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

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

Online dosing control method and system for wastewater treatment

Technical Field

The invention relates to the technical field of thermal power plant wastewater treatment, in particular to an online dosing control method and system for wastewater treatment.

Background

The waste water treatment process of the thermal power plant is a multivariable nonlinear dynamic system with large hysteresis, dynamics and serious interference, and is a complex industrial process which is difficult to control. The realization of the automation of the wastewater treatment of the thermal power plant is a necessary condition for realizing modern treatment and modern management, and is a necessary means for improving the wastewater treatment effect and reducing the cost. In the waste water treatment process of a thermal power plant, processes such as coagulation, flocculation and the like are indispensable important parts, the process is a complex physical and chemical reaction process, the control precision requirement on the types and dosage of added drugs is high, the adding amount of the drugs is judged by operators through the effluent quality in the prior art, a large amount of time and labor are wasted by repeatedly adjusting the adding amount of the drugs in order to meet the effluent quality requirement, the behavior of 'after-knowing feeling' is achieved, and the hysteresis is obvious; and once the addition amount of the medicament is determined, the medicament basically belongs to a long-term constant state, so that the invisible waste of the medicament is caused, and the water outlet index cannot meet the real-time requirement of a user due to the time-varying characteristic of the quality of the incoming water.

Therefore, the chemical adding process of the power plant water treatment shows randomness, hysteresis, nonlinearity and the defects of the traditional treatment method, and the realization of on-line chemical adding control is a problem to be solved urgently in the waste water treatment industry of the thermal power plant.

Disclosure of Invention

The invention aims to provide an online dosing control method and system for wastewater treatment, which overcome the defect that the dosing agent amount is difficult to accurately determine by a traditional dosing mode in the wastewater treatment process and realize a rapid and accurate online dosing control process.

In order to achieve the purpose, the invention provides the following scheme:

an on-line dosing control method for wastewater treatment, comprising:

acquiring the incoming water monitoring index data at the current moment;

inputting the incoming water monitoring index data at the current moment into an online dosing control model to obtain the optimal dosing amount at the current moment; the input of the online dosing control model is the current incoming water monitoring index data; the output of the online dosing control model is the optimal dosing amount at the current moment; the online dosing control model is established according to a principal component analysis algorithm, a genetic algorithm and a neural network model algorithm; the method for establishing the online dosing control model specifically comprises the following steps:

obtaining a training sample; the training samples comprise a plurality of sample pairs; each sample pair comprises a plurality of inputs, one output; the input is the incoming water monitoring index data according with the effluent quality; the output is the optimal dosage corresponding to the incoming water monitoring index data;

processing the sample pairs in the training samples by adopting a principal component analysis algorithm to obtain a principal component vector matrix and the number of principal component components;

establishing a BP neural network model according to the principal component vector matrix and the number of the principal component components; the BP neural network model is a multi-input single-output three-layer model; the input of the BP neural network model is the principal component vector matrix; the output of the BP neural network model is the optimal dosage; the number of input neurons of the BP neural network model is the number of the principal component components;

optimizing the connection weight and the threshold of the BP neural network model by adopting a genetic algorithm to obtain an optimal connection weight and an optimal threshold;

updating the BP neural network model according to the optimal connection weight and the optimal threshold; and the updated BP neural network model is the online dosing control model.

Optionally, the sample pairs are a time series set (X, Y) composed of an input time series signal and an output time series signal;

the input time series signal is an input value of the pair of samples; the input time series signal is X ═ X_ij]_nⅹpI is 1,2, … … n, j is 1,2, … … p, n is the number of the samples of the water inflow monitoring index data in the training sample, and p is the number of the water inflow monitoring index data; the incoming water monitoring index data comprise turbidity values, pH values, ammonia nitrogen contents and COD values;

the output time series signal is an output value of the sample pair; the output time series signal is Y ═ Y_i]_nⅹ1And i is 1,2, … … n, and n is the number of samples of the medicine adding amount data in the training samples.

Optionally, the processing the sample pairs in the training samples by using a principal component analysis algorithm to obtain a principal component vector matrix and the number of principal component components specifically includes:

calculating a correlation coefficient matrix R of the incoming water monitoring index data in the training sample;

calculating a characteristic value according to a characteristic equation of lambda I-R0, wherein the characteristic value is lambda_jJ 1,2, p, and ordering the eigenvalues in order of magnitude, λ₁≥λ₂≥…≥λ_pWherein, I represents an identity matrix;

calculating each of the characteristic values lambda_j1,2, p, and the corresponding feature vector e_jJ is 1,2,. cndot, p; wherein, | | e_j||＝1；

Calculating the cumulative contribution rate according to the characteristic values, selecting the characteristic values with the cumulative contribution rate of 85-95%, and determining the number of the characteristic values with the cumulative contribution rate of 85-95% as the number of the principal component components; the calculation formula of the accumulated contribution rate is as follows:

calculating principal component loads according to the characteristic values and the characteristic vectors; the calculation formula of the principal component load is as follows:

determining a principal component vector matrix according to the principal component load; the principal component vector matrix is: z ═ Z_it]_nⅹm；

Optionally, the correlation coefficient matrixWherein r is_abFor x in the training sample_aAnd x_bOf correlation coefficient r_ab＝r_ba， Is a variable x_aThe average value of the samples of (a),is a variable x_bThe sample mean of (1).

Optionally, the establishing a BP neural network model according to the principal component vector matrix and the number of the principal component components specifically includes:

establishing a BP neural network structure; the number of the principal component components is the number of input neurons of the BP neural network structure, and the principal component vector matrix is the input quantity of the BP neural network structure; the output target of the BP neural network structure is the optimal dosage; the BP neural network structure is a multi-input single-output three-layer model;

initializing a connection weight and a threshold of the BP neural network structure, setting a sample counter and a learning frequency counter to be 1, and determining a minimum error and an iteration frequency; the connection weight comprises a weight from a hidden layer to an input layer and a weight from an output layer to the hidden layer; the threshold comprises a threshold of each neuron node in the hidden layer and a threshold of each neuron node in the output layer;

inputting the c-th sample pair in the training samples into the BP neural network structure, and calculating the input and output of each neuron node in a hidden layer and the input and output of each neuron node in an output layer;

calculating correction errors of each neuron node in the output layer and correction errors of each neuron node in the hidden layer according to the input and the output of each neuron node in the hidden layer and the input and the output of each neuron node in the output layer, and determining errors of the c-th sample pair;

adjusting the connection weight and the threshold according to the error of the c-th sample pair;

judging whether all sample pairs in the training samples are trained;

if not, returning to the step of inputting the c-th sample pair in the training samples into the BP neural network structure, and calculating the input and output of each neuron node in a hidden layer and the input and output of each neuron node in an output layer;

if so, updating the learning times, calculating a global error, and judging whether the global error is smaller than the set minimum error or whether the learning times reach the set iteration times;

if so, establishing a BP neural network model according to the adjusted connection weight and the threshold;

if not, returning to the step of inputting the c-th sample pair in the training samples into the BP neural network structure, and calculating the input and output of each neuron node in the hidden layer and the input and output of each neuron node in the output layer.

Optionally, the optimizing the connection weight and the threshold of the BP neural network model by using a genetic algorithm to obtain an optimal connection weight and an optimal threshold specifically includes:

combining the connection weight value and the threshold value in the BP neural network model as a chromosome to form an individual of a genetic algorithm, and determining the number S of the individuals of an initial population and the number N of genetic iteration;

randomly generating an initialized population, carrying out binary coding on the initialized population, determining the initial population consisting of S individuals, and setting the evolution frequency to be 1;

determining a fitness function of the genetic algorithm; the fitness function is an error function of the BP neural network model;

calculating a fitness function value of each individual in the initial population according to the fitness function;

judging whether the current evolution times reach the set genetic iteration times or not;

if yes, outputting the optimal individual; the optimal individual is the individual with the maximum fitness function value in the initial population;

and if not, increasing the evolution times by 1, performing selection, crossing and mutation genetic operations on the individuals with the maximum fitness function value in the initial population, updating the initial population, and returning to the step to calculate the fitness function value of each individual in the initial population according to the fitness function.

The invention also provides an online dosing control system for wastewater treatment, comprising:

the water inflow monitoring index data acquisition module is used for acquiring water inflow monitoring index data at the current moment;

the optimal dosing quantity acquisition module is used for inputting the incoming water monitoring index data at the current moment into the online dosing control model to obtain the optimal dosing quantity at the current moment; the input of the online dosing control model is the current incoming water monitoring index data; the output of the online dosing control model is the optimal dosing amount at the current moment; the online dosing control model is established according to a principal component analysis algorithm, a genetic algorithm and a neural network model algorithm; the subsystem for establishing the online dosing control model specifically comprises:

the training sample acquisition module is used for acquiring a training sample; the training samples comprise a plurality of sample pairs; each sample pair comprises a plurality of inputs, one output; the input is the incoming water monitoring index data according with the effluent quality; the output is the optimal dosage corresponding to the incoming water monitoring index data;

a principal component vector matrix and principal component number obtaining module, configured to process the sample pairs in the training samples by using a principal component analysis algorithm to obtain a principal component vector matrix and a number of principal component components;

the BP neural network model establishing module is used for establishing a BP neural network model according to the principal component vector matrix and the number of the principal component components; the BP neural network model is a multi-input single-output three-layer model; the input of the BP neural network model is the principal component vector matrix; the output of the BP neural network model is the optimal dosage; the number of input neurons of the BP neural network model is the number of the principal component components;

an optimal connection weight and optimal threshold acquisition module, configured to optimize the connection weight and the threshold of the BP neural network model by using a genetic algorithm to obtain an optimal connection weight and an optimal threshold;

the BP neural network model updating module is used for updating the BP neural network model according to the optimal connection weight and the optimal threshold; and the updated BP neural network model is the online dosing control model.

Optionally, the module for acquiring the number of the principal component vector matrix and the number of the principal component components specifically includes:

a correlation coefficient matrix calculation unit, configured to calculate a correlation coefficient matrix R of the incoming water monitoring index data in the training sample;

a feature value calculation unit for calculating a feature value according to a feature equation | λ I-R | ═ 0, the feature value λ_jJ 1,2, p, and ordering the eigenvalues in order of magnitude, λ₁≥λ₂≥…≥λ_pWherein, I represents an identity matrix;

a feature vector calculation unit for calculating each of the feature values λ_j1,2, p, and the corresponding feature vector e_jJ is 1,2,. cndot, p; wherein, | | e_j||＝1；

A principal component load calculation unit for calculating a principal component load based on the eigenvalue and the eigenvector; the calculation formula of the principal component load is as follows:

the principal component vector matrix determining unit is used for determining a principal component vector matrix according to the principal component load; the principal component vector matrix is: z ═ Z_it]_nⅹm；

Optionally, the BP neural network model establishing module specifically includes:

the BP neural network structure establishing unit is used for establishing a BP neural network structure; the number of the principal component components is the number of input neurons of the BP neural network structure, and the principal component vector matrix is the input quantity of the BP neural network structure; the output target of the BP neural network structure is the optimal dosage; the BP neural network structure is a multi-input single-output three-layer model;

the first initialization unit is used for initializing the connection weight and the threshold of the BP neural network structure, setting a sample counter and a learning frequency counter to be 1, and determining the minimum error and the iteration frequency; the connection weight comprises a weight from a hidden layer to an input layer and a weight from an output layer to the hidden layer; the threshold comprises a threshold of each neuron node in the hidden layer and a threshold of each neuron node in the output layer;

the input and output calculation unit is used for inputting the c-th sample pair in the training samples into the BP neural network structure, and calculating the input and output of each neuron node in a hidden layer and the input and output of each neuron node in an output layer;

a sample pair error determination unit, configured to calculate a correction error of each neuron node in the output layer and a correction error of each neuron node in the hidden layer according to the input and the output of each neuron node in the hidden layer and the input and the output of each neuron node in the output layer, and determine an error of the c-th sample pair;

a connection weight and threshold adjusting unit, configured to adjust the connection weight and the threshold according to the error of the c-th sample pair;

the first judging unit is used for judging whether all sample pairs in the training samples are trained or not;

a first returning unit, configured to, when there is a sample pair in the training samples that is not trained, return to the step of inputting the c-th sample pair in the training samples to the BP neural network structure, and calculate input and output of each neuron node in a hidden layer and input and output of each neuron node in an output layer;

a second judging unit, configured to update the number of learning times, calculate a global error, and judge whether the global error is smaller than the set minimum error or whether the number of learning times reaches the set iteration times when all sample pairs in the training samples are trained;

a BP neural network model establishing unit, configured to establish a BP neural network model according to the adjusted connection weight and the threshold when the global error is smaller than the set minimum error or the learning frequency reaches the set iteration frequency;

and a second returning unit, configured to, when the global error is smaller than the set minimum error and the learning frequency reaches the set iteration frequency, return to the step of inputting the c-th sample pair in the training samples to the BP neural network structure, and calculate input and output of each neuron node in the hidden layer and input and output of each neuron node in the output layer.

Optionally, the module for obtaining the optimal connection weight and the optimal threshold specifically includes:

the second initialization unit is used for combining the connection weight value and the threshold value in the BP neural network model to form an individual of a genetic algorithm, and determining the number S of the individuals of an initial population and the number N of genetic iterations;

the initial population determining unit is used for randomly generating an initial population, carrying out binary coding on the initialized population, determining the initial population consisting of S individuals, and setting the number of evolution times to be 1;

a fitness function determining unit for determining a fitness function of the genetic algorithm; the fitness function is an error function of the BP neural network model;

a fitness function value calculating unit, configured to calculate a fitness function value of each individual in the initial population according to the fitness function;

the third judging unit is used for judging whether the current evolution times reach the set genetic iteration times;

the optimal individual output unit is used for outputting an optimal individual when the current evolution times reach the set genetic iteration times; the optimal individual is the individual with the maximum fitness function value in the initial population;

and the initial population updating unit is used for increasing 1 for the evolution times when the current evolution times does not reach the set genetic iteration times, performing selection, crossing and variant genetic operations on the individuals with the maximum fitness function value in the initial population, updating the initial population, and returning to the step of calculating the fitness function value of each individual in the initial population according to the fitness function.

According to the specific embodiment provided by the invention, the invention discloses the following technical effects:

the invention provides an online dosing control method and system for wastewater treatment, wherein the online dosing control method comprises the following steps: acquiring the incoming water monitoring index data at the current moment, and inputting the incoming water monitoring index data at the current moment into an online dosing control model to obtain the optimal dosing amount at the current moment; the online dosing control model is established by adopting a principal component analysis algorithm, a genetic algorithm and a neural network model algorithm; the invention reduces the dimensionality of the training sample through the principal component analysis algorithm, simplifies the structure of the BP neural network model, improves the speed of the BP neural network model, and further improves the speed of the dosage calculation. According to the invention, the connection weight and the threshold of the BP neural network model are optimized through a genetic algorithm, and the obtained optimal connection weight and threshold are given to the BP neural network model, so that the prediction precision of the BP neural network model is improved and local optimization is not easy to occur. The invention adopts the online dosing control model to overcome the defect that the traditional dosing mode is difficult to accurately determine the dosing dosage in the wastewater treatment process, and realizes the rapid and accurate online dosing control process.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.

FIG. 1 is a schematic flow chart of an on-line dosing control method for wastewater treatment according to an embodiment of the present invention;

FIG. 2 is a schematic flow chart of a method for establishing an online dosing control model according to an embodiment of the present invention;

FIG. 3 is a schematic flow chart of a first-time online dosing control method according to an embodiment of the present invention;

FIG. 4 is a schematic structural diagram of an on-line dosing control system for wastewater treatment according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The invention aims to provide an online dosing control method and system for wastewater treatment, which overcome the defect that the dosing agent amount is difficult to accurately determine by a traditional dosing mode in the wastewater treatment process and realize a rapid and accurate online dosing control process.

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.

The BP neural network model is a multilayer network model based on a gradient descent method, is proposed by Rumelhart and McClelland in 1986, is an algorithm with strong parallel processing capability and self-learning capability, can effectively carry out infinite approximation on complex non-structural problems to solve the problems, and provides a new idea for effectively solving the complex non-structural problems. The key problem of the BP neural network model is the selection of the connection weight and the threshold between the input layer and the hidden layer and the connection weight and the threshold between the output layer and the hidden layer in the network structure, and the selection of the value has important influence on the prediction result.

Genetic Algorithm (GA) is a simulated evolution Algorithm developed in 1969 by Holland's teaching of Michigan university, USA, and summarized by DeJong, Goldberg, etc. It is a randomized and directed search algorithm based on the Darwin theory of evolution and Mendelian theory of genetics, and is based on natural selection and genetic mechanism. The genetic algorithm is a global search algorithm formed by simulating the evolution process of the living of organisms in the natural environment. The genetic algorithm is applied in the process of constructing a model and solving complex problems, such as function optimization, combination optimization, machine learning, pattern recognition and the like, due to the characteristics of strong adaptability, global optimization, simple algorithm, universality and the like.

A Principal Component Analysis (PCA) algorithm is introduced by k.pearson for non-random variables, and then generalized by hotelling in 1933 to the random vector situation. The main idea is to perform dimensionality reduction processing on a high-dimensional data space under the principle of trying to guarantee the minimum data loss. Introducing a PCA algorithm into the BP neural network model, mainly reducing the dimension of a training sample, and simplifying the structure of the BP neural network model.

The invention provides an online dosing control method and system for wastewater treatment, which solve the problems of subjectivity and hysteresis of the existing thermal power plant water treatment dosing system, combine with the BP neural network model to solve the non-linear problem, GA global parameter optimization and the characteristic that the PCA dimension reduction reduces the BP network structure, select the incoming water monitoring index and dosing data with the effluent quality meeting the requirements as training samples, reduce the sample dimension by the PCA, perform the parameter optimization of BP network connection weight and threshold by the GA algorithm, perform the training of the samples by the BP model, finally obtain the online dosing control model based on the principal component analysis-genetic algorithm-BP neural network model algorithm, and realize the intelligent control of the applied dosage by the model.

Fig. 1 is a schematic flow chart of an online dosing control method for wastewater treatment according to an embodiment of the present invention, and as shown in fig. 1, the online dosing control method provided by the embodiment of the present invention specifically includes the following steps:

step 101: and acquiring the incoming water monitoring index data at the current moment. The incoming water monitoring index data mainly comprise turbidity values, PH values, ammonia nitrogen contents and COD values.

Step 102: inputting the incoming water monitoring index data at the current moment into an online dosing control model to obtain the optimal dosing amount at the current moment; the input of the online dosing control model is the current incoming water monitoring index data; the output of the online dosing control model is the optimal dosing amount at the current moment; the online dosing control model is established according to a principal component analysis algorithm, a genetic algorithm and a neural network model algorithm.

Fig. 2 is a schematic flow chart of a method for establishing an online dosing control model according to an embodiment of the present invention, and as shown in fig. 2, the method for establishing an online dosing control model according to an embodiment of the present invention specifically includes the following steps:

step 201: obtaining a training sample; the training samples comprise a plurality of sample pairs; each sample pair comprises a plurality of inputs, one output; the input is the incoming water monitoring index data according with the effluent quality; and the output is the optimal dosage corresponding to the incoming water monitoring index data.

Step 202: and processing the sample pairs in the training samples by adopting a principal component analysis algorithm to obtain a principal component vector matrix and the number of principal component components.

Step 203: establishing a BP neural network model according to the principal component vector matrix and the number of the principal component components; the BP neural network model is a multi-input single-output three-layer model; the input of the BP neural network model is the principal component vector matrix; the output of the BP neural network model is the optimal dosage; and the number of input neurons of the BP neural network model is the number of the principal component components.

Step 204: and optimizing the connection weight and the threshold of the BP neural network model by adopting a genetic algorithm to obtain an optimal connection weight and an optimal threshold.

Step 205: updating the BP neural network model according to the optimal connection weight and the optimal threshold; and the updated BP neural network model is the online dosing control model.

Step 201 specifically includes:

incoming water monitoring in training samplesThe index data mainly comprises turbidity value, PH value, ammonia nitrogen content, COD value, and time sequence signal X ═ X_ij]_nⅹpAnd the input is an online dosing control model, wherein i is 1,2, … … n, j is 1,2, … … p, n is the number of samples of the water inflow monitoring index data in the training samples, and p is the number of the water inflow monitoring index data.

The type of medicine added in the training sample is mainly coagulant, and the time sequence signal Y of the medicine adding amount is [ Y ═ Y%_i]_nⅹ1And as the actual output of the online dosing control model, wherein i is 1,2, … … n, and n is the number of samples of the dosing quantity data in the training samples. The type of the additive is not limited to coagulant, and other agents such as sodium hydroxide, bactericide, coagulant aid and the like can be calculated according to the method of the invention.

Input time series signal x_ijAnd outputting the time-series signal y_iThe formed time sequence set (X, Y) is used as a training sample of an online dosing control model, namely the sample pair is an input time sequence signal X_ijAnd outputting the time-series signal y_iThe time series set of compositions (X, Y).

Step 202 specifically includes:

step 2021: calculating a correlation coefficient matrix R of the incoming water monitoring index data X in the training sample;

wherein r is_abFor x in the training sample_aAnd x_bOf correlation coefficient r_ab＝r_baThe calculation formula is as follows:

wherein,is a variable x_aThe average value of the samples of (a),is a variable x_bThe sample mean of (1).

Step 2022: calculating a characteristic value and a characteristic vector; solving the eigenvalue lambda according to the eigen equation lambda I-R0_jJ 1,2, p, and λ is ordered from large to small₁≥λ₂≥…≥λ_p. Where I denotes an identity matrix. Respectively calculating each characteristic value lambda_j1,2, p, and the corresponding feature vector e_jJ 1,2, p, requiring | | | e_j1 | | |, i.eWherein e_jiIs a vector e_jThe ith component of (a).

Step 2023: calculating the accumulated contribution rate;

the calculation formula of the accumulated contribution rate is as follows:

taking m characteristic values with the accumulated contribution rate of 85-95 percent to obtain lambda₁，λ₂，…，λ_mThe corresponding m principal component components.

Step 2024: calculating the principal component load;

the principal component load calculation formula is as follows:

step 2025: calculating each principal component vector matrix according to the principal component loads; the principal component vector matrix is Z ═ Z_it]_nⅹm：

Step 203 specifically includes:

step 2031: determining a BP neural network structure: according to the number of m principal components obtained by calculation of a principal component analysis algorithm, m BP neural network input neurons are determined, the output target is to control the dosage of the coagulant and is single output, and therefore a multi-input single-output (MISO) three-layer model is adopted.

Step 2032: obtaining a principal component vector matrix Z ═ Z by the principal component analysis algorithm_it]_nⅹmAs an input vector, Y ═ Y_i]_nⅹ1As an output vector, H ═ H₁,h₂,……,h_l]^TIs the output vector of the hidden layer,as the desired output vector, W ═ W_dt]_lⅹmAs a weight from the hidden layer to the input layer, θ ═ θ_d]_l×1For each neuron node of the hidden layer, V ═ w_di]_1ⅹlAnd q is the threshold value of the neuron node of the output layer as the weight value from the output layer to the hidden layer.

Step 2033: initializing a network connection weight and a threshold (the weight from the hidden layer to the input layer, the threshold of each neuron node of the hidden layer, the weight from the output layer to the hidden layer and the threshold of the neuron node of the output layer), setting 1 for a sample counter and a learning time counter, and setting a minimum error and an iteration time.

Step 2034: inputting the c-th sample pair, and calculating the input and the output of each neuron node of the hidden layer and the input and the output of each neuron node of the output layer; input z of the sample pair_c＝[z_c1,z_c2,……,z_cm]And output y_c。

Step 2035: calculating the correction error of each neuron node of the output layer,Correcting errors of each neuron node of the hidden layer to obtain the error of the c-th sample pair of the BP neural network model(6)。

Step 2036: according to the error of the c-th sample pair, the connection weight between the hidden layer and the output layer, the threshold of each neuron node of the output layer, the connection weight between the input layer and the hidden layer and the threshold of each neuron node of the hidden layer are adjusted to obtain a new weight W, V and a new threshold theta ═ theta [ [ theta ] ]_d]_l×1、q。

Step 2037: it is determined whether all sample pairs are trained, if so, step 2038 is performed, otherwise, step 2034 is performed.

Step 2038: updating the learning times and calculating the global errorAnd judging whether the E is smaller than a set minimum error or whether the learning frequency reaches a set iteration frequency, if so, finishing establishing the BP neural network model, and otherwise, executing the step 2034.

Step 204 specifically includes:

step 2041: and combining the connection weight value and the threshold value in the BP neural network model to form a chromosome, so as to form an individual of the genetic algorithm. Determining the initial population size S, the genetic iteration times N and the crossing rate P_cThe rate of variation P_m。

Step 2042: and randomly generating an initialization population, carrying out binary coding on the population to generate S groups of initial populations, and setting the number of evolutionary times to be 1.

And 2043, obtaining the optimal solution of the BP neural network connection weight and the threshold value by adopting the following loop steps. The circulation steps are as follows:

step S1: setting fitness function of genetic algorithm through error function of BP neural network modelC is a constant, and C is a constant,is a time series signal z_itAnd a time series signal y_iAnd in the formed ith group of data states, calculating an expected output result by the BP neural network model, wherein i is 1,2, … …, n.

Step S2: and calculating a fitness function value of each individual in the initial population according to the fitness function, and evaluating each chromosome according to the fitness function value.

Step S3: and judging whether the current evolution times reach the set genetic iteration times, if so, outputting the optimal individual, and ending the step 2043. Otherwise, increasing the evolution times by 1, storing the individual corresponding to the highest fitness function value, performing selection, crossing and mutation genetic operations, updating the initial population, and returning to the step S1.

Further, the genetic manipulation in step S3 includes the specific steps of:

a: the selection operation is to select good individuals in the current group, a roulette selection strategy is generally adopted, the idea is that the probability of each individual being selected is proportional to the magnitude of the fitness function value, and the probability of the g-th individual being inherited to the next generation is:wherein N is the population size, F_gThe fitness function value of the individual g.

B: crossing, i.e. the probability P of crossing two chromosomes paired with each other_cExchanging parts of their genes with each other in a manner such that two new individuals are formed.

C: variance refers to the general rule P of variation_mCertain gene values in the individual code strings are replaced with other gene values to form a new individual.

Fig. 3 is a schematic flow chart of a first-time online dosing control method according to an embodiment of the present invention, as shown in fig. 3, and fig. 3 is a detailed flow description of the flow chart shown in fig. 1 and the flow chart shown in fig. 2.

According to the invention, the parameter value of the incoming water monitoring index monitored in real time is input into the established online dosing control model, so that the dosing amount of the dosing agent at the current moment can be obtained, the online control of the dosing process is realized, the optimal dosing amount can be updated in real time according to the current water quality condition, the adverse effect of the time-varying characteristic of the incoming water quality on the operation process is overcome, the waste of the dosing agent is reduced, the labor and operation cost are reduced, and the water quality requirement of the outgoing water is met in real time.

In order to realize the aim, the invention also provides an online dosing control system for wastewater treatment.

Fig. 4 is a schematic structural diagram of an online dosing control system for wastewater treatment according to an embodiment of the present invention, and as shown in fig. 4, the online dosing control system includes:

the incoming water monitoring index data obtaining module 100 is configured to obtain incoming water monitoring index data at a current time.

The optimal dosing amount acquisition module 200 is used for inputting the current incoming water monitoring index data into an online dosing control model to obtain the current optimal dosing amount; the input of the online dosing control model is the current incoming water monitoring index data; the output of the online dosing control model is the optimal dosing amount at the current moment; the online dosing control model is established according to a principal component analysis algorithm, a genetic algorithm and a neural network model algorithm.

The subsystem for establishing the online dosing control model specifically comprises:

a training sample obtaining module 300, configured to obtain a training sample; the training samples comprise a plurality of sample pairs; each sample pair comprises a plurality of inputs, one output; the input is the incoming water monitoring index data according with the effluent quality; and the output is the optimal dosage corresponding to the incoming water monitoring index data.

A principal component vector matrix and principal component number obtaining module 400, configured to process the sample pairs in the training samples by using a principal component analysis algorithm, so as to obtain a principal component vector matrix and a number of principal component components.

A BP neural network model establishing module 500, configured to establish a BP neural network model according to the principal component vector matrix and the number of the principal component components; the BP neural network model is a multi-input single-output three-layer model; the input of the BP neural network model is the principal component vector matrix; the output of the BP neural network model is the optimal dosage; and the number of input neurons of the BP neural network model is the number of the principal component components.

An optimal connection weight and optimal threshold obtaining module 600, configured to optimize the connection weight and the threshold of the BP neural network model by using a genetic algorithm, so as to obtain an optimal connection weight and an optimal threshold.

A BP neural network model updating module 700, configured to update the BP neural network model according to the optimal connection weight and the optimal threshold; and the updated BP neural network model is the online dosing control model.

The module 400 for acquiring the number of the principal component vector matrix and the number of the principal component components specifically includes:

and the correlation coefficient matrix calculation unit is used for calculating a correlation coefficient matrix R of the incoming water monitoring index data in the training sample.

A feature value calculation unit for calculating a feature value according to a feature equation | λ I-R | ═ 0, the feature value λ_jJ 1,2, p, and ordering the eigenvalues in order of magnitude, λ₁≥λ₂≥…≥λ_pWherein I represents an identity matrix.

A feature vector calculation unit for calculating each of the feature values λ_j1,2, p, and the corresponding feature vector e_jJ is 1,2,. cndot, p; wherein, | | e_j||＝1。

An accumulated contribution rate calculating unit, configured to calculate an accumulated contribution rate according to the feature value, select a feature value with the accumulated contribution rate reaching 85% to 95%, and determine the number of the feature values with the accumulated contribution rate reaching 85% to 95% as the number of the principal component components; the calculation formula of the accumulated contribution rate is as follows:

a principal component load calculation unit for calculating a principal component load based on the eigenvalue and the eigenvector; the calculation formula of the principal component load is as follows:

the principal component vector matrix determining unit is used for determining a principal component vector matrix according to the principal component load; the principal component vector matrix is: z ═ Z_it]_nⅹm；

The BP neural network model building module 500 specifically includes:

the BP neural network structure establishing unit is used for establishing a BP neural network structure; the number of the principal component components is the number of input neurons of the BP neural network structure, and the principal component vector matrix is the input quantity of the BP neural network structure; the output target of the BP neural network structure is the dosing amount; the BP neural network structure is a multi-input single-output three-layer model.

The first initialization unit is used for initializing the connection weight and the threshold of the BP neural network structure, setting a sample counter and a learning frequency counter to be 1, and determining the minimum error and the iteration frequency; the connection weight comprises a weight from a hidden layer to an input layer and a weight from an output layer to the hidden layer; the threshold value comprises a threshold value of each neuron node in the hidden layer and a threshold value of each neuron node in the output layer.

And the input and output calculation unit is used for inputting the c-th sample pair in the training samples into the BP neural network structure, and calculating the input and output of each neuron node in the hidden layer and the input and output of each neuron node in the output layer.

And the sample pair error determining unit is used for calculating the correction error of each neuron node in the output layer and the correction error of each neuron node in the hidden layer according to the input and the output of each neuron node in the hidden layer and the input and the output of each neuron node in the output layer, and determining the error of the c-th sample pair.

And the connection weight and threshold value adjusting unit is used for adjusting the connection weight and the threshold value according to the error of the c-th sample pair.

A first judging unit, configured to judge whether all sample pairs in the training samples are trained.

And the first returning unit is used for returning to the step of inputting the c-th sample pair in the training samples into the BP neural network structure and calculating the input and the output of each neuron node in the hidden layer and the input and the output of each neuron node in the output layer when the sample pair in the training samples is not trained.

A second judging unit, configured to update the learning times, calculate a global error, and judge whether the global error is smaller than the set minimum error or whether the learning times reaches the set iteration times when all the sample pairs in the training samples are trained.

And the BP neural network model establishing unit is used for establishing a BP neural network model according to the adjusted connection weight and the threshold when the global error is smaller than the set minimum error or the learning frequency reaches the set iteration frequency.

And a second returning unit, configured to, when the global error is smaller than the set minimum error and the learning frequency reaches the set iteration frequency, return to the step of inputting the c-th sample pair in the training samples to the BP neural network structure, and calculate input and output of each neuron node in the hidden layer and input and output of each neuron node in the output layer.

The optimal connection weight and optimal threshold obtaining module 600 specifically includes:

and the second initialization unit is used for combining the connection weight value and the threshold value in the BP neural network model as a chromosome to form an individual of the genetic algorithm and determining the number S of the individuals of the initial population and the number N of genetic iterations.

And the initial population determining unit is used for randomly generating an initialization population, carrying out binary coding on the initialized population, determining the initial population consisting of S individuals, and setting the number of evolution times to be 1.

A fitness function determining unit for determining a fitness function of the genetic algorithm; the fitness function is an error function of the BP neural network model.

And the fitness function value calculating unit is used for calculating the fitness function value of each individual in the initial population according to the fitness function.

And the third judging unit is used for judging whether the current evolution times reach the set genetic iteration times.

The optimal individual output unit is used for outputting an optimal individual when the current evolution times reach the set genetic iteration times; and the optimal individual is the individual with the maximum fitness function value in the initial population.

And the initial population updating unit is used for increasing 1 for the evolution times when the current evolution times does not reach the set genetic iteration times, performing selection, crossing and variant genetic operations on the individuals with the maximum fitness function value in the initial population, updating the initial population, and returning to the step of calculating the fitness function value of each individual in the initial population according to the fitness function.

Compared with the prior art, the invention has the beneficial effects that:

1. the dimensionality of the training sample is reduced through a principal component analysis algorithm, the structure of the BP neural network model is simplified, the speed of the BP neural network model is improved, and the speed of the medicine adding amount calculation is further improved.

2. And optimizing the connection weight and the threshold of the BP neural network model through a genetic algorithm, and giving the obtained optimal connection weight and the threshold to the BP neural network model, so that the prediction precision of the BP neural network model is improved, and the BP neural network model is not easy to fall into the local optimal problem.

3. The most important point is that the problem of a hysteresis medicine adding method which only can be repeatedly debugged by depending on effluent quality indexes in the traditional technology is solved, an operator can obtain the optimal value of the medicine adding amount at the current moment without real-time measurement according to the effluent quality indexes, errors caused by subjectivity of the operator are overcome, and adverse effects of time-varying characteristics of the incoming water quality on the operation process are also solved. The process not only reduces the waste of the medicament and the labor cost and the operation cost, but also can meet the water quality requirement of the effluent.

The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.

The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.

Claims (10)

1. An online dosing control method for wastewater treatment is characterized by comprising the following steps:

acquiring the incoming water monitoring index data at the current moment;

inputting the incoming water monitoring index data at the current moment into an online dosing control model to obtain the optimal dosing amount at the current moment; the input of the online dosing control model is the current incoming water monitoring index data; the output of the online dosing control model is the optimal dosing amount at the current moment; the online dosing control model is established according to a principal component analysis algorithm, a genetic algorithm and a neural network model algorithm; the method for establishing the online dosing control model specifically comprises the following steps:

obtaining a training sample; the training samples comprise a plurality of sample pairs; each sample pair comprises a plurality of inputs, one output; the input is the incoming water monitoring index data according with the effluent quality; the output is the optimal dosage corresponding to the incoming water monitoring index data;

processing the sample pairs in the training samples by adopting a principal component analysis algorithm to obtain a principal component vector matrix and the number of principal component components;

establishing a BP neural network model according to the principal component vector matrix and the number of the principal component components; the BP neural network model is a multi-input single-output three-layer model; the input of the BP neural network model is the principal component vector matrix; the output of the BP neural network model is the optimal dosage; the number of input neurons of the BP neural network model is the number of the principal component components;

optimizing the connection weight and the threshold of the BP neural network model by adopting a genetic algorithm to obtain an optimal connection weight and an optimal threshold;

updating the BP neural network model according to the optimal connection weight and the optimal threshold; and the updated BP neural network model is the online dosing control model.

2. The online dosing control method according to claim 1, wherein the sample pairs are time series sets (X, Y) of input and output time series signals;

the input time series signal is an input value of the pair of samples; the input time series signal is X ═ X_ij]_nⅹpI is 1,2, … … n, j is 1,2, … … p, n is the number of the samples of the water inflow monitoring index data in the training sample, and p is the number of the water inflow monitoring index data; the incoming water monitoring index data comprise turbidity value, PH value, ammonia nitrogen content,A COD value;

the output time series signal is an output value of the sample pair; the output time series signal is Y ═ Y_i]_nⅹ1And i is 1,2, … … n, and n is the number of samples of the medicine adding amount data in the training samples.

3. The on-line dosing control method according to claim 1, wherein the processing of the sample pairs in the training samples by using a principal component analysis algorithm to obtain a principal component vector matrix and the number of principal component components specifically comprises:

calculating a correlation coefficient matrix R of the incoming water monitoring index data in the training sample;

calculating a characteristic value according to a characteristic equation of lambda I-R0, wherein the characteristic value is lambda_jJ 1,2, p, and ordering the eigenvalues in order of magnitude, λ₁≥λ₂≥…≥λ_pWherein, I represents an identity matrix;

calculating each of the characteristic values lambda_j1,2, p, and the corresponding feature vector e_jJ is 1,2,. cndot, p; wherein, | | e_j||＝1；

Calculating the cumulative contribution rate according to the characteristic values, selecting the characteristic values with the cumulative contribution rate of 85-95%, and determining the number of the characteristic values with the cumulative contribution rate of 85-95% as the number of the principal component components; the calculation formula of the accumulated contribution rate is as follows:

calculating principal component loads according to the characteristic values and the characteristic vectors; the calculation formula of the principal component load is as follows:

determining a principal component vector matrix according to the principal component load; the principal component vector matrix is:

4. the on-line dosing control method of claim 3, wherein the correlation coefficient matrixWherein r is_abFor x in the training sample_aAnd x_bThe correlation coefficient of (a) is calculated, is a variable x_aThe average value of the samples of (a),is a variable x_bThe sample mean of (1).

5. The online dosing control method according to claim 1, wherein the establishing a BP neural network model according to the principal component vector matrix and the number of the principal component components specifically comprises:

establishing a BP neural network structure; the number of the principal component components is the number of input neurons of the BP neural network structure, and the principal component vector matrix is the input quantity of the BP neural network structure; the output target of the BP neural network structure is the optimal dosage; the BP neural network structure is a multi-input single-output three-layer model;

initializing a connection weight and a threshold of the BP neural network structure, setting a sample counter and a learning frequency counter to be 1, and determining a minimum error and an iteration frequency; the connection weight comprises a weight from a hidden layer to an input layer and a weight from an output layer to the hidden layer; the threshold comprises a threshold of each neuron node in the hidden layer and a threshold of each neuron node in the output layer;

inputting the c-th sample pair in the training samples into the BP neural network structure, and calculating the input and output of each neuron node in a hidden layer and the input and output of each neuron node in an output layer;

calculating correction errors of each neuron node in the output layer and correction errors of each neuron node in the hidden layer according to the input and the output of each neuron node in the hidden layer and the input and the output of each neuron node in the output layer, and determining errors of the c-th sample pair;

adjusting the connection weight and the threshold according to the error of the c-th sample pair;

judging whether all sample pairs in the training samples are trained;

if not, returning to the step of inputting the c-th sample pair in the training samples into the BP neural network structure, and calculating the input and output of each neuron node in a hidden layer and the input and output of each neuron node in an output layer;

if so, updating the learning times, calculating a global error, and judging whether the global error is smaller than the set minimum error or whether the learning times reach the set iteration times;

if so, establishing a BP neural network model according to the adjusted connection weight and the threshold;

if not, returning to the step of inputting the c-th sample pair in the training samples into the BP neural network structure, and calculating the input and output of each neuron node in the hidden layer and the input and output of each neuron node in the output layer.

6. The online dosing control method according to claim 1, wherein the optimizing the connection weight and the threshold of the BP neural network model by using a genetic algorithm to obtain an optimal connection weight and an optimal threshold specifically comprises:

combining the connection weight value and the threshold value in the BP neural network model as a chromosome to form an individual of a genetic algorithm, and determining the number S of the individuals of an initial population and the number N of genetic iteration;

randomly generating an initialized population, carrying out binary coding on the initialized population, determining the initial population consisting of S individuals, and setting the evolution frequency to be 1;

determining a fitness function of the genetic algorithm; the fitness function is an error function of the BP neural network model;

calculating a fitness function value of each individual in the initial population according to the fitness function;

judging whether the current evolution times reach the set genetic iteration times or not;

if yes, outputting the optimal individual; the optimal individual is the individual with the maximum fitness function value in the initial population;

and if not, increasing the evolution times by 1, performing selection, crossing and mutation genetic operations on the individuals with the maximum fitness function value in the initial population, updating the initial population, and returning to the step to calculate the fitness function value of each individual in the initial population according to the fitness function.

7. An on-line dosing control system for wastewater treatment, the on-line dosing control system comprising:

the water inflow monitoring index data acquisition module is used for acquiring water inflow monitoring index data at the current moment;

the optimal dosing quantity acquisition module is used for inputting the incoming water monitoring index data at the current moment into the online dosing control model to obtain the optimal dosing quantity at the current moment; the input of the online dosing control model is the current incoming water monitoring index data; the output of the online dosing control model is the optimal dosing amount at the current moment; the online dosing control model is established according to a principal component analysis algorithm, a genetic algorithm and a neural network model algorithm; the subsystem for establishing the online dosing control model specifically comprises:

the training sample acquisition module is used for acquiring a training sample; the training samples comprise a plurality of sample pairs; each sample pair comprises a plurality of inputs, one output; the input is the incoming water monitoring index data according with the effluent quality; the output is the optimal dosage corresponding to the incoming water monitoring index data;

a principal component vector matrix and principal component number obtaining module, configured to process the sample pairs in the training samples by using a principal component analysis algorithm to obtain a principal component vector matrix and a number of principal component components;

the BP neural network model establishing module is used for establishing a BP neural network model according to the principal component vector matrix and the number of the principal component components; the BP neural network model is a multi-input single-output three-layer model; the input of the BP neural network model is the principal component vector matrix; the output of the BP neural network model is the optimal dosage; the number of input neurons of the BP neural network model is the number of the principal component components;

an optimal connection weight and optimal threshold acquisition module, configured to optimize the connection weight and the threshold of the BP neural network model by using a genetic algorithm to obtain an optimal connection weight and an optimal threshold;

the BP neural network model updating module is used for updating the BP neural network model according to the optimal connection weight and the optimal threshold; and the updated BP neural network model is the online dosing control model.

8. The on-line dosing control system of claim 7, wherein the principal component vector matrix and the principal component number obtaining module specifically include:

a correlation coefficient matrix calculation unit, configured to calculate a correlation coefficient matrix R of the incoming water monitoring index data in the training sample;

a feature value calculation unit for calculating a feature value according to a feature equation | λ I-R | ═ 0, the feature value λ_jJ 1,2, p, and ordering the eigenvalues in order of magnitude, λ₁≥λ₂≥…≥λ_pWherein, I represents an identity matrix;

a feature vector calculation unit for calculating each of the feature values λ_j1,2, p, and the corresponding feature vector e_jJ is 1,2,. cndot, p; wherein, | | e_j||＝1；

An accumulated contribution rate calculating unit, configured to calculate an accumulated contribution rate according to the feature value, select a feature value with the accumulated contribution rate reaching 85% to 95%, and determine the number of the feature values with the accumulated contribution rate reaching 85% to 95% as the number of the principal component components; the calculation formula of the accumulated contribution rate is as follows:

a principal component load calculation unit for calculating a principal component load based on the eigenvalue and the eigenvector; the calculation formula of the principal component load is as follows:

the principal component vector matrix determining unit is used for determining a principal component vector matrix according to the principal component load; the principal component vector matrix is:

9. the online dosing control system according to claim 7, wherein the BP neural network model building module specifically comprises:

the BP neural network structure establishing unit is used for establishing a BP neural network structure; the number of the principal component components is the number of input neurons of the BP neural network structure, and the principal component vector matrix is the input quantity of the BP neural network structure; the output target of the BP neural network structure is the optimal dosage; the BP neural network structure is a multi-input single-output three-layer model;

the first initialization unit is used for initializing the connection weight and the threshold of the BP neural network structure, setting a sample counter and a learning frequency counter to be 1, and determining the minimum error and the iteration frequency; the connection weight comprises a weight from a hidden layer to an input layer and a weight from an output layer to the hidden layer; the threshold comprises a threshold of each neuron node in the hidden layer and a threshold of each neuron node in the output layer;

the input and output calculation unit is used for inputting the c-th sample pair in the training samples into the BP neural network structure, and calculating the input and output of each neuron node in a hidden layer and the input and output of each neuron node in an output layer;

a sample pair error determination unit, configured to calculate a correction error of each neuron node in the output layer and a correction error of each neuron node in the hidden layer according to the input and the output of each neuron node in the hidden layer and the input and the output of each neuron node in the output layer, and determine an error of the c-th sample pair;

a connection weight and threshold adjusting unit, configured to adjust the connection weight and the threshold according to the error of the c-th sample pair;

the first judging unit is used for judging whether all sample pairs in the training samples are trained or not;

a first returning unit, configured to, when there is a sample pair in the training samples that is not trained, return to the step of inputting the c-th sample pair in the training samples to the BP neural network structure, and calculate input and output of each neuron node in a hidden layer and input and output of each neuron node in an output layer;

a second judging unit, configured to update the number of learning times, calculate a global error, and judge whether the global error is smaller than the set minimum error or whether the number of learning times reaches the set iteration times when all sample pairs in the training samples are trained;

a BP neural network model establishing unit, configured to establish a BP neural network model according to the adjusted connection weight and the threshold when the global error is smaller than the set minimum error or the learning frequency reaches the set iteration frequency;

and a second returning unit, configured to, when the global error is smaller than the set minimum error and the learning frequency reaches the set iteration frequency, return to the step of inputting the c-th sample pair in the training samples to the BP neural network structure, and calculate input and output of each neuron node in the hidden layer and input and output of each neuron node in the output layer.

10. The online dosing control system according to claim 7, wherein the optimal connection weight and optimal threshold obtaining module specifically includes:

the second initialization unit is used for combining the connection weight value and the threshold value in the BP neural network model to form an individual of a genetic algorithm, and determining the number S of the individuals of an initial population and the number N of genetic iterations;

the initial population determining unit is used for randomly generating an initial population, carrying out binary coding on the initialized population, determining the initial population consisting of S individuals, and setting the number of evolution times to be 1;

a fitness function determining unit for determining a fitness function of the genetic algorithm; the fitness function is an error function of the BP neural network model;

a fitness function value calculating unit, configured to calculate a fitness function value of each individual in the initial population according to the fitness function;

the third judging unit is used for judging whether the current evolution times reach the set genetic iteration times;

the optimal individual output unit is used for outputting an optimal individual when the current evolution times reach the set genetic iteration times; the optimal individual is the individual with the maximum fitness function value in the initial population;

and the initial population updating unit is used for increasing 1 for the evolution times when the current evolution times does not reach the set genetic iteration times, performing selection, crossing and variant genetic operations on the individuals with the maximum fitness function value in the initial population, updating the initial population, and returning to the step of calculating the fitness function value of each individual in the initial population according to the fitness function.