CN113283481A - Intelligent membrane pollution decision-making method based on knowledge type-two fuzzy - Google Patents

Intelligent membrane pollution decision-making method based on knowledge type-two fuzzy Download PDF

Info

Publication number
CN113283481A
CN113283481A CN202110528334.6A CN202110528334A CN113283481A CN 113283481 A CN113283481 A CN 113283481A CN 202110528334 A CN202110528334 A CN 202110528334A CN 113283481 A CN113283481 A CN 113283481A
Authority
CN
China
Prior art keywords
time
output
membrane
layer
decision
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
Application number
CN202110528334.6A
Other languages
Chinese (zh)
Inventor
魏向义
彭泽东
卜晓军
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Qunzhi Future Artificial Intelligence Technology Research Institute Wuxi Co ltd
Original Assignee
Qunzhi Future Artificial Intelligence Technology Research Institute Wuxi Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Qunzhi Future Artificial Intelligence Technology Research Institute Wuxi Co ltd filed Critical Qunzhi Future Artificial Intelligence Technology Research Institute Wuxi Co ltd
Priority to CN202110528334.6A priority Critical patent/CN113283481A/en
Publication of CN113283481A publication Critical patent/CN113283481A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F3/00—Biological treatment of water, waste water, or sewage
    • 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
    • G06N3/045—Combinations of networks
    • 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
    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F2303/00—Specific treatment goals
    • C02F2303/14—Maintenance of water treatment installations

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Microbiology (AREA)
  • Hydrology & Water Resources (AREA)
  • Environmental & Geological Engineering (AREA)
  • Water Supply & Treatment (AREA)
  • Chemical & Material Sciences (AREA)
  • Organic Chemistry (AREA)
  • Feedback Control In General (AREA)

Abstract

The invention provides a knowledge-based two-type fuzzy membrane pollution intelligent decision method aiming at the problem that the membrane pollution is difficult to accurately decide in the MBR urban sewage treatment process so as to reduce the occurrence of membrane pollution; the method summarizes the prior experience knowledge in the MBR sewage treatment plant, and establishes a membrane pollution decision knowledge base in a fuzzy rule form; initializing parameter design of affiliation layers of the interval two-type fuzzy neural network model by using fuzzy rules, and completing parameter adjustment of the model by using a migration gradient descent algorithm to improve the learning performance of the model; by providing accurate operation suggestions for membrane pollution, the membrane pollution decision-making precision is improved, the membrane pollution is slowed down, and the stable operation of the MBR municipal sewage treatment process is promoted.

Description

Intelligent membrane pollution decision-making method based on knowledge type-two fuzzy
Technical Field
The invention realizes the decision of membrane pollution in the MBR municipal sewage treatment process by using a knowledge-based two-type fuzzy method so as to reduce the occurrence of membrane pollution. A decision-making method based on data and knowledge is applied to a nonlinear MBR sewage treatment system, an operation suggestion is provided for membrane pollution, stable and safe operation of an MBR municipal sewage treatment process is guaranteed, and the method belongs to the technical field of sewage treatment and also belongs to the field of intelligent decision-making. Therefore, intelligent decision making of membrane fouling is of great significance in sewage treatment systems.
Background
The '2020 world water development report' issued by the united nations water system indicates that: the global water consumption is increased by about six times compared with the past 100 years at present, and the value is still steadily increased by 1% per year along with the increase of population, economic development and change of consumption mode. According to the forecast of water conservancy departments, the total water shortage in China will reach 600 billions of cubic meters in 2040 years along with the increase of the total population, and the per capita water resource will be reduced to 1630 cubic meters. The water resource shortage and water pollution crisis are basic problems restricting the sustainable development of social economy and are problems to be solved urgently in the water environment treatment process of China. The recycling of the urban sewage can effectively relieve the problems of water resource shortage, water resource waste and the like. How to improve the urban sewage treatment capacity and level is one of the research focuses on realizing the continuous utilization of water resources, relieving the shortage of water resources and building a water-saving society.
The MBR sewage treatment process is a novel sewage treatment technology combining a high-efficiency membrane separation technology and a traditional activated sludge process, and is recognized by all countries in the world as one of the most potential development technologies in the field of water treatment in the twenty-first century. The MBR technology can achieve higher nutrient removal efficiency and complete biomass retention without secondary clarification, and improves the sewage treatment capacity and efficiency. With the continuous maturity of membrane technology, MBR sewage treatment process is more and more widely applied in municipal sewage treatment. However, the membrane pollution problem is inevitable, which may result in increased operation energy consumption, reduced life of the MBR, and even paralysis of the whole sewage treatment process, which is one of the most widespread problems hindering stable application of the MBR.
With the rapid development of information technology, the establishment of intelligent water treatment plants is called the trend of future development of sewage treatment plants, and how to establish an intelligent decision system becomes a hot problem. Therefore, the research of an intelligent membrane pollution decision method is very important for reducing the occurrence of membrane pollution, ensuring the stable operation of the MBR sewage treatment process and improving the intelligent level of a sewage treatment plant. In the actual operation of the MBR, the decision of membrane fouling is often made through the experience of water plant workers, often based on time periods or values of individual process variables, to derive the cleaning method or replacement speed of the MBR, and is not scientific and comprehensive. Membrane fouling is a dynamic, uncertain process, and operators in different water plants have different preferences and lack reasonable decision-making criteria. The decision-making method of how to comprehensively select membrane fouling using values of multiple process variables has attracted attention. In addition, when the data volume of the membrane fouling decision is insufficient, it is difficult to achieve an accurate decision. How to apply the knowledge of the operators in the actual water plant to the decision algorithm is also an urgent problem to be solved. An effective membrane pollution intelligent decision-making method is designed, and the method has higher research significance and application value for inhibiting the occurrence of membrane pollution and ensuring the long-term stable operation of the MBR municipal sewage treatment process.
Disclosure of Invention
The invention designs a knowledge-based two-type fuzzy membrane pollution intelligent decision method, which summarizes and expresses membrane pollution decision knowledge, adopts a knowledge-based interval two-type fuzzy neural network model, adopts a migration gradient descent algorithm to complete the parameter update of the model, and realizes the accurate decision of membrane pollution.
The invention adopts the following technical scheme and implementation steps:
a data and knowledge based decision method comprising the steps of:
the method comprises the following steps: acquiring variable data of a membrane pollution process:
gather membrane pollution process data through MBR sewage treatment instrument, including 8 key variables: water permeability, water permeability attenuation speed, water production flow, membrane scrubbing air flow, sludge concentration, transmembrane pressure difference, water production turbidity and water permeability recovery rate;
step two: knowledge expression of membrane pollution process:
combining historical data of membrane pollution process variables with decision-making knowledge of actual operation in a water plant, defining the data as a source scene, wherein fuzzy rules about membrane pollution decisions are shown in a table 1;
TABLE 1 fuzzy rules for film fouling decision
Table.1 Membrane fouling decision fuzzy rules
Figure BDA0003067062720000021
According to the MBR membrane system operation manual, the MBR membrane cleaning maintenance manual, the urban sewage treatment and pollution prevention and control technical strategy and other relevant use specifications and laws and regulations, and in combination with the actual cleaning strategy of a sewage treatment plant, the membrane pollution decision-making method is divided into the following methods:
secondly, the operation in the current state is not suitable for more than 4 hours, and the on-line chemical cleaning is needed in time after the overload operation;
secondly, physical cleaning is carried out in the near term;
thirdly, closing the water production system and carrying out off-line cleaning as soon as possible;
fourthly, the water yield is reduced to 260m3Performing off-line cleaning below the/h;
fifth, close the water producing system or increase the aeration rate to 3000m3More than h;
reducing the opening of the valve of the aeration pipeline;
seventhly, increasing the aeration quantity to 4500m3More than h, or reducing the water yield to 260m3H and below;
eighthly, whether the water quality, the water production flow, the membrane scrubbing air quantity and the membrane tank sludge concentration are abnormal is checked;
ninthly, checking a leakage membrane module device and a water production pipeline;
increasing the sludge discharge amount of the membrane tank, increasing the reflux ratio of the membrane tank, and controlling the sludge concentration of the membrane tank to be 8000-charge 12000 mg/L;
Figure BDA0003067062720000031
reducing the water yield of the membrane pool, and controlling the transmembrane pressure difference to be less than 40 kPa;
Figure BDA0003067062720000032
reducing the discharge amount of the residual sludge, checking the sludge reflux ratio, and controlling the sludge concentration of the membrane tank to be more than 6000 mg/L;
Figure BDA0003067062720000033
paying attention to the operation condition of the membrane pool, and performing online chemical cleaning (large cleaning) again within 72 hours;
summarize these decision information into fuzzy rules
Figure BDA0003067062720000034
Wherein,
Figure BDA0003067062720000035
is the water penetration rate of the source scene at time t,
Figure BDA0003067062720000036
is the water permeability decay rate of the source scene at time t,
Figure BDA0003067062720000037
is the water production flow rate of the source scenario at time t,
Figure BDA0003067062720000038
is the amount of membrane scrub gas for the source scene at time t,
Figure BDA0003067062720000039
is the sludge concentration of the source scene at time t,
Figure BDA00030670627200000310
(t) is the transmembrane pressure difference of the source scene at time t,
Figure BDA00030670627200000311
is the water production turbidity of the source scene at time t,
Figure BDA00030670627200000312
is the water permeability recovery rate of the source scene at time t, S represents the source scene,
Figure BDA00030670627200000313
is the linguistic item of the u input variable in the b fuzzy rule of the source scene at the time t,
Figure BDA00030670627200000314
is the output of the b-th fuzzy rule of the source scene at the time t, namely the membrane fouling decision category,
Figure BDA00030670627200000315
the decision result of the linguistic item and the membrane pollution output by the b-th fuzzy rule of the source scene at the time t, u is the number of input variables, b is the number of fuzzy rules, and u is 1, …, 8, b is 1, …, 13;
step three: interval two-type fuzzy neural network based on migration gradient descent:
the interval type two fuzzy neural network has 5 layers which are respectively an input layer, a membership function layer, a rule layer, a back part layer and an output layer, and the mathematical description is as follows:
an input layer: a total of 8 neurons, with the output of the input layer being
ru(t)＝xu(t) (2)
X(t)＝[x1(t),x2(t),x3(t),x4(t),x5(t),x6(t),x7(t),x8(t)] (3)
Wherein r isu(t) is the output of the u-th input neuron of the two-type fuzzy neural network in the t time interval, xu(t) is the input value of the u-th process variable at the time t, X (t) is the input vector of the two-type fuzzy neural network in the interval of the time t, x1(t) is an input value of water permeability at time t, x2(t) is an input value of a permeable rate attenuation rate at time t, x3(t) is an input value of the water production flow at time t, x4(t) is the input value of the amount of membrane scrubbing gas at time t, x5(t) is an input value of sludge concentration at time t, x6(t) is the input value of the transmembrane pressure difference at time t, x7(t) is the input value of the turbidity of the produced water at time t, x8(t) is an input value of the water permeability recovery rate at time t;
membership function layer: a total of 8 × 13 neurons, with the output of the membership function layer being
Figure BDA0003067062720000041
Figure BDA0003067062720000042
Figure BDA0003067062720000043
Figure BDA0003067062720000044
Wherein,
Figure BDA0003067062720000045
is the output of the b-th membership function layer neuron at the u-th input at time t, cub(t) is the center of the mth input, the b-th membership function layer neuron at time t, σub(t) is the width of the mth input, the b-th membership function layer neuron at time t,
Figure BDA0003067062720000046
and
Figure BDA0003067062720000047
respectively an upper bound and a lower bound of the u membership function in the b fuzzy rule of the source scene at the time t, cub(t) and
Figure BDA0003067062720000048
respectively inputting the lower bound and the upper bound of the central value of the b-th membership function layer neuron at the u-th time, wherein e is a natural constant, and is 2.7183, and the interval form of the membership function layer neuron is
Figure BDA0003067062720000049
Figure BDA00030670627200000410
Wherein,
Figure BDA00030670627200000411
and
Figure BDA00030670627200000412
the lower bound and the upper bound of the output of the b-th input membership function layer neuron at the time t are respectively set;
and (3) a rule layer: there are 13 neurons in total, with the output of the rule layer being
Figure BDA00030670627200000413
Figure BDA00030670627200000414
Figure BDA00030670627200000415
Wherein v isb(t) is the output of the b-th regular layer neuron at time t, vb(t) and
Figure BDA00030670627200000416
the lower bound and the upper bound of the output of the b-th rule layer neuron at the time t respectively;
a rear part layer: there are 26 neurons in total, the output of the back-end layer is
Figure BDA0003067062720000051
Figure BDA0003067062720000052
Figure BDA0003067062720000053
Wherein,
Figure BDA0003067062720000054
is the back-part weight for the b-th regular layer neuron and the g-th output layer neuron at time t,
Figure BDA0003067062720000055
is the weight coefficient of the u-th input neuron, the b-th rule layer neuron and the g-th output neuron at the time t, yg(t) and
Figure BDA0003067062720000056
the output lower bound and the output upper bound of the back part layer neuron of the g-th output layer neuron at the time t are respectively, and g is 1, 2, … and 13;
an output layer: there are 13 neurons in total, the output of the output layer is
Figure BDA0003067062720000057
Wherein, yg(t) is the g-th output value, gamma, of the two-type fuzzy neural network in the t-time intervalg(t) is the ratio of the lower bound of the back-part layer neurons of the g-th output layer neuron at time t;
the normalized exponential function model is mainly used for solving the multi-classification problem, and the probability p (y (t) ═ G) is set to represent the probability that the membrane pollution sample is judged to be the class G at the time t, the probability with the maximum output is selected as the final class, and the output of the class G classification is
Figure BDA0003067062720000058
Figure BDA0003067062720000059
Figure BDA00030670627200000510
Wherein h (t) is the output probability vector of the two-type fuzzy neural network model in the t-time interval, y (t) is the classification category of the model at the t time, and k is a normalization function, so that the sum of all probabilities is 1;
in order to improve the learning performance of the interval two-type fuzzy neural network model, a migration gradient descent algorithm is provided, and the knowledge of a source model and the data of a target model are combined in the learning process; the steps of the migration gradient descent algorithm are as follows: in the learning process, extracting a series of knowledge from film pollution decision history data of a source scene, and transferring the knowledge to a target scene;
defining an objective function of the model as
Figure BDA0003067062720000061
Where E (t) is the objective function at time t, D represents the target scene, f0 DIs a regional two-type fuzzy neural network function, x, of the target sceneD(t) is the input of the target model at time t, wnew(t) is the back-end layer output weight parameter of the target model at time t,
Figure BDA0003067062720000062
is the expected output of the target model at time t
wnew(t)＝wold(t)+δ(t)wS(tθ) (21)
Wherein, wold(t) is the back-end layer initial output weight of the target model at time t, wS(t) is the source model back-part layer output weight at time t, δ (t) is the balance parameter at time t, 0<δ(t)<1, representing the relation between a target model and a source model;
in the interval two-type fuzzy neural network model, three variables are updated: the target model back-part layer output weight, the source model back-part layer output weight and the balance parameter are updated according to the following formula
Figure BDA0003067062720000063
Wherein, wS(t +1) is the source model back-end-of-layer output weight at time t +1, η1Is the learning rate of the source model back-end output weights,
Figure BDA0003067062720000064
is the partial derivative, w, of the target function at time t with respect to the source model back-part layer output weightD(t +1) is the target model back-end output weight at time t +1, wD(t) is the target model back-end layer output weight at time t, η2Is the learning rate of the target model's back-end output weights,
Figure BDA0003067062720000065
is the partial derivative of the target function at the time t with respect to the output weight of the target model back-part layer, δ (t +1) is the equilibrium parameter at the time t +1, η3Is the learning rate of the balance parameter,
Figure BDA0003067062720000066
is the partial derivative of the objective function with respect to the balance parameter at time t. In the scheme, in the step one: collecting membrane pollution key process variables through an MBR sewage treatment instrument;
in the second step: the membrane pollution process knowledge represents the combination of membrane pollution process variable historical data and decision-making knowledge of actual operation in a water plant, and the third step is as follows: and (3) modeling the membrane pollution by adopting an interval two-type fuzzy neural network, and updating network model parameters by adopting a migration gradient descent algorithm.
Compared with the prior art, the invention has the following advantages: 1) the invention provides a knowledge-based interval type two fuzzy neural network decision model aiming at the problem that the membrane pollution phenomenon is difficult to accurately identify due to nonlinear and timely complex characteristics in the membrane pollution process, and the construction of an intelligent membrane pollution decision model is completed by converting decision knowledge in the operation process into a fuzzy rule form and designing a membership function by using the fuzzy rule, so that the decision on the membrane pollution is realized; 2) in order to improve the precision of a membrane pollution decision model, the invention designs a migration gradient descent algorithm to carry out self-adaptive updating on model parameters, extracts a series of knowledge from membrane pollution decision history data of a source scene, transfers the knowledge to a target scene to complete the updating of the decision model parameters, realizes the accurate decision of membrane pollution, slows down the occurrence of membrane pollution and promotes the stable operation of the MBR municipal sewage treatment process.
Drawings
FIG. 1 is a block diagram of a knowledge-based two-type fuzzy membrane fouling intelligent decision method.
Detailed Description
For a better understanding and appreciation of the invention, reference will now be made in detail to the embodiments illustrated in the accompanying drawings.
Example 1: referring to fig. 1, a knowledge-based two-type fuzzy membrane fouling intelligent decision method, the method comprising the steps of:
the method comprises the following steps: acquiring variable data of a membrane pollution process:
step two: knowledge expression of membrane pollution process:
step three: the interval type two fuzzy neural network based on migration gradient descending.
According to the on-line sensor who installs in MBR sewage treatment system in this scheme, gather membrane pollution process data, including 8 key variables: water permeability, water permeability attenuation speed, water production flow, membrane scrubbing air flow, sludge concentration, transmembrane pressure difference, water production turbidity and water permeability recovery rate.
In the second step, the knowledge expression of the membrane pollution process is realized, and the method comprises the following steps:
combining historical data of membrane pollution process variables with decision-making knowledge of actual operation in a water plant, defining the data as a source scene, wherein fuzzy rules about membrane pollution decisions are shown in a table 1;
TABLE 1 fuzzy rules for film fouling decision
Table.1 Membrane fouling decision fuzzy rules
Figure BDA0003067062720000071
Figure BDA0003067062720000081
According to the MBR membrane system operation manual, the MBR membrane cleaning maintenance manual, the urban sewage treatment and pollution prevention and control technical strategy and other relevant use specifications and laws and regulations, and in combination with the actual cleaning strategy of a sewage treatment plant, the membrane pollution decision-making method is divided into the following methods:
firstly, the operation in the current state is not suitable for more than 4 hours, and the online chemical cleaning is needed in time after the overload operation;
secondly, physical cleaning is carried out in the near term;
thirdly, closing the water production system and carrying out off-line cleaning as soon as possible;
fourthly, the water yield is reduced to 260m3Performing off-line cleaning below the/h;
fifth, close the water producing system or increase the aeration rate to 3000m3More than h;
reducing the opening of the valve of the aeration pipeline;
seventhly, increasing the aeration quantity to 4500m3More than h, or reducing the water yield to 260m3H and below;
eighthly, whether the water quality, the water production flow, the membrane scrubbing air quantity and the membrane tank sludge concentration are abnormal is checked;
ninthly, checking a leakage membrane module device and a water production pipeline;
increasing the sludge discharge amount of the membrane tank, increasing the reflux ratio of the membrane tank, and controlling the sludge concentration of the membrane tank to be 8000-charge 12000 mg/L;
Figure BDA0003067062720000082
reducing the water yield of the membrane pool, and controlling the transmembrane pressure difference to be less than 40 kPa;
Figure BDA0003067062720000083
reducing the discharge amount of the residual sludge, checking the sludge reflux ratio, and controlling the sludge concentration of the membrane tank to be more than 6000 mg/L;
Figure BDA0003067062720000084
paying attention to the operation condition of the membrane pool, and performing online chemical cleaning again within 72 hours;
summarize these decision information into fuzzy rules
Figure BDA0003067062720000085
Wherein,
Figure BDA0003067062720000086
is the water penetration rate of the source scene at time t,
Figure BDA0003067062720000087
is the water permeability decay rate of the source scene at time t,
Figure BDA0003067062720000088
is the water production flow rate of the source scenario at time t,
Figure BDA0003067062720000089
is the amount of membrane scrub gas for the source scene at time t,
Figure BDA00030670627200000810
is the sludge concentration of the source scene at time t,
Figure BDA00030670627200000811
Figure BDA00030670627200000812
is the transmembrane pressure difference of the source scene at time t,
Figure BDA00030670627200000813
is the water production turbidity of the source scene at time t,
Figure BDA00030670627200000814
is the water permeability recovery rate of the source scene at time t, S represents the source scene,
Figure BDA00030670627200000815
is the linguistic item of the u input variable in the b fuzzy rule of the source scene at the time t,
Figure BDA00030670627200000816
is the output of the b-th fuzzy rule of the source scene at the time t, namely the membrane fouling decision category,
Figure BDA00030670627200000817
the decision result of the linguistic item and the membrane fouling output by the b-th fuzzy rule of the source scene at the time t, u is the number of input variables, b is the number of fuzzy rules, and u is 1, …, 8, b is 1, …, 13.
In the third step, constructing an interval type two fuzzy neural network model based on migration gradient descent, which comprises the following steps:
the interval type two fuzzy neural network has 5 layers which are respectively an input layer, a membership function layer, a rule layer, a back part layer and an output layer, and the mathematical description is as follows:
an input layer: a total of 8 neurons, with the output of the input layer being
ru(t)＝xu(t)； (24)
X(t)＝[x1(t),x2(t),x3(t),x4(t),x5(t),x6(t),x7(t),x8(t)] (25)
Wherein r isu(t) is the output of the u-th input neuron of the two-type fuzzy neural network in the t time interval, xu(t) is the input value of the u-th process variable at the time t, X (t) is the input vector of the two-type fuzzy neural network in the interval of the time t, x1(t) is an input value of water permeability at time t, x2(t) is an input value of a permeable rate attenuation rate at time t, x3(t) is an input value of the water production flow at time t, x4(t) is the input value of the amount of membrane scrubbing gas at time t, x5(t) is an input value of sludge concentration at time t, x6(t) is the input value of the transmembrane pressure difference at time t, x7(t) is the input value of the turbidity of the produced water at time t, x8(t) is an input value of the water permeability recovery rate at time t;
membership function layer: a total of 8 × 13 neurons, with the output of the membership function layer being
Figure BDA0003067062720000091
Figure BDA0003067062720000092
Figure BDA0003067062720000093
Figure BDA0003067062720000094
Wherein,
Figure BDA0003067062720000095
is the output of the b-th membership function layer neuron at the u-th input at time t, cub(t) is the center of the mth input, the b-th membership function layer neuron at time t, σub(t) is the width of the mth input, the b-th membership function layer neuron at time t,
Figure BDA0003067062720000096
and
Figure BDA0003067062720000097
respectively an upper bound and a lower bound of the u membership function in the b fuzzy rule of the source scene at the time t, cub(t) and
Figure BDA0003067062720000098
respectively inputting the lower bound and the upper bound of the central value of the b-th membership function layer neuron at the u-th time, wherein e is a natural constant, and is 2.7183, and the interval form of the membership function layer neuron is
Figure BDA0003067062720000099
Figure BDA0003067062720000101
Wherein,
Figure BDA0003067062720000102
and
Figure BDA0003067062720000103
the lower bound and the upper bound of the output of the b-th input membership function layer neuron at the time t are respectively set;
and (3) a rule layer: there are 13 neurons in total, with the output of the rule layer being
Figure BDA0003067062720000104
Figure BDA0003067062720000105
Figure BDA0003067062720000106
Wherein v isb(t) is the output of the b-th regular layer neuron at time t, vb(t) and
Figure BDA0003067062720000107
respectively, below the output of the b-th regular layer neuron at time tA bound and an upper bound;
a rear part layer: there are 26 neurons in total, the output of the back-end layer is
Figure BDA0003067062720000108
Figure BDA0003067062720000109
Figure BDA00030670627200001010
Wherein,
Figure BDA00030670627200001011
is the back-part weight for the b-th regular layer neuron and the g-th output layer neuron at time t,
Figure BDA00030670627200001012
is the weight coefficient of the u-th input neuron, the b-th rule layer neuron and the g-th output neuron at the time t, yg(t) and
Figure BDA00030670627200001013
the output lower bound and the output upper bound of the back part layer neuron of the g-th output layer neuron at the time t are respectively, and g is 1, 2, … and 13;
an output layer: there are 13 neurons in total, the output of the output layer is
Figure BDA00030670627200001014
Wherein, yg(t) is the g-th output value, gamma, of the two-type fuzzy neural network in the t-time intervalg(t) is the ratio of the lower bound of the back-part layer neurons of the g-th output layer neuron at time t;
the normalized exponential function model is mainly used for solving the multi-classification problem, and the probability p (y (t) ═ G) is set to represent the probability that the membrane pollution sample is judged to be the class G at the time t, the probability with the maximum output is selected as the final class, and the output of the class G classification is
Figure BDA0003067062720000111
Figure BDA0003067062720000112
Figure BDA0003067062720000113
Where h (t) is the output probability vector of the two type fuzzy neural network model in the t time interval, y (t) is the classification category of the model at the t time, and k is the normalization function, so that the sum of all probabilities is 1.
Updating network model parameters by using a migration gradient descent algorithm, extracting a series of knowledge from membrane pollution decision history data of a source scene in a learning process, and transferring the knowledge into a target scene;
defining an objective function of the model as
Figure BDA0003067062720000114
Where E (t) is the objective function at time t, D represents the target scene, f0 DIs a regional two-type fuzzy neural network function, x, of the target sceneD(t) is the input of the target model at time t, wnew(t) is the back-end layer output weight parameter of the target model at time t,
Figure BDA0003067062720000116
is the expected output of the target model at time t
wnew(t)＝wold(t)+δ(t)wS(tθ) (43)
Wherein，wold(t) is the back-end layer initial output weight of the target model at time t, wS(t) is the source model back-part layer output weight at time t, δ (t) is the balance parameter at time t, 0<δ(t)<1, representing the relation between a target model and a source model;
in the interval two-type fuzzy neural network model, three variables are updated: the target model back-part layer output weight, the source model back-part layer output weight and the balance parameter are updated according to the following formula
Figure BDA0003067062720000115
Wherein, wS(t +1) is the source model back-end-of-layer output weight at time t +1, η1Is the learning rate of the source model back-end output weights,
Figure BDA0003067062720000121
is the partial derivative, w, of the target function at time t with respect to the source model back-part layer output weightD(t +1) is the target model back-end output weight at time t +1, wD(t) is the target model back-end layer output weight at time t, η2Is the learning rate of the target model's back-end output weights,
Figure BDA0003067062720000122
is the partial derivative of the target function at the time t with respect to the output weight of the target model back-part layer, δ (t +1) is the equilibrium parameter at the time t +1, η3Is the learning rate of the balance parameter,
Figure BDA0003067062720000123
is the partial derivative of the objective function with respect to the balance parameter at time t.
It should be noted that the above-mentioned embodiments are not intended to limit the scope of the present invention, and all equivalent modifications and substitutions based on the above-mentioned technical solutions are within the scope of the present invention as defined in the claims.

Claims (8)

1. A knowledge-based intelligent membrane pollution decision method based on two-type fuzzy is characterized by comprising the following steps:
the method comprises the following steps: acquiring variable data of a membrane pollution process:
step two: knowledge expression of membrane pollution process:
step three: the interval type two fuzzy neural network based on migration gradient descending.
2. The intelligent knowledge-based two-type fuzzy membrane fouling decision making method according to claim 1, characterized by comprising the following steps: acquiring variable data of a membrane pollution process, which comprises the following steps:
gather membrane pollution process data through MBR sewage treatment instrument, including 8 key variables: water permeability, water permeability attenuation speed, water production flow, membrane scrubbing air flow, sludge concentration, transmembrane pressure difference, water production turbidity and water permeability recovery rate.
3. The intelligent knowledge-based two-type fuzzy membrane fouling decision making method according to claim 1, characterized by the following steps: knowledge expression of membrane pollution process: the method comprises the following specific steps:
combining historical data of membrane pollution process variables with decision-making knowledge of actual operation in a water plant, defining the data as a source scene, wherein fuzzy rules about membrane pollution decisions are shown in a table 1;
TABLE 1 fuzzy rules for film fouling decision
Figure FDA0003067062710000011
According to the MBR membrane system operation manual, the MBR membrane cleaning maintenance manual, the urban sewage treatment and pollution prevention and control technical strategy and other relevant use specifications and laws and regulations, and in combination with the actual cleaning strategy of a sewage treatment plant, the membrane pollution decision-making method is divided into the following methods:
firstly, the operation in the current state is not suitable for more than 4 hours, and the online chemical cleaning is needed in time after the overload operation;
secondly, physical cleaning is carried out in the near term;
thirdly, closing the water production system and carrying out off-line cleaning as soon as possible;
fourthly, the water yield is reduced to 260m3Performing off-line cleaning below the/h;
fifth, close the water producing system or increase the aeration rate to 3000m3More than h;
reducing the opening of the valve of the aeration pipeline;
seventhly, increasing the aeration quantity to 4500m3More than h, or reducing the water yield to 260m3H and below;
eighthly, whether the water quality, the water production flow, the membrane scrubbing air quantity and the membrane tank sludge concentration are abnormal is checked;
ninthly, checking a leakage membrane module device and a water production pipeline;
increasing the sludge discharge amount of the membrane tank, increasing the reflux ratio of the membrane tank, and controlling the sludge concentration of the membrane tank to be 8000-charge 12000 mg/L;
Figure FDA0003067062710000021
reducing the water yield of the membrane pool, and controlling the transmembrane pressure difference to be less than 40 kPa;
Figure FDA0003067062710000022
reducing the discharge amount of the residual sludge, checking the sludge reflux ratio, and controlling the sludge concentration of the membrane tank to be more than 6000 mg/L;
Figure FDA0003067062710000023
paying attention to the operation condition of the membrane pool, and performing online chemical cleaning again within 72 hours;
summarize these decision information into fuzzy rules
Figure FDA0003067062710000024
Wherein,
Figure FDA0003067062710000025
is the water penetration rate of the source scene at time t,
Figure FDA0003067062710000026
is the water permeability decay rate of the source scene at time t,
Figure FDA0003067062710000027
is the water production flow rate of the source scenario at time t,
Figure FDA0003067062710000028
is the amount of membrane scrub gas for the source scene at time t,
Figure FDA0003067062710000029
is the sludge concentration of the source scene at time t,
Figure FDA00030670627100000210
(t) is the transmembrane pressure difference of the source scene at time t,
Figure FDA00030670627100000211
is the water production turbidity of the source scene at time t,
Figure FDA00030670627100000212
is the water permeability recovery rate of the source scene at time t, S represents the source scene,
Figure FDA00030670627100000213
is the linguistic item of the u input variable in the b fuzzy rule of the source scene at the time t,
Figure FDA00030670627100000214
is the output of the b-th fuzzy rule of the source scene at the time t, namely the membrane fouling decision category,
Figure FDA00030670627100000215
is the language term and membrane pollution output by the b-th fuzzy rule of the source scene at the time tAs a result of the decision, u is the number of input variables, b is the number of fuzzy rules, u is 1, …, 8, and b is 1, …, 13.
4. The intelligent knowledge-based two-type fuzzy membrane fouling decision making method according to claim 1, characterized by comprising the following three steps: the interval type two fuzzy neural network based on migration gradient descent specifically comprises the following steps:
the interval type two fuzzy neural network has 5 layers which are respectively an input layer, a membership function layer, a rule layer, a back part layer and an output layer, and the mathematical description is as follows:
an input layer: a total of 8 neurons, with the output of the input layer being
ru(t)＝xu(t)； (2)
X(t)＝[x1(t),x2(t),x3(t),x4(t),x5(t),x6(t),x7(t),x8(t)] (3)
Wherein r isu(t) is the output of the u-th input neuron of the two-type fuzzy neural network in the t time interval, xu(t) is the input value of the u-th process variable at the time t, X (t) is the input vector of the two-type fuzzy neural network in the interval of the time t, x1(t) is an input value of water permeability at time t, x2(t) is an input value of a permeable rate attenuation rate at time t, x3(t) is an input value of the water production flow at time t, x4(t) is the input value of the amount of membrane scrubbing gas at time t, x5(t) is an input value of sludge concentration at time t, x6(t) is the input value of the transmembrane pressure difference at time t, x7(t) is the input value of the turbidity of the produced water at time t, x8(t) is an input value of the water permeability recovery rate at time t;
membership function layer: a total of 8 × 13 neurons, with the output of the membership function layer being
Figure FDA0003067062710000031
Figure FDA0003067062710000032
Figure FDA0003067062710000033
Figure FDA0003067062710000034
Wherein,
Figure FDA0003067062710000035
is the output of the b-th membership function layer neuron at the u-th input at time t, cub(t) is the center of the mth input, the b-th membership function layer neuron at time t, σub(t) is the width of the mth input, the b-th membership function layer neuron at time t,
Figure FDA0003067062710000036
and
Figure FDA0003067062710000037
respectively an upper bound and a lower bound of the u membership function in the b fuzzy rule of the source scene at the time t, cub(t) and
Figure FDA0003067062710000038
respectively inputting the lower bound and the upper bound of the central value of the b-th membership function layer neuron at the u-th time, wherein e is a natural constant, and is 2.7183, and the interval form of the membership function layer neuron is
Figure FDA0003067062710000039
Figure FDA00030670627100000310
Wherein,
Figure FDA00030670627100000311
and
Figure FDA00030670627100000312
the lower bound and the upper bound of the output of the b-th input membership function layer neuron at the time t are respectively set; and (3) a rule layer: there are 13 neurons in total, with the output of the rule layer being
Figure FDA00030670627100000313
Figure FDA00030670627100000314
Figure FDA0003067062710000041
Wherein v isb(t) is the output of the b-th regular layer neuron at time t, vb(t) and
Figure FDA0003067062710000042
the lower bound and the upper bound of the output of the b-th rule layer neuron at the time t respectively;
a rear part layer: there are 26 neurons in total, the output of the back-end layer is
Figure FDA0003067062710000043
Figure FDA0003067062710000044
Figure FDA0003067062710000045
Wherein,
Figure FDA0003067062710000046
is the back-part weight for the b-th regular layer neuron and the g-th output layer neuron at time t,
Figure FDA0003067062710000047
is the weight coefficient of the u-th input neuron, the b-th rule layer neuron and the g-th output neuron at the time t, yg(t) and
Figure FDA0003067062710000048
the output lower bound and the output upper bound of the back part layer neuron of the g-th output layer neuron at the time t are respectively, and g is 1, 2, … and 13;
an output layer: there are 13 neurons in total, the output of the output layer is
Figure FDA0003067062710000049
Wherein, yg(t) is the g-th output value, gamma, of the two-type fuzzy neural network in the t-time intervalg(t) is the ratio of the lower bound of the back-part layer neurons of the g-th output layer neuron at time t;
the normalized exponential function model is mainly used for solving the multi-classification problem, and the probability p (y (t) ═ G) is set to represent the probability that the membrane pollution sample is judged to be the class G at the time t, the probability with the maximum output is selected as the final class, and the output of the class G classification is
Figure FDA00030670627100000410
Figure FDA00030670627100000411
Figure FDA00030670627100000412
Where h (t) is the output probability vector of the two type fuzzy neural network model in the t time interval, y (t) is the classification category of the model at the t time, and k is the normalization function, so that the sum of all probabilities is 1.
5. The intelligent knowledge-based two-type fuzzy membrane fouling decision making method according to claim 1, wherein the step of the migration gradient descent algorithm in the step three is as follows: in the learning process, extracting a series of knowledge from film pollution decision history data of a source scene, and transferring the knowledge to a target scene;
defining an objective function of the model as
Figure FDA0003067062710000051
Where E (t) is the objective function at time t, D represents the target scene, f0 DIs a regional two-type fuzzy neural network function, x, of the target sceneD(t) is the input of the target model at time t, wnew(t) is the back-end layer output weight parameter of the target model at time t,
Figure FDA0003067062710000052
is the expected output of the target model at time t
wnew(t)＝wold(t)+δ(t)wS(tθ) (21)
Wherein, wold(t) is the back-end layer initial output weight of the target model at time t, wS(t) is the source model back-part layer output weight at time t, δ (t) is the balance parameter at time t, 0<δ(t)<1, representing the relation between a target model and a source model;
in the interval two-type fuzzy neural network model, three variables are updated: the target model back-part layer output weight, the source model back-part layer output weight and the balance parameter are updated according to the following formula
Figure FDA0003067062710000053
Wherein, wS(t +1) is the source model back-end-of-layer output weight at time t +1, η1Is the learning rate of the source model back-end output weights,
Figure FDA0003067062710000054
is the partial derivative, w, of the target function at time t with respect to the source model back-part layer output weightD(t +1) is the target model back-end output weight at time t +1, wD(t) is the target model back-end layer output weight at time t, η2Is the learning rate of the target model's back-end output weights,
Figure FDA0003067062710000055
is the partial derivative of the target function at the time t with respect to the output weight of the target model back-part layer, δ (t +1) is the equilibrium parameter at the time t +1, η3Is the learning rate of the balance parameter,
Figure FDA0003067062710000056
is the partial derivative of the objective function with respect to the balance parameter at time t.
6. The intelligent knowledge-based two-type fuzzy membrane fouling decision making method according to claim 1, wherein in the first step: the membrane fouling key process variables are collected by MBR sewage treatment instruments.
7. The intelligent knowledge-based two-type fuzzy membrane fouling decision making method according to claim 1, wherein in the second step: the membrane pollution process knowledge is represented by combining membrane pollution process variable historical data and decision-making knowledge of actual operation in a water plant.
8. The data and knowledge based decision method according to claim 1, characterized by the following steps: and (3) modeling the membrane pollution by adopting an interval two-type fuzzy neural network, and updating network model parameters by adopting a migration gradient descent algorithm.
CN202110528334.6A 2021-05-14 2021-05-14 Intelligent membrane pollution decision-making method based on knowledge type-two fuzzy Pending CN113283481A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110528334.6A CN113283481A (en) 2021-05-14 2021-05-14 Intelligent membrane pollution decision-making method based on knowledge type-two fuzzy

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110528334.6A CN113283481A (en) 2021-05-14 2021-05-14 Intelligent membrane pollution decision-making method based on knowledge type-two fuzzy

Publications (1)

Publication Number Publication Date
CN113283481A true CN113283481A (en) 2021-08-20

Family

ID=77279089

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110528334.6A Pending CN113283481A (en) 2021-05-14 2021-05-14 Intelligent membrane pollution decision-making method based on knowledge type-two fuzzy

Country Status (1)

Country Link
CN (1) CN113283481A (en)

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106406094A (en) * 2016-10-16 2017-02-15 北京工业大学 Interval type-2 fuzzy neural network based dissolved oxygen concentration tracking method in sewage treatment
CN108375534A (en) * 2018-02-06 2018-08-07 北京工业大学 MBR fouling membrane intelligent early-warning methods
CN109133351A (en) * 2018-08-29 2019-01-04 北京工业大学 Membrane bioreactor-MBR fouling membrane intelligent early-warning method
CN110647037A (en) * 2019-09-23 2020-01-03 北京工业大学 Cooperative control method for sewage treatment process based on two-type fuzzy neural network
CN111204867A (en) * 2019-06-24 2020-05-29 北京工业大学 Membrane bioreactor-MBR membrane pollution intelligent decision-making method
CN111415032A (en) * 2020-03-03 2020-07-14 东华大学 Method for predicting production performance of polyester fiber precursor based on E L M-AE of transfer learning
CN111444958A (en) * 2020-03-25 2020-07-24 北京百度网讯科技有限公司 Model migration training method, device, equipment and storage medium
CN111612175A (en) * 2020-05-11 2020-09-01 北京工业大学 Waste mobile phone intelligent pricing method based on fuzzy transfer learning
GB202015382D0 (en) * 2020-09-29 2020-11-11 Smith Graeme Signal processing systems
CN112101402A (en) * 2020-07-22 2020-12-18 北京工业大学 Membrane pollution early warning method based on knowledge fuzzy learning

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106406094A (en) * 2016-10-16 2017-02-15 北京工业大学 Interval type-2 fuzzy neural network based dissolved oxygen concentration tracking method in sewage treatment
CN108375534A (en) * 2018-02-06 2018-08-07 北京工业大学 MBR fouling membrane intelligent early-warning methods
CN109133351A (en) * 2018-08-29 2019-01-04 北京工业大学 Membrane bioreactor-MBR fouling membrane intelligent early-warning method
CN111204867A (en) * 2019-06-24 2020-05-29 北京工业大学 Membrane bioreactor-MBR membrane pollution intelligent decision-making method
CN110647037A (en) * 2019-09-23 2020-01-03 北京工业大学 Cooperative control method for sewage treatment process based on two-type fuzzy neural network
CN111415032A (en) * 2020-03-03 2020-07-14 东华大学 Method for predicting production performance of polyester fiber precursor based on E L M-AE of transfer learning
CN111444958A (en) * 2020-03-25 2020-07-24 北京百度网讯科技有限公司 Model migration training method, device, equipment and storage medium
CN111612175A (en) * 2020-05-11 2020-09-01 北京工业大学 Waste mobile phone intelligent pricing method based on fuzzy transfer learning
CN112101402A (en) * 2020-07-22 2020-12-18 北京工业大学 Membrane pollution early warning method based on knowledge fuzzy learning
GB202015382D0 (en) * 2020-09-29 2020-11-11 Smith Graeme Signal processing systems

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
HAN HONG-GUI等: "Cooperative strategy for constructing interval type-2 fuzzy neural network", 《NEUROCOMPUTING》, vol. 365, pages 249 - 260, XP085805092, DOI: 10.1016/j.neucom.2019.07.004 *
MIN HAN等: "Interval Type-2 Fuzzy Neural Networks for Chaotic Time Series Prediction: A Concise Overview", 《IEEE TRANSACTIONS ON CYBERNETICS》, vol. 49, no. 7, pages 2720 - 2731, XP011720145, DOI: 10.1109/TCYB.2018.2834356 *
李忠鹏: "区间二型模糊神经网络在SCR入口NOx浓度软测量中的应用研究", 《中国优秀硕士学位论文全文数据库 (工程科技Ⅰ辑)》, no. 01, pages 027 - 2617 *

Similar Documents

Publication Publication Date Title
Ye et al. Tackling environmental challenges in pollution controls using artificial intelligence: A review
Farhi et al. Prediction of wastewater treatment quality using LSTM neural network
CN103744293B (en) Wastewater treatment monitoring method and system based on fuzzy neural network
CN102854296B (en) Sewage-disposal soft measurement method on basis of integrated neural network
Han et al. Prediction of activated sludge bulking based on a self-organizing RBF neural network
CN101576734B (en) Dissolved oxygen control method based on dynamic radial basis function neural network
CN108375534A (en) MBR fouling membrane intelligent early-warning methods
CN105844401B (en) The lake and reservoir wawter bloom of case-based reasioning administers complexity and dynamically associates model and decision-making technique
CN113033917B (en) Sewage treatment plant prediction planning operation management method based on peripheral data
CN112101402B (en) Membrane pollution early warning method based on knowledge fuzzy learning
CN106682316A (en) Real-time effluent total-phosphorus monitoring system based on peak radial basis function neural network
CN111652425A (en) River water quality prediction method based on rough set and long and short term memory network
CN109473182A (en) A kind of MBR film permeability rate intelligent detecting method based on deepness belief network
Wang et al. A full-view management method based on artificial neural networks for energy and material-savings in wastewater treatment plants
Ruan et al. Application of fuzzy neural networks for modeling of biodegradation and biogas production in a full-scale internal circulation anaerobic reactor
Yetilmezsoy Applications of soft computing methods in environmental engineering
Elsayed et al. Machine learning classification algorithms for inadequate wastewater treatment risk mitigation
CN117422165A (en) Urban water delivery system water quantity prediction method and system based on low carbon emission
CN109408896B (en) Multi-element intelligent real-time monitoring method for anaerobic sewage treatment gas production
CN115206444A (en) Optimal drug dosage prediction method based on FCM-ANFIS model
Yetilmezsoy et al. Adaptive neuro-fuzzy inference-based modeling of a full-scale expanded granular sludge bed reactor treating corn processing wastewater
Alnajjar et al. Removal efficiency prediction model based on the artificial neural network for pollution prevention in wastewater treatment plants
CN111204867B (en) Membrane bioreactor-MBR membrane pollution intelligent decision-making method
CN113283481A (en) Intelligent membrane pollution decision-making method based on knowledge type-two fuzzy
Wang et al. Artificial intelligence algorithm application in wastewater treatment plants: Case study for COD load prediction

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