CN108408855A - A kind of online Adding medicine control method and system for wastewater treatment - Google Patents

A kind of online Adding medicine control method and system for wastewater treatment Download PDF

Info

Publication number
CN108408855A
CN108408855A CN201810315713.5A CN201810315713A CN108408855A CN 108408855 A CN108408855 A CN 108408855A CN 201810315713 A CN201810315713 A CN 201810315713A CN 108408855 A CN108408855 A CN 108408855A
Authority
CN
China
Prior art keywords
neural network
principal component
sample
input
network model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201810315713.5A
Other languages
Chinese (zh)
Other versions
CN108408855B (en
Inventor
袁照威
刘宁
牟伟腾
齐勇
杨言
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.)
Datang (beijing) Water Engineering Technology Co Ltd
Original Assignee
Datang (beijing) Water Engineering Technology 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 Datang (beijing) Water Engineering Technology Co Ltd filed Critical Datang (beijing) Water Engineering Technology Co Ltd
Priority to CN201810315713.5A priority Critical patent/CN108408855B/en
Publication of CN108408855A publication Critical patent/CN108408855A/en
Application granted granted Critical
Publication of CN108408855B publication Critical patent/CN108408855B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F1/00—Treatment of water, waste water, or sewage
    • C02F1/52—Treatment of water, waste water, or sewage by flocculation or precipitation of suspended impurities
    • C02F1/5209—Regulation methods for flocculation or precipitation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • 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
    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F2209/00—Controlling or monitoring parameters in water treatment
    • C02F2209/001—Upstream control, i.e. monitoring for predictive control
    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F2209/00—Controlling or monitoring parameters in water treatment
    • C02F2209/06—Controlling or monitoring parameters in water treatment pH
    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F2209/00—Controlling or monitoring parameters in water treatment
    • C02F2209/11—Turbidity
    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F2209/00—Controlling or monitoring parameters in water treatment
    • C02F2209/14—NH3-N

Landscapes

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

Abstract

The invention discloses a kind of online Adding medicine control method and system for wastewater treatment, this method includes：Current time water monitoring index data are obtained, and current time water monitoring index data are input in online Adding medicine control model, obtain current time optimal dosage；Wherein, online dosing model is established using Principal Component Analysis Algorithm, genetic algorithm and neural network model algorithm；The present invention reduces the dimension of training sample by Principal Component Analysis Algorithm, improves the speed of BP neural network model, improves the speed that dosage calculates；By the connection weight and threshold value of genetic algorithm optimization BP neural network model, the precision of prediction of BP neural network model is made to improve and be not easy to be absorbed in local optimum；Therefore, overcome dosing method in wastewater treatment process to be difficult to accurately determine the defect of dosing dosage using online Adding medicine control model provided by the invention, realize fast, accurately online Adding medicine control process.

Description

A kind of online Adding medicine control method and system for wastewater treatment
Technical field
The present invention relates to waste water of heat-engine plant processing technology fields, more particularly to a kind of online dosing control for wastewater treatment Method and system processed.
Background technology
Waste water of heat-engine plant processing procedure is the Nonlinear Dynamic system of a multivariable, large time delay, dynamic, serious interference System, is a kind of uncontrollable complex industrial process.Realize that waste water of heat-engine plant processing automation is to realize modernization processing and existing The necessary condition of generationization management is the necessary means for improving water treatment effect, reducing cost.Waste water of heat-engine plant processing procedure In, the techniques such as coagulation, flocculation are indispensable piths, and process is a complicated physical-chemical reaction process, right Dosing type and the control accuracy requirement of dosage are higher, and traditionally additive amount of medicament is that operations staff is sentenced by effluent quality It is disconnected, in order to reach effluent quality requirement by adjusting additive amount of medicament repeatedly, a large amount of time and artificial is wasted, belong to " after Know aftersensation " behavior, there is apparent hysteresis quality；Secondly medicament additive amount once it is determined that after, substantially belong to long term constant state, The invisible waste of drug is not only caused, and due to the time-varying characteristics of water water quality, effluent index cannot be satisfied the real-time of user It is required that.
Therefore the dosing process of water treatment of power plant shows randomness, hysteresis quality, non-linear property and traditional treatment method Defect realizes that online Adding medicine control is one urgent problem to be solved of waste water of heat-engine plant treatment industry.
Invention content
The object of the present invention is to provide a kind of online Adding medicine control method and system for wastewater treatment, overcome waste water Traditional dosing method is difficult to accurately determine the defect of dosing dosage in processing procedure, realizes fast, accurately online Adding medicine control Process.
To achieve the above object, the present invention provides following schemes：
A kind of online Adding medicine control method for wastewater treatment, the online Adding medicine control method include：
Obtain current time water monitoring index data；
The current time water monitoring index data are input in online Adding medicine control model, obtain current time most Excellent dosage；Wherein, the input of the online Adding medicine control model is the current time water monitoring index data；It is described The output of line Adding medicine control model is the current time optimal dosage；The online Adding medicine control model is according to principal component Parser, genetic algorithm and neural network model algorithm are established；The method for building up of the online Adding medicine control model It specifically includes：
Obtain training sample；The training sample includes multiple samples pair；Each sample to include multiple input, one Output；The input is to meet the water monitoring index data of effluent quality；The output is the water monitoring index data Corresponding optimal dosage；
Principal component vector square is obtained to the sample in the training sample to handling using Principal Component Analysis Algorithm The number of battle array and principal component component；
According to the number of the principal component vector matrix and the principal component component, BP neural network model is established；It is described BP neural network model is the three layer model of multiple input single output；The input of the BP neural network model is sweared for the principal component Moment matrix；The output of the BP neural network model is optimal dosage；The input neuron of the BP neural network model Number is the number of the principal component component；
Using genetic algorithm, the connection weight and threshold value of the BP neural network model are optimized, optimal company is obtained Connect weights and optimal threshold；
According to the optimal connection weight and the optimal threshold, the BP neural network model is updated；Updated BP Neural network model is the online Adding medicine control model.
Optionally, the sample is to the time series collection for input time sequence signal and output time series signal composition (X,Y)；
The input time sequence signal is the input value of the sample pair；The input time sequence signal is X= [xij]nⅹp, i=1,2 ... ... n, j=1,2 ... ... p, n are the sample of water monitoring index data described in the training sample Number, p are the number of the water monitoring index data；The water monitoring index data include that turbidity value, pH value, ammonia nitrogen contain Amount, COD value；
The output time series signal is the output valve of the sample pair；The output time series signal is Y= [yi]nⅹ1, i=1,2 ... ... n, n are the number of samples of dosage data in the training sample.
Optionally, described to be led to the sample in the training sample to handling using Principal Component Analysis Algorithm The number of component vectors matrix and principal component component, specifically includes：
Calculate the correlation matrix R of water monitoring index data described in the training sample；
According to characteristic equation | λ I-R |=0, characteristic value is calculated, the characteristic value is λj, j=1,2 ..., p, and press from big To the small sequence sequence characteristic value, λ1≥λ2≥…≥λp, wherein I indicates unit matrix；
Calculate each eigenvalue λj, j=1,2 ..., the corresponding feature vector e of pj, j=1,2 ..., p；Wherein, | | ej| |=1；
According to the characteristic value, contribution rate of accumulative total is calculated, chooses the contribution rate of accumulative total up to 85%~95% characteristic value, And the number of the contribution rate of accumulative total up to 85%~95% characteristic value is determined as to the number of the principal component component；It is described tired Count the calculation formula of contribution rate：
According to the characteristic value and described eigenvector, principal component load is calculated；The calculating of the principal component load is public Formula is：
According to the principal component load, principal component vector matrix is determined；The principal component vector matrix is：Z= [zit]nⅹm；
Optionally, the correlation matrixWherein, rabFor the trained sample X in thisaWith xbRelated coefficient, rab=rba, For Variable xaSample average,For variable xbSample average.
Optionally, the number according to the principal component vector matrix and the principal component component, establishes BP nerve nets Network model, specifically includes：
Establish BP neural network structure；The number of the principal component component is the input nerve of the BP neural network structure The number of member, the principal component vector matrix are the input quantity of the BP neural network structure；The BP neural network structure Output target is optimal dosage；The BP neural network structure is multiple input single output three layer model；
The connection weight and threshold value of the BP neural network structure are initialized, and by Sample Counter and learns counting how many times Device sets 1, determines minimal error and iterations；The connection weight includes hidden layer to weights, the output layer between input layer To the weights between hidden layer；The threshold value includes each neural in the threshold value of each neuron node, output layer in hidden layer The threshold value of first node；
By c-th of sample in the training sample to being input to the BP neural network structure, calculate each in hidden layer The the outputting and inputting of neuron node, each neuron node outputs and inputs in output layer；
According to the outputting and inputting of each neuron node, each neuron node in the output layer in the hidden layer Output and input, calculate output layer in each neuron node correction error, each neuron node of hidden layer correction Error determines the error of c-th of sample pair；
According to the error of c-th of sample pair, the connection weight and the threshold value are adjusted；
Judge sample all in the training sample to whether being all trained to；
If it is not, then return to step by c-th of sample in the training sample to being input to the BP neural network structure, count Calculate that the outputting and inputting of each neuron node in hidden layer, each neuron node outputs and inputs in output layer；
If so, renewal learning number, calculates global error, and judge whether the global error is less than the described of setting Whether minimal error or the study number reach the iterations of setting；
If so, according to connection weight and threshold value after adjustment, BP neural network model is established；
If it is not, then return to step by c-th of sample in the training sample to being input to the BP neural network structure, count Calculate that the outputting and inputting of each neuron node in hidden layer, each neuron node outputs and inputs in output layer.
Optionally, described to use genetic algorithm, the connection weight and threshold value of the BP neural network model are optimized, Optimal connection weight and optimal threshold are obtained, is specifically included：
By in the BP neural network model connection weight and sets of threshold values cooperation be a chromosome, constitute genetic algorithm Individual, and determine the number of individuals S of initial population, genetic iteration times N；
It is random to generate initialization population, and binary coding is carried out to the population after initialization, determine what S individual formed Initial population, juxtaposition evolution number are 1；
Determine the fitness function of the genetic algorithm；The fitness function is the error of the BP neural network model Function；
According to the fitness function, the fitness function value of each individual in the initial population is calculated；
Judge whether reach the genetic iteration number of setting when evolution number；
If so, output optimum individual；The optimum individual is maximum of fitness function value in the initial population Body；
If it is not, number of then evolving increases 1, the maximum individual of fitness function value in the initial population is selected, is handed over Fork, mutation genetic operation, update initial population, and return to step calculates every in the initial population according to the fitness function The fitness function value of individual.
The present invention also provides a kind of online control system for adding drugs for wastewater treatment, the online control system for adding drugs Including：
Water monitoring index data acquisition module, for obtaining current time water monitoring index data；
Optimal dosage acquisition module, for the current time water monitoring index data to be input to online dosing control In simulation, current time optimal dosage is obtained；Wherein, the input of the online Adding medicine control model is the current time Water monitoring index data；The output of the online Adding medicine control model is the current time optimal dosage；It is described online Adding medicine control model is established according to Principal Component Analysis Algorithm, genetic algorithm and neural network model algorithm；It is described The subsystem of establishing of line Adding medicine control model specifically includes：
Training sample acquisition module, for obtaining training sample；The training sample includes multiple samples pair；Each sample To including multiple input, an output；The input is to meet the water monitoring index data of effluent quality；The output is The corresponding optimal dosage of the water monitoring index data；
Principal component vector matrix and principal component component number acquisition module, for using Principal Component Analysis Algorithm, to described Sample in training sample obtains the number of principal component vector matrix and principal component component to handling；
BP neural network model building module, for according to the principal component vector matrix and the principal component component Number, establishes BP neural network model；The BP neural network model is the three layer model of multiple input single output；The BP nerve nets The input of network model is the principal component vector matrix；The output of the BP neural network model is optimal dosage；The BP The number of the input neuron of neural network model is the number of the principal component component；
Optimal connection weight and optimal threshold acquisition module, for using genetic algorithm, to the BP neural network model Connection weight and threshold value optimize, obtain optimal connection weight and optimal threshold；
BP neural network model modification module, for according to the optimal connection weight and the optimal threshold, updating institute State BP neural network model；Updated BP neural network model is the online Adding medicine control model.
Optionally, the principal component vector matrix and principal component component number acquisition module, specifically include：
Correlation matrix computing unit, the correlation for calculating water monitoring index data described in the training sample Coefficient matrix R；
Characteristic value computing unit, for according to characteristic equation | λ I-R |=0, calculate characteristic value, the eigenvalue λj, j= 1,2 ..., p, and by the sequence sequence characteristic value from big to small, λ1≥λ2≥…≥λp, wherein I indicates unit matrix；
Feature vector computing unit, for calculating each eigenvalue λj, j=1,2 ..., the corresponding feature vectors of p ej, j=1,2 ..., p；Wherein, | | ej| |=1；
Principal component LOAD FOR unit, for according to the characteristic value and described eigenvector, calculating principal component load； The calculation formula of the principal component load is：
Principal component vector matrix determination unit, for according to the principal component load, determining principal component vector matrix；It is described Principal component vector matrix is：Z=[zit]nⅹm；
Optionally, the BP neural network model building module, specifically includes：
BP neural network structure establishes unit, for establishing BP neural network structure；The number of the principal component component is The number of the input neuron of the BP neural network structure, the principal component vector matrix are the BP neural network structure Input quantity；The output target of the BP neural network structure is optimal dosage；The BP neural network structure is multi input list Export three layer model；
First initialization unit, connection weight and threshold value for initializing the BP neural network structure, and by sample Counter and study number counter set 1, determine minimal error and iterations；The connection weight includes hidden layer to input Weights, output layer to the weights between hidden layer between layer；The threshold value includes the threshold of each neuron node in hidden layer The threshold value of each neuron node in value, output layer；
Input and output computing unit, for by c-th of sample in the training sample to being input to the BP neural network Structure calculates the outputting and inputting of each neuron node in hidden layer, the input of each neuron node and defeated in output layer Go out；
Sample is to error determination unit, for according to the outputting and inputting of each neuron node, institute in the hidden layer Outputting and inputting for each neuron node in output layer is stated, the correction error of each neuron node in output layer, hidden is calculated The correction error of each neuron node containing layer determines the error of c-th of sample pair；
Connection weight and threshold adjustment unit adjust the connection weight for the error according to c-th of sample pair And the threshold value；
First judging unit, for judging sample all in the training sample to whether being all trained to；
First returning unit, for when having sample to not being trained in the training sample, return to step is by the instruction Practice c-th of sample in sample and calculates the input of each neuron node in hidden layer to being input to the BP neural network structure With output, in output layer, each neuron node outputs and inputs；
Second judgment unit, for when sample all in the training sample is to being all trained to, renewal learning number, Calculate global error, and judge the global error whether be less than setting the minimal error or the study number whether Reach the iterations of setting；
BP neural network model foundation unit, the minimal error or institute for being less than setting when the global error Study number is stated when reaching the iterations of setting, according to the connection weight and threshold value after adjustment, establishes BP neural network Model；
Second returning unit, for being reached less than the minimal error and the study number set when the global error To setting the iterations when, return to step is by c-th of sample in the training sample to being input to the BP nerve nets Network structure, calculate the outputting and inputting of each neuron node in hidden layer, in output layer the input of each neuron node and Output.
Optionally, the optimal connection weight and optimal threshold acquisition module, specifically include：
Second initialization unit, for by the BP neural network model connection weight and sets of threshold values cooperation be one Chromosome constitutes the individual of genetic algorithm, and determines the number of individuals S of initial population, genetic iteration times N；
Initial population determination unit carries out binary system for generating initialization population at random, and to the population after initialization Coding determines that the initial population of S individual composition, juxtaposition evolution number are 1；
Fitness function determination unit, the fitness function for determining the genetic algorithm；The fitness function is The error function of the BP neural network model；
Fitness function value computing unit, for according to the fitness function, calculating in the initial population per each and every one The fitness function value of body；
Third judging unit, for judging whether reach the genetic iteration number of setting when evolution number；
Optimum individual output unit, for when the genetic iteration number for reaching setting when evolution number, Export optimum individual；The optimum individual is the maximum individual of fitness function value in the initial population；
Initial population updating unit, for when the genetic iteration number not up to set when evolution number When, evolution number increases 1, and the maximum individual of fitness function value in the initial population is selected, is intersected, mutation genetic behaviour Make, update initial population, return to step calculates the fitness of each individual in the initial population according to the fitness function Functional value.
According to specific embodiment provided by the invention, the invention discloses following technique effects：
The present invention provides a kind of online Adding medicine control method and system for wastewater treatment, the online Adding medicine controls Method includes：Current time water monitoring index data are obtained, and current time water monitoring index data are input to online In Adding medicine control model, current time optimal dosage is obtained；Wherein, the online Adding medicine control model is using principal component point Analysis algorithm, genetic algorithm and neural network model algorithm are established；The present invention is reduced by Principal Component Analysis Algorithm and is trained The dimension of sample simplifies the structure of BP neural network model, improves the speed of BP neural network model, and then improve dosage The speed of calculating.The present invention is optimal by what is obtained by the connection weight and threshold value of genetic algorithm optimization BP neural network model Connection weight and threshold value assign BP neural network model, to make the precision of prediction of BP neural network model improve and be not easy to fall into Enter local optimum.The present invention overcomes traditional dosing method in wastewater treatment process to be difficult to accurately really using online Adding medicine control model Determine the defect of dosing dosage, realizes fast, accurately online Adding medicine control process.
Description of the drawings
It in order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, below will be to institute in embodiment Attached drawing to be used is needed to be briefly described, it should be apparent that, the accompanying drawings in the following description is only some implementations of the present invention Example, for those of ordinary skill in the art, without having to pay creative labor, can also be according to these attached drawings Obtain other attached drawings.
Fig. 1 is flow diagram of the embodiment of the present invention for the online Adding medicine control method of wastewater treatment；
Fig. 2 is the flow diagram of the online Adding medicine control method for establishing model of the embodiment of the present invention；
Fig. 3 is the flow diagram that the embodiment of the present invention uses online Adding medicine control method for the first time；
Fig. 4 is structural schematic diagram of the embodiment of the present invention for the online control system for adding drugs of wastewater treatment.
Specific implementation mode
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete Site preparation describes, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other Embodiment shall fall within the protection scope of the present invention.
The object of the present invention is to provide a kind of online Adding medicine control method and system for wastewater treatment, overcome waste water Traditional dosing method is difficult to accurately determine the defect of dosing dosage in processing procedure, realizes fast, accurately online Adding medicine control Process.
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, below in conjunction with the accompanying drawings and specific real Applying mode, the present invention is described in further detail.
BP neural network model is a kind of Multi-Layered Network Model based on gradient descent method, it be by Rumelhart and What McClelland was proposed in 1986, which is a kind of with very strong parallel processing capability and self-learning capability Algorithm can effectively to complicated unstructuredness problem infinitely approach to carry out the solution of problem, to efficiently solve Complicated unstructuredness problem provides a new approaches.The critical issue of BP neural network model is to input in network structure The selection of the connection weight, threshold value between connection weight, threshold value and output layer and hidden layer between layer and hidden layer, value Selection has prediction result important influence.
Genetic algorithm (Genetic Algorithm, hereinafter referred to as GA) is the Holland professors of Michigan universities of the U.S. It was proposed in 1969, a kind of simulated evolutionary algorithm formed by the induction and conclusion by DeJong, Goldberg et al..It It is to use for reference natural selection and a kind of randomization of genetic mechanism is oriented searches derived from Darwinian evolution and Mendelian theory Of heredity Rope algorithm.Genetic algorithm is the survival of the fittest in the natural environment of simulation biology, the evolutionary process of the survival of the fittest and formed one Kind full search algorithm.Genetic algorithm is because adaptable, and global optimization, algorithm is simple, it is general the features such as, structure model solution It is applied during challenge, such as function optimizing, Combinatorial Optimization, machine learning, pattern-recognition etc..
Principal component analysis (Principal Components Analysis, hereinafter referred to as PCA) algorithm be by K.Pearson introduces nonrandom variable, after situation vectorial immediately was generalized in 1933 by Hotelling.Its main thought It is attempt to try hard to keep under the minimum principle of loss of data, dimension-reduction treatment is carried out to high-dimensional data space.In BP neural network model PCA algorithms are introduced, dimensionality reduction mainly is carried out to training sample, simplify the structure of BP neural network model.
The present invention proposes a kind of online Adding medicine control method and system for wastewater treatment, and this method is to current thermoelectricity There is subjectivity and hysteresis quality in factory's water treatment dosing system, in conjunction with BP neural network model solution linear problem, GA by no means Global parameter optimizing and PCA dimensionality reductions reduce the characteristic of BP network structures, select effluent quality it is satisfactory come water monitoring refer to Mark and dosing data reduce sample dimension using PCA, carry out BP network connections weights and threshold using GA algorithms as training sample The parameter of value preferably, using BP models carries out the training of sample, finally obtains based on principal component analysis-genetic algorithm-BP nerves The online Adding medicine control model of network model algorithm passes through the control of the implementation dosage of model realization intelligence.
Fig. 1 is flow diagram of the embodiment of the present invention for the online Adding medicine control method of wastewater treatment, such as Fig. 1 institutes Show, online Adding medicine control method provided in an embodiment of the present invention specifically includes following steps：
Step 101：Obtain current time water monitoring index data.Water monitoring index data mainly include turbidity value, PH value, ammonia-nitrogen content, COD value.
Step 102：The current time water monitoring index data are input in online Adding medicine control model, are worked as Preceding moment optimal dosage；Wherein, the input of the online Adding medicine control model is the current time water monitoring index number According to；The output of the online Adding medicine control model is the current time optimal dosage；The online Adding medicine control model is It is established according to Principal Component Analysis Algorithm, genetic algorithm and neural network model algorithm.
Fig. 2 is the flow diagram of the online Adding medicine control method for establishing model of the embodiment of the present invention, as shown in Fig. 2, this hair The online specific following steps of Adding medicine control method for establishing model that bright embodiment provides：
Step 201：Obtain training sample；The training sample includes multiple samples pair；Each sample is to including multiple Input, an output；The input is to meet the water monitoring index data of effluent quality；It is described output for it is described come water monitoring The corresponding optimal dosage of achievement data.
Step 202：Using Principal Component Analysis Algorithm, to the sample in the training sample to handling, obtain it is main at The number of resolute matrix and principal component component.
Step 203：According to the number of the principal component vector matrix and the principal component component, BP neural network mould is established Type；The BP neural network model is the three layer model of multiple input single output；The input of the BP neural network model is described Principal component vector matrix；The output of the BP neural network model is optimal dosage；The input of the BP neural network model The number of neuron is the number of the principal component component.
Step 204：Using genetic algorithm, the connection weight and threshold value of the BP neural network model are optimized, obtained To optimal connection weight and optimal threshold.
Step 205：According to the optimal connection weight and the optimal threshold, the BP neural network model is updated；More BP neural network model after new is the online Adding medicine control model.
Step 201 specifically includes：
Water monitoring index data in training sample include mainly turbidity value, pH value, ammonia-nitrogen content, COD value, time Sequence signal X=[xij]nⅹpAs the input of online Adding medicine control model, wherein i=1,2 ... ... n, j=1,2 ... ... p, n For the number of samples of water monitoring index data in training sample, p is the number of water monitoring index data.
Dosing type in training sample is mainly coagulant, the time series signal Y=[y of dosagei]nⅹ1As The reality output of line Adding medicine control model, wherein i=1,2 ... ... n, n are the number of samples of dosage data in training sample. Coagulant is not limited to for the adding medicine type, the calculating of the medicaments such as other medicaments such as sodium hydroxide, fungicide, flocculation aid It can be calculated in the method in accordance with the invention.
Input time sequence signal xijWith output time series signal yiThe time series collection (X, Y) of composition adds as online The training sample of medicine Controlling model, i.e., the described sample is to for input time sequence signal xijWith output time series signal yiComposition Time series collection (X, Y).
Step 202 specifically includes：
Step 2021：Calculate the correlation matrix R of water monitoring index data X in training sample；
Wherein, rabFor x in training sampleaWith xbRelated coefficient, rab=rba, calculation formula is：
Wherein,For variable xaSample average,For variable xbSample average.
Step 2022：Calculate eigen vector；According to characteristic equation | λ I-R |=0 solves eigenvalue λj, j=1, 2 ..., p, and by sequence sequence λ from big to small1≥λ2≥…≥λp.Wherein I indicates unit matrix.It calculates separately out each Eigenvalue λj, j=1,2 ..., the corresponding feature vector e of pj, j=1,2 ..., p, it is desirable that | | ej| |=1, i.e., Wherein ejiFor vectorial ejI-th of component.
Step 2023：Calculate contribution rate of accumulative total；
The calculation formula of the contribution rate of accumulative total：
It takes contribution rate of accumulative total up to 85%~95% m characteristic value, obtains λ1, λ2..., λmM corresponding principal component point Amount.
Step 2024：Calculate principal component load；
The principal component LOAD FOR formula is：
Step 2025：According to principal component load, each principal component vector matrix is calculated；The principal component vector matrix is Z= [zit]nⅹm：
Step 203 specifically includes：
Step 2031：Determine BP neural network structure：The m principal component being calculated according to Principal Component Analysis Algorithm Number determines that BP neural network input neuron is m, the dosage of output target coagulant in order to control, for single output, therefore uses Multiple input single output (MISO) three layer model.
Step 2032：The principal component vector matrix Z=[z that the Principal Component Analysis Algorithm is obtainedit]nⅹmAs input Vector, Y=[yi]nⅹ1As output vector, H=[h1,h2,……,hl]TFor the output vector of hidden layer,As Desired output vector, W=[wdt]lⅹmAs the hidden layer to the weights between input layer, θ=[θd]l×1For the every of hidden layer The threshold value of a neuron node, V=[wdi]1ⅹlAs the output layer to the weights between hidden layer, q is output layer neuron The threshold value of node.
Step 2033：Initialize network connection weights and the threshold value (hidden layer to weights, the hidden layer between input layer The threshold value of each neuron node, the output layer to weights and output layer neuron node between hidden layer threshold Value), Sample Counter and study number counter all set 1, set minimal error and iterations.
Step 2034：It inputs c-th of sample pair, calculates the outputting and inputting of each neuron node of hidden layer, output layer Each neuron node is output and input；The input z of the sample pairc=[zc1,zc2,……,zcm] and output yc。
Step 2035：The correction of the correction error, each neuron node of hidden layer that calculate each neuron node of output layer misses Difference obtains the error of c-th of sample pair of BP neural network model (6)。
Step 2036：According to the error of c-th of sample pair, adjustment hidden layer between output layer connection weight and output Threshold value, the threshold value of input layer to connection weight and each neuron node of hidden layer between hidden layer of each neuron node of layer, Obtain new weight W, V and threshold θ=[θd]l×1、q。
Step 2037：All samples are judged to whether being all trained to, if all training thens follow the steps 2038, are otherwise held Row step 2034.
Step 2038：Renewal learning number, and calculate global errorJudge E Whether it is less than the minimal error of setting or learns the iterations whether number reaches setting, if it is terminates, establish BP Neural network model, it is no to then follow the steps 2034.
Step 204 specifically includes：
Step 2041：By in BP neural network model connection weight and sets of threshold values cooperation be a chromosome, constitute lose The individual of propagation algorithm.Determine initial population scale S, genetic iteration times N, crossing-over rate Pc, aberration rate Pm。
Step 2042：It is random to generate initialization population, and binary coding is carried out to population, S group initial populations are generated, and It is 1 to set evolution number.
Step 2043, using following circulation step, the optimal solution of BP neural network connection weight and threshold value is obtained.It is described to follow Ring step is：
Step S1：The fitness function of genetic algorithm is set by the error function of BP neural network modelC is constant,For time series signal zitWith time series signal yiInstitute's group At i-th group of data mode under, the desired output being calculated by BP neural network model is as a result, i=1,2 ... ..., n.
Step S2：The fitness function value that each individual in initial population is calculated according to fitness function, according to fitness Functional value assesses every chromosome.
Step S3：Judge whether reach the genetic iteration number of setting when evolution number, if reaching, exports optimal Body, end step 2043.Otherwise, evolution number increases 1, preserves the corresponding individual of highest fitness function value, is selected, handed over Fork, mutation genetic operation, update initial population, return to step S1.
Further, the step S3 genetic manipulation specific steps include：
A：Selection, selection operation is to select excellent individual, generally use roulette selection plan in current group Slightly, thought is exactly that the selected probability of each individual is directly proportional to the size of its fitness function value, and g-th of individual is hereditary It is to follow-on probability：Wherein, N is Population Size, FgFor the fitness function value of individual g.
B：Intersect, to two chromosomes being mutually paired according to the general P of intersectioncIt is exchanged with each other its part base by certain way Cause, to form two new individuals.
C：Variation refers to according to the general P that makes a variationmCertain genic values in individual UVR exposure string are replaced with other genic values, from And form a new individual.
Fig. 3 is the flow diagram that the embodiment of the present invention uses online Adding medicine control method for the first time, as shown in figure 3, Fig. 3 is The detailed process of flow shown in Fig. 1 and flow shown in Fig. 2 describes.
The present invention is directed to the parameter value of the water monitoring index real-time monitored, is entered into the online dosing control of foundation In simulation, you can the dosage of current time adding medicine is obtained, the On-line Control of dosing process is realized, it can be according to current The optimal dosage of water quality situation real-time update overcomes water water quality time-varying characteristics and is adversely affected caused by operational process, subtracted Lack the waste of medicament, reduced artificial and operating cost, meets the water quality requirement of water outlet in real time.
To achieve the above object, the present invention also provides a kind of online control system for adding drugs for wastewater treatment.
Fig. 4 is structural schematic diagram of the embodiment of the present invention for the online control system for adding drugs of wastewater treatment, such as Fig. 4 institutes Show, the online control system for adding drugs includes：
Water monitoring index data acquisition module 100, for obtaining current time water monitoring index data.
Optimal dosage acquisition module 200, for the current time water monitoring index data to be input to online add In medicine Controlling model, current time optimal dosage is obtained；Wherein, the input of the online Adding medicine control model is described current Moment water monitoring index data；The output of the online Adding medicine control model is the current time optimal dosage；It is described Online Adding medicine control model is established according to Principal Component Analysis Algorithm, genetic algorithm and neural network model algorithm.
The subsystem of establishing of the online Adding medicine control model specifically includes：
Training sample acquisition module 300, for obtaining training sample；The training sample includes multiple samples pair；Each Sample is to including multiple input, an output；The input is to meet the water monitoring index data of effluent quality；It is described defeated Go out for the corresponding optimal dosage of the water monitoring index data.
Principal component vector matrix and principal component component number acquisition module 400, for using Principal Component Analysis Algorithm, to institute The sample in training sample is stated to handling, obtains the number of principal component vector matrix and principal component component.
BP neural network model building module 500, for according to the principal component vector matrix and the principal component component Number, establish BP neural network model；The BP neural network model is the three layer model of multiple input single output；The BP god Input through network model is the principal component vector matrix；The output of the BP neural network model is optimal dosage；Institute The number for stating the input neuron of BP neural network model is the number of the principal component component.
Optimal connection weight and optimal threshold acquisition module 600, for using genetic algorithm, to the BP neural network mould The connection weight and threshold value of type optimize, and obtain optimal connection weight and optimal threshold.
BP neural network model modification module 700, for according to the optimal connection weight and the optimal threshold, update The BP neural network model；Updated BP neural network model is the online Adding medicine control model.
The principal component vector matrix and principal component component number acquisition module 400, specifically include：
Correlation matrix computing unit, the correlation for calculating water monitoring index data described in the training sample Coefficient matrix R.
Characteristic value computing unit, for according to characteristic equation | λ I-R |=0, calculate characteristic value, the eigenvalue λj, j= 1,2 ..., p, and by the sequence sequence characteristic value from big to small, λ1≥λ2≥…≥λp, wherein I indicates unit matrix.
Feature vector computing unit, for calculating each eigenvalue λj, j=1,2 ..., the corresponding feature vectors of p ej, j=1,2 ..., p；Wherein, | | ej| |=1.
Contribution rate of accumulative total computing unit chooses the accumulative contribution for according to the characteristic value, calculating contribution rate of accumulative total Characteristic value of the rate up to 85%~95%, and the number of the contribution rate of accumulative total up to 85%~95% characteristic value is determined as described The number of principal component component；The calculation formula of the contribution rate of accumulative total：
Principal component LOAD FOR unit, for according to the characteristic value and described eigenvector, calculating principal component load； The calculation formula of the principal component load is：
Principal component vector matrix determination unit, for according to the principal component load, determining principal component vector matrix；It is described Principal component vector matrix is：Z=[zit]nⅹm；
The BP neural network model building module 500, specifically includes：
BP neural network structure establishes unit, for establishing BP neural network structure；The number of the principal component component is The number of the input neuron of the BP neural network structure, the principal component vector matrix are the BP neural network structure Input quantity；The output target of the BP neural network structure is dosage；The BP neural network structure is multiple input single output Three layer model.
First initialization unit, connection weight and threshold value for initializing the BP neural network structure, and by sample Counter and study number counter set 1, determine minimal error and iterations；The connection weight includes hidden layer to input Weights, output layer to the weights between hidden layer between layer；The threshold value includes the threshold of each neuron node in hidden layer The threshold value of each neuron node in value, output layer.
Input and output computing unit, for by c-th of sample in the training sample to being input to the BP neural network Structure calculates the outputting and inputting of each neuron node in hidden layer, the input of each neuron node and defeated in output layer Go out.
Sample is to error determination unit, for according to the outputting and inputting of each neuron node, institute in the hidden layer Outputting and inputting for each neuron node in output layer is stated, the correction error of each neuron node in output layer, hidden is calculated The correction error of each neuron node containing layer determines the error of c-th of sample pair.
Connection weight and threshold adjustment unit adjust the connection weight for the error according to c-th of sample pair And the threshold value.
First judging unit, for judging sample all in the training sample to whether being all trained to.
First returning unit, for when having sample to not being trained in the training sample, return to step is by the instruction Practice c-th of sample in sample and calculates the input of each neuron node in hidden layer to being input to the BP neural network structure With output, in output layer, each neuron node outputs and inputs.
Second judgment unit, for when sample all in the training sample is to being all trained to, renewal learning number, Global error is calculated, and judges whether global error is less than the minimal error of setting or whether the study number reaches The iterations of setting.
BP neural network model foundation unit, the minimal error or institute for being less than setting when the global error Study number is stated when reaching the iterations of setting, according to the connection weight and threshold value after adjustment, establishes BP neural network Model.
Second returning unit, for being reached less than the minimal error and the study number set when the global error To setting the iterations when, return to step is by c-th of sample in the training sample to being input to the BP nerve nets Network structure, calculate the outputting and inputting of each neuron node in hidden layer, in output layer the input of each neuron node and Output.
The optimal connection weight and optimal threshold acquisition module 600, specifically include：
Second initialization unit, for by the BP neural network model connection weight and sets of threshold values cooperation be one Chromosome constitutes the individual of genetic algorithm, and determines the number of individuals S of initial population, genetic iteration times N.
Initial population determination unit carries out binary system for generating initialization population at random, and to the population after initialization Coding determines that the initial population of S individual composition, juxtaposition evolution number are 1.
Fitness function determination unit, the fitness function for determining the genetic algorithm；The fitness function is The error function of the BP neural network model.
Fitness function value computing unit, for according to the fitness function, calculating in the initial population per each and every one The fitness function value of body.
Third judging unit, for judging whether reach the genetic iteration number of setting when evolution number.
Optimum individual output unit, for when the genetic iteration number for reaching setting when evolution number, Export optimum individual；The optimum individual is the maximum individual of fitness function value in the initial population.
Initial population updating unit, for when the genetic iteration number not up to set when evolution number When, evolution number increases 1, and the maximum individual of fitness function value in the initial population is selected, is intersected, mutation genetic behaviour Make, update initial population, return to step calculates the fitness of each individual in the initial population according to the fitness function Functional value.
Compared with prior art, beneficial effects of the present invention are：
1, the dimension of training sample is reduced by Principal Component Analysis Algorithm, simplifies the structure of BP neural network model, is improved The speed of BP neural network model, and then improve the speed that dosage calculates.
2, by the connection weight and threshold value of genetic algorithm optimization BP neural network model, the optimal connection weight that will be obtained BP neural network model is assigned with threshold value, to make the precision of prediction of BP neural network model improve and be not easy to be absorbed in part most Excellent problem.
3, more most important, solve the hysteresis quality dosing that effluent quality index can only be relied on to debug repeatedly in traditional technology Method, operating personnel need not can obtain out the dosage at current time most according to the real-time measurement of effluent quality index The figure of merit overcomes error caused by operating personnel's subjectivity, while also solving water water quality time-varying characteristics and being made to operational process At adverse effect.The process had not only reduced the waste of medicament, has reduced artificial and operating cost, but also can meet the water of water outlet Matter requirement.
Each embodiment is described by the way of progressive in this specification, the highlights of each of the examples are with other The difference of embodiment, just to refer each other for identical similar portion between each embodiment.For system disclosed in embodiment For, since it is corresponded to the methods disclosed in the examples, so description is fairly simple, related place is said referring to method part It is bright.
Principle and implementation of the present invention are described for specific case used herein, and above example is said The bright method and its core concept for being merely used to help understand the present invention；Meanwhile for those of ordinary skill in the art, foundation The thought of the present invention, there will be changes in the specific implementation manner and application range.In conclusion the content of the present specification is not It is interpreted as limitation of the present invention.

Claims (10)

1. a kind of online Adding medicine control method for wastewater treatment, which is characterized in that the online Adding medicine control method includes：
Obtain current time water monitoring index data；
The current time water monitoring index data are input in online Adding medicine control model, obtain current time it is optimal plus Dose；Wherein, the input of the online Adding medicine control model is the current time water monitoring index data；It is described to add online The output of medicine Controlling model is the current time optimal dosage；The online Adding medicine control model is according to principal component analysis Algorithm, genetic algorithm and neural network model algorithm are established；The method for building up of the online Adding medicine control model is specific Including：
Obtain training sample；The training sample includes multiple samples pair；Each sample to include multiple input, one it is defeated Go out；The input is to meet the water monitoring index data of effluent quality；The output is the water monitoring index data pair The optimal dosage answered；
Using Principal Component Analysis Algorithm, to the sample in the training sample to handling, obtain principal component vector matrix and The number of principal component component；
According to the number of the principal component vector matrix and the principal component component, BP neural network model is established；The BP god Through the three layer model that network model is multiple input single output；The input of the BP neural network model is the principal component vector square Battle array；The output of the BP neural network model is optimal dosage；The number of the input neuron of the BP neural network model For the number of the principal component component；
Using genetic algorithm, the connection weight and threshold value of the BP neural network model are optimized, optimal connection weight is obtained Value and optimal threshold；
According to the optimal connection weight and the optimal threshold, the BP neural network model is updated；Updated BP nerves Network model is the online Adding medicine control model.
2. online Adding medicine control method according to claim 1, which is characterized in that the sample is to for input time sequence The time series collection (X, Y) of signal and output time series signal composition；
The input time sequence signal is the input value of the sample pair；The input time sequence signal is X=[xij]nⅹp, I=1,2 ... ... n, j=1,2 ... ... p, n are the number of samples of water monitoring index data described in the training sample, and p is The number of the water monitoring index data；The water monitoring index data include turbidity value, pH value, ammonia-nitrogen content, COD Value；
The output time series signal is the output valve of the sample pair；The output time series signal is Y=[yi]nⅹ1, I=1,2 ... ... n, n are the number of samples of dosage data in the training sample.
3. online Adding medicine control method according to claim 1, which is characterized in that it is described to use Principal Component Analysis Algorithm, To the sample in the training sample to handling, the number of principal component vector matrix and principal component component is obtained, specific packet It includes：
Calculate the correlation matrix R of water monitoring index data described in the training sample；
According to characteristic equation | λ I-R |=0, characteristic value is calculated, the characteristic value is λj, j=1,2 ..., p, and by from big to small Sequence sort the characteristic value, λ1≥λ2≥…≥λp, wherein I indicates unit matrix；
Calculate each eigenvalue λj, j=1,2 ..., the corresponding feature vector e of pj, j=1,2 ..., p；Wherein, | | ej|| =1；
According to the characteristic value, contribution rate of accumulative total is calculated, chooses the contribution rate of accumulative total up to 85%~95% characteristic value, and will The number of the contribution rate of accumulative total up to 85%~95% characteristic value is determined as the number of the principal component component；The accumulative tribute Offer the calculation formula of rate：
According to the characteristic value and described eigenvector, principal component load is calculated；The calculation formula of the principal component load is：
According to the principal component load, principal component vector matrix is determined；The principal component vector matrix is：
4. online Adding medicine control method according to claim 3, which is characterized in that the correlation matrixWherein, rabFor x in the training sampleaWith xbRelated coefficient, For variable xaSample average,For variable xbSample average.
5. online Adding medicine control method according to claim 1, which is characterized in that described according to the principal component vector square The number of battle array and the principal component component, establishes BP neural network model, specifically includes：
Establish BP neural network structure；The number of the principal component component is the input neuron of the BP neural network structure Number, the principal component vector matrix are the input quantity of the BP neural network structure；The output of the BP neural network structure Target is optimal dosage；The BP neural network structure is multiple input single output three layer model；
The connection weight and threshold value of the BP neural network structure are initialized, and Sample Counter and study number counter are set 1, determine minimal error and iterations；The connection weight include hidden layer between input layer weights, output layer is to hidden Containing the weights between layer；The threshold value includes each neuron section in the threshold value of each neuron node, output layer in hidden layer The threshold value of point；
By c-th of sample in the training sample to being input to the BP neural network structure, each nerve in hidden layer is calculated The the outputting and inputting of first node, each neuron node outputs and inputs in output layer；
According to the outputting and inputting of each neuron node in the hidden layer, each neuron node is defeated in the output layer Enter and export, calculates correction error, the correction error of each neuron node of hidden layer of each neuron node in output layer, Determine the error of c-th of sample pair；
According to the error of c-th of sample pair, the connection weight and the threshold value are adjusted；
Judge sample all in the training sample to whether being all trained to；
If it is not, then for return to step by c-th of sample in the training sample to being input to the BP neural network structure, calculating is hidden Containing the outputting and inputting of each neuron node in layer, each neuron node outputs and inputs in output layer；
If so, renewal learning number, calculates global error, and judge whether the global error is less than the minimum of setting Whether error or the study number reach the iterations of setting；
If so, according to connection weight and threshold value after adjustment, BP neural network model is established；
If it is not, then for return to step by c-th of sample in the training sample to being input to the BP neural network structure, calculating is hidden Containing the outputting and inputting of each neuron node in layer, each neuron node outputs and inputs in output layer.
6. online Adding medicine control method according to claim 1, which is characterized in that it is described to use genetic algorithm, to described The connection weight and threshold value of BP neural network model optimize, and obtain optimal connection weight and optimal threshold, specifically include：
By in the BP neural network model connection weight and sets of threshold values cooperation be a chromosome, constitute genetic algorithm Body, and determine the number of individuals S of initial population, genetic iteration times N；
It is random to generate initialization population, and binary coding is carried out to the population after initialization, it is initial to determine that S individual forms Population, juxtaposition evolution number are 1；
Determine the fitness function of the genetic algorithm；The fitness function is the error letter of the BP neural network model Number；
According to the fitness function, the fitness function value of each individual in the initial population is calculated；
Judge whether reach the genetic iteration number of setting when evolution number；
If so, output optimum individual；The optimum individual is the maximum individual of fitness function value in the initial population；
If it is not, number of then evolving increases 1, the maximum individual of fitness function value in the initial population is selected, intersected, is become Different genetic manipulation updates initial population, and return to step calculates each individual in the initial population according to the fitness function Fitness function value.
7. a kind of online control system for adding drugs for wastewater treatment, which is characterized in that the online control system for adding drugs includes：
Water monitoring index data acquisition module, for obtaining current time water monitoring index data；
Optimal dosage acquisition module, for the current time water monitoring index data to be input to online Adding medicine control mould In type, current time optimal dosage is obtained；Wherein, the input of the online Adding medicine control model is the current time water Monitoring index data；The output of the online Adding medicine control model is the current time optimal dosage；The online dosing Controlling model is established according to Principal Component Analysis Algorithm, genetic algorithm and neural network model algorithm；It is described to add online The subsystem of establishing of medicine Controlling model specifically includes：
Training sample acquisition module, for obtaining training sample；The training sample includes multiple samples pair；Each sample is to equal Including multiple input, an output；The input is to meet the water monitoring index data of effluent quality；The output is described The corresponding optimal dosage of water monitoring index data；
Principal component vector matrix and principal component component number acquisition module, for using Principal Component Analysis Algorithm, to the training Sample in sample obtains the number of principal component vector matrix and principal component component to handling；
BP neural network model building module is used for the number according to the principal component vector matrix and the principal component component, Establish BP neural network model；The BP neural network model is the three layer model of multiple input single output；The BP neural network The input of model is the principal component vector matrix；The output of the BP neural network model is optimal dosage；The BP god The number of input neuron through network model is the number of the principal component component；
Optimal connection weight and optimal threshold acquisition module, for using genetic algorithm, to the company of the BP neural network model It connects weights and threshold value optimizes, obtain optimal connection weight and optimal threshold；
BP neural network model modification module, for according to the optimal connection weight and the optimal threshold, updating the BP Neural network model；Updated BP neural network model is the online Adding medicine control model.
8. online control system for adding drugs according to claim 7, which is characterized in that the principal component vector matrix and it is main at Divide component number acquisition module, specifically includes：
Correlation matrix computing unit, the related coefficient for calculating water monitoring index data described in the training sample Matrix R；
Characteristic value computing unit, for according to characteristic equation | λ I-R |=0, calculate characteristic value, the eigenvalue λj, j=1, 2 ..., p, and by the sequence sequence characteristic value from big to small, λ1≥λ2≥…≥λp, wherein I indicates unit matrix；
Feature vector computing unit, for calculating each eigenvalue λj, j=1,2 ..., the corresponding feature vector e of pj, j= 1,2,...,p；Wherein, | | ej| |=1；
Contribution rate of accumulative total computing unit, for according to the characteristic value, calculating contribution rate of accumulative total, choosing the contribution rate of accumulative total and reach 85%~95% characteristic value, and by the number of the contribution rate of accumulative total up to 85%~95% characteristic value be determined as it is described it is main at Divide the number of component；The calculation formula of the contribution rate of accumulative total：
Principal component LOAD FOR unit, for according to the characteristic value and described eigenvector, calculating principal component load；It is described The calculation formula of principal component load is：
Principal component vector matrix determination unit, for according to the principal component load, determining principal component vector matrix；It is described it is main at Resolute matrix is：
9. online control system for adding drugs according to claim 7, which is characterized in that the BP neural network model foundation mould Block specifically includes：
BP neural network structure establishes unit, for establishing BP neural network structure；The number of the principal component component is described The number of the input neuron of BP neural network structure, the principal component vector matrix are the input of the BP neural network structure Amount；The output target of the BP neural network structure is optimal dosage；The BP neural network structure is multiple input single output Three layer model；
First initialization unit, connection weight and threshold value for initializing the BP neural network structure, and by sample counting Device and study number counter set 1, determine minimal error and iterations；The connection weight include hidden layer to input layer it Between weights, output layer to the weights between hidden layer；The threshold value includes the threshold value of each neuron node in hidden layer, defeated Go out the threshold value of each neuron node in layer；
Input and output computing unit, for by c-th of sample in the training sample to being input to the BP neural network structure, Calculate that the outputting and inputting of each neuron node in hidden layer, each neuron node outputs and inputs in output layer；
Sample is to error determination unit, for according to the outputting and inputting of each neuron node in the hidden layer, described defeated Go out outputting and inputting for each neuron node in layer, calculates the correction error of each neuron node, hidden layer in output layer The correction error of each neuron node determines the error of c-th of sample pair；
Connection weight and threshold adjustment unit adjust the connection weight and institute for the error according to c-th of sample pair State threshold value；
First judging unit, for judging sample all in the training sample to whether being all trained to；
First returning unit, for when having sample to not being trained in the training sample, return to step is by the trained sample C-th of sample be to being input to the BP neural network structure in this, calculates in hidden layer the input of each neuron node and defeated Go out, each neuron node outputs and inputs in output layer；
Second judgment unit, for when sample all in the training sample is to being all trained to, renewal learning number to calculate Global error, and judge whether the global error is less than the minimal error of setting or whether the study number reaches The iterations of setting；
BP neural network model foundation unit, for being less than the minimal error of setting or when the global error When habit number reaches the iterations of setting, according to the connection weight and threshold value after adjustment, BP neural network model is established；
Second returning unit, for being set when the global error reaches less than the minimal error and the study number set When the fixed iterations, return to step is by c-th of sample in the training sample to being input to the BP neural network knot Structure, calculates that the outputting and inputting of each neuron node in hidden layer, each neuron node outputs and inputs in output layer.
10. online control system for adding drugs according to claim 7, which is characterized in that the optimal connection weight and optimal Threshold value acquisition module, specifically includes：
Second initialization unit, for by the BP neural network model connection weight and sets of threshold values cooperation be one dyeing Body constitutes the individual of genetic algorithm, and determines the number of individuals S of initial population, genetic iteration times N；
Initial population determination unit carries out binary coding for generating initialization population at random, and to the population after initialization, Determine that the initial population of S individual composition, juxtaposition evolution number are 1；
Fitness function determination unit, the fitness function for determining the genetic algorithm；The fitness function is described The error function of BP neural network model；
Fitness function value computing unit, for according to the fitness function, calculating each individual in the initial population Fitness function value；
Third judging unit, for judging whether reach the genetic iteration number of setting when evolution number；
Optimum individual output unit, for when the genetic iteration number for reaching setting when evolution number, output Optimum individual；The optimum individual is the maximum individual of fitness function value in the initial population；
Initial population updating unit is used for when described when evolution number is not up to the genetic iteration number set, into Change number and increase 1, the maximum individual of fitness function value in the initial population is selected, is intersected, mutation genetic operation, more New initial population, return to step calculate the fitness function of each individual in the initial population according to the fitness function Value.
CN201810315713.5A 2018-04-10 2018-04-10 A kind of online Adding medicine control method and system for wastewater treatment Active CN108408855B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810315713.5A CN108408855B (en) 2018-04-10 2018-04-10 A kind of online Adding medicine control method and system for wastewater treatment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810315713.5A CN108408855B (en) 2018-04-10 2018-04-10 A kind of online Adding medicine control method and system for wastewater treatment

Publications (2)

Publication Number Publication Date
CN108408855A true CN108408855A (en) 2018-08-17
CN108408855B CN108408855B (en) 2019-02-12

Family

ID=63135066

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810315713.5A Active CN108408855B (en) 2018-04-10 2018-04-10 A kind of online Adding medicine control method and system for wastewater treatment

Country Status (1)

Country Link
CN (1) CN108408855B (en)

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109914120A (en) * 2019-02-01 2019-06-21 东华大学 A kind of design method of dye formulation
CN110223288A (en) * 2019-06-17 2019-09-10 华东交通大学 A kind of Rare-Earth Extraction Process multicomponent content prediction method and system
CN110399668A (en) * 2019-07-17 2019-11-01 西安工业大学 A method of rapidly and accurately solving calutron output characteristics
CN110980898A (en) * 2019-10-11 2020-04-10 浙江华晨环保有限公司 Medicament adding system of water purifying equipment
CN110981021A (en) * 2019-12-23 2020-04-10 中新国际联合研究院 Intelligent dosing system and method for advanced wastewater oxidation treatment based on fuzzy BP neural network
CN111631156A (en) * 2020-06-16 2020-09-08 江苏华丽智能科技股份有限公司 Anti-accumulation dosing method and device
CN113705098A (en) * 2021-08-30 2021-11-26 国网江苏省电力有限公司营销服务中心 Air duct heater modeling method based on PCA and GA-BP network
CN114354455A (en) * 2022-01-17 2022-04-15 北京石油化工学院 Method for online measurement of particle size distribution of battery slurry
CN116360366A (en) * 2023-03-24 2023-06-30 淮阴工学院 Sewage treatment process optimization control method
CN117238389A (en) * 2023-11-14 2023-12-15 江苏海峡环保科技发展有限公司 Fluorine-containing wastewater treatment system and method based on intelligent dosing
CN117275615A (en) * 2023-10-31 2023-12-22 源康(东阿)健康科技有限公司 Intelligent treatment method and system for gelatin production wastewater

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104732276A (en) * 2015-03-18 2015-06-24 国家电网公司 On-line diagnosing method for faults of metering production facility
CN105676814A (en) * 2016-01-11 2016-06-15 大唐环境产业集团股份有限公司 SFLA-SVM-based digital water island online agent adding control method
CN106503802A (en) * 2016-10-20 2017-03-15 上海电机学院 A kind of method of utilization genetic algorithm optimization BP neural network system

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104732276A (en) * 2015-03-18 2015-06-24 国家电网公司 On-line diagnosing method for faults of metering production facility
CN105676814A (en) * 2016-01-11 2016-06-15 大唐环境产业集团股份有限公司 SFLA-SVM-based digital water island online agent adding control method
CN106503802A (en) * 2016-10-20 2017-03-15 上海电机学院 A kind of method of utilization genetic algorithm optimization BP neural network system

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109914120A (en) * 2019-02-01 2019-06-21 东华大学 A kind of design method of dye formulation
CN110223288A (en) * 2019-06-17 2019-09-10 华东交通大学 A kind of Rare-Earth Extraction Process multicomponent content prediction method and system
CN110399668A (en) * 2019-07-17 2019-11-01 西安工业大学 A method of rapidly and accurately solving calutron output characteristics
CN110980898A (en) * 2019-10-11 2020-04-10 浙江华晨环保有限公司 Medicament adding system of water purifying equipment
CN110981021B (en) * 2019-12-23 2022-08-05 中新国际联合研究院 Intelligent dosing system and method for advanced wastewater oxidation treatment based on fuzzy BP neural network
CN110981021A (en) * 2019-12-23 2020-04-10 中新国际联合研究院 Intelligent dosing system and method for advanced wastewater oxidation treatment based on fuzzy BP neural network
CN111631156A (en) * 2020-06-16 2020-09-08 江苏华丽智能科技股份有限公司 Anti-accumulation dosing method and device
CN111631156B (en) * 2020-06-16 2021-11-09 江苏华丽智能科技股份有限公司 Anti-accumulation dosing method and device
CN113705098A (en) * 2021-08-30 2021-11-26 国网江苏省电力有限公司营销服务中心 Air duct heater modeling method based on PCA and GA-BP network
CN114354455A (en) * 2022-01-17 2022-04-15 北京石油化工学院 Method for online measurement of particle size distribution of battery slurry
CN114354455B (en) * 2022-01-17 2023-12-08 北京石油化工学院 Method for online measurement of size distribution of battery slurry
CN116360366A (en) * 2023-03-24 2023-06-30 淮阴工学院 Sewage treatment process optimization control method
CN116360366B (en) * 2023-03-24 2023-12-01 淮阴工学院 Sewage treatment process optimization control method
CN117275615A (en) * 2023-10-31 2023-12-22 源康(东阿)健康科技有限公司 Intelligent treatment method and system for gelatin production wastewater
CN117275615B (en) * 2023-10-31 2024-04-09 源康(东阿)健康科技有限公司 Intelligent treatment method and system for gelatin production wastewater
CN117238389A (en) * 2023-11-14 2023-12-15 江苏海峡环保科技发展有限公司 Fluorine-containing wastewater treatment system and method based on intelligent dosing
CN117238389B (en) * 2023-11-14 2024-01-30 江苏海峡环保科技发展有限公司 Fluorine-containing wastewater treatment system and method based on intelligent dosing

Also Published As

Publication number Publication date
CN108408855B (en) 2019-02-12

Similar Documents

Publication Publication Date Title
CN108408855B (en) A kind of online Adding medicine control method and system for wastewater treatment
Gong et al. Genetic learning particle swarm optimization
Yuan et al. Prediction of concrete compressive strength: Research on hybrid models genetic based algorithms and ANFIS
Floreano et al. Neuroevolution: from architectures to learning
Recknagel ANNA–Artificial Neural Network model for predicting species abundance and succession of blue-green algae
CN105676814B (en) The online Adding medicine control method in digitlization water island based on SFLA SVM
CN108460461A (en) Mars earth shear parameters prediction technique based on GA-BP neural networks
Sarker et al. Evolutionary computation: A gentle introduction
AU2022343574B2 (en) Industrial process soft-measurement method based on federated incremental stochastic configuration network
CN108830035A (en) A kind of novel water process coagulant dosage control method, computer, computer program
CN106991493A (en) Sewage disposal water outlet parameter prediction method based on Grey production fuction
Ferreira et al. Comparing recurrent networks for time-series forecasting
CN108460462A (en) A kind of Interval neural networks learning method based on interval parameter optimization
CN106651001A (en) Needle mushroom yield prediction method based on improved neural network and implementation system
Ji et al. Accuracy versus simplification in an approximate logic neural model
CN107511823A (en) The method of robot manipulating task track optimizing analysis
Chen et al. Dynamic parameter optimization of evolutionary computation for on-line prediction of time series with changing dynamics
Vafamand et al. Multi-objective NSBGA-II control of HIV therapy with monthly output measurement
Castillo et al. Hybrid soft computing models for systems modeling and control
Wróbel et al. Evolving networks processing signals with a mixed paradigm, inspired by gene regulatory networks and spiking neurons
CN111844049B (en) Dexterous hand grabbing control method and device and terminal equipment
Rakshit et al. A recurrent fuzzy neural model of a gene regulatory network for knowledge extraction using invasive weed and artificial bee colony optimization algorithm
Xing et al. Hierarchical genetic algorithm based RBF neural networks and application for modelling of the automatic depth control electrohydraulic system
Kasabov et al. Computational neurogenetic modelling: gene networks within neural networks
Bastos-Filho et al. Routing algorithm based on Swarm Intelligence and Hopfield Neural Network applied to communication networks

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant