CN103488090B - Incinerator hazardous emission control system up to standard and the method for gunz machine learning - Google Patents

Abstract

The invention discloses a kind of incinerator hazardous emission control system up to standard and method of gunz machine learning. It comprises incinerator, intelligence instrument, DCS system, data-interface and host computer; DCS system comprises control station and database; Be connected with DCS system for the intelligence instrument of measuring easy survey variable, DCS system is connected with host computer by data-interface. First host computer carries out pretreatment and obfuscation to training sample, obtain new input matrix, adopt again least square method supporting vector machine to set up model to new input matrix, obtain prediction output, then prediction output is carried out to reverse gelatinization, finally introduce penalty factor and the error margin value of particle cluster algorithm to least square method supporting vector machine and be optimized, the system output being optimized. Whether the present invention has on-line measurement COD, effectively monitor COD exceeds standard, controls the advantages such as strong, the required sample number of COD emission compliance, antinoise and generalization ability is few, computational speed is fast, on-line automatic optimization.

Description

Crowd-sourcing machine learning incinerator harmful substance emission standard-reaching control system and method

Technical Field

The invention relates to the field of pesticide production, in particular to a crowd-sourcing machine learning incinerator harmful substance emission standard control system and method.

Background

China is a big country for producing and using pesticides, the number of pesticide production enterprises reaches about 4100, more than 500 raw pesticide production enterprises exist, and statistical data of the Ministry of agriculture of China shows that the total yield of pesticides reaches 171.1 ten thousand tons in 2008 in 1-11 months. The irrational structure of the agricultural chemical varieties in China further increases the difficulty of environmental management. According to incomplete statistics, the amount of wastewater discharged by the pesticide industry all over the country is about 15 hundred million tons every year. Wherein, the treatment reaches the standard and only accounts for 1 percent of the treated treatment. The incineration method is the most effective and thorough method for treating pesticide residue and waste residue at present and is the most common method for application. The Chemical Oxygen Demand (COD) of waste water after incineration is the most important index for the discharge of harmful substances after the incineration of pesticide waste liquid, but the COD cannot be measured on line, and the off-line measurement needs four or five hours once, so that the working condition change cannot be reflected in time and the actual production cannot be guided. Therefore, COD is seriously out of limits in the actual incineration process.

Disclosure of Invention

In order to overcome the defects that COD can not be measured on line and COD seriously exceeds the standard in the conventional incinerator process, the invention provides the incinerator harmful substance emission standard-reaching control system and method based on swarm intelligence machine learning, and the system and the method have the advantages of on-line measurement of COD, effective monitoring of whether COD exceeds the standard, control of COD emission standard reaching, strong noise resistance and generalization capability, few required samples, high calculation speed, on-line automatic optimization and the like.

The technical scheme adopted by the invention for solving the technical problems is as follows:

the incinerator harmful substance emission standard-reaching control system for crowd-sourcing machine learning comprises an incinerator, an intelligent instrument, a DCS (distributed control system), a data interface and an upper computer, wherein the DCS comprises a control station and a database; on-spot intelligent instrument and DCS headtotail, the DCS system is connected with the host computer, the host computer include:

the data preprocessing module is used for preprocessing the model training samples input from the DCS database, centralizing the training samples, namely subtracting the average value of the samples, and then normalizing the training samples:

calculating an average value: $\stackrel{\‾}{\mathrm{TX}}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{TX}}_i---\left(1\right)$

calculating the variance: ${\mathrm{\σ}}_x^2=\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}({\mathrm{TX}}_i-\stackrel{\‾}{\mathrm{TX}})---\left(2\right)$

and (3) standardization: $X=\frac{\mathrm{TX}-\stackrel{\‾}{\mathrm{TX}}}{{\mathrm{\σ}}_x}---\left(3\right)$

wherein, TX_iThe ith training sample is the data of key variables, Chemical Oxygen Demand (COD) and corresponding operating variables for enabling COD to be discharged when the production is normal, which are collected from a DCS database, N is the number of the training samples,is the mean of the training samples, and X is the normalized training sample. Sigma_xRepresenting the standard deviation, σ, of the training samples² _xRepresenting the variance of the training samples. And the fuzzy equation module is used for fuzzifying the standardized training sample X transmitted from the data preprocessing module. Let there be c in the system of fuzzy equations^*A fuzzy group, the centers of the fuzzy groups k and j are v_k、v_jThe ith normalized training sample X_iMembership mu for fuzzy group k_ikComprises the following steps:

${\mathrm{\μ}}_{\mathrm{ik}}={\left(\underset{j=1}{\overset{c^{*}}{\mathrm{\Σ}}}{\left(\frac{\left|\right|X_i-v_k\left|\right|}{\left|\right|X_i-v_j\left|\right|}\right)}^{\frac{2}{m-1}}\right)}^{-1}---\left(4\right)$

where m is the fractional matrix index required in the fuzzy classification process, and is usually taken as 2, | · |, which is a norm expression.

Using the above membership values or its variants to obtain a new input matrix, for the fuzzy group k, its input matrix variant is:

Φ_ik(X_i,μ_ik)=[1func(μ_ik)X_i]（5）

wherein func (. mu.)_ik) Is a membership value mu_ikDeformation function of, in general, takeexp(μ_ik) Equal, phi_ik(X_i,μ_ik) Denotes the ith input variable X_iAnd membership mu of its fuzzy group k_ikThe corresponding new input matrix. And the least square support vector machine is used as a local equation of the fuzzy equation system and performs optimal fitting on each fuzzy group. Let the ith target output of the model training sample be O_iThe least squares support vector machine equates the fitting problem to the following quadratic programming problem by transformation:

$\min R(w,\mathrm{\ξ})=\frac{1}{2}{w}^{T}w+\frac{1}{2}\mathrm{\γ}{\mathrm{\Σ}}_{i=1}^N{\mathrm{\ξ}}_i^2---\left(6\right)$

the lagrangian function is also defined:

where R (w, ξ) is the objective function of the optimization problem, minR (w, ξ) is the minimum of the objective function of the optimization problem,a non-linear mapping function, N is the number of training samples, ξ = { ξ =₁,…,ξ_NIs a relaxation variable, ξ_iIs the i-th component of the relaxation variable, α_i(i =1, …, N) is the ith component of the corresponding lagrange multiplier, w is the normal vector of the hyperplane of the least squares support vector machine, b is the corresponding offset, and γ is the penalty factor of the least squares support vector machine, the superscript T representing the transpose of the matrix. Mu.s_ikRepresents the ith normalized training sample X_iMembership, Φ, for fuzzy group k_ik(X_i,μ_ik) Denotes the ith input variable X_iAnd membership mu of its fuzzy group k_ikThe corresponding new input matrix. From equations (6), (7) and (8), the output of the fuzzy group k in the training sample i is derived as:

${\hat{y}}_{\mathrm{ik}}=\underset{m=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\α}}_m\×K\⟨{\mathrm{\Φ}}_{\mathrm{im}}(X_m,{\mathrm{\μ}}_{\mathrm{mk}}),{\mathrm{\Φ}}_{\mathrm{ik}}(X_i,{\mathrm{\μ}}_{\mathrm{ik}})\⟩+b---\left(9\right)$

wherein,for the output of the fuzzy group K in the training sample i, K<·>Is the kernel function of a least squares support vector machine, where K<·>Taking a linear kernel function. Mu.s_mkRepresents the m-th training sample X_mMembership, Φ, for fuzzy group k_mk(X_m,μ_mk) Denotes the m-th input variable X_mAnd membership mu of its fuzzy group k_mkCorresponding new input matrix α_mM =1, …, N is the mth component of the corresponding lagrange multiplier.

The output of the final fuzzy equation system is obtained by the gravity center method in the anti-fuzzy method:

${\hat{y}}_i=\frac{{\mathrm{\&Sigma;}}_{k=1}^{c^{*}}{\mathrm{\&mu;}}_{\mathrm{ik}}{\hat{y}}_{\mathrm{ik}}}{{\mathrm{\&Sigma;}}_{k=1}^{c^{*}}{\mathrm{\&mu;}}_{\mathrm{ik}}}---\left(10\right)$

the particle swarm optimization module is used for optimizing a penalty factor and an error tolerance value of a local equation of a least square support vector machine in a fuzzy equation by adopting a particle swarm optimization, and comprises the following specific steps:

① determining the optimization parameters of the particle swarm to be the penalty factor and the error tolerance value of the local equation of the least square support vector machine, the individual number of the particle swarm popsize, the maximum cyclic optimization number iter_maxInitial position r of the p-th particle_pInitial velocity v_pLocal optimum value Lbest_pAnd a global optimum Gbest for the entire population of particles.

Secondly, setting an optimization objective function, converting the optimization objective function into fitness, and evaluating each local fuzzy equation; calculating a fitness function through a corresponding error function, considering that the fitness of the particle with large error is small, and expressing the fitness function of the particle p as follows:

f_p=1/(E_p+1)（11）

in the formula, E_pIs the error function of the fuzzy equation system, expressed as:

$E_p=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{({\hat{y}}_i-O_i)}^2}---\left(12\right)$

in the formula,is the predicted output of a system of fuzzy equations, O_iIs the target output of the fuzzy equation system;

thirdly, according to the following formula, the speed and the position of each particle are circularly updated,

v_p(iter+1)=ω×v_p(iter)+m₁a₁(Lbest_p-r_p(iter))+m₂a₂(Gbest-r_p(iter))

（13）

r_p(iter+1)=r_p(iter)+v_p(iter+1)（14）

in the formula, v_pIndicates the velocity, r, of the update particle p_pIndicating the position of the update particle p, Lbest_pRepresenting the individual optimum value of the update particle p, Gbest being the training sample X corresponding to the ith normalization_iThe predicted value of COD and the variable value of the operation variable for making the COD discharge reach the standard, iter represents the cycle number, omega is the inertia weight in the particle swarm algorithm, m₁、m₂Is the corresponding acceleration factor, a₁、a₂Is [0,1 ]]A random number in between;

and fourthly, for the particle p, if the new fitness is larger than the original individual optimal value, updating the individual optimal value of the particle:

Lbest_p=f_p（15）

⑤ if the individual optimum value Lbest of particle p_pAnd (3) updating the original particle swarm global optimum value Gbest if the particle swarm global optimum value Gbest is larger than the original particle swarm global optimum value Gbest:

Gbest=Lbest_p（16）

⑥ judging whether the performance requirement is satisfied, if yes, ending the optimization to get a group of optimized fuzzy equation local equation parameters, otherwise returning to step ③, continuing the iterative optimization until reaching the maximum iterative times iter_max。

Gbest at the end of the iteration is the training sample X corresponding to the ith normalization_iThe predicted COD value and the operation variable value for making the COD discharge reach the standard.

As a preferred solution: the host computer still include: and the judgment model updating module is used for acquiring on-site intelligent instrument signals according to a set sampling time interval, comparing the obtained actually-measured chemical oxygen consumption with a function forecast value, and adding new data which reaches the standard when the production is normal in the DCS database into the training sample data if the relative error is more than 10% or the actually-measured COD data does not reach the standard.

Further, the host computer still include: the result display module is used for transmitting the COD forecast value and the operation variable value which enables the COD emission to reach the standard to the DCS, displaying the process state at a control station of the DCS, and transmitting the process state information to a field operation station for displaying through the DCS and a field bus; meanwhile, the DCS system takes the obtained operation variable value which enables the COD emission to reach the standard as a new operation variable set value, and automatically executes the COD emission standard-reaching operation. And the signal acquisition module is used for acquiring data from the database according to the set time interval of each sampling.

Still further, the key variables include the flow of waste liquid into the incinerator, the flow of air into the incinerator, and the flow of fuel into the incinerator; the manipulated variables include air flow into the incinerator and fuel flow into the incinerator.

The method for controlling the emission of the harmful substances of the incinerator to reach the standard by the system for controlling the emission of the harmful substances to reach the standard specifically comprises the following steps:

1) determining key variables used by an incinerator harmful substance emission process object according to process analysis and operation analysis, acquiring data of the variables during normal production from a DCS (distributed control system) database as an input matrix of a training sample TX, and acquiring corresponding COD (chemical oxygen demand) and operation variable data for enabling COD emission to reach the standard as an output matrix Y;

2) preprocessing a model training sample input from a DCS database, centralizing the training sample, namely subtracting the average value of the sample, and then normalizing the training sample so that the average value is 0 and the variance is 1. The processing is accomplished using the following mathematical process:

2.1) calculating the mean value: $\stackrel{\&OverBar;}{\mathrm{TX}}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{TX}}_i---\left(1\right)$

2.2) calculating the variance: ${\mathrm{\&sigma;}}_x^2=\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}({\mathrm{TX}}_i-\stackrel{\&OverBar;}{\mathrm{TX}})---\left(2\right)$

2.3) standardization: $X=\frac{\mathrm{TX}-\stackrel{\&OverBar;}{\mathrm{TX}}}{{\mathrm{\&sigma;}}_x}---\left(3\right)$

wherein, TX_iThe ith training sample is the data of key variables, Chemical Oxygen Demand (COD) and corresponding operating variables for enabling COD to be discharged when the production is normal, which are collected from a DCS database, N is the number of the training samples,is the mean of the training samples, and X is the normalized training sample. Sigma_xRepresenting the standard deviation, σ, of the training samples² _xRepresenting the variance of the training samples.

3) And fuzzifying the training sample transmitted from the data preprocessing module. Let there be c in the system of fuzzy equations^*A fuzzy group, the centers of the fuzzy groups k and j are v_k、v_jThe ith normalized training sample X_iMembership mu for fuzzy group k_ikComprises the following steps:

${\mathrm{\&mu;}}_{\mathrm{ik}}={\left(\underset{j=1}{\overset{c^{*}}{\mathrm{\&Sigma;}}}{\left(\frac{\left|\right|X_i-v_k\left|\right|}{\left|\right|X_i-v_j\left|\right|}\right)}^{\frac{2}{m-1}}\right)}^{-1}---\left(4\right)$

where m is the fractional matrix index required in the fuzzy classification process, and is usually taken as 2, | · |, which is a norm expression.

Using the above membership values or its variants to obtain a new input matrix, for the fuzzy group k, its input matrix variant is:

Φ_ik(X_i,μ_ik)=[1func(μ_ik)X_i]（5）

wherein func (. mu.)_ik) Is a membership value mu_ikDeformation function of, in general, takeexp(μ_ik) Equal, phi_ik(X_i,μ_ik) Denotes the ith input variable X_iAnd membership mu of its fuzzy group k_ikThe corresponding new input matrix.

And the least square support vector machine is used as a local equation of the fuzzy equation system and performs optimal fitting on each fuzzy group. Let the ith target output of the model training sample be O_iThe least squares support vector machine equates the fitting problem to the following quadratic programming problem by transformation:

$\min R(w,\mathrm{\&xi;})=\frac{1}{2}{w}^{T}w+\frac{1}{2}\mathrm{\&gamma;}{\mathrm{\&Sigma;}}_{i=1}^N{\mathrm{\&xi;}}_i^2---\left(6\right)$

the lagrangian function is also defined:

where R (w, ξ) is the objective function of the optimization problem, minR (w, ξ) is the minimum of the objective function of the optimization problem,a non-linear mapping function, N is the number of training samples, ξ = { ξ =₁,…,ξ_NIs a relaxation variable, ξ_iIs the i-th component of the relaxation variable, α_iI =1, …, N is the ith component of the corresponding lagrange multiplier, w is the normal vector of the hyperplane of the least squares support vector machine, b is the corresponding offset, and γ is the penalty factor of the least squares support vector machine, the superscript T representing the transpose of the matrix. Mu.s_ikRepresents the ith normalized training sample X_iMembership, Φ, for fuzzy group k_ik(X_i,μ_ik) Denotes the ith input variable X_iAnd membership mu of its fuzzy group k_ikThe corresponding new input matrix. From equations (6), (7) and (8), the output of the fuzzy group k in the training sample i is derived as:

${\hat{y}}_{\mathrm{ik}}=\underset{m=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_m\&times;K\&lang;{\mathrm{\&Phi;}}_{\mathrm{im}}(X_m,{\mathrm{\&mu;}}_{\mathrm{mk}}),{\mathrm{\&Phi;}}_{\mathrm{ik}}(X_i,{\mathrm{\&mu;}}_{\mathrm{ik}})\&rang;+b---\left(9\right)$

wherein,for the output of the fuzzy group K in the training sample i, K<·>Is the kernel function of a least squares support vector machine, where K<·>Taking a linear kernel function. Mu.s_mkRepresents the m-th training sample X_mMembership, Φ, for fuzzy group k_mk(X_m,μ_mk) Denotes the m-th input variable X_mAnd membership mu of its fuzzy group k_mkCorresponding new input matrix α_mM =1, …, N is the mth component of the corresponding lagrange multiplier.

The output of the final fuzzy equation system is obtained by the gravity center method in the anti-fuzzy method:

${\hat{y}}_i=\frac{{\mathrm{\&Sigma;}}_{k=1}^{c^{*}}{\mathrm{\&mu;}}_{\mathrm{ik}}{\hat{y}}_{\mathrm{ik}}}{{\mathrm{\&Sigma;}}_{k=1}^{c^{*}}{\mathrm{\&mu;}}_{\mathrm{ik}}}---\left(10\right)$

4) the method comprises the following steps of optimizing a penalty factor and an error tolerance value of a local equation of a least square support vector machine in a fuzzy equation by adopting a particle swarm optimization, wherein the method specifically comprises the following steps:

① determining the optimization parameters of the particle swarm to be the penalty factor and the error tolerance value of the local equation of the least square support vector machine, the individual number of the particle swarm popsize, the maximum cyclic optimization number iter_maxInitial position r of the p-th particle_pInitial velocity v_pLocal optimum value Lbest_pAnd a global optimum Gbest for the entire population of particles.

Secondly, setting an optimization objective function, converting the optimization objective function into fitness, and evaluating each local fuzzy equation; calculating a fitness function through a corresponding error function, considering that the fitness of the particle with large error is small, and expressing the fitness function of the particle p as follows:

f_p=1/(E_p+1)（11）

in the formula, E_pIs the error function of the fuzzy equation system, expressed as:

$E_p=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{({\hat{y}}_i-O_i)}^2}---\left(12\right)$

in the formula,is the predicted output of a system of fuzzy equations, O_iIs the target output of the fuzzy equation system;

thirdly, according to the following formula, the speed and the position of each particle are circularly updated,

v_p(iter+1)=ω×v_p(iter)+m₁a₁(Lbest_p-r_p(iter))+m₂a₂(Gbest-r_p(iter))

（13）

r_p(iter+1)=r_p(iter)+v_p(iter+1)（14）

in the formula, v_pIndicates the velocity, r, of the update particle p_pIndicating the position of the update particle p, Lbest_pRepresenting the individual optimum value of the update particle p, Gbest being the training sample X corresponding to the ith normalization_iThe predicted value of COD and the variable value of the operation variable for making the COD discharge reach the standard, iter represents the cycle number, omega is the inertia weight in the particle swarm algorithm, m₁、m₂Is the corresponding acceleration factor, a₁、a₂Is [0,1 ]]A random number in between;

and fourthly, for the particle p, if the new fitness is larger than the original individual optimal value, updating the individual optimal value of the particle:

Lbest_p=f_p（15）

⑤ if the individual optimum value Lbest of particle p_pAnd (3) updating the original particle swarm global optimum value Gbest if the particle swarm global optimum value Gbest is larger than the original particle swarm global optimum value Gbest:

Gbest=Lbest_p（16）

⑥ judging whether the performance requirement is satisfied, if yes, ending the optimization to get a group of optimized fuzzy equation local equation parameters, otherwise returning to step ③, continuing the iterative optimization until reaching the maximum iterative times iter_max。

Obtaining Gbest when iteration is terminated, namely the training sample X corresponding to the ith normalization_iThe predicted COD value and the operation variable value for making the COD discharge reach the standard.

As a preferred solution: the method further comprises the following steps: 5) and acquiring on-site intelligent instrument signals according to a set sampling time interval, comparing the obtained actually-measured chemical oxygen demand with a function forecast value, and if the relative error is more than 10% or the actually-measured COD data does not reach the standard, adding new data which reaches the standard when the data is normally produced in the DCS database into training sample data, and updating the model.

Further, in the step 4), transmitting the COD forecast value and the operation variable value which enables the COD emission to reach the standard to the DCS, displaying the process state at a control station of the DCS, and transmitting the process state information to a field operation station for displaying through the DCS and a field bus; meanwhile, the DCS system takes the obtained operation variable value which enables the COD emission to reach the standard as a new operation variable set value, and automatically executes the COD emission standard-reaching operation.

Still further, the key variables include the flow of waste liquid into the incinerator, the flow of air into the incinerator, and the flow of fuel into the incinerator; the manipulated variables include air flow into the incinerator and fuel flow into the incinerator.

The technical conception of the invention is as follows: the invention provides a crowd-sourcing machine-learned incinerator harmful substance emission standard-reaching control system and method, and aims to find out an operation variable value for enabling harmful substance emission to reach the standard.

The invention has the following beneficial effects: 1. establishing an online soft measurement model of quantitative relation between system key variables and chemical oxygen demand emission; 2. quickly finding out the operation condition for reaching the discharge of the chemical oxygen demand.

Drawings

FIG. 1 is a hardware block diagram of the system proposed by the present invention;

fig. 2 is a functional structure diagram of the upper computer according to the present invention.

Detailed Description

The invention is further described below with reference to the accompanying drawings. The examples are intended to illustrate the invention, but not to limit the invention, and any modifications and variations of the invention within the spirit and scope of the claims are intended to fall within the scope of the invention.

Example 1

Referring to fig. 1 and 2, the crowd-sourcing machine learning incinerator harmful substance emission standard control system comprises a field intelligent instrument 2 connected with an incinerator 1, a DCS system and an upper computer 6, wherein the DCS system comprises a data interface 3, a control station 5 and a database 4, the field intelligent instrument 2 is connected with the data interface 3, the data interface is connected with the control station 5, the database 4 and the upper computer 6, and the upper computer 6 comprises: the data preprocessing module is used for preprocessing the model training samples input from the DCS database, centralizing the training samples, namely subtracting the average value of the samples, and then normalizing the training samples:

calculating an average value: $\stackrel{\&OverBar;}{\mathrm{TX}}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{TX}}_i---\left(1\right)$

calculating the variance: ${\mathrm{\&sigma;}}_x^2=\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}({\mathrm{TX}}_i-\stackrel{\&OverBar;}{\mathrm{TX}})---\left(2\right)$

and (3) standardization: $X=\frac{\mathrm{TX}-\stackrel{\&OverBar;}{\mathrm{TX}}}{{\mathrm{\&sigma;}}_x}---\left(3\right)$

wherein, TX_iThe ith training sample is the data of key variables, Chemical Oxygen Demand (COD) and corresponding operating variables for enabling COD to be discharged when the production is normal, which are collected from a DCS database, N is the number of the training samples,is the mean of the training samples, and X is the normalized training sample. Sigma_xRepresenting the standard deviation, σ, of the training samples² _xRepresenting the variance of the training samples. And the fuzzy equation module is used for fuzzifying the standardized training sample X transmitted from the data preprocessing module. Let there be c in the system of fuzzy equations^*A fuzzy group, the centers of the fuzzy groups k and j are v_k、v_jThe ith normalized training sample X_iMembership mu for fuzzy group k_ikComprises the following steps:

${\mathrm{\&mu;}}_{\mathrm{ik}}={\left(\underset{j=1}{\overset{c^{*}}{\mathrm{\&Sigma;}}}{\left(\frac{\left|\right|X_i-v_k\left|\right|}{\left|\right|X_i-v_j\left|\right|}\right)}^{\frac{2}{m-1}}\right)}^{-1}---\left(4\right)$

where m is the fractional matrix index required in the fuzzy classification process, and is usually taken as 2, | · |, which is a norm expression.

Using the above membership values or its variants to obtain a new input matrix, for the fuzzy group k, its input matrix variant is:

Φ_ik(X_i,μ_ik)=[1func(μ_ik)X_i]（5）

wherein func (. mu.)_ik) Is a membership value mu_ikDeformation function of, in general, takeexp(μ_ik) Equal, phi_ik(X_i,μ_ik) Denotes the ith input variable X_iAnd membership mu of its fuzzy group k_ikThe corresponding new input matrix.

And the least square support vector machine is used as a local equation of the fuzzy equation system and performs optimal fitting on each fuzzy group. Let the ith target output of the model training sample be O_iThe least squares support vector machine equates the fitting problem to the following quadratic programming problem by transformation:

$\min R(w,\mathrm{\&xi;})=\frac{1}{2}{w}^{T}w+\frac{1}{2}\mathrm{\&gamma;}{\mathrm{\&Sigma;}}_{i=1}^N{\mathrm{\&xi;}}_i^2---\left(6\right)$

the lagrangian function is also defined:

where R (w, ξ) is the objective function of the optimization problem, minR (w, ξ) is the minimum of the objective function of the optimization problem,a non-linear mapping function, N is the number of training samples, ξ = { ξ =₁,…,ξ_NIs a relaxation variable, ξ_iIs the i-th component of the relaxation variable, α_i(i =1, …, N) is the ith component of the corresponding lagrange multiplier, w is the normal vector of the hyperplane of the least squares support vector machine, b is the corresponding offset, and γ is the penalty factor of the least squares support vector machine, the superscript T representing the transpose of the matrix. Mu.s_ikRepresents the ith normalized training sample X_iMembership, Φ, for fuzzy group k_ik(X_i,μ_ik) Denotes the ith input variable X_iAnd membership mu of its fuzzy group k_ikThe corresponding new input matrix. From equations (6), (7) and (8), the output of the fuzzy group k in the training sample i is derived as:

${\hat{y}}_{\mathrm{ik}}=\underset{m=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_m\&times;K\&lang;{\mathrm{\&Phi;}}_{\mathrm{im}}(X_m,{\mathrm{\&mu;}}_{\mathrm{mk}}),{\mathrm{\&Phi;}}_{\mathrm{ik}}(X_i,{\mathrm{\&mu;}}_{\mathrm{ik}})\&rang;+b---\left(9\right)$

wherein,for the output of the fuzzy group K in the training sample i, K<·>Is the kernel function of a least squares support vector machine, where K<·>Taking a linear kernel function. Mu.s_mkRepresents the m-th training sample X_mMembership, Φ, for fuzzy group k_mk(X_m,μ_mk) Denotes the m-th input variable X_mAnd membership mu of its fuzzy group k_mkCorresponding new input matrix α_mM =1, …, N is the mth component of the corresponding lagrange multiplier.

The output of the final fuzzy equation system is obtained by the gravity center method in the anti-fuzzy method:

${\hat{y}}_i=\frac{{\mathrm{\&Sigma;}}_{k=1}^{c^{*}}{\mathrm{\&mu;}}_{\mathrm{ik}}{\hat{y}}_{\mathrm{ik}}}{{\mathrm{\&Sigma;}}_{k=1}^{c^{*}}{\mathrm{\&mu;}}_{\mathrm{ik}}}---\left(10\right)$

the particle swarm optimization module is used for optimizing a penalty factor and an error tolerance value of a local equation of a least square support vector machine in a fuzzy equation by adopting a particle swarm optimization, and comprises the following specific steps:

① determining the optimization parameters of the particle swarm to be the penalty factor and the error tolerance value of the local equation of the least square support vector machine, the individual number of the particle swarm popsize, the maximum cyclic optimization number iter_maxInitial position r of the p-th particle_pInitial velocity v_pLocal optimum value Lbest_pAnd a global optimum Gbest for the entire population of particles.

Secondly, setting an optimization objective function, converting the optimization objective function into fitness, and evaluating each local fuzzy equation; calculating a fitness function through a corresponding error function, considering that the fitness of the particle with large error is small, and expressing the fitness function of the particle p as follows:

f_p=1/(E_p+1)（11）

in the formula, E_pIs the error function of the fuzzy equation system, expressed as:

$E_p=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{({\hat{y}}_i-O_i)}^2}---\left(12\right)$

in the formula,is the predicted output of a system of fuzzy equations, O_iIs the target output of the fuzzy equation system;

thirdly, according to the following formula, the speed and the position of each particle are circularly updated,

v_p(iter+1)=ω×v_p(iter)+m₁a₁(Lbest_p-r_p(iter))+m₂a₂(Gbest-r_p(iter))

（13）

r_p(iter+1)=r_p(iter)+v_p(iter+1)（14）

in the formula, v_pIndicates the velocity, r, of the update particle p_pIndicating the position of the update particle p, Lbest_pRepresenting the individual optimum value of the update particle p, Gbest being the training sample X corresponding to the ith normalization_iThe predicted value of COD and the variable value of the operation variable for making the COD discharge reach the standard, iter represents the cycle number, omega is the inertia weight in the particle swarm algorithm, m₁、m₂Is the corresponding acceleration factor, a₁、a₂Is [0,1 ]]A random number in between;

and fourthly, for the particle p, if the new fitness is larger than the original individual optimal value, updating the individual optimal value of the particle:

Lbest_p=f_p（15）

⑤ if the individual optimum value Lbest of particle p_pAnd (3) updating the original particle swarm global optimum value Gbest if the particle swarm global optimum value Gbest is larger than the original particle swarm global optimum value Gbest:

Gbest=Lbest_p（16）

⑥ judging whether the performance requirement is satisfied, if yes, ending the optimization to get a group of optimized fuzzy equation local equation parameters, otherwise returning to step ③, continuing the iterative optimization until reaching the maximum iterative times iter_max。

Gbest at the end of the iteration is the training sample X corresponding to the ith normalization_iThe predicted COD value and the operation variable value for making the COD discharge reach the standard.

The upper computer 6 further includes: the signal acquisition module 11 is used for acquiring data from a database according to a set time interval of each sampling;

the upper computer 6 further comprises: and the discrimination model updating module 12 is used for acquiring on-site intelligent instrument signals according to a set sampling time interval, comparing the obtained actual measurement COD with a function prediction value, and if the relative error is more than 10% or the actual measurement COD data does not reach the standard, adding new data which reaches the standard when the DCS database is normally produced into training sample data and updating the model. The key variables include the flow of waste liquid into the incinerator, the flow of air into the incinerator and the flow of fuel into the incinerator; the manipulated variables include air flow into the incinerator and fuel flow into the incinerator.

The system also comprises a DCS (distributed control system), wherein the DCS is composed of a data interface 3, a control station 5 and a database 4; the intelligent instrument 2, the DCS system and the upper computer 6 are sequentially connected through a field bus; the upper computer 6 also comprises a result display module 10 which is used for transmitting the COD forecast value and the operation variable value which enables the COD emission to reach the standard to the DCS, displaying the process state at a control station of the DCS and transmitting the process state information to a field operation station for displaying through the DCS and a field bus; meanwhile, the DCS system takes the obtained operation variable value which enables the COD emission to reach the standard as a new operation variable set value, and automatically executes the COD emission standard-reaching operation.

When the waste liquid incineration process is provided with the DCS system, the functions of obtaining the COD forecast value and the operation variable value for enabling the COD emission to reach the standard are mainly completed on the upper computer by utilizing the real-time and historical databases of the DCS system to detect and store the real-time dynamic data of the sample.

When the waste liquid incineration process is not equipped with a DCS system, the data memory is adopted to replace the data storage function of a real-time and historical database of the DCS system, and the functional system for obtaining the COD forecast value and the operation variable value for enabling the COD emission to reach the standard is manufactured into an independent complete system-on-chip which comprises an I/O element, a data memory, a program memory, an arithmetic unit and a display module and does not depend on the DCS system.

Example 2

Referring to fig. 1 and 2, the crowd-sourcing machine learning incinerator harmful substance emission standard control method specifically comprises the following implementation steps:

1) determining key variables used by an incinerator harmful substance emission process object according to process analysis and operation analysis, acquiring data of the variables during normal production from a DCS (distributed control system) database as an input matrix of a training sample TX, and acquiring corresponding COD (chemical oxygen demand) and operation variable data for enabling COD emission to reach the standard as an output matrix Y;

2) preprocessing a model training sample input from a DCS database, centralizing the training sample, namely subtracting the average value of the sample, and then normalizing the training sample so that the average value is 0 and the variance is 1. The processing is accomplished using the following mathematical process:

2.1) calculating the mean value： $\stackrel{\&OverBar;}{\mathrm{TX}}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{TX}}_i---\left(1\right)$

2.2) calculating the variance: ${\mathrm{\&sigma;}}_x^2=\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}({\mathrm{TX}}_i-\stackrel{\&OverBar;}{\mathrm{TX}})---\left(2\right)$

2.3) standardization: $X=\frac{\mathrm{TX}-\stackrel{\&OverBar;}{\mathrm{TX}}}{{\mathrm{\&sigma;}}_x}---\left(3\right)$

wherein, TX_iThe ith training sample is the data of key variables, Chemical Oxygen Demand (COD) and corresponding operating variables for enabling COD to be discharged when the production is normal, which are collected from a DCS database, N is the number of the training samples,is the mean of the training samples, and X is the normalized training sample. Sigma_xRepresenting the standard deviation, σ, of the training samples² _xRepresenting the variance of the training samples.

3) And fuzzifying the standardized training sample transmitted from the data preprocessing module. Let there be c in the system of fuzzy equations^*A fuzzy group, the centers of the fuzzy groups k and j are v_k、v_jThe ith normalized training sample X_iMembership mu for fuzzy group k_ikComprises the following steps:

${\mathrm{\&mu;}}_{\mathrm{ik}}={\left(\underset{j=1}{\overset{c^{*}}{\mathrm{\&Sigma;}}}{\left(\frac{\left|\right|X_i-v_k\left|\right|}{\left|\right|X_i-v_j\left|\right|}\right)}^{\frac{2}{m-1}}\right)}^{-1}---\left(4\right)$

where m is the fractional matrix index required in the fuzzy classification process, and is usually taken as 2, | · |, which is a norm expression.

Using the above membership values or its variants to obtain a new input matrix, for the fuzzy group k, its input matrix variant is:

Φ_ik(X_i,μ_ik)=[1func(μ_ik)X_i]（5）

wherein func (. mu.)_ik) Is a membership value mu_ikDeformation function of, in general, takeexp(μ_ik) Equal, phi_ik(X_i,μ_ik) Denotes the ith input variable X_iAnd membership mu of its fuzzy group k_ikThe corresponding new input matrix.

And the least square support vector machine is used as a local equation of the fuzzy equation system and performs optimal fitting on each fuzzy group. Let the ith target output of the model training sample be O_iThe least squares support vector machine equates the fitting problem to the following quadratic programming problem by transformation:

$\min R(w,\mathrm{\&xi;})=\frac{1}{2}{w}^{T}w+\frac{1}{2}\mathrm{\&gamma;}{\mathrm{\&Sigma;}}_{i=1}^N{\mathrm{\&xi;}}_i^2---\left(6\right)$

the lagrangian function is also defined:

where R (w, ξ) is the objective function of the optimization problem, minR (w, ξ) is the minimum of the objective function of the optimization problem,a non-linear mapping function, N is the number of training samples, ξ = { ξ =₁,…,ξ_NIs a relaxation variable, ξ_iIs the i-th component of the relaxation variable, α_iI =1, …, N is the ith component of the corresponding lagrange multiplier, w is the normal vector of the hyperplane of the least squares support vector machine, b is the corresponding offset, and γ is the penalty factor of the least squares support vector machine, the superscript T represents the matrixThe transposing of (1). Mu.s_ikRepresents the ith normalized training sample X_iMembership, Φ, for fuzzy group k_ik(X_i,μ_ik) Denotes the ith input variable X_iAnd membership mu of its fuzzy group k_ikThe corresponding new input matrix. From equations (6), (7) and (8), the output of the fuzzy group k in the training sample i is derived as:

${\hat{y}}_{\mathrm{ik}}=\underset{m=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_m\&times;K\&lang;{\mathrm{\&Phi;}}_{\mathrm{im}}(X_m,{\mathrm{\&mu;}}_{\mathrm{mk}}),{\mathrm{\&Phi;}}_{\mathrm{ik}}(X_i,{\mathrm{\&mu;}}_{\mathrm{ik}})\&rang;+b---\left(9\right)$

wherein,for the output of the fuzzy group K in the training sample i, K<·>Is the kernel function of a least squares support vector machine, where K<·>Taking a linear kernel function. Mu.s_mkRepresents the m-th training sample X_mMembership, Φ, for fuzzy group k_mk(X_m,μ_mk) Denotes the m-th input variable X_mAnd membership mu of its fuzzy group k_mkCorresponding new input matrix α_mM =1, …, N is the mth component of the corresponding lagrange multiplier.

The output of the final fuzzy equation system is obtained by the gravity center method in the anti-fuzzy method:

${\hat{y}}_i=\frac{{\mathrm{\&Sigma;}}_{k=1}^{c^{*}}{\mathrm{\&mu;}}_{\mathrm{ik}}{\hat{y}}_{\mathrm{ik}}}{{\mathrm{\&Sigma;}}_{k=1}^{c^{*}}{\mathrm{\&mu;}}_{\mathrm{ik}}}---\left(10\right)$

4) the method comprises the following steps of optimizing a penalty factor and an error tolerance value of a local equation of a least square support vector machine in a fuzzy equation by adopting a particle swarm optimization, wherein the method specifically comprises the following steps:

① determining the optimized parameters of the particle swarm to be the penalty factor and the error tolerance value of the local equation of the least square support vector machine, and the particle swarmNumber of individuals popsize, maximum number of cycles optimization iter_maxInitial position r of the p-th particle_pInitial velocity v_pLocal optimum value Lbest_pAnd a global optimum Gbest for the entire population of particles.

Secondly, setting an optimization objective function, converting the optimization objective function into fitness, and evaluating each local fuzzy equation; calculating a fitness function through a corresponding error function, considering that the fitness of the particle with large error is small, and expressing the fitness function of the particle p as follows:

f_p=1/(E_p+1)（11）

in the formula, E_pIs the error function of the fuzzy equation system, expressed as:

$E_p=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{({\hat{y}}_i-O_i)}^2}---\left(12\right)$

in the formula,is the predicted output of a system of fuzzy equations, O_iIs the target output of the fuzzy equation system;

thirdly, according to the following formula, the speed and the position of each particle are circularly updated,

v_p(iter+1)=ω×v_p(iter)+m₁a₁(Lbest_p-r_p(iter))+m₂a₂(Gbest-r_p(iter))

（13）

r_p(iter+1)=r_p(iter)+v_p(iter+1)（14）

in the formula, v_pIndicates the velocity, r, of the update particle p_pIndicating the position of the update particle p, Lbest_pRepresenting the individual optimum value of the update particle p, Gbest being the training sample X corresponding to the ith normalization_iThe predicted value of COD and the variable value of the operation variable for making the COD discharge reach the standard, iter represents the cycle number, omega is the inertia weight in the particle swarm algorithm, m₁、m₂Is the corresponding acceleration factor, a₁、a₂Is [0,1 ]]A random number in between;

and fourthly, for the particle p, if the new fitness is larger than the original individual optimal value, updating the individual optimal value of the particle:

Lbest_p=f_p（15）

⑤ if the individual optimum value Lbest of particle p_pAnd (3) updating the original particle swarm global optimum value Gbest if the particle swarm global optimum value Gbest is larger than the original particle swarm global optimum value Gbest:

Gbest=Lbest_p（16）

⑥ judging whether the performance requirement is satisfied, if yes, ending the optimization to get a group of optimized fuzzy equation local equation parameters, otherwise returning to step ③, continuing the iterative optimization until reaching the maximum iterative times iter_max。

Gbest at the end of the iteration is the training sample X corresponding to the ith normalization_iThe predicted COD value and the operation variable value for making the COD discharge reach the standard.

The method further comprises the following steps: 5) and acquiring on-site intelligent instrument signals according to a set sampling time interval, comparing the obtained actual measurement COD with a function forecast value, and if the relative error is more than 10% or the actual measurement COD data does not reach the standard, adding new data which reaches the standard when the data is normally produced in the DCS database into training sample data, and updating the model.

6) In the step 4), transmitting the COD forecast value and the operation variable value which enables the COD emission to reach the standard to a DCS system, displaying the process state at a control station of the DCS, and transmitting the process state information to a field operation station for displaying through the DCS system and a field bus; meanwhile, the DCS system takes the obtained operation variable value which enables the COD emission to reach the standard as a new operation variable set value, and automatically executes the COD emission standard-reaching operation. The key variables include the flow of waste liquid into the incinerator, the flow of air into the incinerator, and the flow of fuel into the incinerator; the manipulated variables include air flow into the incinerator and fuel flow into the incinerator.

Claims (2)

1. A crowd-sourcing machine learning incinerator harmful substance emission standard-reaching control system comprises an incinerator, an intelligent instrument, a DCS system, a data interface and an upper computer, wherein the DCS system comprises a control station and a database; the intelligent instrument is connected with the DCS system, the DCS system is connected with the upper computer, and the intelligent instrument is characterized in that: the host computer include:

the data preprocessing module is used for preprocessing the model training samples input from the DCS database, centralizing the training samples, namely subtracting the average value of the samples, and then normalizing the training samples:

calculating an average value: $\stackrel{\&OverBar;}{TX}=\frac{1}{N}\underset{i=1}{\overset{N}{\&Sigma;}}{\mathrm{TX}}_i---\left(1\right)$

calculating the variance: ${\mathrm{\&sigma;}}_x^2=\frac{1}{N-1}\underset{i=1}{\overset{N}{\&Sigma;}}({\mathrm{TX}}_i-\stackrel{\&OverBar;}{TX})---\left(2\right)$

and (3) standardization: $X=\frac{TX-\stackrel{\&OverBar;}{TX}}{{\mathrm{\&sigma;}}_x}---\left(3\right)$

wherein TX is a training sample, TX_iThe ith training sample is the data of key variables, Chemical Oxygen Demand (COD) and corresponding operating variables for enabling COD to be discharged when the production is normal, which are collected from a DCS database, N is the number of the training samples,is the mean value of the training sample, and X is the training sample after standardization; sigma_xRepresenting the standard deviation, σ, of the training samples² _xRepresenting the variance of the training samples;

the fuzzy equation module is used for fuzzifying the standardized training sample X transmitted from the data preprocessing module; let there be c in the system of fuzzy equations^*A fuzzy group, the centers of the fuzzy groups k and j are v_k、v_jThe ith normalized training sample X_iMembership mu for fuzzy group k_ikComprises the following steps:

${\mathrm{\&mu;}}_{ik}={\left(\underset{j=1}{\overset{c^{*}}{\&Sigma;}}{\left(\frac{\left|\right|X_i-v_k\left|\right|}{\left|\right|X_i-v_j\left|\right|}\right)}^{\frac{2}{n-1}}\right)}^{-1}---\left(4\right)$

in the formula, n is a blocking matrix index required in the fuzzy classification process, and is usually taken as 2, | · | | is a norm expression;

using the above membership values or its variants to obtain a new input matrix, for the fuzzy group k, its input matrix variant is:

Φ_ik(X_i,μ_ik)＝[1func(μ_ik)X_i](5)

wherein func (. mu.)_ik) Is a membership value mu_ikIs taken as a deformation function ofOr exp (μ)_ik)，Φ_ik(X_i,μ_ik) Denotes the ith input variable X_iAnd membership mu of its fuzzy group k_ikThe corresponding new input matrix;

the least square support vector machine is used as a local equation of the fuzzy equation system, and optimal fitting is carried out on each fuzzy group; model training sampleIs O_iThe least squares support vector machine equates the fitting problem to the following quadratic programming problem by transformation:

$\min R(w,\mathrm{\&xi;})=\frac{1}{2}{w}^{T}w+\frac{1}{2}{\mathrm{\&gamma;\&Sigma;}}_{i=1}^N{\mathrm{\&xi;}}_i^2---\left(6\right)$

the lagrangian function is also defined:

where R (w, ξ) is the objective function of the optimization problem, minR (w, ξ) is the minimum of the objective function of the optimization problem,a non-linear mapping function, N is the number of training samples, ξ ═ ξ₁,...，ξ_NIs a relaxation variable, ξ_iIs the i-th component of the relaxation variable, α_i(i ═ 1, …, N) is the ith component of the corresponding lagrange multiplier, w is the normal vector of the support vector machine hyperplane, b is the corresponding offset, and γ is the least squares support vector machineThe penalty factor of (2) is superscripted T to express the transposition of the matrix; mu.s_ikRepresents the ith normalized training sample X_iMembership, Φ, for fuzzy group k_ik(X_i,μ_ik) Denotes the ith input variable X_iAnd membership mu of its fuzzy group k_ikThe corresponding new input matrix; from equations (6), (7) and (8), the output of the fuzzy group k in the training sample i is derived as:

${\hat{y}}_{ik}=\underset{m=1}{\overset{N}{\&Sigma;}}{\mathrm{\&alpha;}}_m\&times;K<{\mathrm{\&Phi;}}_{im}(X_m,{\mathrm{\&mu;}}_{mk}),{\mathrm{\&Phi;}}_{ik}(X_i,{\mathrm{\&mu;}}_{ik})>+b---\left(9\right)$

wherein,for the output of the fuzzy group K in the training sample i, K<·>Is the kernel function of a least squares support vector machine, where K<·>Taking a linear kernel function; mu.s_mkRepresents the m-th training sample X_mMembership, Φ, for fuzzy group k_mk(X_m,μ_mk) Denotes the m-th input variable X_mAnd membership mu of its fuzzy group k_mkCorresponding new input matrix α_mM is 1, …, N is the mth component of the corresponding lagrange multiplier;

the output of the final fuzzy equation system is obtained by the gravity center method in the anti-fuzzy method:

${\hat{y}}_i=\frac{{\&Sigma;}_{k=1}^{c^{*}}{\mathrm{\&mu;}}_{ik}{\hat{y}}_{ik}}{{\&Sigma;}_{k=1}^{c^{*}}{\mathrm{\&mu;}}_{ik}}---\left(10\right)$

the particle swarm optimization module is used for optimizing a penalty factor and an error tolerance value of a local equation of a least square support vector machine in a fuzzy equation by adopting a particle swarm optimization, and comprises the following specific steps:

① determining the optimization parameters of the particle swarm to be the penalty factor and the error tolerance value of the local equation of the least square support vector machine, the individual number of the particle swarm popsize, the maximum cyclic optimization number iter_maxInitial position r of the p-th particle_pInitial velocity v_pLocal optimum value Lbest_pAnd the global optimum value Gbest of the whole particle swarm;

secondly, setting an optimization objective function, converting the optimization objective function into fitness, and evaluating each local fuzzy equation; calculating a fitness function through a corresponding error function, considering that the fitness of the particle with large error is small, and expressing the fitness function of the particle p as follows:

f_p＝1/(E_p+1)(11)

in the formula, E_pIs the error function of the fuzzy equation system, expressed as:

$E_p=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\&Sigma;}}{({\hat{y}}_i-O_i)}^2}---\left(12\right)$

in the formula,is the predicted output of a system of fuzzy equations, O_iIs the target output of the fuzzy equation system;

thirdly, according to the following formula, the speed and the position of each particle are circularly updated,

v_p(iter+1)＝ω×v_p(iter)+m₁a₁(Lbest_p-r_p(iter))+m₂a₂(Gbest-r_p(iter))(13)

r_p(iter+1)＝r_p(iter)+v_p(iter+1)(14)

in the formula, v_pIndicates the velocity, r, of the update particle p_pIndicating the position of the update particle p, Lbest_pRepresenting the individual optimum value of the update particle p, Gbest being the training sample X corresponding to the ith normalization_iThe predicted value of COD and the variable value of the operation variable for making the COD discharge reach the standard, iter represents the cycle number, omega is the inertia weight in the particle swarm algorithm, m₁、m₂Is the corresponding acceleration factor, a₁、a₂Is [0,1 ]]A random number in between;

and fourthly, for the particle p, if the new fitness is larger than the original individual optimal value, updating the individual optimal value of the particle:

Lbest_p＝f_p(15)

⑤ if the individual optimum value Lbest of particle p_pAnd (3) updating the original particle swarm global optimum value Gbest if the particle swarm global optimum value Gbest is larger than the original particle swarm global optimum value Gbest:

Gbest＝Lbest_p(16)

⑥ judging whether the performance requirement is satisfied, if yes, ending the optimization to get a group of optimized fuzzy equation local equation parameters, otherwise returning to step ③, continuing the iterative optimization until reaching the maximum iterative times iter_max；

Gbest at the end of the iteration is the training sample X corresponding to the ith normalization_iCOD prediction value and COD discharge complianceOperating the variable value;

the host computer still include:

the judging model updating module is used for acquiring signals of the intelligent instrument according to a set sampling time interval, comparing the obtained actually-measured chemical oxygen consumption with a function forecast value, and if the relative error is more than 10% or the actually-measured COD data does not reach the standard, adding new data which reaches the standard when the DCS database is normally produced into training sample data and updating the model; the result display module is used for transmitting the COD forecast value and the operation variable value which enables the COD emission to reach the standard to the DCS, displaying the values at a control station of the DCS and transmitting the values to a field operation station for displaying through the DCS and a field bus; meanwhile, the DCS system takes the obtained operation variable value which enables the COD emission to reach the standard as a new operation variable set value, and automatically executes the COD emission standard-reaching operation; the signal acquisition module is used for acquiring data from the database according to the set time interval of each sampling; the key variables include the flow of waste liquid into the incinerator, the flow of air into the incinerator and the flow of fuel into the incinerator; the manipulated variables include air flow into the incinerator and fuel flow into the incinerator.

2. A control method implemented by the crowd-sourcing machine-learned incinerator harmful emissions compliance control system of claim 1, characterized in that: the control method comprises the following concrete implementation steps:

1) determining key variables used by an incinerator harmful substance emission process object according to process analysis and operation analysis, acquiring data of the variables during normal production from a DCS (distributed control system) database as an input matrix of a training sample TX, and acquiring corresponding COD (chemical oxygen demand) and operation variable data for enabling COD emission to reach the standard as an output matrix Y;

2) preprocessing a model training sample input from a DCS database, centralizing the training sample, namely subtracting the average value of the sample, and then standardizing the training sample to ensure that the average value is 0 and the variance is 1; the processing is accomplished using the following mathematical process:

2.1) calculating the mean value: $\stackrel{\&OverBar;}{TX}=\frac{1}{N}\underset{i=1}{\overset{N}{\&Sigma;}}{\mathrm{TX}}_i---\left(1\right)$

2.2) calculating the variance: ${\mathrm{\&sigma;}}_x^2=\frac{1}{N-1}\underset{i=1}{\overset{N}{\&Sigma;}}({\mathrm{TX}}_i-\stackrel{\&OverBar;}{TX})---\left(2\right)$

2.3) standardization: $X=\frac{TX-\stackrel{\&OverBar;}{TX}}{{\mathrm{\&sigma;}}_x}---\left(3\right)$

wherein TX is a training sample, TX_iThe ith training sample is the data of key variables, Chemical Oxygen Demand (COD) and corresponding operating variables for enabling COD to be discharged when the production is normal, which are collected from a DCS database, N is the number of the training samples,is the mean value of the training sample, and X is the training sample after standardization; sigma_xRepresenting the standard deviation, σ, of the training samples² _xRepresenting the variance of the training samples;

3) fuzzification is carried out on the training samples transmitted from the data preprocessing module; let there be c in the system of fuzzy equations^*A fuzzy group, the centers of the fuzzy groups k and j are v_k、v_jThe ith normalized training sample X_iMembership mu for fuzzy group k_ikComprises the following steps:

${\mathrm{\&mu;}}_{ik}={\left(\underset{j=1}{\overset{c^{*}}{\&Sigma;}}{\left(\frac{\left|\right|X_i-v_k\left|\right|}{\left|\right|X_i-v_j\left|\right|}\right)}^{\frac{2}{m-1}}\right)}^{-1}---\left(4\right)$

in the formula, m is a blocking matrix index required in the fuzzy classification process, and is usually taken as 2, | · | | is a norm expression;

using the above membership values or its variants to obtain a new input matrix, for the fuzzy group k, its input matrix variant is:

Φ_ik(X_i,μ_ik)＝[1func(μ_ik)X_i](5)

wherein func (. mu.)_ik) Is a membership value mu_ikIs taken as a deformation function ofOr exp (μ)_ik)，Φ_ik(X_i,μ_ik) Denotes the ith input variable X_iAnd membership mu of its fuzzy group k_ikThe corresponding new input matrix;

the least square support vector machine is used as a local equation of the fuzzy equation system, and optimal fitting is carried out on each fuzzy group; let the ith target output of the model training sample be O_iThe least squares support vector machine equates the fitting problem to the following quadratic scale by transformationProblem marking:

$\min R(w,\mathrm{\&xi;})=\frac{1}{2}{w}^{T}w+\frac{1}{2}{\mathrm{\&gamma;\&Sigma;}}_{i=1}^N{\mathrm{\&xi;}}_i^2---\left(6\right)$

the lagrangian function is also defined:

where R (w, ξ) is the objective function of the optimization problem, minR (w, ξ) is the minimum of the objective function of the optimization problem,a non-linear mapping function, N is the number of training samples, ξ ═ ξ₁,...,ξ_NIs a relaxation variable, ξ_iIs the i-th component of the relaxation variable, α_i(i ═ 1, …, N) is the ith component of the corresponding lagrange multiplier, w is the normal vector of the hyperplane of the least squares support vector machine, b is the corresponding offset, and γ is the penalty factor of the least squares support vector machine, the superscript T represents the transpose of the matrix; mu.s_ikRepresents the ith normalized training sample X_iMembership, Φ, for fuzzy group k_ik(X_i,μ_ik) Denotes the ith input variable X_iAnd membership mu of its fuzzy group k_ikThe corresponding new input matrix; from equations (6), (7) and (8), the output of the fuzzy group k in the training sample i is derived as:

${\hat{y}}_{ik}=\underset{m=1}{\overset{N}{\&Sigma;}}{\mathrm{\&alpha;}}_m\&times;K<{\mathrm{\&Phi;}}_{im}(X_m,{\mathrm{\&mu;}}_{mk}),{\mathrm{\&Phi;}}_{ik}(X_i,{\mathrm{\&mu;}}_{ik})>+b---\left(9\right)$

wherein,for the output of the fuzzy group K in the training sample i, K<·>Is a least squares support vector machineKernel function, here K<·>Taking a linear kernel function; mu.s_mkRepresents the m-th training sample X_mMembership, Φ, for fuzzy group k_mk(X_m,μ_mk) Denotes the m-th input variable X_mAnd membership mu of its fuzzy group k_mkCorresponding new input matrix α_mM is 1, …, N is the mth component of the corresponding lagrange multiplier;

the output of the final fuzzy equation system is obtained by the gravity center method in the anti-fuzzy method:

${\hat{y}}_i=\frac{{\&Sigma;}_{k=1}^{c^{*}}{\mathrm{\&mu;}}_{ik}{\hat{y}}_{ik}}{{\&Sigma;}_{k=1}^{c^{*}}{\mathrm{\&mu;}}_{ik}}---\left(10\right)$

4) the method comprises the following steps of optimizing a penalty factor and an error tolerance value of a local equation of a least square support vector machine in a fuzzy equation by adopting a particle swarm optimization, wherein the method specifically comprises the following steps:

① determining the optimization parameters of the particle swarm to be the penalty factor and the error tolerance value of the local equation of the least square support vector machine, the individual number of the particle swarm popsize, the maximum cyclic optimization number iter_maxInitial position r of the p-th particle_pInitial velocity v_pLocal optimum value Lbest_pAnd the global optimum value Gbest of the whole particle swarm;

secondly, setting an optimization objective function, converting the optimization objective function into fitness, and evaluating each local fuzzy equation; calculating a fitness function through a corresponding error function, considering that the fitness of the particle with large error is small, and expressing the fitness function of the particle p as follows:

f_p＝1/(E_p+1)(11)

in the formula, E_pIs the error function of the fuzzy equation system, expressed as:

$E_p=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\&Sigma;}}{({\hat{y}}_i-O_i)}^2}---\left(12\right)$

in the formula,is the predicted output of a system of fuzzy equations, O_iIs the target output of the fuzzy equation system;

thirdly, according to the following formula, the speed and the position of each particle are circularly updated,

v_p(iter+1)＝ω×v_p(iter)+m₁a₁(Lbest_p-r_p(iter))+m₂a₂(Gbest-r_p(iter))(13)

r_p(iter+1)＝r_p(iter)+v_p(iter+1)(14)

in the formula, v_pIndicates the velocity, r, of the update particle p_pIndicating the position of the update particle p, Lbest_pRepresenting the individual optimum value of the update particle p, Gbest being the training sample X corresponding to the ith normalization_iThe predicted value of COD and the variable value of the operation variable for making the COD discharge reach the standard, iter represents the cycle number, omega is the inertia weight in the particle swarm algorithm, m₁、m₂Is the corresponding acceleration factor, a₁、a₂Is [0,1 ]]A random number in between;

and fourthly, for the particle p, if the new fitness is larger than the original individual optimal value, updating the individual optimal value of the particle:

Lbest_p＝f_p(15)

⑤ if the individual optimum value Lbest of particle p_pAnd (3) updating the original particle swarm global optimum value Gbest if the particle swarm global optimum value Gbest is larger than the original particle swarm global optimum value Gbest:

Gbest＝Lbest_p(16)

⑥ judging whether the performance requirement is satisfied, if yes, ending the optimization to get a group of optimized fuzzy equation local equation parameters, otherwise returning to step ③, continuing the iterative optimization until reaching the maximum iterative times iter_max；

Gbest at the end of the iteration is the training sample X corresponding to the ith normalization_iThe COD forecast value and the operation variable value which enables the COD discharge to reach the standard;

the method further comprises the following steps: 5) the judgment model updating module is used for acquiring signals of the intelligent instrument according to a set sampling time interval, comparing the obtained actually-measured chemical oxygen consumption with a function forecast value, and if the relative error is more than 10% or the actually-measured COD data does not reach the standard, adding new data which reaches the standard when the DCS database is normally produced into training sample data and updating the model;

6) transmitting the result of the COD forecast value obtained in the step 4) and the operation variable value which enables the COD emission to reach the standard to a DCS (distributed control system), displaying the result on a control station of the DCS, and transmitting the result to a field operation station for displaying through the DCS and a field bus; meanwhile, the DCS system takes the obtained operation variable value which enables the COD emission to reach the standard as a new operation variable set value, and automatically executes the COD emission standard-reaching operation; the key variables include the flow of waste liquid into the incinerator, the flow of air into the incinerator and the flow of fuel into the incinerator; the manipulated variables include air flow into the incinerator and fuel flow into the incinerator.