CN104613468A - Circulating fluidized bedboiler combustion optimizing control method based on fuzzy adaptive inference - Google Patents

Abstract

The invention discloses a circulating fluidized bedboiler combustion optimizing control method based on fuzzy adaptive inference, and belongs to the technical field of circulating fluidized bed combustion. A circulating fluidized bedboiler combustion optimizing control system comprises a data communication subsystem, a model prediction subsystem and a performance optimization subsystem, wherein the data communication subsystem is in data interaction with OPC server communication software of a DCS; the model prediction subsystem is connected with the data communication subsystem; the performance optimization subsystem is respectively connected with the model prediction subsystem and the data communication subsystem; a boiler efficiency and SO2 and NOx discharge model is established through fuzzy adaptive inference algorithms, the running working condition of a circulating fluidized bedboiler is optimized by selecting a fruit fly optimization algorithm with an optimum reserved strategy, optimal set values of operating variables are provided for a power station DCS basic control layer, and high-efficiency and low-pollutant discharge running of the circulating fluidized bedboiler is realized.

Description

Based on the method for controlling optimized burning in circulating fluid bed boiler of fuzzy self-adaption reasoning

Technical field

The invention belongs to Combustion technology of circulating fluidized field, especially relate to a kind of method for controlling optimized burning in circulating fluid bed boiler based on fuzzy self-adaption reasoning.

Background technology

China is as developing country maximum in the world, and the 3rd, the current production of energy Liang Ju world, primary energy consumption occupies the 2nd.But due to China's rich coal oil starvation weak breath, define the structure of taking as the leading factor with coal, wherein the coal of 50% ~ 60% is used for generating electricity, and thus also becomes the main source of pollution that caused by coal burning thing.Therefore, development Clean Coal Power Generating Technologies, improves generating efficiency, reduces disposal of pollutants, becomes the importance of China's energy strategy.Combustion technology of circulating fluidized is just becoming the emphasis of development clean coal combustion technology owing to having the wide unique advantage of the coal inferior of direct desulfurization and burning in stove, fuel tolerance and the feature such as economy, environmental protection.

Although the advantage that Combustion technology of circulating fluidized has common flour coal stove incomparable in energy-saving and environmental protection, CFBC also can along with a large amount of SO ₂and NO _xgeneration, with SO ₂and NO _xbe that main pollutant causes great harm to environment.Therefore, reduce with SO ₂and NO _xit is the research emphasis that the discharge of main pollutant also becomes CFBB field.

In recent years, under the dual-pressure that profit margin reduction and environmental requirement improve, thermal power plant is in the urgent need to strengthening the competitiveness of enterprise by improving power supplying efficiency and reduction pollutant emission.Boiler is one of large nucleus equipment of power plant 3, and for large-scale thermal power machine group, boiler efficiency often improves 1%, and the efficiency of a whole set of unit can improve 0.3 ~ 0.4 percentage point, and net coal consumption rate can reduce by 0.7% ~ 1%.Therefore, improving boiler efficiency is the key point that large-scale thermal power machine group enhances the competitiveness, and boiler combustion optimization controls with efficient, low pollution emission for target, is a kind of simple, quick, effective energy-saving and emission-reduction measure.

The present invention utilizes fuzzy self-adaption inference method to construct boiler efficiency, SO ₂and NO _xthe forecast model of concentration of emission, and then optimize the optimum setting value obtaining each performance variable, achieve the optimal control of CFBB, significant to energy-saving and emission-reduction.

Summary of the invention

The object of the invention is to propose a kind of method for controlling optimized burning in circulating fluid bed boiler based on fuzzy self-adaption reasoning, its combustion optimization control system of circulating fluidized bed comprises: data communication subsystem, model prediction subsystem, performance optimization subsystem; Wherein, the opc server bitcom interaction data of data communication subsystem and DCS system, model prediction subsystem connection data communication subsystem, performance optimization subsystem is link model predicting subsystem and data communication subsystem respectively; It is characterized in that, comprise the steps:

1) data communication subsystem passes through the opc server bitcom interaction data of RS485 communications protocol and DCS system with OPC user client communication software;

2) model prediction subsystem uses fuzzy self-adaption reasoning algorithm to utilize history data to set up boiler efficiency η respectively _boiler, SO ₂and NO _xemitted smoke model;

3) performance optimization subsystem uses the predicted value of fruit bat algorithm to model prediction subsystem of optimum reserved strategy to be optimized, and obtains the optimum setting value of each performance variable;

4) each performance variable P that subsystem obtains is optimized _a, V _rA, V _rB, SE _a, SE _b, SE _c, SE _doptimum setting value be sent to opc server bitcom by data communication subsystem, instruction is issued field apparatus by DCS system.

Described data communication subsystem passes through the opc server bitcom interaction data of RS485 communications protocol and DCS system with OPC user client communication software; Wherein, input data to comprise: unit load Load, fuel value Q, fuel volatile matter V _ar, primary air pressure P _a, First air right baffle-plate aperture V _rA, First air right baffle-plate aperture V _rB, upper Secondary Air right baffle-plate aperture SE _a, the right baffle opening SE of upper Secondary Air _b, lower Secondary Air right baffle-plate aperture SE _c, the right baffle opening SE of lower Secondary Air _d, flue gas oxygen content ρ (O ₂), unburned carbon in flue dust C _fh, exhaust gas temperature t _py, main steam flow G _gq, steam enthalpy H _gq, feedwater heat content H _gs, burner hearth total blast volume V, SO2 emissions ρ (SO ₂), oxynitride discharge amount ρ (NO _x); Output data comprise: the DCS operational ton P after optimization _a, V _rA, V _rB, SE _a, SE _b, SE _c, SE _d; In data communication subsystem data gatherer process, to the abnormal data collected, adopt Vladimir Romanovskiy criterion to reject, concrete steps are as follows:

Step 1: be provided with n data, determine suspicious data X successively _j, j ∈ [1, n];

Step 2: then calculate the ordered series of numbers mean value deleted after suspicious numerical value and standard deviation $\mathrm{\σ}=\sqrt{\frac{1}{n}\underset{i=1,i\≠j}{\overset{n}{\mathrm{\Σ}}}{(X_i-\stackrel{\‾}{X})}^2};$

Step 3: the residual error calculating suspicious data:

Step 4: according to discriminate | ε _j| >K σ judges whether this suspicious numerical value exists gross error, if existed, then reject, wherein K is test coefficient;

Step 5: for ensureing the continuity of data, utilize difference equation result of calculation to X _jreplace, concrete formula is $X_j^1=X_{j-1}+(X_{j-1}-X_{j-2}),$ for numerical value new on i position.

Described model prediction subsystem uses fuzzy self-adaption reasoning algorithm to utilize history data to set up boiler efficiency η respectively _boiler, SO ₂and NO _xemitted smoke model; First, ρ (O is utilized ₂), G _gq, H _gq, H _gs, the actual history service data such as V uses $N_{o_2}=\frac{V(21-\mathrm{\ρ}\left(O_2\right))}{V_m}$ With ${\mathrm{\η}}_{\mathrm{Boiler}}=\frac{\left[G_{\mathrm{gq}}(H_{\mathrm{gq}}-H_{\mathrm{gs}})\right]}{N_{O_2}{H}_{O_2}}$ Calculate boiler efficiency η _boiler, wherein, for the mole of oxygen consumed in combustion reaction process, mol; , V _mfor the Molar number of oxygen, m ³/ mol; for reaction Heat of Formation, herein value 360kJ/mol.Secondly, obtain 2000 groups of data samples by data communication subsystem, wherein, input sample of data is: Load, Q, V _ar, P _a, V _rA, V _rB, SE _a, SE _b, SE _c, SE _d, ρ (O ₂), C _fh, t _py, exporting data sample is: η _boiler, ρ (SO ₂), ρ (NO _x); Then, fuzzy self-adaption reasoning algorithm A, B and C is used to obtain boiler efficiency η respectively _boiler, SO ₂and NO _xemitted smoke submodel; Specifically comprise the following steps:

Step 1: setting anticipation error and maximum leaf segment are counted H; Determine blurred width α >0; Initialize root node, make its membership function N ₁(x) ≡ 1, degree of depth d=0;

Step 2: the linear dimensions θ on root node ₁use formula (1) solves:

$\left\{\begin{array}{c}{\mathrm{\θ}}_T^{j+1}={\mathrm{\θ}}_T^j+R^{j+1}(Q^{j+1}-{\left({\tilde{p}}^{j+1}\right)}^T{\mathrm{\θ}}_T^j)\\ R^{j+1}=\frac{G_j{\tilde{p}}^j}{1+{\left({\tilde{P}}^j\right)}^T{G}_j{\tilde{P}}^j}\\ G_{j+1}=G_j-G_j-{\tilde{P}}^{j+1}{\left({\tilde{P}}^{j+1}\right)}^T{G}_j{(1+{\left({\tilde{P}}^{j+1}\right)}^T{G}_j{\tilde{P}}^{j+1})}^{-1}\end{array}\right.---\left(1\right)$

In formula (1), for referring to the linear dimensions inside fuzzy rule; G _jfor intermediate variable matrix, and g ₀=β I, β are one and are greater than 10 ⁶positive number, I is unit matrix; J is the time interval; T is matrix or vector transpose; ${\tilde{P}}^j={[\frac{N_{t1}\left(x^j\right)}{\underset{t\∈T}{\mathrm{\Σ}}{N}_t\left(x^j\right)}{\hat{x}}^j,\frac{N_{t2}\left(x^j\right)}{\underset{t\∈T}{\mathrm{\Σ}}{N}_t\left(x^j\right)}{\hat{x}}^j,...,\frac{N_{\mathrm{th}}\left(x^j\right)}{\underset{t\∈T}{\mathrm{\Σ}}{N}_t\left(x^j\right)}{\hat{x}}^j]}^T,$ N _tbe the fuzzy set on Fuzzy subspaee, corresponding membership function is designated as N _t(x); Q ^jfor output valve;

Step 3: process each node on current depth d successively: divide this node, the membership function after computation partition in the new left and right child node produced, according to the linear dimensions on all leaf nodes after formula (1) computation partition;

Step 31: according to formula (2), calculates model corresponding to all input amendment and exports

$\hat{Q}\left(x\right)=\underset{t\∈T}{\mathrm{\Σ}}\frac{N_t\left(x\right)}{\underset{t\∈T}{\mathrm{\Σ}}{N}_t\left(x\right)}{\left({\mathrm{\θ}}_t\right)}^T\hat{x}---\left(2\right)$

In formula (2), N _tx () is membership function, θ _tit is linear dimensions;

Step 32: calculate root-mean-square error RMSE according to formula (3):

$\mathrm{RMSE}=\sqrt{\frac{\underset{j=1}{\overset{M}{\mathrm{\Σ}}}{({\hat{Q}}^j-Q^j)}^2}{M}}---\left(3\right)$

Wherein, M is sample size, if the root-mean-square error that after division, model exports is less than the root-mean-square error dividing front model output, so preserves and this time divide, otherwise this division is invalid, the next node of process current layer;

Step 4: after current layer is disposed, counts exceeded H if the root-mean-square error that model exports is less than anticipation error or leaf segment, obtain η _boiler=[η ₁, η ₂..., η _n] ^t, ρ (SO ₂)=[ρ S ₁, ρ S ₂..., ρ S _n] ^t, ρ (NO _x)=[ρ N ₁, ρ N ₂..., ρ N _n] ^t, then algorithm terminates; Otherwise, make d=d+1, return step 4, continue algorithm.

Described performance optimization subsystem uses the predicted value of fruit bat algorithm to model prediction subsystem of optimum reserved strategy to be optimized, obtain the optimum setting value of each performance variable, optimizing process mainly comprises optimisation strategy, constraints, optimum results three aspects, specifically comprises the following steps:

Step 1: be optimized construction of strategy to optimizing process, proposes 3 kinds of optimal way: 1) SO ₂and NO _xmeet discharge standard condition, boiler efficiency is sought the highest; 2) boiler efficiency and SO ₂discharge meets constraints, NO _xdischarge capacity optimizing is minimum; 3) boiler efficiency and NO _xdischarge meets constraints, SO ₂discharge capacity is sought minimum; Multi-objective optimization is introduced penalty function and is realized, and three kinds of optimal way are corresponding in turn to following each function:

$\min F_1\left(X\right):-{\mathrm{\η}}_B+{\mathrm{\μ}}_1{\left[\max (0,{\mathrm{\ρ}}_{{\mathrm{NO}}_x}-{\mathrm{\ρ}}_{{\mathrm{NO}}_x}^c)\right]}^2+{\mathrm{\μ}}_2{\left[\max (0,{\mathrm{\ρ}}_{{\mathrm{SO}}_2}-{\mathrm{\ρ}}_{{\mathrm{SO}}_2}^c)\right]}^2$

$\min F_2\left(X\right):{\mathrm{\ρ}}_{{\mathrm{NO}}_x}+{\mathrm{\μ}}_1{\left[\max (0,{\mathrm{\η}}_B^c-{\mathrm{\η}}_B)\right]}^2+{\mathrm{\μ}}_2{\left[\max (0,{\mathrm{\ρ}}_{{\mathrm{SO}}_2}-{\mathrm{\ρ}}_{{\mathrm{SO}}_2}^c)\right]}^2$

$\min F_3\left(X\right):{\mathrm{\ρ}}_{{\mathrm{SO}}_2}+{\mathrm{\μ}}_1{\left[\max (0,{\mathrm{\η}}_B^c-{\mathrm{\η}}_B)\right]}^2+{\mathrm{\μ}}_2{\left[\max (0,{\mathrm{\ρ}}_{{\mathrm{NO}}_x}-{\mathrm{\ρ}}_{{\mathrm{NO}}_x}^c)\right]}^2$

In formula, for NO _xthe concentration of emission upper limit, gets new standard 100mg/m ³; for SO ₂the concentration of emission upper limit, 200mg/m ³; for the minimum boiler efficiency allowed, 90%; μ ₁, μ ₂for enough large positive number;

Step 2: the bound of optimizing process adjustable parameter variable is retrained, as shown in table 1:

Table 1

Variable	p A/kP a	V rA/％	V rB/％	SE A/％	SE B/％	SE C/％	SE D/％	ρ(O 2)/％
									Lower limit	3.5	20	30	40	30	20	25	0.5
The upper limit	4.5	70	80	80	70	75	70	3.5

Step 3: the fruit bat algorithm realization optimization aim adopting optimum reserved strategy, step is as follows:

Step 31: initialize n fruit bat colony, represents n adjustable parameter variable in step 2 respectively, arranges the position of each fruit bat colony at random for [X _i, Y _i], iterations is 100, and the random direction of search of food and distance are interval [-1,1];

Step 32: the distance of fruit bat individuality to initial point calculating n population with flavor concentration decision content S _i=1/D _i, S _ibe each adjustable parameter variate-value;

Step 33: the optimizing strategic function value in performance optimization subsystem Optimization Steps 1 is fitness function, namely flavor concentration function, by each S in step 32 _inO is obtained through adaptive nuero-fuzzy inference system model _xconcentration ρ (NO _x), SO ₂concentration ρ (SO ₂) and boiler efficiency η _boiler, substitute into the optimizing strategic function in Optimization Steps 1, try to achieve the flavor concentration value of this fruit bat body position, and retain this fruit bat position, ρ (NO _x), ρ (SO ₂) and η _boilerand flavor concentration value;

Step 34: by iteration optimizing, retains flavor concentration minimum, and each population fruit bat optimal location now and NO _xconcentration, SO ₂concentration, boiler efficiency, and utilize the optimal location of fruit bat to calculate the value of each adjustable parameter;

Step 4: draw optimum results.

The invention has the beneficial effects as follows, the operating mode minimum to efficiency, makes SO ₂and NO _xafter meeting the efficiency optimization of discharge standard condition, boiler efficiency improves about 3.671%, NO _xdischarge also decreases before relatively optimizing; To NO _xdischarge the highest operating mode, make boiler efficiency and SO ₂discharge meets the NO of standard conditions _xthe Emission Optimization, NO _xdischarge reduces 52.67mg/m3; To SO ₂discharge the highest operating mode, make boiler efficiency and NO _xdischarge meets the SO of standard conditions ₂the Emission Optimization, SO ₂discharge reduces 50.55mg/m3; Reach the requirement of environmental regulation.

Accompanying drawing explanation

Fig. 1 is the structured flowchart of the combustion optimization control system of circulating fluidized bed based on fuzzy self-adaption reasoning;

Fig. 2 is boiler efficiency, SO based on fuzzy self-adaption reasoning algorithm ₂and NO _xthe model structure figure of discharge;

Fig. 3 is the fuzzy self-adaption reasoning algorithm flow chart based on combustion optimization control system of circulating fluidized bed;

Fig. 4 is the algorithm flow chart of the fruit bat algorithm adopting optimum reserved strategy.

Detailed description of the invention

The present invention proposes a kind of method for controlling optimized burning in circulating fluid bed boiler based on fuzzy self-adaption reasoning, elaborates to the present invention below in conjunction with the drawings and specific embodiments.

Figure 1 shows that the structured flowchart of the combustion optimization control system of circulating fluidized bed based on fuzzy self-adaption reasoning, described Optimal Control System comprises: data communication subsystem, model prediction subsystem, performance optimization subsystem; Wherein, the opc server bitcom interaction data of data communication subsystem and DCS system, model prediction subsystem connection data communication subsystem, performance optimization subsystem is link model predicting subsystem and data communication subsystem respectively.

Based on the data communication subsystem of the combustion optimization control system of circulating fluidized bed of fuzzy self-adaption reasoning with the opc server bitcom interaction data of OPC user client communication software by RS485 communications protocol and DCS system; Wherein, input data to comprise: unit load Load, fuel value Q, fuel volatile matter V _ar, primary air pressure P _a, First air right baffle-plate aperture V _rA, First air right baffle-plate aperture V _rB, upper Secondary Air right baffle-plate aperture SE _a, the right baffle opening SE of upper Secondary Air _b, lower Secondary Air right baffle-plate aperture SE _c, the right baffle opening SE of lower Secondary Air _d, flue gas oxygen content ρ (O ₂), unburned carbon in flue dust C _fh, exhaust gas temperature t _py, main steam flow G _gq, steam enthalpy H _gq, feedwater heat content H _gs, burner hearth total blast volume V, SO2 emissions ρ (SO ₂), oxynitride discharge amount ρ (NO _x); Output data comprise: the DCS operational ton P after optimization _a, V _rA, V _rB, SE _a, SE _b, SE _c, SE _d.

In data acquisition, rejected the abnormal data collected by Vladimir Romanovskiy criterion, concrete steps are as follows:

Step 1: be provided with n data, determine suspicious data X successively _jj ∈ [1, n];

Step 2: then calculate the ordered series of numbers mean value deleted after suspicious numerical value and standard deviation $\mathrm{\σ}=\sqrt{\frac{1}{n}\underset{i=1,i\≠j}{\overset{n}{\mathrm{\Σ}}}{(X_i-\stackrel{\‾}{X})}^2};$

Step 3: the residual error calculating suspicious data:

Step 4: according to discriminate | ε _j| >K σ judges whether this suspicious numerical value exists gross error, if existed, then reject, wherein K is test coefficient;

Step 5: for ensureing the continuity of data, utilize difference equation result of calculation to X _jreplace, concrete formula is $X_j^1=X_{j-1}+(X_{j-1}-X_{j-2}),$ for numerical value new on i position.

Figure 2 shows that boiler efficiency, the SO based on fuzzy self-adaption reasoning algorithm ₂and NO _xthe model structure figure of discharge.Model prediction subsystem uses fuzzy self-adaption reasoning algorithm A, B and C, utilizes image data, sets up boiler efficiency η _boiler, SO ₂and NO _xdischarge soft-sensing model.Utilize ρ (O ₂), G _gq, H _gq, H _gs, the actual history service data such as V uses with calculate boiler efficiency η _boiler, wherein, G _gqfor main steam flow, kg; H _gqfor steam enthalpy, kJ/kg; H _gsfor feedwater heat content, kJ/kg; V is burner hearth total blast volume, m ³; for the mole of oxygen consumed in combustion reaction process, mol; , V _mfor the Molar number of oxygen, m ³/ mol; for reaction Heat of Formation, herein value 360kJ/mol.Utilize described soft-sensing model, by Load, Q, V _ar, P _a, V _rA, V _rB, SE _a, SE _b, SE _c, SE _d, ρ (O ₂) etc. variable prediction boiler efficiency, SO ₂and NO _xconcentration of emission.

Figure 3 shows that the fuzzy self-adaption reasoning algorithm flow chart based on combustion optimization control system of circulating fluidized bed; First, ρ (O is utilized ₂), G _gq, H _gq, H _gs, the actual history service data such as V uses $N_{o_2}=\frac{V(21-\mathrm{\ρ}\left(O_2\right))}{V_m}$ With ${\mathrm{\η}}_{\mathrm{Boiler}}=\frac{\left[G_{\mathrm{gq}}(H_{\mathrm{gq}}-H_{\mathrm{gs}})\right]}{N_{O_2}{H}_{O_2}}$ Calculate boiler efficiency η _boiler; Secondly, obtain 2000 groups of data samples by data communication subsystem, wherein, input sample of data is: Load, Q, V _ar, P _a, V _rA, V _rB, SE _a, SE _b, SE _c, SE _d, ρ (O ₂), C _fh, t _py, exporting data sample is: η _boiler, ρ (SO ₂), ρ (NO _x); Fuzzy self-adaption reasoning algorithm A, B and C is used to obtain boiler efficiency η respectively _boiler, SO ₂and NO _xemitted smoke submodel; Specifically comprise the following steps:

Step 1: setting anticipation error and maximum leaf segment are counted H; Determine blurred width α >0; Initialize root node, make its membership function N ₁(x) ≡ 1, degree of depth d=0;

Step 2: the linear dimensions θ on root node ₁use formula (1) solves:

$\left\{\begin{array}{c}{\mathrm{\θ}}_T^{j+1}={\mathrm{\θ}}_T^j+R^{j+1}(Q^{j+1}-{\left({\tilde{p}}^{j+1}\right)}^T{\mathrm{\θ}}_T^j)\\ R^{j+1}=\frac{G_j{\tilde{p}}^j}{1+{\left({\tilde{P}}^j\right)}^T{G}_j{\tilde{P}}^j}\\ G_{j+1}=G_j-G_j-{\tilde{P}}^{j+1}{\left({\tilde{P}}^{j+1}\right)}^T{G}_j{(1+{\left({\tilde{P}}^{j+1}\right)}^T{G}_j{\tilde{P}}^{j+1})}^{-1}\end{array}\right.---\left(1\right)$

In formula (1), for referring to the linear dimensions inside fuzzy rule; G _jfor intermediate variable matrix, and g ₀=β I, β are one and are greater than 10 ⁶positive number, I is unit matrix; J is the time interval; T is matrix or vector transpose; ${\tilde{P}}^j={[\frac{N_{t1}\left(x^j\right)}{\underset{t\∈T}{\mathrm{\Σ}}{N}_t\left(x^j\right)}{\hat{x}}^j,\frac{N_{t2}\left(x^j\right)}{\underset{t\∈T}{\mathrm{\Σ}}{N}_t\left(x^j\right)}{\hat{x}}^j,...,\frac{N_{\mathrm{th}}\left(x^j\right)}{\underset{t\∈T}{\mathrm{\Σ}}{N}_t\left(x^j\right)}{\hat{x}}^j]}^T,$ N _tbe the fuzzy set on Fuzzy subspaee, corresponding membership function is designated as N _t(x); Q ^jfor output valve;

Step 3: process each node on current depth d successively: divide this node, the membership function after computation partition in the new left and right child node produced, according to the linear dimensions on all leaf nodes after formula (1) computation partition;

Step 31: according to formula (2), calculates model corresponding to all input amendment and exports

$\hat{Q}\left(x\right)=\underset{t\∈T}{\mathrm{\Σ}}\frac{N_t\left(x\right)}{\underset{t\∈T}{\mathrm{\Σ}}{N}_t\left(x\right)}{\left({\mathrm{\θ}}_t\right)}^T\hat{x}---\left(2\right)$

In formula (2), N _tx () is membership function, θ _tit is linear dimensions;

Step 32: calculate root-mean-square error RMSE according to formula (3):

$\mathrm{RMSE}=\sqrt{\frac{\underset{j=1}{\overset{M}{\mathrm{\Σ}}}{({\hat{Q}}^j-Q^j)}^2}{M}}---\left(3\right)$

Wherein, M is sample size, if the root-mean-square error that after division, model exports is less than the root-mean-square error dividing front model output, so preserves and this time divide, otherwise this division is invalid, the next node of process current layer;

Step 4: after current layer is disposed, counts exceeded H if the root-mean-square error that model exports is less than anticipation error or leaf segment, obtain η _boiler=[η ₁, η ₂..., η _n] ^t, ρ (SO ₂)=[ρ S ₁, ρ S ₂..., ρ S _n] ^t, ρ (NO _x)=[ρ N ₁, ρ N ₂..., ρ N _n] ^t, then algorithm terminates; Otherwise, make d=d+1, return step 4, continue algorithm.

Described performance optimization subsystem uses the predicted value of fruit bat algorithm to model prediction subsystem of optimum reserved strategy to be optimized, obtain the optimum setting value of each performance variable, optimizing process mainly comprises optimisation strategy, constraints, optimum results three aspects, specifically comprises the following steps:

Step 1: be optimized construction of strategy to optimizing process, proposes 3 kinds of optimal way: 1) SO ₂and NO _xmeet discharge standard condition, boiler efficiency is sought the highest; 2) boiler efficiency and SO ₂discharge meets constraints, NO _xdischarge capacity optimizing is minimum; 3) boiler efficiency and NO _xdischarge meets constraints, SO ₂discharge capacity is sought minimum; Multi-objective optimization is introduced penalty function and is realized, and three kinds of optimal way are corresponding in turn to following each function:

$\min F_1\left(X\right):-{\mathrm{\η}}_B+{\mathrm{\μ}}_1{\left[\max (0,{\mathrm{\ρ}}_{{\mathrm{NO}}_x}-{\mathrm{\ρ}}_{{\mathrm{NO}}_x}^c)\right]}^2+{\mathrm{\μ}}_2{\left[\max (0,{\mathrm{\ρ}}_{{\mathrm{SO}}_2}-{\mathrm{\ρ}}_{{\mathrm{SO}}_2}^c)\right]}^2$

$\min F_2\left(X\right):{\mathrm{\ρ}}_{{\mathrm{NO}}_x}+{\mathrm{\μ}}_1{\left[\max (0,{\mathrm{\η}}_B^c-{\mathrm{\η}}_B)\right]}^2+{\mathrm{\μ}}_2{\left[\max (0,{\mathrm{\ρ}}_{{\mathrm{SO}}_2}-{\mathrm{\ρ}}_{{\mathrm{SO}}_2}^c)\right]}^2$

$\min F_3\left(X\right):{\mathrm{\ρ}}_{{\mathrm{SO}}_2}+{\mathrm{\μ}}_1{\left[\max (0,{\mathrm{\η}}_B^c-{\mathrm{\η}}_B)\right]}^2+{\mathrm{\μ}}_2{\left[\max (0,{\mathrm{\ρ}}_{{\mathrm{NO}}_x}-{\mathrm{\ρ}}_{{\mathrm{NO}}_x}^c)\right]}^2$

In formula, for NO _xthe concentration of emission upper limit, gets new standard 100mg/m ³; for SO ₂the concentration of emission upper limit, 200mg/m ³; for the minimum boiler efficiency allowed, 90%; μ ₁, μ ₂for enough large positive number;

Step 2: the bound of optimizing process adjustable parameter variable is retrained, as shown in table 1

Table 1

Variable	p A/kP a	V rA/％	V rB/％	SE A/％	SE B/％	SE C/％	SE D/％	ρ(O 2)/％
									Lower limit	3.5	20	30	40	30	20	25	0.5
The upper limit	4.5	70	80	80	70	75	70	3.5

Step 3: the fruit bat algorithm realization optimization aim adopting optimum reserved strategy, as shown in Figure 4, concrete steps are as follows:

Step 31: initialize n fruit bat colony, represents n adjustable parameter variable in step 2 respectively, arranges the position of each fruit bat colony at random for [X _i, Y _i], iterations is 100, and the random direction of search of food and distance are interval [-1,1];

Step 32: the distance of fruit bat individuality to initial point calculating n population with flavor concentration decision content S _i=1/D _i, S _ibe each adjustable parameter variate-value;

Step 33: the optimizing strategic function value in performance optimization subsystem Optimization Steps 1 is fitness function, namely flavor concentration function, by each S in step 32 _inO is obtained through adaptive nuero-fuzzy inference system model _xconcentration ρ (NO _x), SO ₂concentration ρ (SO ₂) and boiler efficiency η _boiler, substitute into the optimizing strategic function in step 1, try to achieve the flavor concentration value of this fruit bat body position, and retain this fruit bat position, ρ (NO _x), ρ (SO ₂) and η _boilerand flavor concentration value;

Step 34: by iteration optimizing, retains flavor concentration minimum, and each population fruit bat optimal location now and NO _xconcentration, SO ₂concentration, boiler efficiency, and utilize the optimal location of fruit bat to calculate the value of each adjustable parameter;

Step 4: draw optimum results, as shown in the table:

By optimizing each performance variable P that subsystem obtains _a, V _rA, V _rB, SE _a, SE _b, SE _c, SE _doptimum setting value, be sent to opc server bitcom by data communication subsystem, thus instruction is issued field apparatus by DCS system, realize high efficiency for circulating fluidized bed boiler, low pollution emission run.

The above; be only the present invention's preferably detailed description of the invention, but protection scope of the present invention is not limited thereto, is anyly familiar with those skilled in the art in the technical scope that the present invention discloses; the change that can expect easily or replacement, all should be encompassed within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with the protection domain of claim.

Claims (4)

1. based on a method for controlling optimized burning in circulating fluid bed boiler for fuzzy self-adaption reasoning, its combustion optimization control system of circulating fluidized bed comprises: data communication subsystem, model prediction subsystem, performance optimization subsystem; Wherein, the opc server bitcom interaction data of data communication subsystem and DCS system, model prediction subsystem connection data communication subsystem, performance optimization subsystem is link model predicting subsystem and data communication subsystem respectively; It is characterized in that, comprise the steps:

1) data communication subsystem passes through the opc server bitcom interaction data of RS485 communications protocol and DCS system with OPC user client communication software;

2) model prediction subsystem uses fuzzy self-adaption reasoning algorithm to utilize history data to set up boiler efficiency η respectively _boiler, SO ₂and NO _xemitted smoke model;

3) performance optimization subsystem uses the predicted value of fruit bat algorithm to model prediction subsystem of optimum reserved strategy to be optimized, and obtains the optimum setting value of each performance variable;

4) each performance variable P that subsystem obtains is optimized _a, V _rA, V _rB, SE _a, SE _b, SE _c, SE _doptimum setting value be sent to opc server bitcom by data communication subsystem, instruction is issued field apparatus by DCS system.

2. method according to claim 1, is characterized in that, described data communication subsystem passes through the opc server bitcom interaction data of RS485 communications protocol and DCS system with OPC user client communication software; Wherein, input data to comprise: unit load Load, fuel value Q, fuel volatile matter V _ar, primary air pressure P _a, First air right baffle-plate aperture V _rA, First air right baffle-plate aperture V _rB, upper Secondary Air right baffle-plate aperture SE _a, the right baffle opening SE of upper Secondary Air _b, lower Secondary Air right baffle-plate aperture SE _c, the right baffle opening SE of lower Secondary Air _d, flue gas oxygen content ρ (O ₂), unburned carbon in flue dust C _fh, exhaust gas temperature t _py, main steam flow G _gq, steam enthalpy H _gq, feedwater heat content H _gs, burner hearth total blast volume V, SO2 emissions ρ (SO ₂), oxynitride discharge amount ρ (NO _x); Output data comprise: the DCS operational ton P after optimization _a, V _rA, V _rB, SE _a, SE _b, SE _c, SE _d; In data communication subsystem data gatherer process, to the abnormal data collected, adopt Vladimir Romanovskiy criterion to reject, concrete steps are as follows:

Step 1: be provided with n data, determine suspicious data X successively _j, j ∈ [1, n];

Step 2: then calculate the ordered series of numbers mean value deleted after suspicious numerical value and standard deviation $\mathrm{\σ}=\sqrt{\frac{1}{n}\underset{i=1,i\≠j}{\overset{n}{\mathrm{\Σ}}}{(X_i-\stackrel{\‾}{X})}^2};$

Step 3: the residual error calculating suspicious data:

Step 4: according to discriminate | ε _j| >K σ judges whether this suspicious numerical value exists gross error, if existed, then reject, wherein K is test coefficient;

Step 5: for ensureing the continuity of data, utilize difference equation result of calculation to X _jreplace, concrete formula is for numerical value new on i position.

3. method according to claim 1, is characterized in that, described model prediction subsystem uses fuzzy self-adaption reasoning algorithm to utilize history data to set up boiler efficiency η respectively _boiler, SO ₂and NO _xemitted smoke model; First, ρ (O is utilized ₂), G _gq, H _gq, H _gs, the actual history service data such as V uses $N_{O_2}=\frac{V(21-\mathrm{\ρ}\left(O_2\right))}{V_m}$ With ${\mathrm{\η}}_{\mathrm{Boiler}}=\frac{\left[G_{\mathrm{gq}}(H_{\mathrm{gq}}-H_{\mathrm{gs}})\right]}{N_{O_2}{H}_{O_2}}$ Calculate boiler efficiency η _boiler, wherein, G _gqfor main steam flow, kg; H _gqfor steam enthalpy, kJ/kg; H _gsfor feedwater heat content, kJ/kg; V is burner hearth total blast volume, m ³; for the mole of oxygen consumed in combustion reaction process, mol; , V _mfor the Molar number of oxygen, m ³/ mol; for reaction Heat of Formation, herein value 360kJ/mol; Secondly, obtain 2000 groups of data samples by data communication subsystem, wherein, input sample of data is: Load, Q, V _ar, P _a, V _rA, V _rB, SE _a, SE _b, SE _c, SE _d, ρ (O ₂), C _fh, t _py, exporting data sample is: η _boiler, ρ (SO ₂), ρ (NO _x); Then, fuzzy self-adaption reasoning algorithm A, B and C is used to obtain boiler efficiency η respectively _boiler, SO ₂and NO _xemitted smoke submodel; Specifically comprise the following steps:

Step 1: setting anticipation error and maximum leaf segment are counted H; Determine blurred width α >0; Initialize root node, make its membership function N ₁(x) ≡ 1, degree of depth d=0;

Step 2: the linear dimensions θ on root node ₁use formula (1) solves:

$\left\{\begin{array}{c}{\mathrm{\θ}}_T^{j+1}={\mathrm{\θ}}_T^j+R^{j+1}(Q^{j+1}-{\left({\tilde{P}}^{j+1}\right)}^T{\mathrm{\θ}}_T^j)\\ R^{j+1}=\frac{G_j{\tilde{P}}^j}{1+{\left({\tilde{P}}^j\right)}^T{G_j{\tilde{P}}^j}^{}}\\ G_{j+1}=G_j-G_j{\tilde{P}}^{j+1}{\left({\tilde{P}}^{j+1}\right)}^T{G}_j{(1+{\left({\tilde{P}}^{j+1}\right)}^T{G}_j{\tilde{P}}^{j+1})}^{-1}\end{array}\right.---\left(1\right)$

In formula (1), for referring to the linear dimensions inside fuzzy rule; G _jfor intermediate variable matrix, and g ₀=β I, β are one and are greater than 10 ⁶positive number, I is unit matrix; J is the time interval; T is matrix or vector transpose; ${\tilde{P}}^j={[\frac{N_{t1}\left(x^j\right)}{\underset{t\∈T}{\mathrm{\Σ}}{N}_t\left(x^j\right)}{\hat{x}}^j,\frac{N_{t2}\left(x^j\right)}{\underset{t\∈T}{\mathrm{\Σ}}{N}_t\left(x^j\right)}{\hat{x}}^j,...,\frac{N_{\mathrm{th}}\left(x^j\right)}{\underset{t\∈T}{\mathrm{\Σ}}{N}_t\left(x^j\right)}]}^T,$ N _tbe the fuzzy set on Fuzzy subspaee, corresponding membership function is designated as N _t(x); Q ^jfor output valve;

Step 3: process each node on current depth d successively: divide this node, the membership function after computation partition in the new left and right child node produced, according to the linear dimensions on all leaf nodes after formula (1) computation partition;

Step 31: according to formula (2), calculates model corresponding to all input amendment and exports

$\hat{Q}\left(x\right)=\underset{t\∈T}{\mathrm{\Σ}}\frac{N_t\left(x\right)}{N_t\left(x\right)}{\left({\mathrm{\θ}}_t\right)}^T\hat{x}---\left(2\right)$

In formula (2), N _tx () is membership function, θ _tit is linear dimensions;

Step 32: calculate root-mean-square error RMSE according to formula (3):

$\mathrm{RMSE}=\sqrt{\frac{\underset{j=1}{\overset{M}{\mathrm{\Σ}}}{({\hat{Q}}^j-Q^j)}^2}{M}}---\left(3\right)$

Wherein, M is sample size, if the root-mean-square error that after division, model exports is less than the root-mean-square error dividing front model output, so preserves and this time divide, otherwise this division is invalid, the next node of process current layer;

Step 4: after current layer is disposed, if model export root-mean-square error be less than anticipation error or

Leaf segment is counted and has exceeded H, obtains η _boiler=[η ₁, η ₂..., η _n] ^t, ρ (SO ₂)=[ρ S ₁, ρ S ₂..., ρ S _n] ^t, ρ (NO _x)=[ρ N ₁, ρ N ₂..., ρ N _n] ^t, then algorithm terminates; Otherwise, make d=d+1, return step 4, continue algorithm.

4. method according to claim 1, it is characterized in that, described performance optimization subsystem uses the predicted value of fruit bat algorithm to model prediction subsystem of optimum reserved strategy to be optimized, obtain the optimum setting value of each performance variable, optimizing process mainly comprises optimisation strategy, constraints, optimum results three aspects, specifically comprises the following steps:

Step 1: be optimized construction of strategy to optimizing process, proposes 3 kinds of optimal way: 1) SO ₂and NO _xmeet discharge standard condition, boiler efficiency is sought the highest; 2) boiler efficiency and SO ₂discharge meets constraints, NO _xdischarge capacity optimizing is minimum; 3) boiler efficiency and NO _xdischarge meets constraints, SO ₂discharge capacity is sought minimum; Multi-objective optimization is introduced penalty function and is realized, and three kinds of optimal way are corresponding in turn to following each function:

$\min F_1\left(X\right):-{\mathrm{\η}}_B+{\mathrm{\μ}}_1{\left[\max (0,{\mathrm{\ρ}}_{{\mathrm{NO}}_2}-{\mathrm{\ρ}}_{{\mathrm{NO}}_2}^c)\right]}^2+{\mathrm{\μ}}_2{\left[\max (0,{\mathrm{\ρ}}_{{\mathrm{SO}}_2}-{\mathrm{\ρ}}_{{\mathrm{SO}}_2}^c)\right]}^2$

$\min F_2\left(X\right):{\mathrm{\ρ}}_{{\mathrm{NO}}_x}+{\mathrm{\μ}}_1{\left[\max (0,{\mathrm{\η}}_B^c-{\mathrm{\η}}_B)\right]}^2+{\mathrm{\μ}}_2{\left[\max (0,{\mathrm{\ρ}}_{{\mathrm{SO}}_2}-{\mathrm{\ρ}}_{{\mathrm{SO}}_2}^c)\right]}^2$

$\min F_3\left(X\right):{\mathrm{\ρ}}_{{\mathrm{NO}}_2}+{\mathrm{\μ}}_1{\left[\max (0,{\mathrm{\η}}_B^c-{\mathrm{\η}}_B)\right]}^2+{\mathrm{\μ}}_2{\left[\max (0,{\mathrm{\ρ}}_{{\mathrm{NO}}_x}-{\mathrm{\ρ}}_{{\mathrm{NO}}_x}^c)\right]}^2$

In formula, for NO _xthe concentration of emission upper limit, gets new standard 100mg/m ³; for SO ₂the concentration of emission upper limit, 200mg/m ³; for the minimum boiler efficiency allowed, 90%; μ ₁, μ ₂for enough large positive number;

Step 2: the bound of optimizing process adjustable parameter variable is retrained;

Step 3: the fruit bat algorithm realization optimization aim adopting optimum reserved strategy, step is as follows:

Step 31: initialize n fruit bat colony, represents n adjustable parameter variable in step 2 respectively, arranges the position of each fruit bat colony at random for [X _i, Y _i], iterations is 100, and the random direction of search of food and distance are interval [-1,1];

Step 32: the distance of fruit bat individuality to initial point calculating n population with flavor concentration decision content S _i=1/D _i, S _ibe each adjustable parameter variate-value;

Step 33: the optimizing strategic function value in performance optimization subsystem Optimization Steps 1 is fitness function, namely flavor concentration function, by each S in step 32 _inO is obtained through adaptive nuero-fuzzy inference system model _xconcentration ρ (NO _x), SO ₂concentration ρ (SO ₂) and boiler efficiency η _boiler, substitute into the optimizing strategic function in Optimization Steps 1, try to achieve the flavor concentration value of this fruit bat body position, and retain this fruit bat position, ρ (NO _x), ρ (SO ₂) and η _boilerand flavor concentration value;

Step 34: by iteration optimizing, retains flavor concentration minimum, and each population fruit bat optimal location now and NO _xconcentration, SO ₂concentration, boiler efficiency, and utilize the optimal location of fruit bat to calculate the value of each adjustable parameter;

Step 4: draw optimum results.