dissolved oxygen concentration is empirical value,
it is the output valve of the dissolved oxygen concentration of k moment neural network prediction.
Revised the weights of network response surface device by dynamic BP algorithm, formula is as follows:
In formula, η
c1a positive learning rate, η
c1∈ (0,1].
2.7 couples of NPC
2the initial value of weights carry out on-line tuning, the performance index function of definition following formula:
In formula, 1 be the output valve of nitrate is empirical value,
it is the output valve of the nitrate of k moment neural network prediction.
Network response surface device NPC is revised by dynamic BP algorithm
2weights, formula is as follows:
In formula, η
c2a positive learning rate, η
c2∈ (0,1].
3, the k+1 moment, the control of dissolved oxygen concentration and nitrate
3.1 and 3.2 carry out simultaneously below, and 3.4 and 3.5 carry out simultaneously.
The control of 3.1 dissolved oxygen concentrations
Input is the value y of the dissolved oxygen concentration that k moment actual sewage processing procedure exports
1(k) and k moment the 5th oxygen conversion coefficient K of subregion
lathe value u of 5
1(k); Output is the oxygen conversion coefficient K of k+1 moment the 5th subregion
lathe value u of 5
1(k+1).Three layers of BP network are utilized to set up network response surface device NPC
1, the input of neural network is:
z
1(k+1)=[u
1(k),y
1(k)]
T(16)
In formula, u
1k () refers to the oxygen conversion coefficient K of the 5th subregion in k moment
lathe value of 5; y
1k () refers to the value after the dissolved oxygen concentration that actual sewage processing procedure exports.
Being expressed as follows of neural network:
In formula, i
1refer to input layer,
it is input layer i-th
1individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
1refer to neural network output layer neuron,
the transition function of hidden layer,
the weights of hidden layer to output layer.
Output is: u
1(k+1), u
1(k+1) the oxygen conversion coefficient K of the 5th subregion in k+1 moment is referred to
lathe value of 5.
The control of 3.2 nitrates
Input is the value y of the nitrate that k moment actual sewage processing procedure exports
2capacity of returns Q in (k) and k moment
avalue u
2(k); Output is capacity of returns Q in the k moment
avalue u
2(k+1).
Adopt and set up network response surface device with three layers of BP network, the input of neural network is:
z
2(k+1)=[u
2(k),y
2(k)]
T(19)
In formula, u
2k () refers to capacity of returns Q in the k moment
avalue; y
2k () is the value after the nitrate of actual sewage processing procedure output.
Being expressed as follows of neural network:
In formula, i
2refer to neural network input layer neuron,
it is input layer i-th
2individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
2refer to neural network output layer neuron,
the transition function of hidden layer,
the weights of hidden layer to output layer.
Output is: u
2(k+1).U
2(k+1) capacity of returns Q in the k+1 moment is referred to
avalue.
3.3 neural network prediction dissolved oxygen concentration and nitrates
Input is the oxygen conversion coefficient K of k+1 moment the 5th subregion
lathe value u of 5
1and interior capacity of returns Q (k+1)
avalue u
2and the value y of dissolved oxygen concentration that exports of k moment actual sewage processing procedure (k+1)
1the value y of (k) and nitrate
2(k).
Neural network adopts three layers of BP neural network, and it comprises input layer (i
3layer), hidden layer (h
3layer) and output layer (j
3layer), it is input as:
x(k+1)=[u
1(k+1),u
2(k+1),y
1(k),y
2(k)]
T(22)
Being expressed as follows of neural network:
In formula, i
3refer to input layer,
represent input layer i-th
3individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
3refer to neural network output layer neuron,
the transition function of hidden layer,
the weights of hidden layer to output layer.
Output is: the value of the dissolved oxygen concentration of etching system during prediction k+1
with the value of nitrate
obtain predicting output matrix
3.4NPC
1modified weight
On-line tuning is carried out to the weights of nerve network controller.
In formula, 2 are output valves of dissolved oxygen concentration is empirical value,
be the output valve of the dissolved oxygen concentration of k+1 moment neural network prediction, 0.01 is controlled quentity controlled variable weighted value, Δ u
1(k+1) be the oxygen conversion coefficient K of the 5th subregion
lathe difference in adjacent two moment of 5 values, Δ u
1(k+1)=u
1(k+1)-u
1(k).
Revised the weights of network response surface device by dynamic BP algorithm, formula is as follows:
In formula, η
c1a positive learning rate, η
c1∈ (0,1].
3.5NPC
2modified weight
On-line tuning is carried out to the weights of nerve network controller.
In formula, 1 be the output valve of nitrate is empirical value,
be the output valve of the nitrate of k+1 moment neural network prediction, 0.01 is controlled quentity controlled variable weighted value, Δ u
2(k+1) be interior capacity of returns Q
athe difference in adjacent two moment of value, Δ u
2(k+1)=u
2(k+1)-u
2(k).
Network response surface device NPC is revised by dynamic BP algorithm
2weights, formula is as follows:
In formula, η
c2a positive learning rate, η
c2∈ (0,1].
4, the k+1+1 step in k+1 moment is predicted
K in step 3.1 to step 3.3 is added 1 by 4.1, becomes k+1, repeated execution of steps 3.1 to step 3.3, obtains the oxygen conversion coefficient K of the 5th subregion during k+2
lathe value u of 5
1(k+2), interior capacity of returns Q
avalue u
2and obtain the value of dissolved oxygen concentration of prognoses system (k+2)
with the value of nitrate
connect step 4.2 again.
4.2NPC
1modified weight
On-line tuning is carried out to the weights of nerve network controller.
In formula, 2 are output valves of dissolved oxygen concentration is empirical value,
the output valve of the dissolved oxygen concentration of neural network prediction when being k+2,0.01 is controlled quentity controlled variable weighted value, Δ u
1(k+2) be the oxygen conversion coefficient K of the 5th subregion
lathe difference in adjacent two moment of 5 values, Δ u
1(k+2)=u
1(k+2)-u
1(k+1).
Revised the weights of network response surface device by dynamic BP algorithm, formula is as follows:
In formula, η
c1a positive learning rate, η
c1∈ (0,1].
4.3NPC
2modified weight
On-line tuning is carried out to the weights of nerve network controller.
In formula, 1 be the output valve of nitrate is empirical value,
be the output valve of the nitrate of k+2 moment neural network prediction, 0.01 is controlled quentity controlled variable weighted value, Δ u
2(k+2) be interior capacity of returns Q
athe difference in adjacent two moment of value, Δ u
2(k+2)=u
2(k+2)-u
2(k+1).
Network response surface device NPC is revised by dynamic BP algorithm
2weights, formula is as follows:
In formula, η
c2a positive learning rate, η
c2∈ (0,1].
5, the k+1+2 step in k+1 moment is predicted
K in step 3.1 to step 3.3 is added 2 by 5.1, becomes k+2, repeated execution of steps 3.1 to step 3.3, obtains the oxygen conversion coefficient K of the 5th subregion during k+3
lathe value u of 5
1(k+3), interior capacity of returns Q
avalue u
2and obtain the value of dissolved oxygen concentration of prognoses system (k+3)
with the value of nitrate
connect step 4.2 again.
5.2NPC
1modified weight
On-line tuning is carried out to the weights of nerve network controller.
In formula, 2 are output valves of dissolved oxygen concentration is empirical value,
the output valve of the dissolved oxygen concentration of neural network prediction when being k+3,0.01 is controlled quentity controlled variable weighted value, Δ u
1(k+2) be the oxygen conversion coefficient K of the 5th subregion
lathe difference in adjacent two moment of 5 values, Δ u
1(k+3)=u
1(k+3)-u
1(k+2).
Revised the weights of network response surface device by dynamic BP algorithm, formula is as follows:
In formula, η
c1a positive learning rate, η
c1∈ (0,1].
5.3NPC
2modified weight
On-line tuning is carried out to the weights of nerve network controller.
In formula, 1 be the output valve of nitrate is empirical value,
be the output valve of the nitrate of k+2 moment neural network prediction, 0.01 is controlled quentity controlled variable weighted value, Δ u
2(k+3) be interior capacity of returns Q
athe difference in adjacent two moment of value, Δ u
2(k+2)=u
2(k+2)-u
2(k+1).
Network response surface device NPC is revised by dynamic BP algorithm
2weights, formula is as follows:
In formula, η
c2a positive learning rate, η
c2∈ (0,1].
The oxygen conversion coefficient K of the 5th subregion 6, k+1+2 walked
lathe value u of 5
1and interior capacity of returns Q (k+3)
avalue as the input of k+1 moment sewage disposal process, obtain the matrix y (k+1) of the dissolved oxygen concentration in k+1 moment and the value of nitrate.
7, the modified weight of neural network identifier NNI
The weights of online updating NNI, define following performance index function:
In formula: e (k+1) is
with the error of y (k+1),
be the matrix of the output valve of etching system during prediction k+1, y (k+1) is the matrix of actual sewage processing procedure in the output valve in k+1 moment.
Upgrade weights according to the following formula
with
In formula, η is a positive learning rate, η ∈ (0,1].
8, the k in step 3 to step 7 is added n, become k+n, repeat step 3 to step 7, n=1,2,3,4..., circulate.
The present invention compared with prior art, has following obvious advantage and beneficial effect:
The present invention adopts network response surface scheme, control dissolved oxygen concentration and nitrate, although its algorithm is more complicated, but its dynamic property and adaptivity are all better than regulatory PID control algorithm, there is good adaptivity and robustness, solve Model Predictive Control model out of true and fuzzy control shortage self-learning capability, the shortcoming that adaptability is poor, improves control accuracy.While control dissolved oxygen DO, nitrate nitrogen also being controlled, so also just solving the problem that in causing because only controlling dissolved oxygen concentration, capacity of returns increases.Advantage of the present invention is as follows:
(1) Neural Networks Predictive Control Algorithm is adopted, in existing sewage disposal process control technology, Neural Network Control Algorithm is applied to the control of a variable, and successfully solve at this realization utilizing Neural Network Control Algorithm to control Two Variables, and due to sewage disposal process be one complicated, dynamic bioprocesses, not only have non-linear, the feature such as change time large, and there is strong coupling relation between each factor, and the neural network that neuron is formed has powerful non-linear mapping capability and learning functionality, so the control to nonlinear system can be realized well relative to other control method Neural Networks Predictive Control Algorithms.Neural Networks Predictive Control Algorithm is successfully applied in other areas, as Electric Machine Control etc.But relative to the feature of wastewater treatment control procedure, network response surface can play a role better.
(2) dissolved oxygen concentration and nitrate Two Variables is controlled, in sewage disposal process controls, generally control for dissolved oxygen concentration, but the control of nitrate also should be noted that in biochemical reaction process, therefore want the control technology for other, add the control of nitrate here.
To note especially: the present invention just for convenience, employing be to dissolved oxygen concentration S
owith the concentration of nitrate nitrogen S
nOcontrol, this invention is also applicable to sludge reflux amount, sludge discharge etc. equally, carries out controlling all should belong to scope of the present invention as long as have employed principle of the present invention.
Accompanying drawing explanation
Fig. 1 is active sludge wastewater treatment Benchmark Simulation Model of the present invention;
Fig. 2 is network response surface structured flowchart of the present invention, wherein, in figure
with
in p be corresponding with the step number in the prediction k+1 moment in claims, p=1,2;
Fig. 3 is neural network prediction dissolved oxygen concentration Error Graph of the present invention;
Fig. 4 is neural network prediction nitrate Error Graph of the present invention;
Fig. 5 is Dissolved Oxygen concentration Control comparison diagram of the present invention;
Fig. 6 is that nitrate of the present invention controls comparison diagram.
Embodiment
The present invention obtains a kind of method based on network response surface, this control method is by the output of Holy Bible network identifier prediction dissolved oxygen concentration, nitrate, feed back to nerve network controller, the nitrate of the second subregion and the dissolved oxygen concentration of the 5th subregion are controlled.
It is characterized in that, comprise the following steps:
1, process controller;
Active sludge sewage-treatment plant total arrangement comprises biochemical reaction tank and second pond.Biochemical reaction tank part comprises 5 subregions altogether, and the first two subregion is oxygen-starved area, and rear three subregions are aeration zones.In each subregion, all with Q
krepresent flow, Z
krepresent the concentration of each component, Z=(S
i, S
s, X
i, X
s, X
b, H, X
p, S
nO, S
nH, S
nD, X
nD, S
aLK), S
irepresent the not biodegradable organic concentration of dissolubility, unit gCOD.m
-3; S
srepresent dissolubility biodegradable organic concentration, unit gCOD.m
-3; X
irepresent not biodegradable organic concentration, unit gCOD.m
-3; X
srepresent biodegradable organic concentration, unit gCOD.m
-3; X
b, Hrepresent active heterotroph biosolids concentration, unit gCOD.m
-3; X
prepresent the inert concentration that biosolids decay produces, unit gCOD.m
-3; S
nOrepresent the nitrate nitrogen concentration in water outlet, unit gN.m
-3; S
nHrepresent the NH in water outlet
4-N and NH
3the total concentration of-N, unit gN.m
-3; S
nDrepresent the biodegradable organic nitrogen concentration of dissolubility, unit gN.m
-3; XND represents the biodegradable organic nitrogen concentration of graininess, unit gN.m
-3; S
aLKrepresent basicity, unit mol.m
-3.The volume V of two unit in oxygen-starved area
1=V
2, the volume V of three unit in aeration zone
3=V
4=V
5.At the external reflux amount Q being sewage, returning from second pond of the first subregion input
rwith capacity of returns Q in the 5th subregion
asummation Q
1; At the input Q of the second subregion
2the amount Q flowed into by the first subregion
1; The input Q of the 3rd subregion
3the amount Q flowed into by the second subregion
2; The input Q of the 4th subregion
4the amount Q flowed into by the 3rd subregion
3; The input Q of the 5th subregion
5the amount Q flowed into by the 4th subregion
4; The input Q of second pond
fthe amount Q flowed into by the 5th subregion
5with interior capacity of returns Q
apoor Q
5-Q
a; Water outlet Q is divided into after sedimentation in secondary sedimentation tank
e, mud discharging Q
wwith external reflux amount Q
r.Controller 1 controls the dissolved oxygen concentration of the 5th subregion, and controller 2 controls the concentration of the nitrate nitrogen of the second subregion.
2, the k moment, the control of dissolved oxygen concentration and nitrate;
2.1 and 2.2 carry out simultaneously below, and 2.6 and 2.7 carry out simultaneously.
The control of 2.1 dissolved oxygen concentrations
Input is the value y of the dissolved oxygen concentration that k-1 moment actual sewage processing procedure exports
1and the oxygen conversion coefficient K of k-1 moment the 5th subregion (k-1)
lathe value u of 5
1(k-1); Output is the oxygen conversion coefficient K of k moment the 5th subregion
lathe value u of 5
1(k).Three layers of BP network are utilized to set up network response surface device NPC
1, the input of neural network is:
z
1(k)=[u
1(k-1),y
1(k-1)]
T(1)
In formula, u
1(k-1) the oxygen conversion coefficient K of the 5th subregion in k-1 moment is referred to
lathe value of 5; y
1(k-1) value after the dissolved oxygen concentration that actual sewage processing procedure exports is referred to.
Being expressed as follows of neural network:
In formula, i
1refer to input layer, i
1=2,
it is input layer i-th
1individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
1refer to neural network output layer neuron, j
1=1,
the transition function of hidden layer,
the weights of hidden layer to output layer.
Output is: u
1(k), u
1k () refers to the oxygen conversion coefficient K of the 5th subregion in k moment
lathe value of 5
The control of 2.2 nitrates
Input is the value y of the nitrate that k-1 moment actual sewage processing procedure exports
2(k-1) capacity of returns Q and in the k-1 moment
avalue u
2(k-1); Output is capacity of returns Q in the k moment
avalue u
2(k).
Adopt and set up network response surface device with three layers of BP network, the input of neural network is:
z
2(k)=[u
2(k-1),y
2(k-1)]
T(4)
In formula, u
2(k-1) capacity of returns Q in the k-1 moment is referred to
avalue; y
2(k-1) be actual sewage processing procedure export nitrate after value.
Being expressed as follows of neural network:
In formula, i
2refer to neural network input layer neuron, i
2=2,
it is input layer i-th
2individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
2refer to neural network output layer neuron, j
2=1,
the transition function of hidden layer,
the weights of hidden layer to output layer.
Output is: u
2(k), u
2k () refers to capacity of returns Q in the k moment
avalue.
2.3 neural network prediction dissolved oxygen concentration and nitrates
Input is the oxygen conversion coefficient K of k moment the 5th subregion
lathe value u of 5
1(k) and interior capacity of returns Q
avalue u
2the value y of k dissolved oxygen concentration that () and k-1 moment actual sewage processing procedure export
1and the value y of nitrate (k-1)
2(k-1).
Neural network adopts three layers of BP neural network, and it comprises input layer (i
3layer), hidden layer (h
3layer) and output layer (j
3layer), it is input as:
x(k)=[u
1(k),u
2(k),y
1(k-1),y
2(k-1)]
T(7)
Being expressed as follows of neural network:
In formula, i
3refer to input layer, i
3=4,
represent input layer i-th
3individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
3refer to neural network output layer neuron, j
3=2,
the transition function of hidden layer,
the weights of hidden layer to output layer.
Output is: the value of the dissolved oxygen concentration of etching system during prediction k
with the value of nitrate
obtain output valve matrix
2.4 by the oxygen conversion coefficient K of the 5th subregion in k moment
lathe value u of 5
1(k) and interior capacity of returns Q
avalue u
2k (), as the input of sewage disposal process, obtains matrix y (k) at the dissolved oxygen concentration in k moment and the value of nitrate.
The initial weight of 2.5 correction neural network identifier NNI, the performance index function of definition following formula:
In formula: e (k) is
with the error of y (k),
be the matrix of the output valve of etching system during prediction k, y (k) is the matrix of actual sewage processing procedure in the output valve in k moment.
Weights are upgraded by following formula
with
In formula, η is a positive learning rate, η=0.2.
2.6 couples of NPC
1the initial value of weights carry out on-line tuning, the performance index function of definition following formula:
In formula, 2 are output valves of dissolved oxygen concentration is empirical value,
it is the output valve of the dissolved oxygen concentration of k moment neural network prediction.
Revised the weights of network response surface device by dynamic BP algorithm, formula is as follows:
In formula, η
c1a positive learning rate, η
c1=0.5.
2.7 couples of NPC
2the initial value of weights carry out on-line tuning, the performance index function of definition following formula:
In formula, 1 be the output valve of nitrate is empirical value,
it is the output valve that the nitrate nitrogen of k moment neural network prediction is dense.
Network response surface device NPC is revised by dynamic BP algorithm
2weights, formula is as follows:
In formula, η
c2a positive learning rate, η
c2=0.5.
3, the k+1 moment, the control of dissolved oxygen concentration and nitrate
3.1 and 3.2 carry out simultaneously below, and 3.4 and 3.5 carry out simultaneously.
The control of 3.1 dissolved oxygen concentrations
Input is the value y of the dissolved oxygen concentration that k moment actual sewage processing procedure exports
1(k) and k moment the 5th oxygen conversion coefficient K of subregion
lathe value u of 5
1(k); Output is the oxygen conversion coefficient K of k+1 moment the 5th subregion
lathe value u of 5
1(k+1).Three layers of BP network are utilized to set up network response surface device NPC
1, the input of neural network is:
z
1(k+1)=[u
1(k),y
1(k)]
T(16)
In formula, u
1k () refers to the oxygen conversion coefficient K of the 5th subregion in k moment
lathe value of 5; y
1k () refers to the value after the dissolved oxygen concentration that actual sewage processing procedure exports.
Being expressed as follows of neural network:
In formula, i
1refer to input layer, i
1=2,
it is input layer i-th
1individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
1refer to neural network output layer neuron, j
1=1,
the transition function of hidden layer,
the weights of hidden layer to output layer.
Output is: u
1(k+1), u
1(k+1) the oxygen conversion coefficient K of the 5th subregion in k+1 moment is referred to
lathe value of 5.
The control of 3.2 nitrates
Input is the value y of the nitrate that k moment actual sewage processing procedure exports
2capacity of returns Q in (k) and k moment
avalue u
2(k); Output is capacity of returns Q in the k moment
avalue u
2(k+1).
Adopt and set up network response surface device with three layers of BP network, the input of neural network is:
z
2(k+1)=[u
2(k),y
2(k)]
T(19)
In formula, u
2k () refers to capacity of returns Q in the k moment
avalue; y
2k () is the value after the nitrate of actual sewage processing procedure output.
Being expressed as follows of neural network:
In formula, i
2refer to neural network input layer neuron, i
2=2,
it is input layer i-th
2individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
2refer to neural network output layer neuron, j
2=1,
the transition function of hidden layer,
the weights of hidden layer to output layer.
Output is: u
2(k+1), u
2(k+1) capacity of returns Q in the k+1 moment is referred to
avalue.
3.3 neural network prediction dissolved oxygen concentration and nitrates
Input is the oxygen conversion coefficient K of k+1 moment the 5th subregion
lathe value u of 5
1and interior capacity of returns Q (k+1)
avalue u
2and the value y of dissolved oxygen concentration that exports of k moment actual sewage processing procedure (k+1)
1the value y of (k) and nitrate
2(k).
Neural network adopts three layers of BP neural network, and it comprises input layer (i
3layer), hidden layer (h
3layer) and output layer (j
3layer), it is input as:
x(k+1)=[u
1(k+1),u
2(k+1),y
1(k),y
2(k)]
T(22)
Being expressed as follows of neural network:
In formula, i
3refer to input layer, i
3=4,
represent input layer i-th
3individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
3refer to output layer neuron, j
3=2,
the transition function of hidden layer,
the weights of hidden layer to output layer.
Output is: the value of the dissolved oxygen concentration of etching system during prediction k+1
with the value of nitrate
obtain predicting output matrix
3.4NPC
1modified weight
On-line tuning is carried out to the weights of nerve network controller.
In formula, 2 are output valves of dissolved oxygen concentration is empirical value,
be the output valve of the dissolved oxygen concentration of k+1 moment neural network prediction, 0.01 is controlled quentity controlled variable weighted value, Δ u
1(k+1) be the oxygen conversion coefficient K of the 5th subregion
lathe difference in adjacent two moment of 5 values, Δ u
1(k+1)=u
1(k+1)-u
1(k).
Revised the weights of network response surface device by dynamic BP algorithm, formula is as follows:
In formula, η
c1a positive learning rate, η
c1=0.5.
3.5NPC
2modified weight
On-line tuning is carried out to the weights of nerve network controller.
In formula, 1 be the output valve of nitrate is empirical value,
be the output valve of the nitrate of k+1 moment neural network prediction, 0.01 is controlled quentity controlled variable weighted value, Δ u
2(k+1) be interior capacity of returns Q
athe difference in adjacent two moment of value, Δ u
2(k+1)=u
2(k+1)-u
2(k).
Network response surface device NPC is revised by dynamic BP algorithm
2weights, formula is as follows:
In formula, η
c2a positive learning rate, η
c2=0.5.
4, the k+1+1 step in k+1 moment is predicted
K in step 3.1 to step 3.3 is added 1 by 4.1, becomes k+1, repeated execution of steps 3.1 to step 3.3, obtains the oxygen conversion coefficient K of the 5th subregion during k+2
lathe value u of 5
1(k+2), interior capacity of returns Q
avalue u
2and obtain the value of dissolved oxygen concentration of prognoses system (k+2)
with the value of nitrate
connect step 4.2 again.
4.2NPC
1modified weight
On-line tuning is carried out to the weights of nerve network controller.
In formula, 2 are output valves of dissolved oxygen concentration is empirical value,
the output valve of the dissolved oxygen concentration of neural network prediction when being k+2,0.01 is controlled quentity controlled variable weighted value, Δ u
1(k+2) be the oxygen conversion coefficient K of the 5th subregion
lathe difference in adjacent two moment of 5 values, Δ u
1(k+2)=u
1(k+2)-u
1(k+1).
Revised the weights of network response surface device by dynamic BP algorithm, formula is as follows:
In formula, η
c1a positive learning rate, η
c1=0.5.
4.3NPC
2modified weight
On-line tuning is carried out to the weights of nerve network controller.
In formula, 1 be the output valve of nitrate is empirical value,
be the output valve of the nitrate of k+2 moment neural network prediction, 0.01 is controlled quentity controlled variable weighted value, Δ u
2(k+2) be interior capacity of returns Q
athe difference in adjacent two moment of value, Δ u
2(k+2)=u
2(k+2)-u
2(k+1).
Network response surface device NPC is revised by dynamic BP algorithm
2weights, formula is as follows:
In formula, η
c2a positive learning rate, η
c2=0.5.
5, the k+1+2 step in k+1 moment is predicted
K in step 3.1 to step 3.3 is added 2 by 5.1, becomes k+2, repeated execution of steps 3.1 to step 3.3, obtains the oxygen conversion coefficient K of the 5th subregion during k+3
lathe value u of 5
1(k+3), interior capacity of returns Q
avalue u
2and obtain the value of dissolved oxygen concentration of prognoses system (k+3)
with the value of nitrate
connect step 4.2 again.
5.2NPC
1modified weight
On-line tuning is carried out to the weights of nerve network controller.
In formula, 2 are output valves of dissolved oxygen concentration is empirical value,
the output valve of the dissolved oxygen concentration of neural network prediction when being k+3,0.01 is controlled quentity controlled variable weighted value, Δ u
1(k+2) be the oxygen conversion coefficient K of the 5th subregion
lathe difference in adjacent two moment of 5 values, Δ u
1(k+3)=u
1(k+3)-u
1(k+2).
Revised the weights of network response surface device by dynamic BP algorithm, formula is as follows:
In formula, η
c1a positive learning rate, η
c1=0.5.
5.3NPC
2modified weight
On-line tuning is carried out to the weights of nerve network controller.
In formula, 1 be the output valve of nitrate is empirical value,
be the output valve of the nitrate of k+2 moment neural network prediction, 0.01 is controlled quentity controlled variable weighted value, Δ u
2(k+3) be interior capacity of returns Q
athe difference in adjacent two moment of value, Δ u
2(k+2)=u
2(k+2)-u
2(k+1).
Network response surface device NPC is revised by dynamic BP algorithm
2weights, formula is as follows:
In formula, η
c2a positive learning rate, η
c2=0.5.
The oxygen conversion coefficient K of the 5th subregion 6, k+1+2 walked
lathe value u of 5
1and interior capacity of returns Q (k+3)
avalue as the input of k+1 moment sewage disposal process, obtain the matrix y (k+1) of the dissolved oxygen concentration in k+1 moment and the value of nitrate.
7, the modified weight of neural network identifier NNI
The weights of online updating NNI, define following performance index function:
In formula: e (k+1) is
with the error of y (k+1),
be the matrix of the output valve of etching system during prediction k+1, y (k+1) is the matrix of actual sewage processing procedure in the output valve in k+1 moment.
Upgrade weights according to the following formula
with
In formula, η is a positive learning rate, η=0.2.
8, the k in step 3 to step 7 is added n, become k+n, repeat step 3 to step 7, n=1,2,3,4..., circulate.
Based on network response surface method, control the dissolved oxygen concentration of the 5th subregion and the nitrate of the second subregion in activated sludge process transaction module BSM1; Fig. 3 is neural network prediction dissolved oxygen concentration Error Graph; Fig. 4 is neural network prediction nitrate Error Graph; Fig. 5 is Dissolved Oxygen concentration Control comparison diagram; Fig. 6 is that nitrate controls comparison diagram.In Fig. 5, Fig. 6, red curve is the control result of Traditional PID, and blue curve is the result of network response surface.The K of the conventional incremental timestamp device of dissolved oxygen concentration
p, K
i, K
dbe taken as 10 respectively, 2.0,0.5; And nitrate be taken as 20000,4000,500.From Fig. 3 and Fig. 4, the precision of neural network prediction is higher, and error all controls in 0.1 substantially; And having Fig. 5 and Fig. 6 known, the concentration of dissolved oxygen DO and the concentration of nitrate nitrogen also all can control on 2mg/L and 1mg/L preferably.Compared with only controlling dissolved oxygen concentration with employing PREDICTIVE CONTROL, not only dissolved oxygen DO controls in desired range of values, and nitrate nitrogen also controls in expectation value, and control accuracy is higher.And as can be seen from the figure, after reaching steady state (SS), little compared with Traditional PID of the error of network response surface dissolved oxygen concentration, also Two Variables can be controlled preferably in setting value, control accuracy is higher, has good dynamic perfromance, also has stronger antijamming capability.
Claims (1)
1., based on the wastewater treatment forecast Control Algorithm of neural network, it is characterized in that, comprise the following steps:
1), process controller;
Active sludge sewage-treatment plant total arrangement comprises biochemical reaction tank and second pond; Biochemical reaction tank part comprises 5 subregions altogether, and the first two subregion is oxygen-starved area, and rear three subregions are aeration zones; In each subregion, all with Q
krepresent flow, Z
krepresent the concentration of each component, Z=(S
i, S
s, X
i, X
s, X
b, H, X
p, S
nO, S
nH, S
nD, X
nD, S
aLK), S
ithe not biodegradable organic concentration of property, unit gCOD.m
-3; S
srepresent dissolubility biodegradable organic concentration, unit gCOD.m
-3; X
irepresent not biodegradable organic concentration, unit gCOD.m
-3; X
srepresent biodegradable organic concentration, unit gCOD.m
-3; X
b, Hrepresent active heterotroph biosolids concentration, unit gCOD.m
-3; X
prepresent the inert concentration that biosolids decay produces, unit gCOD.m
-3; S
nOrepresent the nitrate nitrogen concentration in water outlet, unit gN.m
-3; S
nHrepresent the NH in water outlet
4-N and NH
3the total concentration of-N, unit gN.m
-3; S
nDrepresent the biodegradable organic nitrogen concentration of dissolubility, unit gN.m
-3; X
nDrepresent the biodegradable organic nitrogen concentration of graininess, unit gN.m
-3; S
aLKrepresent basicity, unit mol.m
-3; The volume V of two unit in oxygen-starved area
1=V
2, the volume V of three unit in aeration zone
3=V
4=V
5; At the external reflux amount Q being sewage, returning from second pond of the first subregion input
rwith capacity of returns Q in the 5th subregion
asummation Q
1; At the input Q of the second subregion
2the amount Q flowed into by the first subregion
1; The input Q of the 3rd subregion
3the amount Q flowed into by the second subregion
2; The input Q of the 4th subregion
4the amount Q flowed into by the 3rd subregion
3; The input Q of the 5th subregion
5the amount Q flowed into by the 4th subregion
4; The input Q of second pond
fthe amount Q flowed into by the 5th subregion
5with interior capacity of returns Q
apoor Q
5-Q
a; Water outlet Q is divided into after sedimentation in secondary sedimentation tank
e, mud discharging Q
wwith external reflux amount Q
r; Controller 1 controls the dissolved oxygen concentration of the 5th subregion, and controller 2 controls the concentration of the nitrate nitrogen of the second subregion;
2), the k moment, the control of dissolved oxygen concentration and nitrate;
2.1 and 2.2 carry out simultaneously below, and 2.6 and 2.7 carry out simultaneously;
The control of 2.1 dissolved oxygen concentrations
Input is the value y of the dissolved oxygen concentration that k-1 moment actual sewage processing procedure exports
1and the oxygen conversion coefficient K of k-1 moment the 5th subregion (k-1)
lathe value u of 5
1(k-1); Output is the oxygen conversion coefficient K of k moment the 5th subregion
lathe value u of 5
1(k); Three layers of BP network are utilized to set up network response surface device NPC
1, the input of neural network is:
z
1(k)=[u
1(k-1),y
1(k-1)]
T(1)
In formula, u
1(k-1) the oxygen conversion coefficient K of the 5th subregion in k-1 moment is referred to
lathe value of 5; y
1(k-1) value after the dissolved oxygen concentration that actual sewage processing procedure exports is referred to;
Being expressed as follows of neural network:
In formula, i
1refer to input layer,
it is input layer i-th
1individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
1refer to neural network output layer neuron,
the transition function of hidden layer,
the weights of hidden layer to output layer;
Output is: u
1(k), u
1k () refers to the oxygen conversion coefficient K of the 5th subregion in k moment
lathe value of 5
The control of 2.2 nitrates
Input is the value y of the nitrate that k-1 moment actual sewage processing procedure exports
2(k-1) capacity of returns Q and in the k-1 moment
avalue u
2(k-1); Output is capacity of returns Q in the k moment
avalue u
2(k);
Adopt and set up network response surface device with three layers of BP network, the input of neural network is:
z
2(k)=[u
2(k-1),y
2(k-1)]
T(4)
In formula, u
2(k-1) capacity of returns Q in the k-1 moment is referred to
avalue; y
2(k-1) be actual sewage processing procedure export nitrate after value;
Being expressed as follows of neural network:
In formula, i
2refer to neural network input layer neuron,
it is input layer i-th
2individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
2refer to neural network output layer neuron,
the transition function of hidden layer,
the weights of hidden layer to output layer;
Output is: u
2(k), u
2k () refers to capacity of returns Q in the k moment
avalue;
2.3 neural network prediction dissolved oxygen concentration and nitrates
Input is the oxygen conversion coefficient K of k moment the 5th subregion
lathe value u of 5
1(k) and interior capacity of returns Q
avalue u
2the value y of k dissolved oxygen concentration that () and k-1 moment actual sewage processing procedure export
1and the value y of nitrate (k-1)
2(k-1);
Neural network adopts three layers of BP neural network, and it comprises input layer (i
3layer), hidden layer (h
3layer) and output layer (j
3layer), it is input as:
x(k)=[u
1(k),u
2(k),y
1(k-1),y
2(k-1)]
T(7)
Being expressed as follows of neural network:
In formula, i
3refer to input layer,
represent input layer i-th
3individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
3refer to neural network output layer neuron,
the transition function of hidden layer,
the weights of hidden layer to output layer;
Output is: the value of the dissolved oxygen concentration of etching system during prediction k
with the value of nitrate
obtain output valve matrix
2.4 by the oxygen conversion coefficient K of the 5th subregion in k moment
lathe value u of 5
1(k) and interior capacity of returns Q
avalue u
2k (), as the input of sewage disposal process, obtains matrix y (k) at the dissolved oxygen concentration in k moment and the value of nitrate;
The initial weight of 2.5 correction neural network identifier NNI, the performance index function of definition following formula:
In formula: e (k) is
with the error of y (k),
be the matrix of the output valve of etching system during prediction k, y (k) is the matrix of actual sewage processing procedure in the output valve in k moment;
Weights are upgraded by following formula
with
In formula, η is a positive learning rate, η ∈ (0,1];
2.6 couples of NPC
1the initial value of weights carry out on-line tuning, the performance index function of definition following formula:
In formula, 2 are output valves of dissolved oxygen concentration is empirical value,
it is the output valve of the dissolved oxygen concentration of k moment neural network prediction;
Revised the weights of network response surface device by dynamic BP algorithm, formula is as follows:
In formula, η
c1a positive learning rate, η
c1∈ (0,1];
2.7 couples of NPC
2the initial value of weights carry out on-line tuning, the performance index function of definition following formula:
In formula, 1 be the output valve of nitrate is empirical value,
it is the output valve of the nitrate of k moment neural network prediction;
Network response surface device NPC is revised by dynamic BP algorithm
2weights, formula is as follows:
In formula, η
c2a positive learning rate, η
c2∈ (0,1];
3), the k+1 moment, the control of dissolved oxygen concentration and nitrate
3.1 and 3.2 carry out simultaneously below, and 3.4 and 3.5 carry out simultaneously;
The control of 3.1 dissolved oxygen concentrations
Input is the value y of the dissolved oxygen concentration that k moment actual sewage processing procedure exports
1(k) and k moment the 5th oxygen conversion coefficient K of subregion
lathe value u of 5
1(k); Output is the oxygen conversion coefficient K of k+1 moment the 5th subregion
lathe value u of 5
1(k+1); Three layers of BP network are utilized to set up network response surface device NPC
1, the input of neural network is:
z
1(k+1)=[u
1(k),y
1(k)]
T(16)
In formula, u
1k () refers to the oxygen conversion coefficient K of the 5th subregion in k moment
lathe value of 5; y
1k () refers to the value after the dissolved oxygen concentration that actual sewage processing procedure exports;
Being expressed as follows of neural network:
In formula, i
1refer to input layer,
it is input layer i-th
1individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
1refer to neural network output layer neuron,
the transition function of hidden layer,
the weights of hidden layer to output layer;
Output is: u
1(k+1), u
1(k+1) the oxygen conversion coefficient K of the 5th subregion in k+1 moment is referred to
lathe value of 5;
The control of 3.2 nitrates
Input is the value y of the nitrate that k moment actual sewage processing procedure exports
2capacity of returns Q in (k) and k moment
avalue u
2(k); Output is capacity of returns Q in the k moment
avalue u
2(k+1);
Adopt and set up network response surface device with three layers of BP network, the input of neural network is:
z
2(k+1)=[u
2(k),y
2(k)]
T(19)
In formula, u
2k () refers to capacity of returns Q in the k moment
avalue; y
2k () is the value after the nitrate of actual sewage processing procedure output;
Being expressed as follows of neural network:
In formula, i
2refer to neural network input layer neuron,
it is input layer i-th
2individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
2refer to neural network output layer neuron,
the transition function of hidden layer,
the weights of hidden layer to output layer;
Output is: u
2(k+1), u
2(k+1) capacity of returns Q in the k+1 moment is referred to
avalue;
3.3 neural network prediction dissolved oxygen concentration and nitrates
Input is the oxygen conversion coefficient K of k+1 moment the 5th subregion
lathe value u of 5
1and the value u of interior capacity of returns Qa (k+1)
2and the value y of dissolved oxygen concentration that exports of k moment actual sewage processing procedure (k+1)
1the value y of (k) and nitrate
2(k);
Neural network adopts three layers of BP neural network, and it comprises input layer (i
3layer), hidden layer (h
3layer) and output layer (j
3layer), it is input as:
x(k+1)=[u
1(k+1),u
2(k+1),y
1(k),y
2(k)]
T(22)
Being expressed as follows of neural network:
In formula, i
3refer to input layer,
represent input layer i-th
3individual neuronic input,
the weights of input layer to hidden layer,
the number of hidden layer neuron,
j
3refer to neural network output layer neuron,
the transition function of hidden layer,
the weights of hidden layer to output layer;
Output is: the value of the dissolved oxygen concentration of etching system during prediction k+1
with the value of nitrate
obtain predicting output matrix
3.4NPC
1modified weight
On-line tuning is carried out to the weights of nerve network controller;
In formula, 2 are output valves of dissolved oxygen concentration is empirical value,
be the output valve of the dissolved oxygen concentration of k+1 moment neural network prediction, 0.01 is controlled quentity controlled variable weighted value, Δ u
1(k+1) be the oxygen conversion coefficient K of the 5th subregion
lathe difference in adjacent two moment of 5 values, Δ u
1(k+1)=u
1(k+1)-u
1(k);
Revised the weights of network response surface device by dynamic BP algorithm, formula is as follows:
In formula, η
c1a positive learning rate, η
c1∈ (0,1];
3.5NPC
2modified weight
On-line tuning is carried out to the weights of nerve network controller;
In formula, 1 be the output valve of nitrate is empirical value,
be the output valve of the nitrate of k+1 moment neural network prediction, 0.01 is controlled quentity controlled variable weighted value, Δ u
2(k+1) be interior capacity of returns Q
athe difference in adjacent two moment of value, Δ u
2(k+1)=u
2(k+1)-u
2(k);
Network response surface device NPC is revised by dynamic BP algorithm
2weights, formula is as follows:
In formula, η
c2a positive learning rate, η
c2∈ (0,1];
4) the k+1+1 step in k+1 moment, is predicted
K in step 3.1 to step 3.3 is added 1 by 4.1, becomes k+1, repeated execution of steps 3.1 to step 3.3, obtains the oxygen conversion coefficient K of the 5th subregion during k+2
lathe value u of 5
1(k+2), interior capacity of returns Q
avalue u
2and obtain the value of dissolved oxygen concentration of prognoses system (k+2)
with the value of nitrate
connect step 4.2 again;
4.2NPC
1modified weight
On-line tuning is carried out to the weights of nerve network controller;
In formula, 2 are output valves of dissolved oxygen concentration is empirical value,
the output valve of the dissolved oxygen concentration of neural network prediction when being k+2,0.01 is controlled quentity controlled variable weighted value, Δ u
1(k+2) be the oxygen conversion coefficient K of the 5th subregion
lathe difference in adjacent two moment of 5 values, Δ u
1(k+2)=u
1(k+2)-u
1(k+1);
Revised the weights of network response surface device by dynamic BP algorithm, formula is as follows:
In formula, η
c1a positive learning rate, η
c1∈ (0,1];
4.3NPC
2modified weight
On-line tuning is carried out to the weights of nerve network controller;
In formula, 1 be the output valve of nitrate is empirical value,
be the output valve of the nitrate of k+2 moment neural network prediction, 0.01 is controlled quentity controlled variable weighted value, Δ u
2(k+2) be interior capacity of returns Q
athe difference in adjacent two moment of value, Δ u
2(k+2)=u
2(k+2)-u
2(k+1);
Network response surface device NPC is revised by dynamic BP algorithm
2weights, formula is as follows:
In formula, η
c2a positive learning rate, η
c2∈ (0,1];
5) the k+1+2 step in k+1 moment, is predicted
K in step 3.1 to step 3.3 is added 2 by 5.1, becomes k+2, repeated execution of steps 3.1 to step 3.3, obtains the oxygen conversion coefficient K of the 5th subregion during k+3
lathe value u of 5
1(k+3), interior capacity of returns Q
avalue u
2and obtain the value of dissolved oxygen concentration of prognoses system (k+3)
with the value of nitrate
connect step 4.2 again;
The modified weight of 5.2NPC1
On-line tuning is carried out to the weights of nerve network controller;
In formula, 2 are output valves of dissolved oxygen concentration is empirical value,
the output valve of the dissolved oxygen concentration of neural network prediction when being k+3,0.01 is controlled quentity controlled variable weighted value, Δ u
1(k+2) be the oxygen conversion coefficient K of the 5th subregion
lathe difference in adjacent two moment of 5 values, Δ u
1(k+3)=u
1(k+3)-u
1(k+2);
Revised the weights of network response surface device by dynamic BP algorithm, formula is as follows:
In formula, η
c1a positive learning rate, η
c1∈ (0,1];
5.3NPC
2modified weight
On-line tuning is carried out to the weights of nerve network controller;
In formula, 1 be the output valve of nitrate is empirical value,
be the output valve of the nitrate of k+2 moment neural network prediction, 0.01 is controlled quentity controlled variable weighted value, Δ u
2(k+3) be interior capacity of returns Q
athe difference in adjacent two moment of value, Δ u
2(k+2)=u
2(k+2)-u
2(k+1);
Network response surface device NPC is revised by dynamic BP algorithm
2weights, formula is as follows:
In formula, η
c2a positive learning rate, η
c2∈ (0,1];
6) the oxygen conversion coefficient K of the 5th subregion, k+1+2 walked
lathe value u of 5
1and interior capacity of returns Q (k+3)
avalue as the input of k+1 moment sewage disposal process, obtain the matrix y (k+1) of the dissolved oxygen concentration in k+1 moment and the value of nitrate;
7), the modified weight of neural network identifier NNI
The weights of online updating NNI, define following performance index function:
In formula: e (k+1) is
with the error of y (k+1),
be the matrix of the output valve of etching system during prediction k+1, y (k+1) is the matrix of actual sewage processing procedure in the output valve in k+1 moment;
Upgrade weights according to the following formula
with
In formula, η is a positive learning rate, η ∈ (0,1];
8), by step 3) to step 7) in k add n, become k+n, repeat step 3) to step 7), n=1,2,3,4 ..., circulate.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210011168.3A CN105446132A (en) | 2012-01-13 | 2012-01-13 | Sewage treatment prediction control method based on neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210011168.3A CN105446132A (en) | 2012-01-13 | 2012-01-13 | Sewage treatment prediction control method based on neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105446132A true CN105446132A (en) | 2016-03-30 |
Family
ID=55556469
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201210011168.3A Pending CN105446132A (en) | 2012-01-13 | 2012-01-13 | Sewage treatment prediction control method based on neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105446132A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106096286A (en) * | 2016-06-15 | 2016-11-09 | 北京千安哲信息技术有限公司 | Clinical path formulating method and device |
CN106354019A (en) * | 2016-11-21 | 2017-01-25 | 北京工业大学 | Accurate control method for dissolved oxygen based on RBF neural network |
CN109523094A (en) * | 2018-12-21 | 2019-03-26 | 哈尔滨工程大学 | A kind of ship saline sewage processing control forecasting system and prediction technique based on wavelet neural network |
CN113031445A (en) * | 2021-03-12 | 2021-06-25 | 北京工业大学 | Mechanism model-based robust multivariable control method for sewage denitrification process |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6535795B1 (en) * | 1999-08-09 | 2003-03-18 | Baker Hughes Incorporated | Method for chemical addition utilizing adaptive optimization |
JP2004000986A (en) * | 1996-05-31 | 2004-01-08 | Mitsubishi Electric Corp | Method for controlling biological water treatment apparatus |
CN101387632A (en) * | 2008-10-17 | 2009-03-18 | 北京工业大学 | Soft measurement method for biochemical oxygen demand BOD in process of sewage disposal |
KR20090078501A (en) * | 2008-01-15 | 2009-07-20 | 부산대학교 산학협력단 | Apparatus and method for prediction of influent flow rate and influent components using artificial neural network |
CN101819407A (en) * | 2010-04-02 | 2010-09-01 | 杭州电子科技大学 | Sewage pump station water level prediction method base on neural network |
CN101923083A (en) * | 2009-06-17 | 2010-12-22 | 复旦大学 | Sewage chemical oxygen demand soft measuring method based on support vector machine and neural network |
CN102122134A (en) * | 2011-02-14 | 2011-07-13 | 华南理工大学 | Method and system for wastewater treatment of dissolved oxygen control based on fuzzy neural network |
-
2012
- 2012-01-13 CN CN201210011168.3A patent/CN105446132A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2004000986A (en) * | 1996-05-31 | 2004-01-08 | Mitsubishi Electric Corp | Method for controlling biological water treatment apparatus |
US6535795B1 (en) * | 1999-08-09 | 2003-03-18 | Baker Hughes Incorporated | Method for chemical addition utilizing adaptive optimization |
KR20090078501A (en) * | 2008-01-15 | 2009-07-20 | 부산대학교 산학협력단 | Apparatus and method for prediction of influent flow rate and influent components using artificial neural network |
CN101387632A (en) * | 2008-10-17 | 2009-03-18 | 北京工业大学 | Soft measurement method for biochemical oxygen demand BOD in process of sewage disposal |
CN101923083A (en) * | 2009-06-17 | 2010-12-22 | 复旦大学 | Sewage chemical oxygen demand soft measuring method based on support vector machine and neural network |
CN101819407A (en) * | 2010-04-02 | 2010-09-01 | 杭州电子科技大学 | Sewage pump station water level prediction method base on neural network |
CN102122134A (en) * | 2011-02-14 | 2011-07-13 | 华南理工大学 | Method and system for wastewater treatment of dissolved oxygen control based on fuzzy neural network |
Non-Patent Citations (3)
Title |
---|
逄泽芳 等: "基于神经网络的污水处理预测控制", 《吉林大学学报(工学版)》 * |
韩广 等: "污水处理过程中溶解氧的模糊神经网络控制", 《山东大学学报(工学版)》 * |
韩红桂 等: "动态结构优化神经网络及其在溶解氧控制中的应用", 《信息与控制》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106096286A (en) * | 2016-06-15 | 2016-11-09 | 北京千安哲信息技术有限公司 | Clinical path formulating method and device |
CN106354019A (en) * | 2016-11-21 | 2017-01-25 | 北京工业大学 | Accurate control method for dissolved oxygen based on RBF neural network |
CN106354019B (en) * | 2016-11-21 | 2019-01-22 | 北京工业大学 | A kind of dissolved oxygen accuracy control method based on RBF neural |
CN109523094A (en) * | 2018-12-21 | 2019-03-26 | 哈尔滨工程大学 | A kind of ship saline sewage processing control forecasting system and prediction technique based on wavelet neural network |
CN109523094B (en) * | 2018-12-21 | 2023-07-25 | 哈尔滨工程大学 | Ship salty sewage treatment control prediction system and prediction method based on wavelet neural network |
CN113031445A (en) * | 2021-03-12 | 2021-06-25 | 北京工业大学 | Mechanism model-based robust multivariable control method for sewage denitrification process |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Bashar et al. | Cost effectiveness of phosphorus removal processes in municipal wastewater treatment | |
CN106873379B (en) | Sewage treatment optimal control method based on iterative ADP algorithm | |
CN106698642B (en) | Multi-target real-time optimization control method in sewage treatment process | |
CN108549234B (en) | Multi-objective optimization control method based on dynamic variable values | |
CN103197544B (en) | Sewage disposal process multi-purpose control method based on nonlinear model prediction | |
CN105404151A (en) | Sewage processing process dynamic multi-target optimization control method | |
CN106802563B (en) | A kind of sewage procedure optimization control method based on drosophila optimization and LSSVM | |
CN101777084B (en) | Optimization design method of sewage treatment plant A2/O process | |
CN108445757B (en) | Sewage treatment process optimization control method based on dynamic multi-target particle swarm algorithm | |
CN111367181B (en) | Hybrid drive intelligent judgment control method for sewage treatment system | |
CN108762082B (en) | Sewage treatment process collaborative optimization control system | |
Qiao et al. | Adaptive optimal control for a wastewater treatment plant based on a data-driven method | |
CN115356930B (en) | Multi-objective optimization control system and method in sewage treatment process | |
Do et al. | A design of higher-level control based genetic algorithms for wastewater treatment plants | |
Solon et al. | Process schemes for future energy-positive water resource recovery facilities | |
CN110032755B (en) | Multi-objective optimization method for urban sewage treatment process under multiple working conditions | |
CN105446132A (en) | Sewage treatment prediction control method based on neural network | |
Seco et al. | Plant-wide modelling in wastewater treatment: showcasing experiences using the Biological Nutrient Removal Model | |
Nam et al. | A multi-agent AI reinforcement-based digital multi-solution for optimal operation of a full-scale wastewater treatment plant under various influent conditions | |
CN111797363B (en) | Self-learning track tracking method of sewage treatment system based on data | |
CN101201592A (en) | Control simulation method for waste water treatment process as well as simulation method thereof | |
Nazif et al. | Developing a practical model for the optimal operation of wastewater treatment plant considering influent characteristics | |
Xie et al. | Modeling and multi-objective optimization for ANAMMOX process under COD disturbance using hybrid intelligent algorithm | |
Rajaei et al. | Improving wastewater treatment plant performance based on effluent quality, operational costs, and reliability using control strategies for water and sludge lines | |
Heo et al. | End-to-end autonomous and resilient operability strategy of full-scale PN-SBR system: From influent augmentation to AI-aided optimal control and scheduling |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20160330 |